Spectrophotometric Determination of Caramel Content in Spirits Aged in Oak Casks

Transcription

1 744 BOSCOLO ET AL.: JOURNAL OF AOAC INTERNATIONAL VOL. 85, NO. 3, 2002 TECHNICAL COMMUNICATIONS Spectrophotometric Determintion of Crmel Content in Spirits Aged in Ok Csks MAURICIO BOSCOLO,LUIZ G. ANDRADE-SOBRINHO,BENEDITO S. LIMA-NETO, nd DOUGLAS W. FRANCO 1 Universidde de São Pulo, Instituto de Químic de São Crlos, São Crlos, São Pulo, Brzil, CP 780 MARCIA MIGUEL CASTRO FERREIRA Universidde Estdul de Cmpins, Instituto de Químic, Cmpins, São Pulo, Brzil A new methodology ws developed for determintion of crmel in spirits ged in ok csks. The method is bsed on differences between the electronic spectr of ok queous lcoholic extrcts nd crmel solutions in the sme solvent. The dt were treted by 2 different pproches: the simplest one ws bsed on the plot of crmel concentrtion versus the rtio of bsorbnce t 210 nd 282 nm; the other ws bsed on prtil lest squres (PLS) clibrtion model using the first derivtive of the spectrl dt. Both methodologies were pplied to nlysis of 159 ged spirit smples. The men crmel content of severl Brzilin sugr cne spirits (cchç) nd ll United Sttes whiskies ws smller thn tht of Scottish whiskies nd other brndies from severl countries. Correltion ws good between crmel concentrtions for the sme smple clculted by the 2 methods. The uncertinties following PLS nd the bsorbnce rtio method were 0.01 nd 0.03 g/l, respectively, for smple contining 0.45 g/l crmel. Tretment of UV-VIS spectr by pttern recognition using hierrchicl clustering nlysis nd principl components nlysis llowed discrimintion of the smples s function of their crmel content. It ws possible to distinguish U.S. whiskies from other whiskies, but cler differentition mong Brzilin cchçs s function of their geogrphic origin ws not fesible. Smll crmel quntities s low s 0.08 g/l were clerly detected by these methodologies. Spirits stored in ok csks develop n mber or golden-brown color, which is ttributed to the extrction of phenolic compounds from the ok csks. In mny countries, except the United Sttes (1), ok csks re reused more thn once for mturtion; in some cses, with beverges like sherry, the color intensity of the finl product decreses Received Februry 7, Accepted by JL August 22, Author to whom correspondence should be ddressed; e-mil: dougls@iqsc.sc.usp.br. during the sme period of mturtion. The golden-brown color cn be mimicked by ddition of suitble mount of crmel or pxrette (2), brown syrup, into freshly distilled spirit. However, crmel is minly used by distillers to ensure color consistency of their ged products (3). Indeed, crmel is fctor which decreses the voltile sulfur compound concentrtion in spirits (4). Thus, relible crmel determintion is importnt not only to scertin the beverge crmel content, but lso to detect dultertion of spirits nd ok extrcts. Considering the chemicl composition complexity of crmel (5, 6), its quntittion is not strightforwrd in food product nlysis. A quntittive nlysis of crmel in spirits cn be performed by selective extrction using Mrsh regent s indicted in the AOAC officil method (7). However, this methodology (4) is subject to interference by synthetic dyes nd substnces liberted from nonchrred ok (7). Bsed on the fct tht furfurl (F) is extrcted from ok stves (8) nd 5-hydroxymethylfurfurl (HMF) is the min compound of crmel (9), liquid chromtogrphic (LC) method hs been proposed (10, 11) to detect crmel in spirits bsed on the F/HMF concentrtion rtio. According to this method, this rtio is >1.0 for smples without crmel nd <1.0 for smples contining crmel. However, this pproch cnnot be pplied to the Brzilin sugr cne spirit (cchç), becuse even in nonged crmel-free cchçs, the F/HMF concentrtion rtio is <1.0 for mny smples (12). The cchç production is round 2 billion liters per yer (13), nd there is growing interest in improving product qulity through consistent mturtion procedures nd the estblishment of up-to-dte chemicl qulity control. During our investigtion of the crmel content in ged cchçs, spectrl differences were noted between queous ethnol extrct of ok nd crmel solutions in the sme solvent. Thus, bsed on spectrophotometric mesurements, 2 methodologies were developed to quntitte crmel in spirits ged in ok csks. The first uses the rtio between bsorbnce vlues for 2 different wvelengths (210 nd 282 nm), nd the second is bsed on multivrite clibrtion through prtil lest squres (PLS) using the first derivtive spectr. Pttern recognition by hierrchicl clustering nlysis (HCA) nd principl component nlysis (PCA) ws performed to discriminte the smples s function of their crmel content.

2 BOSCOLO ET AL.: JOURNAL OF AOAC INTERNATIONAL VOL. 85, NO. 3, Tble 1. Recovery of crmel from 5 ged cchçs spiked with crmel using the A 210 /A 282 rtio Smples Added crmel, g/l Averge ± ± ± ± ± ± ± ± 0.10 Figure 1. Electronic spectr of dditive-free sucrose crmel, cchç ged in n ok csk for 18 months, nd mixture of both (smples diluted 10-fold with ethnol wter, 1+1,v/v). Experimentl Smples () Aqueous ethnol solution. A mixture of wter nd ethnol (1 + 1, v/v) ws used to prepre ok extrcts nd dilute smples. (b) Free crmel spirits. Smples of 16 Brzilin crmel-free cchçs were ged in 180 L ok csks to estblish spectrl chrcteristics of the ok extrct. These smples were supplied nd certified by Indústris Müller de Bebids Ltd. (Pirssunung, São Pulo, Brzil)., Missing dt. (c) Commercil smples. A totl of 159 smples were nlyzed: Brzilin cchçs (São Frncisco, Herr Blumenu, Jequity, Milgre de Mins, Vt 45, Prti, Pul, Ypioc, Box 32, Neg Fulo, Muller, Bocin, Ctedrl, Colonil, Morretes, Spupr, Bosco 12, Bosco 24, Bosco 36, Bosco 48, Cpitão ds Geris, Velh Aroeir, Bel Vist, Terr Brsilis, Beij-Flor, Cju, Cn Verde, Dndiz, Kolonile, Slins, Crvlheir, Mngueir, Águs Quentes, Chpéu de Plh, Cbeceir, Abír, Córrego Azul, Rinh, Espírito de Mins, Ancon, Moend de Ouro, Germn, Armzém Vieir, Almbri, Gurmirng, Coqueiro, Berro, Moribondo, Três Coronéis, Vle Verde, Serrote, Rinh d Lvour, Bodocó, Ping de Plmitl, Curupá, Ginete, Bozinh, Setembrin, Tiqur); Scottish whiskies (Crdhu, Chivs Regl, Grnt s, Lphroig, Tiller s, Glenfiddich, Logn, Bllntine s, St. Jmes, Bell s, Johnny Wlker Green Lbel, Johnny Wlker Red Lbel, Pssport, Blck White, Glendro, 100 Pipers, Queen Anne, Robbie Dhu, White Horse, Dlmore, White & Mcky, Glenrothes, Mclln, Blvenie); U.S. whiskies (Old Grnd Dd, Jck Dniel s, Jim Bem, Tble 2. Experimentl nd predicted crmel concentrtions in g/l for PLS model, stndrd error of prediction (SEP), nd correltion coefficient (r) Stndrds Known Predicted Residul Figure 2. Electronic spectr of queous ethnol solutions of chrred nd nonchrred ok extrct. ST ST ST ST ST ST SEP 0.01 R One ltent vrible.

3 746 BOSCOLO ET AL.: JOURNAL OF AOAC INTERNATIONAL VOL. 85, NO. 3, 2002 Tble 3. Comprison of results of crmel obtined from 2 methods (prtil lest squres nd A 210 /A 282 rtio),b Smple PLS A 210 /A 282 Smple PLS A 210 /A 282 Smple PLS A 210 /A 282 Smple PLS A 210 /A 282 BC BC BC BC BC1b BC BC BC BC1c BC13b BC BC BC1d BC BC BC BC1e BC14b BC BC BC BC14c BC33b BC BC2b BC BC33c BC BC2c BC BC BC BC BC16b BC34b BC BC3b BC16c BC BC BC BC16d BC BC BC4b BC BC36b BC BC4c BC BC BC BC BC BC BC BC BC BC BC BC6b BC BC39b BC BC BC21b BC BC BC BC BC40b BC BC8b BC22b BC BC BC8c BC22c BC BC BC8d BC BC BC BC BC23b BC43b BC BC9b BC23c BC BC BC9c BC BC BC BC BC BC45b BC BC10b BC BC45c BC10c BC BC BC BC BC b Letters re different smples of the sme product. Only crmel concentrtions >0.07 g/l were considered s relible dt. The order of smples is not the sme s tht in the text. BC = Brzilin cchç. Four Roses, Erly Times, Evn Willims, Mker s Mrk); Cndin whisky (Cndin Club); Irish whiskies (Jmison, Tullmore Dew). More thn one smple of the sme product ws nlyzed in some cses. The number of cchçs smples by region ws proportionl to region prticiption in Brzilin cchç production. (d) Ok wood smples. Powdered ok woods of Scottish origin (Segrms), both chrred nd nonchrred, were used to prepre queous ethnol extrcts nd to record bsorption spectr. In typicl experiment, 1.0 g ok powder ws dded to ml queous ethnol solution t 60.0 C, nd stirred for 45 min. The mixture ws then filtered, nd the filtrte ws diluted to the desired volume with the queous lcoholic solution. As self-consistency test, 15 queous lcoholic extrcts of nonchrred ok from the United Sttes, Polnd, Spin, Czechoslovki, Frnce, nd Scotlnd were lso used for stndrd recognition nlysis. (e) Crmels. Crmel solutions were prepred with 7 commercilly vilble sugrs (sucrose). In typicl preprtion, 20.0 g sugr ws heted bove the melting point for 10 min to obtin drk-colored pste. The crmel (10.0 g) ws trnsferred to ml volumetric flsk contining the queous lcoholic solution, nd the volume ws djusted to ml. The bsorptivity ws clculted for ech smple from plot of the crmel concentrtion vs the respective bsorbnce. Spectrophotometric Mesurements The smples were diluted with queous ethnol solution prior to nlysis: for thecchçs; forother beverges. To clculte the smple crmel concentrtion, the bsorbnce

4 BOSCOLO ET AL.: JOURNAL OF AOAC INTERNATIONAL VOL. 85, NO. 3, Tble 4. Comprison between results of crmel obtined from 2 methods (prtil lest squres nd A 210 /A 282 rtio) Smple b PLS A 210 /A 282 Smple b PLS A 210 /A 282 USW SW16b USW SW USW SW17b USW SW17c USW SW USW SW USW SW19b SW SW SW SW SW SW21b SW SW SW SW SW SW SW SW SW SW25b SW SW SW SW SW SW27b SW CW SW CW SW CW2b SW IW SW15b IW1b SW15c IW SW IW2b b Letters re different smples of the sme product. Only crmel concentrtions >0.07 g/l were considered s relible dt. The order of the smples is not the sme s tht in the text. Abbrevitions: USW, U.S. whisky; SW, Scottish whisky; CW, Cndin whisky; IW, Irish whisky. 2-dimensionl spce. The results, qulittive in nture, usully re presented in form of dendogrms, mking it possible to visulize similrities mong smples or vribles. In HCA, the distnces between smples or vribles re clculted, trnsformed into similrity mtrix, S, nd then compred. For ny 2 smples, k nd l, the similrity index is defined s S kl = d d where S kl is n element of S, d mx is the lrgest distnce mong ech pir of smples in the dt, nd d kl is the Eucliden distnce mong the smples k nd l. The PLS regression ws used for crmel quntittion (15) in the first derivtive of spectrl dt. The dvntge of using multivrite method is tht n estimte of more thn one regression coefficient is mde possible for the property being modeled. The PLS method clcultes new set of vribles, clled ltent vribles, tht re generted s liner combintion of the originls nd then used s predictors of concentrtion. The ltent vribles re computed, tking into ccount both the spectrl dt nd the respective crmel content, nd mking use of more informtion t the modeling step. The predictbility of the resulting model cn be ssessed by the leve-one-out cross-vlidtion procedure given by the lowest stndrd error of prediction (SEP), defined s SEP = n i= 1 kl mx ( c c ) where c is the experimentl vlue, is the predicted vlue, nd n is the number of smples used for model building. Once the model ws vlidted, rel smples were tested for their crmel content. The dt were nlyzed with Pirouette sttisticl softwre by Infometrix (Settle, WA). i n i 2 vlues were corrected considering the corresponding dilution fctor. A Hitchi (Sn Jose, CA) U-3501 spectrophotometer, equipped with 1.00 cm long qurtz cell, ws used for spectrl dt cquisition. The spectr were recorded from 210 to 400 nm. Dt Anlysis PCA (14) is widely used to simplify lrge dt sets so tht ptterns nd reltionships cn be redily recognized nd understood. The underlying purpose of the technique is dimension reduction. The smples re mpped through scores nd the wvelengths re mpped by the lodings in new low dimensionl vector spce defined by the principl components. HCA is lso n explortory tool used to confirm the grouping previously identified by PCA. Its primry purpose is to disply the dt so s to emphsize its nturl clusters nd ptterns in Figure 3. Correltion between the A 210 /A 282 rtio nd PLS clibrtion methodologies to determine crmel content in ged spirits.

5 748 BOSCOLO ET AL.: JOURNAL OF AOAC INTERNATIONAL VOL. 85, NO. 3, nd 282 nm (A 210 /A 282 ) ws 4.2 ± 0.5. Chrred nd nonchrred oks of the sme origin do not show ny significnt bsorbnce rtio differences t 210 nd 282 nm from wht occurs in nother region of the spectrum (Figure 2). The ddition of crmel to n queous lcoholic ok extrct leds to some modifiction in the bsorption spectrum (Figure 1), resulting in chnge in the A 210 /A 282 rtio. This feture ws used to clculte the crmel content in ged spirits. The contribution of substnces extrcted from ok csk to the bsorbnce t 282 nm cn be estimted by dividing the bsorbnce t 210 nm by 4.2. Thus, the crmel content cn be estimted by the difference between the experimentlly determined bsorbnce t 282 nm nd tht clculted by the following eqution: CC=DF[A 282 (A 210 /4.2)]/CA 282 Figure 4. PCA scores plot of queous ethnol ok nd 1.7 g/l) nd 109 commercil cchçs from different Brzilin regions: (NE) northest, (SE) southest, (CW) centrl west, nd (S) south. Dt processing: mencentered. Results nd Discussion Nonged, crmel-free cchçs do not present ny significnt bsorption in the UV-VIS region of the spectrum; this is in contrst to queous lcoholic crmel solutions, queous lcoholic ok extrcts, nd mixture of both, s shown in Figure 1. The HMF content of sucrose crmeliztion process ws followed by LC (12) nd, t the sme time, the UV-VIS spectrum of the smple ws recorded. These experimentl dt indicte tht HMF is responsible for t lest 60% of the bsorbnce vlues t 282 nm, which corresponds to the mx for HMF. The sucrose bsorptivity t 282 nm, clculted s n verge vlue of 7 different smples, is 8.0 ± 0.9 L/g cm. As mtter of fct, the ok wood contins furnic compounds such s furfurl, 5H-furnone, 2-furyl-1-propnon, 2-furoic cid, methyl furote, nd HMF (16), mong other clsses of compounds, which contribute to the typicl spectrl chrcteristic of ok extrct solution. The bnd t 282 nm is bsent in crmels with dditives such s sulfites nd mmonium slts becuse HMF is bsent in crmels contining such dditives (6). Therefore, crmels contining sulfites nd mmonium in ged spirits cnnot be detected by LC bsed on the F/HMF rtio or spectrophotometriclly, s they will not show ny significnt spectrum difference from the ok extrct. Thus, if the AOAC method (7) detects the presence of crmel tht ws not detected by spectrophotometric mesurements, crmel with sulfite or mmonium ws likely used to drken the ged spirit. A detiled nlysis of the spectrl fetures of 16 crmel-free cchçs ged in ok csks nd of queous ethnol ok extrcts showed tht the rtio of bsorbnce vlues t where CC is the crmel content, DF is the dilution fctor (10 for cchçs nd 20 for other beverges), A 282 is the bsorbnce t 282 nm, A 210 is the bsorbnce t 210 nm, nd CA 282 is the crmel bsorptivity t 282 nm (8.0 L/g cm). The ccurcy of the method ws evluted by nlyzing smples prepred by dding different mounts of crmel to 5 different cchçs ged in ok csks (Tble 1). These results show tht the uncertinty decreses when the crmel concentrtion increses, which is probbly due to better definition of the bsorption bnd t 282 nm. Consequently, when crmel concentrtions fll to <0.2 g/l, the method is indicted for Figure 5. HCA dendogrm plot of queous ethnol ok nd 1.7 g/l) nd 109 commercil cchçs using normlized spectrl bsorbnce s vribles. Similrity mesurement: Eucliden distnce; dt processing: men-centered; clustering technique: incrementl.

6 BOSCOLO ET AL.: JOURNAL OF AOAC INTERNATIONAL VOL. 85, NO. 3, Figure 6. PCA scores plot of queous ethnol ok nd 1.7 g/l) nd 50 commercil whiskies from United Sttes (USW), Irelnd (IW), Cnd (CW), nd Scotlnd (SW). semiquntittive or qulittive purposes only. However, in the concentrtion rnge of g/l, the uncertinty ws <10%. Due to the complexity of the crmel chemicl composition, we recommend tht clibrtion plot be prepred using crmel stndrds from the sme supplier of the smple whenever possible. A set of 6 dditive-free crmel solutions with known concentrtion ws used to build the PLS clibrtion model. Tble 2 shows the experimentl nd estimted crmel concentrtions obtined by leve-one-out cross-vlidtion. Experimentlly, through successive crmel dditions to n ok extrct solution, it is possible to verify tht concentrtions >0.08 g/l cn be detected by this methodology. The A 210 /A 282 rtio nd the PLS clibrtion model were used to predict the crmel content in 109 cchçs smples nd 50 imported spirits (Tbles 3 nd 4). The negtive vlues indicte the totl bsence of crmel, nd probbly reflect smll vritions in ok wood composition nd, thus, in the spectr. The verges for the crmel content in cchçs by PLS nd A 210 /A 282 rtio were ± nd ± 0.204, respectively. The higher vlue for the sme smple ws bout 2.65 g/l (PLS) nd 2.73 g/l (A 210 /A 282 ). Only 12% of the smples hd men vlue for both methodologies >0.20 g/l. This result cn be explined by the fct tht trditionlly ged cchçs generlly hve light color nd do not require crmel to drken them. The crmel concentrtion dt for U.S. whiskies (5 Bourbons nd 1 Tennessee whiskey) re below the detection limit of the proposed method. Only in one Bourbon (smple USW6) ws the clculted crmel content higher thn the detection limit. This ws expected, becuse crmel ddition is forbidden for stright whiskies. The verges of crmel content in whiskies clculted by PLS nd A 210 /A 282 rtio were ± nd ± 0.305, respectively. The higher vlues for the sme smple were bout 0.97 g/l (PLS) nd 1.10 g/l (A 210 /A 282 ). The PLS clibrtion method gve men vlues smller thn those obtined with A 210 /A 282 methodology. For whiskies with higher vlues, these differences were round 100% s function of unknown fctors. About 74% of the whisky smples showed men vlue >0.20 g/l for both methodologies. Figure 3 shows good correltion between vlues obtined by the 2 methods for the sme smples (y = 1.11x ; R 2 = 0.87). From the explortory nlysis by PCA nd HCA, we clssified 159 smples of spirits into groups s function of spectrl similrities. The PCA results for Brzilin cchçs (Figure 4) indicted tht the 2 first principl components explin 87% of the totl vrince in the dt. The wvelengths in the region of 282 nm re dominnt in PC1, wheres the wvelengths of 240 nm re dominnt for PC2. As crmel ddition incresed in the stndrds of ok queous ethnol solutions, the PC1 scores becme more positive. Thus, smples with high positive scores in PC1 were expected to hve high crmel concentrtions, wheres those with negtive scores probbly hd no crmel dded. The stndrds nd 15 smples of pure ok extrcts re included in dt nlysis for comprison. Note tht these ok smples re seprted, mking distinct group with negtive scores in PC1 nd positive scores in PC2. Figure 7. HCA dendogrm plot of queous ethnol ok nd 1.7 g/l) nd 50 commercil whiskies from United Sttes (USW), Irelnd (IW), Cnd (CW), nd Scotlnd (SW), using normlized spectrl bsorbnce s vribles. Similrity mesurement: Eucliden distnce; dt preprocessing: men-centered; clustering technique: incrementl.

7 750 BOSCOLO ET AL.: JOURNAL OF AOAC INTERNATIONAL VOL. 85, NO. 3, 2002 It ws not possible to distinguish the cchçs through their geogrphic regions bsed on crmel content. This fct suggests no regionl prctices on crmel use to color the cchçs. However, HCA nlysis (Figure 5) shows similrity index of 0.32, indicting tht Brzilin cchçs re seprted into 2 min groups. The first (group A) is more correlted to ok extrcts, wheres the second (group B) is less correlted to ok extrcts nd more to sucrose crmel dded. The smples with negtive PC1 scores in Figure 4 re included in group A, indicting tht both stndrd recognition methodologies re suitble for discriminting the smples s function of the crmel dded to the beverge nd to their extrct chrcteristics. Figure 6 shows the results of explortory nlysis for whiskies. The first 2 principl components cn explin 90% of the totl vrince in the dt. All U.S. whiskies nlyzed were close to the ok stndrds, which do not hve crmel dded (negtive scores in PC1). In Brzil nd in other countries, distillers reuse csks lredy used for mturtion of other beverges such s sherry nd U.S. whiskies (14), thereby decresing the mount of vilble ok extrcts to color the beverges. Of the 50 whisky smples nlyzed, ll 7 U.S. whiskies nd 6 Scottish whiskies hd negtive PC1 scores, indicting the bsence of crmel use. The 15 smples of ok extrct were lso included in the nlysis s comprison. Agin, they pper in seprte group hving negtive scores in PC1, which mens bsence of crmel. For HCA nlysis (Figure 7), s with the cchç smples, the whiskies were seprted into 2 clusters with similrity index of Those with negtive PC1 scores re in the sme group s the ok stndrds (group A), suggesting the bsence or low concentrtion of crmel. Group B contins smples with positive PC1 scores (Figure 6) nd the highest crmel contents. Conclusions The proposed methodologies cn be used for forensic purposes. The results hve demonstrted tht it is possible to identify nd quntitte sucrose crmel in spirits ged in ok csks by UV-VIS spectroscopy. Combining this methodology with the AOAC method for crmel determintion, it thus becomes possible to discriminte between the ddition of sucrose crmel nd strch crmel prepred with sulfite dditives. Mny renowned distilltes were found to hve reltively high crmel content, suggesting tht the use of crmel is more frequent thn expected. Crmel itself does not disqulify the products, if locl lws llow it. However, the customer should be wre of the use of crmel through n indiction on the bottle lbel. Acknowledgments We thnk CAPES nd FAPESP for their finncil support, Indústris Müller de Bebids Ltd. for smples supply, nd Ludmil A. Rmos (Universidde de São Pulo) for helping in erlier experiments. We re indebted to Leif H. Skibsted (University of Copenhgen, Denmrk), Denis De Keukeleire (University of Ghent, Belgium), nd Giuseppe Versini (Istituto Agrrio Sn Michele ll Adige, Itly) for reding nd commenting on this mnuscript. References (1) Mosedle, J.R., & Puech, J.L. (1998) Trends Food Sci. Technol. 9, (2) Crwford, I.M. (1985) Colour of spirits, B. Sc. (Hons.) Thesis in Food Science, University of Strthclyde, Deprtment of Bioscience & Biotechnology, Glsgow, UK (3) Booth, M., Shw, W., & Morhlho, L. (1989) in The Science nd Technology of Whiskies, J.R. Piggott, R. Shrp, & R.E.B. Duncn (Eds), Longmn Scientific & Technicl, Hrlow, Essex, UK, pp (4) Nedjm, M. (1997) Am. J. Enol. Vitic. 48, (5) Pons, I., Grrult, C., Jubert, J.N., Morel, J., & Fenyo, J.C. (1991) Food Chem. 39, (6) Dross, A., Hrdt, R., & Bltes, W. (1987) Fresenius Z. Anl. Chem. 328, (7) Officil Methods of Anlysis (1995) AOAC INTERNA- TIONAL, Githersburg, MD, Method (8) Conner, J.M., Pterson, A., & Piggott, J.R. (1993) J. Sci. Food Agric. 62, (9) Mir, M.V., Grndos, J.Q., de l Serrn, H.L.G., & Mrtinez, M.C.L. (1992) J. Liq. Chromtogr. 15, (10) Grndos, J.Q., Mir, M.V., de l Serrn, H.L.G., & Mrtinez, M.C.L. (1996) Food Chem. 56, (11) Jgnthn, J., & Dugr, S.M. (1999) J. AOAC Int. 82, (12) Nscimento, R.F., Mrques, J.C., Lim-Neto, B.S., & Frnco, D.W. (1997) J. Chromtogr. 782, (13) Lim-Neto, B.S., & Frnco, D.W. (1994) Engrrfdor Moderno 33, 5 8 (14) Beebe, K.R., Pell, R.J., & Sesholtz, M.B. (1998) Chemometrics: A Prcticl Guide, John Wiley & Sons, New York, NY (15) Mrtens, H., & Nes, T. (1989) Multivrite Clibrtion, John Wiley & Sons, New York, NY (16) Pérez-Coello, M.S., Snz, J., & Cbezudo, M.D. (1998) Chromtogrphi 47,