J Sci Food Agric 1998, 77, 121È126

Headspace Concentrations of Ethyl Esters atDiþerent Alcoholic StrengthsJohn M Conner,* Lorraine Birkmyre, Alistair Paterson and John R Piggott

Centre for Food Quality, University of Strathclyde, Department of Bioscience & Biotechnology,204 George Street, Glasgow G1 1XW, UK

(Received 11 November 1996 ; revised version received 22 May 1997 ; accepted 23 September 1997)

Abstract : Ester concentrations in the headspace inÑuence the aroma character ofalcoholic drinks. Activity coefficients for esters showed log-linear decreases asethanol concentration was increased from 17% (v/v), with rates inversely relatedto ester acid chain length. At concentrations below 17% (v/v) the activity coeffi-cient remained constant. This could be related to structural changes in ethanol/water mixtures. Below 17% (v/v), ethanol forms a monodispersed aqueoussolution. Above 17% (v/v), ethanol molecules cluster to reduce hydrophobichydration and esters partition into these ethanol-rich clusters, where the lowerester interfacial tension reduces the free energy of mixing and hence the activitycoefficient. The increased solubility of the ester reduced the headspace concentra-tion of the esters, and hence total ester content may not be a good indicator oftheir Ñavour impact. 1998 SCI.(

J Sci Food Agric 77, 121È126 (1998)

Key words : ethanol ; water structure ; wine ; beer ; spirits

INTRODUCTION

Esters are key Ñavour-active molecules in alcoholicdrinks. The esters considered important for beer Ñavourinclude isoamyl acetate and ethyl esters with acid chainlengths of two to eight carbons (Meilgaard and Peppard1986). In wines ethyl esters have up to 12 carbons in thealiphatic chain (Ribereau-Gayon 1978) and in spirits upto 18 carbons (Nyka� nen and Suomalainen 1983). Thearoma character of ethyl esters changes from fruity tosoapy as aliphatic chain length increases (Ribereau-Gayon 1978). Comparison of sensory thresholds forindividual esters in beers (Engan 1981), model wines(Ribereau-Gayon 1978) and spirits diluted to 9É4%ethanol (v/v) (Salo et al 1972) and 23% ethanol (v/v)(Perry 1989) showed that ethanol concentration hadlittle e†ect on the odour thresholds. For ethyl hexano-ate odour thresholds in a model wine and water were

* To whom correspondence should be addressed.Contract/grant sponsor : Biotechnology and Biological Sci-ences Research Council, UKContract/grant sponsor : Chivas and Glenlivet Group.

the same (0É037 mg litre~1). The threshold increased onaddition of sugar (0É056 mg litre~1) and was markedlyhigher in a dry white wine (0É85 mg litre~1).

The e†ect of alcohols and other cosolvents on theaqueous solubility of nonpolar drugs has been investi-gated (Yalkowsky et al 1976). The ability of a cosolventto solubilise any drug was found to be inversely pro-portional to its interfacial tension against a referenceliquid hydrocarbon. Ethanol, which is miscible withboth hydrocarbons and ethyl esters, was the most effi-cient solubiliser of the solvents considered.

Sensory thresholds are generally expressed as solu-tion concentrations. In previous studies on dilutedspirits, Conner et al (1994a,b) have shown that inter-actions between hydrophobic aroma componentsreduced headspace concentrations or partition coeffi-cients. Medium- and long-chain esters, alcohols andaldehydes contributed in relation to their solubility oractivity coefficient (Conner et al 1994b). Changes inthresholds may therefore arise from changes in theheadspace partition coefficient of a compound either asa result of changes in solubility, or from interactionwith other solution components.

1211998 SCI. J Sci Food Agric 0022È5142/98/$17.50. Printed in Great Britain(

122 J M Conner et al

Headspace concentrations may be expressed relativeto the solution concentration (partition coefficient) orrelative to the headspace concentration over the puresolute (activity). From the latter, the activity coefficientof a dissolved solute can be calculated and can berelated to many solution properties. Prediction of activ-ity coefficients is a major goal of chemical engineeringresearch for the determination of vapourÈliquid andliquidÈliquid equilibria, and of pharmaceutical researchfor the determination of solubility (Grant and Higuchi1990).

Several empirical or semi-empirical methods haveevolved for the estimation of the aqueous solubility oforganic compounds. The UNIFAC procedure wasdesigned to predict activity coefficients of mixtures frompure component data using group activity coefficients(Yalkowsky and Banerjee 1992). UNIFAC is anacronym for UNIQUAC Functional-group ActivityCoefficient and derives from the UNIQUAC (UniversalQuasi Chemical) equation based on GuggenheimÏsquasi-chemical theory of liquid mixtures (Guggenheim1967). In the UNIQUAC model the expression of theactivity coefficient of a component in a liquid mixture iscomposed of two parts (Grant and Higuchi 1990). TheÐrst is a combinatorial part that reÑects the excessentropy of mixing due to di†erences in geometry of themolecules of the mixture. The second represents theresidual contribution that reÑects the excess enthalpy ofmixing due to the interaction between molecules. Theapproach involves two successive correlation steps.First, non-linear regression analysis obtains estimates ofUNIQUAC parameters that give the best Ðt to theexperimental quantity, ie activity coefficient of com-pounds in a mixture. These parameters are then used toestimate or predict the activity coefficient of the sub-stance in another solution. The UNIQUAC model isapplied to complete molecules. Fredenslund et al (1975)extended the concept to the functional groups in a mol-ecule to provide a new group contribution method forpredicting activity coefficients in liquid mixtures. In thegroup contribution approach the molecule is consideredto be composed of groups, each of which is assumed toact independently of the rest of the molecule, and makeits own contribution to a speciÐc thermodynamic pro-perty. The advantage of the group contributionapproach is its versatility, since a multitude of com-pounds are accessible from only a few groups. Like theUNIQUAC model, the UNIFAC method proceeds intwo steps. First, the experimental activity coefficientdata are reduced by regression to give UNIFAC param-eters characterising the interactions between pairs offunctional groups. Then, activity coefficients in othersystems can be predicted using parameters appropriateto the various functional groups. The UNIFAC inter-action parameters have since been determined for manyfunctional groups therefore eliminating the Ðrst step(Yalkowsky and Banerjee 1992). Parameters have also

been incorporated into computer programmes whichcan perform these calculations and so predict activitycoefficients in complex mixtures.

To study whether di†erences in aroma threshold canbe ascribed to solution or aroma compound inter-actions, we determined the activity coefficients of ethylesters with alkyl chain lengths of between two and 12carbons at a range of aqueous ethanol concentrations.As a comparison the activity coefficients of esters werecalculated for each alcohol concentration using theUNIFAC procedure.

MATERIALS AND METHODS

Components

Ethanol was HPLC grade (Rathburn Ltd, Walkerburn,UK) and ethyl esters 99% pure (Sigma Chemical, Poole,UK). Water was distilled and Ðltered using a Millipore-Q system.

Activity coefficient determinations

The activities of ester in solution were obtained bychromatographic determinations of headspace concen-trations and activity coefficients calculated as describedpreviously (Conner et al 1994a).

Glass vials (20 ml), Ðtted with PTFE lined siliconesepta in plastic screw caps, were Ðlled with 10 ml ali-quots of standard ethanolic solutions of esters, pre-viously diluted with water to the desired Ðnal ethanolconcentration. After equilibration in a 25¡C water bathfor at least 30 min, a 2É5 ml sample of headspace waswithdrawn using a 5 ml gas tight syringe, preheated to50¡C. Only one headspace injection was made per vialand samples were analysed in duplicate on a CarloErba Mega series gas chromatograph (CE InstrumentsLtd, Crawley, UK) using a Ñame ionisation detector.Peak areas were calculated using Chromperfect integra-tion software (Justice Innovations, Mountain View,California, USA). On-column injection used a0É55 mm ] 0É5 m ultimetal retention gap (ChrompakLtd, London, UK) with an external gas tight septum. A0É75 mm ] 30 m Supelcowax column (df\ 1 lm)(Supelco Ltd, Poole, UK) was used with helium carriergas at 10 ml min~1. The column was held at 50¡C for3 min after injection, increasing to 220¡C at18¡C min~1. Pure solute headspace injection followedthe same procedure but used a 100 ll syringe for themore volatile esters.

Computer predictions of activity coefficients

UNIFAC predictions of ester activity coefficients weremade using the Justemix programme, version 2 (Biosoft,

Headspace concentrations of ethyl esters 123

Cambridge, UK). Predictions were based on 2] CH3 ,and and used interactionCH2COO (n [ 1)] CH2

parameters derived from classical vapourÈliquid equi-libria data.

Measurement of interfacial tension

Interfacial tensions were measured using a Torsionbalance (White Electrical Instruments Ltd, MalvernLink, UK) using a platinum ring with a circumferenceof 4 cm. The balance was tested by measuring thesurface tension of the water used in solution prep-aration before each measurement. Mean surface tensionwas 73 mNm~1 (published value of 72É8 mNm~1 at20¡C) with a standard deviation of 0É5 mNm~1.Samples, 10 ml aqueous ethanol plus 10 ml ester oralkane, were equilibrated at 25¡C for 1 h before measur-ing.

RESULTS AND DISCUSSION

Headspace concentrations of ethyl esters

Headspace concentrations of ethyl esters were plottedagainst their solution concentrations at ethanol concen-trations typical of a number of alcoholic drinks. Thosechosen were 5% (beer), 10% (wine), 17% (fortiÐed wine),23% (sensory assessment of distilled drinks) and 40%ethanol (v/v) (bottle strength of distilled drinks). Ethylhexanoate is shown in Fig 1. All the esters investigatedfollowed a similar pattern with little change in gradientat 5, 10 and 17% ethanol (v/v), a decrease at 23%ethanol (v/v) and marked decrease at 40% ethanol (v/v).

Fig 1. The relationship between headspace and solution con-centrations for ethyl hexanoate at 5%, 10%, 17%, 23% and

40% ethanol (v/v).

The solubilising power of ethanol for ethyl esters

The relationship between ethanol concentration andester activity coefficient was further investigated withadditional determinations at 0É5, 30, 35, 50, 60 and 80%ethanol (v/v). In studies on non-polar drugs the solu-bilising power of the co-solvent, p, was calculated fromthe plot of log solubility against volume fraction ofethanol (Yalkowsky et al 1976). Using activity coeffi-cients (Fig 2) the regression equation becomes

log cf\ log c

w[ p f

cwhere log is the activity coefficient in the ethanolc

fwater mixture ; log is the activity coefficient in waterc

wand is the volume fraction of ethanol. This equationf

cassumes a semilog decrease in the activity coefficient ofester with increasing ethanol concentration. Howeverfor longer chain esters, ie decanoate and dodecanoatelog showed no signiÐcant change from 0É5 to 17%c

fethanol (v/v). Consequently, there are two possible esti-mates of p, the solubilising power of ethanol. The Ðrstestimate used data from the entire volume fractionrange and the second used data only for the concentra-tion range of 17 to 80% ethanol (v/v). Only minor dif-ferences are observed between the two estimates forlower homologues. For acetate, using all points gave anestimate of 1É71 (standard deviation of regression line(SD)\ 0É13) while the restricted range gave 2É0(SD\ 0É15). Di†erences in p were more marked fordecanoate and dodecanoate, increasing from 6É1 and 6É3(SD\ 0É43) using all points, to 6É8 and 7É1 (SD \ 0É37)using the restricted range. Comparison of R2 values forthe relationship showed no change for acetate (0É98) butan improvement using the restricted range for deca-noate (0É96 to 0É99) and dodecanoate (0É94 to 0É96).

Whichever method is used, the results are similar tothose reported for nonpolar drugs. p was dependent on

Fig 2. The e†ect of increasing ethanol concentration on theactivity coefficient of ethyl esters. Lines were Ðtted by calcu-

lated linear regression using all points.

124 J M Conner et al

the change in the surface free energy at the molecularinterface between the solute and the solvent as cosol-vent increased, with contributions from interactionswith both the hydrocarbon and ester portions of themolecule (Yalkowsky et al 1976). For mixed solvents,interactions involving the hydrophobic portion of thesolute were the major factors that determined the solu-bilising power of the cosolvent. Consequently, moleculeswith the largest hydrophobic surface area were the mosta†ected by the addition of cosolvent. Increasing ethanolconcentration lowered the interfacial tension betweenthe aqueous phase and ethyl esters (Fig 3). The rate ofdecrease was the same for all the esters examined indi-cating that the decrease arises from a change in theaqueous phase. The increase in interfacial tension withthe length of the hydrocarbon chain has also beenreported for alkanes and was attributed to changes inthe density of the alkane (ZograÐ and Yalkowsky 1974).

Relationships between activity coefficient and ester chainlength

The data from Fig 2 were replotted as log activity coef-Ðcient against the number of acid methylene groups(c

f)

at each alcohol concentration. For clarity results at 0É5and 5% (v/v) ethanol have been omitted from Fig 4.Linear regression at each alcohol concentration usedthe general equation

log cf\ log c

o] B@n

where the intercept, is the contribution to thelog co,

activity coefficient of the ester group ; n is the number ofacid methylene groups in the molecule and B@ themethylene group contribution. R2 values were all[0É98. The activity coefficient is the excess or non-idealfunction of mixing solute and solvent and log and B@c

ocan be converted to free energy contributions (2É303 RT

Fig 3. The e†ect of increasing aqueous ethanol concentrationon the interfacial tension of ethyl esters and octane.

Fig 4. The e†ect of ester chain length on the activity coeffi-cients of ethyl esters at 10%, 17%, 23%, 30%, 40%, 50%, 60%and 80% ethanol (v/v). Lines were Ðtted by calculated linear

regression.

Fig 5. The e†ect of increasing ethanol concentration on theexcess free energy contribution of the ethyl ester group

Lines were Ðtted by calculated linear regres-(HÉ É ÉCOOC2H5). sion.

Fig 6. The e†ect of increasing ethanol concentration on theexcess free energy contribution of the acid methylene group.Lines were Ðtted by calculated linear regression using the con-

centration range 17É5È80% ethanol (v/v).

Headspace concentrations of ethyl esters 125

log c). Figures 5 (ester) and 6 (methylene) show howthese group contributions changed with ethanol concen-tration. Ester and methylene group contributions werealso calculated from activity coefficients estimated bythe UNIFAC procedure. The UNIFAC predictionsyield linear results for p and for both ester and methy-lene group contributions. Again the experimental datawere linear for only part of the range studied. For theester group contribution the two lines have similar gra-dients (measured [1É16, SD \ 0É08 ; predicted [1É29,SD \ 0É03) but the measured set shows no signiÐcantdi†erence for points at 0É5, 5, 10 and 17% ethanol (v/v).

The energy contribution of the ester group at 0É5%(v/v) ethanol was calculated to be 6500 Jmol~1, but ascalculated this included a correction factor for the endmethyl group (Davis 1973a). A mean correction value of4770 Jmol~1 was obtained from alkane solubility andwas considered valid for partition studies provided thatthey involved reasonably non-polar solvents. Thisleaves a value of 1730 Jmol~1 attributable to the ethylester group. Published values (Davis 1973b) for thetransfer of OH and COOH groups to water fromalkane were [8500 and [12 500 Jmol~1 respectivelyexcluding correction factors. This decreased to[2134 Jmol~1 for transfer of both OH and COOHfrom octanol and 540 Jmol~1 for transfer of COOHfrom ethyl acetate. The positive free energy contributionsuggested that the ethyl ester group is hydrophobic. Thedecrease with increasing ethanol concentration mayarise from changes in hydration, either of the estergroup or of the end methyl group. For the ester groupthe UNIFAC predicted data produced an almost paral-lel line to the measured data set. This suggested thatthere is a systematic error in the prediction most likelyarising from the selection of the molecular substituents.The predictions were made using 2 ] CH3 , CH2COOand which have a number of possible(n [ 1) ] CH2 ,molecular conÐgurations. The package would thereforegive the same predictions for methyl, ethyl or propylesters.

The energy contribution of each acid methylenegroup at 0É5% ethanol (v/v) was calculated to be3000 Jmol~1. This is close to the values calculated forboth acids and alcohols (3400 to 3500 Jmol~1, Davis1973b). No signiÐcant di†erence was observed for 0É5, 5,10, 17% ethanol (v/v) for the methylene group. The sub-sequent gradient was steeper than that predicted byUNIFAC (measured [0É66, SD \ 0É03 ; predicted[0É44, SD\ 0É01).

Behaviour of ethanol–water mixtures

Both methods of presenting the data showed a plateauat low ethanol concentration, where increasing ethanolconcentration had no e†ect on the activity coefficient ofthe ethyl ester. The UNIFAC predictions however gave

linear results for p, ester and methylene group contribu-tions consistent with even mixing of ethanol and waterat the molecular interface. The experimental patternhowever can be closely related to the structure ofethanolÈwater mixtures (DÏAngelo et al 1994). Com-pressibility and IR measurements have shown that inaqueous solutions containing up to 15% ethanol (v/v),alcohol molecules are monodispersed in water. From 20to 57% ethanol (v/v), there is a progressive aggregationof alcohol molecules reducing the hydrophobic hydra-tion of the alkyl chain. Above 57% ethanol (v/v), thehydrogen bonded network of water is lost and the solu-tion becomes water monodispersed in ethanol. The con-centration at which the activity coefficient of estersstarted to decrease falls in the region of the Ðrst criticalpoint reported by DÏAngelo et al (1994). This indicatesthat for the plateau region, at low ethanol concentra-tions, the solution is essentially aqueous with theethanol and ester monodispersed. Above this Ðrst criti-cal point, ethanol self associates to form “pseudo-micellesÏ. The more rapid decrease in the excess freeenergy of the methylene group could be explained if anyincreasing proportion of the hydrophobic ester wasincorporated into these aggregates. This would continueuntil water loses its hydrogen bonded network com-pletely and mixes into solution as a single molecule andthe solution becomes essentially ethanolic. No distinctbreak point is observed for this second critical point ineither Fig 2 or 6. This indicates that, for ethyl esters, thechange from the ethanol rich “pseudo-micellesÏ to anethanolic solution has little e†ect on the partitioning ofthe ester with the headspace.

The addition of urea and salts are known to a†ect thecritical micelle concentrations and micelle size of non-ionic surfactants (Elworthy et al 1968). Urea, whichincreases the aqueous solubility of non-electrolytes, alsoincreases the concentration at which n-propanol formsmicelle-like aggregates (Hawlicka and Grabowski 1995).If the aggregation of ethanol in aqueous solution is likethat of non-ionic surfactants and n-propanol, then thepresence of other solutes, particularly if they alter thedegree of structure within the water, could indirectlya†ect the activity coefficients of ethyl esters.

CONCLUSIONS

For alcoholic beverages containing less than 17%ethanol no signiÐcant di†erences in activity coefficientswere observed with changes in ethanol concentration.In this range solutions are essentially aqueous and sodi†erences in aroma thresholds between beers, winesand fortiÐed wines must result from either sensory ofphysico-chemical interaction with other beverage com-ponents. Above 20% ethanol (v/v) aroma thresholdswould be expected to increase as partitioning into the

126 J M Conner et al

ethanol clusters reduces headspace concentrations.Aroma thresholds for esters in traditional bottle-strength spirits (40% ethanol (v/v)) would be between 2and 40 fold higher than in wines. Reducing bottlingstrength to 30È35% ethanol (v/v), to reduce excise duty,would increase headspace partition coefficients by 2È10-fold and conversely decrease aroma thresholds. Suchchanges would most a†ect longer chain esters. Thesoap-like aroma notes of these esters are associated withimmature character in distillates (Piggott et al 1992) andso a reduction in perceived spirit quality might resultfrom the reduced bottling strength.

ACKNOWLEDGEMENTS

This work is funded by the Biotechnology and Bio-logical Sciences Research Council of the UK and theChivas and Glenlivet Group. Peter Halling is thankedfor access to the Justemix programme for activity coeffi-cient determinations.

REFERENCES

Conner J M, Paterson A, Piggott J R 1994a Agglomeration ofethyl esters in model spirit solutions and malt whiskies. JSci Food Agric 64 45È53.

Conner J M, Paterson A, Piggott J R 1994b Interactionsbetween ethyl esters and aroma compounds in model spiritsolutions. J Agric Food Chem 42 2231È2234.

DÏAngelo M, Onori G, Santucci A 1994 Self-association ofmonohydric alcohols in water : compressibility and infraredabsorption measurements. J Chem Phys 100 3107È3113.

Davis S S 1973a Determination of the thermodynamics of themethyl group in solutions of drug molecules. J PharmPharmac 25 1È12.

Davis S S 1973b Determination of the thermodynamics ofhydroxyl and carboxyl groups in solutions of drug mol-ecules. J Pharm Pharmac 25 982È992.

Elworthy P H, Florence A T, MacFarlane C B 1968 Solu-bilisation by Surface-Active Agents. Chapman & Hall,London, UK, pp 13È60.

Engan S 1981 Beer composition : volatile substances. In :Brewing Science (Vol 2), ed Pollock J R. Academic Press,London, UK, pp 93È157.

Fredenslund A, Jones R L, Prausnitz, J M 1975 Group contri-bution estimation of activity coefficients in nonideal liquidmixtures. Am Inst Chem Eng J 21 1086È1099.

Grant D R, Higuchi T 1990 Solubility Behaviour of OrganicCompounds. John Wiley & Sons, New York, USA, pp 307È354.

Guggenheim E A 1967 T hermodynamics. North-HollandPhysics Publishing, Amsterdam, The Netherlands, pp 170È219.

Hawlicka E, Grabowski R 1995 The inÑuence of urea on thehydrophobic phenomena in aqueous solutions of alcohols.Chem Phys L etts 236 64È70.

Meilgaard M C, Peppard T L 1986 The Ñavour of beer. In :Food Flavours Part B. T he Flavour of Beverages, edsMorton I D & MacLeod A J. Elsevier Science Publishers,Amsterdam, The Netherlands, pp 99È170.

Nyka� nen L, Suomalainen H 1983 Aroma of Beer, W ine andDistilled Beverages. Reidel Publishing, Dordecht, TheNetherlands, pp 131È199.

Perry D R 1989 Odour intensities of whisky compounds. In :Distilled Beverage Flavour, eds Piggott J R & Paterson A.Ellis Horwood, Chichester, UK, pp 200È207.

Piggott J R, Conner J M, Clyne J, Paterson A 1992 The inÑu-ence of non-volatile constituents on the extraction of ethylesters from brandies. J Sci Food Agric 59 477È482.

Ribereau-Gayon P 1978 Wine Ñavour : In : Flavour of Foodsand Beverages, eds Charalambous G & Inglett G E. Aca-demic Press, New York, USA, pp 355È380.

Salo P, Nyka� nen L, Suomalainen H 1972 Odor thresholdsand relative intensities of volatile aroma components in anartiÐcial beverage imitating whisky. J Food Sci 37 394È398.

Yalkowsky S H, Banerjee S 1992 Aqueous Solubility : Methodsof Estimation for Organic Compounds. Marcel Dekker, NewYork, USA, pp 41È127.

Yalkowsky S H, Valvani S C, Amidon G L 1976 Solubility ofnonelectrolytes in polar solvents IV: nonpolar drugs inmixed solvents. J Pharm Sci 65 1488È1494.

ZograÐ G, Yalkowsky S H 1974 Interfacial properties of polarliquids against nonpolar phases. J Pharm Sci 63 1533È1536.