Microchemical Journal 103 (2012) 49–61

Contents lists available at SciVerse ScienceDirect

Microchemical Journal

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

Analysis of vanilla extract by reversed phase liquid chromatography using water richmobile phases

Barry K. Lavine ⁎, Desire T. Corona, Undugodage Don Nuwan Tharanga PereraDepartment of Chemistry, Oklahoma State University, Stillwater, OK 74078-3071, United States

⁎ Corresponding author.E-mail address: bklab@chem.okstate.edu. (B.K. Lavin

0026-265X/$ – see front matter © 2012 Elsevier B.V. Aldoi:10.1016/j.microc.2012.01.004

a b s t r a c t

a r t i c l e i n f o

Article history:Received 5 January 2012Accepted 11 January 2012Available online 17 January 2012

Keywords:Water rich mobile phasesVanilla extractVanillinLC–MSRPLC

Vanillin is themajor constituent of vanilla extract, a flavoring ingredient used in food and beverages. Natural vanillaextract prepared from the bean of the tropical orchid, Vanilla planifolia, is expensive due to the limited supply of thevanilla bean. For this reason, synthetic vanilla extracts are widely used. Synthetic vanilla extracts are less complexandusually contain vanillin, ethyl vanillin, andother related compounds that are prepared from inexpensive startingmaterials. Several liquid chromatographicmethods have been developed to quantitate coumarin, vanillin, and ethylvanillin in vanilla extract. The use ofwater richmobile phases in reversed phase liquid chromatography (RPLC), e.g.,1% butanol inwaterwith 0.2% acetic acidwith C18, C8, and cyanopropyl columns, has been investigated as a potentialmethod to characterize the composition of synthetic vanilla extracts. Better resolution is achieved in the separationof vanillin compounds when hydrophobic alcohols are used as organic modifiers. This can be attributed to butanolpartitioning into the bonded phase, which provides a more extended ordered surface increasing the contact sur-face area of the stationary phase and thereby increasing the selectivity of the separation. Usingwater richmobilephases, constituents of vanilla extract in 36 commercial products obtained from stores in the local area wereidentified demonstrating the efficacy of the proposed RPLC method.

© 2012 Elsevier B.V. All rights reserved.

1. Introduction

Vanillin (4-hydroxy-3-methoxybenzaldehyde) is themajor constitu-ent of vanilla extract, a flavoring ingredient used in food and beverages.Natural vanilla extract prepared from the bean of the tropical orchid,Vanilla planifolia, is expensive due to the limited supply of the vanillabean. For this reason, synthetic vanilla extracts are widely used. Syn-thetic vanilla extracts are less complex and usually contain vanillin,ethyl vanillin and other related compounds that are prepared from in-expensive starting materials.

Several reversed phase liquid chromatographic (RPLC) methodshave been developed to quantitate coumarin, vanillin, and ethyl van-illin in vanilla products [1–4]. These RPLC methods, which employedisocratic or gradient elution, methanol, acetonitrile or tetrahydrofu-ran as the organic mobile phase modifier and diode array or massspectrometer detection, are not able to separate weakly retainedcompounds such as vanillic acid and isovanillin. The use of waterrich mobile phases has been investigated in this study as a potentialisocratic method to fully characterize the composition of synthetic va-nilla extract. Water rich mobile phases in RPLC are not generally usedbecause of the long retention times involved. The problem of long

e).

l rights reserved.

retention times can be addressed using hydrophobic alcohols suchas butanol in low quantities (approximately 1% v/v) as the organicmodifier.

In an effort to gain insight into the relationship between stationaryphase solvation and selectivity and to improve the chromatographic

Fig. 1. Chemical structures of the compounds comprising the vanillin test mixture.

Fig. 2. Chromatograms of the vanillin testmixture. Elution orderwas vanillic acid, isovanillin,vanillin, o-vanillin, ethyl vanillin and coumarin. A) Methanol–water mobile phase (25%methanol in water with 0.2% acetic acid) that yielded the best separation on the BDSC18 column at ambient temperature. B) Butanol–water mobile phase (1.0% butanol inwater with 0.2% acetic acid) that yielded the best separation of the test mixture on theBDS C18 column at ambient temperature.

Fig. 3. Plot of ln k′ versus ϕ for each vanillin test mixture compound

50 B.K. Lavine et al. / Microchemical Journal 103 (2012) 49–61

separation of water soluble compounds, Lavine [5] in a previous studyhad investigated the use of short and medium chained length alcohols(methanol, n-propanol, n-butanol, and n-pentanol) as mobile phasemodifiers in RPLC to determine their impact on resolution. A widerange of mobile phase compositions was evaluated because of thelarge effect exerted by solvent strength on selectivity. Evidence waspresented to support the view that an increase in the hydrophobicityof the organic modifier increases the selectivity of the C18 alkyl bondedphase by increasing its contact surface area due to improved wetting,which is a result of the greater hydrophobicity of the organic mobilephase modifier used.

Using aqueous rich mobile phases (e.g., 1% butanol in wateracidified with 0.2% acetic acid), an RPLC and liquid chromatography/mass spectrometry (LC/MS) method capable of separating coumarin,vanillin, and derivatives of vanillin (e.g., vanillic acid, isovanillin,o-vanillin, ethyl vanillin) on an alkyl bonded C18 stationary phasecolumn has been developed. Snyder's solvent strength model [6,7]was used to provide a better understanding of the factors that influencethe separation process and to offer insight into the retention mecha-nism when butanol is used as the organic mobile phase modifier onC18 and C8 alkyl bonded phases. Because of the better resolutionachieved in the separation of the vanillin compounds when butanolwas used as the organic mobile phase modifier, constituents of vanillaextract purchased from local supermarkets in the area were identifiedthat hitherto had not been previously reported in the literature. Fur-thermore, vanilla extracts purchased from local supermarkets couldbe divided into distinct groups based on their composition asreflected by their RPLC profiles.

2. Experimental

The six retention probes used in this study, vanillic acid, isovanillin,o-vanillin, ethyl vanillin, vanillin and coumarin were purchased fromAldrich and were used as received. The chemical structures of thesesix compounds are shown in Fig. 1. Stock solutions of the vanillin com-pounds were prepared by weighing and dissolving the correspondingamount of the compound in methanol (Thermo-Fisher) followed by di-lution to the appropriate working concentration (10−4 M) using dou-bly distilled water (prepared by a Barnstead NanoPure II System,Barnstead International, Dubuque, IA). If stored at 4°C, the working so-lutions were stable for approximately one month. Since these com-pounds are weakly retained by alkyl bonded stationary phases, it wasnecessary to use water as the primary solvent to prepare the vanillintest mixtures. If a stronger solvent such as methanol were used to pre-pare the samples, the testmixtureswould not have been deposited onto

using as mobile phases methanol in water with 0.2% acetic acid.

Fig. 4. Plot of ln k′ versus ϕ for each vanillin test mixture compound using as mobile phases butanol in water with 0.2% acetic acid.

51B.K. Lavine et al. / Microchemical Journal 103 (2012) 49–61

the head of the column as a thin plug during sample injection, with theresult being increased band broadening.

Thirty-six vanillin extract products were obtained from local su-permarkets in the area. Sample preparation consisted of pipetting250 μl of vanillin extract into a 25 ml volumetric flask followed by

Fig. 5. Chromatograms of the vanillin test mixture. Elution order was vanillic acid, isovanil(30% methanol in water with 0.2% acetic acid) that yielded the best separation on the BDSin water with 0.2% acetic acid) that yielded the best separation on the BDS cyanopropyl bondein water with 0.2% acetic acid) that yielded the best separation on the BDS C8 column at ambiacid) that yielded the best separation on the BDS cyanopropyl column at ambient temperature

the addition of 25% methanol in water solution with 1% v/v/aceticacid to volume and sonication of the flask. Methanol and butanol,the alcohols used as organic modifiers in this study, were purchasedfrom Thermo-Fisher and Aldrich. Glacial acetic acid, which was usedto acidify the mobile phase, was obtained from Pharmaco. Doubly

lin, vanillin, o-vanillin, coumarin, and ethyl vanillin. A) Methanol–water mobile phaseC8 column at ambient temperature. B) Methanol–water mobile phase (2.5% methanold phase column at ambient temperature. C) Butanol–water mobile phase (0.75% butanolent temperature. D) Butanol water mobile phase (2.5% butanol in water with 0.2% acetic.

52 B.K. Lavine et al. / Microchemical Journal 103 (2012) 49–61

distilled water (purchased from Burdick & Jackson) was used to pre-pare all mobile phases, which was then filtered with 0.45 μm poresize filters (Varian). Varian Nylon 66 filters were used to remove par-ticulate matter from the mobile phase. A transfer pipette was used toprepare mobile phases containing butanol because of the small vol-ume of organic modifier used in the preparation of these solutions.Each mobile phase solution was degassed prior to use.

The chromatographic studies were performed on two LC instru-ments: (1) Varian High Performance Liquid Chromatograph equippedwith a BDS-Hypersil C18, C8, or Cyanopropyl column (100× 4.6 mm,Particle Diameter 5 μm, Thermo-Fisher), Shimadzu column oven,ProStar reciprocating pump, and diode array detector, and (2) SHI-MADZU 2010 LC–MS equipped with a 2010 Premier C18 column (100×4.6 mm, Particle Diameter is 3 μm, Shimadzu), columnoven, and twode-tectors (diode array and electro-spray mass spectrometer). The deadtime of each column was determined by injecting different solutions(methanol, methanol–water, water, or KNO3) onto the BDS or Premierecolumn, and the dead volume (1.1 ml for the Premiere column and1.2 ml for the BDS column) was used in all retention factor calculations.All k′ values determined in this studywere averages of triplicate determi-nations, and deviations in individual k′ valueswere never greater than 1%.

All mobile phases were percolated through the column at a flowrate of 1.0 ml/min (Varian Prostar HPLC) or 0.3 ml/min (Shimadzu

Fig. 6. Log–log plots for the vanillin test mixture compounds using

LC/MS) for several hours to ensure reproducible solvation of thebonded phase. These flow rates were used in all of the studies becauseof the desire to develop an RPLC or LC–MS analysis method for theanalysis of vanillin extract. For the same reason, the temperatureof the column was ambient. The injection volume was 5 μl. For theLC/MS runs, the ESI interface was operated in the negative ionmode, the nebulizing nitrogen gas flow rate for the LC–MS interfacewas set at 1.5 l/min, and the temperature of the curved desolvationline was set at 200 °C. Full-scan spectra were recorded from 50 to500 m/z at a scan time of 500 amu/s.

3. Results and discussion

3.1. Methanol versus butanol as the mobile phase modifier

A series of chromatograms were run to illustrate the advantages ofusing butanol as an organic modifier for the separation of vanillin,derivatives of vanillin and coumarin. The test mixture consisted ofsix compounds (see Fig. 1). Fig. 2A shows a chromatogram of thevanillin test mixture using the methanol–water mobile phase (25%methanol in water with 0.2% acetic acid) that yielded the best sep-aration on the BDS C18 column at ambient temperature. Fig. 2Bshows a chromatogram of the same test mixture on a butanol water

methanol in water with 0.2% acetic acid as the mobile phase.

53B.K. Lavine et al. / Microchemical Journal 103 (2012) 49–61

mobile phase (1.0% butanol in water with 0.2% acetic acid) thatyielded the best separation on the BDS C18 column at ambient temper-ature. The six components of the vanillin testmixture can be separatedby the butanol–water mobile phase but not by the methanol–watermobile phase.

To better understand the factors that contribute to better separationswhen butanol is used as the organic mobile phase modifier, it was nec-essary to examine retention factor data for each test mixture compoundon the BDS C18 column in a systematic manner. In RPLC, the factorswhich influence the separation process can be understood by a thoroughanalysis of k′ data obtained for a set of compounds using Snyder's sol-vent strength plots (see Eq. (1)), where ln kw is the logarithm of the re-tention factor for the compound in a purely aqueous medium, B is theslope of the plot, which is dependent on the solute and the chromato-graphic system investigated (mobile and stationary phase), and Φ isthe volume percentage of organicmodifier in themobile phase. Snyder'ssolvent strength plots can also be used to predict selectivity and resolu-tion for a particular separation over a narrow range of Φ [8].

1nk 0 ¼ 1nkw−BΦ ð1Þ

Fig. 3 shows a plot of ln k′ versus ϕ for each compound of thevanillin test mixture eluted from the column using as mobile phases

Fig. 7. Log–log plots for the vanillin test mixture compounds usin

methanol in water with 0.2% acetic acid. Snyder solvent strengthplots were generated using five methanol–water–acetic acid mobilephases: 15%, 20%, 25%, 30%, and 35% methanol in water with 0.2%acetic acid. (For this study, it was not possible to generate retentiondata using methanol–water mobile phases with less than 15% methanolbecause of difficulties encountered in eluting each component of thetest mixture from the column.) The six compounds studied exhibitedclassical RPLC hydrophobic behavior. Plots of ln k′ versus organic mod-ifier concentration were linear for each test mixture compound investi-gated. Retention time decreases as the concentration of methanolincreases because the stationary phase is saturated with methanol(i.e., the organic modifier) over the mobile phase composition rangeinvestigated.

Fig. 4 shows Snyder solvent strength plots generated for each testmixture compound using at least twelve butanol–water–acetic acidmobile phases with the concentration of butanol varying from 0.75%to 4% v/v. All ln k′ plots are bilinear, with the break occurring at thesame mobile phase composition (2.5% butanol in water with 0.2%acetic acid) for each compound. Scott [9] has shown that saturation ofthe C18 stationary phase by butanol occurs at 2.5% v/v. Therefore, thebreak in each plot could be indicative of a change in the structure ofthe stationary phase [10]. The first linear segment of each plot denotedas “low butanol concentration” may correspond to a simultaneous

g butanol in water with 0.2% acetic acid as the mobile phase.

54 B.K. Lavine et al. / Microchemical Journal 103 (2012) 49–61

change in both the mobile and stationary phases, whereas the secondlinear segment of each plot denoted as “high butanol concentration”may correspond to classical RPLC behavior — that is, a change in thecomposition of the mobile phase with increasing organic modifier(i.e., butanol) concentration, with the stationary phase remainingunchanged because it has been saturated by butanol.

All of the butanol ln k′plotswere reproducible. The same resultswereobtained whether we started at higher organic modifier concentrationand moved towards lower concentration or vice-versa. Clearly, thebreak in the ln k′ plots cannot be attributed to a conformational effectinvolving the folding of the C18 chains.

For each vanillin test mixture compound, the computed kw valuein the regression equation developed from the “high alcohol concen-tration” butanol data is four to seven times less than the correspond-ing kw value in the regression equation developed from the methanolwater data. This result is not surprising because these kw values donot represent true kw values. Each represents what the capacity factorwould be, if the conformation and composition of the stationaryphase in pure water were the same as in the organic aqueous mix-tures used to generate these plots. Differences in these kw values reflectdifferences in the solvation of the C18 alkyl bonded phase by methanol

Fig. 8. Log–log plots for the vanillin test mixture compounds usin

and butanol. These differences cannot be attributed to uncertainty inthe least squares fitting of the data, which can be as high as 80% [11].

Schoemaker's solubility parameter model has been used by otherworkers to describe nonlinear behavior in ln k′ versus ϕ plots[12,13]. An Eϕ0.5 term, where E is a regression coefficient, is addedto Snyder's solvent strength model to describe curvature at low con-centrations (ϕb0.2) of organic modifier. However, the functionalform of the butanol data in the ln k′ versus ϕ plots shown suggeststhat a bilinear model is more appropriate to fit this data, which is sup-ported by residual plots of the regression. Furthermore, the data inthese butanol plots is for ϕb0.04. The degree of structure exhibitedby these plots for the mobile phase composition range (ϕb0.2) inves-tigated cannot be attributed to differences in the solubility parame-ters of the solute, mobile phase and stationary phase.

Water rich mobile phases were also used to investigate C8 andcyanopropyl bonded phase columns. Fig. 5A shows the chromato-gram of the vanillin test mixture using the methanol–water mobilephase (30% methanol in water with 0.2% acetic acid) that yieldedthe best separation on the BDS C8 column at ambient temperature,and Fig. 5B shows the chromatogram of the vanillin test mixtureusing the methanol water mobile phase (2.5% methanol in water

g methanol in water without acetic acid as the mobile phase.

1.04

3.35

3.81 5.25

6.12

6.75

1.20

4.36

4.92

7.25 8.

43

9.08

(A) 2.75% Butanol, C18column

(B) 3.75% Butanol, C8column

Fig. 10. Chromatograms of the vanillin test mixture. Elution order was vanillic acid,isovanillin, vanillin, o-vanillin, ethyl vanillin and coumarin. A) Butanol in water mobilephase without acetic acid (2.75% butanol in water) that yielded the best separation of thevanillin test mixture on the BDS C18 column. B) Butanol–water mobile phase (3.75% butanolin water) without acetic acid that yielded the best separation on the BDS C8 column.

Fig. 9. Log–log plots for the vanillin test mixture compounds using butanol in water without acetic acid as the mobile phase.

55B.K. Lavine et al. / Microchemical Journal 103 (2012) 49–61

with 0.2% acetic acid) that yielded the best separation on the BDS cya-nopropyl bonded phase column at ambient temperature. Fig. 5Cshows the chromatogram of the vanillin test mixture using the buta-nol water mobile phase (0.75% butanol in water with 0.2% acetic acid)that yielded the best separation on the BDS C8 column at ambienttemperature, and Fig. 5D shows the chromatogram of the vanillintest mixture using the butanol water mobile phase (2.5% butanol inwater with 0.2% acetic acid) that yielded the best separation on theBDS cyanopropyl column at ambient temperature. When comparedto C18, C8 yielded comparable results whereas the test mixture waspoorly resolved on the cyanopropyl bonded using either methanol–water or butanol–water mobile phases.

3.2. C18 versus C8

Snyder solvent strength plots were generated for methanol usingfive methanol–water–acetic acid mobile phases: 15%, 20%, 25%, 30%,and 35% methanol in water with 0.2% acetic acid and for butanol usingat least twelve butanol–water–acetic acid mobile phases with the con-centration of butanol varying from 0.75%,−3% v/v. The methanol inwater and butanol in water mobile phases yielded plots similar to C18.Therefore, log–log plots of retention [14] for C18 and C8 alkyl bondedphase columns were constructed to better understand the differencesin behavior of these two alkyl bonded phases. Linear correlations withslopes of unity in such plots are termed “homoenergetic” and are indic-ative of the same retention mechanism for the two columns, whereaslinear correlations with slopes different from unity are termed “home-oenergetic” and are indicative of similar but nonidentical thermody-namic retention behavior. Log–log plots are shown in Fig. 6 formethanol–water and in Fig. 7 for butanol–water. It is evident from an

1.20

4.36

4.92

7.25 8.

43

9.083.

20

4.32

4.93

7.20

8.41

8.97

4.07

4.29

4.92

7.16

8.37

8.93

4.45

4.92

7.01

8.40

8.96

(A) pH 2.92 (B) pH 3.57

(C) pH 4.07 (D) pH 6.0

Fig. 11. Chromatograms of the vanillin test mixture on the BDS C18 column using 2.75% butanol in water mobile phases at A) pH 2.92 (0.2% acetic acid), B) pH 3.57, C) pH 4.07, andD) pH 6 (no acetic acid).

Fig. 12. Chromatograms of the vanillin test mixture on the BDS C8 column using 2.75% butanol in water mobile phases at A) pH 2.92 (0.2% acetic acid), B) pH 3.57, C) pH 4.07, and D)pH 6 (no acetic acid).

56 B.K. Lavine et al. / Microchemical Journal 103 (2012) 49–61

57B.K. Lavine et al. / Microchemical Journal 103 (2012) 49–61

examination of these plots that C18 and C8 exhibit homeoenergetic be-havior for both methanol–water and butanol water mobile phases.However, the slopes of the regression lines for methanol–water arecloser to unity than butanol–water.

To better understand the reasons for homeoenergetic retentionbehavior as opposed to homoenergetic retention behavior, log–logplots of retention were generated for methanol–water and butanol–water mobile phases without acetic acid. Vanillic acid was excludedas it coeluted with the dead marker when acetic acid is not presentin the mobile phase. Log–log plots are shown for methanol–water inFig. 8 and are shown for butanol–water in Fig. 9. The slopes of the re-gression lines for butanol–water are comparable to methanol–water.Furthermore, the slopes of the regression lines for methanol–watermobile phases with and without acetic acid are comparable. Fromthe four sets of log–log plots generated, pH has a greater effect on re-tention with butanol–water mobile phases. This suggests that pH willhave a greater impact on the separation of the vanillin test mixturewhen butanol is used as the organic mobile phase modifier.

3.3. Controlling selectivity through pH

Fig. 10A shows a chromatogram of the vanillin test mixture usingthe butanol in water mobile phase without acetic acid (2.75% butanolin water) that yielded the best separation of the vanillin test mixtureon the BDS C18 column. Fig. 10B shows a chromatogram of the sametest mixture using the butanol–water mobile phase (3.75% butanol inwater) without acetic acid that yielded the best separation on the BDSC8 column. A comparable but faster separation of the vanillin test

Fig. 13. Chromatograms of the vanillin test mixture on BDS C18 column using 3.75% butanol in(no acetic acid).

mixture compounds was achieved using butanol–water mobile phaseswhen acetic acid was not present in the mobile phase. Because vanillicacid coelutes with the dead marker when acetic acid is not present inthe mobile phase, the adjustment of pH in the butanol–water mobilephase is crucial to ensure that vanillic acid does not coelute with thedead marker while ensuring adequate (baseline) resolution for allcomponents. By judiciously controlling the pH of the mobile phase,the separation of vanillic acid, isovanillin, and vanillin in the test mix-ture can be tuned by controlling the retention time of vanillic acidthrough deprotonation of its carboxylic acid group and the retentionof vanillin and isovanillin through its interactions with unreacted sila-nol groups. This will allow for the use of mobile phases with higherconcentrations of butanol resulting in faster separations.

Fig. 11 shows the separation of the vanillin test mixture on the BDSC18 columnusing 2.75% butanol inwatermobile phases at pH2.92 (0.2%acetic acid), pH 3.57, pH 4.07, and pH 6 (no acetic acid), and Fig. 12shows the separation of the same test mixture using the same mobilephases on the BDS C8 column. Figs. 13 and 14 show the separation ofthe same test mixture using a 3.75% butanol in water mobile phase atpH 2.92 (0.2% acetic acid), pH 3.57, pH 4.07, and pH 6 (no acetic acid)on the BDS C18 and BDS C8 columns respectively. The amount of butanolin each of these mobile phases corresponds to the composition of thebutanol in water mobile phase without acetic acid that yielded thebest separation of the vanillin test mixture on the BDS C18 and C8 alkylbonded phases. By controlling the pH of the mobile phase in the rangeof 3.5 to 4.0, problems associated with the co-elution of vanillic acidwith the dead marker (which occurs at higher pH) or overlap withisovanillin (which occurs at lower pH) were obviated. Judicious control

water mobile phase at A) pH 2.92 (0.2% acetic acid), B) pH 3.57, C) pH 4.07, and D) pH 6

Fig. 14. Chromatograms of the vanillin test mixture on BDS C8 column using 3.75% butanol in water mobile phase at A) pH 2.92 (0.2% acetic acid), B) pH 3.57, C) pH 4.07, and D) pH 6(no acetic acid).

58 B.K. Lavine et al. / Microchemical Journal 103 (2012) 49–61

of pH of the butanol in water mobile phase provided greater benefit forthe separation of the vanillin test mixture on C18. For this reason, the3.75% butanol in water at pH 4.03 on the BDS C18 column was used tocharacterize extract of vanilla in commercial products obtained fromsuper markets in the local area.

3.4. Surveying products containing extract of vanilla

Table 1 lists 36 commercial products (both foreign and domesticbrands) containing extract of vanilla which were analyzed using theRPLC method developed as part of this study. Vanillic acid, vanillin,ethyl vanillin and coumarin were found to be present in some or allof these brands. As coumarin has been shown to cause hepatoxicity inanimals, this result is of importance to the general public. The presenceof vanillic acid in some of these products can be attributed to oxida-tion of vanillin suggesting problems with the initial formulation, theshelf life of the product or some combination of the two.

Fig. 15 shows a principal component (PC) plot [15] of the 36 com-mercial products containing extract of vanilla. Each product wasrepresented by 13 unique retention time windows identified basedon the shape of the peak and its retention time. Thus, each chromato-gram for PCA was represented as a data vector, x=(x1, x2, x3, … x13)where each vector component was the peak area of one of the 13peaks. Five products appeared as outliers in the PC plot: Caravelle(artificial extract), Best Choice (real vanilla extract), Always Save,Singing Dog, and Clover Valley (real vanilla extract). After deletingthese five products from the analysis with the remaining 31 samplesreanalyzed, four distinct clusters were detected using principal compo-nent analysis (see Fig. 16). Cluster 1 consisted of 12 real extracts of va-nilla: Spice Club, Nielsen Massey, Watkins (real extract), Rodelle,

McCormick (Real Extract), Great Value (Real Extract), Market Pantry,Simply Organic, Blackbay, Now, Danncy Vainilla, and Archer Farms.Cluster 2 consisted of 9 products: 6 artificial, 2 real and 1 combinationof real and artificial vanilla extract. Molina, Griffins, Francelin, Molino,Watkins (artificial extract),Mi Huerta, Adams, Alicante, andMcCormick(artificial extract). Cluster 3 consisted of 4 artificial extracts of vanilla:SAC, Viola, Dr. Oetke, and Aroma Pasta Vaneli, and cluster 4 consistedof 4 artificial and real extracts of vanilla: Clover Valley (artificialextract), Great Value (artificial extract), Mama Lycha (combinationof artificial and real extract), and Universal.

Chromatograms representative of each cluster are shown in Fig. 17.The chromatograms of the so-called outliers are shown in Fig. 18. Evi-dently, there ismore than one class of real vanilla extract using chemicalconstitution as the variable to differentiate between extracts. Further-more, several products containing artificial vanillin extract have chro-matographic profiles similar to those of real vanilla extract. As for theso-called outliers, there was no pattern evident in the data as 3 werereal vanilla extracts and 2 were artificial vanilla extracts.

Several compounds not part of the original vanillin test mixtureand consequently designated as unknown in the vanilla extract chro-matograms obtained using the Varian Prostar HPLC were identified byLC/MS. The results of the LC/MS study are summarized in Table 2. Allreported compounds are consistent with the reported molecularweights, retention times, and the observed isotope patterns. The com-pounds listed in this table had not been previously reported as con-stituents of extract of vanilla. However, they had been reported atparts per million levels in vanilla beans [16]. In all likelihood, the de-tection of these compounds in vanilla extract is linked to the use ofwater rich mobile phases, which yielded better separations for thesewater soluble compounds.

Table 1Vanilla extract products analyzed by RPLC.

Product Vanillic acid Isovanillin Vanillin O-vanillin Ethyl vanillin Coumarin

Danncy Vainilla Detected Not detected Detected Not detected Detected DetectedMolino (real extract) Detected Not detected Detected Not detected Detected Not detectedCaravelle (artificial) Detected Not detected Detected Not detected Detected Not detectedWatkins (artificial) Detected Not detected Detected Not detected Detected Not detectedAdams (artificial) Detected Not detected Detected Not detected Detected Not detectedMi Huerta (artificial+real) Not detected Not detected Detected Not detected not detected Not detectedClover valley (artificial) Detected Not detected Detected Not detected Detected not detectedGreat value (artificial) Detected Not detected Detected Not detected Detected DetectedGriffins (artificial) Detected Not detected Detected Not detected Detected Not detectedGreat value (real extract) Detected Not detected Detected Not detected Not detected Not detectedBest choice (real extract) Not detected Not detected Detected Not detected Not detected Not detectedMc Cormick (real extract) Detected Not detected Detected Not detected Not detected Not detectedRodelle (real extract) Detected Not detected Detected Not detected Not detected Not detectedSAC Detected Not detected Detected Not detected Not detected DetectedWatkins (real extract) Detected Not detected Detected Not detected Not detected Not detectedAlways save (artificial) Not detected Not detected Detected Not detected Detected Not detectedMolina (real extract) Detected Not detected Detected Not detected Detected Not detectedViola Detected Not detected Detected Not detected Detected Not detectedArcher farms (real extract) Detected Not detected Detected Not detected Detected Not detectedBlack Bay Real Detected Not detected Detected Not detected Not detected Not detectedMama Lycha (artificial)+(real extract) Not detected Not detected Detected Not detected Detected Not detectedAlicante (artificial) Not detected Not detected Detected Not detected Detected Not detectedFrancelin Not detected Not detected Detected Not detected Detected Not detectedUniversal (real extract) Not detected Not detected Detected Not detected Detected Not detectedDr Oetke Not detected Not detected Detected Not detected Detected Not detectedNielsen Massy Detected Not detected Detected Not detected Not detected Not detectedSinging dog (real extract) Not detected Not detected Detected Not detected Not detected Not detectedSpice club (real extract) Detected Not detected Detected Not detected Detected Not detectedSimply organic (real extract) Detected Not detected Detected Not detected Not detected Not detectedClover valley (real extract) Not detected Not detected Detected Not detected Not detected Not DetectedMarket pantry (real extract) Detected Not Detected Detected Not Detected Detected Not DetectedAroma pasta vaneli Detected Not Detected Detected Not Detected Detected DetectedFrontier (real extract) Not Detected Not Detected Detected Not Detected Detected Not DetectedGriffins Cartificial Detected Not Detected Detected Not Detected Detected Not DetectedNow (real extract) Detected Not Detected Not Detected Not Detected Detected Not DetectedMcCormick (artificial) Not Detected Not Detected Detected Not Detected Not Detected Not Detected

59B.K. Lavine et al. / Microchemical Journal 103 (2012) 49–61

4. Conclusions

The use of water rich mobile phases offers advantages for theseparation of water soluble andweakly retained compounds in RPLC in-cluding shorter retention times, improved separation of congeners andbetter tuning of HPLC separations.

Fig. 15. PC plot of the 36 extracts of vanilla products. Each product was represented by 13unique retention timewindows identified based on the shape of the peak and its retentiontime. Outliers are circled.

Acknowledgments

BKL and NP acknowledge helpful discussions with Professor Ziad ElRassi of Oklahoma State University. Financial support of the NationalScience Foundation by the establishment of an REU site at OklahomaState University (Award #0649162) is gratefully acknowledged byDTC and BKL.

Fig. 16. PC plot of the 31 products (with 5 outliers removed). Four distinct clusters ofpoints are evident in the PC plot.

Fig. 17. Chromatograms of the clusters detected by principal component analysis.

Fig. 18. Chromatograms of the outliers detected by principal component analysis.

60 B.K. Lavine et al. / Microchemical Journal 103 (2012) 49–61

Table 2LC/MS analysis of unknown chromatographic peaks.

Product Acetovanillone Anisyl formate p-hydroxy-benzaldehyde 2-Phenylethanol Glucose

Caravelle Detected Detected Not detected Not detected DetectedDanncy Vainilla Not detected Not detected Detected Detected DetectedGreat value (artificial) Not detected Not detected Detected Detected Not detectedUniversal (real extract) Not detected Not detected Detected Detected DetectedSinging dog (real extract) Not detected Not detected Detected Detected DetectedWatkins (real extract) Not detected Not detected Detected Detected DetectedAdams double strength Not detected Not detected Not detected Not detected Detected

61B.K. Lavine et al. / Microchemical Journal 103 (2012) 49–61

References

[1] L.S. de Jager, G.A. Perfetti, G.W. Diachenko, Determination of coumarin, vanillin,and ethyl vanillin in vanilla extract products: liquid chromatography/massspectrometry method development, J. Chromatogr. A 1145 (2007) 83–88.

[2] A.K. Sinha, S.C. Verma, U.K. Sharma, Development and validation of an RP-HPLCmethod for quantitative determination of vanillin and related phenolic compoundsin Vanilla planifolia, J. Sep. Sci. 30 (2007) 15–20.

[3] A.M. van Nederkassel, C.J. Xu, P. Lancelin, M. Sarraf, D.A. MacKenzie, N.J. Walton, F.Bensaid, M. Lees, G.J. Martin, J.R. Desmurs, D.L. Massart, Chemometric treatmentof vanillin fingerprint chromatograms — effect of different signal alignmentson principal component analysis plots, J. Chromatogr. A 1120 (2006)291–298.

[4] G. Lamprecht, F. Pichlmayer, E.R. Schmid, Determination of the authenticity ofvanilla extracts by stable isotope ratio analysis and component analysis byHPLC, J. Agric. Food Chem. 42 (1994) 1722–1727.

[5] B.K. Lavine, J.P. Ritter, S. Peterson, Enhancement of selectivity in reversed-phaseliquid chromatography, J. Chromatogr. A 946 (2002) 83–90.

[6] K. Valko, L.R. Snyder, J.L. Glajch, Retention in reversed phase liquid chromatographyas a function of mobile phase composition, J. Chromatogr. 656 (1993) 501–511.

[7] N. Chen, Y.K. Zhang, P.C. Lu, Effect of molecular structure on the log kw index andlinear S-log kw correlation in reversed phase liquid chromatography, J. Chromatogr.633 (1993) 31–38.

[8] J. Dolan, L.R. Snyder, R.G. Wolcott, P. Haber, T. Baczek, R. Kaliszan, Reversed phaseliquid chromatographic separation of complex samples by optimizing temperatureand gradient time, J. Chromatogr. A 857 (1999) 41–68.

[9] R.P.W. Scott, C.F. Simpson, Solute–solvent interactions on the surface of reversedphases, Faraday Symp. Chem. Soc. 15 (1980) 69–82.

[10] A. Tchapla, H. Colin, G. Guiochon, Linearity of homologous series of retentionplots in reversed phase liquid chromatography, Anal. Chem. 56 (1984) 621–626.

[11] M.M. Hseih, J.G. Dorsey, Accurate determination of log kw in reversed phase liquidchromatography — implications for quantitative structure–activity retentionrelationships, J. Chromatogr. A 631 (1993) 63–78.

[12] P.J. Schoenmakers, H.A.H. Billiet, R. Tussen, L. DeGalan, Description of soluteretention over the full range of mobile phase compositions in reversed phase liquidchromatography, J. Chromatogr. 282 (1983) 107–115.

[13] P.J. Schoenmakers, H.A.H. Billiet, R. Tussen, L. DeGalan, Gradient selection inreversed phase liquid chromatography, J. Chromatogr. 149 (1978) 519–528.

[14] W.T. Cooper, L.-Y. Lin, Effect of stationary phase polarity on retention in reversedphase liquid chromatography, Chromatographia 21 (1986) 335–341.

[15] J.E. Jackson, A User's Guide to Principal Component Analysis, Wiley Interscience,NY, 1991.

[16] A. Perez-Silva, E. Odoux, P. Brat, F. Ribeyre, G. Rodriguez-Jumenes, V. Robles-Olvera,M.A. Garcia-Alvarado, Z. Gunata, GC–MS and GC-olfactometry analysis of aromacomponents in a representative organic aroma extract from cured vanilla beans,Food Chem. 99 (2006) 728–735.