Magnetic resonance in food science : defining food by magnetic resonance

Magnetic Resonance in Food Science Defining Food by Magnetic Resonance

Magnetic Resonance in Food Science Defining Food by Magnetic Resonance

Edited by

Francesco Capozzi Department of Food Science, University of Bologna, Cesena, Italy Email: [email protected]

Luca Laghi Department of Food Science, University of Bologna, Cesena, Italy Email: [email protected]

Peter S. Belton School of Chemistry, University of East Anglia, Norwich, UK Email: [email protected]

Proceedings of the meeting XII International Conference on the Applications of Magnetic Resonance in Food Science: Defining Food by Magnetic Resonance held in Cesena, Italy 20–23 May, 2014.

Special Publication No. 349

Print ISBN: 978-1-78262-031-0 PDF eISBN: 978-1-78262-274-1

A catalogue record for this book is available from the British Library

© The Royal Society of Chemistry 2015

All rights reserved

Apart from any fair dealing for the purpose of research or private study for non-commercial purposes, or criticism or review as permitted under the terms of the UK Copyright, Designs and Patents Act, 1988 and the Copyright and Related Rights Regulations 2003, this publication may not be reproduced, stored or transmitted, in any form or by any means, without the prior permission in writing of The Royal Society of Chemistry or the copyright owner, or in the case of reprographic reproduction only in accordance with the terms of the licences issued by the Copyright Licensing Agency in the UK, or in accordance with the terms of the licences issued by the appropriate Reproduction Rights Organization outside the UK. Enquiries concerning reproduction outside the terms stated here should be sent to The Royal Society of Chemistry at the address printed on this page.

The RSC is not responsible for individual opinions expressed in this work.

Published by The Royal Society of Chemistry, Thomas Graham House, Science Park, Milton Road, Cambridge CB4 0WF, UK

Registered Charity Number 207890

Visit our website at www.rsc.org/books

Printed in the United Kingdom by CPI Group (UK) Ltd, Croydon, CR0 4YY, UK

PREFACE

The 2014 edition of the International Conference on Magnetic Resonance in Food was held in Cesena, Italy, between the 20th and the 23rd of May. This edition of the conference included six dedicated sessions on: multiscale definition of food, quantitative NMR (qNMR), foodomics, on-line non-invasive NMR (dedicated to Brian P. Hills), quality and safety, and new developments. The first international conference was held in 1992 at the University of Surrey in Guilford. Maintaining the long tradition of the Food MR Conference, the 2014 edition presented the latest technical innovations and their current and potential applications to the understanding of food, their processing and stability, and their nutritional value. Deployment of MR relaxometry, diffusometry and imaging, in both time-and frequency domain were the focus of the session dedicated to the definition of food matter at the different scales, from the nanoscopic molecular level to the microscopic compartmental scale. New methods for decoupling and hyphenation, providing "quantitative experimental conditions", were presented during the qNMR session. After the considerable interest shown during the last conference, Foodomics was an important theme in the conference. On-line and non-invasive MR makes this spectroscopic technique unique as investigation tool when working on living systems or manufacturing processes, and the dedicated session provided an overview on the state-of-art of this field of application. Finally new or unusual applications found room in the section dedicated to the new developments. We are proud to have offered a great environment for networking and sharing views and experiences with Magnetic Resonance experts from academia and industry who are committed to the utilisation of MR tools to improve our understanding of food systems.

P.Belton F. Capozzi L.Laghi

Contents

Rapid determination of food quality using steady state free precession sequences in TD-MNR spectroscopy L.A. Colnago, T.B. Moraes, T. Monaretto, F.D. Andrade

Quantitative NMR

60 MHz 1H NMR spectroscopy of triglyceride mixtures A. Gerdova, M. Defernez, W. Jakes, E. Limer, C. McCallum, K. Nott, T. Parker, N. Rigby, A. Sagidullin, A. D. Watson, D. Williamson, and E. K. Kemsley

Usefulness of 1H NMR to study the food lipolysis during in vitro digestion B. Nieva-Echevarría, E. Goicoechea, M.J. Manzanos and M.D. Guillén

Quantitative NMR assessment of polysaccharides in complex food matrices E.J.J. van Velzen, S. Dauwan, N. de Roo1, C.H. Grün, Y. Westphal, and J.P.M. van Duynhoven

Quality and Safety

Magnetic Resonance analysis of dairy processing suitable tools for the dairy industry R. Anedda

NMR spectroscopic studies in saffron authenticity and quality (within the frame saffronomics cost action FA1101) R. Consonni, L. R. Cagliani, M. G. Polissiou, E. A. Petrakis, M. Z. Tsimidou, S. Ordoudi

Food NMR optimized for industrial use-an NMR platform concept E. Humpfer, B. Schütz, F. Fang, C. Cannet, M. Mörtter, H. Schäfer, and M. Spraul

A new ultra rapid screening method for olive oil health claim evaluation using selective pulse NMR spectroscopy E. Mellioul, P. Magiatis and K.B. Killday

Profile of the positional distribution of fatty acids in the triacylglycerols as an index of quality for palm oil (or any oil or fat) S. Ng

1

19

31

40

51

65

77

84

93

viii Contents

On-line Non-invasive NMR

1H-NMR relaxometry and imaging to assess fat content on intact pork loins V. Bortolotti, P. Fantazzini, C. Schivazappa, M. Vannini, E. M. Vasini, R. Virgili

Multiscale Definition of Food 19F labelled polyion micelles as diffusional nanoprobes D.W. de Kort, F.J.M. Hoeben, H.M. Janssen, N. Bourouina, J. Mieke Kleijn, J.P.M. van Duynhoven and H.V. As

Double emulsion character with PFG-NMR- methods: WOW and OWO R. Bernewitz, E. Caro, D. Topgaard, H.P. Schuchmann, G. Guthausen

Assessment of TD-NMR and quantitative MRI methods to investigate the apple transformation processes used in the cider-making technology C. Rondeau-Mouro, S. Deslis, S. Quellec, R. Bauduin

Foodomics

A 1H NMR-based metabolomics approach on dietary biomarker research in human urine A. Trimigno, G. Picone, F. Capozzi

1H NMR metabolic profiling of apulian EVOOs: fine pedoclimatic influences in Salento cultivars L. Del Coco, S.A. De Pascali, F.P. Fanizzi

Addition of essential oils to cows’ feed alters the milk metabolome-NMR spectroscopic studies of “nature’s perfect food” U.K. Sundekilde, M.R. Clausen, J. Lejonklev, M.R. Weisbjerg, M.K. Larsen, and H.C. Bertram

High-resolution magic angle spinning studies of semi-hard Danbo (30+) cheese-impact of processing condition and relation to sensory perception S. Lamichhane, C.C. Yde, L.H. Mielby, U. Kidmose, J.R. Møller, M. Hammershøj and H.C. Bertram

Changes in the 1H NMR metabolic profiling of mussels (Mytilus galloprovincialis) with storage at 0°C V. Aru, M.B. Pisano, P. Scano, S. Cosentino and F.C. Marincola

Applications of 1H-NMR metabolomics: from individual fingerprints to food analysis A. Luchinat and L. Tenori

101

111

120

127

143

154

161

171

181

190

Contents ix

New Developments

Compost biodegradation by 1H magnetic resonance and quantitative relaxation tomography V. Bortolotti, P. Fantazzini, M. Vannini and E.M. Vasini

1H NMR spectroscopy of lipoproteins-when size matters F. Savorani and S.B. Engelsen

Subject Index

203

211

224

RAPID DETERMINATION OF FOOD QUALITY USING STEADY STATE FREE PRECESSION SEQUENCES IN TD-NMR SPECTROSCOPY

L.A. Colnago1, T.B. Moraes2, T. Monaretto3, F.D. Andrade1

1Embrapa Instrumentação, Rua XV de Novembro 1452, São Carlos-SP, 13560-970, Brazil. 2Instituto de Física de São Carlos, Universidade de São Paulo, Avenida Trabalhador São-Carlense 400, São Carlos-SP, 13566-590, Brazil. 3Instituto de Química de São Carlos, Universidade de São Paulo, Avenida Trabalhador São-Carlense 400, São Carlos-SP, 13566-590, Brazil.

1 INTRODUCTION

The use of time-domain NMR spectroscopy (TD-NMR) in food science began more than 40 years ago with the introduction of the small benchtop NMR analyzer.1 Since then, TD-NMR has become one of the most robust, rapid, cost-effective and versatile tools in the food industry. Earlier TD-NMR applications were primarily based on quantitative analysis using the intensity of free induction decay (FID) and/or spin echo signals.1-3 In the last two decades, the use of relaxometry and/or diffusometry methods have expanded the application TD-NMR in food science exponentially.2,3

The majority of these applications use the Carr-Purcell-Meiboom-Gill (CPMG) pulse sequence.1-3 This sequence is very robust4, rapid and yields an exponential decay that is dependent upon the transverse relaxation time (T2). 1-3 Therefore, CPMG has been used as an all-purpose sequence in TD-NMR applications and is a standard pulse sequence present in commercial and homemade TD-NMR spectrometers. CPMG has been used to study food products such as oilseeds, fresh meat, fish, and fruit, as well as industrialized and packaged food products.1,5,6

The longitudinal relaxation time (T1) measurements using inversion-recovery (IR) or progressive saturation pulse sequences have rarely been used in food analysis due to the length of experiment time.2,6 Pulsed field gradient spin-echo (PFGSE) pulse sequences are the second most used pulse sequence in TD-NMR applications.2 PFGSE has been used to measure the water self-diffusion coefficient, water mobility, and droplet size in several food products. However, PFGSE requires an additional hardware accessory that is not available for all TD-NMR spectrometers. Thus, there is an effort towards the development and implementation of rapid TD-NMR analytical methods that meet the growing demand for tools of quality assessment. Accordingly, we have been developing steady-state free precession (SSFP) pulse sequences for TD-NMR spectroscopy since 2000.7 SSFP sequences have been used in quantitative analysis similarly to analyses performed with FID or spin echo.7,8 However, the signal-to-noise ratio (SNR) with SSFP is much higher than that obtained with FID or echo in the same

2 Magnetic Resonance in Food Science: Defi ning Food by Magnetic Resonance

average time.7,8 Moreover, SSFP sequences can also be used in fast flow (online) quantitative measurements of liquid or solid samples.9,10 The theory for quantitative analysis using the amplitude of an SSFP signal is presented in section 2.1. Further advantages of SSFP sequences are: the dependence of the transient signals on two relaxation times (T1 and T2), the data are collected in a length of time similar to CPMG and it does not require special hardware and therefore can be implemented on any modern TD-NMR spectrometer. 1,6,8,11 The theory for the evolution of the NMR signal submitted to a train of pulses (SSFP sequence) is presented in section 2.2.

2 THEORY

2.1 Amplitude of the NMR signal in the SSFP regime

SSFP sequences have been used to improve the SNR in pulsed NMR spectroscopy since 1958.12 It is a simple pulse sequence consisting of a train of radiofrequency pulses (rf) with the same phase and flip angle ( ), and the time between pulses (Tp) is shorter than T2 (Tp < T2) (Figure 1).

Figure 1 Diagram of the SSFP pulse sequence, where n is number of rf pulses.

In 1966, Ernst and Anderson derived the analytical solution for the SSFP regime.13 They showed that the SSFP signal is composed of FID and echo signals. The echo component (M-) immediately preceding the pulse is given by equations 1 through 3, and the FID (M+) component is given by equations 4 through 6.

M x- = M0 (1-E1)[E2 sinq sinF]

D(1)

(2)

(3)

(4)

(5)

DEEEEM

M z

]cos)cos1()cos()[1( 22210 (6)

DEEEM

M y

]sincossin)[1( 22210

DEEEEM

M z

)](coscoscos1)[1( 22210

xx MM

DEEM

M y

]sin)cos1)[(1( 210

Rapid Determination of Food Quality Using Steady State Free Precession 3

where ])cos)(cos[()]cos1)(cos1[( 22121 EEEEED , with the precession

angle , offset frequency , and relaxation components

and .With these equations it is possible to calculate the magnitude of the magnetization in the xyplane after the nth rf pulse, assuming Tp << T1.14

(7)

Therefore, the amplitude of the SSFP signal is dependent upon the flip angle , precession angle pT and T1/T2 ratio.7,14 The magnetization goes to null , when

or

(8)

where n is an integer. Figure 2 shows the dependence of the NMR signal amplitude upon the precession angle and frequency offset for = 45o and 90o, Tp = 0.3 ms, T1 = 150 ms and T2 = 50 ms, according to equation 7. For , the magnetization is minimal because the FID and echo components are dephased by 180º, resulting in destructive interference. For , the FID and echo are in phase and the constructive interaction creates a maximum signal intensity when = 90o.7

Figure 2 Dependence of the normalized SSFP signal amplitude upon the precession angle and frequency offset when = 45º and 90º, Tp = 0.3 ms, T1 = 150 ms and T2 = 50 ms.

Equations 1 through 6 show that the behavior of the magnetization in the SSFP regime is complex and depends on a series of experimental parameters, such as , , , Tp, T1 and T2.However, these analytical descriptions do not include the effect of other parameters on the SSFP signal, such as Tp variation and phase alternation. To fully describe the SSFP phenomenon we have numerically simulated (Matlab)15 the

= t 0ref )exp( 11 TTE p

)exp( 22 TTE p

| M |= M0 | sin(q ) | 2-2cosF(1+ cosq )(1- cosF)+ (1- cosq )2T1 / T2

0|| M

0cos22

F= n2p

F = n2p)12( n


influence of the all the above parameters based on the rotation matrix and the method of the sum of isocromats16, in which the Lorentzian distribution was assumed9.

(9)

and

(10)

where .Figure 3 shows the numerical simulations for the TD-NMR signals after reaching the steady state regime. The time necessary to reach the steady state is discussed in section 2.2. The steady state signals were simulated using T1 = 100 ms, T2 = 50 ms, T2* = 0.5 ms, = 90o, a frequency offset of 8.333 (Figures 3A to D) and 6.666 (Figure 3E) KHz and various Tpvalues. In Figure 3, the pulse is observed in the center of the window (t = 0). The FID component after the pulse is on the right side of t = 0, and the echo component is on the left side of the pulse. Figure 3A shows the NMR signal for Tp = 5T1. This figure shows an FID signal with maximum amplitude. With this pulse repetition rate, the echo signals are not observed. Figure 3B shows the NMR signal for Tp = T2. In this condition the NMR signal is in the SSFP regime and is composed of an FID and an echo between the pulses. The FID signal has higher amplitude than the echo signal. The FID amplitude in Figure 3B is lower than that of the FID in Figure 3A because Tp<5T1, which does not allow the return of the magnetization to thermal equilibrium. Figures 3C to E depicts more than one period between the pulses, in the interval of -1.5 to 1.5 ms. Figure 3C shows two periods for Tp = 2.9T2* in which the FID and echo signals have similar amplitudes, and the FID decays faster than T2* compared to the FID decay in Figures 3A and B. This faster decay is due to the partial destructive interaction between the FID and echo in the center of the SSFP signals. Figures 3D and E show the SSFP signals for Tp = 0.3 ms < T2*. In this condition the overlap between the FID and the echo signal is maximal, yielding a special SSFP regime, known as Continuous Wave Free Precession (CWFP).6,10 The amplitude of CWFP signal is strongly dependent upon pT , as shown in Figure 2. Figure 3D depicts the maximum CWFP signal when =5 with a frequency offset of 8.333 KHz (constructive interference) and Figure 3E depicts the minimal CWFP signal when =4 with a frequency offset of 6.666 KHz (destructive interaction). According to equation 11, the magnitude of the CWFP signal when = 90o and = 5 is dependent upon the T1/T2 ratio. 1,6,11

(11)

2*2

20

*2

)(1)/()0(

TTg

inomTTT 22*

2

111

02 /1 BT inom

2

1

0

1||

TT

MM ss


Figure 3 NMR signals simulated numerically using T1 = 150 ms, T2 = 50 ms, T2* = 0.5 ms and several Tp values. A) Tp = 5T1, B) Tp = T2, C) Tp = 2.9T2*, D) and E) Tp < T2*. The frequency offset is 8.333 KHz (A to D) and 6.666 KHz (E).

The magnitude of the CWFP signal is not dependent on the pulse repetition rate, as in conventional pulse sequences (Figure 3A and B). Instead, it depends on the T1/T2 ratio (equation 11), and the repetition time can be short (Tp << T1, T2 < T2*) and without saturation (Figure 3D). Therefore, thousands of CWFP signals can be averaged during one T1 period, thereby enhancing the SNR by one order of magnitude in the same average time used for FID or echo signals.7,8 The magnitude of CWFP signals has been used in quantitative analysis, in conventional benchtop spectrometers and in online measurements using long Halbach and superconducting magnets1,6,9.


2.2 Transient SSFP signal

The evolution of the NMR signal, submitted to a train of pulses (SSFP sequence) has been used in several applications in TD-NMR. In 1977, Kronenbitter and Schwenk proposed the use of the transient SSFP signal to measure T1 and T2.17 The method consists of two steps: First, the measurements of the T1/T2 ratio, by measuring the maximum amplitude of the SSFP signal as a function of the flip angle ( ) to obtain the optimum ( opt), and second, the use of

opt to measure the time constant (T*), equation 12, for the evolution of the SSFP signal, yielding the T1+T2 value. With these two measurements it is possible to determine both relaxation times.17

(12)

In 2006, Venâncio T. et al reported that is not necessary to use two steps to measure both relaxation times when the transient SSFP signal is obtained with = 90o and

.11

Figure 4 shows the evolution of the magnitude of the CWFP signal from the first pulse to the stationary regime (|Mss|). This signal undergoes two transient regimes before the steady state is reached. The first transient regime (dark grey) shows an alternation of the amplitude between even and odd pulses followed by signal decay, with the time constant of T2*. When the alternations subside, the signal reaches a quasi-stationary state (light grey). The decay of the quasi-stationary state to the stationary state (white) is governed by the time constant T*. For = 90o, equation 12 is reduced to:

(13)

Figure 4 Evolution of the CWFP signal magnitude from the first pulse to the stationary state (Mss).

)cos1()cos1(2

*21

21

TTTT

T

)12( n

21

212*

TTTT

T


Upon rewriting equations 11 and 13, we obtain the following:

and (14)

Therefore, upon measuring the magnitude of the signal after the first pulse || 0M , the magnitude of the CWFP signal and T*, it is possible to calculate the relaxation times in a single scan experiment using equation 14.11 The T* value is calculated by fitting the T* decay with an exponential decay function. The T1 and T2 values are obtained with single CWFP experiments are similar to those obtained by Inversion recovery (T1) and CPMG (T2)pulse sequences.11

When T1 ~ T2 there is only a small difference in amplitude between the quasi-stationary state and the stationary state of the CWFP signal. This might yields, a T* with large error when the CWFP signal has low SNR.4

To solve this problem we proposed the use of a Carr-Purcell sequence, using 90o-refocusing pulses, also known as CP-CWFP (Figure 5).4 The only difference between CWFP and CP-CWFP sequences is the addition of a pulse, which separates the CWFP pulse train (Figure 1) by the time interval Tp/2. The effect of this modification is shown in Figure 6. The CP-CWFP signal intensity decays to a minimum value (quasi-stationary state) and then increases to the same amplitude observed in the CWFP regime. As shown in Figure 6A, the amplitude variation during T* is much more pronounced in CP-CWFP than in the CWFP sequence. This results in improved fitting of T* for a sample when T1 ~ T2.Conversely, CP-CWFP signals yield a small difference in amplitude during T* when T1 >> T2 (Figure 6B). In this case, T* of CWFP can be fitted with minimal error. When T1 > T2(Figure 6C) CWFP and CP-CWFP have similar amplitude variations during T*.

Figure 5 Diagram of CP-CWFP pulse sequence.

In addition to measuring the relaxation times, the CWFP or CP-CWFP signals have been used to obtain qualitative and quantitative information from food products using uni- and multivariate analyses.1,5,18-20 The ratio |Mss|/M0 of the CWFP signal has a higher correlation with intramuscular fat content and water loss during cooking (cooking loss) in beef than T2 measured by CPMG.18,19

|Mss|/M0 also has a higher univariate correlation with animal sex and genetics than CPMG.20

However, the multivariate analysis results in similar beef classification (sex and genetics) using the full CPMG and CWFP data sets.20

0

*

1 ||2

MM

TT

ss

0

*

2 ||1

2

MM

TT

ss

|| ssM


Figure 6 Experimental CWFP and CP-CWFP signal intensities for tap water (A), hotdog sausage (B) and mayonnaise (C) samples.


CWFP and CP-CWFP sequences are a useful alternative to CPMG because the analysis can be performed in the same amount time, they do not require special hardware and, therefore, the sequences can be implemented in any modern TD-NMR spectrometer. The scripts for the CWFP and CP-CWFP sequences for the Minispec spectrometer (BRUKER) are described in the appendix I.

3 COMPARISON BETWEEN CPMG, CWFP AND CP-CWFP ANALYSES OF FRESH AND PROCESSED FOOD PRODUCTS

To demonstrate the potential of CWFP and CP-CWFP analyses in food science we compared them with CPMG, using a benchtop spectrometer, SLK 100, Spinlock (Córdoba, Argentine). Figure 7 shows the CPMG, CWFP and CP-CWFP signals for ripe and unripe grapes and bananas. Figures 7A and B show the CPMG and CWFP/CP-CWFP signals of ripe (brix = 18) and unripe (brix = 10) grapes. The CPMG decays for both grapes were similar. However, the ripe grapes, with a high sugar content, shows a more rapid CPMG decay (1.0 s) than the unripe grapes (1.2 s), with low sugar content, as expected.21 Similar results for ripe and unripe grapes were also observed for CWFP and CP-CWFP signals (7B). The small and large variations in the amplitude during T* for CWFP and CP-CWFP, respectively, and similar |Mss| amplitudes (0.37 and 0.4, respectively) indicate that T1 ~ T2 for both grapes. Conversely, the CPMG, CWFP and CP-CWFP signals (Figures 7C and D) of ripe and unripe bananas are much more distinct. The decay of the CPMG signal (7C) of unripe bananas (0.20 s) is longer than the decay of ripe bananas (0.09 s). This is observed elsewhere and is related to the large difference between the consistencies of bananas in these two conditions.22

This difference between ripe and unripe bananas was also observed in CWFP and CP-CWFP signals (Figure 7D). The variation in the amplitude of T* decay and |Mss| (0.40 and 0.24, respectively) indicates that T1 > T2 in ripe bananas and that T1 >> T2 in unripe bananas. Similar results are also observed for processed food products. Figure 8 shows the CPMG, CWFP and CP-CWFP signals for hotdog sausages (two brands) (Figures 8A and B) and regular and light mayonnaises (Figures 8C and D). The CPMG, CWFP and CP-CWFP signals (Figures 8A and B) of the two hotdog sausages were similar. The duration of the CPMG signal is longer for brand I sausage (0.062 s) than brand II (0.056 s), indicating a minimal difference in the composition of the two sausages. The T* and |Mss| amplitude (0.16) of the two sausages indicate that T1 >> T2.Figures 8C and D depict the CPMG, CWFP and CP-CWFP signals of the regular (33% fat) and light (24% fat) mayonnaise samples from the same brand. Although the fat content in the light mayonnaise is 30% lower, the duration of the CPMG signal is only slightly longer (0.15 s) than in regular mayonnaise (0.14 s).7 However, the T* and Mss of CWFP and CP-CWFP analyses are more sensitive to the fat content in mayonnaise. The |Mss| of regular mayonnaise is lower (0.22) than the light mayonnaise (0.28), thus, reflecting the difference in fat content. It is possible to process the full CWFP and CP-CWFP data using multivariate analysis; this would provide more information about the products and improved calibration and classification models compared to univariate analysis.5,7,20


Figure 7 NMR signals intensities obtained for ripe and unripe grapes (A and B) and bananas (C and D) using CPMG, CWFP and CP-CWFP pulse sequences.

Figure 8 NMR signal intensities obtained from two brands of hotdog sausages (A and B), and regular and light mayonnaise (C and D) using CPMG, CWFP and CP-CWFP pulse sequences.


4 CONCLUSION

Given the results presented, we conclude that the SSFP pulse sequences, using CWFP and/or CP-CWFP regimes, are alternatives and/or complementary methods to CPMG. The SSFP sequences are more efficient than CPMG for samples in which the change in T1 is greater than T2. Furthermore, these sequences can be implemented in modern TD-NMR spectrometers.

Acknowledgements

This work was supported by FAPESP (process # 2011/11160-3, 2012/20247-8, 2013/03770-1 and 2011/14099-3) and CNPq (Brazilian agencies). We are grateful to Professor Eduardo Ribeiro de Azevedo (IFSC- University of São Paulo) and Dr. Márcio Fernando Cobo (Bruker BioSpin) , for the development of CWFP and CP-CWFP pulse sequences for minispec.

References

1 L.A. Colnago, R.B.V. Azeredo, A. Marchi-Netto, F.D. Andrade, T. Venâncio, Magn. Reson. Chem., 2011, 49, S113.

2 J. Van Duynhoven, A. Voda, M. Witek, H. Van As, Annu. Rep. NMR Spectrosc., 2010, 69, 145.

3 F. Dalitz, M. Cudaj, M. Maiwald, G. Guthausen, Prog. Nucl. Magn. Reson. Spectrosc.,2012, 60, 52.

4 F.D. de Andrade, A. Marchi-Netto, L.A. Colnago, Talanta, 2011, 84, 84. 5 F.M.V. Pereira, A.P. Rebellato, J.A.L. Pallone, L.A. Colnago, Food control, 2014 (in

press), DOI:10.1016/j.foodcont.2014.02.028 6 L.A. Colnago, F.D. Andrade, A.A. Souza, R.B.V. Azeredo, A.A. Lima, L.M. Cerioni, T.

M. Osán, D.J. Pusiol, Chem. Eng. Technol., 2014, 37,191.7 R.B.V. Azeredo, L.A. Colnago, M. Engelsberg, Anal. Chem., 2000, 72, 2401. 8 R.B.V. Azeredo, L.A. Colnago, A.A. Souza, M. Engelsberg, Anal. Chim. Acta, 2003,

478, 313. 9 R.B.V. Azeredo, M. Engelsberg, L.A. Colnago, Phys. Rev. E, 2001, 64, 16309. 10 L.A. Colnago, M. Engelsberg, A.A. Souza, L.L. Barbosa, Anal. Chem., 2007, 79, 1271. 11 T. Venâncio, M. Engelsberg, R.B.V. Azeredo, N.E.R. Alem, L.A. Colnago, J. Magn.

Reson., 2005, 173, 34. 12 H.Y. Carr, Phys. Rev., 1958, 112, 1693. 13 R.R. Ernst, W.A. Anderson, Rev. Sci. Instrum., 1966, 37, 93. 14 A. Schwenk, Prog. NMR Spectrosc., 1985, 17, 69. 15 MATLAB; version 7.10.0 (R2010a), MathWorks Inc., Natick, Massachusetts, EUA,

2010.16 P. Shkarin, R. G. S. Spencer, Concepts Magn. Reson., 1996, 8, 253.17 J. Kronenbitter, A. Schwenk, J. Magn. Reson., 1977, 25, 147.18 C.C. Correa, L.A. Forato, L.A. Colnago, Anal. Bioanal. Chem., 2009, 393, 1357. 20 F.M.V. Pereira, S.B. Pflanzer, T. Gomig, C.L. Gomes, P.E. Felício, L.A. Colnago,

Talanta, 2013a, 108, 88. 20 P.M. Santos, C.C. Correa, L.A. Forato, R.R. Tullio, G.M. Cruz, L.A. Colnago, Food

control, 2014, 38, 204. 21 F.M.V. Pereira, A.S. Carvalho, L.F. Cabeça, L.A. Colnago, Microchem. J., 2013b, 108,

14.


22 F.Z. Ribeiro, L.V. Marconcini, I. B. Toledo, R.B.V. Azeredo, L.L. Barbosa, L.A. Colnago, J. Sci. Food Agric., 2010, 90, 2052.

APPENDIX I

#-CWFP final version- uses signal averaging for processing data# #Eduardo Ribeiro de Azevedo: [email protected]#

program setup();

#----------------------Get main parameter from parameter table------------#

par; # starting parameter definition # scans (16); rd (1.000000000); gain (66); dbw (20000.000000); abw ("broad"); off_comp ("off"); det_mode ("complex"); magn_mode ( "PSD" ); # phase sensitive magnitude detection # dig_res ("fast"); endpar; # end of parameter definition #

return(TRUE);

#----------------------Application configuration table--------------------#

program config();

int temp_int; real temp_real; char temp_string[128];

strcpy( temp_string, get_text(CALIBRATION_FILE,"fname" )); if(ERROR) set_conf (CI_INPUT,TRUE,"File Name","F:\usuarios\Eduardo\Sequencias"); else set_conf (CI_INPUT,TRUE,"File Name",temp_string); endif;

temp_real = get_real(CALIBRATION_FILE,"tauus"); if(ERROR) set_conf (CI_INPUT,TRUE,"Delay tau [us]: ","300"); else set_conf (CI_INPUT,TRUE,"Delay tau [us]",temp_real); endif;

temp_int = get_int(CALIBRATION_FILE,"npi"); if(ERROR) set_conf (CI_INPUT,TRUE,"Number of 90 deg. pulses in cwfp","100"); else set_conf (CI_INPUT,TRUE,"Number of 90 deg. pulses in cwfp",temp_int); endif;

temp_int = get_int(CALIBRATION_FILE,"flag_cpcwfp"); if(ERROR) set_conf (CI_SELECT,TRUE,"Activate CP-CWFP",FALSE); else if(temp_int==1) set_conf(CI_SELECT,TRUE,"Activate CP-CWFP",TRUE);

else set_conf(CI_SELECT,TRUE,"Activate CP-CWFP",FALSE); endif; endif;

temp_int = get_int(CALIBRATION_FILE,"savesig"); if(ERROR) set_conf (CI_SELECT,TRUE,"Save Signal",FALSE); else if(temp_int==1) set_conf(CI_SELECT,TRUE,"Save Signal",TRUE);

else set_conf(CI_SELECT,TRUE,"Save Signal",FALSE); endif;


endif;

set_conf (CI_TEXT,TRUE,"File Name and Delays"); get_conf ("Options","FID/ECHO aquisition",0); if (ESC) goto escape; endif;

print_line (CALIBRATION_FILE,"fname", tst_conf (CI_INPUT,0)); print_line (CALIBRATION_FILE,"tauus", ator ( tst_conf (CI_INPUT,1) )); print_line (CALIBRATION_FILE,"npi", atoi (tst_conf (CI_INPUT,2) )); print_line (CALIBRATION_FILE,"flag_cpcwfp", tst_conf (CI_SELECT,0)); print_line (CALIBRATION_FILE,"savesig", tst_conf (CI_SELECT,1));

label escape; return(TRUE);

#--------------------Initialize parameters for pulse sequence-------------#

program measure(); real acq, tau, tauus, pw, pw1, rdt,d1,gainr; real x_array[3204800], y_array[3204800],yi_array[3204800]; int npi, cnt, nsc, ndp,flag_cpcwfp,save_flag; int ph90[20], ph90cp[20],phrc[20]; charhlp fname[256], name[256], name1[256];

#---specific parameters ------------#

strcpy ( fname, get_text(CALIBRATION_FILE,"fname" )); acq = 20/1000.0; # fixed acq time of 20 us with asd# tauus = get_real(CALIBRATION_FILE,"tauus"); npi = get_int(CALIBRATION_FILE,"npi"); flag_cpcwfp = get_int(CALIBRATION_FILE,"flag_cpcwfp"); save_flag = get_int(CALIBRATION_FILE,"savesig");

#---global parameters ------------#

pw = get("90P"); rdt = get ("RDT"); d1 = get_rd; gainr = get_gain; nsc=get_scans;

print_line(RESULTBOX, "acq = ", acq, " ms. " );

#--- Experiment Messages------------#

print_line( RESULTBOX, "------------------------" ); print_line( RESULTBOX, " " );

if(flag_cpcwfp==1)

print_line( RESULTBOX, "(CP-CWFP is activated (CP-CWFP)");

else

print_line( RESULTBOX, "Doing Standard CWFP experiment");

endif

print_line( RESULTBOX, " " );

print_line( RESULTBOX, "------------------------" );


# ------------convert variables units for calculatations---------------- # tau=tauus/1000; # convert to ms # pw1 = pw/1000; # convert to ms #

#-----------------initialize counters------------------------------------# cnt = 0;

#----Check for time limits------#

if ((tau - pw1/2.0 - acq/2 ) < rdt) beep;

print_line(RESULTLINE, "Error, aquisition starts inside dead time. Please, increase tau to be greater than: ", round((pw1/2.0 + acq/2 + rdt)*1000), " us. " ); goto stopseq; endif;

if ((pw > 30)) beep;print_line(RESULTLINE, "Pulse lenght exceed the maximum limit. Please, decrease de pulse length to less than 30 us"); goto stopseq; endif;

if (npi > 30000) beep;

print_line(RESULTLINE, "Too many pulses, please reduce npi to less than 30000"); goto stopseq; endif;

#------------------- phase cycling --------------------------------------#

ph90[0] = 0; ph90cp[0] = 0; phrc[0] = 0; ph90[1] = 90; ph90cp[1] = 90; phrc[1] = 90; ph90[2] = 180; ph90cp[2] = 180; phrc[2] = 180; ph90[3] = 270; ph90cp[3] = 270; phrc[3] = 270; ph90[4] = REDO; ph90cp[4] = REDO; phrc[4] = REDO;

#-------------------Start Actual pulse sequence---------------------------#

pulses;

sd (1000e-3); # first delay for minimum time durantion of the sequence #

cta;

if(flag_cpcwfp==1)

ssp ( pw, ph90cp); # 90 degree pulse for CPCWFP # sd(rdt);

asd (acq, phrc); # acq/2 delay is added to compensate for an extra acq/2 delay in the first pulse of the cwfp loop #

sd ( tau/2.0 - pw1/2. - rdt - acq ); else

ssp ( pw, ph90); # 90 degree pulse for CWFP # sd(rdt); asd (acq, phrc); # data acquisition #


sd ( tau - pw1/2. - rdt - acq ); # acq/2 delay is added to compensate for an extra acq/2 delay in the first pulse of the cwfp loop #

endif

ploop(npi-1) # CWFP loop#

ssp ( pw, ph90); # 90 degree pulse for CWFP # sd ( tau/2.0 - pw1/2. - acq/2.); asd (acq, phrc); # data acquisition # sd ( tau/2.0- pw1/2. - acq/2.); cnt=cnt+1; # loop counter #

endploop;

endpulses; # end of pulse sequence #

measure; # begin of data evaluation #

#---------------------------------Saving echoes am------------------------# sig_abscissa(-1,-1,x_array); sig_ordinate(-1,-1,y_array); # write real part of the echoes amplitudes in y_array# sig_swap; sig_ordinate(-1,-1,yi_array); # write imaginary part of the echoes amplitudes in yi_array# sig_swap;

#-------------------Saving decay curve------------------------------------#

if (save_flag == 1)

strcpy (name,fname); strcat(name, ".dat"); file_name (ASCII_FILE, name); ndp = data_points( -1, -1 ); if(flag_cpcwfp==1) ndp=ndp-1;

endif print_table(ASCII_FILE, x_array, y_array, yi_array, ndp);

#-------------------Saving acquisition parameters-------------------------#

# Show in Resultbox#

print_line( RESULTBOX, "---------------------------------------------" ); print_line( RESULTBOX, "Data file: ",fname); print_line( RESULTBOX, "---------------------------------------------" ); if(flag_cpcwfp==1) print_line( RESULTBOX, " Sequence: ","CP-CWFP"); else print_line( RESULTBOX, " Sequence: ","Standard CWFP"); endif


print_line( RESULTBOX, "---------------------------------------------" ); print_line( RESULTBOX, " Parameters" ); print_line( RESULTBOX, "---------------------------------------------" ); print_line( RESULTBOX, " " ); print_line( RESULTBOX, "pw = ", pw," us");

print_line( RESULTBOX, "acq = ", acq," ms"); print_line( RESULTBOX, "tau = ", tauus," us"); print_line( RESULTBOX, "d1 = ", d1," s"); print_line( RESULTBOX, "number of scans = ", nsc); print_line( RESULTBOX, "number of echoes = ", npi); print_line( RESULTBOX, "receiver gain = ", gainr);

# save in name_parameters name_parameters.txt file#

strcpy (name1,fname); strcat(name1, "_parameters.txt"); file_name (ASCII_FILE, name1); print_line( ASCII_FILE, "--------------------------------------------" ); print_line( ASCII_FILE, "Data file: ",fname); print_line( ASCII_FILE, "--------------------------------------------" ); if(flag_cpcwfp==1) print_line( ASCII_FILE, " Sequence: ","CP-CWFP"); else print_line( ASCII_FILE, " Sequence: ","Standard CWFP"); endif print_line( ASCII_FILE, "--------------------------------------------" ); print_line( ASCII_FILE, " Parameters" ); print_line( ASCII_FILE, "--------------------------------------------" ); print_line( ASCII_FILE, " " ); print_line( ASCII_FILE, "pw = ", pw," us");

print_line( ASCII_FILE, "acq = ", acq," ms"); print_line( ASCII_FILE, "tau = ", tauus," us"); print_line( ASCII_FILE, "d1 = ", d1," s"); print_line( ASCII_FILE, "number of scans = ", nsc); print_line( ASCII_FILE, "number of echoes = ", npi); print_line( ASCII_FILE, "receiver gain = ", gainr);

endif

#--------------------finishing------------------------------------------# beep;if ( ESC ) print_line( CONFIRMBOX, "USER INTERRUPT !" ); return( FALSE ); endif; label stopseq; return( TRUE );

Quantitative NMR

60 MHz 1H NMR SPECTROSCOPY OF TRIGLYCERIDE MIXTURES

A. Gerdova,1 M. Defernez,2 W. Jakes,2,3 E. Limer,2,4 C. McCallum,2,5 K. Nott,1 T. Parker,2,3 N. Rigby,2 A. Sagidullin,1 A. D. Watson,2 D. Williamson,1 and E. K. Kemsley2 1Oxford Instruments Industrial Analysis, Tubney Woods, Abingdon, Oxford OX13 5QX, UK 2Institute of Food Research, Norwich Research Park, Norwich NR4 7UA, UK 3School of Chemistry, University of East Anglia, Norwich Research Park, Norwich NR4 7TJ, UK 4Oriel College, University of Oxford, Oxford OX1 4EW, UK 5Worcester College, University of Oxford, Oxford OX1 2HB, UK

1 INTRODUCTION

Triglycerides (more properly triacylglycerols, abbreviated to TAGs) are esters of glycerol comprising a glyceride backbone with three fatty acids. Natural chemical diversity is conferred by the fact the three acyl residues need not be the same. Triglycerides are of great economic and nutritional importance: edible oils and fats consist almost entirely of triglycerides. Food triglycerides have fatty acid chains ranging in length from 4 to 24. These chains may contain 0 (saturated), 1 (mono-unsaturated), 2 or 3 (collectively, poly-unsaturated) carbon-carbon double bonds. The chain length 18 is particularly abundant: for example the average fatty acid composition of olive oil is 75.5% by weight1 of C18:1 (referring to the chain length of the fatty acid component with a single double bond) and 7.5% by weight C18:2 (two double bonds). The chain length C16 is also relatively common, with olive oil at 11.5%w/w C16:0. Despite the fact that certain chain lengths are dominant, different edible oils and fats have markedly different triglyceride compositions. Any technique capable of analysing triglyceride mixtures is therefore able to establish key compositional properties of pure oils and fats. In addition, such techniques may potentially be able to detect the addition of one oil to another, as occurs when a high-value oil such as olive oil is fraudulently adulterated with a cheaper substitute. There are numerous ways of analysing the TAG content of oils and fats, in particular HPLC2 and GC3. However the presence of many hydrogens in the different environments resulting from various double bond configurations implies a role for proton NMR. Indeed high-field 1H NMR has been used extensively to study TAGs, edible oils,4-7 and mixtures of edible oils.8 Here, we outline recent results on triglyceride mixtures obtained using a new, low-field 1H NMR spectrometer called Pulsar. Developed at Oxford Instruments (Oxford Instruments, Tubney Woods, Oxford, UK), the Pulsar is based on permanent magnets rather than the superconducting magnets standard in modern high-field instruments. It has a field strength of 1.4 T, corresponding to a 1H Larmor frequency of approximately 60 MHz, is cryogen-free and has a bench-top footprint. Compared to early NMR instruments of similar field strengths the Pulsar gains from improved hardware, electronics, and numerically-intensive chemometrics-based software.


Figure 1 Simulated 1H NMR data for ethyl crotonate for 60 and 300 MHz. Expressed in Hz, panel A, the scale for 60 MHz data is relatively compressed. Expressed in ppm, panel B, the more usual representation of NMR data, the same simulation shows broader peaks for 60 than 300 MHz over a common ppm axis Operating at such a low-field has implications for the resulting spectra. Figure 1 shows simulated spectra for ethyl crotonate at 60 and 300 MHz. In the upper panel (Figure 1A) the spectra are displayed on a frequency (Hz) scale. This emphasizes how the 60 MHz spectrum appears compressed compared to the 300 MHz spectrum; linewidths here are the same. For real rather than simulated spectra, the Pulsar line width (FWHM) is better than 1 Hz, comparable with ~0.4 to 0.8 Hz for high-field spectrometers. To aid comparison between data obtained from spectrometers operating at different field strengths it is usual practice to display spectra on the field-independent chemical shift scale,

Quantitative NMR 21

expressed as parts per million (ppm). Displayed on such a scale (Figure 1B), the 60 MHz peaks appear broadened compared to high-field spectra. Typically, the peaks in low-field spectra are more overlapped, which means that resolving individual resonances can be more challenging. Inter-peak spacings may also change in ways that might seem surprising at first sight. Comparisons with existing libraries of high-field data, useful for identifying peaks,9 must therefore be done with care.

Figure 2 60 MHz 1H NMR spectrum of linola, a type of linseed oil having low alpha-linolenic content. The inset shows an illustrative triglyceride having 3 distinct acyl residues, C18:0 (saturated), C18:1 (mono-unsaturated) and C18:3 (tri-unsaturated). Arrows link spectral and associated structural features. This Pulsar spectrum has been subject to reference deconvolution The 60 MHz 1H NMR spectrum (Figure 2) of an exemplar seed oil, edible linseed oil (linola), shows a number of characteristic peaks. The glyceride backbone contributes a peak at ~4.1 ppm due to hydrogens attached to the two end carbons on the backbone. This peak is present for all triglycerides. A double bond –CH=CH– in the acyl chain contributes an ‘olefinic’ peak at ~5.2 ppm. Note that the central carbon from the glyceride backbone also contributes to this peak. Towards the lower ppm range is the bis-allylic peak at ~2.7 ppm, arising from protons attached to a carbon sandwiched between two double bonds, =CH–CH2–CH=. This peak is an important flag for poly-unsaturated TAGs. The zone around 2.0 ppm contains contributions from the carboxyl end of the chain, –OCO–CH2–, and double bond allylic groups –CH2–CH=. The dominant peak at ~1.3 ppm is from methylene bridges, –CH2–, remote from carbon double bonds. Finally, the terminal –CH3 methyl group contributes a peak at ~0.9 ppm. The position of this peak is sensitive to the proximity of a nearby C=C double bond, which to a good approximation in edible oils and fats occurs in linolenic acid (actually, the -linolenic isomer) which has three double bonds.10 Therefore, the terminal methyl peak assumes great significance as a marker of triply-unsaturated fatty acid chains. In 1H NMR, the area of a peak is proportional to the number of contributing protons, offering a direct route to quantitative analysis. This is a distinct advantage over the competing technique of infrared spectroscopy in which quantitation generally requires a calibration step.


Furthermore, since the glyceride peak appears for every TAG molecule, it provides an in-built normalization. Quantification of triglycerides using peak areas has a long history.6, 8, 11 Despite the attractions, peak area quantitation faces several challenges. Reliable peak areas can be difficult to extract due to overlapping peaks, baseline distortion and poor phase correction. An alternative is to follow the infrared example and use chemometric methods that analyse entire spectral profiles.12, 13 These methods are particularly suited to the classification of a dataset into groups, even when to the eye the underlying spectra are quite hard to tell apart. Authentication questions are naturally of this type: ‘is a set of supplied edible oils of the same type as an existing set?’, for instance. The outcome is then a decision (‘authentic’ or ‘non-authentic’) rather than a composition table, which is the more natural outcome of peak area estimation. In this work we will demonstrate examples of both peak area estimation and chemometric methods applied to 60 MHz 1H NMR data acquired from triglyceride systems. In addition, sample preparation and data acquisition protocols are kept as simple and cheap as possible, simulating high-throughput regimes consistent with the use of low-field NMR in a screening or quality control capacity. We now describe three ‘case studies’ that reflect this philosophy.

2 METHODS AND RESULTS

2.1 Composition of Edible Oils and Complex Foods Twenty four edible oils (18 different types, including some mixtures) from local supermarkets were each combined with approximately equal amounts of non-deuterated chloroform to give 50:50 mixtures in standard 5 mm disposable NMR tubes. The chloroform is useful in this context both to reduce sample viscosity, giving reduced line width, and to provide a reference ppm value. Data was acquired on the Pulsar with a sample temperature of 37 °C using 16 scans with an acquisition time of 30 seconds per scan, giving an overall data acquisition time of ~10 min and a trivial sample preparation step. The resulting spectra were processed using peak area estimation coded in-house in Matlab (The Mathworks, Cambridge, UK). The relationships between the different peaks mean that they must be calculated in sequence, beginning with the terminal methyl group peak linked to tri-unsaturated acyl chains, so-called omega-3, followed by the bis-allylic peak, followed by the olefinic peak. At each stage the estimate depends on the stage before, meaning that errors tend to accumulate and render estimations of saturated content least robust of all. Figure 3 shows the results of peak area estimation for poly-unsaturated, mono-unsaturated, omega-3 and saturated fatty acid contents. These results are plotted against GC-FID data acquired in the conventional way following a methyl esterification of the test oils, and ~1 hour GC-FID run per oil sample. The highest omega-3 content is found to be in hemp oil, followed by walnut then rapeseed. The highest poly-unsaturated content is, in descending order, hemp, walnut then grapeseed. The highest unsaturated level is rice bran oil. The key point, however, is that the NMR results tally very well with the GC-FID data, validating the low-field NMR method and the peak area data analysis. The apparently greater scatter in the saturated data points is due in part to the chain of calculation mentioned above, but is also an artefact of the different plot scales. Developments are already in progress that will improve the agreement between the two approaches.

Quantitative NMR 23

Figure 3 Comparison of 24 edible oils analysed by GC-FID (horizontal axis) and 60 MHz 1H NMR. The NMR values rely on peak area estimation. Note the different scales The ‘simple sample preparation’ paradigm was pushed to the extreme in an attempt to extend the approach to much more complex foods. Here, ‘complex foods’ means multi-component food products, rather than simple homogeneous ingredients. For this study, we took a complex food, such as a pork pie, homogenised it in a blender, and mixed the resulting paste with chloroform for a few minutes. The resultant mixture was filtered to remove particulate material and the filtrate analysed by low-field NMR using the same parameters as for the edible oils. Once again the data was processed using peak area measurement. The results for the saturated fat content of 13 disparate products are shown in Figure 4 plotted against the so-called ‘typical’ values as stated on the product label. What is striking from these results is the acceptable level of agreement. The choice of foods was uncompromising: apart from pork pies, the analysis included: crisps, which are hard and brittle; salami, which contains macroscopic pieces of fat; and cakes topped with flakes of almond. The TAG extraction protocol was deliberately crude. The results show scatter when compared to label values, but this is likely to reflect sampling effects given the highly heterogeneous nature of the food products. Clearly, the broad level of agreement indicates this to be a valid avenue for further study, and the next step will be to compare label and NMR values with those from GC-FID.


Figure 4 Saturated fat content as determined by 60 MHz 1H NMR versus the label saturated fat value for a variety of complex foods. RS = raspberry slice (4); CR=croissant; MP=sweet mince pie; SR=sausage roll; PP=pork pie (3); AS=almond slice (4); SE=scotch egg (3); HC=hand-cooked crisps; SC=standard crisps; TM=taramasalata; HM=hummous; SM=salami; CS=chicken samosa. The number in brackets is the number of replicates 2.2 Authentication of Edible Oils

Another interesting avenue for future exploration is the potential for detecting adulteration of edible oils with lard, an issue of concern to certain faith groups. Lard, derived from pigs, is relatively cheap, making it a candidate adulterant for padding more expensive oils. Lard contains large high levels of unsaturated and mono-unsaturated fatty acids: the average fatty acid composition is 24% 16:0, 14% 18:0, 43% 18:1 and 9% 18:2 (plus other small contributions).1 Adulteration of commonly used oils such as sunflower (high in 18:2), olive (high in 18:1) and rapeseed (high in 18:1 and 18:3) should be relatively easy to detect. Figure 5A shows a region of the spectra of lard and sunflower oil. The spectra are normalised to give the same maximum height for the glyceride peak at ~4.2 ppm. The higher levels of poly-unsaturated fatty acid in sunflower oil are presaged by the larger olefinic peak at a chemical shift of ~5.2 ppm and apparent from the larger bis-allylic peak at ~2.7 ppm. We subsequently analysed a set of lard:sunflower oil mixtures by combining them with (deuterated) chloroform and submitting them to low-field NMR using the same settings as above. Even the simple strategy of integrating just the bis-allylic peak is sufficient to show that low-field 1H NMR is a viable candidate for detection of lard adulteration of important edible oils (Figure 5B).

Quantitative NMR 25

Figure 5 Lard:sunflower oil mixtures. Panel A shows the 60 MHz 1H NMR spectra for pure lard and pure sunflower oil, on a common scale, for the region 1.7 – 5.7 ppm. The bis-allylic peak at ~ 2.7 ppm is a useful discriminator. Panel B shows the integrated bis-allylic peak area as a function of lard: sunflower oil composition (%w/w). These spectra have been subject to reference deconvolution More challenging is the detection of the adulteration of relatively expensive olive oil with hazelnut oil. This is a consequence of the fact that the compositions of these two oils are very similar, even though hazelnut oil is typically listed as containing higher levels of poly-unsaturated fatty acids.2 Twenty extra virgin olive oils and ten hazelnut oils were purchased from local supermarkets, mixed with chloroform and analysed by low-field NMR broadly as described above. For each pure oil, the mono-unsaturated fraction is plotted against the poly-unsaturated fraction as calculated by peak area estimation (Figure 6). For both oil types, the range of poly-unsaturated values spans ~10% w/w and that of mono-unsaturated ~18% w/w, in keeping with the considerable spread shown in literature values. Despite the variation in levels, however, combining these two simple peak area measurements is enough to divide the oils into two disjoint groups.


Figure 6 Mono-unsaturated versus poly-unsaturated levels for hazelnut and olive oils from 60 MHz 1H NMR spectra, calculated using peak areas. The two oils are seen to form two distinct groups on the basis of this simple measurement Even the simple ratio of the integrated olefinic to glyceride peak areas is sufficient to distinguish hazelnut oil (ratio > 1.6) from olive oil (ratio < 1.6), and shows systematic differences with mixtures of the two oils. But peak areas alone fail to capture the full complexity of the spectra, for example the small but systematic changes in the dominant methylene peak at ~1.3 ppm with changes in the mixture composition. To capture this additional level of detail, chemometric methods are required that utilise information from across the entire spectrum. We have studied the adulteration of olive oil with hazelnut oil using low-field NMR,14 developing a PLS regression model of the %w/w of olive oil present in mixtures of hazelnut and olive oils. Using this approach, we were able to establish a limit of detection of 11.2% w/w hazelnut oil in olive oil using low-field NMR data.14 This outcome compares well with results from high-field studies8, 15, 16 and with infrared.14 A larger range of 10 different edible oils has also been examined. Four or five independent samples of each oil type were analysed, preparing each by mixing with deuterated chloroform (50:50 proportions) and acquiring spectra under the conditions quoted above. PCA was used to visualise the differences between the spectra obtained (Figure 7). For the most part, the different oils are clearly differentiated using only the first two principal components, illustrating the discriminatory power of chemometric methods applied to low-field NMR data. The plot also offers some guidance on the likely success (or to some extent, the limit of detection) of spotting one oil mixed with another. For example, it would probably be relatively difficult to detect low levels of sunflower oil in corn oil. That may not be much of a fraud concern, but detecting sunflower oil in olive oil should be more successful as well as being more of a real-world interest.

Quantitative NMR 27

Figure 7 PCA plot derived from the NMR spectra of 10 edible oils, comprising olive, palm, coconut, walnut, rice bran, rapeseed (canola), linseed, corn (maize), sunflower and peanut (ground nut). Each dot is an independent oil sample. The ellipses are simply to identify members of an oil group and have no statistical significance 2.3 Authentication of Meat Low-field NMR is sensitive to different triglycerides and can discriminate between naturally occurring mixtures of triglycerides in the form of edible oils. Since edible animal fat is composed mainly of TAGs, it is natural to speculate that low-field NMR may be able to differentiate between different meats by means of the different TAG make-up of the fat component. This is not a new idea: as long ago as 1938 Paschke used chemical means to investigate mixtures of different meat types based on their TAG content.17 Much more recently, high-field 1H NMR has already been used to determine the triglyceride composition of different meats.18, 19 Testing methods for meat species is currently highly topical in the wake of the European horse meat scandal of late 2013, especially methods that might form the basis of a rapid screening protocol. We purchased fresh beef and fresh pork from local supermarkets and butchers. Horse meat was purchased either from meat suppliers in the UK (frozen only) or from butchers and supermarkets in France and Belgium (fresh and frozen). To extract the triglyceride component, a small piece of meat was homogenised in a blender, then left to steep in chloroform (either deuterated or non-deuterated). The resultant mixture was filtered, and the filtrate introduced to standard 5 mm NMR tubes. A variety of scans and acquisition times were used: data acquisition was performed at two sites on two instruments and the results combined. Comprehensive details covering the horse meat and beef studies will be published elsewhere.10 Figure 8 shows a principal components plot from an analysis based around concatenated olefinic, bis-allylic and terminal CH3 spectral regions only. A training dataset (76 beef extractions from 19 independent samples, 62 horse extractions from 19 samples, data not shown) underpins the ellipse shown in the figure. This ellipse is the line of constant Mahalanobis distance from the beef group centre-of-mass. It delineates an ‘authentic beef’ region (p=0.001). Independent test data (91 beef extracts from 31 samples, 16 horse extracts from 6 samples) are displayed in Figure 8. All but one of the beef test data points lie within


the ellipse, meaning that all but one of the beef extracts are correctly classified as authentic. Test data for horse (16 extracts from 6 samples) all lie outside the ellipse and are thus all correctly classified as non-authentic, i.e. not beef. Figure 8 in addition includes data from pork (104 extractions from 28 samples) which also lie entirely outside the ellipse and are therefore also correctly classified as non-authentic. We observe there is most spread in the horse data, which is likely due to the disparate sources of horse meat.

Figure 8 First versus second principal component plot for test data comprising beef (91 extractions from 31 different samples, ), horse (16 extractions from 6 different samples, ) and pork (104 extractions from 28 different samples, ). The ellipse delineates the beef group according to separate training data This analysis was framed to have a ‘beef’ versus ‘not beef’ outcome, that is, to function as an authentication protocol for raw beef. However, preliminary work has also been conducted to develop a methodology with a wider-ranging identification outcome (‘beef’, ‘pork’, ‘horse’, etc.) and we conclude with Figure 9, which shows the additional promise of the 1H NMR approach to also distinguish lamb from the other major red meat types. 3 CONCLUSION Low-field 60 MHz 1H NMR spectroscopy, as opposed to more widely known relaxometry, is able to extract TAG spectra with linewidths in the Hz domain that are comparable to those from much more expensive high-field instruments. Low-field spectra exhibit higher levels of peak overlap than high-field, but with care peak area estimations can still be used to extract useful information from TAG spectra, for example on the saturated fat content of complex foods or to differentiate one edible oil from another. To make the most of the spectral information requires chemometric methods, which are less sensitive to peak overlap. Using these approaches, low-field 1H NMR offers a viable method for testing the authenticity of meat and the detecting the adulteration of edible oils with one another. The case studies presented here rely on sample protocols that are deliberately rapid and inexpensive. Sample

Quantitative NMR 29

preparation methods of this type, combined with bench-top NMR, therefore provide a rapid screening technology relevant to triglyceride-rich mixtures such as edible oils and fats.

Figure 9 Preliminary results on classification of meat samples into beef, lamb, horse and pork based on canonical variates modelling using PLS scores Acknowledgements The authors acknowledge the support of the UK’s Technology Strategy Board (Project Number 101250) and the Biotechnology and Biological Sciences Research Council (Grant Number BBS/E/F/00042674). References 1 H. D. Belitz, W. Grosch and P. Schieberle, Food chemistry, 4 edn., Spinger, Berlin

Heidelberg, 2009. 2 M. Lisa, M. Holcapek and M. Bohac, J. Agric. Food Chem., 2009, 57, 6888-6898. 3 N. K. Andrikopoulos, Food Reviews International, 2002, 18, 71-102. 4 M. D. Guillen and A. Ruiz, J. Sci. Food Agric., 2003, 83, 338-346. 5 M. D. Guillen and A. Ruiz, Eur. J. Lipid Sci. Technol., 2003, 105, 688-696. 6 G. Knothe and J. A. Kenar, Eur. J. Lipid Sci. Technol., 2004, 106, 88-96. 7 G. H. Fang, J. Y. Goh, M. Tay, H. F. Lau and S. F. Y. Li, Food Chem., 2013, 138, 1461-

1469. 8 G. Vigli, A. Philippidis, A. Spyros and P. Dais, J. Agric. Food Chem., 2003, 51, 5715-

5722. 9 R. M. Alonso-Salces, M. V. Holland and C. Guillou, Food Control, 2011, 22, 2041-

2046. 10 W. Jakes, A. Gerdova, M. Defernez, A. D. Watson, C. McCallum, E. Limer, I. J.

Colquhoun, D. Williamson and E. K. Kemsley, submitted to Food Chemistry, 2014. 11 L. F. Johnson and J. N. Shoolery, Anal. Chem., 1962, 34, 1136-1139. 12 P. Dais and E. Hatzakis, Anal. Chim. Acta, 2013, 765, 1-27.


13 T. M. Alam and M. K. Alam, in Annual Reports on NMR Spectroscopy, Vol 54, ed. G. A. Webb, Academic Press Ltd-Elsevier Science Ltd, London, 2005, vol. 54, pp. 41-80.

14 T. Parker, E. Limer, A. D. Watson, M. Defernez, D. Williamson and E. K. Kemsley, Trends in analytical chemistry : TRAC, 2014, 57, 147-158.

15 L. Mannina and A. P. Sobolev, Magn. Reson. Chem., 2011, 49, S3-S11. 16 L. Mannina, M. D'Imperio, D. Capitani, S. Rezzi, C. Guillou, T. Mavromoustakos, M. D.

M. Vilchez, A. H. Fernandez, F. Thomas and R. Aparicio, J. Agric. Food Chem., 2009, 57, 11550-11556.

17 B. Paschke, Zeitschrift fur Untersuchung der Lebensmittel, 1938, 76, 476-478. 18 M. L. He, S. Ishikawa and H. Hidari, Asian-Australasian Journal of Animal Sciences,

2005, 18, 1655-1661. 19 A. B. Lisitsyn, I. M. Chernukha and A. N. Ivankin, Scientific Journal of Animal Science,

2013, 2, 124-131.

USEFULNESS OF 1H NMR TO STUDY THE FOOD LIPOLYSIS DURING IN VITRO DIGESTION

B. Nieva-Echevarría, E. Goicoechea, M.J. Manzanos and M.D. Guillén*

Food Technology, Faculty of Pharmacy, Lascaray Research Center, University of the Basque Country (UPV/EHU), Vitoria, Spain. *[email protected]

1 INTRODUCTION

In the last years a great deal of attention is being paid to food digestion process. In fact, obesity and related health diseases are an increasing problem in public health care, especially in the western countries. Additionally, nutritional quality of foodstuffs, as well as their safety and health properties, are subjects of great interest either for consumers, which claim for healthier food, or for the food industry, interested in designing it. After food intake, the first important step before food components absorption is digestion; in this step, changes in these components can also occur. Due to this, knowledge of the physico-chemical events that take place during digestion in general, and during lipid digestion in particular, is of paramount importance in order to optimize the nutritional value of foodstuffs and to manage lipid release and absorption in the gastrointestinal tract.1

In this context, in vitro models to study food digestion have been widespread used and several protocols can be found in literature.2 However, the knowledge on lipid digestion is limited and the extent of lipolysis under in vitro conditions is scarcely characterized. Precisely in this field, the measure of lipid hydrolysis rate is a crucial task to understand the digestive process and to know the changes undergone by food lipidic components during this step. In consequence, methodologies able to quantify the lipolytic products generated and to determine the lipolysis level reached are required. Until now, in in vitro digestion studies, the most commonly employed methodology to measure lipid hydrolysis rate, is the direct titration of the released fatty acids with NaOH by means of a pH stat titration apparatus.3 Likewise, chromatographic techniques such as High Performance Liquid Chromatography (HPLC), High Performance Thin Layer Cromatography (HPTLC) or Gas Chromatography followed by Mass Spectrometry (GC/MS) have also been employed to quantify some individual lipolytic products. Nevertheless, these above-mentioned methodologies present some limitations, either offering limited quantitative information on the different hydrolysis products generated, unspecificity, or involving many preparation steps; 4,5 therefore, further research in methodological developments able to monitor food lipolysis advance is required. In this paper, the usefulness of 1H NMR for the development of a new methodology to evaluate food lipolysis during in vitro digestion will be outlined.


2 METHOD AND RESULTS

2.1 Samples for Digestion

Farmed European sea bass specimens were purchased from a local supermarket. This kind of sample was selected due to its complex lipid profile. After cleaning, filleting, and skinning, fish fillets were homogenized in a grinder and samples for digestion were made of 4.5 g of minced fish flesh.

2.2 In Vitro Digestion

Fish samples were submitted to a three-step in vitro digestion procedure, following a previously proposed protocol for fed state,6 but slightly modified (Figure 1). This simulates digestive processes occurring in mouth, stomach, and small intestine, by adding sequentially simulated digestive juices prepared artificially in accordance with the original model. All the reagents for the preparation of the digestive juices were acquired from Sigma-Aldrich (St. Louis, MO, USA).

Figure 1 Schematic representation of the in vitro digestion procedure followed in this study

In order to obtain samples digested to different degrees of lipolysis, two different in vitro digestion experiments were carried out by varying some experimental factors such as the concentration of bile in the bile juice (15 and 30 g/l) and the presence of lipase from Aspergillus niger (100 U/ml) in the simulated gastric juice.

2.3 Lipid Extraction

Fish lipids were extracted using dichloromethane as solvent (CH2Cl2, HPLC grade, Sigma-Aldrich, St. Louis, MO, USA). Before digestion, the lipids from minced fish flesh were extracted using CH2Cl2 in a proportion of 1:2 (w/v) and assisted by an ultrasonic bath for 1 h. After digestion, a liquid-liquid extraction using the same solvent in a proportion of 2:3 (v/v) was performed. Afterwards, solvent was eliminated by means of a rotary evaporator under reduced pressure at room temperature, in order to avoid lipid oxidation. Finally, three lipid extracts were obtained: S0 sample, corresponding to fish lipids before digestion, and S1 and

Quantitative NMR 33

S2 samples, corresponding to those obtained after the two different in vitro digestion experiments.

2.4 1H NMR Spectra Acquisition

The 1H NMR spectra of fish lipid extracts were recorded on a Bruker Avance 400 spectrometer operating at 400 MHz using the same conditions as in a previous study.7 For each sample, 200 μl of the lipid extract was mixed with 400 μl deuterated chloroform (CDCl3) which contains 0.2% of non-deuterated chloroform and a small proportion of tetramethylsilane (TMS), used as internal reference (Cortec, Paris, France). The mixture was introduced into a 5 mm diameter tube and the spectrum of each sample was doubly recorded. The acquisition parameters used were the following: spectral width 6410 Hz, relaxation delay 3 s, number of scans 64, acquisition time 4.819 s and pulse width 90º. The relaxation delay and acquisition time allow the complete relaxation of the protons, being possible their use for quantitative purposes. The 1H NMR spectra were plotted at a fixed value of absolute intensity to be valid for comparative purposes.

2.5 Qualitative Changes Observed in the 1H NMR Spectrum of Fish Lipid Extracts Due to the Advance of Triglyceride Hydrolysis During In Vitro Digestion

The 1H NMR spectra of fish lipid extracts before (S0) and after (S1 and S2) in vitro digestion were studied in detail in order to identify qualitative changes due to the progression of lipolysis, if any.

Figure 2 1H NMR spectra of S0, S1 and S2 samples

In fact, qualitative differences among the spectra of the three samples can be clearly observed in Figure 2. Some spectral signals are present in the three spectra whereas others, present in the spectrum of S0 sample, tend to disappear or show chemical shift variations in the spectra of digested samples (S1 and S2). In addition, new signals appear in the spectra of S1 and S2 samples. These changes are more significant in the spectrum of S2 sample and they are due to the occurrence of signals related to the different products coming from triglyceride hydrolysis. The most significant changes due to the progression of lipolysis can be noticed in three spectral regions: those ranging approximately from 3.50 to 5.30 ppm (Region I), from 2.25 to 2.45 ppm (Region II), and from 1.50 to 1.80 ppm (Region III). Due to lipases regiospecificity, the expected pathway of triglyceride (TG) hydrolysis during digestion yields firstly 1,2-diglyceride (1,2-DG) and a fatty acid (FA), and in a second step, 2-monoglyceride (2-MG) and a second FA. A complete hydrolysis can also be achieved, after isomerization of 2-MG into 1-monoglyceride (1-MG). To properly identify the proton signals


of the spectra subject of study, 1H NMR spectra of pure standard compounds of TG, 1,2-DG, 1,3-DG, 2-MG, 1-MG, and FA with different chain length and unsaturation degree were also acquired. Table 1 summarizes the assignment of the signals commented in the present study. It has to be pointed that the hydrolysis of phospholipids was not taken into account, because they were present in very low proportions in comparison with TG (up to 1:500 mol/mol).

Table 1 Chemical shift assignments of characteristic 1H NMR signals in CDCl3corresponding to protons supported on triglycerides, partial glycerides and fatty acids

Signal (ppm) Type of protons Compound A1 1.61 -OCO-CH2-CH2- triglycerides*

1.62 -OCO-CH2-CH2- 1,2-diglycerides* 1.63 -OCO-CH2-CH2-,

COOH-CH2-CH2- 1-monoglycerides and fatty acids*

1.64 -OCO-CH2-CH2- 2-monoglycerides*A2 1.69 -OCO-CH2-CH2- triglycerides

(EPA and ARA acyl groups) 1.72 COOH-CH2-CH2- EPA and ARA acids

B1 2.26-2.36 -OCO-CH2- triglycerides**2.33 -OCO-CH2- 1,2-diglycerides**2.35 -OCO-CH2-,

COOH-CH2- 1-monoglycerides and fatty acids**

2.38 -OCO-CH2- 2-monoglycerides**B2 2.37-2.41 -OCO-CH2-CH2- triglycerides (DHA acyl groups)

2.39-2.44 COOH-CH2-CH2- DHA acid C 3.65 ROCH2-CHOH-CH2OH 1-monoglyceridesD 3.73 ROCH2-CH(OR’)-CH2OH 1,2-diglyceridesE 3.84 HOCH2-CH(OR)-CH2OH 2-monoglyceridesF 3.94 ROCH2-CHOH-CH2OH 1-monoglyceridesG 4.18 ROCH2-CHOH-CH2OH 1-monoglyceridesH 4.22 ROCH2-CH(OR’)-CH2OR’’ triglyceridesI 4.28 ROCH2-CH(OR’)-CH2OH 1,2-diglyceridesJ 4.93 HOCH2-CH(OR)-CH2OH 2-monoglyceridesK 5.08 ROCH2-CH(OR’)-CH2OH 1,2-diglyceridesL 5.27 ROCH2-CH(OR’)-CH2OR’’ triglycerides

Abbreviations: DHA: docosahexaenoate; EPA: eicosapentaenoate; ARA: arachidonate; *except for DHA, EPA and ARA acyl groups; **except for DHA acyl groups.

2.5.1 Qualitative Changes in the Spectral Region I (from 3.50 to 5.30 ppm). In Figure 3 is shown an enlargement of the 1H NMR spectral region between 3.50 to 5.30 ppm of the samples subject of study. Specific signals related to the protons in the glycerol backbone of TG (signal H and L centered at 4.22 and 5.27 ppm) can be observed in the spectrum of S0 sample. As hydrolysis advances, these signals decrease (spectra of S1 and S2 samples), showing the lowest intensity in the spectrum of S2 sample. On the other hand, specific signals related to the same protons but supported on 1,2-DG (signals D, I and K) and 2-MG (signals E and J) appear, showing the highest intensity in the spectrum of S2 sample. It is also noteworthy, although in very low intensity, the presence of signals F and G related to 1-MG in the spectrum of S2 sample, which evidences the isomerization of 2-MG to 1-MG and thus, the possible occurrence of complete hydrolysis of TG under in vitro conditions.

Quantitative NMR 35

These results indicate that the lipolysis degree reached in S2 sample can be considered as total due to the nearly absence of signals related to TG, whereas that reached in S1 sample is only partial because TG signals remain clearly observable in its spectrum.

Figure 3 Enlargement of the 1H NMR spectral region between 3.50 and 5.30 ppm of S0, S1 and S2 samples 2.5.2 Qualitative Changes in the Spectral Region II (from 2.25 to 2.45 ppm). In the spectral region ranging from 2.25 to 2.45 ppm, another two signals related to TG can be identified in the spectrum of S0 sample (Figure 4); signal B1 centered at 2.31 and due to protons of methylenic groups in -position in relation to the carbonyl group of all acyl groups except those of DHA, and signal B2 centered at 2.38 ppm and due to methylenic groups in - and -positions in relation to the carbonyl group of DHA acyl groups in TG.

Figure 4 Enlargement of the 1H NMR spectral region between 2.25 and 2.45 ppm of S0, S1 and S2 samples

As can be observed in Figure 4, great differences can also be observed when comparing the three spectra subject of study. As lipolysis progresses, signals B1 and B2 related to TG gradually decreases, nearly disappearing in the totally lipolyzed sample, whereas the intensity of signals centered approximately at 2.34 and 2.41 ppm increases. These new signals correspond to the overlapping of signals B1 and B2 of DG, MG and FA which appear at higher chemical shifts than those corresponding to triglycerides.


2.5.3 Qualitative Changes in the Spectral Region III (from 1.50 to 1.80 ppm). In Figure 5 is shown an enlargement of the spectral region ranging from 1.50 to 1.80 ppm where signals A1 and A2 due to methylenic protons in -position in relation to the carbonyl group appear.

Figure 5 Enlargement of the 1H NMR spectral region between 1.50 and 1.80 ppm of S0, S1 and S2 samples As can be observed in the figure, slight differences in the chemical shift and multiplicity of both signals can be noticed when comparing the spectra of fish lipids before and after digestion. Therefore, before digestion, signals A1 and A2 generated by protons supported on TG are centered at 1.61 and 1.69 ppm (spectrum of S0 sample), but after digestion, these signals appear at higher chemical shifts, evidencing the occurrence of hydrolysis products in S1 and S2 samples. Nevertheless, due to the high overlapping of these proton signals (Table 1), no further information about the nature of the newly formed products can be obtained by the study of this spectral region. 2.6 Quantitative Determination of Lipid Hydrolysis Products Generated During In Vitro Digestion The fact that the areas of the 1H NMR spectral signals of the different kinds of protons are proportional to the number of protons that generate them, and that the proportionality constant (Pc) is the same in all cases, can be used for quantitative purposes. Taking this into account, the number of moles of a component X (Nx) in a mixture can be determined from the area of a signal due exclusively to protons of this component (AX) and the number of protons (H) that generate this signal: Nx = Pc*(AX/H) (Eq.1) Therefore, the quantification of the different kinds of compounds arising from triglycerides hydrolysis can be carried out by developing equations in which the area of corresponding spectral signals are involved. In the case of hydrolysis products that generate specific non-overlapped signals, such as 1-MG, 2-MG, and 1,2-DG, the number of moles can be easily determined by using the area of signals F, E or J, and K, respectively. For quantification of the number of moles of TG and FA, signal H and B can be used. Nevertheless, as the proton signals of the different hydrolysis products are highly overlapped in these spectral regions,

Quantitative NMR 37

before applying the above-mentioned equation a previous correction of the signal area is required to avoid overestimation. To this aim, the area corresponding to the signal which overlaps is subtracted from the total area integrated in the spectrum, taking into account the number of moles of the compound and the number of protons that generates each signal. Once the number of moles of each kind of molecules present in the sample is known, the molar percentage of acyl chains (AC) joined to each compound in relation to the total number of acyl chains plus fatty acids can be determined using the following equations: ACTG% = 100(3NTG)/(3NTTG + 2N1,2-DG + N2-MG + N1-MG + NFA) (Eq.2) AC1,2-DG% = 100(2N1,2-DG)/(3NTTG + 2N1,2-DG + N2-MG + N1-MG + NFA) (Eq.3) AC2-MG% = 100(N2-MG)/(3NTTG + 2N1,2-DG + N2-MG + N1-MG + NFA) (Eq.4) AC1-MG% = 100(N1-MG)/(3NTTG + 2N1,2-DG + N2-MG + N1-MG + NFA) (Eq.5) FA% = 100(NFA)/(3NTTG + 2N1,2-DG + N2-MG + N1-MG + NFA) (Eq.6) This way to quantify hydrolysis products is usually employed in lipolysis studies in order to assess the extent of the hydrolysis reaction4. Data obtained for the three samples subject of study are represented in Figure 6.

Figure 6 Molar percentages of acyl chains supported on the different glycerides and of fatty acids present in the samples subject of study and determined by 1H NMR The quantitative results obtained totally agree with qualitative information extracted from the simple observation of the different spectral regions above-mentioned; in S2 sample, almost all the triglyceride molecules have undergone a process of hydrolysis whereas in S1 sample a considerable amount remains intact. Finally, the degree of lipolysis advance during in vitro digestion can be assessed from these quantitative results. In fact, the hydrolysis level of the samples can be expressed either as the relative disappearance of the substrate by means of the molar percentage of acyl chains joined to remaining triglycerides (100 - ACTG%), or as the relative appearance of the reaction product, this is the molar percentage of fatty acids (FA%) present in lipid samples. 3 CONCLUSIONS In short, this study evidences the usefulness of 1H NMR to provide very valuable qualitative and quantitative information about food lipolysis processes during in vitro digestion. By means of this technique it is possible to discriminate samples regarding their lipolysis level,


by simple observation of those proton signals appearing in three specific spectral regions. Furthermore, different equations can be developed to quantify all the products that may be present during lipid digestion, and thus, to estimate the advance degree of lipolysis reaction. In contrast with other methodologies, 1H NMR allows a global study of the sample, providing simultaneously detailed information on all the lipolytic products present, in a simple and fast way, and without any chemical modification of the sample. Acknowledgments This work has been supported by the Spanish Ministry of Economy and Competitiveness (MINECO, AGL2012-36466), by the Basque Government (EJ-GV, GIC10/85-IT-463-10) and by the Unit for Education and Research “Food Quality and Safety” (UPV/EHU-UFI-11/21). All authors participate in the COST Action FA1005 INFOGEST. B. N-E. thanks the UPV/EHU for a predoctoral fellowship. References 1 D. J. McClements, E. A. Decker and Y. Park, Crit. Rev. Food Sci., 2009, 49, 48. 2 S. J. Hur, B. O. Lim, E. A. Decker and D. J. McClements, Food Chem., 2001, 125, 1. 3 Y. Li, M. Hu and D. J. McClements, Food Chem., 2011, 126, 498. 4 A. Helbig, E. Silletti, E. Timmerman, R. J. Hamer and H. Gruppen, Food Hydrocolloid.,

2012, 28, 10. 5 L. Sek, C. J. H. Porter, A. M. Kaukonen and W. N. Charman, J. Pharm. Pharmacol., 2002,

54, 29. 6 C. H. M. Versantvoort, A. G. Oomen, E. Van de Kamp, C. J. M. Rompelberg and A. J. A.

M. Sips, Food Chem. Toxicol., 2005, 43, 31. 7 N. P. Vidal, M. J. Manzanos, E. Goicoechea and M. D. Guillen, Food Chem., 2012, 135,

1583.

QUANTITATIVE NMR ASSESSMENT OF POLYSACCHARIDES IN COMPLEX FOOD MATRICES

E.J.J. van Velzen1, S. Dauwan1, N. de Roo1, C.H. Grün1, Y. Westphal1, J.P.M. van Duynhoven1,2

1Unilever R&D, Vlaardingen, The Netherlands 2Wageningen University, Wageningen, The Netherlands

1 INTRODUCTION

In many food products polysaccharides are a critical ingredient for providing stable rheological properties. Food polysaccharides show a wide structural diversity and already provide benefits when formulated at low levels. Taken together with strong matrix interactions these factors complicate the quantitative assessment of polysaccharide in complex product formulations. Most analytical approaches described so far rely on qualitative identification of polysaccharides and quantification based on monosaccharide analysis. NMR has typically been applied for qualitative purposes1 2 3 but recently also progress has been made in semi-quantification4. Whereas NMR presents an ideal tool for quantification of low-molecular weight species5, its deployment for absolute quantitative assessment of polysaccharides is less straightforward. This is primarily due to broadened and overlapping lineshapes, which compromises NMR signal integration. By hydrolysis of polysaccharides spectral resolution is improved, which allows for straightforward, rapid and absolute quantification6, without the use of analytical standards. This procedure however discards the secondary and tertiary structure of polysaccharides, hence compromising their identification.

Figure 1 Schematic depiction of the analytical strategy for the quantification of polysaccharides in food products

We have therefore developed and validated a hybrid NMR approach which is schematically depicted in Figure 1. We first carried out an extensive isolation procedure to remove background of bulk ingredients such as salt, lipids, proteins and low molecular weight


compounds7. The polysaccharides in the isolate were then identified by means of spectral analysis of the 1H NMR mixture spectrum. For monosaccharide quantification we relied on and adapted6 Saeman hydrolysis8 procedure of the polysaccharide mixture followed by absolute monosaccharide quantification. Here we compared the use of a internal standard with PULCON, a relatively novel approach where the principle of reciprocicity9 is exploited to quantify by means of an external standard10 11. By combining information on identified polysaccharides and monosaccharide composition we were able to achieve absolute quantification of polysaccharides in full product formulations.

2 METHODS

2.1. Isolation of Polysaccharides from Complex Food Matrices

The full procedure to isolate the polysaccharide fraction from a full food product has been described elsewhere12. In short, approximately 25 g of product is dissolved in 350 ml of hot (90 °C) water. The dissolved material is mixed at 100 rpm at 80 °C for 30 min in a water bath and centrifuged while being hot using 500 ml Beckman polycarbonate bottles and a Beckman Avanti J-25 centrifuge in a pre-heated (80 °C) JA-10 rotor at 18,000×g for 5 minutes. The centrifuge was thermostated at 40 °C. After centrifugation, the fat layer was removed and the pH of the supernatant was adjusted to pH 6.0. Starch was enzymatically degraded by amylase treatment for one hour, and subsequently the pH was raised to 7.0 by adding sodium hydroxide. Dissolved proteins were hydrolyzed by protease treatment, hydrolysis products, salt and other small molecules were removed by dialysis against demineralised water. The dialysate sample was concentrated by rotary evaporation, subsequently dissolved polysaccharides were precipitated by adding ethanol. The precipitate was recovered by centrifugation, the pellet was washed with aqueous ethanol and ethanol. Ethanol and residual fat were removed by washing with acetone and heptane, while placed in an ultrasonic bath. Finally, the sample was dried at 30 °C.

2.2. Hydrolysis of Isolated Polysaccharides

Hydrolysis of polysaccharides is an adapted6 version of the well known Seaman8 hydrolysis procedure. In short, 3 mg of the food sample was weighed in a 20 ml head space vial. For the presolubilization 0.5 ml of 72 (w/w)% D2SO4 in D2O was added to the sample. The sample was capped with a silicone/PTFE seal and stirred in a water bath at 30 °C for 60 min. After the presolubilization step, 3.1 ml of D2O was added to the sample till the final concentration of 10 (w/w) % D2SO4 in D2O was reached. Note that all fractions were accurately weighed on an analytical scale. The sample was capped and incubated at an oven temperature of 100 °C for 90 min. After the hydrolysis, the samples were cooled down to room temperature and 1 ml of maleic acid standard solution was added to the sample. If the sample contained insoluble matter, the sample was centrifuged at 15000 x g during 5 min. The final solution was transferred into 3 mm NMR tubes and measured with NMR.

2.3. NMR

1D 1H 600 MHz NMR spectra were recorded with a zg30 pulse sequence (without water suppression) on a Bruker Avance III 600 MHz NMR spectrometer equipped with a cryopobe. The internal probe temperature was set to 290 K. 64 scans were collected with a relaxation delay of 60 sec and an acquisition time of 4 sec. The data were processed in TopSpin 3.2

Quantitative NMR 41

software. Manual baseline correction and phase correction were applied after the Fourier Transformation with an exponential window function and a line broadening factor 0.3. The chemical shift axes of the spectra were then referenced against the single CH resonance of maleic acid at 5.9 ppm.

2.4. Quantification

The molar content ( ) of a monosaccharide i (i=1 ...Q) in a sample A is defined as the sum of its individual - and forms. The total series of monosaccharide’s present in A is denoted as ( ) with molar monomer concentrations. Firstly, the “normalized” 1H signalintegrals (j=1...P) from all - and forms are calculated from the “raw” signal integrals( ), the receiver gain ( ), the number of protons ( ), and the length of the 90° pulse ( ) according to Equation 1:

(1)

Here, the concatenation of will result in a ( ) row vector (Equation 2).

(2)

Secondly, together with the PULCON calibration factor and a ( ) designmatrix , the final row vector can be calculated from according to Equation 3. Note that consists of ones and zeros. The ones (in the column direction) represents the connected - and forms of each of the Q monomers.

(3)

The PULCON factor is derived from an external standard B and can again be determined from the 1H NMR spectrum using the receiver gain ( ), the number of protons ( ), andthe length of the 90° pulse ( ). Together with the known molar concentration ( and the raw signal integral , is calculated according to Equation 4:

(4)

can be estimated from a single experiment but can also be calculated as the mean from multiple measurements.

Note that the filling factor of the coil, temperature, field strength, number of transients and instrumental settings between the sample measurement (A) and the external standard measurement (B) are (assumed) similar and that the linearity of the receiver is constant.

Concatenation of row vectors derived from samples in the dataset finally results in a ( ) matrix with Q monomer concentrations (mM). The quantification was done with MATLAB (version R2009a, The Mathworks). For the quantification of monosaccharides handmade MATLAB scripts were developed.


In order to translate the monomer concentrations to polymer concentrations we make use of the monosaccharide stoichiometry derived from the polysaccharide standards. Note that this approach is only applicable when the identity of the polysaccharides in the samples is exactly known.

In this approach, is the product of a ( ) matrix of polymer concentrations ( ) and a stoichiometry matrix ( ) of size ( ) according to Equation 5. Actually is the molar concentration of the repeating unit as (row-wise) defined in .

(5)

If is known, then can be estimated by Alternating Least Squares (ALS) optimization with non-negativity constrains (Equation 6). In the optimization process the error ( ) is minimized.

(6)

can also be expressed in weight concentrations, ( ), by multiplying with themolecular weights ( , size ( )) of the repeating units (Equation 7).

(7)

Where is the notation for the Hadamard product and is a ( ) column vector of ones.

The quantification of polysaccharides was done with the ALS code from the PLS toolbox 4.2 (2007, Eigenvector Research).

3 RESULTS AND DISCUSSION

3.1. Isolation and Identification of Polysaccharides in Food Products

Two consumer food products were subjected to an elaborate isolation procedure12 (Figure 2). The procedure ensured that background of salt, fat, starch and proteins was efficiently removed and that a polysaccharide fraction was obtained of approximately 250 mg from 28 g of food product.

Upon dissolving the fraction in D2O a 1H NMR spectrum could be obtained where signals of different polysaccharides can be discerned. These signals could be identified using spectra in a home-built library of common food polysaccharides (Figure 3).

Quantitative NMR 43

Figure 2 Schematic depiction of the procedure for isolating the polysaccharide fraction from food products12 3.2. Optimisation of Monosaccharide Quantification We departed from an adapted version of the well known Saeman procedure to hydrolyse polysaccharides to monosaccharides. Although this method was already tailored to NMR analysis, several adaptations needed to be made to enable full hydrolysis the food polysaccharides under study here. First, the solubilisation temperature needed to be raised from ambient to 30 0C. Secondly, the D2SO4 level needed to be fixed at 10% to obtain a good compromise between hydrolysis efficiency and reliable signal integration. In 10% D2SO4 mannose, galactose, glucose, and rhamnose showed well resolved spectra, without significant overlap. The ratios between the anomeric protons were according to theoretical expectations. For glucoronic acid, however, we could not observe different anomers, but glucoronic acid lactone as its reversible hydrolysis product13. The hydrolysates of four typical food polysaccharides are presented in Figure 4 and show well resolved signals of the anomeric protons of the aforementioned monosaccharides and the glucoronic acid lactone. The isolated spectral positions of the anomers and the glucoronic acid lactone hydrolysis product allows for straightforward assessment of monosaccharide content in the polysaccharide hydrolysate 3.3. Validation of Monosaccharide and Polysaccharide Quantification

For validation of the quantitative assessment of total polysaccharide content the recovery was determined of the four gums and cellulose (reference sample) as they are available as industrial ingredients. The five samples were subjected to the hydrolysis procedure, monosaccharide levels in the hydrolysate were determined by NMR and overall polysaccharide level was determined using known ratios of monosaccharides within the respective polysaccharides (compiled in the stoichiometry matrix ). As is shown in Figure 5, for both the Internal Standard and PULCON method equivalent results were obtained, no offset and proportional bias were observed.


Figure 3 600 MHz 1H-NMR spectroscopy. 1H-NMR spectra of product A (A), product B (B) high-acyl gellan gum (C), LBG (D), and xanthan gum (E). Gellan gum was identified by the signals indicated by squares, LBG by the signals as indicated by triangles, and xanthan gum by the signals as indicated by circles

Quantitative NMR 45

Figure 4 Part of the 600 MHz 1H NMR spectrum of the hydrolysates of (A) Locust Bean Gum (LBG), (B) xanthan gum, (C) high acyl gellan gum (D) low acyl gellan gum in 10 (w/w)% D2SO4. Annotation of anomeric protons are: Gal1: -galactose, Gal2: -galactose, Man1: -mannose, Man2: -mannose, Glc1: -glucose, Glc2: -glucose, GlcA: glucoronic acid, GlcL: glucuronic acid -lactone, Rha1: -rhamnose, Rha2: -rhamnose


Figure 5 Comparison of polysaccharide levels as determined with PULCON (horizontal) and by using an Internal Standard (vertical) In Table 1 also the confidence intervals of the polysaccharide levels for the different materials are presented. One can observe that average levels and confidence intervals are rather similar for the IS and PULCON methods and that these are also in line with results obtained by the classical methanolysis/TFA method where HPLC is used for monosaccharide quantification14. Table 1 Polysaccharide content based on the IS and PULCON NMR approach and the classical methanolysis/TFA method14. Relative errors are expressed in terms of 95% confidence intervals (CI) Internal Standard PULCON Methanolysis/TFA w/w% CI w/w% CI w/w% CI Cellulose 103 6 102 8 LBG 78 4 83 3 77 4 Xanthan 64 5 65 6 67 7 High Acyl Gellan 46 8 42 11 57 10 Low Acyl Gellan 66 8 64 10 75 10 3.4. Quantitative Assessment of Polysaccharides in Complex Food Products Saeman hydrolysis was applied on a polysaccharide extract isolated from a food product comprising four different polysaccharides at low levels. To decompose the dataset into (i) a subset of polysaccharide concentrations ( ) and (ii) the monosaccharide stoichiometry ( ), ALS was used with non-negativity constraints on and equality constraints on . First was calculated in the food samples, after that was divided by (Equations 5 and 6) and resulted in the matrix of the food samples. The quantitative results of food samples are obtained from the PULCON and the IS methods (Table 2A and 2B, respectively). The concentrations in showed similar results for the two methods. The relative error cannot be

Quantitative NMR 47

calculated for food samples, because repeated measurements are needed. Since in the hydrolysates the small signal of glucoronic acid was below the limit of quantification, no xanthan levels could be reported. Table 2 Concentration of polysaccharides in food samples in w/w% calculated with (A) PULCON and (B) Internal Standard method. A LBG Xanthan Gellan Glucose Product A 34.76 >LOD, <LOQ 25.29 0 Product B 18.20 >LOD, <LOQ 0 46.96 B LBG Xanthan Gellan Glucose Product A 34.98 >LOD, <LOQ 29.81 0 Product B 17.63 >LOD, <LOQ 0 45.58 4 CONCLUSION In conclusion, an analytical method has been designed, refined, implemented and validated for quantification of polysaccharides in food products. The approach can quantify polysaccharides in the w/w% range with 10% precision. Further efficiency can be gained by automation of the hydrolysis procedure and by implementation of (automated) spectral deconvolution of the monosaccharide 1H NMR spectrum15 16 5. Acknowledgements Juliën Boelhouwer and Raymond Baris are gratefully acknowledged for skilful execution of the isolation procedure and expert expertise input. References

1 D. Lowicki, A. Czarny and J. Mlynarsk (2013). NMR of carbohydrates. In (42 ed., pp. 383-419).

2 A. Spyros and P. Dais (2012). NMR spectroscopy in food analysis. In (pp. 1-343). 3 H. N. Cheng and T. G. Neiss, Polym. Rev., 2012, 52, 81-114. 4 E. van Velzen, N. de Roo, R. Poort, L. van Adrichem, K. Brunt, H. Schols, Y. Westphal,

L. Mariani, C. Grun and J. van Duynhoven (2013). In Magnetic Resonance in Food Science: Food for Thought (pp. 156). Royal Society of Chemistry.

5 J. Van Duynhoven, E. van Velzen, E. and D. M. Jacobs, Ann. R. NMR S., 2013, 80, 181-236.

6 A. C. de Souza, T. Rietkerk, C. G. Selin and P. P. Lankhorst, Carbohyd. Polym., 2013, 95, 657-663.

7 C. Grun, P. Sanders, M. van der Burg, E. Schuurbiers, L. van Adrichem, E. van Velzen, N. de Roo, K. Brunt, Y. Westphal, H.A. Schols, submitted., 2014.

8 J. F. Saeman, W. E. Moore, R. L. Mitchell and M. A. Millett, Tappi J., 1954, 37, 336-343.

9 D. I. Hoult and R. E. Richards, J. Magn. Reson. (1969), 1976, 24, 71-85. 10 R.D. Farrant, J.C. Hollerton, S.M. Lynn, S. Provera, P.J. Sidebottom and R.J. Upton,

Magn. Reson. Chem., 2010, 48, 753-762.


11 O. Frank, J. Kreissl, A. Daschner and T. Hofmann, J. Agric. Food Chem., 2014, 62, 2506-2515.

12 C. Grun, P. Sanders, M. van der Burg, E. Schuurbiers, L. van Adrichem, E. van Velzen N. de Roo, K. Brunt, Y. Westphal and H.A. Schols, Food Chem., 2015, 166, 42.

13 R. Wang, T. L. Neoh, T. Kobayashi, Y. Miyake A. Hosoda, H. Taniguchi and S. Adachi, Bioscience, biotechnology, and biochemistry, 2010, 74, 601-605.

14 G. A. De Ruiter, H. A. Schols, A. G. Voragen and F. M. Rombouts, Anal. Biochem., 1992, 207, 176-185.

15 D. M. Jacobs, E. van Velzen and V. Mihaleva (2013). Evaluation of Approaches for Quantitative Targeted Profiling of Complex Compositions using 1D 1H NMR Spectroscopy. In Magnetic Resonance in Food Science: Food for Thought (pp. 3-13). Royal Society of Chemistry.

16 V. V. Mihaleva, S. P. Korhonen, J. van Duynhoven, M. Niemitz, J. Vervoort and D. M. Jacobs, Anal. Bioanal. Chem., 2014 1-12.

Quality and Safety

MAGNETIC RESONANCE ANALYSIS OF DAIRY PROCESSING: SUITABLE TOOLS FOR THE DAIRY INDUSTRY

R. Anedda

Porto Conte Ricerche S.r.l., SP 55 Porto Conte-Capo Caccia, Tramariglio, 07041 Alghero, SS, Italy

1 INTRODUCTION

The considerable development of both instrumental hardware and data analysis tools has promoted, over the last few decades, a widespread diffusion of Magnetic Resonance facilities at the academic and industrial level. In the dairy sector, although an important contribute to the understanding of molecular dynamics in heterogeneous systems as seen by NMR has been provided by early studies on dairy-related systems,1-9 the application of Magnetic Resonance to dairy industry has been somewhat hindered by the intrinsic complexity and heterogeneity of commercial products. The diverse applications of Magnetic Resonance techniques to dairy research have been previously reviewed,10-15 and can be roughly classified into four different categories, depending on the specific method exploited, namely spectroscopy, relaxometry, diffusometry, and imaging. To complete this picture, however, a rapidly emerging field of research, NMR-based metabolomics, should be added nowadays. As it will be detailed in the following, the aforementioned categories can be variably combined in several cases aiming at a more comprehensive description of the system under study. In this chapter the different Magnetic Resonance approaches for characterizing dairy products and processes will be reviewed, with particular focus toward the effects of processing procedures and the most recent developments and practical applications for quality assessment and authentication.

2 MILK PROTEIN SOLUTIONS AND GELS

An important methodological work on the MR characterization of dairy products has been conducted on model systems such as pure milk protein solutions or their mixtures, e.g. casein dispersions and gels,3,7,16-24 whey proteins solutions and gels,2,25-30, cheese analogs and imitations.31-35

One of the main achievements of NMR-based work on milk protein mixtures has been the description of molecular events that are promoted by heat treatments. In this regard, among milk proteins, whey proteins have been a subject matter of the NMR research more than caseins and this can be rationalized by taking into account several considerations. First of all, physicochemical characteristics of caseins show a remarkably high heat stability and therefore caseins are generally believed to be less susceptible to denaturation with respect to


whey proteins.36 In fact, NMR studies have shown that native caseins possess a rather rigid structure.37 Nonetheless, it was demonstrated by early 1H NMR studies that reversible spectral modifications of caseins are temperature dependent, especially at temperature higher than 60°C, increasing temperature likely leading to increased mobility of parts of the protein in the micelle.37 Secondly, the three-dimensional structures of caseins still remain quite vague.38 Third, strong hydrophobic interactions between milk components favour the formation of large colloidal aggregates, known as casein micelles, where different forms of caseins ( S1, S2, , ), calcium and phosphates are embedded in surprisingly stable structures, making high-resolution NMR analyses less informative; nonetheless some NMR spectroscopic studies of -casein have demonstrated that regular -helical secondary structure of some caseins may be responsible of the remarkable stability of these systems.38 2.1 Whey proteins On the other hand, whey proteins are known to undergo structural changes at the mild temperatures usually reached in the most common cheesemaking processes, such as milk thermization or pasteurization, also leading to macroscopic effects. Furthermore, it is commonly accepted that most structural changes occurring in milk proteins at temperatures below 60°C have a reversible nature, and therefore they reversibly affect the phisicochemical properties of such systems,39 while irreversible modifications affect such systems at higher temperature.2,4 -lactoglobulin ( -LG) is the most concentrated whey protein in milk, representing approximately 20% of the total proteins in bovine milk and 50% of the proteins in whey. Moreover, -LG has been the subject of relevant interest because its structural and dynamical properties can significantly affect dairy processing especially when thermal treatments are carried out during cheese production and storage. NMR work revealed that -LG undergo irreversible denaturation at temperature ranges between 70°C and 80°C, the process being strongly pH dependent, but several structural modifications are already observable at temperature higher than 50°C at pH 2.27 Both one-dimensional and two-dimensional 1H NMR spectroscopy have shown their usefulness in the characterization of temperature and pH dependence of thermally induced unfolding and denaturation of -LG, especially exploiting the deuterium exchange method. Such an approach is based on the dissolution of the native protein in deuterated water, where the labile protein protons (NH, OH and SH functional groups) that are accessible to the solvent are able to undergo chemical exchange with the solvent protons. Of course, labile protons in structured hydrophobic regions of the protein and labile protons that are H-bonded in the secondary structure of the macromolecule are not able to exchange, unless some unfolding takes place due to external stimuli. By this method, coupled to 1D and 2D TOCSY, NOESY and DQF-COSY proton NMR analyses, Belloque and Smith27 defined the assignment and secondary structure ( -sheet and -helix regions) of -LG at pH 2, and discussed the stepwise changes of the typical

-barrel induced in the system when increasing the temperature from 35°C up to 75°C, at pH 2. The authors observed an increased structural flexibility of -LG at early stages of heating (< 55°C), as evidenced by the loss of some NH backbone signals in the spectra, especially from strand E (Figure 1). Considering the slow proton exchange rate at acidic pH, they concluded that such protons are expected to be exposed to the solvent for long times even at the first stages of heating, which in turn implies that the protein undergoes either denaturation or is in equilibrium with the unfolded form. At the same pH, higher temperature (75°C) produces gelling of the solution at protein concentrations of 5-10%, suggesting a stepwise and reversible opening of the rigid parts (strands BCD and FGH) of the barrel and hydrophobic bonding. It is worth noting that the abovementioned gelation process was found

Quality and Safety 53

to be significantly favoured, at similar temperatures, by increasing the pH to neutral values (pH 7.4).

Figure 1 Ribbon picture of a -LG monomer produced using VMD40 starting from Protein Data Bank accession code 1BEB.41 Gelation of a mixture of a 5% - and -LG was not observed when it was treated at high pressure (200 MPa), although it was observed that -LG lost its native structure at that pressure at ambient temperature.26 Structural changes due to -LG large unfolding and formation of fibrils were observed in 1-2 wt% protein solutions at pH 2 by proton NMR coupled to Atomic Force Microscopy at temperatures higher than 60°C.28 Several authors faced the problem of the uselessness of high-resolution NMR spectroscopy for the description of the molecular steps involving the transition from monomers to large aggregates or gels of -LG. Indeed, being the width of NMR signals regulated by the rotational motion of the molecular species from which the signal is generated, NMR signals of large molecular weight aggregated species disappear from spectrum, making the exploitation of NMR spectroscopy of liquid samples less informative. Such limitation can be at least partially overcome by exploiting NMR relaxation measurements. Changes in water proton transverse relaxation of thermally aggregated Bovine serum albumin were explained considering the simple concepts of diffusion and chemical exchange.42 Lambelet et al. measured the spin-spin relaxation of 4-18 wt% -LG solutions at temperatures from 20°C to 90°C and pH values in the range from 3.5 to 8.2 By such approach, it has been possible to observe the effects on T2 of the heat-induced -LG conformational changes. It was shown that the first perceivable changes in 1/T2-Temperature sigmoidal-like curves can be observed at approximately 40°C, and from 60°C a steep increase of relaxation rate was observed when increasing temperature. Also the role of pH was quantified, at higher pH the inflection point of the curve being shifted at lower temperature.2 For more practical considerations, one must consider that the same heat-induced changes described above are likely to occur during common dairy practices, such as low temperature long time pasteurization (LTLT) of milk, where milk is heated at 62°C-65°C for at least 30 min.43 It is also worth noting that heat treatments (thermization, pasteurization) are generally carried out on milk at its natural pH of about 6-7, when heat treatments have an effect on protein structure at lower temperatures than at lower pH values.2 Multidisciplinary studies on -LG, in which low-field NMR relaxometric observations were complemented with dynamic light scattering, circular


dicroism and fluorescence, led to the conclusion that heat-induced changes of -LG are mainly due to protein aggregation rather than unfolding.44 Whether protein unfolding and aggregation takes place or not during heat-treatments, certainly the morphology and state of these macromolecules are supposed to be subjected to a modification during common milk or cheese processing. It should also be noted that even if the aforementioned structural changes (i.e. unfolding, denaturation, gelation and aggregation) are likely to occur in milk, milk is certainly a more complex system than a solution of pure -LG, and more complex interactions with other molecular components are therefore expected. For example, milk is known to undergo gelation or coagulation when heated, the proteins -LG and -casein having a key role on determining the kinetics of gelling.45 For instance, it has been observed that the addition of caseins (sodium caseinate, casein micelles) or salts (KCl, CaCl2) to whey protein isolates solutions leads to a concentration- and temperature-dependent increase of the measured spin-spin relaxation rates,4 which shows a steep increase at temperatures between 60°C and 80°C. It has been also suggested that heat treatments of dyalised skim milk may induce aggregation of -LG and -casein.4 However, interactions between denatured -LG and -casein are known to affect structure development in many cheeses.46 2.2 Caseins and milk Early work described in great detail chemical and diffusive exchange processes in pure protein model solutions (lysozyme) and in skim milk, and explained the role of the morphology and state of macromolecular aggregates in determining the observed spin-spin relaxation process.47 It was quantitatively demonstrated that the influence of water molecules strongly interacting with the protein network, whose rotational correlation times are supposed to be significantly affected by water-protein interactions, cannot explain the remarkable increase in transversal relaxation rates observed in protein solutions and reconstituted skim milk. It was concluded that chemical exchange dominates proton relaxation rate in such protein solutions, and that proton relaxation measurements are not able to provide information on the state of water (bound or free) or on particular sites where water molecules reside. It was also shown by several authors that the effect of chemical exchange on T2 of milk can be mainly ascribed to soluble whey proteins and lactose rather than to micellar caseins.47,48 This was explained by considering that due to the very short relaxation time of casein protons (roughly 50 s) and the high hydration potential of casein in micelles, micellar protein protons can have only a negligible chemical shift contribution to relaxation. The wide NMR work on caseins has included several methodological approaches, ranging from spectroscopy, employed to solve structural details of micelles or of relevant segments of caseins,37,38,49-51 to relaxometry, which gives indications on molecular rotational and translational mobilities, chemical exchange and interactions of water with macromolecules,3,7,47,48,52 and diffusometry, mainly exploited to study water diffusive exchange phenomena.16,17,23 As far as the effects of thermal treatments are discussed, not as much NMR work as for whey proteins has been directed to caseins. 1H NMR spectroscopy studies on casein micelles (from skim milk), in the absence of whey proteins and lactose, demonstrated that conformational changes observed at temperature higher than 60°C are believed to be reversible, as the original spectrum restored when cooling the sample to room temperature after the heat treatment.37 Other relaxometric studies provided information on mixtures of casein micelles and whey proteins. In that case, while no change in T2 was registered for casein micelles in a 12% (w/v) solution after heating at 90°C for 30 min, transverse relaxation time increased when solutions of 6% and 12% (w/v) of casein micelles were mixed with 3% (w/v) of whey protein isolate and heated at temperatures higher than


60°C. Similar results were obtained when heating dialysed skim milk.4 More recent 1H NMR studies, combined with other complementary analytical techniques, suggested the chaperone activity of bovine s-Casein in preventing the aggregation of -Lactalbumin and -LG, although the experiments were carried out at temperature lower than 60°C, when aggregation is not expected according to the abovementioned relaxation measurements.50 3 NMR RELAXOMETRY ON CHEESE AND CHEESE ANALOGUES A wealth of information is provided by NMR relaxometric investigations on cheese imitations and commercial cheeses. The complication introduced with respect to the above discussed NMR investigations on dairy solutions and gels is that proton relaxation in cheese is usually multiexponential and faster than the corresponding relaxation rate in the bulk water. Proper interpretation of NMR relaxation behaviour of protons in cheeses can provide very useful indications on the state of macromolecules and other species having exchangeable protons and on sample morphology,47 although in literature discussions based on the number of “states” of water molecules (more or less bound to proteins) or on the number of “sites” where water molecules dynamically reside are often reported. This may in part arise from an incorrect interpretation of the molecular mechanisms that give rise to the observed distributions and in part might be due to the need to explain complex concepts by making use of more practical analogies. This latter explanation would certainly account for the literature concerning practical aspects of food processing, such as that involving dairy industry. When observing protons, at least two different relaxation components from fat and water protons should be expected in a commercial cheese. Only when skimmed milk coagulation is investigated the signal from fat can be obviously neglected, but the signal from water protons shows a multiexponential relaxation after syneresis.6 It has been demonstrated that monoexponential relaxation adequately describes proton T2 in skim milk and curd before syneresis due to fast diffusion of water molecules that sample all possible environments (compartments) in the NMR time scale.6,53 After syneresis, relaxation curve becomes bi-exponential, and the two water protons populations were ascribed to water associated to protein and serum (bulk) water. It has been therefore suggested that NMR relaxometry is able to quantify serum water and water entrapped in the casein network, i.e. to quantitatively monitor syneresis in cheese.6 A similar NMR application for the analysis of curd for quality control and authentication would be particularly relevant for assessing thermal treatments to which milk has been subjected before cheesemaking, since it is known that heat-denatured whey proteins may act as structural elements in cheese,54 leading to peculiar textural changes (increasing firmness) in curd, that in turn causes differential syneresis abilities when the milk used has been heat-treated. Other studies on cheese analogues have been carried out aiming at correlating relaxation time constants with moisture or fat content in dairy products,31,32 and a linear correlation (R2>0.95) was actually found between moisture content and the value of the transverse relaxation time constant T2 (15 ms T2 22 ms) associated with one of the two populations observed. Such studies were carried out on cheese analogues (also referred to as cheese imitations), which are supposed to have similar texture as cheese, but allow a much more precise control over composition since they are manufactured by combining known percentages of rennet casein, salts, vegetable or milk fats, organic acids and water. Since gaining differential information from fats and water is crucial especially for commercially available dairy products, for instance in order to monitor cheese rheology during the ripening process, both relaxation measurements55 and magnetic resonance imaging studies9 of commercial cheeses have been devoted specifically to this end. Interestingly, time-domain


NMR (TD-NMR) investigations recently exploited a combined T1-T2 relaxation sequence coupled with a calibration routine and multivariate analysis of data and allowed a rapid (< 2min) estimation of water and fat content in cow’s milk cheese of different origins, cheese makers, heat treatments of milk and ripening stages, following a procedure that does not require any sample extraction or preparation.55 It is worth noting that the robustness, low cost and user friendliness of TD-NMR instrumentation make the aforementioned method very appealing for dairy industry. This NMR-chemometrics combination for quantitative analysis of fat and water in cheese is certainly more reliable than that proposed by earlier works.31,32 However, going back to the focus of the present chapter, i.e. the effect of processing and cheesemaking practices on the NMR features of the final dairy products, the aforementioned quantitative applications cannot be considered sufficiently useful, since water and fat content of cheese widely vary among different cheese samples, and cannot in principle be directly associated with heat treatments of milk or curd, nor with other dairy practices. It is clear that a reliable method that is sensitive to reveal the effects of dairy practices on the final product should provide other information than fat and moisture quantification. On the contrary, the experimental output should reflect cheese microstructures and molecular dynamics of cheese components and, if possible, it should allow to follow the evolution of such characteristics with time. 3.1 The diffusive and chemical exchange model and practical perspectives for dairy industry One of the most debated issues in describing relaxation in heterogeneous systems like cheese has been the interpretation of the multiexponential recovery of the nuclear magnetization to thermal equilibrium. It is discussed here how this behaviour, far from representing a direct sensor of the number of molecular classes present in the system under study (e.g. fat protons and water protons), arises from the interaction of observed nuclei with the microstructure (spatial heterogeneity) of cheese in the NMR measurement time scale. This can be usually explained by considering a combination of diffusive and chemical exchange, i.e. the dynamics of an observed nucleus within a molecule that experiences, during a time lapse set by its translational velocity and the measurement time, the influence of a surface that acts as a relaxation sink (e.g. protons of dinamically oriented water molecules close to a macromolecule) or undergo chemical exchange with the labile protons of slowly reorienting compounds (e.g. side chain protons of whey proteins or lactose in cheese, which are characterized by different relaxation rates and chemical shift than bulk water protons).47 The general mathematical form of the diffusive and chemical exchange model has been first described by Fedotov et al.,56 then revised by Carver and Richards,57 and refined by Belton et al.58 and Hills et al.59 Such model and some of its applications to foods, including cheese, have been recently reviewed by Brosio et al.60 Some of the features that are most relevant to the present discussion will be briefly recalled in the following. First of all, the model allows to define the limiting condition for the observation of a multiexponential decay of the relaxation curve as:47

(1)

Where is the characteristic dimension of the heterogeneity (e.g. the micelle radius in milk),

is the difference between the relaxation rates of nuclei in two different sites and D the self-diffusion coefficient of the molecule to which the observed nuclei are bound (e.g. water in milk or cheese). For example, in the case of NMR relaxometric analyses of water protons in milk, when the diffusion of water toward the protein surface is faster than the surface


relaxation rate, the relaxation curve will be mono-exponential, the system is in the “fast diffusive exchange” or “surface limited” regime. On the contrary, the “slow diffusion regime”, also referred to as “diffusion limited regime”, is defined by the condition where water diffusion is slower than the surface relaxation rate. In the latter case, the relaxation process will be multi-exponential. It turns out that the limit size of heterogeneity for a surface limited relaxation is approximately <<70 m.60,61 Beyond being influenced by water diffusion and surface relaxivity, in common spatially heterogeneous systems such as dairy products, the NMR relaxation of water protons can be influenced by chemical exchange with labile protons of macromolecules and other molecules, e.g. casein, whey proteins and lactose present in milk and cheese. The exchange of protons between chemically shifted sites causes a typical dispersion curve when transversal relaxation rates are plotted against the reciprocal interpulse spacing of a Carr-Purcell-Meiboom-Gill sequence (CPMG), the transverse relaxation at long pulse spacings being even faster at high spectrometer frequency.59 The effect of chemical exchange on relaxation, and consequently the origin of the above mentioned dispersion curve, can be understood if a nucleus jumping between two sites a and b having different chemical shifts is considered. In fact, since nuclear precession frequencies change from a to b in the two sites, the dephasing rate changes as well when the nucleus exchanges from one site to another. When the interpulse spacing is small enough in CPMG experiments compared with the exchange rate (or better with the mean lifetime of a nucleus in each site) the relaxation rate will not be influenced by chemical exchange; on the other hand, when interpulse spacing is large the observed nuclei will have exchanged repeatedly between two pulses, and the echo will reflect the significant dephasing due to the chemical shift differences between the two sites. It is easy to understand that such effect will be dependent on magnetic field, being more important at high fields.59 In order to consider the effect of chemical exchange between bulk water protons and labile protein protons, at long pulse spacings, the Swift-Connick’s equation can be used:47,57-59,62

(2)

where 1/T2,obs is the observed transverse relaxation rate, Pbulk and T2,bulk are the fraction of water protons not affected by the presence of the proteins (“bulk” or “free” water) and the corresponding transverse relaxation time constant, respectively; Pexch and T2,exch are the fraction of water protons directly interacting with protein and their T2; (kexch)-1

is the inverse of the proton exchange rate, i.e. the average lifetime of exchangeable protons in each site; is the chemical shift difference between the two sites. Equation (2) represents the general formula for the mathematical interpretation of chemical exchange in heterogeneous materials and foods. If the exchange rate is very fast, for long interpulse spacings in CPMG experiments, i.e. kexch>> ( , (T2,exch)-1) , then equation (2) reduces to:

(3)

While for short CPMG interpulse spacings, equation (3) becomes:

(4)


Therefore, the mathematical model proposed allows a complete description of the morphological and dynamical processes in the food sample analyzed. As far as dairy products such as cheese are concerned, the observation of a multiexponential decay of the magnetization is mainly regulated by diffusion of water (and fat) molecules and size heterogeneity of the sample, while T2 dispersion curves observed as a function of pulse spacings in CPMG experiments can account for chemical exchange phenomena. (T2,obs)-1 vs CPMG interpulse spacings can be plotted for each relaxation component in those systems that show multiexponential decay, and mean T2 values and area fractions associated to each component and corresponding to each interpulse delay can be derived. It was suggested that diffusion through magnetic field gradients, also generated by the peculiar morphology at the water-protein interfaces, can result in a monotonic decrease of relaxation time constants and in a variation of the area fraction associated to each component as a function of CPMG pulse spacing.47 Therefore, for what discussed above, there seems not to exist a strict and consistent one-to-one relationship between the number (and relative percentages) of observed T2 populations and the number (and quantity) of molecular classes present in the sample (e.g. protons from fat, water and sugars in cheese), nor a correlation exists between the number of relaxation components and different sites where water (or fats) reside. However, the 1/T2,obs values should be more likely associated to sample microstructure and morphology, and to molecular dynamics within the system under investigation. In this sense, NMR relaxation studies of dairy products could reveal, in a rapid and non destructive way, molecular characteristics of similar dairy products that go far beyond their composition. To be more exact, NMR relaxation is expected to be able to reveal differences between two similar dairy products even when they have the same proximal composition. For certain aspects, dairy industry would certainly benefit the application of NMR relaxation analysis for quality assurance and authentication purposes, since the information provided by such an approach is supposed to be very sensitive to even subtle variations of the samples analyzed. Just to make few practical examples, Gianferri et al., in two different reports,63,64 found two different T2 values of the water entrapped in the casein network and associated this difference to the different drainage of mozzarella cheese in the sample preparation steps before NMR analysis. Nevertheless, the same authors found consistent trends upon aging in the T2 profiles of Mozzarella cheeses produced by different cheese-makers, suggesting that, given a standardized preparation protocol is followed, a specific relaxometric profile can be associated to a specific dairy product (i.e. to all milk processing procedures leading to a certain cheese); Kuo et al. found different relaxation behaviour (changes in water mobility, according to the authors) in pasta filata and non-pasta filata Mozzarella cheese during the first 10 days of storage,65 and explained the different changes observed in the T2 populations they associated to two different samples depending on whether the two cheeses were frozen or not.66 Mulas et al.found consistent and differential relaxometric features (T2 distributions) that describe water protons relaxation in sheep’s milk cheese depending on whether it was manufactured from raw or heat-treated milk (thermized or pasteurized).67 Some more comments to the work by Mulas et al. on Fiore Sardo cheese will be presented in the next section. 3.2 The case of Fiore Sardo PDO cheese Sardinian ewe’s milk cheese Fiore Sardo provides a useful (real) model system to test the sensitivity of NMR methods to reveal the effects of cheesemaking processes on morphological characteristics and molecular dynamics of dairy products. The manufacture of Fiore Sardo cheese is quite straightforward, since it basically consists in renneting raw milk


at about 36°C by means of lamb rennet paste. Curd is left to drain, salted and the obtained cheese is ripened in cellars at low temperature (10-15°C). Since Fiore Sardo cheese is made with uncooked paste, the only heat treatment this product is subject to is mild heating of milk at 34-36°C. According to what discussed above about dairy proteins, and from the wealth of information reported in literature, it is expected that the only structural modifications on proteins are induced by enzymatic activity (renneting) on milk.68 The resulting cheese, already after about 3 months, is characterized by peculiar sensorial traits, and is described as firm, crumbly, and with a floury and grainy texture.69 The paste of mature Fiore Sardo cheese usually brakes into flakes, similarly to what happens with other hard cheeses made from raw milk (e.g. Parmigiano Reggiano PDO). Also cheese eyes formation is expected to be regulated by the indigenous microflora present in milk, and from ripening conditions. It is also known that heat treatments of milk intended for cheese production influence its cheesemaking properties, and also influence body as well as textural and sensorial profiles of the final cheese product.70 Producers of Fiore Sardo, represented by the Consortium for the Protection of Fiore Sardo Cheese (CPFSC), feel the urgent need for safeguarding the original cheesemaking protocol, and are particularly concerned about the presence in the market of counterfeit Fiore Sardo made from pasteurized or thermized milk. To this aim, we have recently developed a Magnetic Resonance Imaging method that allows to discriminate Fiore Sardo cheese from other Sardinian sheep’s milk cheeses manufactured from heat-treated (HT) milk.67 This study is ongoing, and aims at developing standardized analytical procedures, based on the characterization of peculiar NMR relaxometric profiles and MR images analysis, able to differentiate raw milk cheeses from their HT milk counterparts. Briefly, the MRI method is based on the investigation of multiexponential CPMG curves of cheese samples having the same ripening time and manufactured from raw or HT milk. Two relaxation components were found for all cheeses, and attributed to water protons since fat protons were suppressed by a proper MRI preparation sequence: the first population (1P) centred at about 9 ms, and the second one (2P) centered at about 35 ms. The peculiar difference observed between raw milk and HT milk cheeses consists in a different area fraction (A%) of the two populations, the 1P always showing a significantly higher A% in HT milk cheeses with respect to the raw milk counterparts (for 1P A%HT 80%; A%RAW 40%; for 2P A%HT 20%; A%RAW 60%). Results obtained on commercial cheeses were confirmed by analyzing Fiore Sardo samples provided by the CPFSC, manufactured by the same operator and from the same milk and differing only for the heat treatment on milk. Interestingly, consistent results revealed by analyzing the commercial samples and the CPFSC cheeses suggest that heat treatment of milk is the major source of product differentiation, which therefore assumes a role of dominance over other variables (different producing areas, milk composition and cheese making procedures). It is worth noting that the observed area fractions are not necessarily a quantification of water in different sites, but are a result of all the dynamical processes occurring in cheese in the NMR time scales analyzed. Clearly, NMR transverse relaxation is revealing differences at the microscopic level between the raw milk and HT milk cheeses analyzed. The results obtained can be discussed according to the diffusive and chemical exchange model, since it is clear from the dynamics involved that 1P is mainly influenced by chemical exchange between water protons and labile protein protons. The analysis of the relaxation distributions obtained for the Fiore Sardo cheeses in light of the diffusive and chemical exchange model described above provided information on microstructures and dynamics involved at molecular scale, such as the size of the heterogeneity, the chemical exchange rate between water protons and labile protein protons, and water diffusive exchange constant.67

Both T1 and T2 NMR relaxation time constants are affected by protein denaturation and aggregation phenomena.42,44 Changes in T2 relaxation time have been often associated to


reduced mobility of water entrapped in the matrix of denatured and aggregated proteins.71 For what stated above, a reasonable hypothesis that may explain the different T2 distribution exhibited by HT (pecorino) and raw milk (Fiore Sardo PDO) cheeses could be formulated considering heat-treatment induced complex changes in milk proteins (dissociation, misfolding and aggregation, formation of complexes). It is realistically presumable that protein changes induce variation in water compartmentalization, favoring 1P with respect to 2P. Protein misfolding and aggregation likely lead to a denser and thicker network that water experiences in cheese paste,72 and protein aggregates in HT milk cheese increase the amount of interstitial water that strongly interact with protein protons.39 It may therefore be speculated that the observed differential relaxometric behaviour between HT milk and raw milk cheeses is due to the effects of pasteurization and thermization on whey proteins, that are more susceptible to structural and dynamical changes than caseins, or to their combination to form a denser packed structure in the three-dimensional network of cheese. However, further investigation is in progress to specifically address this issue since, although there is a wealth of information in literature in this sense, we have not yet given proper experimental evidence to this reasonable hypothesis. Some more remarks can be also made on the fitting algorithm used by Mulas et al. to deconvolute the multiexponential decay of CPMG curves obtained by MRI analysis of Fiore Sardo. It should be noted that all experimental parameters were adjusted so as to meet the guidelines for multiexponential analysis of relaxation data, as previously suggested for biomedical research and comprehensively reviewed.73-75 It should be noted that several criteria and methods have been used and compared for fitting T2 relaxation data, e.g. discrete methods, maximum entropy methods,6 stretched exponential models,32 Non-Negative-Least-Squares algorithm (NNLS),73-75 the latter being very robust and therefore the most frequently adopted in the biomedical field. In general, it would be advisable to exploit both discrete methods and NNLS when dealing with unknown systems. For said above, the two relaxation components found by Mulas et al. in the investigation on Fiore Sardo cheese can be reasonably considered to be real relaxation components rather than artifacts due to the data analysis procedure.67 In fact, it could be questioned that it is difficult for MRI to characterize short relaxation components centred at T2 values close to the CPMG interpulse spacing (i.e. close to the first echo time). Previous reports73-75 have demonstrated that appropriately setting experimental parameters (e.g. the echo time, i.e. the CPMG interpulse spacing) and acquiring MRI data with a high signal-to-noise-ratio (SNR>100) can satisfactorily solve the problem. If so, then the appearance of similar relaxation components would be expected at different CPMG interpulse spacings. Mulas et al. showed T2 distributions derived from MRI analysis of the same cheeses using interpulse spacings of 3.3 ms and 7.9 ms, evidencing only not relevant deviations with respect to each other in terms of number of components and mean T2 values.67 Figure 2 shows the T2 distributions of two sheep’s milk cheeses obtained by fitting the CPMG decay acquired with a benchtop NMR (4.7 T magnetic field, 20 MHz proton resonance frequency) and interpulse delay of 50 s. Such analysis further supports the existence of two T2 populations found by MRI analysis in the sheep’s milk cheeses analyzed, and substantiates the results of other studies carried out on different cheeses such as Mozzarella,63-66 Grana Padano,76 and other hard cheeses,77 in which the existence of similar populations has been previously uncovered.


Figure 2 An example of transverse proton relaxation spectra (region between 2 and 50 ms) obtained by analyzing two sheep’s milk hard cheeses with a 4.7 T magnetic field (20 MHz) benchtop NMR. CPMG sequence with interpulse spacing of 50 s was used, TR=5s, number of scans=16, 8k points, temperature = 40°C. Fitting of signal decay was performed by means of inverse Laplace Transformation (CONTIN). As a consequence of the foregoing statements, experimental evidences that would allow to quantitatively describe the differences between two cheeses are certainly to be sought in the transversal relaxation time constant T2. In this sense, this NMR parameter potentially assumes a great practical value. Moreover, it could be speculated that the fast relaxing T2 component found in many cheeses (Mozzarella,63-66 Fiore Sardo,67 Pecorino,67 Grana Padano,76 model hard cheeses,77 Parmigiano Reggiano),78 centred at about 5-10 ms and certainly highly influenced by the dynamics of chemical exchange between water protons and labile protein protons, can be associated to the subtle structural changes induced by heat treatments in whey proteins. Aiming to widen available NMR tools and to better describe diffusive and chemical exchange processes, a combination of transversal relaxation parameters T2 with the longitudinal one (T1) or with water self-diffusion coefficient (D) in 2D correlation methods have been proposed for cheese characterization.53,79 Such methods could have interesting practical developments also for inline analysis, providing that a faster data acquisition, an accurate quantitative analysis and proper resolution can be achieved80. T1-T2 and D-T2 maps have already successfully described molecular differences between different cheeses.53 Such measurements made the distinction between fat and water populations possible, and evidenced that the contribution of water protons in two dimensional (D-T2) maps of cheeses is much more informative than the fat protons, the former being significantly sample dependent and the latter similar in all dairy samples.53 In fact, fat relaxation in dairy products is widely dispersed, which leads to the observation of broad T2 distributions.12 However, more detailed NMR studies on the lipid profiles in cheeses are currently in progress in our laboratory. Such investigation could certainly shine a clearer light on peculiar features associated to cheese samples obtained by following different cheesemaking procedures.


4 CONCLUSION Over the years, Magnetic Resonance studies have allowed a detailed characterization of dairy products and their evolution in time, from raw materials (milk and related products) to ripened commercial cheeses. NMR is able to provide information on molecular processes occurring at very different time scales and from different spatial arrangements. Both high-resolution NMR spectroscopy, relaxometry, and diffusometry, have described the effect of dairy processing on the final physicochemical characteristics of dairy products, from dairy solutions and gels, to cheese analogues and commercial cheeses. The aforementioned NMR studies demonstrate that transverse relaxation of protons in cheese is a suitable tool to monitor the evolution of microstructural and dynamic features of dairy products during storage and aging (or ripening). In particular, this chapter focused on the effect of heat treatments on microstructure and molecular dynamics of dairy systems and cheeses in particular. Thanks the technological progresses, the wealth of fundamental and technological information on cheese processes, the optimization of theoretical background for the interpretation of experimental NMR results, together with the availability of cost-effective benchtop NMR instruments, a the more widespread diffusion of NMR instruments in the industry and within quality assurance and certification authorities is expected in the near future. References 1 P.T. Callaghan, K.W. Jolley, R.S. Humprey, J. Colloid Interf. Sci., 1983, 93, 521. 2 P. Lambelet, R. Berrocal and F. Ducret, J. Dairy Res., 1989, 56, 211. 3 S.P.F.M. Roefs, H. Van As, T. Van Vliet, J. Food Sci., 1989, 54, 704. 4 P. Lambelet, R. Berrocal and F. Renevey, J. Dairy Res., 1992, 59, 517. 5 M. Rosenberg, M. McCarthy and R. Kauten, J. Dairy Sci., 1992, 75, 2083. 6 C. Tellier, F. Mariette, J. Guillement and P. Marchal, J. Agric. Food Chem., 1993, 41,

2259. 7 T. Van Vliet and P. Walstra, J. Food Eng., 1994, 22, 75. 8 S.L. Duce, M.H.G. Amin, M.A. Horsfield, M. Tyszka and L.D. Hall, Int. Dairy J., 1995,

5, 311. 9 R. Ruan, K. Chang, P.L. Chen, R.G. Fulcher and E.D. Bastian, J. Dairy Sci., 1998, 80, 9. 10 J. Belloque and M. Ramos, Trends Food Sci. Technol., 1999, 10, 313. 11 F. Mariette in Magnetic Resonance in Food Science: Latest Developments, ed. P.S.

Belton, A.M. Gil, G.A. Webb and D. Rutledge, The Royal Society of Chemistry, Cambridge, 2003, p 209.

12 F. Mariette in Modern Magnetic Resonance, ed. G.A. Webb, Springer Netherlands, 2006, p 1697.

13 F. Mariette in Modern Magnetic Resonance, ed. G.A. Webb, Springer Netherlands, 2006, p 1801.

14 A.D. Maher and S. J. Rochfort, Metabolites, 2014, 4, 131. 15 J. van Duynhoven, A. Voda, M. Witek and H. van As, Annu. Rep. NMR Spectrosc.,

2010, 69, 145. 16 F. Mariette, D. Topgaard, B. Jonsson and O. Soderman, J. Agric. Food Chem., 2002, 50,

4295. 17 A. Metais, M. Cambert, A. Riaublanc and F. Mariette, J. Agric. Food Chem., 2004, 52,

3988. 18 R. Colsenet, F. Mariette and M. Cambert, , J. Agric. Food Chem., 2005, 53, 6784. 19 A. Metais, M. Cambert, A. Riaublanc and F. Mariette, Int. Dairy J., 2006, 16, 344.


20 A. Gottwald, L.K. Creamer, P.L. Hubbard and P. Callaghan, J. Chem. Phys., 2005, 122, 34506.

21 R. Hinrichs, S. Bulca and U. Kulozik, Int. J. Dairy Technol., 2007, 60, 37. 22 S. Le Feunteun and F. Mariette, Macromolecules, 2008, 41, 2071. 23 S. Le Feunteun, M. Ouethrani, F. Mariette, Food Hydrocolloids, 2012, 27, 456. 24 S. Salami, C. Rondeau-Mouro, J. Van Duynhoven, F. Mariette, Food Hydrocolloids,

2013, 31, 248. 25 H. Li, C.C. Hardin and E.A. Foegeding, J. Agric. Food Chem., 1994, 42, 2411. 26 N. Tanaka, S. Kunugi, Int. J. Biol. Macromol., 1996, 18, 33. 27 J. Belloque and G.M. Smith, J. Agric. Food Chem., 1998, 46, 1805. 28 L.N. Arnaudov, R. de Vries, H. Ippel and C.P.M. van Mierlo, Biomacromolecules, 2003,

4, 1614. 29 F.P. Duval, J.P.M. van Duynhoven and A. Bot, J. Am. Oil Chem. Soc., 2006, 83, 905. 30 M.H. Oztop, M. Rosenberg, Y. Rosenberg, K.L. McCarthy and M. J. McCarthy, J. Food

Sci., 2010, 75, E508. 31 M. Budiman, R.L. Stroshine and O.H. Campanella, J. Texture Stud., 2000, 31, 477. 32 M. Budiman, R.L. Stroshine and P. Cornillon, J. Dairy Res., 2002, 69, 619. 33 N. Noronha, E. Duggan, G.R. Ziegler, E.D. O’Riordan, M.O. Sullivan, Int. Dairy J.,

2008, 18, 641. 34 J.M. Arimi, E. Duggan, M.O. Sullivan, J.G. Lyng, E.D. O’Riordan, J. Food Eng., 2008,

89, 258. 35 J.M. Arimi, E. Duggan, M.O. Sullivan, J.G. Lyng, E.D. O’Riordan, Food Chem., 2010,

121, 509. 36 P.F. Fox and P.L.H. McSweeney in Dairy Chemistry and Biochemistry, Blackie

Academic & Professional, Thomson Science, London, 1998, p 368. 37 H.S. Rollema and J.A. Brinkhuis, J. Dairy Res., 1989, 56, 417. 38 P.S. Bansal, P.A. Grieve, R.J. Marschke, N.L. Daly, E. McGhie, D.J. Craik, P.F.

Alewood, Biochem. Biophys. Res. Commun., 2006, 340, 1098. 39 J.N. de Wit and G. Klarenbeek, J. Dairy Sci., 1984, 67, 2701. 40 W. Humphrey, A. Dalke and K. Schulten, J. Molec. Graphics, 1996, 14, 33. 41 S. Brownlow, J.H. Morais Cabral, R. Cooper, D.R. Flower, S.J. Yewdall, I. Polikarpov,

A. CT. North and L. Sawyer, Structure, 1997, 5, 481. 42 B.P. Hills, S.F. Takacs and P.S. Belton, Mol. Phys., 1989, 67, 919. 43 A.L. Kelly, N. Datta and H.C. Deeth, in Thermal Food Processing: New Technologies

and Quality Issues, Chapter 9, page 267. 44 L. Indrawati, R.L. Stroshine and G. Narsimhan, J. Sci. Food Agric., 2007, 87, 2207. 45 H. Singh, Int. J. Dairy Technol., 2004, 57, 111. 46 A.L. Kelly, N. Datta and H.C. Deeth, in Thermal Food Processing: New Technologies

and Quality Issues, Chapter 9, page 290. 47 B.P. Hills, S.F. Takacs and P.S. Belton, Food Chem., 1990, 37, 95. 48 F. Mariette, C. Tellier, G. Brulè and P. Marchal, J. Dairy Res., 1993, 60, 175. 49 M.C.A. Griffin and G.C.K. Roberts, Biochem. J., 1985, 228, 273. 50 P.E. Morgan, T.M. Treweek, R.A. Lindner, W.E. Price and J.A. Carver, J. Agric. Food

Chem., 2005, 53, 2670. 51 L.T. Kakalis, T.F. Kumosinski and H.M. Farrell jr, Biophys. Chem., 1990, 38, 87. 52 A. Le Dean, F. Mariette and M. Marin, J. Agric. Food Chem., 2004, 52, 5449. 53 Y.Q. Song, Prog. Nucl. Magn. Reson. Spectrosc., 2009, 55, 324. 54 A.L. Kelly, N. Datta and H.C. Deeth, in Thermal Food Processing: New Technologies

and Quality Issues, Chapter 9, page 291.


55 A. Castell-Palou, C. Rossellò, A. Femenia, S. Simal, Food Bioprocess Technol., 2013, 6, 2685.

56 V.D. Fedotov, F.G. Miftakhutdinova and S.F. Murtazin, Biophyzica, 1969, 14, 873. 57 J.P. Carver and R.E. Richards, J. Magn. Reson. 1972, 6, 89. 58 P.S. Belton and B.P. Hills, Mol. Phys., 1987, 61, 999. 59 B.P. Hills, S.F. Takacs and P.S. Belton, Mol. Phys., 1989, 67, 903. 60 E. Brosio, M. Belotti and R. Gianferri in Food Science and Technology: New Research,

ed. L.V. Greco and M.N. Bruno, Nova Science Publishers, Inc., Hauppauge NY, 2008, p 323.

61 P.S. Belton, B.P. Hills and E.R. Rimbaud, Mol. Phys., 1988, 63, 825. 62 T.J. Swift, R.E. Connick, J. Chem. Phys., 1962, 37, 307. 63 R. Gianferri, M. Maioli, M. Delfini, E. Brosio, Int. Dairy J., 2007, 17, 167. 64 R. Gianferri, V. D’Aiuto, R. Curini, M. Delfini, E. Brosio, Food Chem., 2007, 105, 720. 65 M.I. Kuo, S. Gunasekaran, M. Johnson and C. Chen, J. Dairy Sci., 2001, 84, 1950. 66 M.I. Kuo, M.E. Anderson and S. Gunasekaran, J. Dairy Sci., 2003, 86, 2525. 67 G. Mulas, T. Roggio, S. Uzzau and R. Anedda, J. Dairy Sci., 2013, 96, 7393. 68 P.F. Fox and P.L.H. McSweeney in Dairy Chemistry and Biochemistry, Blackie

Academic & Professional, Thomson Science, London, 1998, p 347. 69 M.F. Scintu, A. Del Caro, P.P. Urgeghe, C. Piga and R. Di Salvo, J. Sens. Stud., 2010,

25, 577. 70 H. Singh, A. Waungana, Int. Dairy J., 2001, 11, 543. 71 B. Halle, Philos. Trans. R. Soc. Lond. Ser. B-Biol. Sci., 2004, 359, 1207. 72 M.F. Morales-Celaya, C. Lobato-Calleros, J. Alvarez-Ramirez, E.J. Vernon-Carter,

LWT-Food Sci. Technol., 2012, 45, 132. 73 F.R.E. Fenrich, C. Beaulieu and P.S. Allen., NMR Biomed., 2001, 14, 133. 74 S.J. Graham, , P.L. Stanchev and M.J. Bronskill, Magn. Reson Med. 1996, 35, 370. 75 D. Laule, I.M. Vavasour, S.H. Kolind, D.K.B. Li, T.L. Traboulsee, G.R.W. Moore and

A.L. MacKay, Neurotherapeutics, 2007, 4, 460. 76 S. de Angelis Curtis, R. Curini, M. Delfini, E. Brosio, F. D’Ascenzo, B. Bocca, Food

Chem., 2000, 71, 495. 77 B. Chaland, F. Mariette, P. Marchal and J. de Certaines, J. Dairy Res., 2000, 67, 609. 78 A. Bordoni, G. Picone, E. Babini, M. Vignali, F. Danesi, V. Valli, M. Di Nunzio, L.

Laghi and F. Capozzi, Magn. Reson. Chem., 2011, 49, 561-570. 79 S. Godefroy, P.T. Callaghan, Magn Reson. Imaging, 2003, 21, 381. 80 D. Bernin, D. Topgaard, Curr. Opin. Colloid Interface Sci., 2013, 18, 166.

NMR SPECTROSCOPIC STUDIES IN SAFFRON AUTHENTICITY AND QUALITY (WITHIN THE FRAME SAFFRONOMICS Cost action FA1101)

R. Consonni*, L. R. Cagliani*, M. G. Polissiou§, E. A. Petrakis§, M. Z. Tsimidou#, S. Ordoudi#

*Institute for Macromolecular Study, v. Bassini 15, 20133 Milan, Italy.§ Laboratory of Chemistry, Department of Food Science and Human Nutrition, AgriculturalUniversity of Athens, Iera Odos 75,11855, Athens, Greece. #Aristotle University of Thessaloniki, School of Chemistry, Laboratory of Food Chemistry and Technology, 54124, Thessaloniki, Greece.

1 INTRODUCTION

Saffron, the most expensive spice in the world market, is according to the trade standard ISO 36321 obtained from the pistils of Crocus sativus L. flowers after drying. Drying is carried out by producers by rather traditional treatments. The latter differ comparatively from each other according to geographical origin2. The high market value of the material is mainly related to labor cost as the cultivation is not mechanized yet, despite the many centuries of known use in foods and ethnopharmacology. Saffron comes from only a few areas of the world, Iran being the major producer in Asia. India comes next whereas growing is the interest in neighboring to them countries (Afghanistan, China). In Europe, the use of pistils of crocuses, which are now studied as possible progenitors of the sterile triploid C. sativus, is evidenced in the frescoes of Akrotiri (Santorini, Greece) and ancient Greek and Roman literature. Production is currently coming from producer associations in Kozani region (Greece) and Castilla la Mancha (Spain). The respective products (Krokos Kozanis and Azafrán de Castilla la Mancha) have been registered as PDO products. A revival of the cultivation is observed in certain regions of Italy and in particular Sardinia, Abruzzo and Tuscany received PDO stamp (Turri - Villanovafranca - S. Gavino Monreale, L’Aquila and S. Gimignano) while small farmers are present elsewhere. Among Maghreb countries, Morocco presents activity in organized saffron production. The interest in cultivation is expanded within and beyond Europe though the limitation of legally traded C. sativus corms is an obstacle. Plant material from nurseries has not been always found effective in daughter corm yield, number of flowers per corm and length of pistils. COST ACTION FA11001 SaffronOMICS (2011-2015) according to the Memorandum of Understanding (www.saffronomics.org) is a concerted cross-project and a multidisciplinary approach that succeeded so far in bringing together experts, who jointly address issues of authenticity, quality control and origin of saffron among the many other objectives. High throughput techniques such as NMR, that facilitate the examination of plant metabolites, are also aimed for saffron within the SaffronOMICS frame. NMR techniques present a great


potential to establish new criteria for the above-mentioned issues beyond those proposed by ISO 3632 trade standard and literature applications of other spectroscopic techniques (e.g. UV-Vis, fluorescence, Near Infrared, Mid Infrared). ISO 3632 established quality criteria for authenticity, quality and origin control using spectroscopic methods, exclusively based on spectrophotometric measurements. NMR and other spectroscopic techniques, hereafter presented, are largely adopted because of their fast and nondestructive properties.

2 SPECTROSCOPIC INVESTIGATIONS OF SAFFRON

2.1. UV-Vis Spectrometry UV-Vis spectrometry has been applied to tentatively quantify quality characteristics for the commercial categories of saffron in filaments, cut filaments and powder form (Table 1).

Table 1 Saffron quality characteristics and specifications according to ISO 3632-11.

Quality characteristics Specifications Filaments and cut filaments Powder

Moisture and volatile matter content,%, max 12 10

E1% 257nm on dry basis, min (due to the absorbance of picrocrocin) Category I Category II Category III

70 55 40

70 55 40

E1% 330 nm, on dry basis (due to the absorbance of safranal) Min Max

20 50

20 50

E1% 440nm, on dry basis, min (due to the absorbance of crocins) Category I Category II Category III

200 170 120

200 170 120

These quality criteria are set for the three major attributes of saffron as spice, i.e. (a) the coloring strength due to the presence of a group of water soluble apocarotenoids, the crocins; (b) the aroma strength expressed as safranal and (c) the flavour strength expressed as picrocrocin. Expression of quantitative results is as E 1% at max. Chemical structures of the major compounds responsible for the above-mentioned quality attributes are given in Fig. 1. The major crocin is trans-4-GG, representing more than 60% of the total crocetin esters4. In the case of superior quality products, crocetin esters represent 20 - 37 % w/w of dry weight of saffron4,5. Picrocrocin content4,6 (7 - 27 % w/w) together with that of total crocins normally account for 50% of saffron dry weight. It is evident that the product is quite rich in these secondary metabolites currently appreciated for their beneficial properties against Alzheimer and cardiovascular diseases7, gastric disorders, depression etc. A typical spectrum of an aqueous extract is given in Fig.2.


Figure 1 Chemical structures of picrocrocin (I), safranal (II) and major esters of trans- (III) and cis-crocetin (IV) present in C. sativus polar extracts. Nomenclature is by Carmona et al.3


Figure 2 Typical UV-Vis spectrum of saffron aqueous extracts

Authenticity issues reported in the same trade standard are related with methods for ensuring the absence of artificial colorants (yellow and red synthetic acidic ones). These colorants are isolated after a well-defined protocol (Scheme 1) and are then characterized by TLC (screening) and RP-HPLC coupled with a diode array detector (DAD).

Scheme 1 Flow diagram of the sample pretreatment for the detection of artificial colorants in saffron, according to ISO 3632 specifications8

Adulterants like Sudan dyes, or other plant material such as gardenia, safflower and buddleja flowers are not considered in the trade standard. Origin issues are not addressed either in the


ISO 3632 specifications or in any national legislation. To this extent there is an obvious lack of support of PDO products by objective means. Advances in knowledge on saffron composition along with new available analytical tools and global market trends have led a substantial body of international research to focus on the improvement of the ISO standard specifications and methods. For example, it is well documented by means of chromatographic analysis that absorbance values at 257 nm and 330 nm are not specific for picrocrocin and safranal, respectively3,9,10. Alternative protocols for the spectrophotometric estimation of picrocrocin and safranal content have been recently proposed by Alonso and co-workers11,12 based on calculation of the pic index or extraction with less polar solvents, respectively. The ISO-proposed procedure for the extraction of saffron active metabolites seems to also attract the interest of researchers. Apart from solid-liquid extraction with different kinds of solvents13-15, nano-emulsions16 or molecularly imprinted polymer solid-phase extraction17 have been proposed the last years. In all cases, UV-Vis absorption values were monitored during optimization experiments. Considering the detection of artificial colorants in saffron, the latest version of the ISO 3632-2 trade standard8 suggests an alternative sample pretreatment protocol in case “the HPLC chromatograms are unacceptably contaminated by peaks due to the natural pigments of saffron”. This protocol is actually a modification of that described few years earlier by Zalacain et al.18. The original procedure allows direct detection of the exogenous dyes by means of 2nd derivative UV-Vis spectra instead of applying the ISO TLC method. The method was found inappropriate for erythrosine and carminic acid due to low recoveries during sample pretreatment. To partially overcome the limitations of detecting pH-sensitive dyes such as erythrosine, Ordoudi and Tsimidou19 suggested that fluorescence properties of suspected saffron extracts could be investigated. On the basis of measurements at 532 nm excitation/548 nm emission wavelengths, traces of erythrosine in saffron (0.04 mg/Kg) were detected even in the presence of other synthetic dyes. It is worth noting that so far, applications of fluorescence spectroscopy to saffron analysis are extremely rare20. 2.2. Infrared and Raman spectroscopy Infrared and Raman spectroscopy are versatile, non-destructive analytical techniques that provide spectral fingerprints of many analytes. Both vibrational techniques are rapid, inexpensive and require minimal or no sample preparation21-24. The available studies report their successful applications for characterization and verification of the quality and authenticity of saffron. Over the last few decades, Fourier transform Infrared spectroscopy (FT-IR) has been used to characterize and assess quality of saffron. Tarantilis et al. reported the FT-IR spectra of crocetin esters (crocins) and their derivatives, di-methyl-crocetin (DMCRT) and crocetin (CRT),25 where characteristic absorbance bands in the spectral regions from 1706 to 1664 cm-

1 (C=O stretching vibrations) and from 1243 to 1228 cm-1 (C-O stretching vibrations) were observed. Several studies followed, examining quality parameters and authentication of saffron by means of Mid Infrared or Near Infrared spectroscopy. Fourier transform Mid Infrared spectroscopy (FT-MIR) in conjunction with multivariate analysis has been employed for the geographical determination of 250 saffron samples from four countries, i.e. Greece, Iran, Italy and Spain26. FT-MIR spectra were recorded either in diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) mode for powdered saffron samples or in transmission mode, using ZnSe windows, with reference to non-polar (diethyl ether) saffron extracts. The latter proved more suitable for the geographical identification of the samples. In particular, the application of canonical discriminant analysis (DA) to the spectroscopic data of diethyl ether extracts in the region 2000-700 cm-1 resulted in 77.2% correct classification


of samples. The second derivative of the same spectral region provided even better results, as 93.6% of the samples were correctly classified. Additionally, the first two discriminant functions explained the 93.4% of the total variance, with function 1 differentiating Italian samples from all the others; function 2 accounted for the discrimination among the three remaining groups of samples. Italian samples were best characterized by the spectral region associated with the ester carbonyl group, at around 1746 cm-1. This clearly appears because the Italian samples came from Sardinia, where stigmas are traditionally wetted with extra virgin olive oil and thus triglycerides occur in those samples. The process is known as “feidatura” and takes place before drying the stigmas in order to enhance their preservation and appearance. The spectral region assigned to C=C stretching at around 1600 cm-1 and the band at 1670 cm-1 that is attributed to the aldehyde carbonyl group of safranal were responsible for the differentiation among the samples from Greece, Iran and Spain. FT-MIR proves to be an efficient technique for dealing with saffron fraud issues as revealed so far within the framework of Saffronomics (COST Action FA1101). Very recently, the application of FT-MIR combined with multivariate analysis was reported to be very useful for monitoring storage effects and detection of deterioration of saffron due to ageing27. A total of 52 saffron samples were used, including both fresh and aged samples, split in reference and test sets. The spectra were recorded in transmission mode using KBr discs. Principal component analysis (PCA) was carried out by considering spectroscopic data of selected characteristic bands in the region 1800-900 cm-1, with the first two latent variables accounting for the 95% of the total variance (PC1 = 63% and PC2 = 32%). To investigate correlations among the data obtained by HPLC analysis and the PC score values, multiple linear regression (MLR) analysis was also performed, suggesting 35.7% and 52.3% of the variance in PC1 and PC2 values, respectively, due to the variance in the levels of all the major apocarotenoids occurring in saffron samples. Also, picrocrocin content represented 25.5% and 39.9% of the observed variance, respectively. The study concludes that the band at 1028 cm-1, linked with the presence of glucose moieties, as well as the intensities in the region 1175-1157 cm-1, which relate to the breakage of glycosidic bonds, are characteristic for detecting deterioration of commercial saffron. The potential of FT-NIR spectroscopy has been examined for the determination of chemical composition along with the geographical discrimination of 111 saffron samples originated in Greece, Iran and Spain28. Near-infrared spectroscopic data were acquired in reflectance mode and principal component regression (PCR) was performed using reference data obtained by UV-Vis spectrophotometry and HPLC with diode array detection (HPLC-DAD). The results obtained by calibrating and validating the corresponding PCR models indicated the ability of FT-NIR combined with multivariate analysis to determine moisture and volatile content, coloring strength, (250 nm) and (330 nm) as well as the content of the five main crocetin esters and picrocrocin. DA was carried out to separate the samples according to their origin, providing high percent recognition for each group; 100% for Iranian samples and approximately 95% for Greek and 88% for Spanish ones. Additionally, it was reported that Iranian samples were the most different, while Greek and Spanish samples appeared more similar. The first and the second discriminant functions explained 83.8% and 12.2% of the total variance, respectively. Fourier transform Raman (FT-Raman) spectroscopy has provided significant information for the characterization of saffron-related carotenoids over the past years. The FT-Raman analysis of crocetin esters, DMCRT and CRT has revealed two main peaks near 1540 and 1166 cm-1 assigned to C=C and C-C stretching modes, respectively25. FT-Raman spectra are complementary to those obtained by FT-IR concerning the structural information provided for these compounds. FT-Raman has also been employed for the characterization of the cis-trans carotenoids contained in saffron29. The spectra recorded for the major isomers isolated from saffron, all-trans- and 13-cis-Crocetin-di-( -D-gentiobiosyl)


ester, presented substantial differences. A strong single peak at 1535 cm-1 is present for the all-trans isomer, whereas for the 13-cis isomer this peak shifts to 1547 cm-1 and another peak is observed at 1581 cm-1. For the all-trans isomer a medium peak at 1165 cm-1 and a weak peak at 1209 cm-1 are also observed, while for the 13-cis isomer there is another characteristic peak at 1138 cm-1 and the intensity of the peak at 1166 cm-1 is decreased. A more recent study involved dispersive Raman spectroscopy coupled with multivariate regression for the determination of crocetin esters content and coloring strength of saffron30. In this quantitative approach, Raman spectra of 114 ground saffron samples from four different countries (i.e. Greece, Iran, Italy and Spain) were used. Calibration and validation sample sets were extracted and calibration models using partial least squares (PLS) regression were developed for determining both parameters, in the spectral region 1700-955 cm-1. The number of PLS factors used for the quantification of crocetin esters was six (accounting for 98% of the model variability), while four factors were used for the coloring strength (accounting for 96% of the model variability). PLS models were validated using leave one out cross validation procedure (r = 0.97, RMSECV = 1.09 for crocetin esters and r = 0.93, RMSECV = 14.5 for coloring strength). Both models allowed accurate predictions, compared with HPLC-DAD and UV-Vis reference data, indicating that the suggested approach can be used for rapid screening of saffron quality. 2.3. NMR spectroscopy NMR spectroscopy is a well-known spectroscopic technique largely employed because of its intrinsic structural characterization properties. Beyond this feature, the possibility to detect several classes of chemical compounds within a single experiment, without the need of any chemical sample derivatization encodes additional advantages to NMR spectroscopy thus encouraging the application in food analysis. The amount of data obtained from the NMR spectra could be easily handled by the multivariate statistical approaches, broadly appeared in the last years, allowing clustering of samples according to several aims. As already pointed out, a part from the chemical characterization of the chemical components, other information about origin, ageing and unwanted components are strongly required in order to increase the quality requirements of foods. The first NMR data about a group of constituents of saffron, was reported by Wittwer31 in late 1975 on isolated glycosyl esters of crocetin from saffron. The evidence of a geometrical isomer of crocin was reported by Speranza32, highlighting the spectroscopic characterization of 13-cis crocins together with the most abundant all trans-crocins in a Greek saffron sample. In this work, crocins were previously isolated by HPLC and further structurally characterized by NMR, thus confirming the hypothesized structure. These isomers were successively characterized by means of other spectroscopic techniques like UV-Vis and FT Raman33. Pfister and co-workers34 elucidated the structure of two new glycosyl esters of crocetin in both Crocus sativus L. and Gardenia jasminoides Ellis, being crocetin ( -gentiobiosyl)( -neapolitanosyl) and crocetin di( -neapolitanosyl) ester. The structural characterization of crocetin glycosides was always the main aim of other studies. Van Calsteren35 investigated the HPLC extracted and purified crocetin derivatives from Gardenia jasminoides and Crocus sativus by UV-Vis and NMR spectroscopies. In this work the authors confirmed the structural characterization of different carotenoids, in particular extracts dissolved in both DMSO and methanol/benzene mixture, four glycosides were characterized for gardenia: a) all-trans-crocetin di( -D-gentiobiosyl) ester b) all-trans-crocetin -D-gentiobiosyl- -D-glucosyl ester c) all-trans-crocetin mono( -D-gentiobiosyl) ester and d) 13-cis-crocetin -D-gentiobiosyl- -D-glucosyl ester. Additionally, six glycosides were characterized for saffron: a) all-trans-crocetin di( -D-gentiobiosyl) ester b) all-trans-crocetin ( -D-gentiobiosyl- -D-glucosyl ester c) all-trans-crocetin di( -D-glucosyl) ester d) all-trans-crocetin mono( -D-


gentiobiosyl) ester e)13-cis-crocetin di( -D-gentiobiosyl) ester f) 13-cis-crocetin -D-gentiobiosyl- -D-glucosyl ester. All these derivatives were also supported by mass spectrometry analysis. Interestingly, crocetin and corresponding glycosyl esters were isolated from a wild Crocus species, Crocus haussknechtii Boiss, and investigated by different techniques, including NMR36. It was found a similar carotenoid composition with C. sativus, thus suggesting this species as a potential source of saffron. During these last years, the chemical analysis of mixtures in food extracts moved towards metabolomics rather than a classical approach based on chemical isolation followed by structure elucidation. This choice relies on the easier sample preparation, and especially on the possibility to detect several classes of chemical compounds in their original ratio simultaneously. In this direction, appeared quite recently the NMR works of Yilmaz37,38, based on the metabolic fingerprinting of saffron extracts with the aim to distinguish among authentic Iranian saffron and commercial samples obtained from retail stores in Denmark, Sweden and Turkey. The reported data suggested the possibility to obtain the desired discrimination by PCA and Parallel Factor analysis by using mono and two dimensional NMR data respectively. Unfortunately the authors did not report a detailed metabolite content, like different crocetin esters, and only partial resonance assignment was performed. Analysis of NMR spectra for different groups of samples revealed the presence of food additive (E1518) in one group, while in other groups the presence of bio-adulterants, like C. sativus stigmata, Curcuma longa and Carthamus tinctorius flowers were detected. Within the frame of SaffronOMICS COST Action FA1101, our group started an NMR based metabolic profiling study of saffron with different aims. We have analyzed saffron extracts in different solvents and from different origins in order to put light in the quality determination of this valuable food product. Here we present the preliminary study performed in water extracts of Italian PDO saffron samples, with the aim of metabolite characterization. Additionally, commercial saffron bought in Italian markets and samples bought in other countries were analysed by comparison.

3 METHODS AND RESULTS

3.1 NMR analysis A total number of 20 saffron samples were investigated by 1H NMR: 8 Italian PDO, 4 from other countries (1 from Thailand, 1 from Morocco, 1 from Turkey and 1 from India) and 8 commercial saffron bought in Italy. Concerning the Italian PDO samples, all harvested in 2005, 3 were from Consortium of L’Aquila, 1 from S. Gimignano, 3 from Sardinia and 1 was from Florentine hills (producers refused the obtained registration as PDO product). Samples from other countries were bought in 2006 in local markets while the 8 Italian commercial saffron samples were directly bought in stores in 2006 and had no indication of origin or harvesting year on the label. All samples were stored in the same conditions (in dark at room temperature) up to the spectral data acquisition. Two replicates were taken for each sample, prepared in double after sample homogenization to reduce sample variability and check the NMR measurement reproducibility. About 8 mg of saffron were dissolved in 600 L of deuterated water. Samples were then centrifuged at 12100 rcf for 10 minutes and 500 L of the supernatant was used for the NMR analysis. All 1H-NMR spectra have been recorded on a Bruker DMX 500 spectrometer (Bruker Biospin GmbH Rheinstetten, Karlsruhe, Germany) operating at 11.7 T and equipped with a 5-mm reverse probe with z-gradient. Spectra were recorded at 300 K, with a spectral width of 7500 Hz and 32K data points. Solvent suppression was achieved by applying a presaturation scheme with low power radiofrequency irradiation


for 1.2 s. Spectra were Fourier transformed without any resolution enhancement function and manually corrected for phase and baseline with ACD/Spec Manager (ACD Labs, version 11, Toronto, Canada) software. Spectra in water solution were referenced to trimethylsilyl (2,2,3,3-2H4) propionate (TSP) external standard. Good spectra alignment for bucket integration was obtained using the -glucose signal at 5.23 ppm. Spectra were reduced to integrated regions (buckets) of 0.04 ppm width by an intelligent bucketing procedure, covering the spectral region between 11.26 and 0.50 ppm, followed by manual correction of buckets for shifted signals. Complete spectrum area was used for calibration after exclusion of residual water region between 4.61 and 5.03 ppm. NMR data were imported into SIMCA-P+ 12 (Umetrics, Umea, Sweden) for Principal Component Analysis (PCA) and Orthogonal Projection to Latent Structures-Discriminant Analysis (OPLS-DA) using “pareto” as data pretreatment. All models were cross validated. 1H NMR spectrum of PDO saffron in water solution, showed the dominant resonances of the largely abundant components content: at low field the singlet of the aldehydic proton of 2,6,6-trimethyl-1-cyclohexene-1-carboxaldehyde moiety at 9.98 ppm, typically observed for picrocrocin and the group of broad signals between 6 and 7.4 ppm relative to double bonds of crocetin glycosides, these latter scarcely dissolved in this solvent. At high field the 1H spectrum is dominated by the very intense methyls of picrocrocin at 1.21, 1.23 and 2.15 ppm; other signals of picrocrocin were occurring at 1.59, 1.87, 2.37, 2.75, and 4.20 ppm as confirmed by TOCSY and HSQC experiments. Glycosidic moieties could be identified in the anomeric region of 1H NMR spectra, indicating the presence of both glucosyl and gentiobiosyl esters of crocetin, overlapped in a broad signal at 5.64 ppm, free and -glucose at 5.23 and 4.64 ppm respectively while at 4.62 ppm glucosyl of picrocrocin. These assignments were confirmed by comparison with HSQC saffron spectrum and those of reference substances. Samples bought in Thailand, Turkey, India and Morocco resulted depleted of picrocrocin content. In particular, those from India and Morocco present a specific spin system, consisting of signals at 1.14, 3.63 and 3.72 ppm. This unknown compound is largely present in these samples and further investigations are in progress. The integrated buckets of all spectra constituted the data matrix for multivariate statistical analysis. Initially, PCA was performed by considering all saffron samples: by scoring the first two PC’s, a clear differentiation was achieved. All Italian PDO and commercial saffron bought in Italy grouped on the right side of the score plot. Conversely, samples of other countries resulted in the opposite direction and in particular saffron samples from India and Morocco clustered on the top while samples from Turkey and Thailand on the bottom. The corresponding loading plot highlighted sugars (buckets at 3.80, 3.69, 3.74 and 3.97 ppm) and unknown compounds (buckets at 1.11 and 1.90 ppm) as the characteristic metabolites for these latter samples: conversely, all PDO and commercial samples bought in Italy resulted strongly characterized by buckets at 1.17, 1.84, 2.10, 3.36 and 3.44 ppm corresponding to picrocrocin.


Figure 3 PCA score plot representing the first two PC’s of all saffron samples (R2=90.7% with Q2=77.4%)

Figure 4 OPLS-DA score plot of commercial and PDO samples (R2X=75.4%, R2Y=82.2%, Q2=62.1%)

After this explorative PCA was performed with all samples, a two-class OPLS-DA was implemented by considering PDO and commercial saffron bought in Italy, to highlight compositional differences between them. From the score plot represented in Fig. 4, a clear differentiation was achieved. The analysis of the S-plot, highlighted NMR signals responsible for the separation between the two groups of samples; in particular picrocrocin resulted the characteristic compound for the PDO samples (buckets at 1.17, 1.84, 2.10 and 9.95 ppm), indicating the high quality of these products.


4 CONCLUSIONS Saffron is a very valuable spice and for this reason it is exposed to large quality manipulation, including plant, colorant and chemical additions. As pointed out previously, ISO regulations do not address origin issues and this is no longer acceptable, especially in the view of PDO products and open trade markets, that strongly require analytically supported origin declarations. Moreover, plant tissues containing apocarotenoids, like Gardenia jasminoides Ellis, could impair the spectrophotometric quantifications. In order to compensate this evident limitation, advanced spectroscopic techniques such as the abovementioned ones need to be applied since they have already been recognized as objective analytical tools able to detect frauds, verify authenticity and even establish origin of PDO products. References

1 International Organization for Standardization. Saffron (Crocus sativus L.) specifications. Geneva (Switzerland): ISO 2011, ISO/TS 3632-1.

2 S. A. Ordoudi and M. Z. Tsimidou, Production Practices and Quality Assessment of Food Crops, R. Dris, S. M Jain, (Eds)., Kluwer Academic Publ. Dordrecht, Netherlands 2004, 209-260.

3 M. Carmona, A. Zalacain, A. M. Sánchez, J. L. Novella and G. L. Alonso, J. Agric. Food Chem., 2006a, 54, 973.

4 A. M. Sánchez, M. Carmona, M. Prodanov and G. L. Alonso, J. Agric. Food Chem., 2008, 56, 7293.

5 M. Lage and C. L. Cantrell, Scientia Horticulturae, 2009, 121, 366. 6 C. P. Del Campo, M. Carmona, L. Maggi, C. D. Kanakis, E. G. Anastasaki, P. A.

Tarantilis and G. L. Alonso, J. Agric. Food Chem., 2010, 58, 1305. 7 S. H. Alavizadeh and H. Hosseinzadeh, Food Chem. Toxicol. 2014, 64, 65. 8 International Organization for Standardization, Saffron (Crocus sativus Linnaeus) Test

methods. Geneva (Switzerland): ISO 2010, ISO/TS 3632-2. 9 A. M. Sánchez, M. Carmona, M. Prodanov and G. L. Alonso, J. Agric. Food Chem.,

2008, 56, 7293. 10 P. A. Tarantilis, G. Tsoupras and M. G. Polissiou, J. Chromatogr. A,1995, 699, 107. 11 C. P. Del Campo, M. Carmona, L. Maggi, C. D. Kanakis, E. G. Anastasaki, P. A.

Tarantilis, M. G. Polissiou and G. L. Alonso, J. Agric. Food Chem., 2010, 58, 1305. 12 L. Maggi, A. M. Sánchez, M. Carmona, C. D. Kanakis, E. Anastasaki, P. A. Tarantilis,

M. G. Polissiou and G. L. Alonso, Food Chem., 2011, 127, 369. 13 A. M. Sani and S. Mohseni, Nutr. Food Sci., 2014, 44, 2. 14 A. Kyriakoudi, A. Chrysanthou, F. Mantzouridou and M. Z. Tsimidou, Anal. Chim.

Acta, 2012, 755, 77. 15 O. Orfanou and M. Z. Tsimidou, Food Chem. 1996, 57, 463. 16 B. Mokhtari and K. Pourabdollah, Indian J. Chem. Techn., 2013, 20, 222. 17 S. A. Mohajeri, H. Hosseinzadeh, F. Keyhanfar and J. Aghamohammadian, J. Sep. Sci.,

2010, 33, 2302. 18 A. Zalacain, S. A. Ordoudi, I. Blázquez, E. M. Díaz-Plaza, M. Carmona, M. Z.

Tsimidou and G. L. Alonso, Food Addit. Contam., 2005, 22, 607. 19 S. A. Ordoudi and M. Z. Tsimidou, Food Addit. Contam. Part A, 2011, 28, 417. 20 V. Vickackaite, A. Romani, D. Pannacci and G. Favaro, Int. J. Photoen. 2004,6, 175. 21 G. Downey, TRAC-Trend. Anal. Chem., 1998, 17, 418. 22 L. M. Reid, C. P. O’Donnell and G. Downey, Trends Food Sci. Tech., 2006, 17, 344. 23 D. I. Ellis, V. L. Brewster, W. B. Dunn, J. W. Allwood, A. P. Golovanov and R.

Goodacre, Chem. Soc. Rev., 2012, 41, 5706.


24 D. Cozzolino, Appl. Spectrosc. Rev., 2012, 47, 518. 25 P. A. Tarantilis, A. Beljebbar, M. Manfait and M. G. Polissiou, Spectrochim. Acta A,

1998, 54, 651. 26 E. Anastasaki, C. Kanakis, C. Pappas, L. Maggi, C. P. del Campo, M. Carmona, G. L.

Alonso and M. G. Polissiou, Eur. Food Res. Technol., 2010, 230, 571. 27 S. A. Ordoudi, M. De Los Mozos Pascual and M. Z. Tsimidou, Food Chem., 2014, 150,

414. 28 A. Zalacain, S. A. Ordoudi, E. M. Díaz-Plaza, M. Carmona, I. Blázquez, M. Z. Tsimidou

and G. L. Alonso, J. Agric. Food Chem., 2005, 53, 9337. 29 M. K. Assimiadis, P. A. Tarantilis and M. G. Polissiou, Appl. Spectrosc., 1998, 52, 519. 30 E. G. Anastasaki, C. D. Kanakis, C. Pappas, L. Maggi, A. Zalacain, M. Carmona, G. L.

Alonso and M. G. Polissiou, J. Agric. Food Chem., 2010, 58, 6011. 31 F. Wittwer and H. Pfander, Helv. Chim Acta, 1975, 58, 1608. 32 G. Speranza and G. Dadà, Gazzetta Chimica Italiana, 1984,114, 189. 33 M. K. Assimiadis, P. A. Tarantilis and M. G. Polissiou, Spectros. Biol. Mol.: Modern

Trends, 1997, 495. 34 S. Pfister, P. Meyer, A. Steck and H. Pfander, J. Agric. Food Chem. 1996, 44, 2612. 35 M. R. Van Calsteren, M. C. Bissonnette, F. Cormier, C. Dufresne, T. Ichi, J. C. Y. Le

Blanc, D. Perreault and I. Roewer, J. Agric. Food Chem., 1997, 45, 1055. 36 T. Radjabian, A. Saboora, H. Naderimanesh and H. Ebrahimzadeh, J. Food Sci,

Technol.- Mysore, 2001, 38, 324. 37 A. Yilmaz, N. T. Nyberg, P. Molgasrd, J. Asili and J. W. Jaroszewski, Metabol., 2010, 6,

511. 38 A. Yilmaz, N. T. Nyberg and J. W. Jaroszewski, Anal. Chem., 2011, 83, 8278.

FOOD NMR OPTIMIZED FOR INDUSTRIAL USE - AN NMR PLATFORM CONCEPT

E. Humpfer, B. Schütz, F. Fang, C. Cannet, M. Mörtter, H. Schäfer, and M. Spraul

Bruker BioSpin GmbH, Rheinstetten, Germany

1 INTRODUCTION

At current state of the art, conventional quality, safety and authenticity control in food is based on targeted strategies, where predefined analysis targets (e.g. chemical compounds, ingredients, chemical parameters) are identified and quantified. Detected deviations of the results and related ratios from established reference ranges, e.g. defined by official regulations, may then support conclusions on quality issues and frauds in problematic samples. While very successful and widely accepted in food analysis, this approach fails if a food matrix is altered (intendedly or unintendedly) in a way that concentrations of compounds which are covered by the food analysis portfolio remain unaffected. The traditional answer to this challenge is to develop respective new diagnostic assays and include them into the portfolio of available analysis parameters. As a consequence in order to keep confidence in food materials and products, an ever increasing effort in testing would be needed resulting in a respective increase of parameter and sample coverage and hence ever increasing resources, time and costs. However more recently, as a promising alternative out of this dilemma, high resolution 1H-NMR has found its way into routine food analysis. It offers several key advantages: NMR in food needs little sample preparation (often just adding a buffer). It is an inherently quantitative method with a large dynamic range (typically 1:100000). NMR data of intact food matrices are extremely information rich such that it needs just one measurement in order to get access to information on a large parameter portfolio (e.g. several hundreds of spectral lines in a juice or wine spectrum). Under screening conditions, i.e. 15 – 20 samples per minute, NMR can be performed under extremely cost efficient conditions considering cost per sample and parameter. Under well-defined and correctly implemented instrumental specifications and SOPs, NMR generates extremely reproducible and fully quantitative data. Such, a spectrum a food matrix can be regarded as an extremely reproducible comprehensive and unique chemical fingerprint. 1H-NMR can be regarded as a primary method for quantification of compounds even under the conditions found in complex spectroscopic fingerprint of a food matrix. All relevant parameters and factors which are necessary to calculate a specific concentration out of the 1H-NMR -spectrum are directly given and accessible by the physics of the NMR-experiment, the chemical information of the molecule investigated and eventually by an internal or external reference [4].


Statistical methods and models for untargeted analysis of spectral fingerprints have been demonstrated to allow for extraction of latent information on e.g. authenticity. For example, the origin of a sample (country or even region) is often an important aspect in authenticity control. Related NMR methods have been reported for e.g. juice [1], for wine [4] and for edible oil [3]. Further parameters addressed by NMR methods and related to authenticity include variety, vintage or the type of production (e.g. usage of oak in wine making) to name a few. Another advantage of the non-targeted analysis is the possibility to detect atypical deviations of any kind, irrespective of whether the deviation is due to a known or even unknown compound (e.g. melamine in milk-power in [3]). For this purpose, typically hundreds or even thousands of 1H-NMR-variables are compared against an existing reference database. The statistical approaches for classification and verification rely on the distinctive fingerprint of the 1H-NMR -profiles of the samples which cannot produced by artificial chemical design. Hence, such a mega-parametric and fully quantitative fingerprinting method is an ideal concept for industrial quality and authenticity control. However, there is no a priori knowledge on the fingerprints of natural samples. Therefore it is important to note that acquisition of reference spectral databases of hundreds if not thousands of authentic samples covering all relevant aspects of a particular food control application is the key requirement for full exploitation of the target free NMR concept. While there is a vast spectrum of publications on the principal possibilities of NMR in food, implementation of related methods in industrial labs has to address certain specific challenges and requirements:

Efficiency: The method needs to run at low cost per sample and parameter. Hence, it must allow for screening of many samples covering automatic determination of many parameters per measurement without the need of an NMR expert. Validity: Any method used must be fully validated in order to be acceptable. Method quality and reliability must be independent of operator. Scalability: A preferred solution should be scalable in terms of number of samples, number of NMR instruments employed and number of types of samples. Sustainability: Solutions must be sustainable in order to avoid that developments for the same problem are repeated again and again. Solutions should be seamlessly transferable from one instrument to another instrument of similar specification or subsequent instrument generation. Completeness: A solution and related data bases need to cover all relevant aspects (e.g. possible origins) the reliability and quality of the offered parameter portfolio relies on.

In order to successfully transfer a published method into a fully automated industrial food control method, it needs substantial expertise on NMR and related data analysis as well as on the particular field of food control. Furthermore, since an industrial lab might not have access to spectroscopic and model data used in publications, it has to create its own reference databases for all types of samples which it wants to investigate and develop the standard operating procedures (SOPs), quantification and statistical methods again by its own. Especially, the collection of thousands of authentic reference samples is a long-term project, given the access to such samples.

In this article we describe a strategy for a platform-concept which will cover the named aspects efficiency, validity, scalability, sustainability, completeness and the possibility to re-use already developed solutions.


2 STANDARDIZED NMR-PLATFORM

The potential of 1H-NMR -technology as fingerprinting-method opens a new era in analytical methodology but at the same time it raises additional requirements. Highly reproducible NMR-fingerprints are enabling the development of statistical models for comparison (e.g. identity analysis), classification (e.g. determination of origin) or verification (e.g. detection of a-typical deviations). The reproducibility of such a fingerprint is directly correlated to strict and fixed standard operating procedures concerning sample collection, storage, preparation and NMR-measurement. Furthermore, the statistical analyses have to be based on authentic reference databases which have to be collected. Therefore access to such authentic reference samples covering the relevant part of the world possibly even over a long time period is mandatory which is hardly achievable for one single company, laboratory or other institution. Instead, the setup of international partnerships or consortia seems important: Several partners, if needed from all over the world, should collect authentic samples from the region they are representing. Depending on the size of such a database, the preparation and measurement of the samples might not be possible by a single NMR lab anymore. At the same time, several partners may want to use the resulting databases and derived solutions in their own lab. Hence, all acquired data need to be fully exchangeable between laboratories, i.e. the spectral fingerprint of a sample needs to unique, absolutely reproducible and completely independent of the measuring laboratory in order to allow for global data pooling. In such a constellation, each partner of such a consortium may contribute a (local) aspect to a global project and gains the complete solution covering all aspects. To fulfil these requirements, all relevant aspects which have influence on the final 1H-NMR-fingerprint needs to be standardized, starting with the SOPs for sample collection and sample preparation. A standardized NMR-platform is mandatory to ensure highest comparability regarding the acquired NMR-spectra. Also taking into account the robotics for automation and validated and automated software solutions for the spectral analysis, a concept for a standardized NMR-environment for food-quality control will contain following layers:

Layer 1 (NMR-platform): Strictly specified NMR-system based on industry standard NMR technology components. Layer 2 (Automation): industry standard automation (e.g. sample-changer, robot for automated sample preparation, automated optimization of acquisition parameters). Layer 3 (SOPs for applications): application specific SOPs and experimental methods. Layer 4 (Data analysis for applications): application specific automated data analysis and reporting procedures.

The combination of Layer 3 and 4 form a complete application (from sample to result). For layer 1, a fixed magnet field strength (e.g. 400 MHz) is mandatory, since 1H-NMR -fingerprints are only completely comparable (including shift and coupling patterns) at the same field strength. The automation layer 2 ensures an easy-to-use system enabling high throughput with reduced number for sources of errors. Barcode-handling is recommended in order to ensure safe identification of the samples without the possibility of mix-ups. Automated sample preparation will outmatch manual sample preparation regarding precision and costs and operator-dependency can be eliminated. The same applies for automated routines for the optimization of the experimental NMR conditions as automated tuning-and-matching, automated shimming and pulse-calibration. Layers 1 and 2 provide a standardized interface for sample-type specific applications of layers 3 and 4.


Figure 1 Layered platform concept for standardized NMR-measurements

Two NMR-systems standardized with respect to layers 1 and 2 will only generate the very same fingerprint, if the prepared NMR-tube is exactly the same and if the NMR SOPs are identical. The standard operation procedures in layer 3 for each application have to ensure that sample handling, preparation, the setup of the NMR-system and the experiments used are robust, reproducible, easy-to-use with reduced risk of errors. As a result of layer 3, the same sample will be represented by the same NMR-fingerprint, independently on the performing laboratory. Finally, software solutions in layer 4 can be developed based on a robust data acquisition. Efficient method developments for quantification and statistical analyses is now possible and resulting software can be re-used in different laboratories. A complete application (e.g. fruit juice analysis) contains SOPs for layer 3 and automated solutions for data analysis in layer 4. The standardization of layers 1 and 2 guarantees that such a complete application can be easily deployed to other labs which operate on systems with equivalent equipment. The solutions based on such a standardized, layered platform-concept will immediately show the following advantages:

Possibility of fast scalability in terms of number of labs, applications and parameters. Platform allows for implementation of coexisting different applications even from different consortia. Possibility for multiple labs to contribute to data bases enabling “complete solutions” regarding coverage of aspects (e.g. databases covering the whole world). Possibility for centralized method development by special expert labs. Possibility to roll-out methods on platforms available in partner labs. Concept of contribution and gain, e.g. contribute to common data bases and share solutions from partner labs. Validated applications in terms of accuracy and precision are also directly valid when distributed to other labs. Solutions can be used even by non-NMR-experts, if automation is available in all four layers (efficiency). Solutions become sustainable in terms of long-term reusability.

Sharing of know-how and focus on dedicated expertise can extremely accelerate the development of such NMR-based methods. There is no need for NMR-experts or expert


teams for the interpretation of spectroscopic data and the development of statistical routines or quantification methods on each site any more. Instead, expertise can be bundled which leads to more efficient and valid method development, since expert knowledge is not needed in each lab using NMR methods on several different fields and topics. The easy plug-in of applications, developed in other labs or institutions, will boost new partnerships, since the technological barrier is not there anymore. Once an application has been developed by an institution or a consortium, it can be rolled-out, distributed or even sold to other laboratories compensating for the initial investment needed to create the application.

3 PROOF OF CONCEPT: STANDARDIZED FOODSCREENER & APPLICATIONS

A platform for 1H-NMR -based food quality and authenticity control, the FoodScreener™-platform, has been introduced by Bruker BioSpin, recently. It is based on an Avance 400 MHz Spectrometer with a 5 mm inverse probe-head which is optimized for proton detection. It is equipped with an automated sample changer, which is able to read barcode-labelled NMR-tubes. Automation is controlled by the Micro-LIMS (Laboratory Information Management System) SampleTrack™. Automated routines for tuning-and-matching, shimming, temperature adjustment and pulse-calibration complete the entire automation process such that there is no need for manual interaction. This system forms the standardized platform according layer 1 (spectrometer) and layer 2 (automation). Each user of the FoodScreener™-platform can now develop sample-type specific applications having the possibility for easily rolling out the solution to other labs. Bruker has demonstrated this principle (from sample to result) for the screening of fruit juices and wines – an application for honey samples is currently in development. The fruit juice screening (named SGF-Profiling™ [1]) started in 2008 and currently contains more than 20.000 database samples, measured on four different NMR-spectrometers. The SOPs in layer 3 define for each type of fruit (e.g. orange, apple) and for each type of product (direct juice, concentrate, puree) a standardized preparation. The acquisition based on robust NMR-experiments including a 1D-noesypresat and a 2D-JRES-experiment takes 15 minutes per sample. Immediately after the measurement, the automated data analysis and reporting software starts the interpretation of the data including the quantification of more than 40 compounds and statistical analyses based on more than 50 models (e.g. determination of origin, estimation of fruit content or non-targeted detection of deviations of any kind). The Wine-Profiling™ by Bruker BioSpin was started in 2011 and up to now more than 12.000 reference samples have been measured on 8 different spectrometers at 4 different laboratories and partners. Preparation is automated including buffer addition and pH-adjustment. Automated NMR-acquisition takes about 20 minutes and uses 8-fold suppression for water and ethanol [5]. The analysis routines on layer 4 are currently able to detect and quantify more than 50 parameters and more than 20 statistical models are used for the determination of grape variety or year of vintage and the non-targeted detection of deviations of any kind. The principle of fingerprinting is shown in figure 2: the very same sample was prepared and measured in three different laboratories. The resulting 1H-NMR-profiles are highly reproducible even for the smallest peaks of the spectrum.


Figure 2 Excerpts of 1H-NMR-fingerprints of one wine sample prepared and measured in three different laboratories using the standardized FoodScreener™-platform . Each single NMR-peak is highly reproducible in all three labs.

The FoodScreener™-platform is installed in several labs using the same applications including automated analysis and reporting. Such, the laboratories can focus on the biological or chemical interpretation of the results which leads to an efficient and cost-effective operation. Since the applications for fruit juice and wine are extensively validated regarding accuracy and precision (validated by spiking, comparison on thousands of reference values from conventional analyses, participation in international ring-tests) at Bruker, the installed applications at partner labs fulfil the same quality standards since the very same SOPs are implemented there.

4 CONCLUSION AND OUTLOOK

In this article we describe a layered platform-concept for NMR-based food quality and authenticity analysis. The standardization of an NMR-platform equipped with automation for barcode-based sample-changing, preparation together with automated preconditioning of the NMR-spectrometer and acquisition provides a well-defined interface for specialized applications for different types of samples. Complete applications consist of robust and easy-to-use SOPs for sample handling, preparation and measurement and automated routines for the data analysis leading to quantifications and statistical results. This enables the distributions of applications to other labs of partners or customers. Solutions become sustainable, scalable and efficient and can be applied at a long-term basis. This is the prerequisite for the creation of consortia in which several partners from the whole world contribute inside a network of standardized NMR-systems to a solution which covers all global aspects.

References

1 E. Humpfer, H. Schaefer, B. Schuetz, M. Moertter, M. Spraul and P. Rinke, Magn.Reson. Food Sci., 2009


2 F. Longobardi, A. Ventrella, C. Napoli, E. Humpfer, B. Schütz, H. Schäfer, M.G. Kontominas and A. Sacco, Food Chem., 2012, 130, 177

3 D. Lachenmeier E. Humpfer, F. Fang, B. Schütz, P. Dvortsak, C. Sproll and M. Spraul, J. Agric. Food Chem., 2009, 57, 7194

4 R. Godelmann, F. Fang, E. Humpfer, B. Schütz, M. Bansbach, H. Schäfer and M. Spraul, J. Agric. Food Chem., 2013, 61, 5610

5 Y. Monakhova, H. Schäfer, E. Humpfer, M. Spraul, T. Kuballa and D.W. Lachenmeier, Magn. Reson. Chem., 2011, 49, 734

A NEW ULTRA RAPID SCREENING METHOD FOR OLIVE OIL HEALTH CLAIM EVALUATION USING SELECTIVE PULSE NMR SPECTROSCOPY

E. Melliou1,2, P. Magiatis1,2and K.B. Killday3

1Laboratory of Pharmacognosy and Natural Products Chemistry, Faculty of Pharmacy, University of Athens, Panepistimiopolis Zografou, 157 71 Athens, Greece, 2Olive Center, University of California, Davis CA 95616 3Bruker BioSpin, Billerica, MA, USA

1 INTRODUCTION

The traditional Mediterranean diet, which is continuously attracting the interest of the scientific community for its health protecting activities, is based on the daily consumption of olive oil as the major source of lipids.1-3 One of the most important class of constituents of olive oil are the secoiridoid polyphenolic derivatives which present an increasing potential for health protection.4,5 The European Union legislation based on the scientific opinion of EFSA6

has permitted specific health claims related to the levels of specific phenolic compounds found in olive oil (5 mg per 20 g dose or 250 mg/Kg). The key compounds that are responsible for the recognized health claim “protection of blood lipids from oxidative stress” are hydroxytyrosol, tyrosol and their derivatives. For this reason the accurate measurement of the levels of those compounds in olive oil is very important. As of today there is no officially adopted method for their measurement because of well known technical difficulties. Hydroxytyrosol and tyrosol are found in olive oil mainly as the following esterified derivatives: oleacein (3,4-DHPEA–EDA), monoaldehydic form of oleuropein aglycon (3,4-DHPEA-EA), oleocanthal (p-HPEA-EDA) and the monoaldehydic form of ligstroside aglycon (p-HPEA-EA) which possess significant biological activities, as previously summarized.7,8 There are several works concerning the chromatographic analysis of those compounds (HPLC-UV or LCMS)9-12 but their accuracy is questionable because as we have recently described7 oleocanthal and oleacein react with methanol and water which are commonly and officially13 used for the extraction of polyphenols and as constituents of the mobile phase during their analysis, leading to the formation of several artifacts and making the analysis very difficult. To overcome these problems that make the chromatographic analysis complicated and questionable we recently developed a simple and rapid method using quantitative NMR (qNMR) including a simple extraction step to increase the concentration of the analytes and reduce the bulk lipid matrix.8 Nuclear Magnetic Resonance (NMR) spectroscopy is well suited for quantitative analyses of complex chemical mixtures. 1D 1H NMR typically provides the highest sensitivity with excellent linear response to component concentrations. Quantitation of key trace analytes in the presence of very strong signals from the bulk matrix can however be problematic or even impossible, depending on concentration, using typical broad band excitation. This is due to dynamic range limitations of the analog to digital converter (ADC), especially on older instruments. These limitations can be overcome by the use of band selective shaped pulses to excite only the region containing the minor analytes while excluding the regions containing


strong matrix signals which would exceed the ADC range. Region selective 1D 1H-NMR methods for quantitation of aldehydes in honey and terpenes in olive oil have been described.14 These utilize double pulsed field gradient spin echo (DPFGSE) sequences with band-selective refocusing pulses. In cases where the selectively excited region contains two or more J-coupled spins, significant anti-phase magnetization can occur, reducing the integrated signal intensities of these relative to uncoupled spins. This coupling evolution can be removed by utilizing the recently reported “perfect echo” (PE) sequence.15 In this framework, we envisaged the utilization of selective excitation with a double pulsed field gradient perfect echo method (SELDPFGPE) to analyze the aldehydes in olive oil without the need for concentration of these analytes. The method was developed targeting all the four major secoiridoid derivatives of hydroxytyrosol and tyrosol. An analysis of the aldehyde region in Sicilian extra virgin olive oils utilizing DPFGSE has been reported,16 although the authors did not identify oleocanthal, oleacein, or the aglycons of oleuropein and ligstroside as the observed components nor were the components quantitated. The developed method in combination with the previously reported qNMR method8 was applied to the study of 100 commercial olive oil samples from all the major brands available in supermarkets in California offering a good estimation of the average levels of the secoiridoid aldehydes that are available to the consumers. The varieties presenting the highest concentrations of the studied compounds were recognized and presented herein.

2 MATERIALS AND METHODS

2.1 Extra Virgin Olive Oil samples

The commercial extra virgin olive oil samples used in the study were obtained from olives (Olea europaea L.). The samples were purchased from major super markets in the San Francisco and Sacramento area in November 2013. The samples were bottled in 2013 and were coming from the 2012-2013 harvest season. 40 samples were produced in California, 26 in Italy, 11 in Greece, 7 in Spain, 1 in orocco, 1 in rgentina, 1 in Chile, 1 in France and 12 were labelled as Mediterranean mixtures.

2.2 Reference Compounds

Oleocanthal, oleacein, oleuropein aglycon and ligstroside aglycon were isolated from an olive oil extract as previously described7,8 and their purity was >98%. Syringaldehyde (98% purity) used as internal standard (IS) was purchased from Sigma–Aldrich (Steinheim, Germany). IS solution for extracted oil was prepared in acetonitrile at a concentration of 0.5 mg/mL and kept in refrigerator. Prior to use the IS solution was left to come to room temperature. All NMR solvents used throughout the experiments were obtained by Sigma–Aldrich.

2.3 NMR Analysis of Olive Oil without Extraction

225 mg olive oil (ca 250 L) were mixed with 500 L CDCl3 containing syringaldehyde as internal standard (50 g/mL) and transferred to a 5mm NMR tube. The DPFGPE sequence was executed utilizing a 2.6 ms 180 degree reburp selective refocusing pulse, affording a 2400 Hz excitation window from 11 to 7 ppm. Data from 16 scans were collected for a total experiment time of 3 min. The spectra were phase corrected automatically using TopSpin software (Bruker). Accurate integration was performed manually for the peaks of interest. The experiments were performed using a Bruker Avance 600 MHz with cryoprobe.


2.4 Calibration Curves and Quantitation

Calibration curves were prepared by addition of known quantities of oleocanthal, oleacein, oleuropein aglycon or ligstroside aglycon to a selected commercial olive oil that was found to be free of all secoiridoid derivatives and following the above described measurement method. The standard compounds were dissolved in CDCl3 containing I.S. and then mixed with the oil. The quantitation was based on the integration ratio between the aldehydic proton signal of syringaldehyde at 9.81 ppm and the aldehydic protons of oleocanthal at 9.23 ppm, oleacein at 9.21 ppm, oleuropein aglycon at 9.50 ppm and ligstroside aglycon at 9.49 ppm.

2.4.1 Standard and spiked solutions. Stock standard solutions of oleocanthal, oleacein, oleuropein aglycon and ligstroside aglycon were prepared in CDCl3 at the 3 mg/mL level and were kept in refrigeration. Prior to use the stock solution were let to come to room temperature. Spiked olive oil samples were prepared to give concentrations of each analyte at 5, 20, 75, 150 and 300 mg/Kg by mixing appropriate volumes of the stock standard solutions with 225 mg of olive oil and CDCl3 containing I.S. The mixture was homogenized using a vortex mixer for 30 sec and then transferred to a 5 mm tube for NMR analysis.

2.5 Method Validation

The method was checked for the linearity, accuracy [evaluated as the relative percentage error % (Er%), defined as (assayed concentration nominal concentration)/(nominal concentration)×100], sensitivity [evaluated as the limits of Detection (LOD) and Quantitation (LOQ)].

2.5.1 Linearity. Spiked olive oil samples were prepared to give concentrations of oleuropein aglycon and ligstroside aglycon at the 10, 20, 40, 80, 160 and 320 mg/Kg and were analyzed for the determination of the linearity. The relationship of the integration ratio of the analytes versus the internal standard and the corresponding concentration of the spiked olive standards was determined by linear regression analysis.

2.5.2 Accuracy. Spiked olive oil samples at three concentration levels of both aglycons, 20, 75 and 150 mg/Kg, were analyzed in order to determine the accuracy of the method.

2.5.3 Limits of detection and quantitation. The LOD and LOQ were determined running six blank samples of olive oil free of secoiridoids and measuring the background response at the chemical shift of each analyte. A signal-to-noise (S/N) ratio of 3:1 and 10:1 were used for the calculation of the LOD and LOQ, respectively.

2.6 Olive Oil Extraction and Sample Preparation and NMR Spectral Analysis of Extracted Oil.

The analysis of extracted oil was performed as previously described.8 Briefly: Olive oil (5.0 g) was mixed with cyclohexane (20 mL) and acetonitrile (25 mL) and the mixture washomogenized using a vortex mixer for 30 sec and centrifuged at 4,000 rpm for 5 min. A part of the acetonitrile phase (25 mL) was collected, mixed with 1.0 mL of a syringaldehyde solution (0.5 mg/mL) in acetonitrile and evaporated under reduced pressure using a rotary evaporator. The residue was dissolved in CDCl3 (750 L) and an accurately measured


volume of the solution (550 L) was transferred to a 5 mm NMR tube.1H-NMR spectra were recorded at 600 MHz (Bruker Avance 600) with cryoprobe. Typically, 50 scans were collected into 32K data points over a spectral width of 0-16 ppm with a relaxation delay of 1 s and an acquisition time of 1.7 s.Prior to Fourier transformation an exponential weighing factor corresponding to a line broadening of 0.3 Hz was applied.The spectra were phased corrected and integrated automatically using TopSpin software (Bruker). Accurate integration was performed manually for the peaks of interest.


3.1 Method Development.

3.1.1 Selection of NMR solvent. The selection of CDCl3 as solvent for NMR analysis of olive oil was based on the observation that it does not react with the analytes and that it presents a well resolved set of peaks corresponding to the aldehydic protons of the studied compounds between 9.1 and 9.8 ppm. It should be emphasized that methanol and water which are commonly11,12 and officialy13 used for the extraction of phenolics from olive oil react immediately with the dialdehydic form of oleocanthal or oleacein leading to the corresponding acetals or hemiacetals.7 It should be emphasized that a large number of compounds identified in HPLC-UV or LCMS chromatograms of olive oil extracts are obviously artifacts produced by that type of reactions. All previous studies using those type of solvents for quantitative analysis should be reconsidered. Other studied deuterated solvents like acetonitrile, acetone or CD2Cl2 gave overlapping signals of the analytes and were considered as not appropriate.

3.1.2 Selection of internal standard. The choice of syringaldehyde as internal standard and of CDCl3 as solvent for the NMR measurement was based on reasons explained previously.7

3.1.3 NMR spectral analysis of target compounds in extra virgin olive oil. The spectrum region between 9.1 and 9.8 ppm in the spiked oil that was used for the method development was clearly resolved making feasible the integration of the corresponding peaks and their comparison with the peak of the internal standard. Oleuropein aglycon (3,4-DHPEA-EA) and ligstroside aglycon (p-HPEA-EA) were quantified by integrating doublets at 9.50 ppm and at 9.49 ppm respectively. Oleocanthal and oleacein were measured at 9.23 ppm and 9.21 ppm respectively.

3.1.4 Development of calibration curves and validation. The calibration curves were constructed by the addition of known quantities in a specifically selected olive oil sample which did not contain any of the analytes. The method was validated for accuracy and sensitivity. Linearity: Good linearity was achieved for all analytes for concentration ranging from 20 to 300 mg/Kg, with satisfactory correlation coefficients, r2 >0.995. Accuracy: The estimated accuracy values with the proposed method are within acceptable levels for the four analytes (Er%<10) and the method could be considered as accurate.Sensitivity: The sensitivity of the method as presented by its limit of detection (LOD) and the limit of quantification (LOQ) were found to be 5 mg/Kg and 20 mg/Kg, respectively, for all compounds. 3.1.5 Overlapping limitations. We observed two types of signal overlap which limit the application of the selective excitation method. In the first case, the peaks of oleuropein and


ligstroside aglycons are overlapped by the signal of conjugated aliphatic aldehydes like hexenal (9.5 ppm doublet). The overlapping problem is more serious in the case of aged oils with increased levels of hexenal which is a derivative of oxidative decomposition of lipids. This type of problem is not significant in the case of fresh oils. The second type of overlapping occurs in the 9.20-9.21 ppm region. This is the region where the peaks of oleocanthal and oleacein are observed as two well separated doublets. When the oil contains only those two dialdehydic compounds there is no problem of overlapping (Figure 1). However, if the oil contains compounds like the oleuropein and ligstroside aglycon dialdehyde forms there is overlap that can be revealed only after comparison with the spectrum of the extracted oil (Figure 2).

Figure 1 Comparison of spectra using selective pulse without extraction (up) and after extraction (down). This case does not present overlapping limitations

Figure 2 Comparison of spectra using selective pulse without extraction (up) and after extraction (down). This case presents two types of overlapping limitations


In most studied oils those compounds are found in small quantities and do not create a significant problem, however there are oils with high concentrations which may lead to a false overestimation of oleacein and oleocanthal levels. The overlapping problem can be overridden by 2D qNMR which is currently under investigation.

3.2 Method Application

The selective pulse method presents important advantages in comparison to alternative chromatographic or spectroscopic methods. The oil sample can be quantitatively analysed without any treatment (e.g extraction, derivatization, separation etc), without the use of standards and without the risk of decomposition or isomerization as happens during chromatographic analysis.8 Moreover it is the fastest available method, with a total needed time of less than 5 min for sample preparation, spectrum acquisition, integration and quantitation. However, as explained above, the method presents some overlapping limitations that in some cases may lead to overestimation of some compounds. For this reason, this method is currently useful mainly as a screening tool. More specifically it is the most rapid and easy way to discriminate the oils that do not fulfill the European Union criteria for polyphenol content and health claims. Even if it does not give accurate results in all oils, it can definitely be used to exclude oils from further evaluation. In this framework, during the current study the olive oil samples were first screened with the selective pulse and those presenting concentration close or higher than 250 mg/Kg were re-examined following the previously validated qNMR method8 after extraction of olive oil to achieve accurate measurement. The samples showing lower content were excluded from any further accurate measurement saving significant amount of time and effort. The combined method with first the selective pulse measurement and second the extraction step, was applied to the study of 100 commercial olive oil samples from all the major brands available in supermarkets in California (San Francisco and Sacramento area). The quantification results for each compound together with data about variety and geographic origin of the highest content oils are provided in Table 1. A wide variation concerning the concentrations of all four secoiridoids was recorded. The concentration of each one ranged from non-detectable to 402 mg/Kg and the sum of the four major secoiridoids (D3) from non-detectable to 1232 mg/Kg. More than 50% of the studied samples failed to offer more than 250 mg/Kg of hydroxytyrosol or tyrosol derivatives and only 22 samples (5 Italy, 1 Spain, 1 Greece, 15 California) showed hydroxytyrosol derivatives (oleacein+oleuropein aglycon)> 250 mg/Kg as required by the EU health claim regulation. This result emphasizes the need for appropriate labeling of olive oils. The present study offers a good estimation of the average levels of the secoiridoid aldehydes that are available to the consumers by commercially available oils. One interesting observation concerning the role of the variety on the chemical profile of the olive oil polyphenols was that some varieties showed consistently increased concentration of specific compounds. More specifically, all oils produced exclusively by California Mission variety or even containing a part coming from Mission variety showed high levels of oleuropein aglycon, which is a compound with promising activity against Alzheimer disease.17 The highest concentration was recorded at 397 mg/Kg coming from a Mission sample from Berkeley Olive grove. The cv. Mission from California seems to be highly interesting since in all studied samples the major secoiridoid was oleuropein aglycon. Moreover, in all Mission samples the concentration of oleocanthal and oleacein was lower than that of oleuropein and ligstroside aglycons confirming our previous observation that there are least two distinct biosynthetic pathways leading to the domination of each group of compounds.


Table 1. List of the 22 out of 100 commercial samples showing the highest content in secoiridoid polyphenols and satisfying the EU regulation for health claim. Oils are listed according their total measured content in secoiridoid polyphenols.

Origin Variety

Ole

urop

ein

agly

con

mg/

Kg

Lig

stro

side

ag

lyco

n m

g/K

g

Ole

ocan

thal

m

g/K

g

Ole

acei

n m

g/K

g

California 70% Mission,Leccino, Frantoio

352.5 155.1 314.5 400.6

Greece/Antiparos Koroneiki 95.2 65.1 347.2 281.4

California Manzanillo, Mission 163.3 65.6 231.9 318.1

California/Oroville Mission 397.2 162.5 69.0 69.0 California/Oroville Mission 355.1 52.0 107.9 107.0 California Not mentioned 285.8 103.2 125.2 176.3 California/Yolo Leccino, Pendolino,

Moraiolo, Frantoio 179.4 74.1 352.4 257.7

California Barouni 197.3 133.4 233.5 233.0

Italy Not mentioned 149.5 139.2 402.3 250.4 California Ogliarola, Barese,

Biancolilla, Cerasuola

197.3 89.3 181.1 193.9

Northern California Mission 275.7 82.8 34.7 49.6 California/Yolo Picual, Ascolano,

Koroneiki, Pendolino, Leccino, Frantoio

87.2 47.8 271.1 236.4

California/Yolo Leccino 83.2 41.7 301.3 235.3

Italy Not mentioned 124.9 174.5 266.7 167.6

California Arbequina 13.2 7.0 218.9 277.8

California Arbequina, arbosana, koroneiki

0 1.8 147.6 284.2


California/Marin county

Frantoio, leccino, pendolino, maurino, coratina, leccio del corno

119.9 69.9 135.5 152.8

Italy Coratina 128.6 100.0 287.8 141.9 Spain Not mentioned 195.3 132.4 98.5 67.2California Arbosana 17.6 8.9 167.6 241.4



In most Italian and Greek oils, the dominating compound was oleocanthal and in almost all cases the ratio between oleocanthal and oleacein (index D2=oleacein/oleocanthal) was lower than 1. Interestingly in most of the samples coming from the Spanish varieties Arbequina and Arbosana the D2 index was higher than 1 and oleacein was the dominating derivative. It confirms our previous observation that D2 seems to be dependent on the olive tree variety probably due to genetic reasons and independent of the olive mill procedure.

A final observation is that the selective pulse NMR method offers a new alternative way for direct observation of aldehydes related to lipid oxidation and identification of rancid oils without sensory evaluation. The peak at 9.5 ppm corresponding to conjugated aliphatic aldehydes, like hexenal, could be used as a marker of rancidity. Indeed, oils showing high levels of that peak were evaluated in all cases as rancid oils by a sensory panel (data not presented).

4 CONCLUSION

Although according to their label all the studied samples were considered as extra virgin olive oils, the observed significant variation of the concentration of the bioactive polyphenolic secoiridoids confirms our previous conclusion8 that there is need of a new type of classification of EVOO especially related to possible health claims of those compounds. D3 index is more accurate and specific than the commonly used total polyphenols index (expressed as gallic acid or caffeic acid equivalents) and could become a new standard for the characterization of olive oil healthfulness.

qNMR is a powerful tool for the measurement of the specific polyphenols required by the EU legislation. The application of the selective pulse for that purpose is a step towards the use of NMR as a high throughput screening method that can be routinely used by the olive oil industry for the discrimination and evaluation of hundreds of samples in a single day. Acknowledgments. The authors would like to thank Dan Flynn, Executive Director of UC Davis Olive Center for supporting of this study. We also thank Berkeley Olive Grove for providing a collection of Mission samples; and J. Dallas for technical assistance in the UC Davis NMR facility.

References

1 F. Visioli and E. Bernardini, Curr. Pharm. Design 2011, 17, 786. 2 F. Pérez-Jiménez, J. Ruano, P. Perez-Martinez, F. Lopez-Segura and J. Lopez-Miranda,

Mol. Nutr. Food Res. 2007, 51, 1199. 3 E.N. Frankel, J Agric Food Chem. 2011, 59, 785. 4 R.S.J. Keast, Q. Han, A.B. III Smith, G.K. Beauchamp, P. Breslinand J. Lin, EP2583676,

2013. 5 A.H. Abuznait, H. Qosa, B.A Busnena, K.A. El Sayed and A. Kaddoumi, ACS

ChemNeurosci. 2013, 19, 973. 6 EFSA journal 2011, 9, 2033 7 E. Karkoula, A. Skantzari, E. Melliou and P. Magiatis, J Agric. Food Chem. 2012, 60,

11696. 8 E. Karkoula, A. Skantzari, E. Melliou and P. Magiatis, J. Agric. Food Chem. 2014, 62,

600. 9 A. Bendini, L. Cerretani, A. Carrasco-Pancorbo, A.M. Gómez-Caravaca, A. Segura-

Carretero, A. Fernández-Gutiérrez, and G. Lercker, Molecules 2007, 12, 1679.


10 G. K. Beauchamp, R. S. J. Keast, D. Morel, J. Lin, J. Pika, Q. Han, C. Lee, A.B. Smithand P. A. S. Breslin, Nature 2005, 437, 45.

11 J. Impellizzeri and J. Lin, J. Agric. Food Chem. 2006, 54, 3204. 12 P. Kanakis, A. Termentzi, T. Michel, E. Gikas, M. Halabalaki and A.L. Skaltsounis,

Planta Med. 2013, 79, 1576. 13 International Olive council testing methods. Determination of biophenols in olive oil by

HPLC. COI/T.20/Doc No 29, November 2009 14 F. Rastrelli, E. Schievano, A. Bagno and S. Mammi, Magn. Reson. Chem., 2009, 47, 868. 15 R.W. Adams, C.M. Holroyd, J.A. Aguilar, M. Nilsson and G.A. Morris, Chem. Commun.

2013, 358. 16 A. Rotondo, A. Salvo, D. Giuffrida, G. Dugo and E. Rotondo, Atti Accad. Pelorit.

Pericol. Cl. Sci. Fis. Mat. Nat. 2011, 89, C1A8901002 17 C. Grossi, S. Rigacci, S. Ambrosini, T. Ed Dami and I. Luccarini, PLoS ONE 2013, 8(8):

e71702. doi:10.1371/journal.pone.0071702

PROFILE OF THE POSITIONAL DISTRIBUTION OF FATTY ACIDS IN THE TRIACYLGLYCEROLS AS AN INDEX OF QUALITY FOR PALM OIL (OR ANY OIL OR FAT)

Soon Ng

Department of Chemistry, University of Malaya, 50603 Kuala Lumpur, Malaysia

1 INTRODUCTION

There is a need for a modern method to characterize edible oils and fats. The traditional method is to measure the iodine value (IV), which is an enometric constant that specifies the amount of molecular iodine absorbed by 100 g of the oil/fat. This parameter indicates the degree of unsaturation of the total amount of acylglycerol molecules (mono-, di- and tri-acylglycerols) and free fatty acids present in the oil/fat. Its determination is according to an official method of the American Oil Chemists’ Society. 1 The IV is conventionally used as a quality parameter in the palm oil industry. As an index for characterization the IV has obvious shortcomings or inadequacies, including: (1) a polyunsaturated fatty acid with two or more double bonds contributing proportionately more than a monounsaturated, resulting in an inflated value which does not reflect the number of fatty acid chains; (2) no indication of the presence of saturated fatty acids; (3) not a true measure of the contribution from triacylglycerols if mono- and di-acylglycerols and free fatty acids are present. We have shown that the 13C NMR spectrum of the carbonyl carbons of triacylglycerol molecules (TAG) in palm oil in the solution state clearly depicts the positional distribution of the fatty acids (FA). 2,3 The carbonyl carbons of the acyl groups at the sn-1,3 positions are equivalent on the NMR timescale and appear as a group of distinct peaks centered at 173.04 ppm in CDCl3 solution for the saturated, oleic and linoleic acyl groups (in decreasing order of the values), while those of the acyl groups at the sn-2 position appear in the same sequence and centered at 172.65 ppm. The total integrated intensity of a peak in the group for a given glycerol position is proportional to the concentration of the acyl group for that glycerol position. From the peak areas the composition of the FA at each of the two glycerol positions is calculated. From these two sets of data the overall FA composition is calculated. The three sets of data constitute a complete positional distribution profile of the FA in the oil/fat. Palm oil is a mixture of TAG, of which 19 have been detected in a high temperature gas chromatogram, 4 as shown in Table 1. It is noted that NMR spectroscopy, because of its long timescale, cannot distinguish between the various TAG molecules. In addition, the 13C NMR spectrum cannot distinguish between the saturated FA of similar chain length, such as, palmitic and stearic, so that the spectrum of the carbonyl carbons show only one peak for the saturated FA in palm oil (or any oil or fat) for each glycerol position. The data derived from


the NMR spectrum for a given FA at each glycerol position is therefore the weighted average of all the TAG in the oil/fat. In processing crude palm oil (CPO) for commercial applications, it is refined, bleached and deodorized (RBD) and fractionated to obtain a liquid fraction called RBD palm olein and a solid fraction called RBD palm stearin. These palm products, as expected, differ much in the composition of the TAG. The TAG molecule is identified by an acronym which consists of the first letters of the three FA, given in the order of the glycerol positions, as depicted in the schematic structure for acronym POL (Figure 1). The typical composition of TAG in palm products is shown in Table 1. 4 It is noted that POP and POO and similar TAG: POS, MOP, SOO and MOO, constitute 65% of CPO. Another 27% in CPO comes from six TAG: PLP, POL, PPP and OOO. Variations in the composition of TAG determine the properties of the palm products and hence the quality of the oil/fat.

Figure 1 Schematic structure of the TAG POL

Table 1 Composition of TAG in palm products No. TAG Crude Palm Oil Palm Olein Palm Stearin (1) POP 32.51 29.78 40.9(2) POO 22.64 27.65 5.99(3) PLP 8.91 10.03 5.63(4) POL 8.78 10.71 1.95(5) PPP 6.37 0.20 26.34(6) POS 5.38 5.20 5.85(7) OOO 2.97 3.56 0.54(8) SOO 2.11 2.60 0.49(9) PLS 1.79 2.22 0.86(10) MOP 1.66 1.71 1.51(11) PLL 1.51 2.00 0.36(12) PSP 1.11 4.74(13) MOO 0.88 0.61 0.53(14) OOL 0.82 1.08 0.10(15) LOL 0.74 1.30 0.23(16) MPP 0.57 0.06 2.18(17) SOS 0.54 0.55 0.47(18) MLP 0.39 0.49 0.13(19) PSS 0.22 0.55

Legend: P Palmitin; O Oleic; L Linoleic; S Stearic; M Myristic


The palm olein, palm stearin and palm superolein samples, together with the respective IV data, were provided by the palm oil company, Intercontinental Specialty Fats Sendirian


Berhad, in Port Klang, Selangor, Malaysia. The CPO samples, together with the IV data, were provided by the oil palm plantation, P.T. Musim Mas, in Medan, Sumatra, Indonesia. The IV data were obtained in accordance with official method: AOCS Cd 1d-92. 1 The oil samples were dissolved in CDCl3 (concentration 1:3 v/v) in 5 mm NMR tubes, dagassed and sealed under vacuum. The 13C NMR data were obtained on the JEOL GSX270 spectrometer, using the method reported previously.3 The total intensity of the peaks in the NMR spectra was measured precisely by a curve fitting procedure which was provided in the data processing software of the spectrometer.


Figure 2 shows the typical 13C NMR spectrum of the carbonyl carbons in a sample of CPO in dilute CDCl3 solution. Table 2 shows the positional distribution profile for five CPO samples together with the respective IV data. The variation in the IV is small, but there is significant variations in the positional distribution of the FA. The saturated FA is heavily concentrated in the 1,3-positions, as is expected from the high concentrations of POP, PLP, PPP, POS and similar TAG. The 2-position is nearly 85% unsaturated, as is expected from 15 TAG having oleic or linoleic FA in this position. In all these CPO samples the saturated and oleic FA do not vary much in the overall composition, but the variation of the linoleic FA (6.8 – 11.1%) is significant in its contribution to the IV and clearly plays a major role in causing the small variation in the IV. Sample 4 has the highest IV and on that basis it would be taken to be of higher quality oil, but the distribution profile shows that the higher IV is no doubt the result of the higher concentration of linoleic FA. Comparing samples 4 and 5, on account of the distribution at the 2-position in which sample 5 has higher oleic and lower linoleic FA concentrations, sample 5 can be considered to be of higher quality than sample 4, in spite of the difference in IV. Fig. 3 shows the 13C NMR spectrum of the carbonyl carbons in a sample (6) of RBD palm olein which has IV 57.3. Fig. 4 shows the 13C NMR spectrum of the carbonyl carbons in a sample (7) of RBD palm stearin with IV 34.0. Table 3 shows the distribution profile of these corresponding palm oil fractions. As expected, the palm olein has less saturated but more unsaturated FA than the corresponding palm stearin, hence the higher IV for the palm olein which also has the higher concentration of linoleic FA in the overall composition. It is noted that in both the palm products, the 2-position has more unsaturated FA than the 1,3-positions, and the ratio is higher in the case of the palm stearin. In palm olein the saturated FA is heavily concentrated in the 1,3-positions while the 2-position is 91% unsaturated. In the palm stearin the abundant saturated FA is also concentrated in the 1,3-positions, but the 2-position has nearly five times more saturated FA than in the case of the palm olein. This situation is the result of the refining process which has fractionated a big portion of POP, POS, PPP and PSP to the palm stearin fraction, while the majority of POO, PLP, POL, OOO, SOO, PLS and PLL remains in the palm olein fraction. The overall composition data show that in the palm olein the ratio of saturated FA to unsaturated FA is 47 to 53, so that it can be said that the RBD palm olein is well balanced in saturated and unsaturated FA. It is also noted that in the palm stearin the 2-position has more unsaturation than the dismally low IV would imply. The overall composition data show that the low IV is due in part to the lower concentration of the linoleic FA. This discussion emphasizes that the positional distribution profile contains information that can be useful for specific applications of the palm products.


Figure 2 13C NMR spectrum (67.94 Mhz) of the carbonyl carbons of crude palm oil In CDCl3solution. The peaks centered at 173.04 ppm pertain to the FA esterified at the 1,3 positions and are assigned, in order of decreasing chemical shift (i.e. left to right), to saturated, oleic and linoleic FA. The peaks centered at 172.64 ppm pertain to the FA esterified at the 2-position and are assigned in the same order of chemical shift as for those at the 1,3-positions.

Table 2 The positional distribution of fatty acids in crude palm oil (CPO) for which the iodine values are known.

sn-1,3 positions

%

sn-2position

%

Overall composition

%(1) CPO (IV 52.7)Sat’d FA 70.8 15.5 52.4 Oleic 24.5 64.7 37.9 Linoleic 4.7 19.8 9.7 (2) CPO (IV 52.7) Sat’d FA 70.7 15.3 52.2 Oleic 26.4 68.9 40.6 Linoleic 2.9 15.8 7.2 (3) CPO (IV 52.5) Sat’d FA 71.7 17.5 53.6 Oleic 24.3 63.9 37.6 Linoleic 3.9 18.6 8.8 (4) CPO (IV 53.6) Sat’d FA 67.6 14.6 49.9 Oleic 27.0 62.9 39.0 Linoleic 5.4 22.5 11.1 (5) CPO (IV 52.4) Sat’d FA 71.6 15.9 53.0 Oleic 26.0 68.5 40.2 Linoleic 2.4 15.6 6.8


Figure 3 13C NMR spectrum (67.94 Mhz) of the carbonyl carbons of palm olein (IV 57.3) in CDCl3 solution. The spectral peaks are assigned exactly as described in the caption for Figure 2.

Table 3 The positional distribution of fatty acids in palm olein and the corresponding palm stearin which are depicted in Figures 3 and 4, and in an olive oil.

sn-1,3positions

%

sn-2position

%

Overall composition

%(6) RBD Palm Olein (IV 57.3)Sat’d FA 66.0 8.6 46.9Oleic 26.2 68.2 40.2Linoleic 7.8 23.2 12.9(7) RBD Palm Stearin (IV 34.0) Sat’d FA 81.3 39.9 67.5 Oleic 13.8 44.8 24.1 Linoleic 4.9 15.3 8.4 (8) Olive Oil Sat’d FA 25.0 1.6 17.2 Oleic 61.9 84.1 69.3 Linoleic 8.2 14.3 10.2 cis-Vaccenic 4.9 3.3

Table 3 also shows the distribution profile of a commercial (BARTOLLI) sample of olive oil, which normally has IV in the range 80 – 88. It is noted that the olive oil is 98% unsaturated at the 2-position with a relatively low contribution (14%) from the linoleic FA, and is 75% unsaturated at the 1,3-positions. This distribution profile for olive oil can be taken to be the bench-mark for assessment of the quality of an oil. It is also noted that the palm olein sample is 91% unsaturated at the 2-position where the linoleic FA contribution is relatively high (23%). There is an obvious difference in the oil quality between the olive oil and the RBD palm olein in terms of the positional distribution of the FA. To narrow this difference, it is desirable for the palm olein to have more oleic and less linoleic FA at the 2-position. This can be achieved if the RBD palm olein undergoes subsequent fractionation operations to remove most of the TAG such as PLP, PLS and PLL, while retaining the TAG such as POO, POL, OOO and SOO in the “enriched” RBD palm olein (or superolein).


Figure 4 13C NMR spectrum (67.94 Mhz) of the carbonyl carbons of palm stearin (IV 34.0) in CDCl3 solution. The spectral peaks are assigned exactly as described in the caption for Figure 2. This sample of palm stearin and the palm olein sample depicted in Figure 3 were fractionated from the same CPO.

4 CONCLUSIONS

The positional distribution profile of fatty acids for an oil/fat permits an assessment of the relative importance of the composition of the fatty acids at the sn-1,3 and sn-2 positions and the overall compositon in relation to the properties of the oil/fat. It provides useful information for food applications and the nutritional benefits, namely: (1) the nature of the fatty acids, (2) the chain length, (3) the glycerol position, for which the sn-2 position has special significance. 5 Hence the positional distribution profile of the fatty acids in the triacylglycerols can be an useful index of quality for the oil/fat.

Acknowledgement

This research work was supported by the University of Malaya (research grant A-21003-DA680). References

1 D. Firestone, Official Methods and Recommended Practices of the AOCS, American Oil Chemists’ Society, Champaign, IL, 1998.

2 S. Ng, J. Chem. Soc. Chem. Commum., 1983, 179 – 180. 3 S. Ng, Lipids, 1985, 20, 778 – 782. 4 Data provided by Perkin Elmer (Malaysia) Sdn. Bhd., Petaling Jaya, Selangor, Malaysia,

2009. 5 T. Karupaiah, K. Sundram, Nutrition & Metabolism, 2007, 4:16.

On-line Non-invasive NMR

1H-NMR RELAXOMETRY AND IMAGING TO ASSESS FAT CONTENT ON INTACT PORK LOINS

V. Bortolotti1,2, P. Fantazzini3, C. Schivazappa4, M. Vannini1,2, E. M. Vasini1, R. Virgili4

1DICAM, University of Bologna, Bologna, Italy 2Interdepartmental Research Center “Energy and Environment”, Rimini, Italy 3Department of Physics and Astronomy, University of Bologna, Bologna, Italy 4Stazione Sperimentale per l’Industria delle Conserve Alimentari (SSICA), Parma, Italy

1 INTRODUCTION

In fresh pork loin, the information concerning fat content is very important, both from the consumer as well as from the nutritional point of view. The fat content affects flavour, juiciness, and tenderness1-3: knowing the fat content allows industry to classify the fresh loins accordingly. Fat content in loins is a factor related to moisture variability; therefore on-line, non-destructive technologies that could be used in an industrial environment to predict fat-to-moisture ratio are of special interest for companies from the pork processing sector. The time-domain nuclear magnetic resonance (1H TD-NMR)4, and in particular Relaxometry and Imaging (MRI), has been introduced as a promising alternative to traditional food characterization method due to its rapidity, simplicity, and potential for on-line non-destructive measurements5-14. As an example, by MRI it is possible to predict non-destructively the salt-to-moisture ratio and monitor ham curing9-12. Compared to high resolution NMR, TD-NMR experiments can be obtained by means of permanent magnet technology, which significantly reduces overall system and running costs8.In this paper, original and innovative procedures have been introduced by using NMR Relaxometry and Imaging of 1H nuclei to estimate the fat content in intact pork loins by exploiting the difference between fat and water longitudinal relaxation time (T1) distributions. It is known13,14 that water and fat signals in tissues have well distinguishable values of T1. On this characteristic is based the well-known Short Time Inversion Recovery (STIR) sequence, commonly used to suppress the signal of fat in MRI. A characteristic of our approach is to regard the distribution of the values of T1 as quasi-continuous distributions instead of discrete distributions. The fat-to-moisture ratios of the loins were determined by NMR Relaxometry through the assumption that the signal of fat 1Hnuclei can be distinguished from that of the moisture through the choice of a proper cut-off on the quasi-continuous T1 distributions. The ratios obtained by NMR were then compared with the corresponding ratios obtained by chemical analysis. On a subgroup of samples the fat-to-moisture ratio was determined also by quantitative analysis of images. Innovative Parametrically Enabled Relaxation Filters with Double and multiple Inversion (PERFIDI) sequences that implement band-pass filters13,15,16 were also used.



2.1 Samples

A total of 30 fresh pork loins (muscle longissimus dorsi), coming from 30 carcasses of Italian heavy pigs, were collected during 5 sampling sessions (6 loins per session) from the same commercial slaughterhouse. At 24 h post mortem, a 2.5 cm thick chop near the seventh rib was removed from the whole boneless loin for NMR analyses. Furthermore, two samples (1.5 cm thick) were excised to carry out chemical analyses. The NMR specimen was wrapped with a film that does not give NMR signal. All the samples were stored at 4 °C until NMR measurement and left at room temperature for 2 h to equilibrate the temperature. The NMR measurements of the 6 loins of the same sampling session required a few days (no more than 4 days from the slaughter). For comparison, also 19 loins, immediately frozen (-18 °C) after the sampling (24 h post mortem) and brought to laboratory temperature before NMR measurements, were analyzed with the same procedure used for the loins stored at 4 °C.

2.2 Chemical Analyses

The two specimens taken from the same loin were minced together. The sample was used to estimate fat and moisture content according to AOAC 950.46, and 960.39 official methods respectively17 The results of both analyses were expressed as grams per 100 grams of wet sample. Percent fat-to-moisture ratio, reported as (F/W)CH (%), calculated for the 30 loins, ranged between 2.1% and 12.0%.

2.3 1H TD-NMR Relaxometry analysis

NMR Relaxometry analyses were performed by a home prototype relaxometer, assembled in the LAGIRN laboratories (DICAM department of the University of Bologna) by using the permanent magnet (B0 = 0.2 T) of an ARTOSCAN tomograph (ESAOTE SpA, Genova, Italy), equipped with a portable NMR console and a full size coil up to 12 cm in diameter manufactured by Stelar s.r.l. (Mede, Pavia, Italy). The Inversion Recovery (IR) sequence was used to acquire the experimental relaxation curve of the longitudinal magnetization component with the following parameters: recycle delay 2 s, 64 inversion times selected in logarithmic scale in order to span all the relaxation time range, 512 points acquired on the Free Induction decays, 4 scans to obtain an adequate signal to noise ratio, 90 pulse duration approximately 60 μs. T1 quasi-continuous distributions were obtained by inversion of the experimental multi-exponential relaxation curves by means of UpenWin Software18 applying the algorithm UPen19,20. Starting from the observation that T1 distributions contain two main features (two peaks or a peak and a tail), at long and short relaxation times, which are respectively assigned to water and fat components13,14, each T1 distributions was divided in two parts by the choice of a cut-off. The ratio of the areas under the two features of the T1distribution separated by the cut-off was defined as the IR-NMR fat-to-moisture ratio, reported as (F/W)IR-NMR (%). The area below the distribution on a given T1 range is proportional to the 1H NMR signal, proportional to the number of 1H with T1 in that range.

2.4 MRI analysis

MRI images were acquired by the tomography ARTOSCAN, the standard STIR sequence was used with the following parameters: TR = 2 s, inversion time = 50 ms, TE = 18 ms, slice thickness 3 mm, 4 scans. Each STIR acquisition took about 30 minutes.

On-line Non-invasive NMR 103

Also a preliminary analysis was performed by a sequence based on PERFIDI filters13,15,16.PERFIDI is a family of filters that implements band-pass filters obtained by a linear combination of standard acquisition sequences each one having a preamble of inversion pulses. These filters allow one to strongly attenuate the signal in a selected range of T1 values, while the remaining signal is affected by a computable attenuation. The simplest PERFIDI filter is obtained by subtracting the signal acquired by two elementary PERFIDI blocks characterized by a suitable choice of two different delays. Figure 1 shows the result of the IR PERFIDI sequence, in which two standard IR sequences with a preamble of two inversion pulses are subtracted. The filter suppresses completely the signal at shorter times in the T1distribution of a loin. Also images can be filtered by PERFIDI and a first preliminary example will be shown. To evaluate fat-to-moisture ratios from the images (F/W)MRI the home-made ARTS software was used to count the pixels assigned to fat and to moisture by a proper image segmentation algorithm. The images were analysed to obtain the (F/W)MRI ratio for each loin as the average of 3 sections.

Figure 1 Application of IR PERFIDI filter to loin sample. Blue line is the T1 distribution of a loin sample obtained by IR data, the green and the red lines are the results of simulation and experimental application of a IR PERFIDI filter, respectively, with the following parameters. First sequence: d1= 5 TR, d2= 5 ms; second sequence: D1= 100 ms, d2 = 5 ms, where d1, and D1 are the delays between the first two inversion pulses and d2 is the delay between the second inversion pulse and the beginning of the standard acquisition sequence. The filter largely suppress the shorter times signal, especially for the experimental data.

2.5 Correlation between NMR and chemical data

The fat-to-moisture ratio obtained by NMR (F/W)NMR by both Relaxometry (F/W)IR-NMR and MRI (F/W)MRI should be proportional to (F/W)CH through a proportionality constant that should depend on the number of 1H nuclei per mole of water and fat. A linear relationship is then expected between (F/W)CH and (F/W)NMR. The hypothesis was checked by best fit of chemical and NMR data to Eq. [1], where k is the slope, that should vary with fat composition, and off represents an offset:

CH NMR

F FkW W

+ off [1]



3.1 1H TD-NMR Relaxometry

T1 distributions of 1H obtained by IR measurements of the 30 loins showed the same main features: a small peak or tail in the range 10-150 ms, while the most of the signal is in the peak centred at about 300 ms (Figure 2 and Figure 3).

Figure 2 T1 distribution of fresh pork loins. A magnification of the signals at low T1 is included.

Figure 3 Example of T1 distribution of fresh pork loin with low, intermediate, and high fat content.

The results of (F/W)IR-NMR are based on the assumption that the peak at long T1 is the signal of water, while the signal at short T1 is due to fat, in such a way that by choosing a proper cut-off on the distribution, to divide the two features, the ratio between the areas below the two peaks (the peak or tail at shorter times divided by the peak at longer times) will give the ratio (F/W)IR-NMR of the sample. Figure 4 shows the scatterplot of (F/W)IR-NMR (%) against (F/W)CH (%) obtained by a proper cut-off, chosen individually for each sample.


Figure 4 Scatterplot of NMR Relaxometry data (by IR sequences) versus chemical data.

A positive relationship was found between the two data set. It is worth to note that the best fit straight line does not go to zero for zero (F/W)CH (%). Probably fat by ether extraction does not include those lipids (i.e. phospholipids, lipoproteins) that, on the contrary, give NMR signal. Though encouraging and useful for a rough sub-grouping of loins according to fat content (the level of significance of the regression is high: p < 0.001), the data show a large dispersion, that could be due to the loss of water from samples in the time elapsed between chemical and NMR measurements (moreover samples for chemical analyses were close, but not the same used for NMR analyses).In order to reduce this source of data dispersion, we applied the method also to a set of loins frozen at 24h post mortem and brought to environment temperature only before NMR measurement. Not only the regression is very good in this case, but also there is no difference between individual or fixed cut-off (set at 150 ms for all the distributions), as shown in Figure 5, where R2 is 0.66 and 0.78 for data evaluated by an individual and a fixed cut-off, respectively. A fixed cut-off, valid for all the samples, would greatly simplify the possible implementation of the protocol in the industry.

Figure 5 Scatterplot of NMR Relaxometry data (by IR sequences) versus chemical data for a set pork loins frozen immediately after sampling in order to reduce the water loss.


3.2 MRI sequences

Preliminary MRI images were obtained on 10 fresh loins (two samples from each sampling session) with the standard STIR sequence, frequently used for fat suppression purpose. An example is reported in Figure 6. The scatterplot of (F/W)MRI against the corresponding values obtained by Relaxometry (F/W)IR-NMR is reported in Figure 7.

Figure 6 STIR image with signal of fat suppressed.

Figure 7 Scatterplot of the ratio fat-to-moisture obtained by MRI (by STIR sequences) versus the corresponding values by Relaxometry.

The positive association between MRI and Relaxometry is preserved, but the regression coefficient is low (R2 = 0.47). This result could be occurred because theoretically fat signal should be black (zero signal), but as expected, the STIR sequence does not suppress all fat signal equally. Indeed, for its nature, STIR suppresses the signal with a specific T1, but this is not correct because, as we have shown, the fat has a distribution of T1 values, not a single value. Thanks to this consideration, the use of a more appropriate sequence, which acts as a filter on the signal to be acquired, could lead to a better determination of the F/W ratio. For that reason a low-pass PERFIDI filter sequence, able to suppress the fat signal in all its T1range should give better results. Figure 8 compares a STIR images with a preliminary application of PERFIDI filter on the same section of a loin.


PERFIDI filters, using a preamble of inversion pulses before the signal acquisition with standard sequences, allow to strongly attenuate the signal in selected range of T1 values, while the remaining signal is affected by a computable attenuation. This allows obtaining images with higher signal-to-noise ratio than the standard STIR sequence.

Figure 8 a) STIR image and b) low-pass PERFIDI filter of a section of the loin c)

4 CONCLUSIONS

The analysis of the quasi-continuous distributions of T1 in fresh intact loins allowed us to obtain a significant relationship between F/W ratios determined from Relaxometry and chemical analysis. The causes of data variability may be the sampling way and the loss of water occurring before NMR analysis. Therefore, the regression between NMR and chemical data should improve substantially if the analysis was done at the slaughterhouse. The results of Relaxometry and MRI suggest the application of Imaging sequences that, differently from STIR images, based on the assumption of a single value of T1 for fat, are able to suppress signals on a given interval of T1. For this reason the innovative PERFIDI sequence that implements band-pass filters before signal acquisition was used. In summary, both NMR Relaxometry and Imaging can estimate the fat-to-moisture ratios in loins, in a non-destructive way: Relaxometry gives a global information on the loin, while MRI can also furnishes a spatial distribution of fat. After a calibration, it should be possible to get fat-to-moisture ratios on intact meat samples by means of an NMR apparatus directly installed at the slaughterhouse. Further improvement is expected from the application of PERFIDI filters and also by improving the calibration comparing NMR and Chemical Analysis on animals belonging to the same group. NMR methods could be applied by means of dedicated apparatus to be used directly by the industry at a lower cost, of course lower for Relaxometry than for Imaging. It is worthwhile to note that from laboratory bench top devices to an on-line configuration there is still a long way to go, but these preliminary results give evidence of the industrial prospective, especially knowing the relative low costs of installation and maintenance of low field NMR.

Acknowledgments

This research has received funding from the European Research Council (Project FP6-036245-2, Q-Porkchains: Improving the quality of pork and pork products for the consumers). The authors wish also to thank ITALCARNI, Carpi (MO, Italy).

References 1 X. Fernandez, G. Monin, A. Talmant, J. Mourot and B. Lebret, Meat Science, 1999, 53,

59.2 J.D. Wood, G.R. Nute, R.I. Richardson, F.M. Whittington, O. Southwood, G. Plastow, R.

Mansbridge, N. da Costa, K.C. Chang, Meat Science, 2004, 67, 651.


3 S.M. Lonergan, K.J. Stalder, E. Huff-Lonergan, T.J. Knight, R.N. Goodwin, K.J. Prusa, D.C Beitz, Journal of Animal Science, 2007, 85, 1074.

4 R. Kimmich, NMR Tomography, Diffusometry, Relaxometry, Springer – Verlag Berlin, Heidelberg, 1997.

5 F.M.V Pereira, S. Bertelli Pflanzer, T. Gomig, C. Lugnani Gomes, P. E. de Felício, L.A. Colnago, Talanta, 2013, 108, 88.

6 J.-L. Damez, S. Clerjon, Meat Science, 2013, 95(4), 879. 7 H. Todt, G. Guthause, W. Burk, D. Schmalbein, A. Kamlowski, Food Chemistry, 2006,

96, 436.8 P. M. Santos, C. C. Corrêa, L. A. Forato, R. R. Tullio, G.M. Cruz, L.A. Colnago, Food

Control, 2014, 38, 204. 9 P. Fantazzini, Magn. Reson. Imaging, 2005, 23, 125. 10 P. Fantazzini, V. Bortolotti, C. Garavaglia, M. Gombia, P. Schembri, R. Virgili, C. Soresi

Bordini, Industria Conserve, 2004, 79, 289.11 P. Fantazzini, V. Bortolotti, C. Garavaglia, M. Gombia, S. Riccardi, P. Schembri, R.

Virgili, C. Soresi Bordini, Magn. Reson. Imaging, 2005, 23, 359. 12 P. Fantazzini, M. Gombia, P. Schembri, N. Simoncini, R. Virgili, Meat Science, 2009, 82,

219. 13 V. Bortolotti, P.Fantazzini, M. Gombia, D.Greco, G. Rinaldin, S.Sykora, Journal of

Magnetic Resonance, 2010, 206, 219. 14 H-P. Muller, F. Raudies, A. Unrath, H. Neumann, A.C. Ludolph,J. Kassubek, NMR

Biomedicine, 2011, 24: 17. 15 S. Sykora, P. Fantazzini, 2005, Italian Patent BO2005A000445. 16 S. Sykora, V. Bortolotti, P. Fantazzini, Magn. Reson. Imaging, 2007, 25, 529.17 AOAC. (2002). Official methods of analysis (17th ed.) Association of Official Analytical

Chemists, Arlington, Virginia (USA). Official Method 950.46 for moisture in meat; Official Method 960.39 for ether extract in meat.

18 V. Bortolotti, R. J. S. Brown, P. Fantazzini, UpenWin: a software to invert multi-exponential decay data, [email protected], http://www.unibo.it/PortaleEn/Research/Services+for+companies/UpenWin.htm.

19 G.C. Borgia, R. J. S. Brown, P. Fantazzini, Magn. Reson. Imaging, 2001, 19, 473. 20 P. Fantazzini, R.J.S. Brown, Concepts Magn. Reson. 2005, 27 A, 122.

Multiscale Definition of Food

19F LABELED POLYION MICELLES AS DIFFUSIONAL NANOPROBES

Daan W. de Kort1,5, Freek J.M. Hoeben2,5, Henk M. Janssen2,5, N. Bourouina3,5, J. Mieke Kleijn3,5, John P.M. van Duynhoven1,4,5,* and Henk Van As1,5

1Laboratory of Biophysics, Wageningen University, Dreijenlaan 3, Wageningen, NL 2SyMO-Chem BV, Den Dolech 2, Eindhoven, NL 3Laboratory of Physical Chemistry and Colloid Science, Wageningen University, Dreijenplein 6, Wageningen, NL 4Unilever R&D, Olivier van Noortlaan 120, Vlaardingen, NL 5TI-COAST, Science Park 904, Amsterdam, NL

1 INTRODUCTION

In modern food manufacturing, there is a strong drive to find alternative formulations in which ingredients are replaced by alternatives derived from a sustainable source. It is, however, not a trivial challenge to obtain a product with comparable performance, mainly due to poorly understood structure-function relationships. In order to better understand these relationships, new measurement methods are required to quantify structural properties. Particularly at the sub-micron level, even powerful (electron) microscopy does not allow for reliable quantification of structural features. At this length scale, many details fall below the resolution limit, and the narrow field of view raises questions of representativeness. Furthermore, microscopy methods are invasive and require careful image analysis in order to quantify structural features. For the sub-micron structural characterization of biopolymer hydrogels –a model system for structured foods– quantitative nanoprobe diffusometry is emerging as a powerful method to complement knowledge obtained by microscopy. Physical models are available to describe hindered diffusion of nanoparticles in polymer gels1 and solutions2 in terms of structural length scales and dynamics. These physical models allow quantitative network descriptors to be derived from experimental nanoparticle diffusion data. Quantitative nanoprobe diffusometry has been demonstrated in various biopolymer model systems, including alginate3, kappa-carrageenan4, casein5,6, gelatin7, whey8 and collagen9 gels. Collective diffusive properties of nanoparticles in hydrogels show two effects based on the properties of the polymer matrix. With increasing density of the polymer network, the diffusion coefficient of the nanoparticles is reduced. This effect is often modeled by considering the polymer matrix to act as an "obstruction" for the nanoparticles, which otherwise diffuse freely in the water-continuous phase. Physical models of diffusion of particles in hydrogels allow the derivation of matrix properties such as mesh size and polymer strand thickness. A second, but less often described observation is that in heterogeneous hydrogels, multi-modal diffusion of nanoparticles can be present.4 This can be explained by the presence of micro-domains with different polymer densities, in which particles have different diffusion coefficients. The sizes of these micro-domains are at least of order of the mean-square displacement of the particles over the diffusion observation


window. In the long run, a single diffusion coefficient will again be observed due to particle exchange between domains (central limit theorem). Although nanoparticle diffusion can be measured by very sensitive optical methods, such as fluorescence correlation spectroscopy (FCS) or fluorescence recovery after photobleaching (FRAP), these methods do not measure diffusion of the entire particle ensemble. Besides this, particles have to be fluorescent in order to be observed by FCS and FRAP. Pulsed-field gradient (PFG) NMR, on the other hand, does measure ensemble properties, allowing the observation of multi-modal diffusion due to gel heterogeneity. For PFG NMR, particles do not need to be labeled. However, in 1H NMR, particles have to be observed against a highly protonated background of water and solutes. Because nanoparticles are dosed at low levels in order to prevent perturbation of the network structure, this constitutes a significant measurement challenge. For this reason, we designed 19F labeled nanoparticles for background free NMR observations. We used complex coacervate core micelles (C3Ms), also known as polyion micelles, functionalized with poly(ethylene oxide) (PEO) coronas and 19F labels within their cores. C3Ms are spherical nanostructures with a complex coacervate core, consisting of a complex of positively and negatively charged polyelectrolyte chains, and a neutral, hydrophilic corona.10 A PEO corona solubilizes the particles and prevents attractive or repulsive interactions between the particles and the biopolymer matrix. The size of the micelles depends on the length of the various polymer blocks. 19F is an NMR active isotope with a high gyromagnetic ratio that is otherwise not present in biopolymer systems. Therefore, 19F labeling allows background-free observation of the particles. In this study, we have used PFG NMR diffusometry to measure diffusion of labeled 19F-C3Ms (diameter approximately 30 nm) in a heterogeneous gel model system (kappa-carrageenan). In this model system, multi-modal diffusion of dendrimer nanoparticles (diameter of order 5 nm) was observed.4 We assess the merits of 19F NMR diffusometry for characterization of heterogeneous gels, and compare this to the results obtained by 1H NMR diffusometry.

2 EXPERIMENTAL SECTION

2.1 Design of 19F Labeled Complex Coacervate Micelles

Complex coacervate core-based micelles, functionalized with PEO coronas, were prepared according to procedures described by Bourouina et al.10 Poly(allylamine hydrochloride) (PAH, 15 kDa) was used as the positively charged polyelectrolyte. Diblock copolymer poly(methacrylic acid)-b-poly(ethylene oxide) (PEO-PMAA), which was used as the negatively charged polyelectrolyte, was functionalized with a third block containing –C19F3 groups (PEO-PMAA-19F, 6 kDa). Micelles were formed in water, after which crosslinking of about 25% of the ionic bonds in the coacervate cores was performed. A schematic of the 19F labeled 19F-C3M micelles is presented in Figure 1. More details will be published elsewhere. After crosslinking of the coacervate cores, we measured longitudinal (T1) and transverse (T2) relaxation times. Experiments were performed on a Bruker Avance II spectrometer, equipped with a Bruker diff25 probe, at 7.0 T (300 MHz for 1H and 282 MHz for 19F), in water at 294 K. The probe was equipped with a 10-mm RF insert that could be tuned to both 1H and 19F. Sample volume was chosen as to not exceed the NMR coil volume. T2 was measured by a frequency-domain CPMG experiment with an inter-echo time of 1 ms. In the 1H case, diffusion editing was used to suppress the water signal. The diffusion coefficient of 19F-C3Ms in water at 294 K was determined by dynamic light scattering (DLS). Hydrodynamic diameters could then be calculated through the Stokes-Einstein equation.

Multiscale Defi nition of Food 113

Figure 1 Design of 19F-labeled complex coacervate core (19F-C3M) micelles. PEO-PMAA-19F contains the negatively charged polyelectrolyte PMAA; PAH is the positively charged polyelectrolyte.

2.2 Diffusometry Experiments in Kappa-carrageenan

2.2.1 Preparation of Kappa-carrageenan Gels. Kappa-carrageenan gels were prepared by dissolution of kappa-carrageenan powder (Sigma-Aldrich), NaCl and KCl salts and nanoparticles in water, as described by Lorén et al.4 Kappa-carrageenan weight fraction was varied between 0.25% and 3% in six steps. In all experiments, 19F-C3Ms were dosed at 0.1 wt%; NaCl concentration was kept at 200 mM and KCl concentration at 20 mM. Polymer solutions were transferred into NMR-tubes, where the filling height was chosen not to exceed the linear part of the magnetic field gradient. Gels were allowed to form and stabilize for 24 hours before measurements were performed.

2.2.2 NMR diffusometry. PFG NMR experiments were carried out by stepwise variation of the gradient pulse amplitude, while keeping the diffusion-observation time and gradient pulse width constant. The attenuation of signal intensity as a function of the experimental parameters is described by the Stejskal-Tanner equation.11 In case of multi-modal diffusion of particles in a heterogeneous system, the signal attenuation curve can be fitted with a discrete sum of attenuation exponentials , where is the signal attenuation, Ai the amplitude and Di the diffusion coefficient (m2 s–1) of component i, the gyromagnetic ratio of the observed nucleus (rad T–1 s–1), the effective gradient pulse width (s), g the magnetic field gradient amplitude (T m–1) and the effective diffusion time (s), where (narrow gradient pulse approximation). All PFG NMR experiments were performed on the same Bruker Avance II spectrometer at 7.0 T equipped with a Bruker diff25 gradient probe as described above. This probe generates a maximum field gradient strength of 9.6 T m–1. Sample temperature was kept at 294 K, regulated indirectly through the gradient-coil cooling system to prevent any temperature gradients across the sample.

2.2.2.1 1H (300 MHz) and 19F (282 MHz) DOSY. Diffusion-ordered spectroscopy (DOSY) experiments were performed in water and carrageenan gels. Two sets of DOSY experiments were performed. The first set was a direct comparison of 19F with 1H diffusometry at different gel densities. For these experiments, we used an effective diffusion time of 100 ms and an effective gradient pulse width of 2.7 ms. For 1H, the gradient amplitude was varied in a logarithmic manner between 0.96 and 9.17 T m–1 in 64 steps. A spin-echo (PFG-SE) based experiment


was used in combination with unipolar, trapezoid-shaped gradient pulses. For 19F, gradient amplitude was varied between 0.05 and 9.71 T m–1 for 19F in 64 steps. A stimulated-echo (PFG-STE) based experiment was used in combination with unipolar, trapezoid-shaped gradient pulses. The initial gradient was stronger in the 1H case than in the 19F case in order to effectively suppress the intense 1H signal of water, whose diffusion coefficient is ~100 times higher than that of 19F-C3Ms. For 1H NMR measurements, the NMR signal was averaged 96 times and for 19F NMR 1024 times. Experimental repetition time was set at 2 times T1 of PEO and 19F labels, respectively. The experiments were performed in kappa-carrageenan gels with gel concentrations of 0.25, 0.50, 0.75, 1.0, 2.0 and 3.0 wt%. The second set of measurements was performed to see whether the diffusion-observation time affected the outcome of the experiment, which would be the case if particle exchange between heterogeneous domains takes place on the time scale of the diffusion experiment, or in case particles experience restricted diffusion. These experiments were carried out only for 1H, because it offers more sensitivity than 19F. We used a spin-echo based experiment (PFG-SE) with unipolar, trapezoid-shaped gradient pulses. Attenuation curves were measured for three diffusion times of 60, 300 and 600 ms. Gradient pulse width was 5 ms in all experiments. For =60 ms, the gradient amplitude was stepped logarithmically between 0.68 and 9.71 T m-1, for =300 ms between 0.30 and 4.29 T m-1 and for =600 ms between 0.21 and 3.03 T m-1. These parameters lead to identical sampling of the b-axis (in s m-2) for all diffusion times. Non-zero initial gradient values were chosen to suppress the water signal. In order to avoid effects due to T2 relaxation weighting between the three experiments, which would complicate direct comparison of the attenuation curves, the echo times of the =60 ms and =300 ms experiments were increased by 540 ms and 300 ms, respectively. Hence in all three experiments the 1H NMR signal will experience the same transversal relaxation decay (corresponding to total delay of 600 ms). These experiments were performed on 19F-C3Ms in a 2 wt% carrageenan sample only. 2.2.2.2 1H (300 MHz) diffusion-relaxation correlation spectroscopy (DRCOSY). We performed 1H DRCOSY in order to measure T2 and diffusion coefficients of particles simultaneously. This way, we would be able to see whether T2 of a restricted or more hindered fraction was lower than the T2 of a less hindered particle fraction. Because of the intense background-water signal, time domain CPMG acquisition was problematic. For this reason we reverted to frequency-domain acquisition, while going stepwise through the CPMG dimension. The CPMG train was placed before the PFG block. This way, any remaining water signal due to pulse imperfections would be suppressed during the diffusion experiment. For the CPMG block, we used an echo time of 1 ms and sampled the T2 dimension in 32 steps between 0 and 1800 echoes; Only an even number of 180-degree pulses was used. We used a spin-echo (PFG-SE) based experiment with a gradient pulse width of 5 ms and diffusion time of 100 ms. Gradient amplitude was varied in a logarithmic manner between 0.52 and 9.70 T m–1 in 48 steps. The repetition time of the experiment was made independent of the length of the echo train by inserting a variable compensatory delay after acquisition of the FID. The DRCOSY experiment was performed on 19F-C3Ms in a 2 wt% carrageenan sample only.

2.2.2.3 1H diffusion exchange spectroscopy (DEXSY). We performed 1H DEXSY to assess whether diffusive exchange between particle fractions could be observed. The PFG blocks were spin-echo based (PFG-SE), combined with unipolar, trapezoid-shaped gradient pulses. The diffusion time was set at 40 ms, gradient pulse width at 2.5 ms, and the gradient amplitude was stepped in a logarithmic manner between 0.58 and 9.49 T m-1. The exchange


time was set at 400 ms. The DEXSY experiment was performed on 19F-C3Ms in a 2 wt% carrageenan sample only.

2.2.3 Data Analysis. Bootstrap data resampling12 was used to estimate confidence intervals in diffusion coefficients and relaxation times. Bootstrap resampling also effectively stabilizes 2D Laplace inversion under high noise levels. Bootstrap resampling was implemented as transformation of subsequent random sub-selections of data points, and summation of the resulting intensity spectra or correlation maps. Alternatively, addition of noise has been used as an analogous method to estimate errors in PFG NMR data13, but we prefer bootstrap resampling because it does not manipulate the data.

2.2.3.1 Analysis of DOSY Data. NMR spectra were obtained through Fourier transformation of FIDs and subsequent phasing using standard procedures. The 19F and PEO–1H resonance lines were integrated to obtain their attenuation curves. We then prepared 1000 resampled attenuation curves, which were successively fitted with 1–2 attenuation exponentials through SplMod.14,15 Confidence intervals were estimated by calculating percentiles from the bootstrapped parameter distributions.

2.2.3.2 Analysis of DRCOSY and DEXSY Data. NMR spectra were obtained through Fourier transformation of FIDs and subsequent phasing. The 1H-PEO resonance line was integrated to obtain the two-dimensional experimental response surface. Subsequently, the data were resampled 100 times and successively fitted through Fast 2D Laplace Inversion (FLI).16 For all fits, the regularization parameter was fixed at a value appropriate for the non-resampled dataset. Resampled correlation maps were summed to obtain an average map, in which most noise artifacts had disappeared. 3 RESULTS AND DISCUSSION 3.1 19F Labeled Complex Coacervate Core Micelles Attaching –C19F3 units to the negatively charged PEO-PMAA electrolyte lead to the formation of stable and water-soluble crosslinked 19F-C3Ms with a molecular mass of 453 kDa of which 4.0 wt% 19F. From DLS data, we calculated a particle diameter of 31 nm. PFG NMR diffusion experiments show a self-diffusion constant of 1.42 10-11 m2/s and the mono-exponential attenuation curve indicates that the particles are essentially monodisperse. For 1H NMR, the particles are observed from their prominent PEO signal (3.6 ppm). At 300 MHz, we measured T1(PEO)=507±25 ms and T2(PEO)=343±34 ms. The 19F NMR spectrum showed a major signal at –71 ppm, with T1=620±20 ms and T2=11±1 ms. We conclude that the current particle design allows for NMR diffusometry on the 101–102 ms timescale, both for 1H and 19F. Since the T2 value of the 19F labels is relatively short due to their position within the micellar cores, stimulated-echo based experiments were used to prevent signal loss due to id T2 decay. We note that by adapting the current 19F-C3M design, the 19F transversal relaxation times can be prolonged by increasing the internal mobility of the label. 3.2 Diffusometry in Kappa-carrageenan Gels 19F and 1H-PEO DOSY attenuation curves of 19F-C3Ms in kappa-carrageenan gels at different gel densities are presented in Figure 2. 1H measurements start at higher b-values in order to effectively remove the contribution of water. A measurement on a gel sample without 19F-


C3Ms showed no background signal of the carrageenan matrix, due to the short T2 of the immobile kappa-carrageenan polymer chains. All 1H signal is therefore arising from 19F-C3Ms. In the 1H case, noise levels are significantly lower than in the 19F case, but on the other hand, 19F DOSY allows for background-free observations in any system and does not require a strong initial gradient for water suppression.

Figure 2 1H PFG-SE and 19F PFG-STE attenuation curves of 19F-C3Ms in kappa-carrageenan gels of different concentrations. Note that the 1H measurements start at higher b values because of water suppression.

After fitting attenuation exponentials to the experimental data, we found that the first component of a 2-component fit of 1H data showed good overlap with a 1-component fit of 19F data at all measured concentrations. It can be seen in Figure 3 that fitting a 1-component model to both 19F and 1H data does not give good overlap. Fitting 2 components to 19F data did not yield a satisfactory result, because the fastest component predicts a higher diffusion coefficient than that of 19F-C3Ms in water (not shown). Apparently, in 19F DOSY, noise levels are high with respect the intensity of the second component, so that the second component cannot be found. Besides the effects of noise, the second component could have a slightly shorter T2 because it is in a more crowded environment, which would have a significant effect on its intensity: T2 of the 19F labels (11 ms) is of the same order as the time that the spins are in the transverse plane during the diffusion experiment. Although 1H DRCOSY of 19F-C3Ms in 2 wt% kappa-carrageenan shows that the two components do not differ significantly in T2 relaxation time (Figure 4), T2 of the PEO groups is an order of magnitude longer than that of the 19F labels and therefore relatively long with respect to the duration of the diffusion experiment. Therefore, small differences in T2 between the two components would only have a significant weighting effect on the 19F echoes. Lorén et al.4 hypothesized that bi-modal diffusion of dendrimer nanoparticles in kappa-carrageenan gels can not reflect the heterogeneous sub-micron structure of the gel, because the mean-square displacement of dendrimer nanoparticles during the PFG NMR measurement (~1 m) is higher than the largest microstructural features in the gel (~ 10 nm). Therefore, microstructural effects should have averaged out on the timescale of the NMR diffusion experiment and bi-modal diffusion should reflect two truly different diffusion coefficients, e.g. due to differences in density of domains that are larger than the mean-square displacement of the particles during the diffusion experiment. This explanation is straightforward, but somewhat unsatisfactory in that there is no direct microscopic evidence for the existence of such domains, and that if these domains indeed exist, we would expect to observe diffusive particle exchange between them. A DEXSY experiment (400 ms exchange time), however, did not reveal any diffusive exchange between the components (Figure 5, left panel). Also, we did not observe a change in the attenuation curves upon changing the


diffusion time (Figure 5, right panel). This indicates that, if indeed two domains exist which differ in network density, they are large enough (micron-scale) to prevent strong exchange contributions and dependency of PFG attenuation curves on diffusion time.

Figure 3 1H and 19F attenuation curves at different gel concentrations were fitted with one and two components. Good overlap is seen between a 1-component fit of 19F and a 2-component fit of 1H data, but not between a 1-component fit of 19F and a 1-component fit of 1H data.

In a separate 19F NMR diffusometry study of 19F labeled dendrimers in the same kappa-carrageenan gels, we found a slow-diffusing particle fraction for particles with a hydrodynamic diameter of 7 nm, but not for dendrimers with a hydrodynamic diameter up to 5 nm. Besides this, the intensity of the second component was <10 %: much lower than in the 19F-C3M case.17 If the two components truly represent two different diffusion coefficients, i.e., related to micro-domains with different polymer densities, we would expect that theirintensity ratio would be independent of particle size. However, at identical gel concentration, the intensity for the second component of C3Ms is 40%, It might be that the second component does not represent a separate diffusion coefficient after all. The particle-size dependency of the intensity of the second component can be better explained by viewing the network mesh as a heterogeneous sieve, through which the particles have to pass when moving between two points in the system. Larger particles will experience stronger sieving, leading to a more pronounced non-Gaussian displacement distribution at the intermediate timescale where the average displacement of the particles in of the order of the network heterogeneity.18 This more pronounced non-Gaussian displacement distribution would lead to a higher apparent intensity of a second component.


Figure 4 1H DRCOSY map of 19F-C3Ms in 2 wt% kappa carrageenan shows two components with comparable T2 relaxation time.

Figure 5 Left panel: 1H DEXSY (40 ms mixing time) of 19F-C3Ms in 2 wt% kappa carrageenan shows no off-diagonal peaks, indicating that no diffusive exchange takes place between the domains. Right panel: 1H PFG-SE attenuation curves obtained at three different diffusion times (but all at equal echo time) show no significant differences.

The finding that the attenuation curves do not differ between diffusion times of 60 and 600 ms, might be due to the fact that the mean-square displacements of the nanoparticles in this time interval varies only a factor of 3. Although this is significant, it also means that the particles are still probing the same length scale at 60 and 600 ms and therefore it would be plausible not to see strong changes between the attenuation curves.

4 CONCLUSION Functionalized complex coacervate core micelles fulfill the demand for monodisperse particles with diameter in the 10-50 nm size range that can be used for NMR diffusometry. 19F-C3Ms functionalized with a PEO corona and internally labeled with 4.0 wt% 19F allow 1H and 19F NMR diffusion experiments with diffusion-observation times in the 10-100 nm range. Although sensitivity for 19F is lower than for 1H, 19F diffusometry can offer significant advantages in compositionally complicated matrices in which it is difficult to separate the 1H signal of the particles from the background.


We measured diffusion of 19F-C3Ms in heterogeneous kappa-carrageenan gels. The PFG attenuation curves can be fitted with two components. The current data support two interpretations of the second component that is observed in the 1H PFG decays. On the one hand, the fact that attenuation curves are apparently invariant between 60 and 600 ms diffusion time points to the existence of two separate diffusion coefficients, reflecting the existence of domains at the micron-scale with different mesh sizes, in which the particles have different diffusion coefficients. On the other hand, the fact that the intensity of the second component is significantly lower for dendrimer nanoparticles at the same gel concentrations can be better explained by viewing the network mesh as a heterogeneous sieve. If the diffusion path of the particles is of the order of the network heterogeneity, there is a non-Gaussian distribution of particles displacements, which becomes more pronounced when the particles are larger with respect to the smallest pores.

Acknowledgments

We thank Yi-Qiao Song (Schlumberger-Doll Research, USA) for providing the FLI algorithm. This research received funding from the Netherlands Organization for Scientific Research (NWO) in the framework of the Technology Area COAST.

References

1 L. Masaro and X. Zhu, Prog. Polym. Sci., 1999, 24, 731–775. 2 L.-H. Cai, S. Panyukov, and M. Rubinstein, Macromolecules, 2011, 44, 7853–7863. 3 D. Bernin, G.-J. Goudappel, M. van Ruijven, A. Altskär, A. Ström, M. Rudemo, A.-M.

Hermansson, and M. Nydén, Soft Matter, 2011, 7, 5711–5716. 4 N. Lorén, L. Shtykova, S. Kidman, P. Jarvoll, M. Nydén, and A.-M. Hermansson,

Biomacromolecules, 2009, 10, 275–284. 5 S. Salami, C. Rondeau-Mouro, J. van Duynhoven, and F. Mariette, J. Agric. Food

Chem., 2013, 61, 5870–5879. 6 S. Salami, C. Rondeau-Mouro, J. van Duynhoven, and F. Mariette, Food hydrocolloids,

2013, 31, 248–255. 7 J. Hagman, N. Lorén, and A.-M. Hermansson, Biomacromolecules, 2010, 11, 3359–

3366. 8 R. Colsenet, O. Söderman, and F. Mariette, Macromolecules, 2006, 39, 1053–1059. 9 T. Stylianopoulos, B. Diop-Frimpong, L. L. Munn, and R. K. Jain, Biophys. J., 2010, 99,

3119–3128. 10 N. Bourouina, M. A. Cohen Stuart, and J. M. Kleijn, Soft Matter, 2014. 11 E. O. Stejskal and J. E. Tanner, J. Chem. Phys., 1965, 42, 288–292. 12 B. Efron, Ann. Statist., 1979, 7, 1–26. 13 J. S. Alper and R. I. Gelb, J. Phys. Chem., 1990, 94, 4747–4751. 14 R. W. W. van Resandt, R. H. Vogel, and S. W. Provencher, Rev. Sci. Instrum., 1982, 53,

1392. 15 R. H. Vogel, SplMod Users Manual, Version 3, Heidelberg, 1988. 16 Y. Q. Song, L. Venkataramanan, M. D. Hürlimann, M. Flaum, P. Frulla, and C. Straley,

J. Magn. Res., 2002, 154, 261–268. 17 D.W. de Kort et al., submitted 18 N. Lorén, M. Nydén, and A.-M. Hermansson, Adv. Colloid Interface Sci., 2009, 150, 5–

15.

DOUBLE EMULSION CHARACTERISATION WITH PFG-NMR METHODS: WOW AND OWO

Richard Bernewitz1, Esteban Caro2, Daniel Topgaard3, Heike P. Schuchmann1, Gisela Guthausen2 1 Institute of Process Engineering in Life Sciences, Section I: Food Process Engineering 2 Pro2NMR, IBG2 and IMVM, KIT, Karlsruhe, Germany 3 Physical Chemistry, Lund University, Lund, Sweden

1 INTRODUCTION

Double emulsions are of high scientific interest, especially in the field of functional foods. In a first approach, these dispersed systems are described by structural parameters, chemical composition and diffusion parameters. Progress has been made in the characterisation of double emulsions by means of diverse NMR methods with respect to droplet size distribution (DSD), disperse phase ratios (DPR) and molecular exchange (ME) of WOW- and OWO-double emulsions. However, both double emulsion types exhibit different NMR-properties, such that the PFG-NMR methods have to be adapted for the special needs of these systems. Different NMR sequences and data processing steps are compared for OWO and WOW double emulsions. Methods to determine DSD, DPR and ME – relevant for description und understanding of release - are presented and compared with findings from alternative measuring techniques.

2 DROPLET SIZE DETERMINATION IN WOW AND OWO EMULSIONS

Since years, pulsed field gradient NMR is used to determine droplet size distributions in emulsions. 1, 2 Measurements on double emulsions3-6 are more demanding as two chemically similar phases have to be discriminated. This discrimination can no longer be achieved by chemical shift separation or relaxation7-10 in most of the cases. Diffusion properties have to be exploited to separate inner and outer water and oil phases in WOW and OWO double emulsions, respectively. However, the success of full separation of the three phases is a prerequisite for quantitative data interpretation. The outer most phases Wo and Oo are characterised by quasi-free diffusion, which can be modelled by one or more apparent diffusion coefficients Dapp. Of course, Dapp depend on the substances and additionally on the concentration of the dispersed phase, as the molecular motion is geometrically hindered by the presence of droplets. In the inner most phase, the Brownian motion is restricted by the droplets inner surface, leading to restricted diffusion. This picture leads to two models for the description of double emulsions in pulsed field gradient NMR (PFG-NMR), which differ by the role of molecular


exchange on the time scale of the experiment between similar phases, here Wo and Wi. The first approach neglects the exchange, leading to a two compartment model, in which the outer water phase is described by an apparent diffusion coefficient, the inner water phase contains the information about DSD and is modelled according to Murday, Cotts and Packer.1, 2 In the second approach, the exchange between similar phases is taken into account in a similar way as it was done in NMR examination of tissue. The molecules are assumed to diffuse between the phases through a barrier, leading to the introduction of residence times.11-16 The situation is more complex in OiWOo emulsions, as the diffusion of oil cannot be characterised by a single diffusion coefficient, but rather by a distribution. This distribution is convoluted by the effect of geometrical hindrance in the double emulsion. Additionally, the difference of effective diffusion coefficients of Oi and Oo is not as pronounced as for WiOWo emulsions, such that the data analysis is more complicated, but can be solved analytically, when knowing the properties of the conjugated single emulsion.17 Inner DSD of WiOWo (Fig. 1a)18) and OiWOo (Fig. 1b)17) emulsions are compared in Figure 1 as obtained by PFG-STE measurements. The PFG-NMR data of Fig. 1a) was analysed according to the two-compartment model (see inset). The Wi-DSD size range is in good agreement with the Wi size range extracted of CLSM raw data.19 The inset of Fig. 1b) shows PFG-NMR data of a model OiWOo emulsion. The inner DSD was obtained while considering a distribution of Dapp in Oi and hindered diffusion in Oo

17. Additional experiments showed that molecular exchange is negligible in the investigated double emulsion. This procedure led to an Oi-DSD similar to the O-DSD of the corresponding OW single emulsion, which is available during the double emulsion’s production.

Figure 1 Inner DSD of WiOWo and OiWOo double emulsions determined by PFG-NMR. a) Wi-DSD of a WiOWo emulsion. The black line represents the Wi-DSD obtained fromPFG-NMR and ( ) the Wi-DSD from CLSM raw data. The inset shows the PFG-NMR water data ( ) together with the fit according to the two compartment model. b) Oi-DSD of a model OiWOo (black line) and the O-DSD (grey line) of the corresponding OW emulsion. The inset shows the oil signal attenuation of the OiWOo ( ) and the corresponding OW ( ) emulsion. In both cases inner DSDs can be determined and compared with results from complementary techniques.

3 DISPERSED PHASE RATIO IN WOW AND OWO EMULSIONS

Apart from the DSD as the main structural parameter of emulsions, the chemical composition of the phases in multiple emulsions is of interest. In contrast to the intermediate phase, the


relative amount of inner and outer phase can change during storage and during double emulsion production due to the mechanical energy intake. To quantify the phase ratios by means of NMR, oil and water can be separated via their chemical shift difference, even in low field NMR spectroscopy.20 As most NMR investigations of emulsions are performed by pulsed field gradient sequences, relaxation and diffusion effects have to be taken into account, when the 1H-NMR amplitudes are quantified in terms of concentrations of the diverse phases – additionally to the number of 1H per molecule. Overall, the following phenomena have to be considered when deducing DPR’s from PFG-experiments:

Restricted diffusion inside the droplets – depending on the inner DSD, hindered diffusion in the phases containing droplets – depending on DPR and DSD, molecular exchange on the NMR time scale between the inner and outer most phases– depending on the recipe of the emulsion, the longitudinal and transverse relaxation times of all phases, the (apparent) diffusion coefficient of the phases (water and oil), the number of 1H atoms per molecule.

In the case of WOW emulsions, diffusion contrast has been exploited between inner and outer water phases. Additionally, a correction for transverse and longitudinal relaxation has been performed.17 Of course, the procedure to be chosen depends on both DSD’s - because the diffusion properties depend on it - as well as on the molecular exchange properties which are discussed in the next section. In the case of a double emulsion with negligible exchange on the time scale of the NMR diffusion time, the signal amplitude at low gradient amplitude reflects the total 1H content of water and oil, the signal amplitude at large gradients is proportional to the inner dispersed phase. Fig. 2a)17 shows the phase ratios j of a WiOWo double emulsion with initial Wi = 0.15, O = 0.35 and Wo = 0.5 (dashed lines), where O is a constant during the emulsification process. It can be seen, that due to the D- and T1,2 correction of the data, changes in O as a function of are minimised. As ME is negligible for small , j at = 50 ms are the most reliable values. The same principle is valid for OiWOo double emulsions. The molecular weight distribution in vegetable oils leads to a blurred diffusion contrast, such that Oi and Oo signal cannot be distinguished clearly. However, if the signatures of the outer and inner phases are known, the OWO PFG-NMR signal can be reconstructed while keeping the signal ratio variable, which is directly linked to the DPR (Fig. 2b). The best reconstruction of the signal attenuation from that of the corresponding single emulsion was achieved with x = 0.08, instead of initial x = 0.33. This indicates a preservation degree of O1(t)/ O1(t = 0) = 24 %. ME was examined by CLSM experiments and found to be negligible in the DSD determination on this double emulsion.17

4 MOLECULAR EXCHANGE IN WOW AND OWO EMULSIONS

Multiple emulsions enable the encapsulation of active agents and other sensitive molecules in the inner phases to prevent unwanted side reactions and early digestion. Retard effects can be adjusted, depending on the double emulsions’ composition. It is therefore of major importance to know parameters and mechanisms of release and exchange between phases. About the mechanisms, diverse conceptions can be found.21-37 The most discussed mechanisms are molecular diffusion and diffusion of molecules within micellar structures, formed by emulsifier molecules. Some NMR investigations aim therefore to shade some light onto this important question.


Figure 2 Determination of DPR with PFG-NMR. a) DPR j of a WiOWo emulsion (j: Wi ( ), Wo ( ), and O ( )), extracted from spectrally resolved PFG-NMR signal for different diffusion times . The black symbols represent the uncorrected data points. The grey symbols show the corrected data, considering T1 and diffusion effects. b) The signal reconstruction ( ) of a OiWOo ( ) emulsion describes the signal attenuation when considering the above listed aspects, and the DPR can be determined.

As already mentioned, the exchange can be observed in conventional PFG experiments and was reported especially for stimulated echo experiments.12, 15, 16 Another approach uses paramagnetic relaxation agents and the observation of their interaction with molecules in the different phases. These relaxation agents contribute to longitudinal as well as to transverse relaxation such that a measurement of chemical shift resolved relaxation rates allows determining the concentration of the relaxation agents in the corresponding phases. When additionally taking into account the solubility of the diverse relaxation agents in the hydrophilic and hydrophobic phases, detailed conclusions can be drawn also concerning the question whether molecular diffusion or micellar transport is present. The relaxation experiments point strongly to an exchange process on a molecular rather than a micellar basis. Both, a WiOWo

38 and a OiWOo17 were investigated. In both cases, molecular diffusion

has been found to be the dominant diffusion mechanism for molecular exchange phenomena of relaxation agents. This conclusion can be drawn because the relaxation rates of both, oil and water phases were enhanced. Due to the distance (r) dependence of the dipolar part of paramagnetic relaxation, which is proportional to 1/r6, the enhancement of the intermediate phase is only possible, when the molecules are to a considerable amount within a close neighbourhood of the paramagnetic molecules. Additionally, filter exchange spectroscopy (FEXSY39) can be applied to investigate the diffusion of molecules between inner and outer phases (ME). Due to the diffusion filter at the beginning of the FEXSY sequence (Fig. 3), the signal of the outer phase of a double emulsion can be suppressed, while the signal of the inner phase is retained by the restriction of the molecules in the droplet. While increasing the mixing time in a measurement on a WOW emulsion, an increasing amount of outer water is observed. This can only be explained by exchange of water between the inner and outer phases.


Figure 3 The Filter EXchange SpectroscopY (FEXSY) sequence allows filtering out the outer phase’s signal by applying a diffusion filter in the beginning of the sequence. By varying the mixing time tm, diffusion from the inner to the outer phase can be observed. FEXSY was applied to WiOWo emulsions with varying xanthan concentration in Wo. The surface active xanthan is usually used as a thickener and was already found to influence water residence times in a WiOWo in PFG-NMR experiments.15 FEXSY measurements prove insight into molecular exchange for example by the increase of the ratio S(g1)/S(g3) with increasing mixing time tm (Fig. 4). g1 and g3 denote the first and the third gradient step, respectively. The mentioned change in the ratio S(g1)/S(g3) is also observed in the present case.

Figure 4 FEXSY spectra of WOW emulsions with a) and b) 0.1 w-% and c) and d) 1.0 w-% Xanthan in Wo. The water peak at C = 4.6 ppm depends on the mixing time. a) and c) show spectra at short mixing time tm = 0.2 s, whereas b) and d) show the corresponding spectra at tm = 0.4 s. The diffusion of water molecules from Wi to Wo is proven for both concentrations due to the increase of the ratio S(g1)/S(g2).


The findings about molecular exchange in the diverse NMR experiments are supported by CLSM investigations (Confocal Laser Scanning Microscopy), where the diffusion of dye molecules across the hetero-phase can be observed in both, WOW and OWO emulsions.17, 38 From both instrumental methods, NMR and CLSM, the kinetics of exchange can be derived. In the first case, the equilibration of relaxation agent concentration can be observed as well as the exchange of molecules in concentration equilibrium by PFG-NMR. In the second case, dye concentration differences can be measured time-resolved, thus leading to the observation of a concentration equilibration process. Both complementary methods show that a double emulsion cannot be thought of as a static structure, but rather the possibility of release and molecular exchange has to be considered. It should be pointed out, however, that not all double emulsions show exchange on the NMR time scale. Depending on the recipe, i.e. especially the emulsifier system, the residence times vary. The addition of gelation agents like xanthan, which is also surface active, leads to a variation of residence times. Therefore, some effort should be made to characterise the emulsion system and the emulsifiers in detail.

5 CONCLUSIONS

Double emulsions offer a variety of new possibilities in product development. In the last years, progress has been made in their analytical characterisation and description. In this work, diverse NMR methods were summarised to characterise both double emulsion types, WOW and OWO. PFG-NMR is a powerful tool to investigate the structural parameters of double emulsions. The inner DSD and DPR can be determined after the production. NMR relaxometry, using paramagnetic relaxation agents, delivers information about diffusion mechanisms on the molecular scale, which reveals deep insights into molecular diffusion phenomena.

6 ACKNOWLEDGEMENT

DFG is kindly acknowledged for financial support of the instrumental facility Pro2NMR. We would like to thank Lydia Schütz for the production of the emulsions.

References

1 K. J. Packer and C. Rees, J. Colloid Interf. Sci., 1972, 40, 206–218. 2 J. S. Murday and R. M. Cotts, J. Chem. Phys., 1968, 48, 4938-4945. 3 K. G. Hollingsworth and M. L. Johns, J. Colloid Interf. Sci., 2006, 296, 700-709. 4 M. L. Johns, Curr Opin Colloid In, 2009, 14, 178-183. 5 I. Lönnqvist, B. Haakansson, B. Balinov and O. Soederman, J. Colloid Interf. Sci., 1997,

192, 66-73. 6 R. Mezzenga, B. M. Folmer and E. Hughes, Langmuir, 2004, 20, 3574-3582. 7 G. J. W. Goudappel, J. P. M. van Duynhoven and M. M. W. Mooren, J. Colloid Interf.

Sci., 2001, 239, 535-542. 8 J. P. M. van Duynhoven, G. J. W. Goudappel, G. van Dalen, P. C. van Bruggen, J. C. G.

Blonk and A. P. A. M. Eijkelenboom, Magnetic Resonance in Chemistry, 2002, 40, S51-S59.


9 J. P. M. van Duynhoven, B. Maillet, J. Schell, M. Tronquet, G. J. W. Goudappel, E. Trezza, A. Bulbarello and D. van Dusschoten, European Journal of Lipid Science and Technology, 2007, 109, 1095-1103.

10 M. A. Voda and J. P. M. van Duynhoven, Trends Food Sci. Technol., 2009, 20, 533-543. 11 S. M. Schoberth, N.-K. Bär, R. Krämer and J. Kärger, Anal. Biochem., 2000, 279, 100-

105. 12 J. P. Hindmarsh, J. Su, J. Flanagan and H. Singh, Langmuir, 2005, 21, 9076-9084. 13 J. Pfeuffer, U. Flögel, W. Dreher and D. Leibfritz, NMR Biomed., 1998, 11, 19-31. 14 J. Pfeuffer, U. Flögel and D. Leibfritz, NMR Biomed., 1998, 11, 11-18. 15 X. Z. Guan, K. Hailu, G. Guthausen, F. Wolf, R. Bernewitz and H. P. Schuchmann, Eur.

J. Lipid Sci. Tech., 2010, 112, 828-837. 16 F. Wolf, L. Hecht, H. P. Schuchmann, E. H. Hardy and G. Guthausen, Eur. J. Lipid Sci.

Tech., 2009, 111, 723-729. 17 R. Bernewitz, U. S. Schmidt, H. P. Schuchmann and G. Guthausen, J. Colloids and

Surfaces A: Physicochemical and Engineering Aspects, 2014, in press. 18 R. Bernewitz, Dissertation, Karlsruher Institut für Technologie (KIT), 2013. 19 S. Schuster, R. Bernewitz, G. Guthausen, J. Zapp, A. M. Greiner, K. Köhler and H. P.

Schuchmann, Chem. Eng. Sci., 2012, 81, 84–90. 20 G. Guthausen and A. Kamlowski, Magnetic Resonance in Food Science: Challenges in a

Changing World, Nordic House, Reykjavik, Iceland, 2009. 21 M. F. Ficheux, L. Bonakdar, F. Leal-Calderon and J. Bibette, Langmuir, 1998, 14, 2702-

2706. 22 N. Garti, A. Aserin and Y. Cohen, J. Control. Release, 1994, 29, 41-51. 23 T. Higuchi, J. Pharm. Sci.-Us., 1963, 52, 1145-&. 24 R. G. Stehle and W. I. Higuchi, J. Pharm. Sci.-Us., 1972, 61, 1922-1930. 25 S. Magdassi and N. Garti, J. Control. Release, 1986, 3, 273-277. 26 J. Weiss, C. Canceliere and D. J. McClements, Langmuir, 2000, 16, 6833-6838. 27 J. Weiss and D. J. McClements, Langmuir, 2000, 16, 5879-5883. 28 J. R. Avendano-Gomez, J. L. Grossiord and D. Clausse, J. Colloid Interf. Sci., 2005, 290,

533-545. 29 M. Stambouli, J. R. Avendano-Gomez, I. Pezron, D. Pareau, D. Clausse and J. L.

Grossiord, Langmuir, 2007, 23, 1052-1056. 30 N. Garti and A. Aserin, Adv. Colloid Interfac., 1996, 65, 37-69. 31 M. Verbrugghe, P. Sabatino, E. Cocquyt, P. Saveyn, D. Sinnaeve, P. Van der Meeren

and J. C. Martins, Colloid Surface A, 2010, 372, 28-34. 32 H. J. Bart, H. Jungling, N. Ramaseder and R. Marr, J. Membrane Sci., 1995, 102, 103-112. 33 A. G. Kopp, R. J. Marr and F. E. Moser, Institution of Chemical Engineers Symposium

Series, 1978, 54, 279-290. 34 L. X. Wen and K. D. Papadopoulos, Coll. Surf. A, 2000, 174, 159-167. 35 G. M. Tedajo, M. Seiller, P. Prognon and J. L. Grossiord, J. Control. Release, 2001, 75,

45-53. 36 R. T. Hamilton and E. W. Kaler, J. Phys. Chem., 1990, 94, 2560-2566. 37 A. T. Florence and D. Whitehill, Int J Pharm, 1982, 11, 277-308. 38 R. Bernewitz, F. Dalitz, K. Köhler, H. P. Schuchmann and G. Guthausen, Micropor.

Mesopor. Mat., 2013, 178, 69-73. 39 I. Aslund, A. Nowacka, M. Nilsson and D. Topgaard, J. Magn. Reson., 2009, 200, 291-

295.

ASSESSMENT OF TD-NMR AND QUANTITATIVE MRI METHODS TO INVESTIGATE THE APPLE TRANSFORMATION PROCESSES USED IN THE CIDER-MAKING TECHNOLOGY

C. Rondeau-Mouro1,2*, S. Deslis1,2, S. Quellec1,2, R. Bauduin3

1 IRSTEA, UR TERE, 17 Avenue de Cucillé, CS 64427, F-35044 Rennes, France 2 Université européenne de Bretagne, France 3 IFPC, Domaine de la Motte, F-35650 Le Rheu France * Corresponding author

1 INTRODUCTION

Extraction of the apple juice is one of the important operations in the cider-making technology1. It requires two unit operations: the apple grinding and the pressing of the ground apples. The grinding is a mechanical shredding of the fruit tissues that leads to the formation of a pulp (milled material), while the pressing is a mechanical operation which involves a cell burst or just a permeabilization of the cell wall to enable the juice flow out of the ground apples. Depending on the fruit quality (variety, ripeness, mealiness) and the extracting equipment, the yield of extracted juice varies strongly, from 50 to 75% for apple pure juice. Previous studies have highlighted some treatments of ground apple capable to increase the juice yields like traditional and microwave heating 2, Pulsed Electric Field (PEF) application 3 or pectinase treatment 4. A more classical technic consists in breaking up the press-cake and carrying out on a second pressing to permit the alternation of pressing and relaxation phases conducive to cell burst and juice flow (empirical observations). However the behavior of the parenchyma at cellular scale during grinding and pressing is largely unknown. NMR relaxation parameters as spin-spin (T2) and spin-lattice (T1) relaxation times have been shown to be very relevant to study the microstructure of plant tissues5. It is well established now that the T2 relaxation time is related to the water status in cell compartments and depends on the water interaction and chemical exchange with solutes and macromolecules present in cells. Relaxation times vary also with diffusional exchange between compartments through the cell permeable membranes, which result, in the case of slow diffusional exchange, to a multi-exponential NMR signal, reflecting water compartmentalization 5a. This phenomenon has been studied in different plant cells like in apple 6, tomato 7 or potato 8. The mathematical processing of the NMR signal measured on apple parenchyma gives four T2 components which are supposed to correspond each to a water compartment defined by his water content (relative intensity of the component) and by the environment of water which can interact with macromolecules but also solubilize solutes (T2 value). The attribution of each component to a subcellular compartment is still a subject of debate. Nevertheless, in the case of apple, the last studies tend to account for four components associated to water in cell walls, in extracellular spaces, in cytoplasm and in vacuole 9 . Magnetic Resonance Imaging (MRI) has also been used to study plant tissues 5b, 10. This non-destructive and non-invasive technique allows the recording of images (T1-weighted, T2-weighted, microporosity maps, intensity maps) offering information about the internal


structure of intact plant, their evolution under the plant development or under a specific stress 7b, 11. Many MRI studies have focused on the quality and defects of fruits12. Recent developments in MRI allowed the recording of multiple spin-echo MSE-MRI maps which allow the assessment of spatially resolved multi-exponential T2 in apples13. Moreover, it was also possible to extract from T2 maps quantitative information of microporosity 14. This method exploits the susceptibility effects induced by small gas bubbles on the MRI relaxation signal. The combination of NMR and MRI technics is an efficient way to monitor agricultural practices and industrial processes like ripening15, cooking16 or freezing17. Indeed the association of these two methods enables a study at different scales of the structure and the tissue integrity 18 and seems to be convenient for the study of juice extraction processes. The aim of the study was to understand the determinants that contribute to the extraction yield optimization. To characterize the impact of the grinding and pressing processes on apple tissues, TD-NMR relaxometry and quantitative MRI were performed not only on intact fruits but also on ground apples before and after the pressing step. NMR and MRI techniques allowed accessing multi-exponential transverse relaxation times (T2) providing insights on water status and distribution at the subcellular level and apparent microporosity maps giving information about gas distribution. These preliminary experiments aimed at developing and validating TD-NMR and MRI acquisitions planned for the characterization of cider apples harvested during a limited period in autumn. Therefore for practical reasons, measurements were first carried out on two apple cultivars usually stocked at cold temperatures and controlled atmosphere for an annual consumption: Reinette d’Armorique and Golden Delicious. For this last apple type, special ripening conditions were used to get mealy apples rather difficult to press.


2.1 Materials

The two apple cultivars were provided by IFPC (Institut Français de Production Cidricole, Le Rheu, France). The Reinette d’Armorique apples were harvested in September and October then stocked at 4°C. The Golden Delicious apples were harvested early October then stocked at 0 - 1°C under controlled atmosphere (replacing the oxygen by a gas stopping maturation of fruits). The Golden Delicious were left at least one week at ambient temperature before analyses to obtain overripe Golden apples expected to be more difficult to press compared to Reinette apples. Three kinds of sample were studied: apple parenchyma, milled material from the apple grinding and pressed material. The milled material was produced thanks to a fruit milling machine (Stossier V., Österreich, Austria) operating at 450 round/min. Then, the milled material was pressed on a small laboratory high-pressure press (model HP5, 5 L, Hafico, Fischer and Co., Dusseldorf, Germany) to obtain the crude juice. The hydraulic pressure was set at 8 bars during 10 min. on 3kg of apples. Each sample analyzed by NMR and MRI was dried in an oven at 103°C during 24h to estimate their water content. Table 1 gathers the water content for each sample analyzed. For the parenchyma samples, the sampling protocol consisted of cutting a 1 cm thick slice in the equatorial region of the fruit. In order to compare MRI and NMR results, the slices sampled for NMR in the equatorial region of each apple corresponded to the recorded MRI slices and the sampling of the cylinders has been carried out at the same locations as the MRI ROIs were drawn (figure 1).


Table 1 : Water content (expressed in % of the total humid weight of the three different samples studied by NMR and MRI).

Samples Reinette d’Armorique Overripe Golden Parenchyma 85.7 ± 1.7 % 84.7 ± 1.9 % Milled material 84.5 ± 1.0 % 83.4 ± 0.6 % Pressed material 80.3 ± 1.1 % 80.7 ± 0.8 %

Figure 1 Example of the sampling procedure used for NMR analyses based on MRI images.

For NMR, cylinders of 8 mm in diameters were cut out of the parenchyma of the apple slice. Samples were gently wiped to remove water from the broken cells and then placed in NMR tubes which were closed with caps. For the NMR samples of milled material, pieces of crushed parenchyma were selected to avoid the sampling of apple pericarp or seeds. For MRI analyses, fruits were placed on a specially designed support and marked, making it possible to select the equatorial slice for NMR measurements at the same position that the virtual slice. Milled materials were placed in a closed sampling flask inside the same support.

2.2 NMR relaxometry

NMR relaxometry measurements were performed on a 20 MHz (0.47 T) spectrometer (Minispec PC-120, Bruker, Karlsruhe, Germany) equipped with a thermostatted probe at 20°C. T2 relaxation times were measured using the Carr-Purcell-Meiboom-Gill (CPMG) sequence. 16000 and 18000 echoes were recorded for the pressed materials and the intact and milled parenchyma respectively with a 90°-180° pulse spacing of 0.2 ms. Data were averaged over 16 acquisitions. The recycle delay was adjusted at 10 s for each sample to avoid saturation of the magnetization. Spin-spin relaxation data were analysed with the following model :

offset)T/texp(I)t(I j24à1j

j (1)

The curves were fitted with two different methods: the discrete method Marquardt19 and the MEM method20. As the results given by both methods were consistent, only the results from the Marquardt method are presented here.

2.3 MRI experiments

The MRI measurements were carried out on a whole-body 1.5 Tesla scanner (Magnetom Avanto, Siemens, Erlangen, Germany) with maximum imaging gradients of 40 mT/m. The


MRI system was equipped with an emission/reception 15-channel knee coil to acquire data for whole fruits and milled materials. The temperature of the thermally insulated Faraday cage was set at 19 ± 1°C. The median equatorial planes of each fruit as well as the median vertical plane of milled material flasks were imaged with the following geometrical parameters: matrix size=128² pixels, field of view=152² mm² and slice thickness=10 mm and 5 mm for the whole fruit and the milled materials respectively. The repetition time (TR) was set at 10 s. for all measurements. For the whole fruits and the milled materials, T2 relaxation maps were obtained from multi-spin echo sequence (MSE) with 512 consecutives echoes, ranging from 6.5 to 3328 ms.T2* relaxation maps were obtained for the whole fruits from a multi-gradient echo sequence (MGE) with 12 consecutive echoes ranging from 2.42 to 18.15 ms (TE1=2.42ms then 1.43 ms spaced out). For each whole fruits, two ROIs were defined manually from the MSE images that corresponded to a homogeneous region of apple parenchyma. The mean value of these ROIs were computed for all the MSE amplitude images of the series and the decreasing T2 signal obtained was fitted via the Levenberg-Marquardt method (implemented on Scilab softwater) using a tri-exponential model. The T2 and T2* maps were created by a pixel-by-pixel mono-exponential fitting of the magnitudes of MSE and MGE images series via Levenberg-Marquardt algorithm implemented on Scilab software. The microporosity was then calculated pixel-by-pixel from T2 and T2* values to generate the microporosity map14.


3.1 T2 relaxation times of apple parenchyma

TD-NMR of apple parenchyma have been performed on two cultivars : Reinette d’Armorique and overripe Golden Delicious. For each of them, the relaxation decays could be fitted by a multi-exponential model (equation 1). Four T2 relaxation peaks were identified similarly to previous studies9, 13 as shown in Table 2. Despite the different cultivars and analysis temperatures, values obtained for the last component were very similar for each apple type with a T2(4) of 1178-1250 ms and I(4) of 79-81%. For the other T2 components, the values obtained in the present study were comparable to those of Adriaensen et al. on Reine de Reinettes13. The Granny Smith apples analyzed by Sibgatullin et al. 9 differed from the others with higher T2 values (480ms versus 363-384ms, 105 ms and 31 ms versus 65-88 ms and 9-13 ms for T2(3), T2(2) and T2(1) respectively). The intensities were rather similar for all the apples with values of 3-5%, 3-6% and 10-13% for the components 1, 2 and 3 respectively. The gap between the T2 values could originate from various reasons, including their ripening stage, known to influence the cell size but also the starch, sugar and water contents in fruits. Moreover, despite the different temperatures of measurement, the fitting method used by Sibgatullin et al. to analyze the NMR signal differed from ours9. Nevertheless this comparison demonstrated that, on the basis of NMR relaxation times, it was rather difficult to distinguish between the various apple cultivars. The four T2 components were tentatively assigned to four populations of protons in subcellular and extracellular compartments of the apple tissues. In order to propose an assignment of the various T2 values measured by NMR, it is important to understand that the proton population of “free” water should be characterized by a long T2 value close to 2 s. whereas a short T2 time would mean that water is probably in strong interaction (chemical exchange) with compounds like macromolecules. The relative intensity of a specific T2 value can also help to the assignment since it is proportional to the content of protons characterized


Table 2. T2 and relative intensity values of four cultivars of apple measured by NMR

Golden Delicious (20°C) Reinette d’Armorique (20°C) Reine des Reinettes (19°C)13 Granny Smith (25°C)9

Component T2 (ms) I (%) T2 (ms) I (%) T2 (ms) I (%) T2 (ms) I

(%)

1 13 ± 1 3.1 ± 0.5 9 ± 1 4.7 ± 0.7 10 ± 2 3.0 ± 0.3 31 5

2 79 ± 13 5.2 ± 0.7 65 ± 12 4.1 ± 0.6 88 ± 13 3.1 ± 0.4 105 6

3 363 ± 21 12.2 ± 1.6 384 ± 15 11.8 ± 1.0 376 ± 27 13 ± 1 480 10

4 1178 ± 4 79.4 ± 2.6 1187 ± 35 79.0 ± 1.0 1228 ± 39 81 ± 1 1250 79

by this T2 value. Regarding our results, the components 1 and 4 were easily assigned to water in cell walls and in vacuole respectively. Indeed in apple about 80% of the water is known to be localized in the vacuole which is in agreement with the relative intensity of the fourth component which represented about 80% of the signal. Moreover water present in cell walls should strongly interact with the cell wall macromolecules and then was waited to be characterized with a short T2 relaxation time. The water content in cell walls in not known but we could estimate the polysaccharide content which was around 1-2% of the total humid weight of parenchyma21. The mass density of water (two protons on the total water molecular weight) being two times higher than for glucose (12 protons on the total glucose molecular weight), the NMR mass intensity for water in cell walls should at least be twice that of cell wall protons, then compared to the total mass of samples, the water content in cell walls should totalize at least 2-4% of the total NMR signal, which corresponded to the signal amplitude for T2(1). The remaining compartments (extracellular spaces and cytoplasm) are more difficult to differentiate. Nevertheless considering that the majority of the extracellular water is contained in the middle lamella in the form of pectin gel, it seems logical to affirm that the T2 value of the extracellular water is lower than the T2 value of the water in cytoplasm.

Figure 2 T2 (A and B) and I(T2) maps (C and D) of Golden (A and C) and Reinette (B and D) apples.


Information about the water distribution in apples could also be obtained by MRI that is a very relevant method to spatially and quantitatively resolve T2 relaxation times13. Figure 2 shows the T2 and intensity maps processed using a mono-exponential model and recorded on Golden and Reinette apples. The heterogeneity of the apple tissues reflected by various contrasts is well demonstrated in these maps. A pattern of radial variation was observed, with a progressive increase in T2 values from the centre of the fruit (~300 ms) to the region near the cuticle (~500 ms). The T2 map of Reinette apple was the brightest in relation with highest T2 values. For the two cultivars, the highest T2 values were localized in the parenchyma under the cuticle of the apple. The vascular bundles were visible on each image and were characterized by weaker T2 values (~ 300 ms) and higher intensities (~ 1000u.a.) for the two cultivars. Vascular bundles are known to be composed of small cells and lignified cell walls, therefore characterized by lower T2 values than the surrounding tissues and the contrast should generally emphasize dry bundles. True and false fruits were distinguishable on each map but with difficulty on the T2 map of the Golden apple. For the Reinette apple, the true fruit seemed to be composed of two different tissues on T2 maps: a tissue with a weak T2 value (318±35ms) around the endocarp and a tissue with a higher T2 value (407±17ms) which separates the precedent tissue from the false fruit. These distinct tissues are not identifiable on Golden maps. Mean values of T2 and intensity were calculated on the basis of four fruits per variety and are exposed on Table 2.

Table 2. Mean values of T2 and intensity for Golden and Reinette apples measured in MRI

Variety T2 (ms) Intensity (u.a.)

True fruit False fuit True fruit False fuit Golden 297 ± 27 357 ± 27 840 ± 11 815 ± 27 Reinette 368 ± 23 445 ± 11 863 ± 14 829 ± 11

Golden true and false fruits have lower T2 values than Reinette fruits but globally their intensity were similar. The T2 map observed for Golden agreed with previous results from Barreiro et al.12a, 22, who have correlated lower T2 values with the apple mealiness, specific textural attribute that was waited for overripe Golden in the present study. With the aim of determining the contributions from the different subcellular water compartments (vacuole, cytoplasm and cell wall), the acquisition of multiple spin-echo MSE-MRI maps has been performed for each apple cultivar based on the recent sequence developed by Adriaensen et al.13. The optimum fitting of the MRI data has been performed using a tri-exponential function. It should be precised that the long echo train and short echo time required to cover the complete T2 range estimated using TD-NMR, was not possible to implement in MRI due to a lower number of more spaced echoes inherent to the technic. Therefore, the fast relaxing component was hindered in the MRI experiments by its short relaxation time and its small contribution to the total MRI signal. Table 3. T2 and relative intensity values obtained by MRI

Golden Delicious (20°C) Reinette d’Armorique (20°C)

Component T2 (ms) I (%) T2 (ms) I (%)

2 56 ± 35 5.5 ± 3.5 51 ± 22 4.8 ± 1.8 3 190 ± 70 17.2 ± 0.6 212 ± 48 17.8 ± 1.4 4 426 ± 36 77.4 ± 3.1 502 ± 18 77.5 ± 3.2


Mean values of each parameter measured by MRI are provided in Table 3. For each apple, the multi-exponential MRI signal consisted of one short T2 component (~ 50 ms) with low relative intensity close to 5 %, an intermediate T2 component (120 to 260 ms) with 17% of signal intensity and a long T2 component (400 to 520 ms) representing 77% of the signal. As for the mono-exponential fitting, overripe Golden displayed the lower T2(3) and T2(4) values.It was possible to compare NMR and MRI T2 results by sampling NMR samples at the same place that ROIs on T2 maps (see figure 1 in Materials and Methods). Compared to NMR results, the T2 values measured by MRI differed in particular for components (3) and (4) (Figure 3). This was already observed and interpreted by Adriaensen et al. 13 who have shown that the relationship between MRI and NMR values depended both on intrinsic T2 values and on characteristics of the heterogeneity of apple tissue in term of chemical exchange and microporosity13. Thus, the MRI T2 values were underestimated compared to the corresponding NMR T2 value and the phenomenon was amplified for high T2 values. This trend was confirmed in this study for the two apple cultivars with the biggest difference for the fourth component (Figure 3). The gap between MRI and NMR results was therefore of 752 ms and 685 ms for Golden and Reinette apples respectively. The relative intensities were almost similar for the two technics with a slight overestimation in MRI for the third component (IIRM 17.5% and IRMN 12.0%).

Figure 3. NMR-MRI I0 and T2 relationshi plots for Reinette and Golden parenchyma at 20°C.

Given these NMR and MRI results, it appeared that Golden and Reinette cultivars could only be distinguished on the basis of the T2(4) value obtained by MRI (mean values of 426 ms and 502 ms respectively). Indeed on the Figure 3, segregation between black and white points was visible on the y axis as the T2(4) of Golden samples seemed the most underestimated. The weak T2 values on T2 maps for the Golden variety was then mainly explained by a weaker T2(4) compared to the Reinette apples, but only measurable by MRI, very sensitive to susceptibility inhomogeneities. Thus the microporosity, large source of inhomogeneities, may play a key role to influence the T2 values. As already shown in tomatoes and apples, the more important the microporosity is the more underestimated should be the T2.13-14 Reconstruction of microporosity maps (Figures 4) has been performed based on a multigradient echo sequence (MGE) and record of T2*. Reinette and Golden parenchyma were characterized respectively by 23±1% and 26±1% of microporosity, thus confirming the lower T2(4) value measured for Golden apples. These values agreed with those obtained in different previous studies using MRI (Royal Gala apple – microporosity = 21±2%14), X-


microtomography (Braeburn apple – microporosity = 22±3%23) or accurate gravimetric approach (Braeburn apple – microporosity = 22±3%24).

Figure 4. Microporosity maps of Golden (A) and Reinette (B) apples.

3.2 Impact of the grinding and pressing processes

In the cider-making technology, the grinding of apple is realized in order to facilitate the extraction of apple juice while pressing. The milled materials obtained after grinding were slightly poorer in water (1-2%) than parenchyma because of the projection of a small quantity of water during the process. A way to understand the impact of grinding on the hydration status of cells is to study by NMR the water distribution in the milled materials. For the two cultivars, the data could be fitted using a tetra-exponential as for the intact parenchyma. Compared to intact parenchyma, the T2 values of milled materials differed significantly for the components (3) and (4), which decreased significantly from values close to 370 ms and 1200 ms for parenchyma to values around 220 ms and 730 ms in milled materials (Table 4). In the same time, the relative intensity of (4) decreased from 80% to 23-28% at the benefit of the relative intensity of (3) and (2).

Table 4 T2 and relative intensity values measured by NMR at 20°C on milled materials made with Golden and Reinette apples

Milled materials of Golden Delicious

Milled materials of Reinette d’Armorique

Component T2 (ms) I (%) T2 (ms) I (%)

1 14 ± 1 4.1 ± 1.5 13 ± 2 5.8 ± 1.6

2 73 ± 8 20.9 ± 2.3 68 ± 1 24.1 ± 1.8 3 230 ± 17 47.0 ± 3.6 225 ± 8 46.7 ± 1.8 4 731 ± 17 27.8± 3.9 734 ± 13 23.3± 1.5

These variations are illustrated on figure 5 (full bars for parenchyma and dotted bars for milled materials). They proved that water redistribution occurred during grinding with a probable migration of water from the vacuole toward cytoplasm, extracellular spaces and cell walls. These phenomena were very similar in Golden and Reinette apples. However it is likely that grinding treatment affected strongly the cell structure and then the initial compartments observed in parenchyma have certainly been changed. The previous T2 assignation 9 to specific water compartments is probably not reliable after grinding.


The reduction in T2 values was not so obvious using MRI as shown in figure 6 compared to figure 2. This result originated from the large heterogeneity of ground samples and the loosening of the subcellular structure. The grinding achieved, the milled materials underwent a pressing process to extract the apple juice. The pressed materials have been analyzed by NMR taking different samples at the top, middle and bottom of the “apple cake” which was 8 cm height. No difference in water content neither T2 relaxation times have been found between these samples. Therefore mean values were calculated and presented in table 5, together with the four T2 components measured by TD-NMR. The pressing process induced a decreasing of T2 values for components (2), (3) and (4) except for T2(4) of Golden which remained relatively constant even after pressing. The main changes observed on relative intensities is the loosening of 26-27% for the component (3) and 3-7% for the component 4 while the components (2) and (1) increased of 20-27% and 6-8% respectively. Figure 5 illustrates the comparison between T2 values and relative intensities measured on apple parenchyma, milled materials and pressed materials. The T2 value of components (2), (3) and (4) were found lower for Reinette which showed a difference of 152 ms for the fourth component compared to Golden. For Reinette, the relative intensity of T2(2) was slightly weaker at the benefit of the first component.

Figure 5. T2 and relative intensity values of the four components measured by TD-NMR for Golden (black frame) and Reinette (grey frame) for parenchyma (full bars), milled materials (dotted bars) and pressed materials (hatched bars)


Figure 6. T2 maps of milled materials from Golden (A) and Reinette (B) apples

Table 5. Water content, T2 and relative intensity values of pressed samples for Golden and Reinette.

Variety Pressed materials of Golden Delicious

80.3 ± 1.1

Pressed materials of Reinette d’Armorique

80.7 ± 0.8 Water content (%) Components T2 (ms) I (%) T2 (ms) I (%)

1 15 ± 2 10.2 ± 1.0 14 ± 3 13.9 ± 2.6 2 53 ± 6 47.5 ± 3.1 43 ± 5 43.8 ± 2.2 3 157 ± 17 21.1 ± 2.2 133 ± 19 22.3 ± 3.0 4 720 ± 45 21.2 ± 2.1 568 ± 49 20.0 ± 2.2

At this stage of the process it is difficult to envisage the cellular integrity and even to consider the subcellular compartmentation of water. However, by using the assignment proposed by Sibgatullin et al.9, the pressing process seemed to enrich the second component assigned to the extracellular spaces. Therefore at the end of the pressing process, about 50-45% of the water remaining in the milled materials seems to be localized into the extracellular spaces. As mentioned in the experimental section, the Golden apples were overripe compared to Reinette. This protocol has been chosen to get Golden apples more difficult to press. However the pressing yield (around 50%) as well as the final water content in pressed materials (80-81%) was nearly the same for the two cultivars. The reason why T2(4) was higher for overripe Golden is actually under investigation. Our objective is to link this particular water distribution with cellular deterioration (microscopy) but also with the chemical composition (sugar, ions…) and the mechanical properties of the pressed materials.

4 CONCLUSION

We demonstrated in this study that TD-NMR and MRI are reliable tools providing information about the heterogeneous structure of apple parenchyma tissue as well as the subcellular water distribution at the origin of the multi-exponential T2 signal measured. The four relaxation T2 components determined by TD-NMR were linked to cell compartments characterized with different water T2 values and relative intensities. These values were interpreted using the known subcellular water content and the modification of the water distribution between the compartments due to grinding and pressing. Therefore, while the grinding process induced the water evacuation from the vacuole to the rest of the cell,

A B


increasing the hydration state of cell walls, the pressing seemed to fulfil the extra-cellular part of tissues. The TD-NMR measurements have been performed on mature Reinette and overripe Golden apples with the aim of getting distinct textural properties regarding the pressing process. The T2 values indicated that the distinction between the two apple cultivars was very difficult using TD-NMR while T2 maps, processed using a mono-exponential fitting, were in agreement with lower T2 values for overripe Golden, in coherence with the higher mealiness of these apples. However, the relation between mealiness and NMR T2 values was not so obvious if we consider the contribution of the fruit apparent microporosity, a large source of inhomogeneities which induced decreasing of T2 measured in MRI13. On the other hand, when milled materials of apple were pressed to extract the apple juice, the longer T2 value was higher for overripe Golden apples. Our results indicated that complementary investigations should be necessary using biochemical quantitative analyses and mechanical tests to understand the water distributions in pressed materials.

Acknowledgements The authors are indebted to the Regional Council of Britany and Rennes Métropole for financial support (OPTIPRESS project). They also acknowledge the Pôle Agronomique Ouest (PAO, Rennes, France) for project coordination and the Agro-scans laboratory (PRISM Research Platform, Rennes, France) for access to the NMR facilities (ISO 9001 certified).

References

1 Lequéré, J.-M.; Bauduin, R.; A., B., Elaboration des jus de pomme et des cidres. Techniques de l’Ingénieur 2010, 1-12.

2 (a) McLellan, M. R.; Kime, R. L.; Lind, L. R., Electroplasmolysis and other treatments to improve apple juice yield. Journal of the Science of Food and Agriculture 1991, 57 (2), 303-306; (b) Gerard, K. A.; Roberts, J. S., Microwave heating of apple mash to improve Juice yield and quality. Lebensmittel-Wissenschaft Und-Technologie-Food Science and Technology 2004, 37 (5), 551-557.

3 Turk, M. F.; Vorobiev, E.; Baron, A., Improving apple juice expression and quality by pulsed electric field on an industrial scale. LWT-Food Sci. Technol. 2012, 49 (2), 245-250.

4 Ribeiro, D. S.; Henrique, S. M. B.; Oliveira, L. S.; Macedo, G. A.; Fleuri, L. F., Enzymes in juice processing: a review. International Journal of Food Science and Technology 2010, 45 (4), 635-641.

5 (a) Hills, B. P.; Duce, S. L., The influence of chemical and diffusive exchange on water proton transverse relaxation in plant tissues. Magnetic Resonance Imaging 1990, 8 (3), 321-331; (b) Van As, H., Intact plant MRI for the study of cell water relations, membrane permeability, cell-to-cell and long distance water transport. J. Exp. Bot. 2007, 58 (4), 743-756; (c) Van As, H.; van Duynhoven, J., MRI of plants and foods. Journal of Magnetic Resonance 2013, 229, 25-34.

6 (a) Snaar, J. E. M.; Vanas, H., Probing water compartments and membrane-permeability in plant-cells by H-1-NMR relaxation measurements. Biophysical Journal 1992, 63 (6), 1654-1658; (b) Hills, B. P.; Remigereau, B., NMR studies of changes in subcellular water compartmentation in parenchyma apple tissue during drying and freezing. International Journal of Food Science and Technology 1997, 32 (1), 51-61.

7 (a) Marigheto, N. A.; Moates, G. K.; Furfaro, M. E.; Waldron, K. W.; Hills, B. P., Characterisation of Ripening and Pressure-Induced Changes in Tomato Pericarp Using NMR Relaxometry. Applied Magnetic Resonance 2009, 36 (1), 35-47; (b) Musse, M.;


Cambert, M.; Mariette, F., NMR Study of Water Distribution inside Tomato Cells: Effects of Water Stress. Applied Magnetic Resonance 2010, 38 (4), 455-469.

8 Mariette, F.; Brannelec, C.; Vitrac, O.; Bohuon, P. In Effet du procédé de friture sur la répartition et l'état de l'eau mesurée par RMN et IRM, Les produits alimentaires et l'eau, Agoral 99, Nantes, Edition Tec & Doc: Nantes, 1999.

9 Sibgatullin, T. A.; Anisimov, A. V.; de Jager, P. A.; Vergeldt, F. J.; Gerkema, E.; Van As, H., Analysis of diffusion and relaxation behavior of water in apple parenchymal cells. Biofizika 2007, 52 (2), 268-276.

10 (a) Van As, H., MRI of water transport in intact plants: characteristics and dynamics. Comparative Biochemistry and Physiology a-Molecular & Integrative Physiology 2006, 143 (4), S42-S42; (b) Van As, H.; Scheenen, T.; Vergeldt, F. J., MRI of intact plants. Photosynth. Res. 2009, 102 (2-3), 213-222.

11 (a) Van der Weerd, L.; Claessens, M.; Efde, C.; Van As, H., Nuclear magnetic resonance imaging of membrane permeability changes in plants during osmotic stress. Plant Cell and Environment 2002, 25 (11), 1539-1549; (b) van der Weerd, L.; Claessens, M.; Ruttink, T.; Vergeldt, F. J.; Schaafsma, T. J.; Van As, H., Quantitative NMR microscopy of osmotic stress responses in maize and pearl millet. J. Exp. Bot. 2001, 52 (365), 2333-2343.

12 (a) Barreiro, P.; Ortiz, C.; Ruiz-Altisent, M.; Ruiz-Cabello, J.; Fernandez-Valle, M. E.; Recasens, I.; Asensio, M., Mealiness assessment in apples and peaches using MRI techniques. Magnetic Resonance Imaging 2000, 18 (9), 1175-1181; (b) Hernandez-Sanchez, N.; Hills, B. P.; Barreiro, P.; Marigheto, N., An NMR study on internal browning in pears. Postharvest Biol. Technol. 2007, 44 (3), 260-270; (c) Melado, A.; Barreiro, P.; Val, J.; Blanco, A.; Ruiz-Cabello, J.; Rodriguez, I., Non destructive assessment of watercore in apples using MRI. disorder detection with HR-MAS. 2010; p 178; (d) Chen, P.; McCarthy, M. J.; Kauten, R., NMR for Internal Quality Evaluation of Fruits and Vegetables. Transactions of the ASAE 1989, 32 (5), 1747-1753; (e) Tu, K.; De Baerdemaeker, J.; Deltour, R.; De Barsy, T., Monitoring post-harvest quality of Granny Smith apple under simulated shelf-life conditions: Destructive, non-destructive and analytical measurements. Int J Food Sci Tech 1996, 31 (3), 267-276.

13 Adriaensen, H.; Musse, M.; Quellec, S.; Vignaud, A.; Cambert, M.; Mariette, F., MSE-MRI sequence optimisation for measurement of bi- and tri-exponential T2 relaxation in a phantom and fruit. Magnetic Resonance Imaging (0).

14 Musse, M.; De Guio, F.; Quellec, S.; Cambert, M.; Challois, S.; Davenel, A., Quantification of microporosity in fruit by MRI at various magnetic fields: comparison with X-ray microtomography. Magnetic Resonance Imaging 2010, 28 (10), 1525-1534.

15 (a) Musse, M.; Quellec, S.; Cambert, M.; Devaux, M. F.; Lahaye, M.; Mariette, F., Monitoring the postharvest ripening of tomato fruit using quantitative MRI and NMR relaxometry. Postharvest Biol. Technol. 2009, 53 (1-2), 22-35; (b) Shaarani, S. M.; Cardenas-Blanco, A.; Amin, M. H. G.; Soon, N. G.; Hall, L. D., Monitoring Development and Ripeness of Oil Palm Fruit (Elaeis guneensis) by MRI and Bulk NMR. International Journal of Agriculture and Biology 2010, 12 (1), 101-105.

16 Shaarani, S. M.; Nott, K. P.; Hall, L. D., Combination of NMR and MRI quantitation of structure and structure changes for convection cooking of fresh chicken meat. Meat Sci. 2006, 72 (3), 398-403.

17 Kotwaliwale, N.; Curtis, E.; Othman, S.; Naganathan, G. K.; Subbiah, J., Magnetic resonance imaging and relaxometry to visualize internal freeze damage to pickling cucumber. Postharvest Biol. Technol. 2012, 68, 22-31.


18 Marcone, M. F.; Wang, S. A.; Albabish, W.; Nie, S. P.; Somnarain, D.; Hill, A., Diverse food-based applications of nuclear magnetic resonance (NMR) technology. FoodResearch International 2013, 51 (2), 729-747.

19 Marquardt, D. W., Algorithm for least-squares estimation of non-linear parameters. Current Contents/Engineering Technology & Applied Sciences 1979, 27, 14-14.

20 Mariette, F.; Davenel, A.; Marchal, P.; Chaland, B., A study of water by NMR and MRI in dairy processes. Rev. Inst. Fr. Pet. 1998, 53 (4), 521-525.

21 Redgwell, R. J.; MacRae, E.; Hallett, I.; Fischer, M.; Perry, J.; Harker, R., In vivo and in vitro swelling of cell walls during fruit ripening. Planta 1997, 203 (2), 162-173.

22 Barreiro, P.; Ruiz-Cabello, J.; Fernandez-Valle, M. E.; Ortiz, C.; Ruiz-Altisent, M., Mealiness assessment in apples using MRI techniques. Magnetic Resonance Imaging 1999, 17 (2), 275-281.

23 Mendoza, F.; Verboven, P.; Mebatsion, H. K.; Kerckhofs, G.; Wevers, M.; Nicolai, B., Three-dimensional pore space quantification of apple tissue using X-ray computed microtomography. Planta 2007, 226 (3), 559-570.

24 Drazeta, L.; Lang, A.; Hall, A. J.; Volz, R. K.; Jameson, P. E., Air volume measurement of 'Braeburn' apple fruit. J. Exp. Bot. 2004, 55 (399), 1061-1069.

Foodomics

A 1H NMR-BASED METABOLOMIC APPROACH ON DIETARY BIOMARKER RESEARCH IN HUMAN URINE

Alessia Trimigno1, Gianfranco Picone1, Francesco Capozzi1

1 Department of Agro-Food Science and Technology, University of Bologna, Piazza Goidanich 60, 47521, Cesena (FC), Italy

1 INTRODUCTION

The term 'metabolomics' is frequently used to define “the comprehensive analysis of the whole metabolome under a given set of conditions” [1]. This definition implies that this 'omic' science is capable of giving an overview of the whole metabolic profile of a specific biological matrix, to wit its 'metabolome'. The analysis of the so-called metabolome has become particularly important in human studies, allowing the identification of typical and atypical metabolites in determined biofluids and tissues. This is clearly fundamental for medical investigations, since it could be possible to assess the presence or absence of certain diseases and syndromes or risk conditions. Another widespread use of the metabolomic approach is the research on specific dietary effects on the metabolic profile. Nowadays, nutrition aims to find new techniques in order to assess the real benefits and drawbacks of determined dietary patterns and to advice people for the best nutritional intake possible. The employment of metabolomic techniques has been proved useful in the assessment of the effects of particular dietary patterns on the molecular profile of human biofluids, especially urine. This was done both in intervention studies and observational studies. Obviously, in intervention studies it is easier to assess the direct effect of the consumption of certain foods or food groups, though in real life the conditions are quite different. Therefore, it is necessary to take the research further and investigate the presence of determined food molecules in fluids or tissues from individuals without a controlled dietary intervention. Many molecules were proven to be related to the consumption of particular foods, especially in urine, since this biofluid is influenced at the larger extent by the dietary intake, because it resents less of homeostasis, unlike other biological samples such as plasma. These molecules are often referred to with the term “biomarkers”, defined as “characteristic that is objectively measured and evaluated as an indicator of normal biological processes, pathogenic processes or pharmacological responses to a therapeutic intervention” (National Institute of Health – NIH – Biomarkers Definitions Working Group). [2] In this field, they are specific metabolites occurring in a biofluid in relation to the acute or chronic intake of specific foods or to their metabolism by human and the hosted gut microflora. Diet is characterised by the concomitant ingestion of many different types of food and may be more or less variable day by day. This variance alters the metabolic profile of biofluids: as it was said, some metabolites may be contained in a more constant amount and others may


vary highly in their relative concentrations. To understand the possible effect of a certain dietary patterns, it is firstly necessary to make sure that the individuals are aware of their diet. In effect, the information given by means of nutritional questionnaires, such as 24 hour recalls (food diaries) and FFQs (Food Frequency Questionnaires) might not be completely reliable. For example, McKeown et al. [3] demonstrated that the correlation between 24h nitrogen excretion and its dietary intake from FFQ was as low as 0.25. Metabolomics, can therefore play a useful role in obtaining reliable robust data by proving, with the presence of specific biomarkers in the analysed biological samples, the compliance of people investigated to the given diet or the actual correspondence to the stated diet.

2 1H-NMR SPECTROSCOPY

Non-targeted spectroscopic techniques, such as NMR or MS (mass spectroscopy) are usually employed in metabolomic studies as they allow the direct and quick observation and discovery of many biomarkers. Proton Nuclear Magnetic Resonance (1H-NMR) has some advantages with respect to MS. It does not require any prior knowledge about the chemistry of the compounds present in the analysed biofluids and requires minimal sample preparation, with a consequent very little sample manipulation. 1H-NMR spectroscopy is also very quick and can function with small sample volumes [4], allowing the simultaneous detection of all metabolites, present at levels above the sensitivity limits, independently of their different physical and chemical properties, like hydrophobicity or acidity [5]. It is, though, less sensitive than MS and can show up only molecules appearing in concentrations greater than 0.1 mmol/L, a concentration which is more than sufficient for metabolomic studies. However, the low sensitivity weakness is more than compensated by the possibility to combine different NMR experiments to discover the chemical structure of unknown compounds, even though the corresponding standards are not available. Usually, indeed, at first a 1-D experiment is employed, generally using a standard water presaturation pulse sequence. This especially hold for urine samples, which do not require any special sequence for suppression of broad signals due to macromolecules, such as the CPMG (Carr–Purcell–Meiboom–Gill) spin-echo pulse sequence, mainly used for spectra acquisition on plasma. Through these 1-D techniques it is possible to acquire spectra with many different signals arising from the molecules present in the sample. The standard protocol requires the application of an exponential line-broadening multiplication, zero filling and Fourier transform of the free induction decays (FIDs). Spectra are then adjusted for phase and baseline distortions, as well as calibrated to the frequency of a reference signal (usually TSP at 0.0 ppm). After these pre-statistical processing procedures, spectra are generally normalised to account for dilution factors. This is nowadays most frequently done by means of a probabilistic quotient normalisation (PQN) algorithm. [6] After normalisation, the whole set of spectra may undergo multivariate statistical analysis. When necessary, signal misalignments due to pH effects are corrected by algorithms that recursively proceed with a segment-wise peak-alignment procedure. [7, 8] In order to find spontaneous sample clustering, and to highlight possible outliers, untargeted multivariate techniques like principal component analysis (PCA) are employed, evidencing patterns of covariant molecules expressed as orthogonal principal components. After the recognition of contingent patterns in the PC space, supervised multivariate techniques, such as PLS or PLS-DA, are applied to evidence hindered spectral differences among sample categories.

Foodomics 145

Partial Least Squares (PLS) technique is a method employing linear regression to correlate X, a set of predictor variables, to Y, the independent variables. PLS can deal with highly collinear data like spectral matrices. Whilst PCA finds a subspace containing the maximum variance of X, PLS aims to obtain a small subspace describing X and predicting Y. Usually, in metabolomics problems, PLS is used to classify samples. Class labels will function as the independent variable vector Y and this method is called PLS-DA, where DA stands for 'Discriminant Analysis'. PLS-DA will try to improve group separation among samples using the information contained in the class vector. Since PLS-DA is supervised, though, there is a risk of overfitting, therefore it's usually necessary and advisable to employ validation techniques such as cross-validation and evaluate the performance of the models for example by means of the number of misclassifications. Once the statistical techniques have assessed groupings and claimed the differences among patterns to be due to particular spectral regions and integrals, it will be necessary to assign those peaks to the corresponding molecules, in order to understand their origin and cause. Generally, molecular assignments is performed by consultation of literature or of online or software databases (e.g. The Human Metabolome Database, HMDB, hmdb.ca). When there is no information in previous studies or in databases about the resonances found, the best way to understand their origin is through 2-dimensional NMR experiments. These can be of various types, such as total correlation spectroscopy (TOCSY), two-dimensional 1H-1H correlation spectroscopy (COSY), heteronuclear single quantum coherence (HSQC) or 2-D J-Resolved (JRES) homonuclear 1H spectra. These techniques use various pulse sequence and can show the correlation of various resonances, highlighting their molecular formula and dimension.

3 URINE

Metabolomics employing 1H-NMR spectroscopy usually analyses biological fluids in order to highlight a typical human metabolic profile. The most employed biofluids are blood, urine and plasma, though also body tissues and vaginal mucus were analysed. Urine and plasma samples are complementary in giving a global image of the metabolic state of an organism, giving insight, for example, on underlying patterns of metabolic processes controlled by many different types of cells and tissues. Plasma is a representation of the metabolic content of the sample at the exact moment of its collection, thus it illustrates the lipoprotein content and the concentration of metabolites controlled by our bodies. Urine, instead, gives an average pattern of polar molecules, excreted by our bodies in variable concentrations in relation to homeostatic control. [9] Urine is therefore more complex, with a greater metabolites variability. [10] This type of biofluid, in addition, has less tecnical problems during NMR spectra acquisition. Its main problem is dilution, thus freeze-drying and subsequent dissolving in deuterated buffer is the general approach to concentrate this type of sample, in order to acquire spectra more quickly. [11] Other advantages rising from the use of urine instead of plasma are the facts that it can be collected promptly and non-invasively. [12] Its variability in composition is very important, since this can be influenced by many different environmental factors such as daily variation or oestral cycle in women. Moreover, humans show a greater intersubject metabolic variation than that found in lab animals, due to a higher number of genetic and external factors. A way to reduce some of these external influences is the following of accurate interventions on diets. Later, specific statistical techniques like chemometrics are employed to determined precisely the slight metabolic effects related to changes in diet and nutrition. [13]


4 NUTRIMETABONOMICS STUDIES

Nutritional metabonomics or nutrimetabonomics is more and more employed for research on molecular interactions between diet and the global metabolic system. The study of nutritional data, later associated and related to metabolomic data regarding biofluids, is usually done through the investigation of partecipants' diets by means of nutritional diaries (24 hour recalls) or Food Frequency Questionnaires (FFQ). The approach of researching for biomarkers proving the intake of a determined dietary component or pattern can be applied, in effect, both in interventional and observational studies, where subjects can lead free, ad-libitum diets. In epidemiological studies carried out on a large number of subjects, where dietary habits are collected through FFQ or 24 hour recalls, it is possible to use metabolomics to evaluate also nutritional data.

Figure 1 The metabolomic approach in nutritional studies. Through the investigation of biofluid spectra, it is possible to assess the presence of particular food biomarker and validate dietary questionnaires.

In studies instead focusing on the research of a particular biomarker that might prove the adherence to a determined intervention diet or the chronic intake of a specific food component, nutritional intervention and metabolic phenotyping must be employed and biomarkers must be validated by the use of a large number of subjects studied. Interventional research has been obviously the most employed, since effects can be shown more easily, avoiding a lot of noise from the ingestion of many different types of food. The majority of these studies were performed on healthy adult individuals, whilst just some were carried out on healthy children, cancer patients, metabolic sydrome or hypertensive subjects or pregnant women. The study duration varied between just two days to 8 months and urine was the most employed sample, for the reasons previously stated. The metabolomic technique mainly employed was H-NMR spectroscopy, again for its demonstrated advantages. [14]

Foodomics 147

5 FOOD-DERIVED BIOMARKERS

5.1 Intervention Studies

Many recent studies have been carried out using both non-targeted metabolome fingerprinting or metabolite profiling in order to assess the presence of food biomarkers in short-term and acute dietary intervention researches.

The principles of this type of study are the following: acute exposure to determined food products and collection of post-prandial urine non-targeted metabolic profiling with NMR techniques employment of data-mining techniques such as PCA to determine if subjects can be discrimined by the intake or not of the specific food consumed identification of metabolites discriminating among subjects and investigation on their relation with the researched food product

[15]

NMR spectroscopy has been proved effective in many nutrimetabonomics intervention study. The first approach is to use one-dimensional techniques to acquire general spectra of biofluids. Usually urine is employed, for the aforementioned reasons and for the fact that it doesn't require any special suppression sequence for large molecules like proteins and lipoproteins, thing instead necessary for plasma samples, which are generally analysed by means of the CPMG pulse sequence. For this reason, usually a NOESYPRESAT sequence is employed with urine. Vàzquez-Fresno et al. [16] employed this pulse sequence in order to investigate the presence of wine biomarkers in urine samples. Urine spectra were then compared by ANOVA with a post-hoc Fisher's LSD test and Pearson's correlation. By means of the ANOVA some spectral bins were found statistically different among the groups (wine, dealcoholised wine and gin intake). This research demonstrated the existence of different types of biomarkers from wine consumption: those from direct food composition (mannitol and tartrate) and those from endogenous alteration due to wine intake (BCAA metabolites). Together with these, gut microbiota markers were found and showed peculiar effects due to alcohol intake, highlighting the need for further research to understand the consequences of this consumption to the assimilation of polyphenols.

Many studies have also employed firstly one-dimensional pulse sequences to highlight possible differences between control and treated individuals. Those studies usually took visual inspection of spectra to see if major differences were already evident. [17, 18, 19, 8] In all of these cases some slight differences were evident among spectra due to the dietary intervention: Van Dorsten et al. [19] showed changes in the profile of hippurate after tea consumption, whilst Edmand et al. [8] saw that four singlet peaks were accountable for differences between a high and a low cruciferous vegetable diet, though most of the spectral difference was accountable to interindividual effects. Another possible approach is the one employed by Lehtonen et al. [20], in which urine spectra were visually compared to the ones of the food extract, in this case of lingonberries, in order to find out if already metabolic similarities were clear, but this was not the case. Personal fingerprint is always most accountable for the differences in metabolic profiles, so its effect is generally mitigated by investigating all the spectral variations direction by means of unsupervised multivariate data analysis techniques statistical, in particular by PCA. In effect, in all the aforementioned studies the first data-mining technique applied was principal component analysis. This method allows to highlight the possible presence of


outliers (e.g. spectra with baseline problems or interfering substantial signals like ethanol) and of samples clustering due to external factors to be removed (e.g. gender effects) or to the desired dietary effects. Wang et al. [18] for example, found through PCA that one specific food metabolite was influencing PCA scoreplots (i.e. TMAO after fish consumption) and this was unrelated to the dietary intervention, which focused on chamomile, thus it needed to be removed. They also found out a gender effect on the scoreplots and this factor also necessitated removal in order to mainly focus on the treatment effects. Van Dorsten et al. [19], instead, after removing the resulting PCA outliers, saw how this technique already highlighted clusters for the 3 different treatments analysed (black tea, green tea and control with caffein). When PCA shows groupings, it is possible to understand which molecules are responsible for this clustering through the inspection of the loadings. For example, Van Dorsten et al. [19] saw that the clusterings were caused by an increase in hippurate, citrate and glucose. Edmands et al. [8] found out that the four singlets already showing differences in spectral visualisation were also the main loadings responsible for sample groupings in PCA scoreplots. After a first inspection through PCA, supervised techniques such as PLS or PLS-DA are usually employed, as in all of the reported studies. In some cases these methods are applied to the whole spectra, whilst in other cases it is advisable to just concentrate on some particular regions that have less noise and are supposed to show more differences. This was the case for the research of Van Dorsten et al. [19], where PLS-DA was performed in the aliphatic and aromatic region of the spectra, to further discriminate among samples. Here, in the aromatic region the consumption of tea, either green or black, showed a distinct increase in hippuric acid in contrast with caffeine treatment spectra. In addition, also 1,3-dihydroxyphenyl-2-sulphate was proven to be due to tea consumption. For what concerned the aliphatic region, instead, where signals from endogenous metabolites are found, there were differences between green and black tea like a greater increase in dimethylamine, citrate, glycine, N-acetyl glycoproteins and pyruvate in green tea and a higher concentration of beta-hydroxybutyrate, glucose and methionine in black tea. In many cases, though, the observable differences are related to resonances which haven't been previously assigned in literature or in databases. In this cases the usual approach is to perform bidimensional NMR techniques in order to gain information on the molecular composition and structure of those unknown metabolites. One of the most applied technique is TOCSY, total correlation spectroscopy, which visualises cross peaks of coupled protons, even if the respective nuclei are connected by a chain of couplings. TOCSY is therefore very important in the identification of spin couplings networks and employs repetitive pulse series in the mixing period, in order to cause isotropic mixing. If this is longer, the polarisation will spread to a greater number of bonds. Winning et al., for example, used this technique to identify an unknown biomarker for onion intake. In this way, they assessed that the unassigned resonance was probably of the CH3 protons in dimethyl sulfone. This molecule derives from the oxidation of dimethyl sulfoxide and it was previously proved to be deriving from dietary sources, gut microbiota metabolism and methanethiol metabolism. A variation of the TOCSY method is the so-called STOCSY (Statistical Total Correlation SpectroscopY) [21] which is more often used to identify metabonomic biomarkers. This technique uses the multicollinearity of the intensity variables in a spectral matrix in order to create pseudo-2-D spectra which represent the correlations of the peak intensities in the sample.

Foodomics 149

This is one of the 2-D techniques employed by Edmands et al. [8] In order to assign the peaks of the unknown biomarker for cruciferous vegetables consumption. In this way they found out that the resonances were consistent with low molecular weight coumpounds containing sulfur or nitrogen and chirally centered. These might rise from SMCSO, a sulfur containing amino acid typical of CVs, and its metabolites. To further prove this hypothesis they synthetised the molecule and analysed it with the same 2-D NMR techniques, also after spiking into human urine. Some resonances, though, still remained unassigned, but since they were highly correlated to this molecule, they are probably related to its metabolites. SMCSO, also known as methiin, has been reported to have some biological activity (i.e. antihyperlipidemic, antidiabetic, antimicrobial, antigenotoxic), thus this might be use as an effective marker for CV consumption and for the investigation of its properties in epidemiological studies. Another 2-D technique which is frequently used in this type of study is COSY (Correlation SpectroscopY). As the name says, it can help identifying coupled spins, by using a single RF pulse followed by and evolution time, a second pulse and a measurement time. The resulting spectra is usually calculated for a single isotope, which is mostly hydrogen, and shows its frequencies along both axes. Two types of peaks are visualised by COSY: the diagonal peaks and the cross peaks. The first ones show the same frequency in both axis, thus appear along the graph diagonal and correspond to the 1-D experiment peaks; the second ones, instead have different frequency in the two axis and so are not on the diagonal, they correspond to couplings of nuclei pairs. Using cross peaks it is therefore possible to find out which atoms are connected. 2D homonuclear 1H-1H COSYwas for example employed by Solanky et al. [13] to further confirm the metabolites responsible for discrimination between a control diet, a soy protein diet and a miso diet. Other 2-D techniques, instead, employ heteronuclear through-bond correlation methodologies, which give rise to signals due to the coupling between different nuclei, for example, the most used are 1H and 13C. One of the most widespread heteronuclear techniques is HSQC (Heteronuclear single-quantum correlation spectroscopy). This method can find correlation between different nuclei separated by just one bond. The resulting spectra shows one peak for each pair of coupled nuclei, with coordinates formed by the chemical shift of those atoms. This technique transfers magnetisation from the first nucleus (1H usually) to the second one (13C generally) with a special pulse sequence called INEPT. Then the magnetisation evolves and transfers back to the first nucleus in order to be observed. This method was used by Winning et al. [22] and confirmed the previously stated TOCSY findings for the assignment of an unknown molecule. It was also employed by Edmands et al. [8] as one of the 2D techniques employed to assign unknown resonances to SMCSO, as before stated. In some cases, though, researches prefer to couple 1D NMR techniques to other metabolomics strategies such as mass spectrometry. Daykin et al. [23], who researched on the effects of black tea consumption on the metabolic profile, coupled NMR to HPLC to analyse the urine sample that contained most of the unassigned signals. Since this method was not that successful, they then also linked the two methodology with MS (HPLC-NMR-MS) in order to find the molecular components for these signals. Combining the information from the three techniques it was then possible to assign it to 1,3-dihydroxyphenyl-2-O-sulphate, a metabolite probably derived by enzymatic conjugation of sulphate and pyrogallol, thus deriving from black tea compounds.


5.2 Observational Studies

Other studies have focused on the analysis of dietary data and their link to metabolic profiles. In most of these cases, therefore, subjects analysed were able to eat freely-chosen diets. Using the nutritional meta-data (24 h recalls or FFQ), it was then possible to divide individuals in classes in relation to the habitual consumption of specific foods. This will be then useful to find possible biomarkers for the habitual consumption of determined food products or for a particular dietary pattern. The intake of specific foods can be grouped in patterns, summarising some exposure categories from products very rarely consumed to daily employed foods, such as coffee. In this studies, in order to create a robust model for specific pattern biomarkers, it is necessary to have many subjects analysed. Once subjects are divided into higher vs. lower frequency of consumption for specific food products, multivariate data analysis and classification can be employed. For example, one of the most employed techniques to find potential dietary biomarkers is the assessment of the value for class discrimination using random forest margin values. Another widespread methodology is the one using area under the receiver operating characteristic curve values in order to classify samples in a robust and reliable manner. Using these techniques, it is then possible to evaluate the classification efficiency for each food product in analysis: more frequently consumed foods, obviously, will show better results. These products can then be used to find possible biomarkers for their habitual ingestion [15] analysing biofluids from the subjects examined, especially by means of NMR spectroscopy. In this case, thus, NMR becomes a supporting method to prove the robustness of the dietary information obtained by traditional techniques. For these studies, usually 1-D common NMR techniques are employed, followed by multivariate data analysis. In 2011, O'Sullivan et al. [24] tried to highlight dietary patterns and their link to specific metabolic profiles. Three-days estimated dietary records were employed to assess food intake. Then foods were grouped in 33 classes, using previous data, relating to nutrients and cooking use. Cluster analysis was then carried out on standardised food intake values in order to visualise dietary patterns. Three main clusters were found: one characterised by a healthier diet, one, on the contrary, with unhealthy nutritional characteristics and the last one which could be identified as a traditional Irish nutrition. Multivariate data analysis (PCA and PLS-DA) was carried out on urine NMR spectroscopic data to find outliers and aggregations among samples, due to the dietary clusters found. In this way it is possible to see many different metabolites and link them to the related food intake. It was shown how, for example, levels of phenylacetylglutamine were positevly linked to vegetable intake, particularly higher in the first dietary cluster, to with the 'healthy' one. The same was the case with urinary O-acetylcarnitine concentration and red-meat consumption, typical of the traditional Irish diet pattern. These two might therefore be considered potential biomarkers for specific food intakes. Phenylacetylglutamine, in addition, derives from the conjugation of phenylacetyl-coenzyme A and glutamine: the first might derive from dietary intake of phenylacetate, present especially in plant-derived foods. Other studies showed population-based nutrimetabolomics differences in urine, stating how subjects from different geographical origins have specific metabolic profiles. Analysing urine from four different populations (China, Japan, USA and UK), by 1H-NMR spectroscopy, two different patterns were found and it was hypothesised that this was due to specific dietary patterns, though this needs to be proved further by food consumption information. [25] This showed how epidemiological studies can show by NMR spectroscopy many information on

Foodomics 151

the analysed subjects, as far as showing also in untargeted analysis difference dietary patterns.A study from Lenz et al. [26] also mimicked a typical clinical study population which is not required to follow a restrict diet. 1H-NMR spectra of urine were acquired using a NOESYPRESAT pulse sequence. In this case visual inspection, especially, and PCA showed distinct outliers, due, for example, to the consumption of paracetamol. This proves again how NMR can easily highlight non-compliance to the study requirements (abstinence from common drugs). In this research, unconventionally, more outliers were found from spectral visual inspection: in effect, some people which consumed fish (highlighted by TMAO presence in spectra) or chewing gums (evident from mannitol resonances) were found only by the visualisation of spectra, whilst they remained clustered with other participants in PCA scoreplots. This highlights once again the great importance and capability of NMR techniques in giving evidence of many dietary and environmental information about study participants, proving its reliability for this type of studies. Another part of this study tried to find differences in two population groups: one Swedish and one British. This investigation found a typical dietary biomarker for the Swedish diet: TMAO due to fish consumption and also saw how a particular high-protein dietary regimen (Atkins diet) influences the metabolome with a distinct signal for taurine. Yang et al. [27] tried a different generic approach coupling SPE fractionation with both mono- and bi-dimensional NMR techniques to analyse human urine. Five SPE fractions were obtained and analysed by the following NMR sequences: 1D-NOESY, 1H-1H COSY 2D, 1H-1H TOCSY 2D, 1H-13C HSQC 2D and 1H-13C HMBC 2D. This last 2D technique is similar to HSQC and stands for Heteronuclear Multiple-Bond Correlation spectroscopy. It can detect the correlation between pairs of different nuclei over larger bond ranges (2-4 bonds), whilst HSQC was focusing on single-bond separation. The combination of all these NMR techniques allowed the recognition of more than 70 metabolites, mostly in the first fraction, corresponding to polar molecules, of which urine is particularly rich. This experiment thus gave a practical methodology to identify different metabolites in urinary specimens and assign them by means of various NMR techniques, proving how this spectroscopic strategy can quickly allow an insight of the metabolic profile of complex matrices.

Type of study Food consumed References Observational study vegetables O'Sullivan et al.

Short term intervention study cruciferous vegetables Edmands et al.

Short term intervention study tea Van Dorsten et al.

Short term intervention study; Acute intervention study

black tea Van Dorsten et al.;

Daykin et al.

Short term intervention study green tea Van Dorsten et al.

Observational study meat O'Sullivan et al.; Bertram et al. [28] Short term intervention study fish Wang et al.

Observational study chewing gums Lenz et al.

Short term intervention study milk Bertram et al. 2007 [28]

Short term intervention study fruits Heinzmann et al.

Short term intervention study dark chocolate Martin et al., 2009 [29]

Short term intervention study high protein diet Rasmussen et al., 2012 [30]

Table 1 Some of the reported studies and the relative foods of which biomarkers can be spotted.


6 CONCLUSIONS

In recent years, metabolomics approaches have been increasingly applied to nutritional studies in order to investigate the possible links between determined food consumption and metabolic and health state. In this paper the two main nutrimetabolomic strategies were highlighted, both greatly aided by the use of NMR spectroscopy. The first one requires a direct intervention on a sample population in order to assess the direct effects on the metabolome after the intake of a particular food product or category. This is particularly useful when specific foods need to be evaluated, especially if they supposedly have beneficial or damaging effects on human health. In this way, their metabolite alteration can be visualised on NMR spectra of biofluid such as urine, and subsequently, the altered pathway or the reason for this metabolic variation can be hypothesised and tested. In the other technique, instead, observational studies in free-eating populations are carried out. At first nutritional data is analysed in order to find possible dietary patterns, then these have to be proved by the inspection of biofluid spectra, in order to find molecules capable in discriminating among the dietary behaviours. This approach is important when the effect of a global diet on human health needs to be evaluated or if there is true adherence to the reported dietary information. Even though further research is needed to validate research protocols and the food biomarkers found, it is clear how these methodologies can give great insight in human metabolism and its relation to both diet and health status. It is clear, though, how NMR spectroscopy techniques can be very useful in different nutrimetabonomics approaches, thanks to the quickness, high-throughput efficacy and high-reproducibility of its methods. In addition, those strategies can allow the elucidation of the holistic profile of metabolites without any kind of a priori knowledge or selection. In this way it is possible to gain a lot of information about the metabolome and its variation, also in cases were nothing in particular is been known or expected. Future research will therefore benefit of the combination of nuclear magnetic resonance techniques in many different metabolomic studies, allowing also the identification of specific molecular biomarkers.

References

R. Goodacre, S. Vaidyanathan, W. B. Dunn, G. G. Harrigan, and D. B.Kell, TrendsBiotechnol., 2004, 22(5), 245-252 N. J. Serkova and C. U. Niemann, Expert Rev. Mol. Diagn., 2006, 6(5), 717-731 N. M. McKeown, N. E. Day, A. A. Welch, S. A. Runswick, R. N. Luben, A. A. Mulligan, A. McTaggart, and S. A. Bingham, Am. J. Clin. Nutr., 2001, 74(2), 188-196. C. Zuppi, I. Messana, F. Forni, C. Rossi, L. Pennacchietti, F. Ferrari, and B. Giardina, Clin. Chim. Acta, 1997, 265(1), 85-97. P. Bernini, I. Bertini, C. Luchinat, P. Nincheri, S. Staderini and P. Turano, J. Biomol. NMR, 2011, 49(3-4), 231-243. F. Dieterle, A. Ross, G. Schlotterbeck, and H. Senn, Anal. Chem., 2006, 78(13), 4281-4290.S. S. Heinzmann, I. J. Brown, Q. Chan, M. Bictash, M. E. Dumas, S. Kochhar, J. Stamler, E. Holmes, P. Elliott, and J. K. Nicholson, Am. J. Clin. Nutr, 2010, 92(2), 436-443.W. M. Edmands, O. P. Beckonert, C. Stella, A. Campbell, B. G. Lake, J. C. Lindon, E.

Foodomics 153

Holmes, and N. J. Gooderham, J. Proteome Res., 2011, 10(10), 4513-4521. A. D. Maher, S. F. Zirah, E. Holmes and J. K. Nicholson, Anal. Chem., 2007, 79(14), 5204-5211. S. Kochhar, D. M. Jacobs, Z. Ramadan, F. Berruex, A. Fuerholz and L. B. Fay, Anal. Biochem., 2006, 352(2), 274-281. E. M. Lenz, J. Bright, I. D. Wilson, S. R. Morgan and A. F. P. Nash, J. Pharmaceut. Biomed., 2003, 33(5), 1103-1115. N. G. Psihogios, I. F. Gazi, M. S. Elisaf, K. I. Seferiadis, and E. T. Bairaktari, NMR Biomed., 2008, 21(3), 195-207. K. S. Solanky, N. J. Bailey, B. M. Beckwith-Hall, S. Bingham, A. Davis, E. Holmes, J. K. Nicholson and A. Cassidy, J. Nutr. Biochem., 2005, 16(4), 236-244. R. Llorach, M. Garcia-Aloy, S. Tulipani, R. Vazquez-Fresno, and C. Andres-Lacueva, J.Agr. Food Chem., 2012, 60(36), 8797-8808. M. Beckmann, A. J. Lloyd, S. Haldar, G. Favé, C. J. Seal, K. Brandt, J. C. Mathers and J. Draper, P. Nutr. Soc., 2013, 72(03), 352-361. R. Vázquez-Fresno, R. Llorach, F. Alcaro, M. Á. Rodríguez, M. Vinaixa, G. Chiva-Blanch, R. Estruch, X. Correig, and C. Andrés Lacueva, Electrophoresis, 2012, 33(15),2345-2354. K. S. Solanky, N. J. Bailey, B. M. Beckwith-Hall, A. Davis, S. Bingham, E. Holmes, J. K. Nicholson and A. Cassidy, Anal. Biochem., 2003, 323(2), 197-204. Y. Wang, H. Tang, J. K. Nicholson, P. J. Hylands, J. Sampson and E. Holmes, J. Agr. Food Chem., 2005, 53(2), 191-196. F. A. Van Dorsten, C. A. Daykin, T. P. Mulder, and J. P. Van Duynhoven, J. Agr. Food Chem., 2006, 54(18), 6929-6938. H. M. Lehtonen, A. Lindstedt, R. Järvinen, J. Sinkkonen, G. Graça, M. Viitanen, H. Kallio, and A. M. Gil, Food Chem., 2013, 138(2), 982-990.O. Cloarec, M. E. Dumas, A. Craig, R. H. Barton, J. Trygg, J. Hudson, C. Blancher, D. Gauguier, J. C. Lindon, E. Holmes and J. Nicholson, Anal. Chem., 2005, 77(5), 1282-1289H. Winning, E. Roldán-Marín, L. O. Dragsted, N. Viereck, M. Poulsen, C. Sánchez-Moreno, M. P. Cano and S. B. Engelsen, Analyst, 2009, 134(11), 2344-2351. C. A. Daykin, J. P. V. Duynhoven, A. Groenewegen, M. Dachtler, J. M. V. and T. P. Mulder, J. Agr. Food Chem., 2005, 53(5), 1428-1434. A. O'Sullivan, M. J. Gibney and L. Brennan, Am. J. Clin Nutr. , 2011, 93(2), 314-321. E. Holmes, R. L. Loo, J. Stamler, M. Bictash, I. K. Yap, Q. Chan, T. Ebbels, M. De Iorio, I. J. Brown, K. A. Veselkov, M. L. Daviglus, H. Kesteloot, H. Ueshima, L. Zhao, J. K. Nicholson and P. Elliott, Nature, 2008, 453(7193), 396-400. E. M. Lenz, J. Bright, I. D. Wilson, A. Hughes, J. Morrisson, H. Lindberg and A. Lockton, J. Pharmaceut. Biomed., 2004, 36(4), 841-849. W. Yang, Y. Wang, Q. Zhou and H. Tang, Sci. China Ser. B, 2008, 51(3), 218-225. H. C. Bertram, C. Hopp, B. O. Petersen, J. Ø. Duus, C. Mølgaard, K. F. Michaelsen, Br.J. Nutr., 97, 758-763F.-P. J. Martin, S. Rezzi, E. Peré Trepat, B. Kamlage, S. Collino, E. Leibold, J. Kastler, D. Rein, L. B. Fay and S. Kochhar, J. Proteome Res., 8, 5568-5579 L. G. Rasmussen, H. Winning, F. Savorani, H. Toft, T. M. Larsen, L. O. Dragsted, A. Astrup and S. B. Engelsen, Nutrients, 4, 112-131

1H NMR METABOLIC PROFILING OF APULIAN EVOOS: FINE PEDOCLIMATIC INFLUENCES IN SALENTO CULTIVARS

Laura Del Coco, Sandra A. De Pascali, Francesco P. Fanizzi*

Di.S.Te.B.A., Prov.le Lecce-Monteroni, University of Salento, Lecce, Italy. Email: *[email protected]

1 INTRODUCTION

Extra virgin olive oils (EVOOs) are characterized by countless factors including their cultivar, terroir, milling systems and last but not least the dedication of the producer. Apulia Region is the leader within Italy and one of the main Southern Italian regions producing extra virgin olive oil. However, differences in microclimatic, pedoclimatic and ecological conditions contribute to induce differences in taste and nutritional properties of the extra virgin olive oils [1]. EVOO beneficial effects on human health, such as the reduction of coronary heart disease risk factors, the prevention of several types of cancer and the modification of immune and inflammatory responses, are well known [2,3]. Benefits are mainly due to both the elevated oleic acid content and the antioxidant properties of EVOO minor components [4], such as phytosterols, carotenoids, tocopherols and hydrophilic phenols. Among the others, antioxidant phenolic compounds play an important role in the olive oil blending [5], since they are related not only to the nutraceutical but also to the sensory EVOO qualities [6]. In this regard, a metabolomic approach was applied to assess the potential effects of cultivar and cultivation area on samples of monovarietal extra virgin olive oils (Olea europaea L.). Metabolomic studies have increased dramatically over the past few years and many researches attempt to process complex metabolic data sets, originating from high throughput analytical methods (such as gas chromatography, GC-MS, as well as nuclear magnetic resonance, NMR and/or mass spectrometry, MS) on environmental and food research [7,8,9]. We are currently involved in several studies aimed to assess intercultivar and intracultivar variations of apulian EVOOs by 1H NMR metabolic profiling. Our previous investigations include 1H NMR analysis of genetically characterized apulian EVOOs [10]; NMR metabolomic studies on monocultivar and blend Salento EVOOs from young and secular olive trees [11]; monitoring of Apulia olive oil production chain by 1D and 2D NMR spectroscopy [9]; 1H NMR spectroscopy and multivariate analysis studies of monovarietal EVOOs as a tool for modulating Coratina-based blends [5] and 1H NMR based comparison of apulian EVOOs with those commercially available as Italian products in the USA [12]. Here we report on 93 authentic EVOO samples collected during the harvesting period 2012-2013 from different microareas of Salento, Lecce, Italy: 26 monocultivar Cellina di Nardò, 32 monocultivar Ogliarola salentina and 35 blend Cellina/Ogliarola. The attention was focused on the two most representative cultivars of Salento area, Cellina di Nardò and Ogliarola

Foodomics 155

Leccese, as they are the basis of “Terra d’Otranto” PDO production. 1H nuclear magnetic resonance (NMR) spectroscopy and chemometrics (OPLS DA) were used to investigate potential differences of Salento EVOOs, originating from specific microareas, related to the chemical composition of major and minor components. As an example of microarea effect the distance from the sea was taken into account to find potential influence of microclimate on the specific studied cultivars.

2 MATERIAL AND METHODS

2.1. Sample Collection The authentic EVOO samples (93) were collected during the harvesting period 2012-2013 from different microareas of Lecce province (Le, Italy): 26 monocultivar Cellina di Nardò; 32 monocultivar Ogliarola Leccese; 35 blend Cellina/Ogliarola. The different olive oil samples were labelled as declared by farmers. The sampling area is geographically located within the Salento peninsula, between the Adriatic and Ionian seas. Salento is an even territory which is quite flat. It can be subdivided into a northern area which is flat and intensely cultivated (Tavoliere of Lecce) and a southern area, which is a low plateau that slopes down towards the sea (Murge Salentine) (Figure 1 and Table 1).

Figure 1 A. Map of Italy, the circle indicates the southern area of the Italian heel. B. Salento peninsula, numbers indicate district area of each town [coastal: 1. Rauccio (Lecce), 2. San Cataldo (Lecce), 3. Vernole, 4. Melendugno, 5. Nardò, 6. Galatone, 7. Racale, 8. Ugento, 9. Presicce, 10. Salve, 11. Morciano di Leuca, 12. Alessano, 13. Tricase; inner: 14. Salice salentino, 15. Veglie, 16. Arnesano, 17. Località Cupa, 18. Monteroni di Lecce, 19. Caprarica, 20. Calimera, 21. Galatina, 22. Cutrofiano, 23. Giurdignano, 24. Palmariggi, 25. Matino, 26. Botrugno].

2.2. Chemicals All chemical reagents for analysis were of analytical grade. CDCl3 (99.8 atom %D) and tetramethylsilane, TMS (0.03 v/v %) were purchased from Armar Chemicals (Switzerland).


Table 1 List of districts from which EVOO samples originate.

ID Coastal District ID Inner District 1 Rauccio (Lecce) 14 Salice Salentino 2 San Cataldo (Lecce) 15 Veglie 3 Vernole 16 Arnesano 4 Melendugno 17 Località Cupa 5 Nardò 18 Monteroni di Lecce6 Galatone 19 Caprarica 7 Racale 20 Calimera 8 Ugento 21 Galatina 9 Presicce 22 Cutrofiano 10 Salve 23 Giurdignano 11 Morciano di Leuca 24 Palmariggi 12 Alessano 25 Matino 13 Tricase 26 Botrugno

2.3. 1H NMR Spectroscopy For NMR sample preparation ~140 mg of olive oil was dissolved in deuterated chloroform (CDCl3 with TMS as internal standard) adjusting the mass ratio of olive oil:CDCl3 to 13.5%:86.5%. 600 μL of the prepared mixture was transferred into a 5 mm NMR tube. NMR spectra were recorded on a Bruker Avance III spectrometer (Bruker, Karlsruhe, Germany), operating at 400.13 MHz for 1H observation and a temperature of 300.0 K, equipped with a BBO 5 mm direct detection probe incorporating a z axis gradient coil. NMR spectra were acquired using Topspin 2.1 (Bruker). Automated tuning and matching, locking and shimming using the standard Bruker routines ATMA, LOCK, and TopShim were used to optimize the NMR conditions. Experiments were run in automation mode after loading individual samples on a Bruker Automatic Sample Changer, (BACS-60), interfaced with the software IconNMR (Bruker). Two different 1H NMR experiments were performed for each sample: a standard one-dimensional 1H ZGNMR experiment and a one-dimensional 1H NOESYGPPS NMR pulse sequence with suppression of the strong lipid signals (20 frequencies), in order to enhance signals of minor components present in EVOOs (Bruker). Spectra were obtained by the following conditions: zg pulse program (for 1H ZGNMR) 64 K time domain, spectral width 20.5555 ppm (8223.685 Hz), p1 12.63 s, pl1 1.00 db, 16 repetitions; noesygpps1d.comp2 pulse program (for 1H NOESYGPPS NMR) 32 K time domain, spectral width 20.5555 ppm (8223.685 Hz), p1 12.63 s, pl1 1.00 db, 32 repetitions.

2.4. NMR Data Reduction and Preprocessing NMR data were processed using Topspin 2.1 (Bruker) and visually inspected using Amix 3.9.13 (Bruker, Biospin). 1H NMR spectra were obtained by the Fourier Transformation (FT) of the FID (Free Induction Decay), applying an exponential multiplication with a line-broadening factor of 0.3 Hz. The resulting 1H NMR spectra were manually phased and baseline corrected using the Bruker Topspin software. Chemical shifts were reported with respect to the TMS signal set at 0 ppm. 1H NMR spectra were segmented in rectangular buckets of fixed 0.04 ppm width and integrated, using the Bruker Amix software. Bucketing of 1H ZGNMR spectra (BUCKET-1) and 1H NOESYGPPSNMR spectra (BUCKET-2) were obtained within the range 10.0 - 0.5 ppm (BUCKET-1) and 10.0 - 5.6 ppm (BUCKET-2), respectively. In both cases, the spectral region between 7.60 and 6.90 ppm was discarded because of the peak due to residual solvent signal. The remaining buckets were then

Foodomics 157

normalized to total area to minimize small differences due to total olive oil concentration and/or acquisition conditions among samples. A third data set named BUCKET-3 was generated combining BUCKET-1 and BUCKET-2 in one matrix (1 line per olive oil sample).

2.5 Multivariate statistical analysis (MVA) Multivariate statistical analysis and graphics were obtained using Simca-P version 13.0.2 (Umetrics, Sweden). PCA, (unsupervised) and supervised (OPLS-DA) pattern recognition methods were performed to examine the intrinsic variation in the data set. The Partial Least Squares Discriminant Analysis (PLS-DA) and the Orthogonal Partial Least Squares Discriminant technique (OPLS-DA) are the most recently used for the discrimination of samples with different characteristics (such as cultivars and/or geographical origin) as shown in several recent studies of metabolomics [12,13,14]. OPLS-DA is a modification of the usual PLS-DA method which filters out variation that is not directly related to the response. So, the further improvements made by the OPLS-DA resides in the ability to separate the portion of the variance useful for predictive purposes from the not predictive variance (which is made orthogonal). Furthermore, OPLS-DA focuses the predictive information in one component, facilitating the interpretation of spectral data. The robustness and predictive ability of the OPLS-DA models for discrimination purposes were tested by cross-validation [15]. The R2(cum) and Q2(cum) are the two parameters that describe the goodness of the models. The former (R2) explains the total variations in the data, whereas the latter (Q2) is a cross validation parameter, which indicates the predictability of the model.


OPLS-DA was performed on Cellina di Nardò (26 samples) and Ogliarola Leccese (32 samples) using the statistical models for classification purposes of blend samples (35 Cellina/Ogliarola samples). OPLS-DA in classification studies is a maximum separation projection of data and is most useful when dealing with two classes as it shows which variables are responsible for class separation. In this way potential biomarkers may be found by looking at the most positive and negative loadings in the S-line plot. OPLS-DA, applied at the two most representative cultivars of Salento area (Cellina di Nardò and Ogliarola Leccese), gives a good model (1 predictive + 2 orthogonal) with R2=0.661 and Q2=0.448. The predictive variation, t1, explains 9.01% of all variation in the data and the uncorrelated variation, to1 (orthogonal variation), corresponds to 2.22%. Significant differences for the two cultivars concerned concentrations of some molecules of the unsaponifiable fraction of EVOOs (such as minor components) and fatty acids. The performance classification of OPLS-DA for blend EVOOs (Cellina/Ogliarola samples) in the score plot tPS[1] vs. toPS[1] (Figure 2) shows a clear separation of the two groups. The S-line plot reveals the extent to which each variable is either up or down regulated when going from one to another cultivar, representing in a one-dimensional plot the discriminating features. Analysing the loadings, Cellina samples showed higher levels of molecules having signals at 9.64 and 6.64 ppm, attributed to aldehydic and dialdehydic forms of oleuropein and ligstroside [16], and lower levels of molecules having signals in the range H 6.8 - 5.5, that could be ascribed to carotenoids. Therefore, this model could also offer a method for blend classification by investigating the degree of overlap for blend samples with the monovarietal ones, providing that they were obtained in the same relatively small geographical area such as for Salento EVOOs.


Figure 2 left: OPLS-DA CV-scoreplot for blend Cellina/Ogliarola samples (1 predictive and 2 orthogonal, R2=0.661 and Q2=0.448); right: S-Line Plot for the model between Cellina and Ogliarola EVOOs. The relevance to the model is indicated by the signal amplitude.

In order to evaluate microarea effect, for the studied Cellina and Ogliarola EVOO samples we have taken into account the distance from the sea to find potential influence of mesoclimate. Chemometrics (OPLS DA) was used to investigate how the distance from the sea can affect the olive oil quality of a specific cultivar. To eliminate the intrinsic variation of genotype, one cultivar at time was considered. Therefore, 32 samples of Ogliarola Leccese and 24 samples of Cellina di Nardò were submitted to statistical analysis. Two new subcategories were built inside each group: one category consisted of samples from districts having a cross border with the sea, the other category was made of samples from internal areas [17]. San Cataldo (Lecce), Parco Rauccio (Lecce), Morciano di Leuca, Ugento, Presicce, Racale, Salve, Tricase, Galatone, Alessano, Melendugno, Nardò, Vernole are the coastal areas, while Palmariggi, Calimera, Monteroni, Arnesano, Località Cupa, Matino, Caprarica, Galatina, Veglie, Botrugno, Cutrofiano, Giurdignano, Salice Salentino are considered as inner districts. Salento peninsula, being characterized by a dense partitioning in small towns, offers the opportunity to correlate generally the particular pedoclimatic condition with the district area of each considered town (http://www.salento.com/il-salento/i-comuni). OPLS-DA (Figure 3) applied for each of the two most representative cultivars of Salento area (Ogliarola Leccese and Cellina di Nardò) give models with 1 predictive and 1 orthogonal components, (R2=0.629 and Q2= 0.012, R2=0.665 and Q2= 0.001, for Ogliarola and Cellina, respectively). R2 is a quite good parameter for both the studied models, while a low value of Q2 indicates that although not predictive the models remain still very good as descriptive. Also in this case, a certain degree of variation in metabolite content was observed from the S-line plot analysis. The S-line plot was used to visualize NMR signals that influence the separation of the groups. Higher polyphenols and aldehydes content was found (signals at 9.64, 9.28, 9.24 ppm for aldehydes, 6.64, 6.76 ppm for phenolic compounds) for EVOOs coming from coastal areas, as well as for saturated fatty acid content (methilenic protons at 1.26 ppm). On the contrary, higher polyunsaturated fatty acid content (signals at 2.02 and 1.3 ppm assigned to allylic and methylenic protons of both linolenic and linoleic acids) was found for samples coming from inner areas of Salento peninsula. Interestingly, these results parallel the already reported studies on the adaptive response to water stress,

Foodomics 159

being water stress condition effects similar to those exhibited in this work by samples obtained from olive trees closer to the sea. Among the agronomical factors affecting EVOO chemical composition (cultivar, the system of extraction, and the conditions of processing and storage), most studies report a decrease of total phenol and polyphenols content and oxidation stability as the amount of supplied water increases [2]. As far as we know, this finding, obtained in our case for Cellina and Ogliarola cultivars, is the first reported evidence of microarea effect, which relates the distance from the sea on the EVOO characteristics.

Figure 3 OPLS-DA score plot for Ogliarola Leccese (A) and Cellina di Nardò (B) EVOOs. OPLS DA was used to investigate how the distance from the sea can affect the olive oil quality of a specific cultivar. Numbers are referred to districts listed in Table 1.

4 CONCLUSIONS 1H nuclear magnetic resonance (NMR) spectroscopy and chemometrics (OPLS DA) were used to investigate potential differences of Salento EVOOs, originating from specific microareas, related to the chemical composition of major and minor components. OPLS-DA not only differentiates oils between monocultivars and blends but also according to other features such as the sea distance. Higher polyphenols and aldehydes content was found for EVOOs coming from coastal areas, as well as for saturated fatty acid content. On the contrary, higher polyunsaturated fatty acid content was found for samples coming from inner areas of Salento peninsula. These results parallel the reported studies on adaptive response of olive trees and therefore EVOO production to water stress, being the characteristics of oils related to water stress condition similar to those exhibited in this work by samples obtained from olive orchard closer to the sea. Acknowledgments This work was supported by Apulia region project grant (PIF mis. 124 Filiera Olivicola 100% Pugliese JonicoSalentina). We thank Coldiretti Lecce, Agricola Nuova Generazione soc. coop. Agricola and Dr Carmelo Buttazzo for providing and organizing EVOO samples.


References

1 Mannina, A. P. Sobolev, Magn. Reson. Chem. 2011, 49, S3–S11. 2 W.C. Willett, F. Sacks, A. Trichopoulou, G. Drescher, A. Ferro-Luzi, E. Helsing, D.

Trichopoulos, Am. J. Clin. Nutr., 1995, 61, 1402S–1406S. 3 L. Lipworth, M.E. Martinez, J. Angell, C.C. Hsien, D. Trichopoulos, Prev. Med. 1997,

26, 81–190. 4 R.W. Owen, A. Giacosa, W.E. Hull, R. Haubner, G. Wurtele, B. Spiegelhalder, H.

Bartsch, Lancet Oncol. 2000, 1, 107–112. 5 L. Del Coco, S.A. De Pascali, F.P. Fanizzi, Foods, 2014, 3, 238-249. 6 M. Tsimidou, Ital. J. Food Sci. 1998, 10, 99–116. 7 A. Spyros, P. Dais, NMR spectroscopy in food analysis, RSC Food Analysis

Monographs, 2013, ISBN: 978-1-84973-175-1. 8 S.A. De Pascali, A. Coletta, L. Del Coco, T. Basile, G. Gambacorta, F.P. Fanizzi, Food

Chem., 2014, 161, 112–119. 9 L. Del Coco, S.A. De Pascali, V. Iacovelli, G. Cesari, F.P. Schena, F.P. Fanizzi, Eur. J.

Lipid Sci. Technol., 2014, 116, DOI: 10.1002/ejlt.201400139. 10 P. Papadia, L. Del Coco, I. Muzzalupo, M. Rizzi, E. Perri, G. Cesari, V. Simeone, D.

Mondelli, F.P. Schena, F.P. Fanizzi, J. Am. Oil Chem. Soc., 2011, 88, 1463–1475. 11 L. Del Coco, S.A. De Pascali, F.P. Fanizzi, Food and Nutrition Sciences, 2014, 5, 89-95. 12 L. Del Coco, F.P. Schena, F.P. Fanizzi, Nutrients, 2012, 4, 343–355. 13 F. Longobardi, A. Ventrella, A. Bianco, L. Catucci, I. Cafagna, V. Gallo, P. Mastrorilli,

A. Agostiano, Food Chem., 2013, 141, 3028–3033. 14 L. Mannina, F. Marini, M. Gobbino, A.P. Sobolev, D. Capitani, Talanta, 2010, 80, 2141-

2148. 15 H. Eastment, W. Krzanowski, Technometrics, 1982, 24, 73–77. 16 S. Christophoridou, P. Dais, L.H. Tseng, M. Spraul, J. Agric. Food Chem. 2005, 53,

4667–4679. 17 S. Lanteri, C. Armanino, E. Perri, A. Palopoli, Food Chem., 2002, 76, 4, 501–507. 18 M. Patumi, R. D’Andria, V. Marsilio, G. Fontanazza, G. Morelli, B. Lanza, Food Chem.,

2002, 77, 27-34.

ADDITION OF ESSENTIAL OILS TO COWS’ FEED ALTERS THE MILK METABOLOME – NMR SPECTROSCOPIC STUDIES OF “NATURE’S PERFECT FOOD”.

U.K. Sundekilde1, M.R. Clausen1, J. Lejonklev1, M.R. Weisbjerg2, M.K. Larsen1, and H.C. Bertram1.

1Department of Food Science, Aarhus University, Denmark 2Department of Animal Science, Aarhus University, Denmark

1 INTRODUCTION

Milk is an important component in the food throughout the life, as milk is a central source of minerals and is well-balanced with respect to the nutritional value of carbohydrates, fats, proteins, minerals and vitamins. Accordingly, milk is often described as an almost perfect food.1,2 Milk contains lipids, carbohydrates, proteins, immunoglobulins, peptides, amino acids, nucleotides, vitamins, minerals, and other metabolites. Many factors influence the milk composition; feed, seasonal variation, genetics, health, and number of and stage of lactation.3–7 There is a clear correlation between the milk composition and e.g. its nutritional value,8 technological properties,9,10 and content of functional ingredients11 – all of which collectively is covered by the term milk quality. For a number of years, breeding programs have been in place to actively improve milk quality. The breeding strategies employed focused on maximizing parameters directly related to the payment plans, and were neither specifically related to nutritional nor technological quality. Technological properties of milk, e.g. milk coagulation (rennet- and acid-induced) and milk stability during processing, is also related to the milk composition.9,12,13 Additionally, a large focus has been on the effect of feed on milk quality. It has been shown that changes in the feeding regime can introduce changes in the fatty acid composition of milk.3,14–16 The milk flavour can also be altered through the feeding.17–19 Presently, less attention has been paid to changes in the milk metabolite profile through alterations in feeding regime. The milk metabolome encompass the low molecular weight compounds present in milk and originate from multiple cell types or metabolic pathways, and the different sources of metabolites may contribute to the variability of milk metabolites. Volatile compounds such as terpenes can be transferred from feed to milk and affect milk flavour.20,21 Terpenes are common in animal feed, especially herbs and their transfer to milk is well established.22 Mono- and sesquiterpenes can be transported to the milk through both respiratory and gastrointestinal exposure.17 Nuclear magnetic resonance (NMR) is a useful technique for the analysis of metabolites in solution and has proven to be an excellent metabolomics technique for the study of food items, as a result of its ability to assess a metabolic fingerprint of an individual sample or specimen.23–25 In bovine milk, NMR-based metabolomics has previously been used to identify small molecules in several studies; biochemical variability in early and late lactation milk,26 association between milk metabolites and the coagulation process,9 biomarkers for


elevated somatic cell count,6 biomarkers for ketosis milk adulteration,27 and control of geographic region.28 The use of NMR metabolomics to assess the milk metabolome and in dairy research in general has recently been reviewed.29,30 In the present study NMR spectroscopy was used to identify how changes in the feeding regime could cause changes in the milk metabolome. More specifically, we wanted to examine if addition of oregano (Origanum vulgare) and caraway (Carum carvi) essential oils in the feed, which are known to introduce differences in the milk volatile compounds immediately upon addition, would also affect the milk metabolite profile.


2.1 Animals

Fifteen Danish Holstein cows were included in the study. The animals were divided into five groups, a control group and a group for each of the four essential oil treatments; low caraway (Carum Carvi), high caraway, low oregano (Origanum vulgare), and high oregano. Each group included one animal in first lactation, one animal in second lactation, and one animal in third or fourth lactation. Days in milk ranged from 46 to 142 days at the start of the experiment. The total mixed ratio used as animal feed was composed by, per kg dry matter, 350 g corn silage, 300 g grass silage, 120 g barley, 110 g soy bean meal, 100 g rape seed meal and 20 g minerals, with a dry matter concentration of 49.3% ± 0.4. Feed consumption and milk production were recorded during the experiment.31 Caraway or oregano oils (New Directions, Hampshire, UK) were added to the feed at 0.2 g oil per kg dry matter feed for the low level, and 1.0 g oil per kg for the high level. These levels correspond to an herb content of 1 or 5% in the feed. Milk samples for NMR spectroscopy was obtained at two occasions by equal mixing of evening milk day 23 with morning milk day 24, and evening milk day 24 and morning milk day 25. Within six hours after morning milking the mixed milk was pasteurized (72°C, 15 seconds) and frozen at -20°C prior to NMR spectroscopy. Subsamples of the pasteurized milk were analysed for content of fat, protein and lactose using a Milkoscan 4000 (Foss Electric, Denmark).

2.2 NMR Spectroscopy

Prior to NMR spectroscopy the skimmed milk samples were thawed and thoroughly shaken to homogenize the samples. The samples were then filtered to remove residual lipids and protein using Amicon Ultra 0.5 mL 10 kDa (Millipore, USA) spin filters at 10,000 g for 15 min. 400 L filtered sample was added 200 L D2O containing sodium trimethylsilyl-[2,2,3,3-2H4]-1-propionate (TSP; Sigma-Aldrich, Brøndby, Denmark) as an internal chemical shift reference. Two technical replicates of each sample were prepared for 1H NMR spectroscopy. The sample sequence was randomized prior to acquisition. 1H NMR spectroscopy was performed at 298 K on a Bruker Avance III 600 spectrometer, operating at a 1H frequency of 600.13 MHz, and equipped with a 5-mm 1H TXI probe (Bruker BioSpin, Rheinstetten, Germany). Standard one-dimensional spectra were acquired using a single 90 pulse experiment with a relaxation delay of 5 s. Water suppression was achieved by irradiating the water peak during the relaxation delay, and a total of 64 scans were collected into 32 K data points spanning a spectral width of 12.15 ppm. NMR signals have been assigned in accordance with existing literature.29

Foodomics 163

All 1H NMR spectra were referenced to the TSP signal at 0 ppm. The data were multiplied by a 0.3 Hz line-broadening function prior to Fourier transformation. The proton NMR spectra were phase- and baseline corrected manually using Topspin 3.2 (Bruker BioSpin).

2.3 Multivariate data analysis.

The 1H NMR spectra were aligned using Icoshift by co-shifting of the whole spectra according to the lactose doublet at 5.23 ppm.32 The proton NMR spectra were subdivided into 0.01 ppm bins reducing each spectrum to 960 separate variables. Principal component analysis (PCA) and orthogonal projection to latent squares discriminant analysis (OPLS-DA) was performed in order to identify differences in the metabolite profiles. The OPLS-DA models were cross validated using segmentation with 9 splits, where each technical replicate was placed in the same segment. Q2, the per cent of variation of the training set predicted by the model according to cross validation, and root mean square error of cross validation (RMSECV) were measures of the model robustness. Covariance was investigated by analysis of OPLS-DA regression coefficients back-transformed to original data and colour coded by the loading weights.33 Test for OPLS-DA model statistical significance was performed by ANOVA of the cross-validated residuals 34. PCA and OPLS-DA analyses were performed using SIMCA-P+ 13 (Umetrics AB, Sweden). Alignment by Icoshift, binning, and analysis of OPLS-DA regression coefficient plots were performed in MATLAB 7.13 (MathWorks Inc., Natick, MA, USA).

3 RESULTS

Figure 1 Spectra of milk samples (n = 30). The region 10-5.3 ppm is enlarged 100 times and the region 3.2-0 ppm is enlarged 25 times compared to the central region 5.3-3.2. For assignments see Table 1.

The obtained 1H NMR spectra (n=30) of the milk samples are shown overlaid in Figure 1. The proton assignments are listed in Table 1. A total of 32 metabolites were assigned (Figure 1 and Table 1).29


Table 1 List of chemical shift values and proton assignments for milk metabolites.

Metabolite 1H chemical shift Assignment Butyrate 0.90 CH3 Isoleucine 0.93 -CH3 Valine 0.99, 1.04 CH3, CH3 Ethanol 1.17 CH3 Fucose 1.25 CH3 Lactate 1.32 CH3 Alanine 1.48 CH3 Ornithine 1.8 -CH2 Acetate 1.92 CH3 N-acetyl hexosamine 2.04 CH3 N-acetyl hexosamine 2.06 CH3 Methionine 2.15 -CH2, S-CH3 Acetone 2.24 CH3 Glutamate 2.36 -CH2 Carnitine 2.44 CH2 Citrate 2.52, 2.72 CH2, CH2 Creatinine 3.05 CH3 Malonic acid 3.11 CH2 Choline 3.18 3×CH3 Carnitine 3.21 3×CH3 Lactose Multiple1 – Phosphocholine 3.58 N-CH2 Phosphocholine 4.16 O-CH2 Glycerophosphocholine 4.32 N-CH2 Galactose 4.57 CH-1Glucose 4.65 CH-1Galactose 1-phosphate 5.38 CH-1Glucose 1-phosphate 5.51 CH-1Orotate 6.2 CHFumarate 6.52 CHHippurate 7.54, 7.64, 7.84 CH2-3,5; CH-4; CH2-2,6 Adenine 8.12, 8.13 CH-8, CH-2 Formate 8.45 CH

Foodomics 165

Figure 2 PCA scores plot of the fifth and sixth principal components of Pareto scaled NMR data from 15 milk samples measured in duplicates. Cows were fed oregano (o) or caraway ( ). Control samples (+). The most abundant metabolites detected in milk by 1H NMR spectroscopy included lactose, citrate and carnitine (Figure 1). The overlaid spectra show some variability and to identify if these differences could be associated with addition of essential oils in the feed, multivariate data analysis was performed. A PCA model based on the metabolite profiles of all 30 (6 oregano, 6 caraway, and 3 control samples analysed in duplicates) show a clustering along principal component (PC) 5 and PC 6 (Figure 2). The first four principal components do not show any clustering according to treatments. PC5 and PC6 in total explain 4.3% of the variance (Figure 2). Thus, an effect of addition of essential oils to the feed is detected in the milk by NMR spectroscopy, but the major part of the variability in the milk metabolomes was not altered as a result of addition of essential oils in the feed (Figure 2). In order to identify the changes introduced in the milk metabolome by addition of oregano and caraway, PCA and OPLS-DA models were generated on the data individually according to treatment (Figure 3 and 4, respectively). Changes in the milk metabolome upon addition of oregano to the feed are shown in Figure 3. A PCA model shows a tendency for clustering (Figure 3A) along PC6, which accounts for 1.44% of the explained variance. An OPLS-DA model of the unit variance scaled data would simplify interpretation as uncorrelated variation for this clustering is discarded. Thus, an OPLS-DA model is shown in Figure 3B and Figure 3C (Q2=0.97). The predictive component encompasses 5.41% of the total variation in the dataset (Figure 3B). Covariance was investigated by analysis of OPLS-DA regression coefficients back-transformed to original data and colour coded by the loading weights (Figure 3C). The coefficient plot shows that N-acetyl hexosamine ( 2.04 ppm), citrate ( 2.52 ppm), creatinine ( 3.05 ppm), and orotate ( 6.51 ppm) were significantly increased in milk samples from cows fed essential oils derived from oregano, whereas N-acetyl hexosamine ( 2.06 ppm), choline ( 3.18 ppm), lactose ( 3.5-4 ppm, 4.46 ppm, and 5.23 ppm), phosphocholine ( 4.16 ppm), orotate ( 6.27 ppm), galactose 1-phosphate ( 5.38 ppm), and glucose 1-phosphate ( 5.51 ppm) were significantly decreased in milk from cows


fed essential oils derived from oregano (Figure 3C). However, as the loading weights of lactose and citrate is low, their importance for the discriminative ability of the OPLS-DA model is insignificant. Changes in the milk metabolome upon addition of caraway to the feed are shown in Figure 4. A PCA model shows a tendency for clustering along PC3 and PC5 (Figure 4A), which account for 6.29% and 1.46% of the explained variance, respectively. An OPLS-DA model of the unit variance scaled data is shown in Figure 4B and Figure 4C (Q2=0.88).

Figure 3 (A) PCA score plot of the first and sixth principal components of unit variance scaled NMR data from 9 milk samples measured in duplicates of cows fed oregano (o), control samples (+). (B) OPLS-DA score plot of a model of the unit variance scaled data in (A). (C) OPLS-DA coefficients plot of the model shown in (B) back-transformed. Each variable has been coloured according to the OPLS-DA loadings.

The predictive component encompasses 6.99% of the total variation in the dataset (Figure 4B). The coefficient plot shows that the metabolites N-acetyl hexosamine ( 2.06 ppm), glutamate ( 2.36 ppm), citrate ( 2.52 and 2.72 ppm), creatinine ( 3.05 ppm), choline ( 3.18 ppm), carnitine ( 3.21 ppm), and hippurate ( 7.54 ppm, 7.64 ppm, and 7.84 ppm) were significantly different between control and caraway samples. In addition, the coefficient plot indicated resonances in the region 3-4 ppm varied between the different feeding regimes (Figure 4C). This region is dominated by resonances from the milk carbohydrates, most importantly lactose. However, the proton resonances in this region do not all have the same direction in the coefficient plot, which could indicate variation in metabolites other than

Foodomics 167

lactose in this region or the variability could be noise arising from small shifts between samples. Citrate, creatinine, and choline were significantly increased in milk samples from cows fed essential oils derived from caraway, whereas N-acetyl hexosamine, glutamate, carnitine, and hippurate were significantly decreased in milk samples from cows fed essential oils derived from caraway (Figure 4C).

Figure 4 (A) PCA score plot of the third and fifth principal components of unit variance scaled NMR data from 9 milk samples measured in duplicates of cows fed caraway ( ), control samples (+). (B) OPLS-DA score plot of a model of the unit variance scaled data in (A). (C) OPLS-DA coefficients plot of the model shown in (B) back-transformed. Each variable has been coloured according to the OPLS-DA loadings

4 DISCUSSION

Milk composition is affected by the diet.3 There has been a large amount of research into the transfer of volatile compounds from feed into milk, whereas little effort has been put into research on the effect of diet on the milk metabolite profile. The present study pursued to identify differences in the bovine milk metabolome by using NMR spectroscopy upon addition of substances in the feed known to able to transfer into milk, and thereby eliciting differences in milk quality by an explorative approach. Principal component analysis (PCA) was performed on all milk samples, which included control samples and samples, where oregano and caraway essential oils have been added to the feed of the cow, and it showed that the three groups of samples could be distinguished from each other, indicating significant


changes between the sample groups. In combination with supervised OPLSDA models, the differences induced by changes in the feeding regime were shown. During the experiment dry matter intake and milk production were measured and analysed.31 No effects of feeding regime on dry matter intake and milk production were observed.31 Changes in citrate levels in milk is known to take place as a result of changes in the milk lipid fraction.35,36 However, we did not observe changes in fatty acid profile of the milk samples as a result of addition of oregano and caraway essential oils (data not shown). Creatinine content of milk has been shown to be affected by feed, as inclusion of urea in the feed elevated the level of creatinine in the milk.37 Furthermore, cows on an urea-supplemented diet showed an increase in orotate excretion in the urine, but did not show elevated levels of orotate in milk.38 Orotate was originally identified as vitamin B13.39 However, evidence has emerged that orotate is not a vitamin, and can accordingly be produced by the animal. Orotate is known to be synthesized in the mammary gland,40 and to participate in the biosynthesis of pyrimidine nucleotides.41 Interestingly, in the present study orotate levels were shown to increase upon inclusion of oregano in the feed. The carbohydrates N-acetyl glucosamine and N-acetyl galactosamine, are components of oligosaccharides and glycosylation of proteins. Results from an experiment where lactating mothers were given stable isotope labelled [13C]galactose indicated that a significant part of galactose intake is transferred directly to the mammary gland for use in the biosynthesis of carbohydrates.42 How the N-acetylated carbohydrates in milk are altered by addition of essential oils in the feed remains unclear, but are nonetheless an interesting finding, due to the fact that oligosaccharides are found in high concentration in human milk, whereas bovine milk only has low amounts of oligosaccharides. Identification of a source of milk oligosaccharides is an important step in securing a high quality infant formula with a high resemblance to human milk. It has been suggested that choline may be a limiting nutrient for milk production and addition of rumen-protected choline in the feed has been extensively studied.43–45 Abomasally infusion of choline resulted in significantly increased choline concentration in bovine milk.46 Milk is isotonic with blood, and lactose is responsible for 50% of the osmotic pressure of milk. Thus, in order to maintain the osmotic pressure, milk with a low level of lactose has a high level of inorganic salts or other compounds.47 Thus, if changes in the feed result in influx of additional compounds or result in altered milk yield, the level of lactose in milk is dynamically changed to match the osmotic pressure of milk to blood. In the present study the addition of essential oils derived from herbs to the feed mimics the difference between organic and conventional farming, as organically breed cows have an increased intake of herbs, grasses, and clover. Thus, we would expect to observe similar differences in the milk metabolome as observed in conventional and organic cows. The level of glucose 1-phosphate, hippurate, and carnitine in milk has previously been shown to be affected by the diet as the compounds were found in significantly different concentrations in organic and conventional farming.48 Moreover, hippurate has been proposed as a marker for organic farming, due to the fact that it is increased in milk from organic cows compared with conventional farming.48 In conclusion, our results highlight the usefulness of the exploratory approach of NMR-metabolomics in assessing the differences introduced into milk by alterations in the feeding regime. Acknowledgements The authors thank the Danish Research Council FTP for financial support through the project “Advances in Food quality and Nutrition Research through implementation of metabolomic

Foodomics 169

strategies”. Furthermore, The Danish Ministry of Food, Agriculture and Fisheries and the Sund of velsmagende mælk consortium, consisting of Arla Foods, DLF Trifolium and Aarhus University are thanked for financial support.

References

1 N. W. Solomons, Nutr. Rev., 2002, 60, 180. 2 M. DuPuis, Nature’s Perfect Food: How Milk Became America's Drink, New York

University Press, New York, 2002. 3 T. C. Jenkins and M. A. McGuire, J. Dairy Sci., 2006, 89, 1302. 4 M. J. Auldist, B. J. Walsh, and N. A. Thomson, J. Dairy Res., 1998, 65, 401. 5 A. J. Buitenhuis, U. K. Sundekilde, N. a Poulsen, H. C. Bertram, L. B. Larsen, and P.

Sørensen, J. Dairy Sci., 2013, 96, 3285. 6 U. K. Sundekilde, N. A. Poulsen, L. B. Larsen, and H. C. Bertram, J. Dairy Sci., 2013,

96, 290. 7 J. M. L. Heck, H. J. F. van Valenberg, J. Dijkstra, and A. C. M. van Hooijdonk, J. Dairy

Sci., 2009, 92, 4745. 8 K. E. Kliem and D. I. Givens, Annu. Rev. Food Sci. Technol., 2011, 2, 21. 9 U. K. Sundekilde, P. D. Frederiksen, M. R. Clausen, L. B. Larsen, and H. C. Bertram, J.

Agric. Food Chem., 2011, 59, 7360. 10 H. Harzia, K. Kilk, I. Jõudu, M. Henno, O. Kärt, and U. Soomets, J. Dairy Sci., 2012, 95,

533. 11 D. Barile, N. Tao, C. B. Lebrilla, J. D. Coisson, M. Arlorio, and J. B. German, Int. Dairy

J., 2009, 19, 524. 12 M. J. Auldist, K. A. Johnston, N. J. White, W. P. Fitzsimons, and M. J. Boland, J. Dairy

Res., 2004, 71, 51. 13 A. Wedholm, L. B. Larsen, H. Lindmark-Månsson, A. H. Karlsson, and A. Andrén, J.

Dairy Sci., 2006, 89, 3296. 14 K. J. Shingfield, C. K. Reynolds, G. Hervás, J. M. Griinari, A. S. Grandison, and D. E.

Beever, J. Dairy Sci., 2006, 89, 714. 15 N. A. Poulsen, F. Gustavsson, M. Glantz, M. Paulsson, L. B. Larsen, and M. K. Larsen,

J. Dairy Sci., 2012, 95, 6362. 16 T. Slots, G. Butler, C. Leifert, T. Kristensen, L. H. Skibsted, and J. H. Nielsen, J. Dairy

Sci., 2009, 92, 2057. 17 J. Lejonklev, M. M. Løkke, M. K. Larsen, G. Mortensen, M. A. Petersen, and M. R.

Weisbjerg, J. Dairy Sci., 2013, 96, 4235. 18 G. Urbach, J. Dairy Sci., 1990, 73, 3639. 19 M. Yayota, M. Tsukamoto, Y. Yamada, and S. Ohtani, J. Dairy Sci., 2013, 96, 5174. 20 J. G. Bendall, J. Agric. Food Chem., 2001, 49, 4825. 21 G. Tornambe, A. Cornu, P. Pradel, N. Kondjoyan, A. P. Carnat, M. Petit, and B. Martin,

J. Dairy Sci., 2006, 89, 2309. 22 S. Prache, A. Cornu, J. L. Berdagué, and A. Priolo, Small Rumin. Res., 2005, 59, 157. 23 J. C. Lindon, J. K. Nicholson, and J. R. Everett, Annu. Reports Nmr Spectrosc., 1999, 38,

1. 24 T. N. Kolokolova, O. Y. Savel’ev, and N. M. Sergeev, J. Anal. Chem., 2008, 63, 04. 25 D. S. Wishart, Trends Food Sci. Technol., 2008, 19, 482. 26 M. S. Klein, M. F. Almstetter, G. Schlamberger, N. Nurnberger, K. Dettmer, P. J.

Oefner, H. H. D. Meyer, S. Wiedemann, and W. Gronwald, J. Dairy Sci., 2010, 93, 539. 27 H. Xin and R. Stone, Science, 2008, 322, 1310. 28 P. Mazzei and A. Piccolo, Food Chem., 2012, 132, 1620.


29 U. K. Sundekilde, L. B. Larsen, and H. C. Bertram, Metabolites, 2013, 3, 204. 30 A. Maher and S. Rochfort, Metabolites, 2014, 4, 131. 31 J. Lejonklev, Aarhus University, 2013. 32 F. Savorani, G. Tomasi, and S. B. Engelsen, J. Magn. Reson., 2010, 202, 190. 33 O. Cloarec, M. E. Dumas, J. Trygg, A. Craig, R. H. Barton, J. C. Lindon, J. K.

Nicholson, and E. Holmes, Anal. Chem., 2005, 77, 517. 34 L. Eriksson, J. Trygg, and S. Wold, J. Chemom., 2008, 22, 594. 35 W. Banks, J. L. Clapperton, A. K. Girdler, and W. Steele, J. Dairy Res., 1984, 51, 387. 36 D. L. Palmquist, A. Denise Beaulieu, and D. M. Barbano, J. Dairy Sci., 1993, 76, 1753. 37 P. Susmel, M. Spanghero, B. Stefanon, and C. R. Mills, Livest. Prod. Sci., 1995, 44, 207. 38 T. Motyl, W. Barej, and H. Leontowicz, Arch. Tierernahr., 1986, 36, 551. 39 A. F. Novak and S. M. Hauge, J. Biol. Chem., 1948, 174, 647. 40 A. A. Ahmed, G. A. Porter, and R. D. McCarthy, J. Dairy Sci., 1978, 61, 39. 41 M. E. Jones, Annu. Rev. Biochem., 1980, 49, 253. 42 C. Kunz, S. Rudloff, and W. Baier, Annu. Rev. Nutr., 2000, 20, 699. 43 R. A. Erdman and B. K. Sharma, J. Dairy Sci., 1991, 74, 1641. 44 J. Sales, P. Homolka, and V. Koukolová, J. Dairy Sci., 2010, 93, 3746. 45 R. L. G. Zom, J. van Baal, R. M. A. Goselink, J. A. Bakker, M. J. de Veth, and A. M.

van Vuuren, J. Dairy Sci., 2011, 94, 4016. 46 K. N. Deuchler, L. S. Piperova, and R. A. Erdman, J. Dairy Sci., 1998, 81, 238. 47 P. F. Fox and P. L. H. McSweeney, Dairy chemistry and biochemistry, Blackie

Academic & Professional Publishers, London, 1998. 48 K. J. Boudonck, M. W. Mitchell, J. Wulff, and J. A. Ryals, Metabolomics, 2009, 5, 375.

HIGH-RESOLUTION MAGIC ANGLE SPINNING STUDIES OF SEMI-HARD DANBO (30+) CHEESE- IMPACT OF PROCESSING CONDITION AND RELATION TO SENSORY PERCEPTION

Lamichhane S., Yde C. C., Mielby L. H., Kidmose U., Møller J. R., Hammershøj M., and Bertram H. C.

Department of Food Science, Aarhus University, Denmark

1 INTRODUCTION

Danbo is a semi-hard, matured Danish cheese made from cow’s milk, buffalo’s milk, or their mixture (Codex Standard 264-1966). Danbo is one of the most commonly consumed cheeses in Denmark, and it is typically manufactured with three different fat contents corresponding to a fat content of 45, 30 and 20% (dry matter basis).1 Due to the proclaimed health benefits of eating a low salt (NaCl) diet, WHO recommends consuming less than 5g salt/day. The fact that Danbo (30+) contains 1.8% salt, it is of public concern to reduce the salt content. In order to achieve the key-hole labeling or low-salt foods the content of a Danbo cheese should be less than 1.25 % salt as recommended from the Danish Ministry of Food, Agriculture and Fisheries. The reduction of salt in cheese products might have an effect on the taste, texture, and microbial flora development of the cheeses and it is therefore important to study how technologies, ingredients, and new cultures can work as alternatives for salt and how these parameters relate in low-salt cheese production. Tremendous improvement in analytical techniques has enabled us to study the chemical, biological composition and physical organization of food samples. Above all, techniques based on magnetic resonance spectroscopy are considered as very informative. In particular, high-resolution NMR spectroscopy-based metabolomics is emerging as a powerful approach in agricultural and food science. Many recent studies have demonstrated the usefulness of high-resolution NMR-based metabolomics to study fruits,2 meat,3 milk,4 processing-induced changes,5 and also to study the relationship between metabolite composition and sensory perception of the food.6 In fact, liquid-state high-resolution NMR spectroscopy remains a very common approach to study foods and has gained popularity in the field of metabolomics. However, this methodology has certain limitations for semi-solid foods such as cheese, as they require chemical, physical and/or biological manipulation (i.e. extraction) before analysis. The extraction process is laborious, time consuming, requires large quantity of starting material, and consequently qualitative or quantitative information may be lost. To overcome these limitations, high-resolution magic-angle spinning (HRMAS) NMR spectroscopy could be an appealing technique, as the analysis is performed directly on the sample (semi-solid foods) without any prior extraction. HRMAS NMR-based metabolomics studies have been applied in determination of the geographical origin of Apulian lamb,7 to measure total n-3 fatty acids content in intact fish muscle,8 to detect the aging process of


Parmigiano Reggiano,9 to assess quality and traceability of mozzarella cheese from Campania buffalo milk10 and several other applications in foodstuffs. 11 Among the HRMAS NMR-based metabolomics applications in food science, only few studies are performed on cheese. The aim of the present study was to elucidate the potential use of proton HRMAS NMR spectroscopy for metabolite profiling of Danbo (30+) cheese and to investigate the effects of different rennet and starter culture combinations and extent of brine-salting period. Furthermore, the HRMAS NMR analysis was combined with sensory profiling in order to elucidate possible relations between the cheese metabolite profile and its sensory attributes.


2.1 Cheese and Sampling

All cheeses (n=18) were produced in the pilot plant at Arla Foods, Brabrand, Denmark. These cheeses were prepared with three combinations of starter culture and rennet type i) Frozen Easy-Set (FES) C1050 + CHY-MAX PLUS (Chr. Hansen, Denmark), ii) FES C1050 + CHY-MAX M (Chr. Hansen, Denmark), and iii) FES C1050 + F-DVS HB-3 (EPS producing culture) + CHYMAX PLUS (Chr. Hansen, Denmark), and brine-salted for 6, 12, and 24 hours, respectively. After production, the cheeses were matured at 130C for 12 weeks. Samples of cheese cubes 1cm x 1cm x 5cm cheese were cut. The cheese cubes from the same cheese were then mixed to ensure an even sample distribution and 4 cubes were packed in plastic boxes and stored at 80C before analysis.

2.2 HRMAS NMR analysis

Cheese samples (10 ±3 mg) were cut to fit the 30 μL disposable inserts (Bruker Biospin, Germany) filled with 15 μL D2O containing 0.05 mg/mL 3-(Trimethylsilyl) propionicacid-d4 sodium salt (TSP) for chemical shift referencing. The insert was put into a 4mm zirconium rotor (Bruker Biospin, Germany) and HRMAS NMR spectra were acquired on Bruker Avance III 600MHz equipped with a 1H/13C/31P MAS probe with gradient (Bruker Biospin, Germany). Cpmgpr1D sequence D-90x

o-(t-180 yo-t) l4; where t is fixed echo time (ms) and l4

is loop of the filter (Bruker Biospin GmbH, Germany) was optimized for the cheese samples (described further in section 3.1) and spectra were acquired at 80C probe temperature with the following parameters; 5 kHz spin rate, 256 scans, a spectral width of 10416 Hz collected into 32k data points, an acquisition time of 2.24 s, and a relaxation delay of 5 s. The Free Induction Decay (FID) obtained was multiplied by 0.3 Hz of line broadening before Fourier transformation. The spectra were referenced to TSP (0 ppm), phased, and baseline corrected using the Topspin 3.0 software (Bruker, Rheinstetten, Germany). Assignments of 1H NMR signals were carried out using Chenomx NMR Suite 7.7 (Chenomx, Canada) and the HMDB database. 12

2.3 Sensory analysis

The sensory descriptive profiling was performed by a trained, sensory panel (n = 8) with general experience in assessing food products. The sensory evaluation of the cheeses consisted of 3 training sessions, and 2 descriptive analysis sessions. The first training session was used to acquaint the assessors with the cheeses and enable them to verbalize and discuss attributes, which they found could differentiate between the cheeses. After verbalizing their own list of attributes the assessors were additionally introduced to a list of attributes used by

Foodomics 173

the cheese dairy Arla Foods, Denmark. Final lists of attributes were generated. During the second and third training session, the assessors were given references to some of the attributes and trained on selected samples. The attributes were discussed at the end of each training session to ensure that the all assessors agreed on the meaning of the used attribute. Descriptive analysis was performed in two repetitions following a block design over two days. The following attributes (Table 1) were assessed on a 15 cm unstructured line scale with anchor points on each side going from very low intensity to very high intensity.

Table 1 Sensory Attributes of Danbo (30+) cheese

Overall modality Attributes Description

Aroma Aroma intensity Overall aroma intensity

Emmental aroma Aroma of semi-mature Emmental cheese

Buttermilk aroma Aroma of buttermilk

Butter aroma Aroma of salted butter

Appearance Yellowness Yellowness (ranging from cold yellow to warm yellow)

Texture finger Hardness finger Force needed to compress the cheese sample

Elasticity finger Ability to restore original shape when compressed by finger

Texture mouth Hardness mouth Front teeth bite

Stickiness After 8 chews

Graininess After 8 chews

Melting velocity After 8 chews

Flavour and taste

Sharpness The characteristic sharp taste of mature Danbo cheese

Sourness Buttermilk sour

Saltiness Salty taste

Emmental flavour Flavour of semi-mature Emmental cheese

Butter flavour Flavour of salted butter

Bitterness Bitter taste

After taste Aftertaste Intensity of the overall aftertaste

Overall hedonic

Harmonic How harmonic you perceive the cheese to be

2.4 Multivariate data analysis

NMR spectra were imported to Matlab R2010b (The Mathworks, Inc., USA) and the following spectral preprocessing steps were performed. Initially, misalignments of the spectra were corrected using the algorithm icoshift based on the correlational shifting of spectral


intervals.13 Unwanted regions (16 to 8.5 ppm, 0.8 to -3 ppm) and the region containing water resonance (4.9 to 4.7 ppm) were removed from the aligned spectra. The spectra were normalized to total area to compensate for differences in sample size. These preprocessed spectra were used as the final input for multivariate data analysis. Principal component analysis (PCA), partial least square discriminant analysis (PLS-DA) and partial least square (PLS)-regression were carried out using the PLS Toolbox (Eigenvector Research, USA) in Matlab R2010b (The Mathworks, Inc., USA). At first, PCA of the preprocessed NMR spectra was performed to visualize general grouping, trend, and outliers in the data. Subsequently, PLS-DA was performed to evaluate the metabolic differences between the pre-defined groups. The spectra were Pareto-scaled before the multivariate modeling. The models were validated by full cross-validation and permutation testing. In addition, integrals of selected 1H NMR resonances were performed by peak integration in MatlabR2010b (The Mathworks, Inc., USA). PLS2 regression was carried out with the NMR spectra as x-variables and the means of different sensory profiles as y-variables.


3.1 1H HRMAS NMR spectroscopic analysis of Danbo (30+) cheese

The 1H HRMAS NMR spectra were obtained on the 18 cheese samples included in the study, and a total of 20 metabolites were assigned. The assignments of these metabolites are presented in Table 2 with numbering scheme in Figure 1. The spectral resolution obtained from cheese using HRMAS NMR spectroscopy is comparable to liquid state NMR spectroscopy. 14 Figure 1 shows a representative 1H HRMAS NMR spectrum of a Danbo (30+) cheese sample in D2O. In particular, the HRMAS NMR spectra contained the resonance signals from organic acids, fatty acids, and amino acids. This wide range of compounds can provide chemical, biological or even physical insight on the cheese. The advantage of using HRMAS NMR is the rapid, reproducible, and minimal sample handling. In addition, the sample remains intact and could potentially be used for further analyses. Initially, results showed that the HRMAS NMR spectra of cheese contained broad fatty acids signals that overlaid the signal from low molecular weight compounds (data not shown). To overcome this complexity, a spin echo sequence of appropriate echo time was applied by T2 relaxation editing techniques. The T2 relaxation reduces the signal from fatty acids or macromolecules and enhances the relative signal resolution from low-molecular-weight compounds. Previously, different relaxation editing techniques have been performed to improve the detection of signals from small metabolites in plasma samples.15 In this study, relaxation editing technique (spin echo) was optimized for cheese samples. Figure 2 shows spin echo spectra obtained using 5 different echo times (100 ms, 150 ms, 250 ms, 300 ms, and 350 ms). By increasing the echo time, the overlap due to the lipid signal was reduced. The lipid signals were not suppressed properly at shorter echo time (150 ms, 200 ms, 250 ms), which most likely is due to multiple component relaxation among different fatty acids.15 However, with increases in the effective echo time (300ms, 350ms), the lipid signals were well suppressed and improved signal resolutions from smaller metabolites were obtained. It should be noted that the major disadvantage of increasing effective echo time is a reduction of signal to noise ratio (S: N). Therefore taking all factors into account, an effective time of 300ms appeared to be optimal for sufficient lipid signal suppression and proper signal resolution for these cheese samples.

Foodomics 175

Figure 1 Representative 1HRMAS NMR spectrum of a Danbo (30+) cheese sample. a)Aliphatic region 0.8 to 4.2 ppm. b) Aromatic region 5.0 to 8.5 ppm. The aromatic region in the spectrum (5.0 to 8.5 ppm) has been magnified two times as compared to the aliphatic region (0.8 to 4.2 ppm). Keys to the figure are given in Table 2.

Table 2 1HNMR Chemical shift assignment for Danbo (30+) cheese

S/N Compound Proton Chemical shift in ppm and multiplicity 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Residual lipid* Residual lipid* Leucine Iso-leucine Valine Lactate Acetoin Alanine Acetate Glutamate* Methionine Oxaloacetate

-Ketoglutarate Asparagine* Taurine Glycine Unsaturated lipid* Tyrosine Phenylalanine Formate

0.89(s) 0.94(s) 0.96(d), 0.97(d), 1.7(m) 0.94(t), 1.02(d), 1.27(m) 0.99(d), 1.05(d), 2.3(m) 1.33(d), 4.12(q) 1.37(d), 2.21(s), 4.42(q) 1.48(d) 1.92(s) 2.10(m), 2.36(m), 2.50(m) 2.14(s), 2.16(m) 2.38(s) 2.45(t), 3.01(t) 2.86(m), 2.96(m) 3.25(t), 3.43(t) 3.57(s) 4.96(s) 6.91(d), 7.2(d) 4.0(dd), 7.34(m), 7.38(m),7.44(m) 8.46(s)

* Tentative assignments with best matched signals


Figure 2 1HRMAS NMR spectrum of a Danbo (30+) cheese sample acquired with different echo times (4.5 to 1.2 ppm). Five different echo time were used a) 100 ms b) 200 c) 250 ms d) 300 ms e) 350 ms. The number of loops (l4) used in the spin echo sequence (D-90x

o-(t-180 yo-

t) l4) were changed to obtain the echo times mentioned above. The spectral region 1.4 to 1.2 ppm is zoomed twice compared to the region 4.5 to 1.35 ppm. 3.2 Multivariate data analysis

In order to study the metabolic changes due to different processing conditions on Danbo (30+) cheese, multivariate data analysis was applied on the obtained HRMAS NMR data. PCA was carried out on the Pareto-scaled normalized 1H NMR spectra to explore the metabolite differences due to different rennet and starter culture combinations and extent of the salting period. A pronounced effect of salting period was observed, while the effects of rennet and starter culture on the metabolite profile of the cheese were less evident (data not shown). PCA performed on the 18 cheese sample brine-salted for 6, 12, and 24 hours showed distinct clustering between 24 hours and 6 hours salted cheese, when compared to samples salted for 12 hours (Figure 3a & b). To characterize the distinct metabolite differences, PLS-DA between cheese samples salted for 24 hours, 12 hours and 6 hours was performed. The PLS-DA results showed a high CV accuracy for separating 24 hours and 6 hours cheese spectra (84%, P < 0.02), while the classification of the brine-salting for 12 hours vs. 6 hours spectra and the brine-salting for12 hours vs. 24 hours spectra was less successful and non-significant. The score plot (Figure 3c) depicts distinct clustering of 24 hours and 6 hours salted cheeses along latent variable one (LV1). The samples salted for 24 hours were characterized by a higher level of lactate, while samples obtained after 6 hours of salting had a higher level of the metabolites acetoin, acetate, alanine, and tyrosine (Figure 3d).

Foodomics 177

Figure 3 PCA and PLS-DA score plots and loadings. a) Score plot of the first and second PC from 18 cheese samples salted for 6, 12, and 24 hours, respectively. 6 hours salted cheese,

12 hours salted cheese and 24 hours salted cheese samples. b) Loading profile of second principal component. c) PLS-DA score plot of the first and second latent variable (LV) from 12 cheese samples salted for 6 and 24 hours. d) Loading profile of second latent variable.

It is well known that during cheese production and maturation the biomolecules (fat, carbohydrates and proteins) are transformed into low-molecular-weight compounds. Many of these low-molecular-weight compounds are responsible for the flavor of the cheese. Lactate is one of the most abundant organic acids in the cheese and the principle metabolite derived from lactose fermentation. The change in lactate content is indicative of pH changes and indirectly linked to texture.16 In previous studies, salt-tolerant lactic acid starter bacteria have shown to decrease the pH of the cheese significantly by fermenting available lactose to lactate.17 In our study, we suggest that the increased level of lactate in samples salted for 24 hours may be associated with the increased lactose metabolism by salt tolerant starter lactic acid bacteria. Likewise, the samples salted for 6 hours had a lower level of lactate; this result might be due to increased consumption of lactate by non-starter microbes or could also be due to further metabolism of lactate to other organic metabolites, such as acetate and formate by heterofermentive pathways.16, 17 Intriguingly, our result also showed an elevated acetate content in samples salted for 6 hours. Acetate is important for flavor in many cheeses. Acetate production in cheese may be due to oxidation of lactate to acetate by non-starter culture bacteria or may also be formed from lactate or amino acids by the starter culture.16 In addition to lactate and acetate, the loading plot in Figure 3d reveals that acetoin is an important metabolite differing among cheese samples salted for 6 and 24 hours, respectively. Acetoin is a pleasant buttery odor and neutral compound that is produced from


decarboxylation of acetoacetate.19 The production of acetoin is related to neutralization of the acidification of the intracellular environment and culture medium.20 Thus, the higher amount of acetate, which is acidic, in samples salted for 6 hours compared to 24 hours could possibly have stimulated a higher acetoin production by the fermentative bacteria. In addition, tyrosine and alanine levels were also elevated in samples salted for 6 hours. Amino acids such as alanine and tyrosine are known to add flavor to the cheese during the ripening process. Interestingly, alanine is known to yield pyruvate and to be converted into diacetyl, which may be fermented to acetoin.20, 21 This fact is in agreement with our findings, as the elevated level of acetoin was evident in samples salted for 6 hours. The loading plot in Figure 3d also shows a higher level of tyrosine in samples salted for 6 hours. Intriguingly, a study by Bu ková etal. 2011 showed that bacterial tyrosine decarboxylation was highly influenced by the salt concentration; the highest activity of bacterial tyrosine decarboxylation occurred at the highest salt concentration applied.22 Thus, we assume that samples salted for 6 hours exhibited lower bacterial tyrosine decarboxylation compared to 24 hours, resulting in a higher level of tyrosine in these cheeses.

3.3 Correlation between sensory attributes and cheese metabolites In order to correlate the cheese metabolites detected by HRMAS NMR with the sensory attributes of the cheeses, PLS2 regression was carried out between NMR data and the means of the sensory profiles, using the integrals of selected 1H NMR resonances identified in the cheese spectra. The result showed that the butter milk aroma and butter aroma were the main sensory attributes separating different combinations of the rennet and starter culture (Figure 4). Intriguingly, a correlation between the cheese content of acetoin and the sensory attributes butter milk aroma (R2 = 0.547), and butter aroma (R2 = 0.524) could be established. Acetoin is known to be the fermented product derived from diacetyl with distinct butter aroma.20 Previously, diacetyl formation in fermented milk products depended on the presence of microaerophilic bacterium species.22 Thus, our finding substantiates that the acetoin content of cheese relates to butter aroma and butter milk aroma as a result of different combinations of the rennet and starter.

4 CONCLUSION

Overall, the present study demonstrated that the combination of HRMAS NMR spectroscopy and multivariate analysis enabled discrimination of semi-hard Danbo cheese samples produced by varying duration of the brine-salting period. In addition, NMR and sensory data could be correlated to unravel that acetoin is the main cheese metabolite describing sensory differences obtained as a result of different rennet’s and starter cultures. Acknowledgement The authors wish to thank the Danish research council FTP for financial support through the project ‘Advances in food quality and nutrition research through implementation of metabolomics technologies (#274-09-0169)’ and the Danish Dairy Foundation, Arla Foods, and ‘Future Food Innovation’ for financial support through the project ‘A pinch of salt’.

Foodomics 179

Figure 4 PLS loading plot illustrating the means of different sensory profiles and selected metabolites in the cheese sample here represents the metabolites and the sensory attributes.

References

1 J.S. Madsen and Y. Ardö, Int Dairy J., 2001, 11, 4. 2 D. Cuthbertson, P.K. Andrews, J.P. Reganold, N.M. Davies and B.M. Lange, J Agr Food

Chem., 2012, 60, 35. 3 Y. Jung, J. Lee, J. Kwon, K.S. Lee, D.H. Ryu and G.S. Hwang. J Agr Food Chem., 2012,

58, 19 4 U. K. Sundekilde, P.D. Frederiksen, M.R. Clausen, L.B. Larsen and H.C. Bertram, J Agr

Food Chem, 2011, 59, 13. 5 R. Beleggia, C Platani, R. Papa, A Di Chio, E. Barros, C. Mashaba, J. Wirth, A.

Fammartino, C. Sautter, S. Conner, J. Rauscher, D. Stewart and L. Cattivelli, J Agr Food Chem., 2011, 59, 17.

6 I.K. Straadt, M.D. Aaslyng and H.C. Bertram, Meat Sci., 2014 96, 1. 7 D. Sacco, M.A. Brescia, A. Buccolieri and A.C. Jambrenghi, Meat Sci., 2005, 71 3. 8 G. Nestor, J. Bankefors, C. Schlechtriem, E. BräNnäS, J. Pickova, C. Sandström, J

Agr Food Chem., 2010, 58, 20. 9 L. Shintu and S. Caldarelli, J Agr Food Chem., 2005, 53, 10. 10 P. Mazzeiand A. Piccolo, 2012, Food Chem., 132, 3. 11 M, Valentini, M. Ritota, C. Cafiero, S. Cozzolino, L. Leita, and P. Sequi, Magn Reson

Chem., 2011, 49, 1. 12 D. S. Wishart, C. Knox, A. C. Guo, R. Eisner, N. Young and B. Gautam, et al. Nucleic

Acids Res., 2009. 37. 13 F. Savorani, G. Tomasi and S.B. Engelsen, J. Magn. Reson., 2010, 202. 14 L. Shintu, F. Ziarelli and S. Caldarelli, Magn Reson Chem., 2004, 42, 4.


15 H. Tang, Y. Wang, J.K. Nicholson and J.C. Lindon, Anal Biochem., 2004, 325, 2 16 C. Marincola, F.C. Savorani, F. Engelsen, S.B. Cosentino, S. Viale, S. Pisano and M.B.

Food Chem., 2013, 141, 3. 17 S. Agarwal, J.R. Powers, B.G. Swanson and S. Chen, S. Clark, J Dairy Sci., 2008, 91, 8. 18 E Lechner, F. Ginzinger, W. Rohm, H and E. Microbiology and fermentation

compounds.,1999, 79 4 19 M. Curic, B.S. Lauridsen, P. Renault and D. Nilsson, App Environ Microb., 1999, 65, 3. 20 Z. Xiao and P. Xu, Crc Rev Microbiol., 2007, 33, 2. 21 D.L. Bars and M. Yvon, J Appl Microbiol., 2008, 104, 1. 22 L. Bu ková, F. Bu ka, E. Pollaková, T. Podešvová and V. Dráb, Int J Food Microbiol.,

2011, 147, 2.

CHANGES IN THE 1H NMR METABOLIC PROFILE OF MUSSELS (Mytilus galloprovincialis) WITH STORAGE AT 0°C

Violetta Aru1, Maria Barbara Pisano2, Paola Scano1, Sofia Cosentino2 and Flaminia Cesare Marincola1

1 Dipartimento di Scienze Chimiche e Geologiche - Università degli Studi di Cagliari, S.S. 554 Bivio Sestu, 09042 Monserrato (Italy) 2Dipartimento di Sanità Pubblica, Medicina Clinica e Molecolare - Università degli Studi di Cagliari, S.S. 554 Bivio Sestu, 09042 Monserrato (Italy)

1 INTRODUCTION

Mussels belonging to the Mytilus galloprovincialis specie are very popular shellfishes harvested in the Mediterranean Sea. These bivalves are highly appreciated globally due to their exceptional nutritional value making them ideal for the human diet. Indeed, mussels are rich in minerals and vitamins (A, B1, B2, B6, B12 and C) (1, 2). Furthermore, mussel fat is rich in polyunsaturated fatty acids (PUFA, 37–48% of total fatty acids, mainly 3) (3). These fatty acids are biologically important and have been associated with a decreased risk of cardiovascular disease (4). Although mussels are available in the market under all the possible forms (frozen, vacuum-packed, pickled, smoked, and canned), most of the mussels are kept alive on ice or refrigerated (2-4°C) until consumed. The shelf life of mussels is limited, primarily due to a variety of microbial and biochemical degradation mechanisms depending on the duration and conditions of storage as well as the initial quality of the product (5, 6). There are many well established traditional analytical techniques and methods available to assess the quality of seafood (7), including sensory evaluation based on quality index method, microbial inspection based on total viable counts, biochemical methods related, and proteome analysis. Although chemical and microbiological methods are useful both for research and product development, they are not practical for routine use, as they require expensive laboratory equipment and trained staff, are destructive, and can be labor intensive and time consuming. In order to surmount the aforementioned disadvantages, fast reliable methods are therefore necessary for assuring freshness specification of the starting material and making sure that the product will not become stale when distributed and displayed. Nuclear Magnetic Resonance (NMR) spectroscopy is a technique able to provide detailed chemical information on a wide range of compounds simultaneously present in a sample. In combination with multivariate analysis, NMR is recognize to make available relevant information on the composition of food in many area of food science such as foodstuffs quality, raw material safety, and authentication (8). The goal of this work was to investigate the NMR metabolic changes in the aqueous extracts of Mytilus galloprovincialis samples stored at 0° over a period of 7 days and to find putative metabolites-markers influenced by the conservation conditions. In order to achieve a more complete characterization of mussel spoilage, the most significant microbial groups were also investigated during storage.


2. MATERIAL AND METHODS

2.1 Chemicals

All solvents used, of the highest available purity, were purchased from Merck (Darmstadt, Germany), while deuterium oxide (D2O, 99.9%) and sodium 3-trimethylsilyl-propionate-2,2,3,3,-d4 (TSP) were acquired from Sigma–Aldrich (Milan, Italy).

2.2 Samples

Live mussels (Mytilus galloprovincialis) were obtained from the local seafood market in Cagliari (Italy) and immediately transported to the laboratory in portable coolers at approximately 4°C. In the laboratory, mussels were washed and scrubbed under running potable water to remove debris; dead mussels or those with broken shells were discarded. The remaining mussels (70 individues) were manually shucked by cutting the adductor muscle with a sterile knife and each sample was put into insulated sterile plastic boxes without ice or water. Mussels were stored for 7 days at 0°C. Microbiological analyses were performed at 0, 2, and 6 days intervals, while the extraction of the hydrosoluble component to be analysed by NMR spectroscopy was carried out on days 0, 3, and 7.

2.3 Microbiological and statistical analyses

A 100 g sample, composed by the tissues and shell liquor from mussels, was mixed with an appropriate volume (1/2) (w⁄v) of 0.1% peptone diluent (Biolife). The mixture was homogenized in a stomacher (Stomacher Lab-Blender 400, PBI, Italy) for 1 min at normal speed. A volume of 70 ml of diluent was added to 30 ml of the homogenate to make a master dilution. After mixing, further decimal dilutions in 0.1% diluent were made and plated in specific media to enumerated microbial groups. The analysis for Escherichia coli was performed according to the International Organization for Standardization, using the five-tube, three-dilution most-probable-number (MPN) method (9). Briefly, 10 ml volumes of the 10-1 dilution were added to each of five tubes of double strength Mineral Modified Glutamate Broth (MMGB, Biolife), 1 ml of the 10-1 dilution was added to each of five tubes of single strength MMGB and 1 ml of the 10-2 dilution was added to a second set of five tubes of single strength MMGB. The inoculated tubes were incubated at 37 °C for 24 h. Tubes showing any sign of a yellow coloration were sub-cultured onto Tryptone Bile X-glucuronide medium (TBX, Oxoid) and incubated at 44°C for 22 h. Tubes from which subcultures yielded blue or blue-green colonies on TBX were deemed positive. MPN tables (10) were used to calculate E. coli numbers per 100 g of flesh and inter-valve liquid (11). The limit of assay sensitivity was a MPN of 20 E. coli cells per 100 g of shellfish. Negative samples were reported as MPN < 20/100 g. Total counts of mesophilic (MMC) and psychrotrophic (PMC) microorganisms were determined by the pour plate method, using Plate Count Agar (PCA, Microbiol, Cagliari, Italy) and incubating at 30 °C for 48 h and at 4 °C for 7 days, respectively. Total Enterobacteriaceae levels were determined by pour plating on Violet Red Bile Glucose Agar (Microbiol) and incubating at 30° C for 24–48 h. Qualitative Vibrio parahaemolyticusanalysis was carried out on 25 g of blended mussels which were added to 225 ml of phosphate buffered peptone water, homogenized and incubated for 24 h at 37 °C. A loopful of peptone water was streaked onto Thiosulphate Citrate Bile Salt Agar (TCBS, Oxoid) and incubated for 24 h at 37 °C. The suspected colony types (yellow and green) were picked out, streaked on to TCBS and incubated at 37 °C for 24 h. Isolates were examined for Gram

Foodomics 183

reaction, cell morphology, moltility, oxidase and catalase reaction and for behavior in TSI (Triple Sugar Iron Agar). They were then grouped according to the criteria of Mossel et al. (12) and identified by API 20E and API 20NE (Biomérieux, Marcy l’Etoile, France). The galleries were incubated at 30 °C for 24-48 h and the API profiles were compared with the analytical Profile Index (Apilab plus version 3.3.3) The presence of food-borne pathogens Salmonella spp. and Listeria monocytogenes was also investigated. Twenty-five grams of sample were diluted in 225 ml of buffered peptone water, homogenized as previously described and incubated for 18 h at 37 °C for the detection of Salmonella spp. Another 25 g of sample was diluted in Half-fraser broth, homogenized and then incubated for 24 h at 30 °C for the detection of L. monocytogenesaccording to the ISO methods (13,14). One-way ANOVA was performed on microbiological data for each storage time. Tukey test for multiple comparisons was used to separate treatment means. All statistics were performed using GraphPad Prism Statistics software package version 3.00 (GraphPad Prism Software Inc., San Diego, CA, USA). Statistical significance was inferred at P < 0.01.

2.4 pH measurement of the samples

The pH of the homogenized samples (100 g in 200 ml of 0.1% peptone diluent) was measured with a HI8520 pH meter (Pool Bioanalysis Italiana, Milan, Italy).

2.5 Sample preparation for NMR analysis

Water-soluble metabolites were extracted from mussels using a methanol/chloroform/water mixture according to the Folch method (15). Each mussel sample was dissolved in 12 mL of a mixture chloroform–methanol (2:1,v/v). After the addition of 4 mL of H2O and centrifugation at 1300 rpm for 1 h at 4 °C, the methanol/water mixture, containing the low molecular weight water-soluble components, was separated from the chloroform fraction, containing the lipid components. The CH3OH/H2O phase was dried using a rotary evaporator. The water-soluble fraction was dissolved in 1.2 mL of a D2O solution of the internal standard (TSP) 0.80 mM. The pH of the final sample was accurately adjusted to 6.52 ± 0.03. Then, an aliquot of 650 μL was placed into a 5 mm NMR tube for NMR analysis.

2.6 NMR measurements

1H NMR spectra were recorded at 300 K using a Varian Unity Inova 500 MHz NMR spectrometer (Agilent Technologies, CA, USA) operating at 499.839 NMR spectra were acquired with a sweep width of 6000 Hz, a 45° pulse, an acquisition time of 1.5 s, a relaxation delay of 4 s, and 256 scans. The residual signal of H2O was suppressed by applying a presaturation technique with low power radiofrequency irradiation for 1.5 s. After Fourier transformation with 0.3 Hz line broadening and a zero-filling to 64 K, 1H spectra were phased and baseline corrected. Spectral chemical shift referencing on the TSP CH3signal at 0.00 ppm was performed in all spectra.

2.7 NMR data preparation and analysis

For multivariate statistical analysis, the NMR spectra were manually phased and baseline corrected using MestReNova (Version 8.1, Mestrelab Research SL). Each NMR spectrum was used to construct a data matrix by subdividing it into regions having an equal bin size of 0.02 ppm over a chemical shift range of 0.5-9.5 ppm. The regions between 4.6 and 5.2 ppm


and 0.5 and -0.5 ppm were excluded for multivariate data analysis because of the signals of water and TSP, respectively. Bins were normalized to the sum of total spectral area to compensate for the overall concentration differences. The final data set was automatically reduced to ASCII files, converted into an Excel file and then imported to SIMCA version 13.0 (Umetrics, Umea, Sweden) for statistical analysis. Data were Pareto scaled and analyzed using principal component analysis (PCA) (16) and orthogonal partial least squares discriminant analysis (OPLS-DA) (17). Unsupervised PCA was used for the overview of all data set, to observe clustering or separation trends and identify outliers, while supervised OPLS-DA was performed to remove variability not relevant to class separation. OPLS-DA is a classification technique that uses class information to maximize the separation between groups of observations. With respect to PCA, this method allows improved interpretation of the variations between discriminated groups, by removing information that has no impact on discrimination. Briefly, OPLS-DA facilitates the separation of the systematic variation in X into two parts, one that is linearly related to Y (predictive information) and one that is unrelated to Y (orthogonal information). The predictive information of Y in X is concentrated in the first predictive component and is associated with the between groups variation, while the variation in X which is unrelated to Y is put in the second and orthogonal component and is linked to the within groups variation. The quality of the model is described by the parameter R2Y and Q2Y, which represent the fraction of the variation of Y-variable and the predicted fraction of the variation of Y-variable, respectively. The statistical significance of R2Y and Q2Y values is estimated through a response permutation testing (18). In this test, the Y-matrix is randomly re-ordered, while the X-matrix is kept constant. This means that the Y-data remain numerically the same, but their positions are shifted by random shuffling. Each time a new OPLS-DA model is fitted using X and the permuted Y matrix, providing a reference distribution of R2Y and Q2Y for random data. An R2Y intercept value less than 0.3-0.4 and a Q2Y intercept value less than 0.05 are indicative of a valid model.

3. RESULTS AND DISCUSSION

3.1 Microbiological analysis

Changes in microbial flora of mussels during storage in ice are shown in Table 1. We identified 58 of the 65 isolates from the samples (Figure 1). The total viable counts of mesophilic and psychrotrophic bacteria significantly increased (P < 0.01) at a similar rate throughout the storage period at 0°°C, both microbial groups showing a slow growth. Enterobacteriaceae are active seafood spoilers that grow rapidly and become predominant in spoiling fish (19). Therefore, the contribution of Enterobacteriaceae to the microflora of mussels and its spoilage potential must be taken into consideration, especially in the case of polluted water. Although Enterobacteriaceae can grow at low temperatures, the storage at 0°C, under our experimental conditions, provided a good control of this group. Indeed, the corresponding levels decreased over time up to being below the detection limit of 100 CFU/g at 6 days of storage.

Pre-harvest contamination with pathogens from the animal/human reservoir (Salmonella,Shigella, E.coli, enteric virus) may pose a risk factor for humans and animals since in some cases a very low infective dose is required to cause illness (20). In the samples under investigation, E. coli concentrations were lower than the limit of 20 MPN per 100 g throughout all storage period, Salmonella and L. monocytogenes were never isolated, while V. parahaemolyticus was present in one sample.

Foodomics 185

Table 1. Microbial changes in Mytilus galloprovincialis samples stored at 0 °C.Days of storage

Microbial parameters* Mesophilic

bacteria Psychrotrophic

bacteria Enterobacteriaceae E. coli**

0 3.08 ± 0.07 2.30 ± 0.05 2.48 ± 0.35 <20 2 3.50 ± 0.17 2.60 ± 0.05 2.48 ± 0.16 <20 6 3.90 ± 0.30 3.08 ± 0.07 <2 <20

Values are means ± S.D of log CFU per gram **Values are MPN per 100 per 100 g using ISO method 16649-3

Figure 1 Distribution % of the species isolated from M. galloprovincialis samples stored at 0 °C

3.2 pH

The change in pH of seafood is usually a good index for quality assessment. Indeed, it is commonly related to the accumulation of lactic acid generated by glycogen in anoxic condition and/or the accumulation of basic substances, such as ammonia and trimethylamine (TMA), mainly derived from microbial developments. Under our experimental conditions, the pH of mussels slightly decreased from 6.56 to 6.17 during 7 days of storage at 0°C, suggesting a prevalent contribution from glycogen degradation.

3.2 NMR Spectroscopy

A representative 1H NMR spectrum of the aqueous phase extracted from mussels is shown in Figure 2. The assignment of the peaks was based on data reported in the literature (21,22). The spectra indicated the presence of numerous metabolites including amino acids, organic osmolytes, organic acids, alcohols, alkaloids, sugars and other carbohydrates.


Figure 2 Typical 1H NMR spectra of the aqueous extract of mussels. Peaks: 1. Branched Chain Aminoacids (Isoleucine, Leucine, Valine), 2. unknown, 3. Lactate, 4. Alanine, 5. unknown, 6. Acetate, 7. Methionine, 8. Succinate, 9.Trimethylamine, 10. Betaine/Taurine/Trimethylamine N-Oxide, 11. Taurine, 12. Glycine, 13. Betaine, 14. Homarine, 15. Glucose, 16. Glycogen, 17. Fumarate, 18. Tyrosine, 19. Phenylalanine, 20. ADP.

An exploratory unsupervised PCA analysis of the overall data set was firstly performed in order to achieve the natural interrelationship (grouping, clustering, or outlier detection) among samples. The model showed noticeable overlaps among samples, thus preventing any class differentiation in terms of storage time, possibly because of the structure variation within each sampling (data not shown). By applying OPLS-DA and using storage days as classifiers, a better separation was obtained in the scores map (Figure 3) with the following parameters resulting from internal cross validation: R2Y 0.981 and Q2 0.704. The goodness of fit and the predictability of this result were subjected to validation to test the possibility of correlation by chance. Thus, the model was subjected to ‘‘Y-scrambling” statistical validation. We randomly permutated the Y–variable, re-built the statistical model, and observed the trends of the predictive power and goodness of fit at each step. Two hundred rounds of such reshuffling gave coherent decreases in both parameters. The result of permutation test indicated that the OPLS-DA model was statistically sound and that the predictability was not due to over-fitting of the data (intercepts: RY2 = 0.101; Q2 = -0.216).

In order to further identify the key metabolic changes occurring during the storage time at 0°C, OPLS-DA models were generated in pairwise comparisons (Figure 4). All models showed satisfactory values of the parameters that evaluate the explained variance and the prediction capability (see figure legend). The S-plots of the OPLS-DA models were used for identification of potential markers of group separation. S-plot is a scatter plot which

Foodomics 187

Figure 3 OPLS-DA scores plot applied to the 1H NMR spectra of the hydrosoluble component of mussels. Each score represents a NMR spectrum at a certain storage time at 0°C: 0 ( ), 3 ( ) and 7( ) days. The quality factors for this models were R2X = 0.935, R2Y= 0.981, and Q2 = 0.704.

combines the covariance and correlation loading profiles arising from the predictive component of the OPLS-DA model. In the S-plot, each point represents a variable (i.e. bin); the X axis represents variable contribution (covariance, p), where the farther the distance the variable points from zero, the more the variable contributes to the difference between twogroups; the Y axis represents variable confidence (correlation, p(corr)), where the farther the distance the variable points from zero, the higher the confidence level of the variable to the difference between two groups. So, the variable points with high correlation and covariance values are located in regions far away from the origin. The variable importance coefficient (VIP) value reflecting the influence of each variable on the classification was used as an additional selection criterion using the “greater than one” rule. In the S-plot, the range of the variables selected is highlighted with a dotted rectangle.

In particular, the time-related metabolic signatures of mussels stored at 0°C were characterized mainly by alterations in the concentration of amino acids, organic acids, and osmolites: taurine, homarine, and betaine were reduced over time, whereas acetate, alanine, glycine, succinate, and TMA were elevated with respect to fresh mussels. TMA, the most commonly used metabolite to evaluate the freshness of fish, is formed from bacterial use of trimethylamine-N-oxide (TMAO), naturally occurring osmoregulating substance found in most marine fish species. This compound is found in very low levels in fresh fish, and its formation is associated with bacterial spoilage (23). Since, as seafood spoils, proteins are broken down to peptides, free amino acids, amines and volatile ammonia (24), the increase of alanine and glycine can be also involved in the microorganism development. Although glucose and lactate seem to be the substrates which are attacked first by the various groups of spoilage bacteria under aerobic and anaerobic conditions, the changes in their levels over storage at 0°C did not show statistical significance.


Figure 4 OPLS-DA scores plots (A-B) and their respective S-plots (C-D) from the 1H NMR spectra of the hydrosoluble component of mussels on 0 ( ), 2 ( ) and 7 ( ) days of storage at 0°C. The quality factors for these models were: R2X = 0.681, R2Y= 0.893, and Q2 = 0.75 (right side); R2X = 0.851, R2Y= 0.882, and Q2 = 0.544 (left side).

4 CONCLUSIONS

The results of the present study indicated that the NMR spectroscopy in conjunction with multivariate analysis is a feasible approach to analyze the time-related changes of the metabolic profile of mussels stored at 0°C. Although the observed modifications are in good agreement with the occurrence of microbiological development, as pointed out by the microbial counts, an accurate interpretation of the spoilage influence on the metabolic fingerprint of mussels is a difficult task. One of the reason is that the changes in the metabolic profile arise from the complex interplay of different biochemical processes, related to the concomitant actions of different microorganisms. Further integration of biological and chemical data is needed to demonstrate the utility of this analytical approach in providing rapid, accurate, and quantitative results on microbiological spoilage.

References

1 P. A. Karakoltsidis, A. Zotos and S.M. Constantinides, J. Food Comp. Anal., 1995, 8,258.

2 K. Vareltzis, Fish. News, 1996, 11, 38.3 E. Orban, G. Di Lena, T. Nevigato, I. Casini, A. Marzetti and R. Caproni, Food Chem.,

2002, 77, 57. 4 P. M. Kris-Etherton, W. S. Harris and J. A. Lawrence, Circulation, 2002, 106, 2747.

Foodomics 189

5 A.E. Goulas, I. Chouliara, E. Nessi, M.G. Kontominas and I.N. Savvaidis, J. Appl. Microbiol., 2005, 98, 752.

6 M. A. Khean, C. C. Parrish and F. Shaidi, J. Agric. Food Chem. 2005, 53, 7067. 7 G. Blafsdbttir, E. Martinsdbttir, J. Oehlenschbger, P. Dalgaard, B. Jensen, I. Undeland,

I.M. Mackie, G. Henehan, J. Nielsen and H. Nilsen, 1997, Vol. 8. 8 L. Laghi, G. Picone and F. Capozzi, Trend Anal. Chem., 2014, 59, 93. 9 ISO/TS 16649-3:2005. Microbiology of food and animal feeding stuffs - Horizontal

method for the enumeration of beta-glucuronidase-positive Escherichia coli - Part 3: Most probable number technique using 5-bromo-4-chloro-3-indolyl-beta-D-glucuronide. Brussels: European Committee for Standardization.

10 T. J. Donovan, S. Gallacher, N. J. Andrew, M.H. Greenwood, J. Graham, J. E. Russel, D. Roberts and R. Lee, Comm. Dis. Publ. Health, 1998, 188, 196.

11 T.L. Maugeri, M. Carbone, M.T. Fera, G.P. Irrera and C. Gugliandolo, J. Appl. Microbiol., 2004, 97, 354.

12 D. A. A. Mossel, J. E. Corry, C. B. Struijk and R. M. Baird in Essentials of the microbiology of foods: a textbook for advanced studies. John Wiley and Sons, Chichester (England), p 699.

13 EN ISO 6579:2004. Microbiology of Food and Animal Feeding Stuffs – Horizontal Method for the Detection of Salmonella spp. Brussels: European Committee for Standardization.

14 ISO/DIS 11290-1: 2001. Microbiology of the food chain -Horizontal method for the detection and enumeration of Listeria monocytogenes and other Listeria spp. -Part 1: Detection method. Brussels: European Committee for Standardization.

15 J. Folch, M. Less and G. H. S. Stanley, J. Biol. Chem., 1957, 226, 497. 16 H. Hotelling, J. Educ. Psychol., 1933, 24, 417.17 M. Bylesjo, M. Rantalainen , O. Cloarec, J.K. Nicholson, E. Holmes and J. Trygg, J.

Chemometr., 2006, 20, 341. 18 L. Eriksson, E. Johansson, N. Kettaneh-Wold, J. Trygg, C. Wikström and S. Wold.

Multi- and megavariate data analysis: Principles and applications. Umetrics AB, Umeå, Sweden, 2006.

19 L. Gram and H.H. Huss, Int. J. Food Microbiol., 1996, 33, 121. 20 G. Teophilo, R. dos Fernandes Vieira, D. dos Prazeres Rodrigues and F. R. Menezes Int.

Microbiol., 2002, 5, 11. 21 S. J. Rochfort, V. Ezernieks, A. D. Maher, B. A. Ingram and L. Olsen, Food Res. Int.,

2013, 54, 1302. 22 W. Tuffnail, G. A. Mills, P. Cary and R. Greenwood, Metabolomics, 2009, 5, 33. 23 T. A. Gill, Food Rev. Int., 1990, 6, 681. 24 L. Gram, L. Ravn, M. Rasch, J. Bartholin Bruhn, A. B. Christensen and M. Givskov, Int.

J. Food Microbiol., 2002, 78, 79.

APPLICATIONS OF 1H-NMR METABOLOMICS: FROM INDIVIDUAL FINGERPRINTS TO FOOD ANALYSIS

Claudio Luchinat1*, Leonardo Tenori2

1Magnetic Resonance Center (CERM), University of Florence, via Luigi Sacconi 6, 50019, Sesto Fiorentino (FI), Italy 2FiorGen Foundation, via Luigi Sacconi 6, 50019, Sesto Fiorentino (FI), Italy *for correspondence: [email protected]

1. Introduction

Metabolites are the final products of cellular activities and their levels in a living organism can change according to genetic or environmental factors. The set of metabolites present by a biological system is called “metabolome”1. The metabolome can be considered the final product of the complex interactions of the genome, transcriptome, proteome and the environment: these interactions can be regarded as a cascade linking the genome to the phenotype1. The metabolome, consisting of low-molecular weight chemical compounds, can be considered an amplified version of gene expression: for this reason the metabolite space represents an optimal level at which to analyze changes in biological systems with high sensitivity2. The most common biological specimens used in metabolomics are serum/plasma and/or urines, because they can be collected with low invasiveness, and are rich in biological information at the systemic level. A number of other biofluids such as saliva3, tissue extract4, cerebrospinal fluid5, bile6, seminal fluid7, amniotic fluid8, synovial fluid9, exhaled breath condensate10,11 and faecal extracts12 can also be studied. These biofluids are easily available and are used for many other clinical analyses. Mass spectrometry (MS) and proton nuclear magnetic resonance spectroscopy (1H-NMR) represent the techniques most commonly employed in metabolomics. The NMR spectrum of a body fluid can be also called metabolic profile, and constitutes a “fingerprint” of the NMR-detectable part of the whole metabolome. From mono-dimensional 1H-NMR spectra it is possible to extract a disease signature using multivariate statistical analysis13,14,15 and this feature could confer to metabolomics a key role in disease diagnosis, prognosis and for monitoring drug therapies. Finally, NMR is a relatively high-throughput methodology that requires minimal sample handling, allowing the simultaneous detection of a large number of different metabolites in a short time. Metabolomics has already proved in the past decade to be useful for the characterization of several pathologies, and offers promises as a clinical tool16, because it is based on the analysis of the measured dynamic changes of a living organism in response to genetic modifications or physiological stimuli such as nutrients, drugs treatment or toxic insults17,18. Metabolomics has provided new information on a wide range of pathologies, such as cancer19, meningitis20, neurological disorders21, cardiovascular diseases22, inborn errors of metabolism23, and coeliac disease14,24.

Foodomics 191

2. The individual human metabolic phenotype

The evidence of the existence of a stable individual metabolic phenotype25,26 (metabotype) opens new interesting perspectives for human health. The metabolic phenotype is defined as a “multiparametric description of an organism in a given physiological state based on metabolomic data”27. The analysis of individual phenotypes over the years demonstrates that the metabolic phenotypes are the result of complex interactions between environmental factors and genetics. An important finding that follows is the fact that it is possible to recognize NMR profiles/spectra as belonging to a unique subject with near 100% accuracy25 even over a period of some years26: the effects of the environment are observable in the sporadic and periodical metabolic fluctuations, while a combination with genetic factors is present in the stable, invariant part of metabolic phenotype. It is well-known that the genotype plays a key role in defining the individual metabolic phenotype, but the latter is also modulated by gut microflora and external stimuli. So individual metabolic phenotypes describe the fact that each individual has an unique adaptive response to the environment. It is also demonstrated that the metabolic phenotype seems to be stable during a period of several years26. Metabolic fluctuations, due to the action of gut microflora and/or external stimuli and lifestyle, have been also identified. These variations can be called “jumps”, “waves” and “spikes”26. The first two terms are probably related to gut microflora: (i) “jumps” are sudden changes in signal intensities that persist for a long period. Jumps have been observed in the hippurate aromatic signals, suggesting changes in the activity and/or composition of the gut microflora. (ii) “Waves” are signals whose variation in intensity is more gradual and persists for a number of days. Waves can be observed in the concentration of hippurate and m-hydroxyphenylpropionic acid. Also these variations could be due to changes of the endogenous gut microflora population. (iii) Spikes are signals that suddenly appear in single spectra having markedly different intensity from the preceding and following days: they are most often caused by sporadic assumption of particular food components, (e.g. high alcohol or meat intake) or particularly intense physical activity26. Applying hierarchical cluster analysis, it was shown26 that genetically related individuals (father and son and twins), are metabolically similar than unrelated individuals. Genetic influence can account for the confusion between metabolic phenotypes of two twins. Moreover, for genetic reasons, identical twins could show higher similarity in gut microbiota than genetically unrelated individuals. Overall, the individual metabolic phenotype can be considered a metagenomic entity that links host metabolic phenotype and gut microbiome, that is, the host is characterized by a gut microflora compatible with the own genome. Characterizing individual metabolic fingerprints may allow researchers to follow phenotype changes as a function of disease outcome and progression, possibly leading to earlier diagnosis and prognosis, creating a new paradigm in personalized medicine.28

3. Metabolomics for diseases fingerprinting

3.1. Coeliac disease Coeliac disease (CD) is a common complex chronic immune-mediated disorder with a known (gluten) environmental trigger. Recent surveys indicate that it may affect, for example, 1 in 105 subjects in the United States29, 1 in 67 Finnish schoolchildren30 and 1 in 230 in Italian school age children31, with seroprevalence of 1-3% in subjects of white European origin. The environmental trigger of coeliac disease is gluten, a protein complex formed by gliadin and


glutenin, which is found in wheat and related grain species like barley and rye, and also in processed food where it is used to enhance food texture and as a stabilizing agent. To date, a limited number of metabolomics studies of coeliac disease are available32, but they clearly show that metabolic differences between healthy individuals and coeliac patients exist. Bertini et al.13, examined adult healthy controls and coeliac patients by 1H NMR profiling of their serum and urine profiles before and after gluten free diet, showing that a metabolic fingerprint for CD can be defined. This fingerprint was found to be made up by three components; one related to malabsorption, one related to energy metabolism, and the third related to alterations in gut microflora and/or intestinal permeability. Using this metabolic fingerprint it was possible to make predictions about the coeliac status with a very good accuracy (ca. 84%). One of the most interesting findings was that the metabolic profile of CD patients reverts to normality after 12 months of a strict gluten-free diet: interestingly, a similar behavior was not found in CD patients when analyzing them from a gut microflora prospective33–35. The main observed differences in serum spectra between CD patients and controls were lower levels of several amino acids (asparagine, isoleucine, methionine, proline, valine), methylamine, pyruvate, creatinine, choline, methylglutarate, lactate, lipids and glycoproteins, and higher levels of glucose and 3-hydroxybutyric acid. A decrease in the level of pyruvate and lactate, and a higher level of glucose in the blood of coeliac patients were observed, probably as a consequence of an impaired glycolysis process. Untreated coeliac subjects often report symptoms of fatigue. In patients on a gluten-free diet, fatigue tends to be reduced and, in fact, it has been proposed that this condition is gluten-related36. In CD patients on a gluten-free diet the levels of glucose and 3-hydroxy-butyric acids was found to revert to normality13. Further, the authors found that CD patients are characterized by higher urine levels of some metabolites related to gut microbiota: indoxyl sulfate (IS), meta-[hydroxyphenyl]propionic acid (m-HPPA), and phenylacetylglycine (PAG). These findings are consistent with the hypothesis that in CD patients the gut microflora of the small bowel is altered, or presents peculiar species with their own microbial metabolome. In a following investigation14, the same research group highlighted again the existence of a metabolic fingerprint for coeliac disease, confirming most of the previously discussed metabolites. In the same study, the analysis of the so-called “potential coeliac patients” (i.e. subjects who have a positive antibody test but no evidence of intestinal injury) showed that the metabolic patterns of overt and potential coeliac patients are similar14 indicating that CD related dysmetabolism precedes the intestinal damage. The authors concluded that, although free from intestinal damage, placing potential CD subjects on a gluten-free diet could be justified because they are experiencing most of the pathological alterations experienced by overt coeliac patients14.

3.2. Breast Cancer Breast cancer is the commonest form of cancer in women. Despite many advances in screening, surveillance and intervention, many individuals will die from progressive, advanced breast cancer. Critical points in the search for new strategies are a focus on tumour biology rather than tumour burden, recognition of breast cancer as a heterogenous disease with distinct subgroups and individualization of clinical assessment and therapy37. Metabolomics might emerge as a valuable complementary clinical tool for improved management of breast cancer patients if it can add value to current approaches. A critical point in the quest for improved management is recognition of breast cancer as a heterogeneous disease with distinct subgroups; better subclassification of disease might refine diagnosis, estimation of prognosis and prediction of treatment sensitivity38.

Foodomics 193

The existence of a breast cancer related metabolic fingerprint has been demonstrated19. Serum metabolomic profile differentiated between healthy individuals, and early and metastatic patients. Preoperative patients were identified with 75% sensitivity, 69% specificity and 72% predictive accuracy. In the same study, a comparison of relapse risk calculated with metabolomics and with Adjuvantionline revealed some discordance. Of 21 patients assessed as high risk by Adjuvantionline, 10 (48%) and 6 (29%) were at high risk by metabolomics in pre- and postoperative settings, respectively. Of 23 low-risk patients by Adjuvantionline, 11 (48%) preoperative and 20 (87%) postoperative patients were at low risk by metabolomics19. The authors hypothesised that micrometastic diseases may account for metabolomic misclassifications. Metabolomics may play a role in subselecting patients with HER2 positive disease with greater sensitivity to HER2 targeted therapy. A pilot study in which NMR serum metabolite profiling was used to explore outcomes and response to treatment in women with metastatic breast cancer was undertoken39. The women had been treated as part of a clinical trial and had been randomized to paclitaxel plus either a targeted anti-HER2 treatment (lapatinib) or placebo. Pre- and on-treatment serum samples were assessed. Baseline profiles did not correlate with outcomes or response, however, in the subgroup of patients with HER2-positive disease treated with paclitaxel plus lapatinib, on-treatment samples had a high predictive accuracy for time to progression (N=22, predictive accuracy = 89.6%) and overall survival (N=16, predictive accuracy = 78.0%)39. In a following study40, the same authors analyzed serum samples from women with metastatic (n = 95) and predominantly oestrogen receptor (ER) negative early stage (n = 80) breast cancer using NMR spectroscopy. Multivariate statistics and a Random Forest classifier were used to create a prognostic model for disease relapse in early patients. Metabolomics correctly distinguished between early and metastatic disease in 83.7% of cases. In the early breast cancer training set (n = 40), a prognostic risk model predicted relapse with 73% predictive accuracy. These results were reproduced in an independent early breast cancer set (n = 40), with 75% predictive accuracy. Disease relapse was associated with significantly lower levels of histidine and higher levels of glucose and lipids, compared with patients with no relapse40. A larger scale validation of these results is ongoing.

3.3. Cardiovascular diseases Cardiovascular diseases (CVD) are considered the leading cause of death for both men and women in the developed world41. The profiling of low molecular weight metabolites like lipids, sugars, nucleotides, organic acids, and amino acids which may serve as substrates and products in metabolic pathways has recently received a great deal of attention in the framework of research on CVD and associated risk factors42,43. In Bernini et al.22, a comprehensive metabonomic analysis of 864 plasma samples from healthy volunteers, was presented. The authors used NMR and multivariate statistical analysis and demonstrated that subjects that are classified as at high or at low risk for CVD using common clinical markers (such as total cholesterol, triglycerides, LDL, HDL) can also be discriminated using NMR metabolomics. This discrimination is not only due to the common markers, but also to other metabolites previously not associated with cardiovascular diseases. In particular they found that “low risk” peoples are characterized by high levels of 3-HB and low levels of threonine and creatinine. The “high risk” peoples are characterized by low levels of -ketoglutarate, dimethylglycine, and serine. As discussed by the authors, all of these metabolites are related to mitochondrial metabolism, reinforcing the idea (already formulated in42) that an alteration of the mitochondrial activity is the basis for the risk of developing cardiovascular injuries22.


In Tenori et al.44, serum samples obtained from patients with heart failure (n = 185) and healthy controls (n =111) were studied using NMR metabolomic analysis. Strikingly, it was found that the heart failure pattern is essentially independent of the severity of the disease (described by NYHA classes45), of its origin (primary or secondary), therapy, and of the time of onset. This pattern is mainly constituted by higher serum concentrations of phenylalanine, tyrosine, isoleucine, creatine, and low serum levels of lactate, citrate, lysine and l-dopa. According to the authors, the fact that this fingerprint predicts the presence of heart failure irrespectively of the severity of the disease open interesting perspective for the screening of asymptomatic subjects in a true predictive medicine setting44. In another recent study46, an ambitious attempt was made in identifying, prior to device implantation, the heart failure patients who will respond and non respond to cardiac resynchronization therapy (pacemaker). The existence of a specific metabolomic fingerprint in heart failure patients compared to healthy controls was again confirmed46 and the two populations were discriminated with high accuracy. This metabolomics fingerprint was similar between patients with ischemic and nonischemic dilated cardiomyopathy. Unfortunately, no significant difference was observed in responder’s metabolomic fingerprint with respect to non-responder ones both at baseline (before implantation) and after 6 months follow up46. 4. Metabolic changes in response to diet

Many dietary constituents are associated with physical health, disease onset, progression or prevention, and general wellbeing27,47. Besides endogenous metabolism, human urine essentially derive from foods, drinks, drugs, and other environmental products. Because different nutritional habits and host/guest interactions with the gut microbiome lead to different metabolic signatures in urine, metabolomics can reveal dietary intake patterns48–51 and represents a powerful tool for the understanding of the outcomes of dietary intervention52,53. The presence of dietary patterns, on the other hand, introduces intrapersonal variability that can constitute a confounding factor. The effects of standardized diet on the metabolic profile have been assessed in the literature, both during short (24 hours)54–56 or prolonged (2 weeks)56 interventions. In the latter, ten healthy volunteers were recruited in a clinical research center for two weeks of standardized diet. The results suggested that 24 hours of dietary standardization could be enough to provide the normalization of the metabolic profile necessary in a clinical setting56. In another study, the results of analyzing the urine of 30 healthy individuals revealed that the metabolic profile in urine is more sensitive than plasma and saliva to acute dietary intake55. However, the presence of an invariant part of the urine metabolomic profile that characterizes individuals irrespectively of the dietary habits has been well established25,26. This finding is reinforced by a comparative study performed in humans and monkeys that showed that dietary habits have little or none effects on the individual metabolic phenotype57. The metabolic composition of human urine is also influenced by country specific dietary habits. In another study48, NMR-based metabolomics was employed to analyze urine samples obtained from residents of Britain and Sweden. Very strong differences were found in the urine samples of the two sub-groups, which were clearly attributable to dietary differences between the two countries48. Application of metabolomics to the discovery of the influence of dietary components and to discover specific diet-related biomarkers has been extremely fruitful in recent years49,58. The effects of three different kinds of diet (mostly meat, minimal meat and vegetarian) on the human metabolic phenotypes was deeply analyzed by means of 1H-NMR metabolomics59. In

Foodomics 195

this study 12 healthy participants consumed each kind of diet continuously for 15 days followed by 7 days washout periods. The results revealed different urine metabolic signatures for the different diets. Higher levels of carnitine, acetylcarnitine, and trimethylamine-N-oxide (TMAO) were found to be associated with a mostly meat diet59. Furthermore, the urinary concentration of creatine (a metabolite that is found mainly in muscle tissue) was much higher in samples taken during the mostly meat diet than in samples from the minimal meat and vegetarian diets. On the other hand, p-hydroxyphenylacetate (a microbial/mammalian co-metabolite) was elevated in the samples taken from those on a vegetarian diet compared with samples taken from those on meat diets, indicating also an alteration of bacterial composition (and/or bacterial metabolism) in response to diet59. The latter finding suggests the use of dietary intervention to affect the activity of the gut microbioma. Because of the essential role played by gut microflora in regulating key physiological activities, such as processing and absorbing nutrients and the metabolism of many xenobiotic compounds60, several study are devoted to explore how modulate its activity. To reach this goal a variety of options is available, including changes in diet, antimicrobial-based intervention, probiotics administration, and faecal microbiota transplantation61. Probiotics microorganisms in particular, alone or in combination with each other, are becoming increasingly popular and are currently available in a wide range of consumer formulation including yoghurts, drinks, capsules or powders62. However only few investigations of the effects of probiotics administration on the metabolic fingerprint are currently available63–68. Preliminary results of an ongoing study in our laboratory suggest that the urinary metabolome is affected by probiotics consumption, though not so strongly to hamper individual recognition. The possibility offered by metabolomics to monitor the fluctuations due to changes in nutritional habits and nutrients intake could open a new frontier for discovery, validation and registration of nutritional claims. Food labels are important tools to educate the population about healthiness and benefits of a particular food69. Because regulatory agencies request strong scientific evidences about the possible effects of nutrients and/or foods on the human health, metabolomics can play an important role in this field. Based on the economic importance of this area, there will be a strong interest in this kind of research for industries and governments, but also for consumers that could benefit from healthier products and from the availability of validated scientific information70. 5. Food profiling

In the last years metabolomics has demonstrated to be a fast and reliable tool to provide consumers with reliable information about the origin and traceability of agro-food products. Through the use of NMR-based metabolomics it is possible to directly evaluate foodstuff in order to obtain information about provenience, and it is also possible to evaluate some chemical features that can be related to potential health benefits and organoleptic properties. 1H-NMR spectroscopy has been extensively used for the analysis of olive oil and it has been established as a valuable tool for its quality assessment and authenticity. To date, a large number of research papers and review articles have been published with regards to the analysis of olive oil, reflecting the potential of the NMR technique in these studies71. As an example of the potential of the technique, in Sacchi et al.72 55 extra virgin olive oil samples from four Italian Regions were analyzed by NMR metabolomics. A predictive model was build with a predictive accuracy of 95%. Wine is another products extensively explored using NMR metabolomics73. The authenticity, the grape variety, the geographical origin, and the year of vintage of different wines produced in Germany were investigated74 using 1H-NMR spectroscopy in combination with several


steps of multivariate statistics. The wine production regions of Germany could be differentiated by grape varieties with a high degree of predictivity. Also, mixtures of wines from different grapes were separated. In addition, the year of the vintage and the wine-producing region were separated with a high degree of differentiation. The authors concluded that each NMR spectrum could be regarded as the individual “fingerprint” of a wine sample, which includes information about variety, origin, vintage, physiological state, technological treatments, and others74. Metabolomics was demonstrated also a useful tool for distinguishing the species and origins of green coffee bean samples of arabica and robusta from six different geographic regions75. By the application of information on signal assignment, significantly different levels of 14 metabolites of green coffee beans were identified as significant for the classification75. Besides geographical origin, metabolomics can be used to assess the relationship between organoleptic properties and chemical composition of food matrices. NMR spectroscopy can be considered a kind of "magnetic tongue" for the characterisation and prediction of the tastes of foods, since it provides a wealth of information in a non-destructive and untargeted manner76. The taste of commercial coffee beans, identified by human sensory tests, has been successfully predicted on the basis of their NMR metabolite profiles using multivariate statistics on the whole spectral profile76. The ripening process of fruits can be followed using metabolomics: in Gil et al.77, mango juice and pulp were followed in time, showing that spectral changes reflect the complex biochemistry of mango ripening, and enlightening the role played by some identified compounds. Some differences observed between the composition of juices and pulps are also discussed. Moreover the authors could show that HR-MAS NMR spectroscopy enables the direct characterization of intact mango pulps, thus allowing the non-invasive study of the overall biochemistry in the whole fruit77. Mixture analysis by NMR was applied to high-throughput fruit juice screening for quality control78. The application introduced in that paper can also be seen as a strong proof-of-principle for other food quality control applications, such as screening milk, wine or beer78. NMR was first used to investigate the properties of milk in the 1950s and has since been employed in a wide range of studies; including properties analysis of specific milk proteins to metabolomics techniques used to monitor the health of dairy cows79. Preliminary results from our laboratory suggest that NMR metabolomics can be successfully used for milk traceability. Regarding tomatoes, it has been demonstrated that 1H-NMR-based metabolomics could be a feasible tool to evaluate the metabolic changes during fruit growing80. Quantitative descriptive analysis and its correlation with NMR-metabolomics has been applied also to evaluate the taste of tomatoes with good results. In fact, a number of sensory predictors can be easily predicted through NMR data81. Moreover, NMR metabolomics can be useful for the analysis of tomato-derived products like ketchup: Vallverdù-Queralt et al.80 analyzed organic and conventional ketchup present on the market finding that organic ketchup has an higher amount of phenolic compounds than conventionally derived. Peaches contain many metabolites with important values for human health and they are a very common fruit in temperate region of the world82. The use of metabolomics can provide information to discover metabolic pathway and human health-promoting metabolites. For this purpose, the metabolic profile of peach has been be followed during development and ripening82, in the effort to understand the functions of key enzymes important for these processes. In another study83, metabolomics was employed to study two varieties of peaches, aiming at understanding plant defensive mechanisms against insect attacks.

Foodomics 197

6. Conclusions

NMR-based metabolomics can be considered a “universal” quantitative analytical technique84. Offering an unbiased and untargeted view of the sample composition, and the possibility to simply quantify multiple compounds simultaneously, it has become the method of choice for studies and quality control of complex natural samples such as biofluids, foods, and plant extracts84. The future challenge is to integrate the different sources of information, in a way that the simultaneous acquisition of metabolic profiles of individuals and of the food they consume will provide a holistic molecular view of health and nutrition.

References

1 O. Fiehn, Plant MolBiol, 2002, 48, 155–171. 2 J. Li, O. Brazhnik, A. Kamal, D. Guo, C. Lee, S. Hoops, and P. Mendes, eds. G.

Harrigan and R. Goodacre, Kluwer Academic Publishers, 2003, pp. 293–309. 3 M. Aimetti, S. Cacciatore, A. Graziano, and L. Tenori, Metabolomics, 2012, 8, 465–474. 4 H. Wu, A. D. Southam, A. Hines, and M. R. Viant, Anal. Biochem., 2008, 372, 204–212. 5 D. S. Wishart, M. J. Lewis, J. A. Morrissey, M. D. Flegel, K. Jeroncic, Y. Xiong, D.

Cheng, R. Eisner, B. Gautam, and D. Tzur, J. Chromatogr. B, 2008, 871, 164–173. 6 G. A. N. Gowda, N. Shanaiah, A. Cooper, M. Maluccio, and D. Raftery, Lipids, 2009,

44, 527–535. 7 A. D. Maher, O. Cloarec, P. Patki, M. Craggs, E. Holmes, J. C. Lindon, and J. K.

Nicholson, Anal. Chem., 2009, 81, 288–295. 8 G. Graça, I. F. Duarte, B. J. Goodfellow, I. M. Carreira, A. B. Couceiro, M. do R.

Domingues, M. Spraul, L.-H. Tseng, and A. M. Gil, Anal. Chem., 2008, 80, 6085–6092. 9 L. Lacitignola, F. P. Fanizzi, E. Francioso, and A. Crovace, Vet. Comp. Orthop.

Traumatol. VCOT, 2008, 21, 85–88. 10 P. Montuschi, D. Paris, D. Melck, V. Lucidi, G. Ciabattoni, V. Raia, C. Calabrese, A.

Bush, P. J. Barnes, and A. Motta, Thorax, 2012, 67, 222–228. 11 I. Bertini, C. Luchinat, M. Miniati, S. Monti, and L. Tenori, Metabolomics, 1–10. 12 G. Le Gall, S. O. Noor, K. Ridgway, L. Scovell, C. Jamieson, I. T. Johnson, I. J.

Colquhoun, E. K. Kemsley, and A. Narbad, J. Proteome Res., 2011, 10, 4208–4218. 13 I. Bertini, A. Calabrò, V. De Carli, C. Luchinat, S. Nepi, B. Porfirio, D. Renzi, E.

Saccenti, and L. Tenori, J ProteomeRes, 2009, 8, 170–177. 14 P. Bernini, I. Bertini, A. Calabrò, G. la Marca, G. Lami, C. Luchinat, D. Renzi, and L.

Tenori, J ProteomeRes, 2011, 10, 714–721. 15 S. Cacciatore, C. Luchinat, and L. Tenori, Proc. Natl. Acad. Sci., 2014, 111, 5117–5122. 16 C. J. Lindon, J. K. Nicholson, and E. Holmes, The Handbook of Metabonomics and

Metabolomics, Elsevier, 2007. 17 J. K. Nicholson, J. C. Lindon, and E. Holmes, Xenobiotica, 1999, 29, 1181–1189. 18 V. Travagli, I. Zanardi, P. Bernini, S. Nepi, L. Tenori, and V. Bocci, IntJ Toxicol, 2010,

29, 165–174. 19 C. Oakman, L. Tenori, W. M. Claudino, S. Cappadona, S. Nepi, A. Battaglia, P. Bernini,

E. Zafarana, E. Saccenti, M. Fornier, P. G. Morris, L. Biganzoli, C. Luchinat, I. Bertini, and L. A. Di, Ann Oncol, 2011.

20 M. Coen, M. O’Sullivan, W. A. Bubb, P. W. Kuchel, and T. Sorrell, Clin.Infect.Dis., 2005, 41, 1582–1590.

21 T. Bartsch, K. Alfke, S. Wolff, A. Rohr, O. Jansen, and G. Deuschl, Neurology, 2008, 70, 1030–1035.


22 P. Bernini, I. Bertini, C. Luchinat, L. Tenori, and A. Tognaccini, J. Proteome Res., 2011, 10, 4983–4992.

23 S. H. Moolenaar, U. F. H. Engelke, and R. A. Wevers, Ann. Clin. Biochem., 2003, 40, 16–24.

24 M. Sellitto, G. Bai, G. Serena, W. F. Fricke, C. Sturgeon, P. Gajer, J. R. White, S. S. K. Koenig, J. Sakamoto, and D. Boothe, PloS One, 2012, 7, e33387–e33387.

25 M. Assfalg, I. Bertini, D. Colangiuli, C. Luchinat, H. Schafer, B. Schutz, and M. Spraul, Proc.Natl.Acad.Sci.U.S.A, 2008, 105, 1420–1424.

26 P. Bernini, I. Bertini, C. Luchinat, S. Nepi, E. Saccenti, H. Schafer, B. Schutz, M. Spraul, and L. Tenori, J ProteomeRes, 2009, 8, 4264–4271.

27 C. L. Gavaghan, E. Holmes, E. Lenz, I. D. Wilson, and J. K. Nicholson, FEBS Lett., 2000, 484, 169–174.

28 I. Bertini, C. Luchinat, and L. Tenori, Pers. Med., 2012, 9, 133–136. 29 J. A. Murray, C. V. Dyke, M. F. Plevak, R. A. Dierkhising, A. R. Zinsmeister, and L. J.

Melton, Clin. Gastroenterol. Hepatol., 2003, 1, 19–27. 30 M. Mäki, K. Mustalahti, J. Kokkonen, P. Kulmala, M. Haapalahti, T. Karttunen, J.

Ilonen, K. Laurila, I. Dahlbom, and T. Hansson, N. Engl. J. Med., 2003, 348, 2517–2524. 31 C. Catassi, I. M. Rätsch, E. Fabiani, S. Ricci, F. Bordicchia, R. Pierdomenico, and P. L.

Giorgi, Acta Paediatr., 1995, 84, 672–676. 32 Calabrò, Antonio, E. Gralka, C. Luchinat, E. Saccenti, and L. Tenori, Autoimmune Dis.,

2014, 2014. 33 I. Nadal, E. Donant, C. Ribes-Koninckx, M. Calabuig, and Y. Sanz, J. Med. Microbiol.,

2007, 56, 1669–1674. 34 E. Sánchez, E. Donat, C. Ribes-Koninckx, M. Calabuig, and Y. Sanz, J. Clin. Pathol.,

2010, 63, 1105–1111. 35 R. Di Cagno, M. De Angelis, I. De Pasquale, M. Ndagijimana, P. Vernocchi, P. Ricciuti,

F. Gagliardi, L. Laghi, C. Crecchio, M. E. Guerzoni, M. Gobbetti, and R. Francavilla, BMC Microbiol., 2011, 11, 219.

36 F. C. Jordá and J. L. Vivancos, J. Clin. Gastroenterol., 2009, 1. 37 C. Oakman, L. Tenori, L. Biganzoli, L. Santarpia, S. Cappadona, C. Luchinat, and L. A.

Di, IntJ Biochem. Biol, 2010. 38 C. Oakman, L. Tenori, S. C. S, C. Luchinat, I. Bertini, and A. D. Leo, Curr. Breast

Cancer Rep., 2012, 4, 249–256. 39 L. Tenori, C. Oakman, W. M. Claudino, P. Bernini, S. Cappadona, S. Nepi, L. Biganzoli,

M. C. Arbushites, C. Luchinat, I. Bertini, and A. Di Leo, Mol. Oncol., 2012, 6, 437–444. 40 L. Tenori, C. Oakman, P. G. Morris, E. Gralka, N. Turner, S. Cappadona, M. Fornier, C.

Hudis, L. Norton, C. Luchinat, and A. Di Leo, Mol. Oncol., 2014. 41 R. B. D’Agostino, M. W. Russell, D. M. Huse, R. C. Ellison, H. Silbershatz, P. W.

Wilson, and S. C. Hartz, AmHeart J, 2000, 139, 272–281. 42 S. H. Shah, J. R. Bain, M. J. Muehlbauer, R. D. Stevens, D. R. Crosslin, C. Haynes, J.

Dungan, L. K. Newby, E. R. Hauser, G. S. Ginsburg, C. B. Newgard, and W. E. Kraus, Circ.Cardiovasc.Genet., 2010, 3, 207–214.

43 H. L. Kirschenlohr, J. L. Griffin, S. C. Clarke, R. Rhydwen, A. A. Grace, P. M. Schofield, K. M. Brindle, and J. C. Metcalfe, Nat.Med., 2006, 12, 705–710.

44 L. Tenori, X. Hu, P. Pantaleo, B. Alterini, G. Castelli, I. Olivotto, I. Bertini, C. Luchinat, and G. F. Gensini, Int. J. Cardiol., 2013, 168, e113–115.

45 C. Raphael, C. Briscoe, J. Davies, W. Z. Ian, C. Manisty, R. Sutton, J. Mayet, and D. P. Francis, Heart, 2007, 93, 476–482.

Foodomics 199

46 L. Padeletti, P. A. Modesti, S. Cartei, L. Checchi, G. Ricciardi, P. Pieragnoli, S. Sacchi, M. Padeletti, B. Alterini, P. Pantaleo, X. Hu, L. Tenori, and C. Luchinat, J. Cardiovasc. Med. Hagerstown Md, 2014.

47 S. A. Bingham, Public Health Nutr., 2002, 5, 821–827. 48 J. B. E M Lenz, J. Pharm. Biomed. Anal., 2004, 36, 841–9. 49 A. J. Lloyd, M. Beckmann, G. Favé, J. C. Mathers, and J. Draper, Br. J. Nutr., 2011, 106,

812–824. 50 K. S. Solanky, N. J. Bailey, B. M. Beckwith-Hall, S. Bingham, A. Davis, E. Holmes, J.

K. Nicholson, and A. Cassidy, J. Nutr. Biochem., 2005, 16, 236–244. 51 C. Zuppi, I. Messana, F. Forni, F. Ferrari, C. Rossi, and B. Giardina, Clin. Chim. Acta

Int. J. Clin. Chem., 1998, 278, 75–79. 52 M.-B. S. Andersen, Å. Rinnan, C. Manach, S. K. Poulsen, E. Pujos-Guillot, T. M.

Larsen, A. Astrup, and L. O. Dragsted, J. Proteome Res., 2014, 13, 1405–1418. 53 A. O’Sullivan, M. J. Gibney, and L. Brennan, Am. J. Clin. Nutr., 2011, 93, 314–321. 54 E. M. Lenz, J. Bright, I. D. Wilson, S. R. Morgan, and A. F. P. Nash, J. Pharm. Biomed.

Anal., 2003, 33, 1103–1115. 55 L. B. Marianne C Walsh, Am. J. Clin. Nutr., 2006, 84, 531–9. 56 J. H. Winnike, M. G. Busby, P. B. Watkins, and T. M. O’Connell, Am. J. Clin. Nutr.,

2009, 90, 1496–1496. 57 E. Saccenti, L. Tenori, P. Verbruggen, M. E. Timmerman, J. Bouwman, J. van der Greef,

C. Luchinat, and A. K. Smilde, PloS One, 2014, 9, e106077. 58 S. S. Heinzmann, C. A. Merrifield, S. Rezzi, S. Kochhar, J. C. Lindon, E. Holmes, and J.

K. Nicholson, J. Proteome Res., 2012, 11, 643–655. 59 C. Stella, B. Beckwith-Hall, O. Cloarec, E. Holmes, J. C. Lindon, J. Powell, F. van der

Ouderaa, S. Bingham, A. J. Cross, and J. K. Nicholson, J Proteome Res, 2006, 5, 2780–2788.

60 H. Curtis, G. Dirk, K. Rob, A. Sahar, H. Badger Jonathan, T. Chinwalla Asif, H. Creasy Heather, M. Earl Ashlee, G. FitzGerald Michael, and S. Fulton Robert, 2012.

61 C. J. Walsh, C. M. Guinane, P. W. O’Toole, and P. D. Cotter, FEBS Lett., 2014. 62 S. S., in Probiotics, ed. E. Rigobelo, InTech, 2012. 63 R. Vázquez-Fresno, R. Llorach, J. Marinic, S. Tulipani, M. Garcia-Aloy, I. Espinosa-

Martos, E. Jiménez, J. M. Rodríguez, and C. Andres-Lacueva, Pharmacol. Res. Off. J. Ital. Pharmacol. Soc., 2014, 87, 160–165.

64 Y.-S. Hong, K. S. Hong, M.-H. Park, Y.-T. Ahn, J.-H. Lee, C.-S. Huh, J. Lee, I.-K. Kim, G.-S. Hwang, and J. S. Kim, J. Clin. Gastroenterol., 2011, 45, 415–425.

65 Y.-S. Hong, Y.-T. Ahn, J.-C. Park, J.-H. Lee, H. Lee, C.-S. Huh, D.-H. Kim, D. H. Ryu, and G.-S. Hwang, Arch. Pharm. Res., 2010, 33, 1091–1101.

66 F.-P. J. Martin, Y. Wang, N. Sprenger, I. K. S. Yap, T. Lundstedt, P. Lek, S. Rezzi, Z. Ramadan, P. van Bladeren, L. B. Fay, S. Kochhar, J. C. Lindon, E. Holmes, and J. K. Nicholson, Mol. Syst. Biol., 2008, 4, 157.

67 F.-P. J. Martin, Y. Wang, N. Sprenger, I. K. S. Yap, S. Rezzi, Z. Ramadan, E. Peré-Trepat, F. Rochat, C. Cherbut, P. van Bladeren, L. B. Fay, S. Kochhar, J. C. Lindon, E. Holmes, and J. K. Nicholson, Mol. Syst. Biol., 2008, 4, 205.

68 F.-P. J. Martin, N. Sprenger, I. K. S. Yap, Y. Wang, R. Bibiloni, F. Rochat, S. Rezzi, C. Cherbut, S. Kochhar, J. C. Lindon, E. Holmes, and J. K. Nicholson, J. Proteome Res., 2009, 8, 2090–2105.

69 S. Agarwal, S. Hordvik, and S. Morar, Toxicology, 2006, 221, 44–49. 70 P. J. H. Jones, N.-G. Asp, and P. Silva, J. Nutr., 2008, 138, 1189S–1191S. 71 P. Dais and E. Hatzakis, Anal. Chim. Acta, 2013, 765, 1–27.


72 R. Sacchi, L. Mannina, P. Fiordiponti, P. Barone, L. Paolillo, M. Patumi, and A. Segre, J. Agric. Food Chem., 1998, 46, 3947–3951.

73 Y.-S. Hong, Magn. Reson. Chem., 2011, 49, S13–S21. 74 R. Godelmann, F. Fang, E. Humpfer, B. Schütz, M. Bansbach, H. Schäfer, and M.

Spraul, J. Agric. Food Chem., 2013, 61, 5610–5619. 75 F. Wei, K. Furihata, M. Koda, F. Hu, R. Kato, T. Miyakawa, and M. Tanokura, J. Agric.

Food Chem., 2012, 60, 10118–10125. 76 F. Wei, K. Furihata, T. Miyakawa, and M. Tanokura, Food Chem., 2014, 152, 363–369. 77 A. M. Gil, I. F. Duarte, I. Delgadillo, I. J. Colquhoun, F. Casuscelli, E. Humpfer, and M.

Spraul, J. Agric. Food Chem., 2000, 48, 1524–1536. 78 M. Spraul, B. Schütz, E. Humpfer, M. Mörtter, H. Schäfer, S. Koswig, and P. Rinke,

Magn. Reson. Chem., 2009, 47, S130–S137. 79 A. D. Maher and S. J. Rochfort, Metabolites, 2014, 4, 131–141. 80 A. Vallverdú-Queralt, A. Medina-Remón, I. Casals-Ribes, M. Amat, and R. M. Lamuela-

Raventós, J. Agric. Food Chem., 2011, 59, 11703–11710. 81 A. Malmendal, C. Amoresano, R. Trotta, I. Lauri, S. De Tito, E. Novellino, and A.

Randazzo, J. Agric. Food Chem., 2011, 59, 10831–10838. 82 V. A. Lombardo, S. Osorio, J. Borsani, M. A. Lauxmann, C. A. Bustamante, C. O.

Budde, C. S. Andreo, M. V. Lara, A. R. Fernie, and M. F. Drincovich, Plant Physiol., 2011, 157, 1696–1710.

83 D. Capitani, A. P. Sobolev, A. Tomassini, F. Sciubba, F. R. De Salvador, L. Mannina, and M. Delfini, J. Agric. Food Chem., 2013, 61, 1718–1726.

84 C. Simmler, J. G. Napolitano, J. B. McAlpine, S.-N. Chen, and G. F. Pauli, Curr. Opin. Biotechnol., 2014, 25, 51–59.

New Developments

COMPOST BIODEGRADATION BY 1H MAGNETIC RESONANCE AND QUANTITATIVE RELAXATION TOMOGRAPHY

Villiam Bortolotti1,2, Paola Fantazzini3, Marianna Vannini1,2, Ester Maria Vasini1

1 Department of Civil, Chemical, Environmental and Materials Engineering, University of Bologna 2 CIRI Energy and Environment, University of Bologna 3 Department of Physics and Astronomy, University of Bologna

1 INTRODUCTION

About one-third of worldwide food produced for human consumption is lost or wasted, entailing both a waste of energy and emissions of greenhouse gas. The production of food waste (hereinafter FW) covers all the food life cycle: from farming phase, up to both industrial manufacturing and processing, and retail and household1. A best practice is to use wastes from one industrial sector as input for other industrial sectors, to minimize world wastes. One of the most used practice of reusing food residues is the production of compost. The collection and composting of organic waste from both households and industries has long been recognized as a valuable contribution to waste management. The first step towards a “zero-waste” concept, in particular in the food processing sector, is the identification, quantification and characterization of residues1. For this reason, considerable efforts has been focused towards the definition of compost maturity in order to decide on the quality of a compost. Therefore the principal requirement of a compost for its safe use, e.g. in soil, is a high degree of maturity, which implies a stable organic matter (OM) content and the absence of phytotoxic compounds and plant or animal pathogens. Over the last decades, research has been focused on the complex interaction amongst physical, chemical and biological factors that occur during composting. The control of parameters such as bulk density, porosity, particle size, nutrient content, Carbon to Nitrogen (C/N) ratio, temperature, pH, moisture and oxygen supply, have demonstrated to be key factors for composting optimisation because they influence the optimal microbial development and OM degradation2. Generally, the degradation kinetic is accompanied by the development of a temperature profile that indicates the different phases of the process. The composting process evolves in three main phases: (i) an initial mesophilic phase, where mesophilic bacteria and fungi degrade compounds such as sugars, amino acids, proteins, etc., increasing quickly the temperature; (ii) a thermophilic phase, where fats, cellulose, hemicellulose and some lignin are degraded; (iii) a cooling or maturing phase, characterized by a decrease of the microbial activity and, hence, of the temperature2. Maturity is not described by a single property so it is best assessed by taking in account two or more parameters. Physical characteristics such as colour, odour and temperature give a general idea of the decomposition stage reached, but they give poor information on the


degree of maturation. Also the C/N ratio, trend of C/N and pH during composting are usually analysed, but they don’t form a complete set of standard criteria2. Magnetic Resonance Relaxometry (MRR) and Imaging (MRI) of hydrogen nuclei are optimal candidates to give information on the degree of degradation and maturity of compost, especially when there is needed a careful non-invasive and non-destructive characterization. Because many foods are proton-rich, e.g., from water, fat, carbohydrates, and proteins, MRR is a common tool to gain information about the composition and internal structure of foods, permitting also to monitor the compositional and structural modifications when they undergo natural or artificial processes. Since the relaxation times (longitudinal T1 and transverse T2) variations change the image contrast, MRI can easily monitor the structural changes in foods during processing and storage3. MRI can detect and show internal variations in the water content, as well as changes in water interaction with cellular tissues4. This is especially important for compost where the moisture has a role in structure changes, nutrient transport and heat conductivity. Moisture content is usually assessed ‘off-line’ by dry weight measurements, which, however, does not differentiate the water available for microorganism activity from the water bound to the substrate unavailable to microorganisms5. Water is both produced by and required for microbial activity, as it is necessary to support the metabolic processes of the microbes6. Water provides the medium for chemical reactions, transports nutrients, and allows the microorganisms to move about7. The use of a moisture mass or volume ratio is a convenient and generally accepted method of describing the moisture status of materials, but as Miller8 has pointed out, this approach provides only little insight into the availability of moisture. However, consistent moisture prediction is still a challenging, as it depends on density and compost maturity9. In this work typical domestic food wastes have been reused as organic resources to produce compost for household. Typical MRR parameters as T1 and T2, and 1H signal amplitude have been related to water content, and to micro and molecular structure of the compost itself, to follow the compost phases. Furthermore we used Quantitative Relaxation Tomography (QRT), a method to measure locally the relaxation times by means of MRI images. QRT images are relaxation times maps, where each image pixel is proportional to the relaxation time values10. This technique allowed us to locally follow the FW biodegradation over time and to evaluate the maturity and quality parameters. In this first, but promising study, both MRR techniques seem to well describe the stages of composting, as observed with other techniques, in addition giving information on local degradation of compost components.


2.1 Samples

The compost analysed in this study was a mix of household organic waste with coffee grounds, fruits waste, food waste (bread, cheese and cereals) and loose grass (see Table 1), in proper percentages to obtain a C/N ratio around 25, which is an acceptable C/N ratio to obtain a good compost mix (Compost Maturity Index11). To obtain the percentages of Table 1, the Klickitat County calculator of the Washington state (www.KlickitatCounty.org) has been used. A laboratory scale compost reactor (Figure 1) suitable for MRR measurements was created to both ensure oxygenation and immobility of the compost sample; the last is essential to take MRI images over time (and therefore to obtain QRT maps). The reactor was cylindrical and

New Developments 205

non-magnetic, with dimensions suitable to fit both the MRR and MRI (12 cm in diameter) coils used in the experiments. Holes on the reactor surface ensure the oxygenation of the sample, which was provided using a laboratory extractor fan. Another identical reactor was used for temperature analyses.

Table 1 Percetages of the domestic organic wastes used.

Waste Percentage on the total mass (%) Fruit residues 78 Food residues 7 Loose grass 4

Coffee grounds 11

Figure 1 Pictures of the reactor (d=10 cm, h= 13,5 cm) used for MRR/MRI measurments. Notice the special supports to ensure the same position in the MRR/MRI coils during each measure and the holes to ensure oxygenation.

2.2 Temperature and pH measurements

A mass of approximately 500 g of compost mix was stored, at room temperature conditions (23-25 C approximately constant), in the dedicated reactor, to detect the sample temperature, using two temperature probes for the sample and one for the room temperature. For pH analyses the standard procedure was adopted: a certain compost amount (a few grams) was mixed with a quantity of distilled water (1:5 w/w) and using a pH probe, pH was determined. Both temperature and pH measurements were carried out over a period of time of two months starting from the formation of the compost mix. Measurements were roughly daily in the first two weeks, then they became less frequent as biodegradation changes slowed down. 2.3 MRR and QRT measurements

A mass of the same compost, approximately equal (in laboratory conditions, oxygenation and room temperature) to the one used in temperature measurements, has been introduced in the reactor. This sample has been used for both MRR and QRT measurements. MRR data were collected to obtain both T1 and T2 relaxation times distributions using a permanent 0.2 Tesla magnet (ESAOTE SPA, Genova, Italy), a MRR console and a coil both manufactured by Stelar (Mede, PV, Italy). The sequences used were Saturation Recovery


(SR) for T1 and CPMG for T2. In the former, 64 Free Induction Decay signals log-spaced of 512 points were acquired, starting from 0.005 ms. Each scan was repeated 4 times in order to obtain a good signal to noise ratio. The measure was repeated for three different echo times: 100 ms, 250 ms, 500 ms. Each CPMG contained 3000 echoes and was repeated 4 times. For both SR and CPMG the duration of the 90 pulse was 60 μs. The repetition time was in all cases 4000 ms. MRI measurements were performed at 30 °C by means of ARTOSCAN™ (Esaote SpA., Genova, Italy), a tomograph consisting of a 0.2 T permanent magnet (15 mT/m maximum gradient intensity), corresponding to about 8 MHz for protons. A noteworthy aspect is that the low-magnetic field instrument gives a relatively low signal-to-noise ratio, but it has the advantage of reducing the signal dephasing due to the distribution of magnetic fields within the sample, produced by the magnetic susceptibility differences between the matrix framework and the fluid in the pores. The quality of low-field MRI quantitative measurements was already confirmed previously (Borgia et al., 2001). In each MRI measurement, three axial sections of 5 mm were imaged; gap between slice of 0.2 mm; pixel size of 0.47×0.47 mm2 were used. Depending on the signal to noise ratio, a number of excitations between 4 and 10 were used. QRT was performed both using Saturation Recovery (SR) to obtain a T1 map and the Spin Echo (SE) sequences for the T2 map. To create a map, a set of images was acquired, varying the inversion times in the case of SR and the echo times in the case of SE. For T2 maps, ten multislice SE sequences were acquired, using a fixed repetition time (TR=6000 ms) and a variable echo time (TE), from 10 ms to 3308 ms. For T1 maps, eleven multislice SR sequences were used with fixed TE=12 ms, TR variable from 180 to 9160 ms and inversion time equispaced on the logarithmic scale from 10 ms up to 9000 ms. 2.4 MRR and QRT data processing

Relaxometry data have been processed using the in-house UpenWin software (distributed by University of Bologna12), a Windows program specifically designed to not provide distribution details that are not supported by data, which might be misinterpreted, for example, as physically meaningful resolved pore compartments. The weighted geometric mean (T(1,2)g) was used as a characterizing parameter to summarize distributions13. Images have been processed using an home-made image analysis software developed in C+ and the library MFC, called ARTS.

3 RESULTS AND DISCUSSION 3.1 Temperature and pH

In Figure 2 the registered temperature (cross symbols) values during composting time and the reference room temperature (dotted line) are shown. The typical temperature trend during composting can be observed: an initial rise from ambient conditions, until the achievement of a maximum value, and then a slow return to lower values. Temperatures did not increase over 30 C, probably due to the dimensions of the laboratory-scale reactor and to the high degree of aeration, that was approximately of 200 m3/h. As Sundberg et al.14 observed, increased aeration increases both cooling and oxygen concentration, two process conditions that accelerate the increase of pH during composting and therefore resulting in an increase of the decomposition rate. In our sample, during the first week of composting, pH was around 4, followed by an increase up to values above 5.


3.2 MRR

During the composting process all the T1 and T2 distributions were wide, from a few ms to a few seconds (Figure 3-4). Over the first days of composting both relaxation times gradually increased and then they slowly decreased. In Figure 5 T1g and T2g are plotted versus the measurement time, it is possible to appreciate the trend described above. This trend appears similar to that of the temperature (Figure 2). The increase of relaxation times during the first stage of the degradation process is also due to the formation of new water released by the microbial activity. This new water is not strongly bound and therefore has longer relaxation times (Figure 3-4). Summarizing, during the first 6 days temperature increased, pH was acidic ( 4) and an increase of both relaxation times appeared, probably due to the release of new free water and to the increase of temperature. During this first stage simple compounds were degraded and temperature started increasing. Then a stage during which temperature decreased followed, pH moved towards pH > 5 and relaxation times, after the achievement of their maximum values, started to decrease. During these stages the degradation rate is high, in fact a loss of over 40% of mass was measured in three weeks. This can be observed in Figure 6 where in the semi-log plots of the MRR signal intensity and mass loss are plotted: they both decrease but with different rate. This different trend is consistent because the loss of mass detects the loss of OM (by degradation), water (by evaporation) but not the presence of new water released. On the contrary MRR signal is due to the balance from the loss of water (by evaporation) and the presence of new water released. In general it is preferable to identify a thermophilic phase with temperatures over 45 C , but in this laboratory-scale reactor no thermophilic phase has been detected, probably because of the high aeration rate which raised pH and decreased temperature. Nevertheless the degradation rate was anyway high14. After that, a cooling phase followed, during which approximately another 40 % of mass was lost in 50 days, meaning the presence of a slower degradation rate (data not shown).

Figure 2 Temperature and pH data vs the time of composting; for temperature both sample (x) and room (---) data are represented, pH is represented with the (•) symbols.


Figure 3 T1 distributions of the compost sample during composting time.

Figure 4 T2 distributions of the compost sample during composting time.

Figure 5 T (1,2)g during composting time, each point is the geometric mean time of the distribution.


Figure 6 Semi-logarithmic plot of the mass loss and the MRR signal intensity during composting time. The straight lines are the linear fit obtained on data until the 19th day of measure, both for mass loss and MRR signal. (In this case the MRR signal is the one obtained only from saturation recovery measures. The one from CPMG has a similar trend, not shown).

3.3 QRT

In Figure 7 T1 maps obtained by QRT created during the composting time are shown. In the first row, filtered T1 images are shown, where different thresholds in a) and b) are chosen in order to highlight the degradation phenomenon. In these regions local degradation phenomena are acting and higher values of T1 (on the order of 2 s) are revealed, likely meaning the release of free water by microbial activity. These phenomena happened immediately in few regions (detectable only with QRT) and then expanded until the twelfth day when a strong and global degradation was reached and relaxation times were uniformly high. This first stage of degradation lasted approximately 20 days, but slow degradation phenomena still acted, until the end of the cooling phase. Similar results have been obtained on T2 maps (data not shown). This preliminary QRT analysis well underlines the possibility of this technique to detect locally different kinetics rates, identifying fast and slow degradation phenomena.

4 CONCLUSIONS In summary, non-invasive and non-destructive MRR/MRI techniques can follow compost degradation globally, by means of the weighted geometric means of T1 and T2, showing a trend which is almost comparable to that of temperature. At the same time local information on degradation of compost is obtained by QRT maps, highlighting the presence of fast and slow degradation regions where specific components show their own particular behaviour. Matching magnetic resonance techniques with other standard analyses (Temperature, pH, C/N) could be of great importance to fully describe the stages of composting and define the degree of compost maturity.


Figure 7 First row (a): 700 ms < T1 < 3000 ms in the T1 maps during composting time; second row (b): 50 ms < T1 < 700 ms in the T1 maps during composting time, in (a) areas highlighted by circles represent components with high relaxation times; in (b) the brighter areas have T1 > 700 ms.

References

1 N. Mirabella, V. Castellani and S. Sala, J. Cleaner Prod, 2013, 65, 28-41. 2 M. P. Bernal, J. A. Alburquerque and R. Moral, Bioresour. Technol., 2009, 100 (22),

5444-5453. 3 M. F. Marcone, S. Wang, W. Albabish, S. Nie, D. Somnarain and A. Hill, Food Res. Int.,

2013, 51, 729-747. 4 P. Butz, C. Hofmann and B.Tauscher, Journal of Food Science, 2005, 70, R131–R141. 5 V. Bellon-Maurel, O. Orliac and P. Christen, Process Biochem., 2003, 38 (6), 881-896. 6 S.G Hall, D. Aneshansley and L.P. Walker, American Society of Agricultural

Engineering, Chicago, IL, 1995, Paper no. 953563. 7 R. Rynk, On-Farm Composting Handbook. Northeast Regional Agricultural Engineering

Service, Cooperative Extension. Ithaca, New York, 1992. 8 F. C. Miller, Microb. Ecol., 1989, 18, 59-71. 9 J.M. Agnew and J.J. Leonard, Compost Science & Utilization, 2003, 1, 238-264. 10 G. C. Borgia, V. Bortolotti and P. Fantazzini, J. Appl. Phys., 2001, 90 (3), 1155-1163. 11 CCQC, 2001. Compost Maturity Index. California Compost Quality Council. 12 V. Bortolotti, R.J.S. Brown and P. Fantazzini, UpenWin: a software to invert multi-

exponential decay data, 2009, Distributed by University of Bologna: [email protected].

13 G.C. Borgia, V. Bortolotti, R.J.S. Brown and P. Fantazzini, Magn. Reson. Imaging, 1996, 14, 895-897.

14 C. Sundberg and H. Jönsson, Waste Manag., 2008, 28 (3), 518–526.

1H NMR SPECTROSCOPY OF LIPOPROTEINS – WHEN SIZE MATTERS

F. Savorani and S.B. Engelsen

Spectroscopy and Chemometrics, Department of Food Science, Faculty of Science, University of Copenhagen, Rolighedsvej 30, 1958 Frederiksberg C, Denmark

1 INTRODUCTION

Lipoproteins are a wide class of biological structures represented by micellar aggregates of biomolecules that allow the fat content to be transported through the hydrophilic blood environment.

Chylomicrons, the largest lipoprotein structures, are assembled in the intestinal mucosa to carry the dietary fats through the intestinal barrier and into the blood stream by active exocytosis. The fat load of the chylomicrons, primarily in the form of triglycerides, is later distributed to the tissues by hydrolysis through the extracellular enzyme lipoprotein lipase (LPL) which is present in many tissues. The released fatty acids can then be taken up by the adipose tissue (for storage) or by skeletal muscle (as energy substrate) or the fatty acids can bind to serum albumin. The lifetime of the chylomicrons is limited to a postprandial state, since their function rapidly vanishes when all meal’s fat has been absorbed into the circulatory system.

The general structure of plasma lipoprotein particles is sketched in Figure 1. Plasma lipoproteins are characterized by an external phospholipidic monolayer membrane, giving the micelle its hydrophilic characteristic, whose surface is intercalated with free cholesterol and apo-proteins of different nature. The internal core is composed by triglycerides and in minor part (but greatly varying among sub classes) by esterified cholesterol. They can be roughly divided into five main sub-classes that mainly differ for their i) micellar size, ii) density (related to the percentage of their triglycerides content), and iii) specific apo-protein which determines their biological role. The liver, and more specifically its cells, hepatocytes, are the factories in which lipoprotein micelles are assembled and provided with their specific apo-proteins and fat content. They are then released into the blood stream to reach the target organs and their cells, distributing their fat and cholesterol content and therefore becoming smaller and more dense throughout their journey.


Figure 1 Plasma Lipoprotein general structure. The main constituents are annotated

The lipoproteins VLDL (very low density lipoproteins), IDL (intermediate density lipoproteins, LDL (low density lipoproteins) and HDL (high density lipoproteins) are a group of differently sized particles which function in the transport of endogenous triglycerides and cholesterol from the liver to other tissues. Just like the chylomicrons the VLDL transport the triglycerides (and cholesterol esters) synthesized by the liver into the capillaries of adipose tissue or skeletal muscle. Here the VLDL remnants are decreased in size and enriched in cholesterol esters and then called IDL particles. In humans, about half of the circulating IDL particles are removed quickly by the liver by a process called endocytosis, within 2-6 h, whereas the other half remain in the circulation much longer where they play a central role in cholesterol transport to the tissues. HDL particles are secreted from the liver or intestinal cells and have a key function in the reverse cholesterol transport, from the tissues to the liver. They remove excess cholesterol from the membranes of tissue cells or from the surfaces of other lipoprotein particles. Disorders in the metabolism of LDL and HDL are critical to the development of atherosclerosis and coronary heart disease (CHD) some of the leading causes of death in the western countries.1 It is thus of great importance to be able to profile the size of the lipoproteins in the blood stream.

Table 1 describes the 5 main plasma lipoprotein sub-classes sorted by decreasing size and increasing density. Content of triglycerides, proteins and cholesterol are reported along with the apo-proteins that mainly characterize each sub-class.

The “gold standard” method for separating and measuring plasma lipoproteins and lipoprotein is fractionation by density gradient ultracentrifugation (UC) followed, most frequently, by colorimetric determination of the metabolite of interest2. However, this results in a time-consuming and labour-intensive method that does not allow high-throughput clinical screenings and, as a consequence, has so far prevented the assessment of postprandial


lipoprotein metabolism in large clinical trials. On the contrary, Proton Nuclear Magnetic Resonance Spectroscopy (1H-NMR) has, in the last 30 years, developed into a optimal analytical alternative that could nowadays overcome all the limits of the UC approach. The approach is based on the NMR technique’s unique sensitivity to molecular rotational and translation diffusion which in turn can be translated into size.3

Table 1 Lipoproteins main subclasses sorted by their size

Class Name Size

diameter

(nm)

Density

(g/mL)

triglycerides

& cholesterol

esters (%)

Protein

(%)

Cholesterol

(%)

Main

Apo-

proteinsChylomicrons 100-1000 <0.95 84 <2 8 ApoB48

ApoC-III

ApoC-I&II VLDL 30–80 0.95–1.006 50 10 22 ApoC-III

ApoB100

ApoE IDL 25–50 1.006–

1.019

31 18 29 ApoB100

ApoE

ApoC LDL 18–28 1.019–

1.063 8 25 50 ApoB100

HDL 5–15 >1.063 4 33 30 ApoA-I

ApoA-II

The first 1H NMR study of human blood was performed by Bock (1982)4 who analysed her own blood before and after ingestion of alcohol. A representative human blood spectrum (plasma) is shown in Figure 2 including assignments of the major metabolites observed. The figure shows two superimposed spectra of two human subjects having a drastically different lipoprotein profile. The most prominent metabolites in the NMR spectrum of blood are fat (methylene- and methyl-moieties from the lipoproteins), lactate and glucose.

A comprehensive list of metabolites in the human serum metabolome can be found in the research articles by Nicholson et al. (1995)5 and by Psychogios et al. (2011)6. Otvos7, 8 and co-workers have been pioneering the 1H NMR analysis of blood for studying the lipoprotein profiles using an approach in which the lipid signals are deconvoluted into individual lipoproteins and lipoprotein subclasses. If deconvoluted successfully their quantification is straight forward. Later a chemometric approach to the subject, which does not rely on peak deconvolution, was introduced by Dyrby et al.9-11, Savorani et al.12 and most recently by Jacobs et al.13 - all with promising results. The advantage of the latter approach is a more robust quantification, but it requires a larger calibration database of coherent NMR plasma spectra and UC lipoprotein size profile determinations.

When analysing lipoproteins, NMR spectroscopy represents the only analytical platform capable of obtaining signals from their constituent molecules while still structured in the physiological micellar form. Indeed, being non-destructive both in sample preparation and during sample analysis, NMR spectroscopy has a key role on the study of the lipoprotein


particle profile and the lipoprotein metabolism, since it is able to measure the signals from the different subclasses avoiding the biased constraint of a manual fractionation as it happens for the UC reference method. Other successful metabolomics platforms such as LC-MS and GC-MS require that the micellar state is broken and thus they lose the important size information. While it is the chemical composition of the constituent triglycerides and cholesterol esters that allow NMR to separate the different signal arising from the lipoprotein protons, it is the size that determines the diffusion behaviour of each lipoprotein subclass, enabling a small but efficient separation among them.11

Figure 2 Human plasma (heparin) 600 MHz NMR superimposed spectra acquired at 310 K using a cryoprobe. Two extremely different lipoprotein profiles are shown In comparison to the UC reference method, NMR spectroscopy boasts drastically shorter times for the analysis (minutes instead of about 1 day) and much smaller samples size (100-500 μl plasma instead of several ml) which makes it well suited for high-throughput analysis. Furthermore, the manual slicing requested for separating the lipoprotein subclasses in UC, results in larger analytical errors and reproducibility problems which are significantly reduced when NMR analysis is adopted. Indeed, the NMR approach provides a much higher reproducibility. On the downside, the NMR signals from the lipoprotein main constituents, the triglycerides, give rise to very broad overlapping signals, combined by all the contributing lipoprotein subclasses, which cannot be efficiently solved by classical statistical methods. In order to disentangle such overlapping contributions, several approaches have been attempted, all with good results. A “classical” approach has been for several years to deconvolute the overlapped signals fitting several underlying curves according to the expected number of subclasses.1 More recently, also because of the increased computational power of modern computers, a chemometric approach on the full featured, full resolution NMR spectra has been attempted with great success. The slightly different peak shapes and shifts are modelled in a multivariate fashion providing detailed information on the lipoprotein composition in a continuous fashion which is presumably a better representation of the biological distribution. It is also possible to sensitise the NMR measurements to translational diffusion (2D Diffusion Edited experiments) which can be modelled to achieve an even better definition of the lipoprotein particle distribution.10


In the latter approach the di usion coe cients (D) are obtained and the hydrodynamic radii of the particles (R) obtained from the Stokes–Einstein equation:

where k is the Boltzmann constant, T is the temperature and is the plasma viscosity. It is common practice in blood clinical tests to ask the patient to come to the analytical laboratory in the morning hours in a fasting status, meaning that normally no food is allowed after the last dinner. Thus, the withdrawn blood spot samples is representative of a fasting status and only after it has been taken the subject is allowed to have breakfast, whose very meaning is, indeed, to break the fasting status! However, we spend the majority of our life in a postprandial status and such spot samples cannot tell us anything about the ability of a subject to deal with a fat, protein or carbohydrate load. Moreover, chylomicrons are rarely investigated (practically absent in fasting blood) but they actually represent our usual postprandial status much better than the other lipoprotein sub-classes. Their concentration, but also the clearance ability of a subject to restore the pre-meal metabolic levels, can tell us much more about a subject’s health status than what can be derived from a “classical” clinical assay on fasting blood. In order to disclose this immense potential, an inexpensive and rapid analytical method is essential and NMR, assisted by multivariate data analysis, can definitely play a central role in future large scale clinical trials addressing important issues such as obesity and the development of diabetes II. 2 MATERIALS AND METHODS Nowadays, many metabolomics investigation are run worldwide in cohorts with increasing size and biological variance. Plasma represents, together with urine, the most studied biofluid and the enormous amount of acquired data might play a crucial role in future applications that goes beyond the current aim of each single project generating the data. This, we postulate, is particularly true with respect to lipoprotein profiles hidden in the NMR spectra. However, if we want to unleash this potential, it is of utmost importance that Standard Operating Procedures (SOPs) are defined for both sample collection, storage, preparation and analysis. The variance introduced to the data because of a variation in the SOPs can easily be larger than any biological variance investigated, in particular the delicate and overlapped nature of the lipoprotein signals. So far, plasma measurements with regard to lipoprotein profiling have been like the wild west, which mean that important efforts with the cumbersome UC reference method has been lost and that data fusion and calibration transfers made impossible. This should be avoided in future by adopting common rigorous SOPs. Importantly, a new SOP should not only focus on fasting plasma but allow for post prandial dynamic sampling and measurements which much better describe how the organism is able to cope with the metabolic challenges introduced with a meal, a bioactive compound or a session of exercise. This is especially true for the lipoprotein particle profiling, that represents a very dynamic scenario and cannot exhaustively be described with a single spot blood sample. In the following a couple of real experiments, used as examples, are given in which the lipoprotein particle profile has resulted fundamental for achieving and understanding the results.


2.1 New insight from a -glucan intervention study

14 normo-cholesterolemic subjects were investigated in a randomised, blinded, cross-over 3-week intervention study aiming at proving the hypocholesterolemic effects of 3.3 g -glucan/day from three equally sized but structurally different oat and barley -glucans.14 Individuals were set in a free-living condition and maintained their habitual diet with a few restrictions with respect to fibre intake, alcohol and exercise. Blood plasma samples were collected as fasting (0 h) and postprandial (2 and 4 h) after a -glucan or control test meal at the start (day 1) and end (day 22) of each fibre treatment period for a total of 6 samples/subject for each of the 4 different meals (3 enriched in -glucans and a control one). Plasma samples were prepared in heparin containing centrifuge tubes and stored at -80 °C until NMR analysis. For each plasma sample colorimetric kits were used to obtain total and LDL cholesterol as well as plasma triglycerides (TG) total concentration for both fasting and postprandial status.

Once thawed, 300 l of each plasma sample were transferred to 5 mm NMR tubes containing 300 l phosphate buffer (pH 7.4), trimethylsilyl propionate (TSP) as reference signal and 10% D2O for the lock signal. NMR FID’s were recorded on a Bruker Avance III 600 spectrometer (Bruker Biospin GmbH) operating at a Larmor frequency of 600.13 MHz for protons, equipped with a double tuned cryo-probe (TCI) set for 5 mm sample tubes and a cooled autosampler (SampleJet). Spectra were acquired at 310 K and fixed receiver gain (RG) value previously optimized, using the Carr–Purcell Meiboom–Gill (cpmg) experiment as described by Beckonert et al. (2007).15 Each FID was collected using a total of 128 scans. FID’s were then zerofilled to 64 K points and apodised by 0.3 Hz Lorentzian line broadening prior to Fourier transformation. Each obtained spectrum was baseline- and phase-corrected automatically. Spectral data were imported into Matlab (Mathworks, Inc.) into a data table sized 328 x 65536 before data preprocessing. The spectral area chosen for multivariate data analysis ranged between 0.0 and 8.0 ppm with exclusion of the 4.4–4.9 ppm region dominated by the residual water signal. To correct for spectral misalignment the entire dataset was globally aligned with respect to the -d-glucose signal around 5.25 ppm using the icoshift algorithm.16 For correcting absolute intensity variation among spectra, 1-Norm and TSP area normalisation approaches were evaluated but eventually the aligned non-normalised data was selected as the most unbiased data block.

2.2 Postprandial blood sampling: chylomicron triglycerides

For this study 18 young and slightly overweight male subjects (age: 27.2 ± 4.0; BMI: 25.4 ± 2.2 kg/m2) where monitored up to 7 hours after an oral meal selected in turn out of four test iso-caloric meals enriched in fibre content. Blood samples were collected in EDTA centrifuge tubes starting from the overnight fasting status (T=0 min) and, after fibre enriched meal consumption, breaking their fast, again after T= 30, 60, 120, 180, 240, 300 and 420 minutes for a total of 8 samples/subject/meal summing up to 576 obtained plasma samples. Ultra Centrifugation (UC) + colorimetric measurement assay was performed in order to obtain the “gold standard” values for chylomicron TG content along with some other clinical lipoprotein values.12

The NMR spectra were obtained by mixing 500 l plasma and 60 l D2O and acquiring them at 310 K on a Bruker Avance 500 spectrometer (Bruker Biospin GmbH) operating at a Larmor frequency of 500.13 MHz for 1H (11.75 Tesla) and using a 1H/13C Z-gradient 4 mm diameter, 120 l active volume flow-probe. The FIDs’ were recorded using the zgcppr Bruker


standard pulse sequence. The relaxation delay was set to 5.0 s during which water presaturation was performed at 4.700 ppm. The 90 degree pulses had duration of 8.0 s. The acquisition time for acquiring 32 k data point free induction decays was 1.573 s using a spectral width of 20.8278 ppm. A total of 64 scans were acquired and each free induction decay was zero-filled to 64 k points and apodized by a Lorentzian line-broadening of 0.3 Hz. Phase and baseline correction were performed manually using Bruker Topspin™ 1.3 software. All spectra were subsequently imported into Matlab (Mathworks, Inc.) and then aligned and referenced with respect to the doublet signal of the anomeric proton of -D-Glucose at 5.230 ppm. No interval alignment was performed in order to preserve the natural distribution of the highly overlapping lipoprotein signals located at about 0.80 and 1.30 ppm and representing the triglyceride’s terminal methyl groups and chain methylene ones respectively. 2.3 Partial Least Squares regression using spectral intervals (iPLS) When multivariate calibration is needed, PLS regression17 is by far the primary calibration tool and iPLS represents its interval-based extension. iPLS18 relies on a stepwise algorithm that calculates, validates and compares PLS models of user-defined intervals in the variable space. These intervals can be equally distributed among the variable range or can be user defined and tailored to specific needs. The data set X matrix is therefore divided into i intervals that are modelled for PLS regression with a y vector of reference values (see Scheme I).

Scheme I iPLS (regular) segmentation of the variable (NMR) space and the regression to the response variable (UC) The validated performances (RMSECV) of each interval model i are then displayed and compared each other and with the model using all intervals (global model X) in a visual highly informative plot that allows spotting at-a-glance those spectral regions that are better correlated with the response y variable in the regression equation. This concept works great with NMR spectral data in which the identification of important regions, towards biomarker discovery, becomes therefore straightforward, enabling also a supervised variable (interval) selection that can help refining the models.19 It is of utmost importance to stress that this segmentation, opposite to Binning, does not lead to a reduction of data but, rather, it provides “an overall picture of the relevant information in different spectral subdivisions, focusing on important spectral regions”.18

218 Magnetic Resonance in Food Science: Defi ning Food by Magnetic Resonance When working with spectroscopic data, especially in the case of NMR spectra of biofluids, it must be taken into account that different data/spectral regions:

1. are often different in chemical quality content 2. may exhibit different dynamics and scale 3. may exhibit different amounts of signals (density) 4. may exhibit differences in horizontal alignments 5. may exhibit differences in baseline noise

All this issues can be better addressed when working with intervals of more homogeneous and contiguous variables but the gained advantages also extend to the multivariate modelling performances since:

1. Interval models use fewer variables 2. Interval models contain fewer interferences 3. Interval methods lead to more parsimonious models 4. Interval methods lead to enhanced model performances and interpretability

All these advantages are therefore a good claim in favour of interval-based chemometrics.19 3 RESULTS AND DISCUSSION 3.1. New insight from a non-confirmatory -glucan intervention study Nutritional intervention studies are increasingly investigated with a metabolomics approach, but still, the traditional univariate statistics plays a dominant role. In the case of this human interventions study, aimed to investigate the hypothesised hypocholesterolemic effect of -glucans of different origin, also a complete metabolomic analysis was conducted in order to confirm and corroborate the results found with the traditional approach and to obtain, thanks to the multivariate insight, new additional and complementary results. 1H NMR spectroscopy, in combination with multivariate data analysis, was used to investigate the full plasma metabolic effects of daily supplementation of mixed linkage -glucans from oat and barley. Both explorative and supervised metabolomics approaches were used. In this case the variance induced by the beta-glucan supplement is by far smaller than the subject and/or the gender one, it becomes very challenging for the metabolomics approach to be able to provide sharp results which very often become of difficult interpretation. However, multivariate data analysis can decompose data into simpler and more interpretable structures: for instance, principal component analysis (PCA) on the NMR plasma spectral data set demonstrated that the main variance among samples was due to subject specific metabolomes and, even more sharply, to gender. Indeed, especially when looking at the lipoprotein main spectral region, the gender specific behaviour was unmistakable (Figure 3A), even considering the variation introduced by the diet specific effect. A supervised discriminatory approach, in which Multilevel PLS-DA20 was used to model pair-wise comparison between control group and each of the three intervention -glucan meals, showed no further significant evidence that the fibre intake was actually perturbing the metabolome and even an interval-based approach confirmed such a finding. Instead, a more focused investigation on the lipoprotein region provided some more interesting and unexpected results. Besides the sharp difference between genders, also a clear relationship of methyl and methylene peaks to BMI values was observed. Even more interestingly, an accidental fat-rich meal of some subjects, spotted as outliers by the explorative analysis,


revealed the existence of a sharpened fat-boosted profile in the lipoprotein region that, in turn, resulted to be more sensitive to the -glucan nutritional intervention (Figure 3B).

Figure 3 Lipoprotein profile differences shown by the -glucan data set: A) gender specific profiles; B) altered profiles as a consequence of a spot fat boosting meal .

In conclusion, explorative metabolomics revealed the existence of subject unique lipoprotein profiles, which especially are dependent on gender and diet. However, no significant blood metabolic exposure and effect markers were identified for intake of -glucans from oat and barley as studied by NMR metabolomics. These results are in line with the somehow conflicting results on the cholesterol-lowering properties of cereal -glucans,21 and do not fully support the hypocholesterolemic effects of -glucans; nevertheless, a larger potential was observed for oat -glucan, presumably due to its higher solubility and viscosity. The multivariate metabolomic approach, still in perfect agreement with the results obtained using classical univariate statistics,22 has therefore provided a wider insight on the metabolic and dietary aspects involved, that may help on fine tuning the experimental design of future similar studies for overcoming the spotted problems and increasing their meaningfulness.

3.2 Postprandial blood sampling

Frequent blood sampling enables to investigate also the capability of TG clearance at a subject level and investigate the dynamic aspects rather than the spot analysis when evaluating individual’s health status12. A research study for which chylomicrons clinical data are available over time represents a perfect opportunity for investigating the potential of a dynamic metabolomic approach. In this study each participant was followed from a fasting status up to 7 hours after a meal to investigate both how his plasma metabolome behaves when a food challenge is faced and the subjective fat clearance rate of participants in different health status. The analysis were performed using both the UC reference method and NMR metabolomics, and aimed at developing a calibration model for predicting the chylomicron triglyceride (TG) content from the NMR spectra. iPLS was used to correlate the NMR spectral profiles with the UC-TG values performing a regression X(NMR) y(UC). The result of such an iPLS model is shown in Figure 4A.


Figure 4 iPLS model and Chylomicron TG calibration: A) iPLS informative plot in which the performance of each interval model (all built using 3 LVs’) are represented by bars measuring their RMSECV of prediction and the dashed line represents the error for the global model. The highlighted interval is the one performing the best; B) Calibration curve of NMR-TG validated values (y) predicted by the best interval versus UC-TG measured values (x) The highly informative plot in Figure 4A shows at-a-glance that at least 2 intervals, both containing signals from fatty acids (compare with figure 2), are performing better that the model built for the whole spectral range and the one highlighted has the better performance in terms of Root Mean Square Error of cross validation (RMSECV). This interval contains the complex features resulting from the contribution of all -CH2- chain protons of all fatty acids of the different lipoprotein sub-classes. When a calibration curve is plotted for its PLS model built with 3 Latent Variables (LVs’) the line depicted in Figure 4B is obtained, showing an impressing high correlation (r2 = 0.92) with the UC-TG measured values with a cross validated error RMSECV = 0.156 mmol/l. Although it is very high, the squared Pearson’s coefficient obtained with this regression might have been even better if not limited by the higher analytical error caused by the UC approach which was reported to be UC-TG = 0.55 mmol/l. The described method is therefore a relatively simple multivariate model that facilitates parsimonious and accurate prediction of chylomicron lipids from NMR spectra of plasma. This hereto unachieved performance, both in terms of analytical speed and accuracy, discloses the possibility to perform large scale clinical and nutritional trials creating new opportunities for research in lifestyle diseases and obesity. To determine an individual's ability to clear postprandial lipids or the capacity of a food ingredient to diminish lipid uptake, TG-rich lipoproteins and, in particular, chylomicrons TG content can be measured after an oral load. The clearance profiles obtained using either the NMR predicted TG values or the UC-TG measured values are shown in Figure 5. Figure 5A and B show a very good agreement between NMR predictions and UC measurements. Also another important information can be inferred: the 5 represented subjects can easily be divided into two groups, one capable to restore the blood homeostasis after 7 hours and the other one (subj. 4 and 5) being much slower in clearing the fat load. As a matter of fact, these two groups might represent, because of their dynamics, a way for assessing an individual’s health status more that the bare absolute concentration of certain metabolites and this possibility must be further investigated.


Figure 5 TG clearance profiles for 5 subjects: A) PLS predicted chylomicron TG values from NMR spectra B) Ultracentrifugation chylomicron TG (UC-TG) reference values

4 OUTREACH

In conclusion, the new NMR-based high-throughput method for lipoprotein profiling has great potential for future nutritional metabolomics studies focused on developing stratified nutrition for different populations such as elderly, people in risk of poverty, etc… The new iPLS-based method enables extraordinarily fast, inexpensive, and robust prediction of absorption kinetics of chylomicrons. The high-throughput nature of the new lipoprotein profiling method will allow real-time measurements of the return to normal homeostasis after a food challenge and thus provide a much better understanding of food digestion and health than the current static methods. It creates new opportunities for research in lifestyle diseases and obesity, becoming a valuable tool in nutritional research for assessment of absorption of exogenous diet-derived lipids.

In metabolomics only NMR can measure lipoprotein profiles. NMR is unique in this sense by being sensitive to physical phenomena such as molecular tumbling and diffusion.3 Basically, the lipoprotein particle measurement by NMR provides a smooth quantitative measurement over a broad range of lipoprotein sizes which has to be calibrated to an accepted reference method (UC). However, so far the lipoprotein measures with NMR have only been made with relative small and relative homogeneous cohorts. In order to validate the current calibrations, the method have to be reinforced by new cohorts spanning more variance and we currently have plans to reinforce our previous cohort with a large cohort of Danish elderly in a one-year-long intervention study that includes exercise sessions and dietary supplements. On the longer term, the continuous nature of the NMR profiles might even be applicable across different animal species, although a comparison between some common species shows that it might be difficult because of the drastically different lipoprotein profiles (see Figure 6).


Figure 6 Comparison of NMR spectra of plasma from human, rat (both acquired at 500 MHz) and porcine blood (acquired at 600 MHz). References 1 M. Ala-Korpela, Progress in Nuclear Magnetic Resonance Spectroscopy, 1995, 27, 475. 2 M. J. Chapman, S. Goldstein, D. Lagrange and P. M. Laplaud, Journal of Lipid

Research, 1981, 22, 339. 3 F. Savorani, M. A. Rasmussen, M. S. Mikkelsen and S. B. Engelsen, Food Research

International, 2013, 54, 1131. 4 J. L. Bock, Clinical Chemistry, 1982, 28, 1873. 5 J. K. Nicholson, P. J. D. Foxall, M. Spraul, R. D. Farrant and J. C. Lindon, Analytical

Chemistry, 1995, 67, 793. 6 N. Psychogios, D. D. Hau, J. Peng, A. C. Guo, R. Mandal, S. Bouatra, I. Sinelnikov, R.

Krishnamurthy, R. Eisner, B. Gautam, N. Young, J. G. Xia, C. Knox, E. Dong, P. Huang, Z. Hollander, T. L. Pedersen, S. R. Smith, F. Bamforth, R. Greiner, B. McManus, J. W. Newman, T. Goodfriend and D. S. Wishart, PLoS One, 2011, 6.

7 J. D. Otvos, E. J. Jeyarajah and D. W. Bennett, Clinical Chemistry, 1991, 37, 377. 8 J. D. Otvos, E. J. Jeyarajah, D. W. Bennett and R. M. Krauss, Clinical Chemistry, 1992,

38, 1632. 9 M. Dyrby, M. Pedersen, S. B. Engelsen, L. Nørgaard and U. Sidelmann, in Magnetic

Resonance in Food Science: Latest Developments, eds. G. A. Webb, P. S. Belton and D. N. Rutledge, The Royal Society of Chemistry, Cambridge (UK), 2002, pp. 101

10 M. Dyrby, M. Petersen, A. K. Whittaker, L. Lambert, L. Nørgaard, R. Bro and S. B. Engelsen, Analytica Chimica Acta, 2005, 531, 209.

11 M. Petersen, M. Dyrby, S. Toubro, S. B. Engelsen, L. Nørgaard, H. T. Pedersen and J. Dyerberg, Clinical Chemistry, 2005, 51, 1457.

12 F. Savorani, M. Kristensen, F. H. Larsen, A. Astrup and S. B. Engelsen, Nutrition & Metabolism, 2010, 7.

13 V. V. Mihaleva, D. B. van Schalkwijk, A. A. de Graaf, J. van Duynhoven, F. A. van Dorsten, J. Vervoort, A. Smilde, J. A. Westerhuis and D. M. Jacobs, Analytical Chemistry, 2014, 86, 543.

14 M. S. Mikkelsen, F. Savorani, M. A. Rasmussen, B. M. Jespersen, M. Kristensen and S. B. Engelsen, Food Research International.

15 O. Beckonert, H. C. Keun, T. M. D. Ebbels, J. G. Bundy, E. Holmes, J. C. Lindon and J. K. Nicholson, Nature Protocols, 2007, 2, 2692.


16 F. Savorani, G. Tomasi and S. B. Engelsen, Journal of Magnetic Resonance, 2010, 202, 190.

17 S. Wold, H. Martens and H. Wold, Lecture Notes in Mathematics, 1983, 973, 286. 18 L. Nørgaard, A. Saudland, J. Wagner, J. P. Nielsen, L. Munck and S. B. Engelsen,

Applied Spectroscopy, 2000, 54, 413. 19 F. Savorani, M. A. Rasmussen, Å. Rinnan and S. B. Engelsen, in Data Handling in

Science and Technology, ed. M. Federico, Elsevier, 2013, vol. Volume 28, pp. 449. 20 L. Ståhle and S. Wold, Journal of Chemometrics, 1987, 1 185. 21 L. Brown, B. Rosner, W. W. Willett and F. M. Sacks, American Journal of Clinical

Nutrition, 1999, 69, 30. 22 S. Ibrugger, M. Kristensen, M. W. Poulsen, M. S. Mikkelsen, J. Ejsing, B. M. Jespersen,

L. O. Dragsted, S. B. Engelsen and S. Bugel, Journal of Nutrition, 2013, 143, 1579.

Subject Index Apple ................................................... 127 Aroma ................................................. 171 Automation ........................................... 77 Biomarker ...................................143; 161 Blood ...........................................161; 211 Butter .................................................. 171 Cancer ................................................. 190 Cheese ...........................................51; 171 Cholesterol .......................................... 211 Coeliac ................................................ 190 Compost .............................................. 203 CPMG ................................................... 51 Crocetin ................................................. 65 Cultivar .......................................127; 154 Cwfp ........................................................ 1 Danbo .................................................. 171 Diet ..............................................143; 190 Diffusion ........ 51; 84; 111; 120; 127; 211 Diffusometry ....................................... 111 Digestion ............................................... 31 Enterobacteriaceae .............................. 181 Fingerprint ....................................77; 190 Gut ..............................................143; 190 HR-MAS ............................................. 171 Hydrolysates ......................................... 40 Hydrolysis .......................................31; 40 Imaging ............................................... 101 Industrial ............................................... 77 Isolation ................................................ 40 Lipoprotein .......................................... 211 Metabolome ............... 143; 161; 190; 211 Metabolomics .... 143; 154; 161; 171; 190;

211 Microbiota ...................................143; 190 Microporosity ...................................... 127 Milk .............................. 51; 161; 171; 190 Monocultivar ....................................... 154 Monosaccharides .................................. 40 MRI .................................... 101; 127; 203 Mussels ............................................... 181

Nanoparticles ...................................... 111 Nutrimetabonomics ............................ 143 Obesity ................................................ 211 Oil ............................ 19; 84; 93; 154; 161 Oleuropein ............................................ 84 Olive ........................ 19; 84; 93; 154; 190 Oregano .............................................. 161 Organic .............. 161; 171; 181; 190; 203 Owo .................................................... 120 Palm ...................................................... 93 PFG-NMR .............................. 1; 111; 120 Plasma ................................. 143; 190; 211 Polysaccharides .................................... 40 Polyunsaturated fat ....................... 19; 154 Pork ............................................... 19; 101 Proteins ................................................. 51 Quantitative ................................ 1; 31; 40 Reactor ................................................ 203 Relaxation . 1; 31; 51; 101; 111; 120; 127;

171; 203 Relaxometry ..................51; 101; 127; 203 Ripening .............................................. 190 Saffron .................................................. 65 Saturated fat ............................ 19; 93; 154 Saturation ................................................ 1 Seafood ............................................... 181 Sequences ............................................... 1 Spoilage .............................................. 181 Storage ................................................ 181 Sunflower .............................................. 19 Triglycerides ........................... 19; 31; 211 Unsaturated fat ...................................... 93 Urine ........................................... 143; 190 Vegetables .......................................... 143 Vegetarian ........................................... 190 Whey ..................................................... 51 Wine ...................................... 77; 143; 190 Wow .................................................... 120 Xanthan ................................................. 40

Documents

Magnetic resonance in food science : defining food by magnetic resonance