Microchemical Journal 117 (2014) 242–249

Contents lists available at ScienceDirect

Microchemical Journal

j ourna l homepage: www.e lsev ie r .com/ locate /mic roc

Chemometric approach to optimize the operational parameters of ESI forthe determination of contaminants of emerging concern in aqueousmatrices by LC-IT-TOF-HRMS

Keila Letícia Teixeira Rodrigues ⁎, Ananda Lima Sanson, Amanda de Vasconcelos Quaresma,Rafaela de Paiva Gomes, Gilmare Antônia da Silva, Robson José de Cássia Franco AfonsoInstitute of Exact and Biologic Sciences, Chemistry Department, Federal University of Ouro Preto — UFOP, 35400-000 Ouro Preto, MG, Brazil

⁎ Corresponding author. Tel.: +55 31 91594240.E-mail addresses: keilaleticia17@hotmail.com (K.L.T. R

ananda_lima@hotmail.com (A.L. Sanson), amanda.qui.uforafaela_p_gomes@hotmail.com (R.P. Gomes), gilmare@gmrobsonafonso@iceb.ufop.br (R.J.C.F. Afonso).

http://dx.doi.org/10.1016/j.microc.2014.06.0170026-265X/© 2014 Elsevier B.V. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 24 March 2014Received in revised form 12 June 2014Accepted 12 June 2014Available online 30 June 2014

Keywords:Contaminants of emerging concernLC-IT-TOF-HRMSESIMultivariate optimizationDoehlert designKohonen neural network

Contaminants of emerging concern are organic compounds used in large quantities by the society for various pur-poses. Theyhave shownbiological activity at low concentrations,which gives great environmental relevance. Thedifficulty to detect and quantify contaminants of emerging concern in the environment stimulates the develop-ment of appropriate analytical methods. In this work a chemometric approach to positive and negativeelectrospray ionization (ESI) optimization for the simultaneous determination of contaminants of emergingconcern in water samples by liquid chromatography-ion trap-time of flight-high resolution mass spectrometry(LC-IT-TOF-HRMS) was applied. Three types of phase modifiers were used: formic acid, ammonium hydroxideand formic acid/ammonium formate. The effects of operational parameters such as mobile phase modifier con-centrations, mobile phase flow rate, heating block temperature and drying gas flow rate were evaluated by the24− 1 fractional factorial experimental design, resolution IV, in the screening phase and byDoehlert experimentaldesign. Initial factorial experimental design studies indicated that the phasemodifier ammoniumhydroxide wasmore efficient compared to the other evaluated modifiers. It provided higher ion intensities to the majority ofanalytes. Doehlert experimental design allowed finding a region indicative of the optimum experimental condi-tions formost analytes. The best experimental condition observedwas 3.5mMammoniumhydroxide concentra-tion; 0.0917 mL/min of mobile phase; 300 °C heating block temperature; and drying gas at 200 kPa. Theseoptimized parameters resulted in decreased detection limits of the method. The optimized method was appliedto the evaluation of water samples coming from the Rio Doce basin— Minas Gerais/Brazil utilizing multivariateexploratory techniques such as principal component analysis and Kohonen neural network. In this way, the useof chemometric approach showed to be a promisingway to optimize the simultaneous determination of twenty-one contaminants of emerging concern in aqueous matrices by LC-IT-TOF-HRMS using ESI.

© 2014 Elsevier B.V. All rights reserved.

1. Introduction

The contaminants of emerging concern are the indicators of anthro-pogenic activity and are associated with a diverse set of organic com-pounds that are used in large quantities by the society for variouspurposes [1–3]. The growing interest in these substances occurs mainlybecause they may have biological activity at low concentrations, whichgives them great environmental relevance [4–7]. Many analyticalmethods are being developed and refined to detect and quantify them[8–12]. A multi-residue analytical method is advantageous to reducecost and time while simultaneously obtaining information on the

odrigues),p@gmail.com (A.V. Quaresma),ail.com (G.A. da Silva),

occurrence of a broad number of compounds [13]. Therefore, a system-atic study to optimize the analytical method to the simultaneous detec-tion of contaminants of emerging concern is important.

Optimization of both ionization processes and ion transportation isof crucial importance in order to achieve high sensitivity, low detectionlimits and acceptable accuracy in liquid chromatography–mass spec-trometry (LC–MS) analysis [14]. The amount of ions reaching the detec-tor depends on the efficiency of ionization promoted by the interfaceused, which can be improved by adjusting their operational parameters[15–17].

Chemometric methods provide powerful tools for designing oroptimizing experiments and to statistically process the data with thepurpose of obtaining the maximum information [18–21]. Multivariatestatistical methods most used in chemistry can be conveniently classi-fied according to how one decides which experiments are to be execut-ed. All methods require the user to supply minimum and maximumvalues for each factor that defines the experimental domain to be

243K.L.T. Rodrigues et al. / Microchemical Journal 117 (2014) 242–249

investigated during the optimization procedure. Optimization proce-dures are frequently performed by experimental designs. The mostcommonly used designs to determine response surfaces are the fulland fractional factorial, primarily in the screening step, followed by op-timization with more complex central composite, Box–Behnken,Doehlert and mixture designs [22–24].

In thiswork thepossible parameters and their interactions that couldinfluence the ionization efficiency of electrospray ionization (ESI) sourceon the systematic study of simultaneous detection of, initially, twenty-five contaminants of emerging concern were investigated. Thecompounds studied included hormones, phthalates, pharmaceuticalcompounds, detergents by-products and plastic additives. The 24 − 1

fractional factorial experimental design, resolution IV, for screeningand Doehlert experimental design for optimization were employed.The parameters evaluated were the type of mobile phase modifier(formic acid, ammonium hydroxide and formic acid/ammonium for-mate), phase modifiers concentrations, mobile phase flow rate, heatingblock temperature and drying gas flow rate; this procedure has not beenpreviously described in the literature. The optimized method was ap-plied to the evaluation of water samples coming from the Rio Docebasin — Minas Gerais/Brazil and the results treated by multivariate ex-ploratory techniques such as principal component analysis (PCA) [25]and Kohonen neural network (Self-Organising Maps, SOM) [26], forthe determination of contamination profile.

2. Materials and methods

2.1. Chemicals and reagents

The solid-phase extraction (SPE) employed 500 mg Strata-X car-tridges (Phenomenex). LC grade methanol and the ethyl acetate werepurchased from J. T. Backer (Phillipsburg, USA) andwaterwas producedusing an ion exchange purification system (TKA Wasseraufberei-tungssysteme, Niederelbert, Germany). The buffers formic acid and am-monium hydroxide were purchased from Synth®. Ammonium formateand the standards of acetylsalicylic acid, acetaminophen, azithromycin,bezafibrate, cimetidine, ciprofloxacin, clarithromycin, diclofenac, diltia-zem, gemfibrozil, ibuprofen, miconazole, naproxen, ranitidine, sulfa-methoxazole, trimethoprim, caffeine, bisphenol-A, bis-(2-ethylhexyl)phthalate, diethyl phthalate, 4-nonylphenol, 4-octylphenol, estrone,17α-ethinylestradiol and 17β-estradiol were purchased from SigmaAldrich (St. Louis, MO, USA).

The stock solutionswere composed of amixture of the contaminantsof emerging concern standards, at a concentration of 1000 μg/L, and theappropriateworking standard solutionswere prepared inmethanol andstored in amber glass-polyethylene stopper bottles at 4 °C. For the eval-uation by the 24 − 1 fractional factorial experimental design and by theDoehlert experimental design was used a working standard solution ata concentration of 50 μg/L.

Table 1Coded and decoded values (between parentheses) of modifiers and concentrations (mM)tested by fractional factorial design for screening of the operating parameters of LC-IT-TOF-HRMS for the determination of the contaminants of emerging concern using ESI.

Type of mobile phase modifier Concentration (mM)

Formic acid −1 (2.6) 0 (14.4) 1 (26.1)Ammonium hydroxide −1 (1.5) 0 (3.0) 1 (4.5)Formic acid/ammonium formate −1 (26.1/1.6) 0 (104.4/16.7) 1 (182.7/31.8)

Note: (−1) = lowest level; (+1) = highest level; (0) = central point.

2.2. Instrumentation

Liquid chromatography was performed on a Shimadzu Prominencesystem equipped with a high-pressure binary solvent delivery system(LC-20AD) and a SIL 20AC auto-sampler, according to operational con-ditions described in § 2.2 in the Supplementary information.

The high resolution mass spectrometry (HRMS) detection was per-formed using a Shimadzu LC-IT-TOF-HRMS, a tandem ion trap (IT)and a time-of-flight (TOF) sequential mass spectrometer, workingat high resolution (10,000 FWHM) and high mass accuracy (b5 ppm)in the following conditions: electrospray ionization (ESI) at −3.5 kV(negative) and +4.5 kV (positive) with nebulizer gas at 1.5 L/min,curved desorption line (CDL) interface at 200 °C, octapole ion accumu-lation time of 100 ms and MS scan in the range 100–800 m/z.

2.3. Optimization procedures

The 24 − 1 fractional factorial design with central point was used inthe screening step in order to assess the parameters that influence thesystem (supplied in Fig. S.1 in the Supplementary information) and todelimit the experimental area that should be explored further. To refinethe experimental region, the significant effects were investigatedby using the Doehlert experimental design. This approach allowsextracting maximum information from the system being investigatedin a more efficient way.

The criterion for the selection of variableswas based on the influencethat they would provide the signal strength of the system studied, con-sidering the function of each parameter. The mobile phase compositioncan influence the chromatographic separation of analytes and it couldalso be important at the ionization. Other instrumental parameterssuch as: mobile phase flow rate, heating block temperature and dryinggas flow rate could contribute to increase the detectability of themethod. So, all these variables can and/or should be studied. In thework the type of mobile phase modifier (formic acid, ammonium hy-droxide and formic acid/ammonium formate), phasemodifiers concen-trations, mobile phase flow rate, heating block temperature and dryinggas flow rate was investigated.

Three fractional factorial experimental designs were employed inthe screening step, one for each type of mobile phasemodifier. Themo-bile phase concentration varied according to Table 1 and other parame-ters varied likewise for all three designs. Mobile phase flow rate variedfrom 0.10 to 0.30 mL/min, heating block temperature varied from 200to 300 °C and drying gas flow rate varied from 100 to 200 kPa. Each pa-rameter was varied across a high and low settingwith triplicates in cen-tral point.

The results of the fractional factorial designs demonstrated that theparameters evaluated had significant effects on the response of mostcontaminants of emerging concern studied. From these, it was selectedthat the most appropriate mobile phase modifier and the parameterspreviously studied were again investigated utilizing the Doehlert exper-imental design as follows:mobile phasemodifier concentrations (1.00 to6.00 mM, in 5 levels), mobile phase flow rate (0.05 to 0.30 mL/min, in 7levels), heating block temperature (200 to 300 °C, in 3 levels) anddryinggas flow rate was kept constant at 200 kPa. The number of experimentswas 17, including five replicates of the central point to estimate the re-peatability. Table 2 shows the applied Doehlert experimental design.

The analytical curve fitting of the method was conducted using theoptimized conditions pointed out by the Doehlert design. The adjust-ment of the curve relating the chromatographic peak areas of analyteswas obtained with increasing concentrations. Theworking standard so-lutions used were 1, 5, 10, 30, 50,100, 150 e and 200 μg/L.

The ion chromatogram was divided into appropriate operationalsegments. High-resolution scan mass spectra were obtained in all seg-ments and a selected ion monitoring (SIM) was supplied in Table S.1in the Supplementary information.

2.4. Sample analysis

2.4.1. Sample collectionThe surface water samples were collected monthly from the Rio

Doce basin — Minas Gerais/Brazil at twenty-four sampling points

Table 2Coded and decoded values (between parentheses) for the operational variables of LC-IT-TOF-HRMS for the optimization experiments determined according Doehlert experimen-tal design: x1— ammoniumhydroxide concentrations (mM), x2 —mobile phase flow rate(mL/min) and x3 — heating block temperature (°C).

Experiment x1 x2 x3

1 1 (6.00) 0 (0.1750) 0 (250)2 0.5 (4.75) 0.866 (0.3000) 0 (250)3 0.5 (4.75) 0.289 (0.2167) 0.817 (300)4 −1 (1.00) 0 (0.1750) 0 (250)5 −0.5 (2.25) −0.866 (0.0500) 0 (250)6 −0.5 (2.25) −0.289 (0.1333) −0.817 (200)7 0.5 (4.75) −0.866 (0.0500) 0 (250)8 0.5 (4.75) −0.289 (0.1333) −0.817 (200)9 −0.5 (2.25) 0.866 (0.3000) 0 (250)10 0 (3.50) 0.577 (0.2583) −0.817 (200)11 −0.5 (2.25) 0.289 (0.2167) 0.817 (300)12 0 (3.50) −0.577 (0.0917) 0.817 (300)13a 0 (3.50) 0 (0.1750) 0 (250)

Note: a = Central point with five replicates.

244 K.L.T. Rodrigues et al. / Microchemical Journal 117 (2014) 242–249

located in different villages and towns, as shown in Fig. 1. The samplingspots were coded with the same code used by the “Águas de Minas”monitoring project, of the Instituto Mineiro de Gestão das Águas —

IGAM, an agency of the state government responsible for planning andpromoting directed actions for preserving the quantity and quality ofwater of Minas Gerais/Brazil. The IGAMwas responsible for the samplecollecting of this work.

Fig. 1. Map of sampling stations in the Rio Doce basin — Minas Gerais/Brazil. Source: Adaptetrimestre/doce-2otrim-2013.pdf.

The sampling was conducted in July of 2012 (dry period — D) andJanuary of 2013 (rainy period — R). The collected samples were ran-domly taken at the water surface, preferably at the main river channel,in glass flasks, transferred to 1 L amber glass bottles and preserved bythe addition of 1% v/v HPLC-grade methanol. The samples weretransported to the laboratory in cooling boxes and prepared for analysisafter.

2.4.2. Sample preparationThe first step of sample preparation consisted of consecutive filtra-

tions of the water samples under vacuum through 8 μm blue bandpassand 1.2 μm glass-fiber filters, to remove suspended particulate matterand to avoid clogging of the SPE cartridge. The sample pHwas adjustedto 2.0 by the addition of 30% (v/v) HCl and the analytes were extractedusing SPE cartridges (Strata-X, 500 mg), preconditioned with 5 mL ofethyl acetate followed by 5mL ofmethanol and 5mL ofwater. The sam-ples (usually 500 mL) were then loaded onto the cartridges at a ratelower than 5 mL/min. After that, the cartridges were dried for 20 minunder vacuum and eluted with 10 mL of ethyl acetate. The extracts col-lected in amber glass flaskswere dried under nitrogen and resuspendedin 500 μL of methanol, concentration factor of 1000. The solutions weretransferred to sealed cap vials and analyzed by ESI-LC-IT-TOF-MS.

The extraction recovery and the matrix effect on the ion signals ofthe analytes in the ESI source were determined accordingly, asdescribed in § 2.4.2 in the Supplementary data.

d from http://www.igam.mg.gov.br/images/stories/mapoteca/Mapas/qualidade/2013/2_

245K.L.T. Rodrigues et al. / Microchemical Journal 117 (2014) 242–249

2.4.3. Software and data treatmentThe data acquisition and peak integration were performed using the

software LCMS solution from the Shimadzu Corporation.Optimization calculations were performed using spreadsheets,

available at the website of the Laboratory of Theoretical and AppliedChemometrics, State University of Campinas [27].

The software used to perform the multivariate analysis of Kohonenneural network was freely available on the Internet at page http://www.cis.hut.fi/projects/somtoolbox/ [28]. The data set was organizedinto a matrix of 45 samples (45 lines) and 17 variables (17 columns),corresponding to the contaminants detected in the samples. Beforeprocessing, the entire data set was autoscaled for all variables. Thispre-processing step ensures that all variables have the same level of im-portance, allowing users to assess the significance of all variables in thesamples [29]. The Kohonen maps were created and initialized linearly.The Kohonen neural network was trained with the data using thebatch training algorithm, the neighborhood function used in trainingwas the Gaussian, the structure was hexagonal and the shape of themapwas planar. During the data training, architectures with several or-derswere tested (from 5× 5 to 10× 10) for the evaluation of the groupsof samples and the architecture that had the best sample distribution ingroups (which was more informative) was chosen.

For comparison, a PCA analysis was also performed using the sameset of data through the computing environment GNUOctave 3.6.4, free-ly available on the Internet at page http://www.gnu.org/software/octave/; before data processing, the entire data set was autoscaled forall variables.

3. Results and discussion

3.1. Optimization procedures

Analysis of data gained from screening experiments allowed an in-sight into the basic operation of the ESI source when exposed to deter-mined conditions according to 24 − 1 fractional factorial design. Table 3shows the levels of the variables that presented better performance(increased peak areas) for each mobile phase modifier used.

A comparison among the best assays for eachmobile phasemodifieris shown in Fig. 2. It can be verified that all the twenty-five compoundswere detected at least in oneof themobile phasemodifiers studied. Am-monium hydroxide proved to be the best modifier since twenty-threecompounds were detected while with the formic acid and formic acid/ammonium formate modifiers only sixteen and seventeen compounds,respectively, were detected. In addition, ammonium hydroxide modifi-er showed higher area intensities to themajority of the compounds. So,themobile phasemodifier ammoniumhydroxidewas themost efficientcompared to the other ones.

Table 4 shows that all variables investigated in the 24 − 1 fractionalfactorial design with central point using ammonium hydroxide as mo-bile phase modifier revealed to be significant for most analytes studiedin the system, with significant coefficients at a 95% confidence level.

According to Table 4, the mobile phase flow rate was the variablewith the highest influence on the systemusing the ESI, tomost analytes.

Table 3Coded and decoded values (between parentheses) of the operational parameters studied:x1—mobile phasemodifiers concentrations (mM), x2—mobile phaseflow rate (mL/min);x3— heating block temperature (°C) and x4— drying gas flow rate (kPa), that provided thehighest peak areas on the determination of the majority of contaminants of emergingconcern by LC-IT-TOF-HRMS in the screening experiments, determined according 24 − 1

fractional factorial design, for themobile phasemodifiers formic acid, ammoniumhydrox-ide and formic acid/ammonium formate buffer.

Mobile phase modifier x1 x2 x3 x4

Formic acid −1 (2.6) −1 (0.1) −1 (200) −1 (100)Ammonium hydroxide +1 (4.5) −1 (0.1) +1 (300) +1 (200)Formic acid/ammonium formate −1 (26.1/1.6) −1 (0.1) −1 (200) −1 (100)

This parameter influences the formation of the droplets during sprayingprocess, which promotes the transfer into the gas phase ions, resultingin greater efficiency of ionization. Still, a lower flow favors the formationof spray and improves the solvent dying efficiency. The heating blocktemperature was the variable that less influenced the system.

Doehlert design was applied for the optimization of the factorsaffecting ESI source performance. Analysis of variance (ANOVA) wasperformed on the models to determine the statistical significance ofthe coefficients, related to the parameters mobile phase modifier con-centrations, mobile phase flow rate, heating block temperature andinteraction between themselves. ANOVA calculations or the total varia-tion in response calculations, examine the overall significance of eachterm in the model compared to the residual error. Terms found tohave a probability value of less than 0.05 are considered to besignificant.

Table 5 resumes the ANOVA results obtained from the Doehlert ex-perimental design to the contaminants of emerging concern evaluated.

ANOVA coefficients for all elements were statistically valid with 95%confidence level, however most of them showed lack of fit at the sameconfidence level (only acetaminophen, sulfamethoxazole and trimetho-prim exhibited an acceptable modeling).

Although most models have not been adjusted, it was possible tofind a region indicative of the optimum experimental conditions formost analytes. The best conditions for most analytes were observedwhen these were subjected to the conditions of assay 12: 3.5 mM am-monium hydroxide concentration, 0.0917 mL/min of mobile phaseflow rate, 300 °C heating block temperature and drying gas flow rateat 200 kPa.

Fig. 3 compares the results of screening experiments determined ac-cording to 24 − 1 fractional factorial experimental design resolution IVwith the Doehlert experimental design in the operational conditionsthat provided higher peak areas for most analytes and shows theimprovement for majority of emerging contaminants studied whenthe response surface methodology was used.

The adjusted analytical curve equations for twenty-one contami-nants of the twenty-three emerging contaminants of concern detectedin the ammonium hydroxide modifier based on the multioptimizedLC-IT-TOF-HRMS procedure are shown in Table S.2 in the Supplementa-ry information. Linear and quadratic models with the determination co-efficients (R2) ranging from 0.990 to 0.998, with concentrations from1.0 to 200 μg/L, according to the contaminant evaluated were found.

3.2. Sample analysis

After the analysis of the samples collected (results supplied inTable S.3 and Table S.4 in the Supplementary information) by themultioptimized LC-IT-TOF-HRMS procedure the PCA method for con-tamination profile investigation was applied. However, the necessityof 15 PC to represent 99.28% of sample variability was found.

In this way, the SOMalgorithmwas applied, since it provides an easeviewing and interpreting of data, compared with other approaches,such as PCA.

A Kohonen neural network with hexagonal grids was obtained afterperforming the multivariate analysis from the data and architectures ofseveral orders were evaluated (from 5 × 5 to 10 × 10) and the arrange-ment 8 × 8 with 64 neurons had the best sample distribution in themap.

Fig. 4 presents the formation of 9 different groups (I to IX) that werecircled. It is important tomention that samples located at the sameneu-ron or at neighboring neurons form groups with similar characteristics.The map of the variables is shown in Fig. 5, where the color bars besidethe maps indicate the intensity of each parameter evaluated; the whitecolors in these bars mean higher values and a higher importance in theformation of the groups for each variable and the black colors meanlower values.

4

6

7

8 910

11

11

12

13

14

15

15

1516

17

18

18

1819

19

1920

21

21

22

23

24

24

253.00E+06

4.00E+06

5.00E+06

6.00E+06

7.00E+06

8.00E+06

Peak

Are

a

(1) clarithromycin

(2) azithromycin

(3) ibuprofen

(4) diethylphthalate

(5) 17- estradiol

(6) acetaminophen

(7) bisphenol A

(8) estrone

(9) 17- ethinylestradiol

(10) 4-octylphenol

(11) cimetidine

(12) miconazole

(13) diclofenac

(14) 4-nonylphenol

(15) trimethoprim

(16) bezafibrate

(17) naproxen

1 2

34 4

5

56 6

7 7

10

1112

1213

13

16

16 17

1819

2020

2123 23

2425

25

0.00E+00

1.00E+06

2.00E+06

Formic acid Formic acid / ammonium formate

Ammonium hydroxide

Mobile phase

(18) dietilhexilftalato

(19) acetylsalicylic acid

(20) sulfamethoxazole

(21) ranitidine

(22) gemfibrozil

(23) ciprofloxacin

(24) caffeine

(25) diltiazem

Fig. 2. Peak area of analytes (standard solution at a concentration of 50 μg/L) obtained in the best assay performed in the 24 − 1 fractional factorial designwith central point for each phasemodifier evaluated: formic acid— Essay 1: x1=2.6mM, x2=0.1 mL/min, x3=200 °C and x4=100 kPa; formic acid/ammonium formate— Essay 1: x1= 26.1/1.6mM, x2=0.1 mL/min,x3 = 200 °C and x4 = 100 kPa and ammonium hydroxide— Essay 2: x1 = 4.5 mM, x2 = 0.1 mL/min, x3 = 200 °C and x4 = 200 kPa. Definitions: x1 —mobile phase modifiers concen-trations (mM), x2 — mobile phase flow rate (mL/min); x3 — heating block temperature (°C) and x4 — drying gas flow rate (kPa).

246 K.L.T. Rodrigues et al. / Microchemical Journal 117 (2014) 242–249

Through Figs. 4 and 5 it is possible to evaluate the profile ofmicrocontaminants contamination of the samples. The formation ofnine groups of samples, according to their microcontaminant contents

Table 4p-Values obtained by the analysis of the effects of operational variables ammonium hy-droxide concentration — x1, mobile phase flow rate — x2, heating block temperature —

x3 and drying gas flow rate — x4, proposed by the 24 − 1 fractional factorial design usingammonium hydroxide mobile phase modifier for the optimization of operating parame-ters of LC-IT-TOF-HRMS for the determination of contaminants of emerging concern.Bold values are significant at a confidence level of 0.95.

Emerging contaminant p Parameter

x1 x2 x3 x4

17α-Ethinylestradiol 0.0324 0.0059 0.0490 0.022417β-Estradiol 0.6763 0.0281 0.7814 0.16204-Nonylphenol 0.2129 0.0160 0.6645 0.15394-Octylphenol 0.3151 0.0042 0.9438 0.0172Acetaminophen 0.0437 0.1465 0.7452 0.2073Acetylsalicylic acid 0.0281 0.3684 0.2175 0.1052Bezafibrate 0.0449 0.0167 0.8531 0.6659Bis (2-ethylhexyl) phthalate 0.0023 0.0011 0.8518 0.0608Bisphenol-A 0.0449 0.0167 0.8531 0.6659Caffeine 0.0015 0.0023 0.0823 0.0825Cimetidine 0.0105 0.0050 0.8095 0.1494Diclofenac 0.1318 0.0443 0.4050 0.5112Diethylphthalate 0.0367 0.5205 0.0023 0.4231Diltiazem 0.6307 0.0381 0.8891 0.3275Estrone 0.0025 0.0084 0.0132 0.0033Gemfibrozil 0.0242 0.6286 0.4971 0.1861Ibuprofen 0.0010 0.0001 0.1807 0.0005Miconazole 0.4419 0.0275 0.7719 0.6196Naproxen 0.2415 0.0117 0.9384 0.3348Ranitidine 0.0670 0.0096 0.0977 0.0307Sulfamethoxazole 0.0105 0.0001 0.0028 0.4151Trimethoprim 0.2073 0.0080 0.1025 0.340217α-Ethinylestradiol 0.0324 0.0059 0.0490 0.0224

can be noticed. From these formations, three groups were composed byonly one sample due to the fact that RD099R sample (formation IX) wasthe only one contaminated with trimethoprim, RD056D (formation VII)was unique with very prominence caffeine values, and also presentinghigh values of gemfibrozil, naproxen and estrone and, lastly, RD053D(formation IV) was the sample that presented the lower levels of allcontaminants studied, except to diclofenac and sulfamethoxazole thatwere presented at intermediate levels within the range detected.

It was also verified that the samples of the rainy period are situatedbasically on the right of sample group map (natural waters) obtainedby Kohonen neural network and the ones of the dry period on theleft (Fig. 4); the samples were separated according to the seasonal

Table 5Contaminants of emerging concern ANOVA results by models proposed by Doehlert ex-perimental design for the optimization of operating parameters of LC-IT-TOF-HRMS.Bold values are significant at a confidence level of 0.95.

Analyte p Parameter Explained variance

Regression model Lack of fit

4-Octylphenol 1.1220 × 10−5 0.0160 0.9681Acetaminophen 4.4970 × 10−5 0.2724 0.9562Acetylsalicylic acid 7.6169 × 10−7 0.0070 0.9826Bezafibrate 0.0294 0.0002 0.8223Bis (2-ethylhexyl) phthalate 0.0003 0.0322 0.9345Bisphenol-A 0.0040 0.0032 0.8747Caffeine 0.0054 0.0009 0.8648Diclofenac 0.1342 0.0001 0.7165Diethylphthalate 0.0027 0.0002 0.8862Gemfibrozil 0.0044 0.0030 0.8954Miconazole 0.0006 0.0011 0.9386Ranitidine 0.0057 0.0482 0.8632Sulfamethoxazole 9.2447 × 10−5 0.5385 0.9483Trimethoprim 3.8019 × 10−8 0.1005 0.9911

1.00E+07

1.50E+07

2.00E+07

2.50E+07

3.00E+07

3.50E+07

Peak

Are

aScreening

Doehlert experimental design

0.00E+00

5.00E+06

Fig. 3. Peak area of contaminants of emerging concern (standard solution at a concentration of 50 μg/L) under the best condition of the screening step compared to optimized conditions byDoehlert experimental design. Screening— essay 2: x1= 4.5mM of ammoniumhydroxide, x2= 0.1 mL/min, x3= 200 °C and x4= 200 kPa; Doehlert design— essay 12: x1= 3.5mMofammonium hydroxide, x2 = 0.917 mL/min, x3 = 300 °C and x4 = 200 kPa.

247K.L.T. Rodrigues et al. / Microchemical Journal 117 (2014) 242–249

conditions. This happened mainly due to greater contamination duringthe rainy period by synthetic hormone 17α-ethinylestradiol and plasti-cizers 4-nonylphenol and 4-octylphenol that maybe are leaching fromthe soil of agricultural crops, since these compounds are also found inthe formulation of various pesticides, second Moreira 2011 [10]. Theprofile of microcontaminants contamination of the samples in the dry

Fig. 4.Map of groups of samples (natural waters) obtained by Kohonen neural netwo

period was characterized by higher contents of bisphenol-A andpharmaceuticals like bezafibrate, diclofenac, ibuprofen, ranitidine andsulfamethoxazole, although a sample of the rainy period (RD077R)presented among the other microcontaminants found, high content ofthis compound, that is probably most used in the dry period is morefrequent because of respiratory diseases.

rk, in organic contamination evaluation of Rio Doce basin — Minas Gerais/Brazil.

Fig. 5.Maps of the distribution of individual variables obtained by Kohonen neural network, in organic contamination evaluation of Rio Doce basin—Minas Gerais/Brazil. The color barsindicate the intensity of themeasured variable; thewhiter the color, themore intense the variable value and the blacker the color the less intense the variable value. 4NF— 4-nonylphenol;4OF— 4-octylphenol; AAS— Acetylsalicylic acid; BEZ— bezafibrate; BFA— bisphenol-A; CAF— caffeine; CIM— cimetidine; DCF— diclofenac; E1— estrone; EE1— 17α-ethinylestradiol;GEN — gemfibrozil; IBU — ibuprofen; NPX— naproxen; PCT— acetaminophen; RAN — ranitidine; SFZ — sulfamethoxazole; TMP — trimethoprim.

248 K.L.T. Rodrigues et al. / Microchemical Journal 117 (2014) 242–249

4. Conclusions

An efficient multioptimization methodology was developed tofind the best operational parameters of ESI for the simultaneous determi-nation of 21 contaminants of emerging concern in aqueous matrices byLC-IT-TOF-HRMSusing24− 1 fractional factorial experimental design, res-olution IV, in the screening phase, and Doehlert experimental design, asthe response surface methodology. The experimental parameter values:mobile phase modifiers concentrations, mobile phase flow rate, heatingblock temperature and drying gas flow rate were optimized to obtainmaximized chromatographic signals for the majority of contaminants ofemerging concern evaluated. These optimized parameters resulted in de-creasing of the detection limit of the method, being applied to themicrocontaminant evaluation in natural water samples from Rio Docebasin (MinasGerais, Brazil). Through thedata treatment byKohonenneu-ral network it was possible to describe the contamination profile of thesebodies of water. Therefore the use of chemometric approaches showed tobe a promising way for the optimization of the simultaneous determina-tion of contaminants of emerging concern in aqueous matrices by LC-IT-TOF-HRMS using ESI; this investigation is inedited.

Acknowledgments

The authorswould like to thank the Foundation for Research Supportof the State of Minas Gerais FAPEMIG (APQ-03864-09 and CEX - APQ-00711-09), theNational Research Council of Brazil (CNPq), the Coordina-tion for Improvement of Higher Level Personnel (CAPES), Financier ofStudies and Projects (FINEP) and the Federal University of Ouro Preto(UFOP) for the financial support of this work.

Appendix A. Supplementary data

Supplementary data to this article can be found online at http://dx.doi.org/10.1016/j.microc.2014.06.017.

References

[1] D.J. Lapworth, N. Baran, M.E. Stuart, R.S. Ward, Emerging organic contaminants ingroundwater: a review of sources, fate and occurrence, Environ. Pollut. 163(2012) 287–303.

[2] Y. Yoon, J. Ryu, B.G. Choi, S.A. Snyder, Occurrence of endocrine disrupting com-pounds, pharmaceuticals, and personal care products in the Han River (Seoul,South Korea), Sci. Total Environ. 408 (2010) 636–643.

[3] R. Rodil, J.B. Quintana, E. Concha-Graña, Emerging pollutants in sewage, surface anddrinking water in Galicia, Chemosphere 86 (2012) 1040–1049.

[4] C. Zwiener, F.H. Frimmel, LC–MS analysis in the aquatic environment and in watertreatment technology—a critical review. Part II: applications for emerging contami-nants and related pollutants, microorganisms and humic acids, Anal. Bioanal. Chem.378 (2004) 862–874.

[5] A. Waye, V.L. Trudeau, Neuroendocrine disruption: more than hormones are upset.Journal of toxicology and environmental health, J. Toxicol. Environ. Health B Crit.Rev. 14 (2011) 270–291.

[6] C.G.A. Silva, C.H. Collins, Applications of highperformance liquid chromatography forthe study of emerging organic pollutants, Quim. Nova 34 (2011) 665–676.

[7] F.F. Sodré, M.A.F. Locatelli, W.F. Jardim, Occurrence of emerging contaminants inBrazilian drinking waters: a sewage-to-tap issue, Water Air Soil Pollut. 206 (2009)57–67.

[8] H.S. Chang, K.H. Choo, B. Lee, S.J. Choi, Themethods of identification, analysis and re-moval of endocrine disrupting compounds (EDCs) in water, J. Hazard. Mater. 172(2009) 1–12.

[9] N. Jonkers, H.P.E. Kohler, A. Dammshäuser, W. Giger, Mass flows of endocrinedisruptors in the Glatt River during varying weather conditions, Environ. Pollut.157 (2009) 714–723.

[10] M. Moreira, S. Aquino, M. Coutrim, J. Silva, R. Afonso, Determination of endocrine-disrupting compounds in waters from Rio das Velhas, Brazil, by liquid chromatogra-phy/high resolution mass spectrometry (ESI-LC-IT-TOF/MS), Environ. Technol. 32(2011) 1409–1417.

[11] M.Di Carro, C. Scapolla, C. Liscio, E.Magi, Development of a fast liquid chromatography–tandem mass spectrometry method for the determination of endocrine-disruptingcompounds in waters, Anal. Bioanal. Chem. 398 (2010) 1025–1034.

[12] L. Viglino, M. Prévost, S. Sauvé, High throughput analysis of solid-bound endocrinedisruptors by LDTD-APCI-MS/MS, J. Environ. Monit. 13 (2011) 583–590.

[13] K. Nödler, T. Licha, K. Bester, M. Sauter, Development of a multi-residue analyticalmethod, based on liquid chromatography–tandem mass spectrometry, for the si-multaneous determination of 46 micro-contaminants in aqueous samples, J.Chromatogr. A 1217 (2010) 6511–6521.

[14] A. Kruve, K. Herodes, I. Leito, Optimization of electrospray interface and quadrupoleion trap mass spectrometer parameters in pesticide liquid chromatography/electrospray ionization mass spectrometry analysis, Rapid Commun. MassSpectrom. 24 (2010) 919–926.

[15] R. Kostiainen, T.J. Kauppila, Effect of eluent on the ionization process in liquidchromatography–mass spectrometry, J. Chromatogr. A 1216 (2009) 685–699.

[16] P. Kebarle, U.H. Verkerk, Electrospray: from ions in solution to ions in the gas phase,what we know now, Mass Spectrom. Rev. 28 (2009) 898–917.

[17] A. Garcia-Ac, P.A. Segura, L. Viglino, C. Gagnon, S. Sauvé, Comparison of APPI, APCIand ESI for the LC–MS/MS analysis of bezafibrate, cyclophosphamide, enalapril,methotrexate and orlistat in municipal wastewater, J. Mass Spectrom. 46 (2011)383–390.

[18] M.M. Ferreira, A.M. Antunes, M.S. Melgo, P.L.O. Volpe, Quimiometria I: calibraçãomultivariada, um tutorial, Quim. Nova 22 (1999) 724–731.

[19] E. Peré-Trepat, M. Petrovic, D. Barceló, R. Tauler, Application of chemometricmethods to the investigation of main microcontaminant sources of endocrine

249K.L.T. Rodrigues et al. / Microchemical Journal 117 (2014) 242–249

disruptors in coastal and harbour waters and sediments, Anal. Bioanal. Chem. 378(2004) 642–654.

[20] N.D. de Sousa, C.J. Ellett, M. Gilbert, A.D. Wright, Use of chemometrics to optimizethe operation of an ion source, Anal. Chem. 80 (2008) 873–877.

[21] L. Luo, Chemometrics and its applications to X-ray spectrometry, X-Ray Spectrom.35 (2006) 215–225.

[22] L.C. Nunes, G.A. da Silva, L.C. Trevizan, Simultaneous optimization by neuro-geneticapproach for analysis of plant materials by laser induced breakdown spectroscopy,Spectrochim. Acta B At. Spectrosc. 64 (2009) 565–572.

[23] G.A. da Silva, F. Augusto, R.J. Poppi, Simultaneous optimization by neuro-genetic ap-proach of a multiresidue method for determination of pesticides in Passiflora alatainfuses using headspace solid phase microextraction and gas chromatography, J.Chromatogr. A 1138 (2007) 251–261.

[24] C.H. Kowalski, G.A. da Silva, R.J. Poppi, H.T. Godoy, F. Augusto, Neuro-geneticmultioptimization of the determination of polychlorinated biphenyl congeners in

humanmilk by headspace solid phase microextraction coupled to gas chromatogra-phy with electron capture detection, Anal. Chim. Acta. 585 (2007) 66–75.

[25] K.R. Beebe, R.J. Pell, M.B. Seasholtz, Chemometrics: A Practical Guide, 1st ed. JohnWiley & Sons Inc., New York, 1998.

[26] T. Kohonen, Self-organizing Maps, 3rd ed. Springer, Berlin Heidelberg, 2001.[27] R.F. Teófilo, M.M.C. Ferreira, Spreadsheet calculations for experimental sched-

ules, available in: http://lqta.iqm.unicamp.br/portugues/Downloads.html#teofilo(Accessed: 3 December 2012).

[28] E. Alhoniemi, J. Himberg, J. Parhankangas, J. Vesanto, Laboratory of Computer andInformation Science, Finland, 2000.

[29] G.A. da Silva, F. Augusto, R.J. Poppi, Exploratory analysis of the volatile profile ofbeers by HS-SPME-GC, Food Chem. 111 (2008) 1057–1063.