9
* To whom correspondence should be addressed. Fax: #45 35 28 32 45; E-mail: se@kvl.dk Lebensm.-Wiss. u.-Technol., 33, 103 } 111 (2000) Prediction of Sensory Texture of Cooked Potatoes using Uniaxial Compression, Near Infrared Spectroscopy and Low Field 1H NMR Spectroscopy A. K. Thybo, I. E. Bechmann, M. Martens and S. B. Engelsen* A. K. Thybo: Danish Institute of Agricultural Sciences, Department of Fruit, Vegetable and Food Science, P. O. Box 102, DK-5792 Aarslev (Denmark) I. E. Bechmann, S. B. Engelsen: The Royal Veterinary and Agricultural University, Department of Dairy and Food Science, Food Technology, Rolighedsvej 30, DK-1958 Frederiksberg C (Denmark) M. Martens: The Royal Veterinary and Agricultural University, Department of Dairy and Food Science, Sensory Science, Rolighedsvej 30, DK-1958 Frederiksberg C (Denmark) (Received April 16, 1999; accepted October 28, 1999) The present work evaluated the ability of uniaxial compression, near-infrared reyectance (NIR) and low xeld pulsed 1 H nuclear magnetic resonance (LF-NMR) in predicting the sensory texture quality of 24 samples of cooked potato by partial least squares regression (PLSR). The best predictions of the sensory texture proxle were found for (1) LF-NMR measures (Carr-Purcell-Meiboom-Gill relaxation) on raw potatoes and (2) uniaxial compression on cooked potatoes combined with the chemical measure dry matter and pectin methylesterase activity. Among the sensory variables, the root mean square error of prediction indicated springiness, xrmness, moistness and chewiness to be better predicted than the geometrical variables reyection from surface, mealiness and graininess. Transverse relaxation times, determined according to bi-exponential xtting, resulted in a fast relaxing (T 21 of 100 ms) common water component for raw and cooked potatoes and a slower relaxing component with T 22 of 250 ms for cooked and 500 ms for raw potatoes. ¹he only covariant NMR parameter was the amount of the slow relaxing component (¹ 22 ) which correlates negatively to dry matter (r"!0.85) and to mealiness (r"!0.77), and positively to moistness (r"0.75). This study clearly demonstrates that LF-NMR (CPMG) relaxation on raw potato samples can be applied as an alternative rapid method for detecting sensory texture of cooked potatoes. ( 2000 Academic Press Keywords: potato; texture; sensory; uniaxial compression; near-infrared re#ectance; low "eld pulsed 1H nuclear magnetic resonance Introduction Texture is very important for the consumer's perception of potato quality. Consumer texture preference of cooked potatoes varies with age, within and between countries and is highly dependent on processing, con"rming the need for sensory texture characterization of cooked po- tatoes. Methods for measurement of the sensory texture pro"le of cooked potatoes have been developed and validated in several studies, grouping the texture pro"le into mechanical, geometrical and moisture dimensions (1}4). Not only consumers are concerned about the tex- ture of potatoes, but also the potato processing industry which produces various products which are highly de- pendent on the rheological properties of the cooked product. Measurement of quality characteristics by sens- ory methods is in general time-consuming and not well- suited for industrial routine control. For this reason, the industry is demanding on-line instrumental methods which are able to predict sensory texture quality of the processed product or, even better, to predict sensory texture quality of the product directly from measure- ments on the raw material. A number of instrumental techniques for texture charac- terization are presently available, but the correlation to the sensory texture attributes of individual products is not always known. For example, double axial compres- sion (texture pro"le analysis, TPA) and uniaxial com- pression are widely used for texture determination of fruit, vegetables, gels, cheeses and potatoes. The relation between these measures and the sensory attributes on cooked potatoes is only sparsely treated in the literature (5,6). Thybo and Martens (4) studied the relationship between TPA, uniaxial compression and sensory text- ure quality of cooked potatoes and found uniaxial compression data (stress and strain at fracture and modulus of deformability) to be better predictors of the 0023-6438/00/020103 #09 $35.00/0 doi:10.1006/fstl.1999.0623 ( 2000 Academic Press All articles available online at http://www.idealibrary.com on 103

Prediction of Sensory Texture of Cooked Potatoes using Uniaxial Compression, Near Infrared Spectroscopy and Low Field1H NMR Spectroscopy

Embed Size (px)

Citation preview

Lebensm.-Wiss. u.-Technol., 33, 103}111 (2000)

Prediction of Sensory Texture of Cooked Potatoes usingUniaxial Compression, Near Infrared Spectroscopy and

Low Field 1H NMR SpectroscopyA. K. Thybo, I. E. Bechmann, M. Martens and S. B. Engelsen*

A. K. Thybo: Danish Institute of Agricultural Sciences, Department of Fruit, Vegetable and Food Science, P. O. Box102, DK-5792 Aarslev (Denmark)

I. E. Bechmann, S. B. Engelsen: The Royal Veterinary and Agricultural University, Department of Dairy and FoodScience, Food Technology, Rolighedsvej 30, DK-1958 Frederiksberg C (Denmark)

M. Martens: The Royal Veterinary and Agricultural University, Department of Dairy and Food Science, SensoryScience, Rolighedsvej 30, DK-1958 Frederiksberg C (Denmark)

(Received April 16, 1999; accepted October 28, 1999)

The present work evaluated the ability of uniaxial compression, near-infrared reyectance (NIR) and low xeld pulsed 1H nuclear magneticresonance (LF-NMR) in predicting the sensory texture quality of 24 samples of cooked potato by partial least squares regression (PLSR).The best predictions of the sensory texture proxle were found for (1) LF-NMR measures (Carr-Purcell-Meiboom-Gill relaxation) on rawpotatoes and (2) uniaxial compression on cooked potatoes combined with the chemical measure dry matter and pectin methylesteraseactivity. Among the sensory variables, the root mean square error of prediction indicated springiness, xrmness, moistness and chewinessto be better predicted than the geometrical variables reyection from surface, mealiness and graininess. Transverse relaxation times,determined according to bi-exponential xtting, resulted in a fast relaxing (T21 of 100 ms) common water component for raw and cookedpotatoes and a slower relaxing component with T22 of 250 ms for cooked and 500 ms for raw potatoes. ¹he only covariant NMRparameter was the amount of the slow relaxing component (¹22 ) which correlates negatively to dry matter (r"!0.85) and tomealiness (r"!0.77), and positively to moistness (r"0.75). This study clearly demonstrates that LF-NMR (CPMG) relaxation onraw potato samples can be applied as an alternative rapid method for detecting sensory texture of cooked potatoes.

( 2000 Academic Press

Keywords: potato; texture; sensory; uniaxial compression; near-infrared re#ectance; low "eld pulsed 1H nuclear magnetic resonance

Introduction

Texture is very important for the consumer's perceptionof potato quality. Consumer texture preference of cookedpotatoes varies with age, within and between countriesand is highly dependent on processing, con"rming theneed for sensory texture characterization of cooked po-tatoes. Methods for measurement of the sensory texturepro"le of cooked potatoes have been developed andvalidated in several studies, grouping the texture pro"leinto mechanical, geometrical and moisture dimensions(1}4). Not only consumers are concerned about the tex-ture of potatoes, but also the potato processing industrywhich produces various products which are highly de-pendent on the rheological properties of the cookedproduct. Measurement of quality characteristics by sens-ory methods is in general time-consuming and not well-

*To whom correspondence should be addressed. Fax: #45 35 28 3245; E-mail: [email protected]

0023-6438/00/020103#09 $35.00/0( 2000 Academic Press All ar

10

suited for industrial routine control. For this reason, theindustry is demanding on-line instrumental methodswhich are able to predict sensory texture quality of theprocessed product or, even better, to predict sensorytexture quality of the product directly from measure-ments on the raw material.A number of instrumental techniques for texture charac-terization are presently available, but the correlation tothe sensory texture attributes of individual products isnot always known. For example, double axial compres-sion (texture pro"le analysis, TPA) and uniaxial com-pression are widely used for texture determination offruit, vegetables, gels, cheeses and potatoes. The relationbetween these measures and the sensory attributes oncooked potatoes is only sparsely treated in the literature(5,6). Thybo and Martens (4) studied the relationshipbetween TPA, uniaxial compression and sensory text-ure quality of cooked potatoes and found uniaxialcompression data (stress and strain at fracture andmodulus of deformability) to be better predictors of the

doi:10.1006/fstl.1999.0623ticles available online at http://www.idealibrary.com on

3

lwt/vol. 33 (2000) No. 2

mechanical sensory attributes than the TPA measures.As uniaxial compression mainly predicts the mechanicalattributes (hardness, "rmness and springiness), thismethod is not well-suited for description of the overalltexture pro"le. By combining the uniaxial compressionwith measures of dry matter and pectin methylesterase,however, approximately 70% of the mechanical, geomet-rical and moistness attributes were described (4). Assess-ment of sensory texture quality is of great importance inthe potato industry. The need for rapid methods hascaused a growing interest in the application of spectro-scopic screening methods, such as nuclear magnetic res-onance (NMR) or near infrared (NIR) spectroscopy, asversatile tools in food analysis. Previous studies showedNIR to be a good predictor of hardness, juiciness andmealiness of peas (7) and adhesiveness and stickiness ofcooked rice (8). Using NIR measurements on steam-cooked potatoes, the sensory attributes moist, waxy, "rmand mealy were better predicted than the attributesgrainy, crumbly and sticky (9). Low-"eld pulsed1H NMR measuring the proton relaxation in the mater-ial has been successfully related to instrumental hardnessof wild rice (10), to instrumental "rmness of stale bread(11), to sensory juiciness and tenderness of porcine meat(12) and to sensory "brousness and toughness of frozencod mince (13).While the conventional sensory quality parameters arethe ultimate measures which are directly interpretable,they are costly and time consuming to perform. Spectro-scopic sensors, on the other hand, have the advantagethat the measurements are rapid, noninvasive and can beinstalled on-line/at-line in a production chain. However,spectroscopic sensors will require multivariate calib-rations to be related to sensory and chemical referenceanalyses. Application of chemometric algorithms such aspartial least squares regression (PLSR) (14) is a moste$cient way to detect correlations and to calibratespectral data to sensory quality attributes.The aim of the present work was: "rstly, to study theprediction of the sensory texture pro"le by variousmethods such as uniaxial compression combined withchemical analysis, NIR and LF-NMR; secondly, to "ndoptimal methods for sensory texture prediction; and"nally, to study the ability of instrumental measurementson raw samples to predict the sensory texture quality ofcooked potatoes. This current study is meant to form thebasis of future work that will enable on-line or at-lineanalysis of potatoes in industrial production to sort theraw material prior to processing. It should be noted,however, that on-line implementation of any of the pro-posed instrumental methods will require a large amountof calibration work including collection of very expensivereference data (sensory or chemical) which may representa large barrier.

Material and Methods

PotatoesTwenty-four samples representing six potato varietiesgrown conventionally (Bintje, Asva, Oleva, Dali, Sava,

10

Folva), four varieties grown organically (Sava, Agria,Folva, Forelle) and one variety, Sava, grown by twovariable organic fertilization methods, were harvested atfull maturity in September 1997. All samples were grownin 1}2 "eld replicates, making a total of 24 samples(Table 1). The samples were density-graded in salt solu-tions and the density fraction representing the most tu-bers was analysed by uniaxial compression, chemicalanalysis, NIR, LF-NMR and sensory evaluation in Octo-ber 1997. The six conventionally grown varieties werestored at 5 3C and 95% RH until May 1998 and analysedusing the same methods. The di!erences between potatovarieties were known to span a texture variation relevantfor cooked potatoes (4), but textural e!ects of organicfertilization methods were not known.

Sensory texture analysisPotatoes were peeled and boiled in water for 20}25 minuntil they were cooked through. All samples were ana-lysed in three replications in a randomized design withsix samples per sensory session. The samples were evalu-ated hot and one tuber per sample was served. A panel often trained assessors evaluated texture by quantitativedescriptive texture analysis, as previously described (2).The sensory attributes included one visual attribute,namely re#ection from surface, and eight oral attributes:hardness, "rmness, springiness, adhesiveness, graininess,mealiness, moistness and chewiness. The attributes wereevaluated on a 1 to 9 point line scale with the anchorpoint &none' on the left side and &very strong' on the rightside, except for chewiness, which ranged from &few chew-ings' to &many chewings'.

Uniaxial compressionUniaxial compression on raw and cooked potatoes wasmeasured on cylinders (d"12 mm, h"10 mm) from 15potatoes, respectively (Fig. 1a), as described previously (4),at a deformation rate of 20 mm/min. Modulus of deforma-bility (E

$), stress (p

&) and strain (e

&) at fracture were cal-

culated and means of 15 repetitions were obtained.

Chemical analysisDry matter (g/100 g) and activity of pectin methylesterase(kmol/min )g) were analysed and calculated on freshweight basis, as described by Thybo and Martens (4).

Near infrared spectroscopyTwo slices of approximately 5 mm were cut transversely 1

4of the length from the stem end of 10 potato tubers(Fig. 1b). Ten slices were measured by NIR in the rawstage immediately after being stamped to circular slices of37 mm in diameter to "t in the standard quartz samplecups closed with a compressible paper disc. Ten sliceswere cooked for 12 min in 350 mL water, cooled to roomtemperature for approximately 3 h, covered with plastic"lm, trimmed to circular slices (d"37 mm) and mea-sured by NIR. No visual after-cooking discoloration was

4

Table 1 Sample information and dry matter content of the 24 potato samples

Dry matter Dry matterSample Growing Storage fraction meanabbreviations Variety method (mo) (g/100 g) (g/100 g)

bi1 Bintje Non-organic 1 20}22 21.9as1 Asva Non-organic 1 18}20 19.8ol1 Oleva Non-organic 1 26}28 26.4da1 Dali Non-organic 1 18}20 18.6sa1 Sava Non-organic 1 20}22 22.2fo1 Folva Non-organic 1 20}22 21.6sa1a Sava Organica 1 16}18 18.1sa1a Sava Organica 1 16}18 17.8fo1a Folva Organica 1 16}18 18.4fo1a Folva Organica 1 16}18 18.6ag1a Agria Organica 1 16}18 17.3ag1a Agria Organica 1 16}18 18.4for1a Forelle Organica 1 16}18 18.0for1a Forelle Organica 1 16}18 18.1sa1b Sava Organicb 1 16}18 18.3sa1b Sava Organicb 1 16}18 18.3sa1c Sava Organicc 1 16}18 18.4sa1c Sava Organicc 1 16}18 18.4bi8 Bintje Non-organic 8 20}22 22.5as8 Asva Non-organic 8 18}20 19.5ol8 Oleva Non-organic 8 26}28 26.8da8 Dali Non-organic 8 18}20 19.8sa8 Sava Non-organic 8 20}22 22.2fo8 Folva Non-organic 8 20}22 21.0

a,b,c representing di!erent organic growing methods

Fig. 1 Positions within the potato tuber of a) cylinders foruniaxial compression or NMR analysis and b) slices for NIRanalysis

lwt/vol. 33 (2000) No. 2

observed. The NIR measurements were made with anNIR Systems Inc. spectrophotometer (model 6500, SilverSprings, Maryland, U.S.A.) equipped with a rotatingsampling device. The spectrophotometer uses a split de-tector system with a silicon (Si) detector between 400 and1100 nm and a lead sulphide (PbS) detector from 1100 to2498 nm. The angle of the illuminating light source was1803 and re#ectance from the tissue surface was mea-sured at a 45 3 angle. The spectra were collected at ambi-ent temperature in log (1/R) units (R"re#ectance) overthe wavelength range 400}2498 nm at 2 nm intervals.The 24 samples were grouped in four fractions and ana-lysed in randomized order by each fraction. Meanspectra of 10 repetitions were calculated.

10

Nuclear magnetic resonanceTwo cylinders (d"13 mm, l"25 mm) were stampedlongitudinally from the stem end from 10 potato tubers(Fig. 1a). Ten cylinders were cooked for 10 min in 175 mLwater. Raw and cooked samples were stored at 5 3C for"ve to nine h and equilibrated to 35 3C for exactly 30 minbefore analysis. The height of the samples was trimmedto weight of approximately 5.00}5.30 g and the exactweight was measured. Measurements of NMR were per-formed by measuring the transverse relaxation time (¹

2)

of the water protons using a MARAN pulsed 1H NMRbenchtop analyser (Resonance Instruments, Witney,U.K.) operating at a frequency of 23.2 MHz and equip-ped with an 18 mm variable temperature probe. The¹

2relaxation was measured as one free induction decay

(FID) signal (16 384 points with 16 ks intervals) and oneCarr-Purcell-Meiboom-Gill (CPMG) pulse sequencewith a q-spacing of 100 ms. In the CPMG sequence thesignal amplitude was measured every second echo (intotal, 4096 data points) and 16 acquisitions were accumu-lated with a repetition time of 8 s. Unless otherwisenoted, the relaxation data are treated in the multivariateanalysis as is, but for extraction of ¹

2values, the CPMG

relaxation curves were least squares "tted to a two com-ponent exponential model (15). The samples weremeasured at 35 3C (magnet temperature) in the samerandomized order as used for the NIR measurements.

Statistical analysisUnivariate statistical di!erences between the potato sam-ples were studied by ANOVA (SAS version 6.12, Cary,

5

lwt/vol. 33 (2000) No. 2

U.S.A.). Multivariate data analysis was performed onmean data using Unscrambler (CAMO A/S version 7.01,Norway). The structure in the sensory, uniaxial compres-sion, NIR and LF-NMR data were revealed by principalcomponent analysis (PCA). Partial least squares regres-sion (PLSR) was used to predict the nine sensory at-tributes (Y-variables) from the instrumental data (X-vari-ables). All FID and CPMG relaxation data were dividedby weight using MatLab (version 5.2, The MathWorksInc., MA, U.S.A.). The sensory, instrumental and chem-ical variables were standardized and full cross validationwas used as validation method. The accuracy of themodels was expressed by root mean square error ofprediction (RMSEP) (14) of the respective sensory at-tributes and correlation coe$cients between measuredand predicted variables were given.

Results and Discussion

Overall sample informationMean values, s

9and the range of variation (min.}max.

values) of each nonspectral variable are shown in Table 2.The dry matter content known to be highly related to thetexture quality of cooked potatoes spanned from 16.2 to26.8 g/100 g. This range in dry matter covered the rel-evant texture variation of cooked table potatoes (16).Table 1 illustrates the sample names, abbreviations, in-formation and the dry matter fraction from which sam-ples for all analyses were taken and the mean dry matterof each sample. It is obvious that the organically grown

Table 2 Mean values (x), standard deviations (s9) and

minimum (min) and maximum (max) values of all cul-tivars for all measured variables

Variables x (s9) min}max

ChemicalDry matter (g/100 g) 19.7 (2.7) 16.2}26.8Pectin methylesterase activity 2.77 (0.89) 1.45}4.96[(kmol/min )g)]

Uniaxial compression (raw samples)Stress at fracture (kPa) 1599 (160) 1248}1909Strain at fracture (-) 0.39 (0.02) 0.34}0.43Modulus of deformability 711 (179) 406}1140(kPa)

Uniaxial compression (cooked samples)Stress at fracture (kPa) 95.0 (23.6) 42.7}133.1Strain at fracture (-) 0.15 (0.02) 0.12}0.20Modulus of deformability 235 (60) 118}340(kPa)

SensoryRe#ection from surface (-) 2.7 (1.7) 1.3}7.9Hardness (-) 4.5 (0.8) 2.9}5.9Firmness (-) 4.6 (0.9) 2.1}5.9Springiness (-) 3.3 (0.8) 1.4}4.4Adhesiveness (-) 3.9 (0.6) 2.7}5.0Graininess (-) 3.2 (1.1) 2.3}6.9Mealiness (-) 3.2 (1.6) 1.7}7.4Moistness (-) 4.6 (1.1) 2.1}6.4Chewiness (-) 4.2 (0.8) 2.9}5.8

[-] indicates that the measure is dimensionless.

10

potatoes have signi"cantly lower dry matter content thanthe traditionally grown potatoes.Univariate data analysis of all variables showed a signi"-cant e!ect of potato samples on sensory, chemical anduniaxial compression data (P(0.05). For all variables,except for a few sensory attributes, a signi"cant e!ect ofstorage time was observed, indicating di$culties for thepanel in remembering the sensory scoring level after7 mo.

Sensory texture proxling of cooked potatoesPrincipal component analysis was applied to the ninesensory attributes to investigate the relevant and inter-pretable structure in the data. Four signi"cant PCs de-scribed 94% of the total validation variance (Fig. 2a, b).The main variation (PC1) in the sensory data classi"edcooked potatoes with respect to mechanical (springinessand "rmness) vs. geometrical (re#ection from starchgrains on surface, mealiness and graininess) properties.Hardness and adhesiveness were negatively correlated tomoistness (PC2), and it seemed that four of the six var-ieties being stored were less hard and more moist thannonstored potatoes. Furthermore, adhesive potatoes hadlow levels of graininess (PC3) and springy potatoes hadlow levels of moistness (PC4). The scores of the potatosamples showed large di!erences between conventionallyand organically grown varieties in PC1 as well as minordi!erences between the organic growing methods withrespect to the texture attributes (PC2). Varietal di!er-ences were described by PC1 and PC3 and the e!ect ofstorage by PC2 and PC4, showing stored potatoes to bemore moist and less hard, adhesive and springy. TheSava variety had the same dry matter level as the mealysample, Bintje (Table 1), but was not evaluated as a mealysample, showing the dry matter content not to be the onlyfactor responsible for the mealiness. The revealed structurein the sensory texture pro"le has been observed previously(1}3) on di!erent data sets, con"rming the interpretationof four relevant dimensions in the sensory data.

Uniaxial compression and chemical dataPrevious results have shown that uniaxial compressiondata (stress, strain and modulus of deformability) weregood predictors of the sensory mechanical properties ofhardness, "rmness and springiness, but bad predictors ofthe geometrical and moistness attributes. Combining theuniaxial compression data with a few chemical analysessuch as dry matter and activity of pectin methylesterase,approximately 70% of the overall texture pro"le wasexplained in a PLSR model (4). In the present study,however, only 64% of the sensory texture variation wasdescribed in a PLSR calibration by uniaxial compres-sion, dry matter and pectin methylesterase activity.

Near infrared measurements of raw and cooked potatoesFigure 3 shows examples of the NIR spectra of two rawand cooked potato samples. Water is one of the bestknown NIR absorbers, and due to their high water

6

Fig. 2 Scores ( . ) and loadings (f ) showing the structure in the sensory texture data after PCA (a) PC1 vs. PC2 and (b) PC3 vs. PC4.Sample abbreviations are explained in Table 1

lwt/vol. 33 (2000) No. 2

content, the NIR spectra of the raw and cooked potatosamples are dominated by the three broad water peaks at970 nm (O-H stretch, second overtone), 1450 nm (O-Hstretch, "rst overtone) and 1940 nm (O-H stretch#O-Hdeformation). The last broad peak at approximately1200 nm is due to second overtones of C-H stretch vibra-tions. When comparing raw and cooked samples, themain variations in the spectra were observed in the visualregion from 425 to about 500 nm, probably due to caro-tenoids and generally a less structured and more absorb-ent NIR region for the cooked samples. In the visualinterval 425}500 nm, the least yellow and mealy varieties,Oleva and Bintje, nonstored and stored, have low ab-sorbancies. The opposite is seen for the more yellowsamples. In the water regions, cooked samples with highdry matter content (Bintje, Oleva, Sava) had lower ab-sorbance than samples with lower dry matter content.This trend was not as clear for the raw samples, however,

10

which is in some contrast to the results obtained byBoeriu et al. (9). For starch, the in#uential wavelengthwas 2100 nm (Fig. 3) and thus relevant for potatoes sinceapproximately 80% of the dry matter is composed ofstarch. Principle component analysis on the NIR data ofthe raw potato samples showed two signi"cant PCsto account for 93% of the total variation (Fig. 4),grouping the samples with respect to nonstored organi-cally grown varieties, nonstored conventionally grownand stored conventionally grown potatoes. This PCAscore pattern was di!erent from the pattern observed inthe PCA plot of the sensory data (Fig. 2a) in which mostof the sensory variation accounted for a variety di!erencein mealy, grainy varieties vs. "rm, springy varieties. Thedi!erence is related to the fact that sensory analysismainly probes the mouthfeel or texture characteristics,while NIR mainly probes total content of water andstarch.

7

Fig. 3 NIR spectra of two potato samples in raw and cookedstage and a potato starch for reference. Sample abbreviationsare explained in Table 1. (**), Oleva raw (ol1); (} } }), Savaraw (sa1c); () ) ) )), Oleva cooked (ol1), very mealy; (* -* - -),Sava cooked (sac) very "rm springy; (00), potato starch

Fig. 4 Score plot of NIR spectra of 24 raw samples after PCA.Sample abbreviations are explained in Table 1. Clusters: 1)organic grown potatoes stored for 1 mo, 2) conventionallygrown potatoes stored for 1 mo and 3) conventionally grownpotatoes stored for 8 mo. For key, see Fig. 3

Fig. 5 CPMG relaxation curves of two potato samples inraw and cooked stage. Sample abbreviations are explained inTable 1.

Fig. 6 Score plot of CPMG relaxation curves of 24 raw samplesafter PCA. Sample abbreviations are explained in Table 1.Cluster (1), organically grown potatoes stored for 1 mo

lwt/vol. 33 (2000) No. 2

NMR measurements of raw and cooked potatoesThe CPMG relaxation of two raw and cooked potatovarieties are presented in Fig. 5. The variation in thesignals between samples in the raw as well as the cookedstage showed a faster relaxing signal for the mealy, grainysamples (with high dry matter content), nonstored andstored Oleva and Bintje. The samples with high levels ofsensory moistness, springiness or "rmness and low drymatter content had slower decreases in signal. Interest-ingly, PCA analysis on the CPMG relaxation curves ofraw samples (Fig. 6) di!erentiated between the samples insimilar groups as seen for the sensory data with respect tothe mealy, grainy varieties of Oleva and Bintje, "rm,springy organically grown varieties and moist samples(Fig. 2a).Bi-exponential "tting (tri-exponential curve "ts werejudged as over"t) of the CPMG relaxation curves of bothraw and cooked potatoes resulted in a common, fastrelaxing component with a ¹

2(¹

21) of approximately

10

100 ms and a slower relaxing component with ¹2

(¹22

)relaxation times of approximately 500 ms and 250 ms,respectively. The di!erence in the ¹

22relaxation between

raw and cooked is tentatively ascribed to starch gelatiniz-ation, providing a more structured water compartmen-talization in the intracellular matrix. The slow relaxingcomponent, ¹

22(approximately 70%), is therefore as-

signed to intracellular water in a relative nonrestrictedsituation in the raw potato matrix and to intracellularwater in a more restricted situation after gelatinization inthe cooked potato matrix. Then, the fast relaxing andconserved component, ¹

21(approximately 30%), might

be assigned to di!usion-hindered extracellular water situ-ated in the vascular tissue (water transport tissue) andto water in the moister inner medulla. The only (butweak) correlation to this component was adhesiveness(r"!0.56). The fact that the amount of the ¹

22com-

ponent in the raw samples is highly but negatively cor-related to the dry matter content (r"!0.85),grainy (r"!0.73), mealy (r"!0.77) and chewiness(r"!0.80), and only positively correlated to springi-ness (r"0.71) and moistness (r"0.75) supports theassignment of this NMR component. Analysis of cooked

8

Tab

le3

Val

idat

ion

ofPLSR

mod

elsfo

rth

epr

edic

tion

ofth

eni

nese

nso

ryte

xtur

eat

trib

utes

from

unia

xial

com

pres

sion

and

chem

ical

data

,NIR

and

NM

Rda

ta.R

MSE

Pgi

ven

inab

solu

teva

lues

,num

ber

ofPC

san

d%

expl

ained

Y-v

aria

nces

are

give

n RM

SEP

No

%Y

var.

Pre

dic

tors

Re#

ection

Har

dne

ssFirm

nes

sSpringi

nes

sA

dhes

iven

ess

Gra

inin

ess

Mea

lines

sM

oist

ness

Chew

ines

sPC

sE

xpl.

aCom

p.#

chem

.0.

660.

520.

430.

430.

560.

450.

720.

620.

532

69.6

bRaw

com

p.#

chem

1.00

0.71

0.63

0.50

0.57

0.51

0.82

0.62

0.51

459

.4N

IR-c

ook

ed0.

990.

640.

660.

570.

510.

720.

870.

580.

573

57.7

NIR

-raw

1.08

0.53

0.62

0.60

0.60

0.84

1.06

0.75

0.59

347

.3C

PM

G-c

ooke

d1.

420.

630.

720.

510.

510.

931.

100.

740.

512

46.4

CPM

G-r

aw0.

910.

640.

530.

410.

430.

720.

680.

560.

425

68.3

FID

-cooke

d1.

770.

720.

880.

670.

611.

161.

520.

980.

692

13.5

FID

-raw

1.40

0.75

0.80

0.58

0.59

0.95

1.00

0.60

0.47

543

.3

aU

nia

xial

com

pres

sion

on

cooke

dsa

mpl

esco

mbi

ned

with

dry

mat

ter

and

pec

tin

met

hyl

este

rase

activi

tybU

nia

xial

com

pres

sion

on

raw

sam

ple

sco

mbin

edw

ith

dry

mat

ter

and

pec

tin

met

hyle

ster

ase

activi

ty

lwt/vol. 33 (2000) No. 2

10

samples resulted in exactly the same trend, but withsigni"cantly lower correlations. For example, the amountof the ¹

22component has a correlation coe$cient of

r"!0.72 to dry matter content and r"0.58 to moist-ness. The reason for the less convincing results on thecooked samples remains unsolved, but is probablyrelated to the fact that this component is now heavilyin#uenced by the gelatinization process and therefore lessdistinct.

Comparison of uniaxial compression, NIR and NMR aspredictors of sensory texture attributesDi!erent data sets were subjected to PLSR using (i)uniaxial compression in combination with dry matterand pectin methylesterase activity, (ii) NIR spectra, (iii)CPMG signals and (iv) FID signals of raw and cookedsamples as predictors for the sensory data. All nine sens-ory attributes were included, as they all were relevant indescribing the sensory variation. Optimal numbers ofPCs, RMSEPs, explained variances and correlation coef-"cients between measured and predicted texture at-tributes are shown in Tables 3 and 4. Using spectral dataas regressors, FID signals gave the highest RMSEPs,illustrating this method to be the poorest method indescribing the sensory texture quality. The RMSEPsshowed the CPMG signals measured on raw samples tobe more accurate predictors of the texture attributes andexplained the texture pro"le to a greater extent (68.3%explained variance) than CPMG signals on cooked sam-ples (46.4% explained variance). This result was furthersubstantiated by calculated lower correlation coe$cientsbetween the amount of the ¹

22component and di!erent

quality attributes of the cooked potatoes when comparedto the raw potatoes. The CPMG signals also seemed tobe better predictors than NIR spectra on raw and cookedsamples, which was indicated by the similarities in thePCA plots of the sensory data (Fig. 2a) and the CPMGsignals (Fig. 6). The reason for NMR being a better probefor potato quality is probably related to the fact that incontrast to NIR re#ectance, NMR is a bulk measure and,perhaps more importantly, NMR is strongly phase andcompartmentalization sensitive and should therefore cor-relate to texture and mouthfeel attributes which in turnare most important sensory variables. Pretreatment ofthe NIR data by multiplicative scatter correction did notimprove the correlation to the sensory data (14). Boeriuet al. (9) found better correlations to the sensory textureattributes ("rm, moist and mealy) of steam cooked po-tatoes from NIR spectra of raw potatoes than obtained inthis study, but calibrations based on the CPMG signalsfrom the raw samples were of almost equal quality(mealiness even slightly better). With respect to nonspec-tral data, uniaxial compression measured on cookedsamples combined with dry matter and pectin methyles-terase activity (69.6% of the explained variance)predicted the texture to the same extent as theCPMG signals on raw samples (68.3% of the explainedvariance). In this study, these two methods were the bestpredictors of the texture. However, the RMSEPs in the0.41}0.91 range and correlation coe$cients between

9

Table 4 Validation of PLSR models for the prediction of the nine sensory texture attributes from uniaxialcompression and chemical data, NIR and NMR data. Coe$cients of correlation between measured and predictedsensory texture attributes are given

Coe$cients of correlation

Predictors Re#ection Hardness Firmness Springiness Adhesiveness Graininess Mealiness Moistness Chewiness

aComp.#chem. 0.93 0.70 0.88 0.83 0.33 0.92 0.88 0.79 0.72bRaw comp.#chem 0.82 0.40 0.70 0.76 0.33 0.89 0.85 0.79 0.75NIR-cooked 0.83 0.50 0.67 0.67 0.54 0.77 0.83 0.82 0.67NIR-raw 0.79 0.69 0.71 0.62 0.25 0.66 0.73 0.67 0.63CPMG-cooked 0.61 0.52 0.60 0.75 0.52 0.58 0.71 0.68 0.74CPMG-raw 0.86 0.54 0.80 0.85 0.70 0.77 0.90 0.84 0.84FID-cooked 0.33 0.28 0.30 0.52 0.07 0.28 0.34 0.36 0.45FID-raw 0.64 0.43 0.55 0.70 0.37 0.58 0.77 0.80 0.79

aUniaxial compression on cooked samples combined with dry matter and pectin methylesterase activitybUniaxial compression on raw samples combined with dry matter and pectin methylesterase activity

Fig. 7 PLSR predicted vs. measured plot (with target line) ofmoistness by (j) CPMG signals (5 PCs) and by (h) uniaxialcompression, dry matter and pectin methylesterase (2 PCs). Forj, r"0.84; for h, r"0.79

Fig. 8 PLSR predicted vs. measured plot (with target line) ofsensory mealiness by (j) CPMG signals (5 PCs) and by (h)uniaxial compression, dry matter and pectin methylesterase (2PCs). For j, r"0.90; for h, r"0.88

lwt/vol. 33 (2000) No. 2

measured and predicted texture attributes of 0.33}0.93indicated that prediction of some of the sensory at-tributes was more reliable than prediction of others.Attributes such as springiness, "rmness, moistness andchewiness were reliably predicted by both methods(RMSEPs of 0.41}0.62) with correlation coe$cients inthe range of 0.72}0.88 (Fig. 7). With respect to the geo-metrical attributes (re#ection from surface, graininessand mealiness), these attributes had higher RMSEPs(0.45}0.91) for both methods, but also high correlationcoe$cients (0.77}0.93) caused by a wide span of theseattributes in high or low scores (Fig. 8). In this experi-ment, hardness was predicted by a very low correlationcoe$cient which was in contrast to previous "ndingsusing uniaxial compression combined with dry matterand pectin methylesterase (4). The attribute adhesiveness,known to be the lowest predictable sensory attribute byuniaxial compression and chemical analysis (4), was con-

11

"rmed in this study too (r"0.33), but in contrast,CPMG signals predicted this attribute better (r"0.70).This study showed that the ¹

22water signals measured

by LF-NMR are di!erent in raw and cooked stages, andthat they are more related to mechanical and moistnessattributes than to geometrical sensory properties. These"ndings were supported by high correlations found be-tween LF-NMR and the mechanical property tendernessin porcine meat (12), hardness of rice (10) and toughnessand "brousness of "sh (13). McComber et al. (17) studiedthe relationship between NMR ¹

2measurements and

the sensory geometrical property mealiness of potatoesfor one mealy and one nonmealy variety only but werenot able to detect signi"cant di!erences. Since there havebeen very few studies on the relationship between low"eld NMR and texture of foods (10}13,16), and the factthat a predictive model of a sensory texture pro"le usinglow "eld NMR has not been addressed, this work will

0

lwt/vol. 33 (2000) No. 2

contribute to new perspectives in using low "eld NMRto predict the textural quality of foods. Findings inthis experiment were based on 24 samples. More sam-ples would have supported the conclusions to a higherextent.

Conclusion

The sensory texture variation of 24 cooked potato sam-ples was grouped into geometrical, mechanical andmoistness dimensions covering a total of nine textureattributes. The prediction of the sensory attributes wasstudied by instrumental methods such as (i) uniaxialcompression combined with dry matter content and ac-tivity of pectin methylesterase, (ii) NIR and (iii) low "eld1H NMR on raw and cooked potato samples. In thisexperiment, the methods were found to di!erentiate be-tween the potato samples. The best predictions werefound for "rstly, CPMG signals of raw samples andsecondly, uniaxial compression on cooked samples sup-plemented by the chemical data, both methods explain-ing about 70% of the total variation. Among the sensoryvariables, springiness, "rmness, moistness and chewinesswere best predicted. For the geometrical attributes meali-ness, graininess and re#ection from surface, and hard-ness, the predictability was not as good. Bi-exponential"tting of the CPMG relaxation curves resulted in a fastrelaxing (¹

21"100 ms) common water component of

raw and cooked potato samples and a slower relaxingcomponent ¹

22of 500 ms for raw and 250 ms for cooked

samples. The amount of the ¹22

component in rawsamples displayed the highest correlation to the sensorytexture pro"le by negative correlation to the geometricalattributes and chewiness as well as positive correlation tothe mechanical attributes and moistness. Accordingly, itseems that CPMG relaxation measured on raw samplescan be applied as an alternative faster method for detect-ing sensory texture quality of cooked potatoes. Theseindications about LF-NMR as a rapid method for pre-dicting the "nal texture quality from raw stage measure-ments are of great importance for the industry.

Acknowledgements

The authors wish to thank the Danish Ministry of Food,Agriculture and Fisheries (under the projects &The Centreof Raw Materials' and &The Research Centre for OrganicAgriculture'), the Danish Food Research Center for Ad-vanced Studies (LMC) and the Danish Directorate forDevelopment (nonfood) for "nancial support. Thanks aredue to the assessors from The Danish Institute of Agri-cultural Sciences for participation in sensory analyses,Elisabeth Kjemtrup for performing the uniaxial compres-sion measurements and to Poul Erik Lvrke and JensPeter M+lgaard for providing the material. In addi-tion, we thank Henrik Toft Pedersen for his technicalassistance with the NMR measurements, Lisbeth G.

11

Thygesen and Helle J. Martens for valuable discussionsand Gilda Kischinovsky for help with the manuscript.

References

1 VAN MARLE, J. T., VAN DER VUURST DE VRIES, R., WILKIN-

SON, E. C. AND YUKSEL, D. Sensory evaluation of thetexture of steam-cooked table potatoes. Potato Research, 40,79}90 (1997)

2 THYBO, A. K. AND MARTENS, M. Development of a sensorytexture pro"le of cooked potatoes by multivariate dataanalysis. Journal of ¹exture Studies, 29, 453}468 (1998)

3 THYBO, A. K. AND MARTENS, M. Analysis of sensory as-sessors in texture pro"ling by multivariate modelling. FoodQuality and Preference (2000a), in press

4 THYBO, A. K. AND MARTENS, M. Instrumental and sensorycharacterization of cooked potato texture. Journal of ¹ex-ture Studies (2000b), in press

5 LEUNG, H. K., BARRON, F. H. AND DAVIS, D. C. Texturaland rheological properties of cooked potatoes. Journal ofFood Science, 48, 1470}1474, 1496 (1983)

6 TRUONG, V. D., WALTER, W. M. JR. AND HAMANN, D. D.Relationship between instrumental and sensory parametersof cooked sweet potato texture. Journal of ¹exture Studies,28, 163}185 (1997)

7 MARTENS, M. AND MARTENS, H. Near-infrared re#ectancedetermination of sensory quality of peas. Applied Spectro-scopy, 40, 303}310 (1986)

8 WINDHAM, W. R., LYON, B. G., CHAMPAGNE, E. T., BAR-

TON, F. E. II, WEB, B. D., MCCLUNG, A. M., MOLDEN-

HAUER, K. A., LINSCOMBE, S. AND MCKENZIE, K. S. Predic-tion of cooked rice texture quality using near-infraredre#ectance analysis of whole-grain milled samples. CerealChemistry, 74, 626}632 (1997)

9 BOUERIU, C. G., YUKSEL, D., VAN DER VUURST DE VRIES,R., STOLLE-SMITS, T. AND VAN DIJK, C. Correlation be-tween near infrared spectra and texture pro"ling of steamcooked potatoes. Journal of Near Infrared Spectroscopy, 6,A291}A297 (1998)

10 RUAN, R. R., ZOU, C., WADHAWAN, C., MARTINEZ, B.,CHEN, P. L. AND ADDIS, P. Studies of hardness and watermobility of cooked wild rice using nuclear magnetic reson-ance. Journal of Food Processing and Preservation, 21,91}104 (1997)

11 SEOW, C. C., TEO, C. H., Staling of starch-based products:A comparative study by "rmness and pulsed NMRmeasurements. Starch/StaK rke, 3, 90}93 (1996)

12 FJELKNER-MODIG, S. AND TORNBERG, E. Water distribu-tion in porcine M. longissimus dorsi in relation to sensoryproperties. Meat Science, 17, 213}231 (1986)

13 STEEN, C. AND LAMBELET, P. Texture changes in frozen codmince measured by low-"eld nuclear magnetic resonancespectroscopy. Journal of the Science in Food and Agriculture,75, 268}272 (1997)

14 MARTENS, H. AND NAES, T. Multivariate Calibration. NewYork: John Wiley and Sons Ltd., pp. 73}163 (1989)

15 BECHMANN, I. E., PEDERSEN, H. T., N+RGAARD, L. AND

ENGELSEN, S. B. Comparative Chemometric Analysis ofTransverse Low-Field 1H NMR relaxation Data. In: BELT-

ON, P. S., HILLS, B. P. AND WEBB, G. A. (Eds), Advances inMagnetic Resonance in Food Science. Cambridge UK: TheRoyal Society of Chemistry, pp. 217}225 (1999)

16 BURTON, W. G. Cooking and processing quality. In: BUR-

TON, W. G. (Ed.), ¹he Potato. Essex: Longman Scienti"cand Technical, pp. 392}405 (1989)

17 MCCOMBER, D. R., HORNER, H. T., CHAMBERLIN, M. A.AND COX, D. F. Potato cultivar di!erences associated withmealiness. Journal of Agriculture and Food Chemistry, 42,2433}2439 (1994)

1