Food Chemistry 154 (2014) 90–101

Contents lists available at ScienceDirect

Food Chemistry

journal homepage: www.elsevier .com/locate / foodchem

Relationships between harvest time and wine composition in Vitisvinifera L. cv. Cabernet Sauvignon 2. Wine sensory properties andconsumer preference

0308-8146/$ - see front matter � 2014 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.foodchem.2013.12.099

⇑ Corresponding author. Tel.: +61 8 83136190; fax: +61 8 83136601.E-mail address: keren.bindon@awri.com.au (K. Bindon).

Keren Bindon ⇑, Helen Holt, Patricia O. Williamson, Cristian Varela, Markus Herderich, I. Leigh FrancisThe Australian Wine Research Institute, P.O. Box 197, Glen Osmond, Adelaide, South Australia 5064, Australia

a r t i c l e i n f o

Article history:Received 10 October 2013Received in revised form 21 December 2013Accepted 29 December 2013Available online 8 January 2014

Keywords:Fruit maturityCabernet Sauvignon wineEthanolAromaVolatilesPLS regressionFruitDark fruitRed fruitGreenVegetativeColourBitternessAstringencySensory descriptive analysisConsumer hedonic test

a b s t r a c t

A series of five Vitis vinifera L. cv Cabernet Sauvignon wines were produced from sequentially-harvestedgrape parcels, with alcohol concentrations between 12% v/v and 15.5% v/v. A multidisciplinary approach,combining sensory analysis, consumer testing and detailed chemical analysis was used to better definethe relationship between grape maturity, wine composition and sensory quality. The sensory attributeratings for dark fruit, hotness and viscosity increased in wines produced from riper grapes, while the rat-ings for the attributes red fruit and fresh green decreased. Consumer testing of the wines revealed that thelowest-alcohol wines (12% v/v) were the least preferred and wines with ethanol concentration between13% v/v and 15.5% v/v were equally liked by consumers. Partial least squares regression identified thatmany sensory attributes were strongly associated with the compositional data, providing evidence ofwine chemical components which are important to wine sensory properties and consumer preferences,and which change as the grapes used for winemaking ripen.

� 2014 Elsevier Ltd. All rights reserved.

1. Introduction whether these wine sensory attributes are defined by changes in

Vitis vinifera L. cv Cabernet Sauvignon wines have been de-scribed as presenting a ‘dichotomy of sensory attributes’ and arenotably distinguished by the presence of both vegetative and fruitycharacteristics (Heymann & Noble, 1987; Preston et al., 2008). Gen-erally it is accepted that the vegetative (or green) attributes can bedominant in Cabernet Sauvignon wines made from earlier-har-vested grapes, while fruity attributes are often more intense inwines made from later-picked fruit, but published evidence for thisis limited. There are surprisingly few sensory studies which haveinvestigated harvest timing, and where the above-mentioned trendin wine sensory characteristics has been reported, it is not neces-sarily consistent across multiple seasons (Heymann et al., 2013).It is therefore of interest to determine whether a ripening-specificsensory profile for Cabernet Sauvignon can be described, and

wine composition as the grapes ripen.Vegetative flavours in certain red wines are thought to arise pri-

marily from the potent grape-derived aroma compound isobutylmethoxypyrazine (IBMP) (Bindon, Varela, Kennedy, Holt, & Herde-rich, 2013a; de Boubee, Cumsille, Pons, & Dubourdieu, 2002; Ryona,Pan, & Sacks, 2009; Sala, Busto, Guasch, & Zamora, 2005). Vegetativeor green flavour attributes can also be derived from C6 volatiles suchas hexanal, hexanol, (E)-2-hexenal and (Z)-3-hexen-1-ol (Escudero,Campo, Fariña, Cacho, & Ferreira, 2007; Kalua & Boss, 2009, 2010).These C6 compounds can be present in the grape berry (Kalua &Boss, 2009, 2010), but primarily derive de novo from the degrada-tion of polyunsaturated fatty acids via the lipoxygenase pathwaywhen cell membranes are disrupted during crushing (Herraiz, Her-raiz, Reglero, Martinalvarez, & Cabezudo, 1990; Joslin & Ough,1978; Roufet, Bayonove, & Cordonnier, 1986). It has also been foundthat the compound dimethyl sulfide can give a cooked vegetablenote to some red wines (San-Juan, Ferreira, Cacho, & Escudero,2011). The contribution of C6-derived compounds to green

K. Bindon et al. / Food Chemistry 154 (2014) 90–101 91

attributes in wines is poorly understood, and is further complicated,since the corresponding esters of C6 alcohols are thought to contrib-ute fruity notes (Forde, Cox, Williams, & Boss, 2011).

The concentration of IBMP is known to decline during grape rip-ening (de Boubee, Van Leeuwen, & Dubourdieu, 2000; Scheiner,Vanden Heuvel, Pan, & Sacks, 2012). However, a close correlationbetween IBMP concentration and vegetative or green characters isnot always observed (Preston et al., 2008; Scheiner et al., 2012)and is likely due to interactions within the wine volatile matrix,so that other volatiles mask the effect on flavour and aroma. As al-ready highlighted, it is also possible that the attribute is caused bythe interaction of IBMP and C6 aldehydes and alcohols (Escuderoet al., 2007).

The suppression of vegetative characters by compounds givingfruity attributes has been reported (Hein, Ebeler, & Heymann,2009; King, Osidacz, Curtin, Bastian, & Francis, 2011). The contribu-tion of esters to fruity aroma attributes has been well demon-strated, and particularly, red berry, raspberry-like aroma has beenindicated as being related to proportionally higher concentrationsof the compounds ethyl butanoate, ethyl hexanoate, ethyl octano-ate, and ethyl 3-hydroxybutanoate, while black berry charactersmay be conferred by ethyl propanoate, ethyl 2-methylpropanoate,and ethyl 2-methylbutanoate (Escudero et al., 2007; Pineau, Barbe,Van Leeuwen, & Dubourdieu, 2009). The enhancement of fruityattributes is also reported to be caused by synergistic interactionof esters with low concentrations of C13-norisoprenoids and di-methyl sulfide (Escudero et al., 2007). A ripening-related decreasein both IBMP and C6 volatiles was recently reported in a series ofVitis vinifera L. cv. Cabernet Sauvignon wines, with concomitant in-creases in volatile esters (Bindon, Varela, Kennedy, Holt, & Herde-rich, 2013b; Bindon et al., 2013a). To further explore the effects ofharvest date and wine alcohol concentration, it was of interest tocompare these observed trends in wine composition with theresulting sensory perception of the wines.

Further to changes in volatile compounds, grape ripening cer-tainly confers changes to non-volatile wine composition, affectingmajor components such as sugars and acid concentration, as wellas secondary metabolites such as anthocyanins and tannins. Inour previous report (Bindon et al., 2013a) wine made from later-harvested grapes was characterised by higher concentrations ofglycerol and yeast-derived mannoprotein, but had lower grape-de-rived polysaccharides. These changes might be expected to con-tribute to wine in-mouth sensory properties, with polysaccharideconcentration having been shown to exert effects on the percep-tion of astringency and bitterness (Vidal et al., 2004a). Furthermore,Cabernet Sauvignon wine made from later-harvested grapes (Bin-don et al., 2013a) was shown to be characterised by higher skintannin concentrations, notably with a reduction in the relative con-tribution of grape seed tannin, resulting in a higher proportion ofgrape skin-derived tannin and mean degree of polymerisation(mDP). In addition, anthocyanin concentration in wine was en-hanced by ripeness, and as such, wine colour density and poly-meric (tannin-bound) colour increased. Similar observations interms of phenolic composition have recently been reported forsequentially-harvested Cabernet Franc wines, with later-harvestedwines having a higher tannin concentration and proportion of epi-gallocatechin, but not a higher mDP (Cadot, Caille, Samson, Bar-beau, & Cheynier, 2012). For the sensory attributes determinedby descriptive analysis in the Cabernet Franc study, later harvestdate was associated most strongly with higher perceived colourintensity, bitterness and astringency.

The question as to whether a target ‘optimal grape ripeness’ fora given grape variety can be defined is important in commercialwinemaking practice. This is because decisions around the timingof harvest are aimed at maximising positive attributes in the wine,minimising negative attributes, and optimising resources during

the season. For Cabernet Sauvignon, green characters in wine aro-ma or flavour are generally considered negatively by wine produc-ers. The commonly-held belief that delaying harvest date canreduce green grape-derived component IBMP is one of the factorswhich contribute to a trend toward higher alcohol content in Cab-ernet Sauvignon, through an increase in total soluble solids in thegrapes. In addition, market demands for strongly fruit-flavoured,intensely coloured wines, with desirable textural attributes, hasalso resulted in extended ripening periods for grapes. The evidencethat delaying harvest can significantly affect these attributes islimited, and for the most part anecdotal. In fact, for warm wine-growing regions, which generally have lower grape colour, and ashorter ripening phase, this practice may result only in higher alco-hol concentration without a significant increase in desirable com-pounds (Cozzolino, Cynkar, Dambergs, Gishen, & Smith, 2010). Theadded pressure of compressed growing seasons, warming climaticconditions, and water deficits associated with climate change maycompound the current world-wide industry trends toward wineswith higher alcohol content (Webb et al., 2012). A pertinent ques-tion, therefore, is whether the sensory attributes of wine madefrom grapes of different ripeness levels support the practice ofdelayed harvest. Perhaps even more important, is whether con-sumers in fact prefer the sensory attributes associated with late-harvest, higher alcohol wines.

In order to address these questions, the aim of the current studywas to determine the effect of harvest stage on the sensory attri-butes of wine, and how the sensory attributes relate to detailed winecompositional data, previously reported (Bindon et al., 2013a,2013b), and liking scores of consumers. The ultimate objective wasto better define optimal grape ripeness in Cabernet Sauvignon.

2. Materials and methods

2.1. Wine production and analysis

Experimental conditions outlining the production of triplicateVitis vinifera L. cv. Cabernet Sauvignon wines have been reportedpreviously (Bindon et al., 2013a). Briefly, grape samples were ob-tained from a commercial vineyard in the Langhorne Creek regionof South Australia, Australia at five different stages of ripeness(H1–H5), in 2010. Harvest dates were on the 16th (H1) and 23rdFebruary (H2), and on the 2nd (H3), 10th (H4) and 17th (H5)March. Grape batches of triplicate 50 kg were crushed and de-stemmed with the addition of 50 mg/L SO2. The pH of the mustswere adjusted to a target of 3.2 using tartaric acid. Total assimila-ble nitrogen was adjusted to 250 mg/L with diammonium phos-phate (DAP). Yeast (Saccharomyces cerevisiae PDM, Maurivin,Sydney, Australia) was inoculated at 200 mg/kg, and fermentationwas carried out on the skins for 7 days in a temperature-controlledroom at 15 �C, and plunged 20 times, twice daily. Thereafter, theferments were drained and pressed and fermentation was com-pleted to <1 g/L glucose and fructose. Following racking off grosslees, 60 mg/L SO2 was added, the wines were acid-adjusted to pH3.5 with tartaric acid, cold-stabilised, and thereafter racked off fin-ing lees. The final SO2 level was adjusted to a total of 80 mg/L, sinceno malolactic fermentation was performed. Wines were filteredthrough a 0.8 lm membrane, bottled under screw-cap and storedat 12–15 �C until chemical and sensory analysis, approximately10 months after bottling. Analysis of wine composition was per-formed as reported previously (Bindon et al., 2013a).

2.2. Sensory descriptive analysis

The sensory analysis was conducted in February–March 2011. Apanel of twelve assessors (four male, eight female) was convened

92 K. Bindon et al. / Food Chemistry 154 (2014) 90–101

for this study from the trained AWRI descriptive analysis panel, allof whom had extensive previous experience in wine descriptivestudies. A consensus-based descriptive methodology was used.Panellists attended three two-hour training sessions to generateand refine appropriate descriptors and their definitions for ratingin formal sessions. During these sessions the panellists assessedwines from the study which represented the full range of sensoryproperties. Wines were evaluated by appearance, aroma and pal-ate. For the second and third training sessions, standards for the ar-oma attributes and a palate standard for salty were presented anddiscussed, and relative ratings between pairs of samples were dis-cussed. A practice booth session was then held, where a subset ofthe wines were rated in duplicate under the same conditions asthose used in the formal sessions, except that constant presenta-tion order was applied, and palate standards for astringent andacidity were also presented. Panel performance was assessed fol-lowing this session and redundant or non-discriminating termsdiscussed with the panel. A final list of 3 appearance attributes, 9aroma attributes and 12 palate terms were developed. Details ofthe attributes selected and the reference standards are listed inTable 1.

Samples were assessed over three days of formal sessions, inwhich wines were presented to panellists in 30 mL aliquots in 3-di-git-coded, covered, ISO standard wine glasses at 22–24 �C. Assess-ment was performed in isolated booths under daylight lighting,with a randomised block design and a random presentation orderwithin each tray of samples across assessors. The assessors werepresented with five trays of three samples per tray per session,so that the 15 wines (five harvest dates � three fermentation rep-licates) were presented in a randomised order across the trays,with the same three samples presented within a tray for each

Table 1Descriptors, definitions and reference standards used in the sensory descriptive analysis.

Descriptors Definition

AppearanceOpacity How much the wine blocks the passage of light, intensity of colourPurple colour The intensity of purple colour in the wineRed colour The intensity of red colour in the wine

AromaOverall fruit

intensityThe overall intensity of fruit aroma of the wine includes plum, chestrawberry, raspberry and blackcurrant

Dark fruit The aroma of dark fruits including blackberry, black cherryRed fruit The aroma of red fruits includes strawberry, red cherry, raspberry

Vanilla The aroma of vanillaFresh green The aroma of fresh green stalks, leafy, fresh grass, tomato leaf, fresh

green chiliesCooked

vegetablesThe aroma of cooked vegetable water, or brine of canned vegetable

Earthy The mulch-like aroma associated with wet earth, forest floor, soil,Sewage/drain The aroma of rotting sewage, dirty drains, stinky waterPungent An irritating sharp sensation in the nose

PalateOverall fruit

flavourThe overall flavour of fruits, includes plum, cherry, blackberry, stra

Dark fruit The flavour of dark fruits, includes plum, blackberry, black cherryRed fruit The flavour of red fruits, includes strawberry, red cherry, raspberryVanilla The flavour of vanillaFresh green The flavour of fresh green stalks, leafy, green capsicum, fresh grassSalty The taste of sodium chlorideAcidity The intensity of acid taste perceived in the mouth or after expectoAstringency The drying and mouth-puckering sensation in the mouthViscosity The perception of the body, weight or thickness of the wine in the m

thin, high = thick, oily)Hotness The intensity of warmth, heat perceived in the mouth or after expeBitter The intensity of bitter taste perceived in the mouth or after expectFruit AT Fruit aftertaste, the lingering fruit flavour perceived in the mouth a

a In 40 mL base wine Yalumba Classic Dry Red 2009 vintage, 2L bag-in-box unless oth

judge. There was a 60 s break between samples on a tray wherepanellists rinsed with water, with the forced rest controlled bythe data acquisition software, and a minimum ten minute breakbetween trays. All samples were expectorated. The wines werepresented three times. The intensity of each attribute was ratedusing an unstructured 15 cm graphic rating line scale, with in-dented anchor points of ‘low’ and ‘high’ placed at 10% and 90%,respectively. Ratings from the scale were measured from 0 to 10.Data was acquired using Fizz sensory software (Version 2.46,Biosystemes, Couternon, France). Panel performance was assessedusing Senstools (OP&P, The Netherlands) and PanelCheck (Matf-orsk) software, to determine the degree of agreement with the pa-nel mean and degree of discrimination across samples forindividual assessors. All assessors were found to be performingto an acceptable standard according to limits based on long-stand-ing experience with panel performance data.

2.3. Consumer acceptance testing

The consumer test was carried out in Sydney, Australia in May2011 by 104 red wine consumers who matched the selection crite-ria of: drinking red wine at least once per week, buying bottled wine$10–$20 from time to time, age 18–65 equally split in three age cat-egories (18–35, 36–50 and 51–65), and an approximately equalnumber of males and females. Recruitment of the consumers wascompleted by a contract sensory research company in Sydney. Con-sumers went to a central location to evaluate five Cabernet Sauvi-gnon wines, each being a single fermentation replicate of eachharvest date, and were paid an incentive fee on completion. Uponarrival, consumers signed an informed consent form outlining thestudy and were then briefed on what they would be required to

Standarda

rry, blackberry,

1 Frozen blackberry (Creative Gourmet)1 Frozen strawberry and 2 frozen raspberries (CreativeGourmet)1/8 tsp Vanilla bean paste with vanilla seeds (Queen)

green capsicum and 10 � 1 cm Pieces of freshly cut grass and 1 � 1 cm2 squareof fresh green capsicum

s 15 mL Juice from canned vegetables (Edgell MixedVegetables)

potato skin, dusty 10 lL Geosmin (62 lg/L), 1 cm2 piece of fresh potato skin10 lg/L 2-Mercaptoethanol solution10 mL Ethanol

wberry, raspberry

7.5 g/L Table salt in filtered waterrating 2 g/L Tartaric acid in filtered water

0.5 g/L Aluminum sulfate in filtered waterouth (low = watery,

ctoratingoratingfter expectorating

erwise stated.

K. Bindon et al. / Food Chemistry 154 (2014) 90–101 93

do during the session. They were also given a demographic ques-tionnaire to complete. Five wines from the different harvest stageswere presented during a one hour session, one at a time at roomtemperature. Wines (30 mL) were presented according to a ran-domised block design in three-digit coded ISO wine tasting glassesat 22–24 �C. Samples were not expectorated. Consumers rated eachwine for overall liking on a nine-point hedonic category scale (‘dis-like extremely’ to ‘like extremely’). After the tasting, consumers an-swered several questions regarding wine usage and attitudes.Assessments were made on paper with a separate questionnairepresented for each wine. Respondents had a forced break of threeminutes between wines and were requested to rinse with water.The consumers’ detailed demographic information is provided inSupplementary information (S3), but briefly, there were 52% fe-males, with 13% of the consumers in each of the age categories18–25 and 26–35, 20% in age category 36–55, 26% in each of theage categories 46–55 and 56–65, and 1% aged 65 and above.

Table 2Mean scores for sensory attributes for which significant harvest treatment effectswere observed (attributes rated from 0 to 10).

H1 H2 H3 H4 H5 5% LSDa

Appearanceb

Opacity⁄ 1.76 2.47 3.27 3.64 4.85 0.28Red colour⁄ 2.51 2.27 2.32 2.06 2.09 0.27Purple colour⁄ 1.56 2.22 2.68 3.02 3.66 0.34

Aromab

Overall fruit⁄ 3.03 3.31 3.25 3.20 3.57 0.34Dark fruit⁄ 1.21 1.55 2.33 2.48 2.97 0.51Red fruit 2.21 2.18 1.73 1.07 1.04 0.31Fresh green 1.51 1.37 1.04 1.08 0.87 0.36Cooked vegetable 1.15 0.77 0.84 0.82 1.23 0.33Earthy⁄ 0.96 0.83 1.24 1.15 1.57 0.51Sewage/Drain⁄ 1.22 0.63 0.88 0.96 1.13 0.55Pungent 1.83 1.93 2.33 2.54 2.86 0.33

Palateb

Overall fruit 2.99 3.44 3.64 3.98 4.29 0.19Dark fruit⁄ 1.10 1.69 2.53 3.18 3.75 0.51Red fruit 2.32 2.20 1.95 1.24 1.14 0.34

2.4. Statistical analysis

Analysis of variance (ANOVA) was carried out using JMP 5.0.1a(SAS Institute, Cary, NC). For the sensory descriptive analysis data,the effects of assessor (Judge), harvest date (Harvest), fermentationreplicate (Wine Replicate) nested within harvest date, Judge⁄Har-vest, Judge⁄Wine Replicate nested within Harvest, and presenta-tion replicate nested within Harvest and Wine replicate wereassessed with Judge treated as a fixed effect. Following ANOVA,Fisher’s least significant difference (LSD) value was calculated(P = 0.05). Principal component analysis (PCA) was conducted onthe mean values averaged over panellists and replicates, usingthe correlation matrix. Partial least squares regression analysis(PLS) was conducted to relate the sensory data (y variables) withthe compositional data (x variables) for the 15 samples, usingPLS2, with all data standardised prior to analysis.

For the consumer liking data, a two factor ANOVA for consumerand wine was conducted. Internal preference mapping was carriedout on the mean liking response for the sample as well as the clus-ter mean data, using the sensory descriptive analysis data as sup-plementary variables. Agglomerative hierarchical cluster (AHC)analysis was performed on the centred and reduced data withthe dissimilarity measured by Euclidean distance and aggregationby Ward’s method. PLS was used to model consumer liking scoreswith sensory data. PLS was conducted using the standardised meandata of the sensory descriptive values and hedonic scores. L-PLSregression was used to relate sensory, consumer liking and socio-demographics data. The statistical software packages used for theanalyses were Panel Check (Version 1.3.2 Nofima), JMP 5.0.1a(SAS Institute Inc.), XLSTAT (Version 2010.3.05 Addinsoft), andThe Unscrambler X (Version 10.1, CAMO Software AS). For all PLSmodels, the optimal number of components for the models wasdetermined from inspection of residual variance explained by eachprincipal component (PC), models were generated using full leave-one-out cross validation and an uncertainty test applied to assessstatistical significance of the x-variables.

Vanilla 0.54 0.91 0.89 1.35 1.30 0.27Fresh green⁄ 1.95 1.86 1.63 1.49 1.40 0.49Acidity 4.49 3.75 4.07 3.71 4.14 0.27Astringency⁄ 2.64 2.58 2.74 2.51 3.08 0.42Viscosity⁄ 1.92 2.25 2.57 2.70 2.95 0.27Hotness 1.53 1.91 2.53 2.71 3.58 0.31Bitter 1.27 1.20 1.36 1.47 1.84 0.33Fruit aftertaste 1.34 1.73 1.73 2.39 2.34 0.26

a Fisher’s least significant difference (LSD) value was calculated following ANOVA(P = 0.05). Significant differences by harvest date were found between treatmentsfor all attributes except vanilla (aroma) and salty (palate) (data not shown).

b Where significant interaction effects of fermentation replicate nested withinharvest date occurred this is indicated alongside the attribute as ‘*’.

3. Results

3.1. Sensory descriptive analysis

A significant effect of harvest date for all wine sensory attri-butes, except vanilla aroma and salty taste, was found followingstatistical analysis by ANOVA (Table 2). For certain attributes, therewere significant harvest effects, but no wine replicate effect whichenabled harvest date means to be directly compared. However,there were significant fermentation replicate effects for many

attributes, namely opacity, red and purple colour, overall fruit, darkfruit, earthy, sewage aromas, dark fruit flavour, fresh green flavour,astringency and viscosity. This does not negate the harvest effectdifferences, but these should be considered when interpretingthe results of harvest date on wine sensory properties. Detail ofindividual means is provided as supporting information (Support-ing information S1 and S2).

Given the significant fermentation replicate effects, PCA wasperformed for the mean data for significant sensory attributes ofthe 15 wines, including all fermentation replicates (Fig. 1). The firsttwo components accounted for 75.8% of the variance in the dataset. The wines from the earlier harvest dates were more highlyrated for red fruit (aroma and palate), red colour and fresh green (ar-oma and palate), and had relatively low scores for overall fruit, va-nilla, fruit aftertaste, purple colour, viscosity, dark fruit, hotness,pungent and bitterness. Wines from the later harvest dates, by con-trast, were rated higher in dark fruit (aroma and palate), overall fruit(aroma and palate), hotness, pungent, opacity, bitter and earthy attri-butes. In general, there was a shift from H1 through to H5 alongPC1.

Separation of the wines along PC2 was primarily on the basis ofacidity, sewage/drain and cooked vegetable scores, with H1 and H5wines separated from the other harvest wines along PC2, beingrated higher for these attributes. Overall, the individual wine rep-licates for each harvest were clustered closely, with the exceptionof H4, where one replicate was rated much higher than the othersin opacity, purple colour, overall fruit, dark fruit, vanilla and earthy,and was also rated lower in sewage/drain (Supporting informationS1).

Regarding astringency, variation in this attribute is not shownclearly in Fig. 1 as indicated by the relatively short vector. The mainseparation of the samples on PC3 (10% of the variance) was on thebasis of this attribute (data not shown), with a similar rating

PC1 (62.8%)

PC2 (13.0%)

Fig. 1. Principal component (PC) analysis biplot of the mean scores of thesignificant (P < 0.05) sensory descriptive analysis data for the 15 wines (H1–H5:harvest 1–5, and their individual fermentation triplicates). Vectors for the sensoryattributes and the circle symbols for the wines are shown, calculated from scores of10 judges � 3 presentation replicates. AT, aftertaste; a, aroma; p, palate.

94 K. Bindon et al. / Food Chemistry 154 (2014) 90–101

observed for astringency across the harvest dates until the finaldate H5, where it was rated significantly higher. Values for astrin-gency ranged from 2.57–2.71 for all replicates of all harvest dates,with the exception of replicate 2 of H3, and the three replicates ofH5, which ranged from 3.05 to 3.20, Supporting information S2.The eigenvalues for higher order PCs were below 1, indicating lim-ited information is obtainable for these PCs.

3.2. Partial least squares regression modelling of wine sensoryattributes and compositional data

The sensory data reported here, and previously published winecompositional data (Bindon et al., 2013a, 2013b) for the wines fromdifferent harvest dates were analysed using PLS, with separatemodels developed for appearance and aroma terms (Fig. 2) andfor palate terms (Fig. 3). The approach taken was not to attemptto generate an optimal predictive model that could be applied tounknown samples, due to the experimental design being limitedto samples from one vineyard only. Rather, the aim was to assessassociations among the sensory and compositional variables. Ini-tially, models including all variables were generated, followed bysubsequent models excluding some measures. The rationale forexcluding chemical variables was related to the high degree of co-correlation of several of the non-volatile measures, notably theindividual acids and the phenolic variables. Chemical variables thatdid not differ significantly across the samples, or which were pres-ent at a concentration:sensory detection threshold ratio of less than0.1 (Ferreira, Lopez, & Cacho, 2000; Francis & Newton, 2005) werealso not included in the models. Separate models were developedfor appearance/aroma and palate as a combined model required alarger number of factors and was more difficult to visualise,although broadly the results of the regression analyses were simi-lar. Regression coefficients for each sensory attribute for both mod-els are provided as supporting documents (Supporting informationS3 and S4). For the appearance and aroma sensory attributes, anumber of non-volatile chemical variables were included in the fi-nal model. These were selected partly based on the above criteria,with the components glycerol, malic acid and tannin included asthey may have contributed to a ‘matrix effect’ due to their relatively

high concentrations. As such, their potential involvement in botharoma and appearance attributes could not be excluded.

Clear separation of wines by harvest date was observed in bothregression models, and most sensory attributes were well mod-elled by the chemistry data. For both PLS models the greatest sep-aration between harvest dates was along the horizontal (Factor 1)axis, with H4 separated from wines of the other harvest datesalong the vertical (Factor 2) axis.

The X and Y loadings plots (Figs. 2 and 3B) enable the visualisa-tion of the associations between sensory attributes and wine com-positional measures. The first two dimensions of the model inFig. 2B account for over 58% of the explained variance for the sen-sory data. In general, as presented above, the attributes red colourand red fruit aroma, as well as fresh green aroma, were rated higherin the earlier-harvest wines and the attributes opacity, purple col-our, dark fruit aroma, and pungent aroma, were rated higher forthe later-harvest wines. The descriptors opacity, purple colour, darkfruit, pungent, and red fruit were very well-modelled by the winecompositional data, as indicated by their position within the outerellipses in the loadings plot, with coefficients of determination forpredicted (calibration) versus measured values for the attributesopacity, purple colour, dark fruit, pungent, and red fruit in the PLSmodel being 0.82, 0.78, 0.76, 0.86 and 0.63, respectively. Red colourand fresh green were modelled moderately well, with coefficientsof determination (calibration) of 0.39 and 0.45 respectively.

Opacity and purple colour were strongly associated with totalanthocyanin, wine colour density and SO2-resistant pigments, allof which are likely to be significant direct contributors for theseattributes rather than simply co-correlating. The volatile com-pounds that were positively associated with dark fruit aroma in-cluded dimethyl sulfide, and multiple esters (Supportinginformation S3). Red fruit aroma, negatively correlated with thedark fruit attribute, was negatively associated with multiple esters,related to the lower ester concentration in earlier-harvested wines.IBMP and C6 alcohol concentrations were negatively associatedwith dark fruit aroma, but positively correlated with red fruit andfresh green aroma. The cooked vegetable, sewage/drain and earthy ar-oma attributes were not well modelled by the wine compositionaldata. For the pungent attribute there was a strong positive associa-tion with wine ethanol content.

The PLS regression analysis of significant palate sensory attri-butes and wine compositional data (Fig. 3) produced a similar sep-aration of wines by harvest date to that seen for appearance andaroma (Fig. 2). All of the palate attributes were well-predicted bythe PLS model, as indicated by R2 values (calibration) greater than0.74 for all descriptors except for the fresh green attribute whichwas somewhat less well-modelled by the chemical measures thanthe other descriptors, with an R2 (predicted vs measured, calibra-tion) of 0.53.

The attributes dark fruit flavour, overall fruit flavour and fruitaftertaste were associated with higher dimethyl sulfide. Red fruitand fresh green palate attributes were associated with a generallyhigher contribution of esters and higher concentrations of IBMPand C6 alcohols. A further observation was that vanilla flavour,which was not significantly different for the wines in terms of ar-oma (Table 2) was positively correlated with fruit aftertaste, andassociated with a higher alcohol, 2-methylbutanol.

As expected, the acidity attribute was associated most stronglywith pH and also titratable acidity. Both astringency and bitternesswere positively associated with total tannin concentration, skintannin concentration (% skin), and tannin mDP, with astringency alsorelated to titratable acidity. Bitterness was also negatively correlatedwith grape-derived polysaccharides, as indicated by the highregression coefficients of polysaccharide-derived galacturonic acid,arabinose, xylose, rhamnose and fucose, in the PLS model(Supporting information S4). The attribute viscosity had a strong

F1 (71%, 50%)

F2 16%, 8%)

DMSAlcohol

pH

TA

VA

Glycerol

MalicIBMP

Me thioAcTannin

Total Anth

SO2 Res PigWCD

EAEP

E2-MP

EBEH

2-MBA

3-MBA

Butanol

2-MB

Z-3-H-1-ol

Opacity

Red colour

Purple colour

Overall fruit

Dark fruit

Red fruitCooked vegetable

Earthy

Sewage/Drain

PungentF1 (71%, 50%)

F2 16%, 8%)

Fresh green

A

B

Fig. 2. Scores plot and X and Y loadings plot from PLS regression for appearance and aroma terms using selected chemical compositional data. (A) Scores plot for wines fromfive harvest dates; (B) X and Y loadings for PLS2 regression. X loadings (chemical variables) are shown in grey, Y loadings (sensory attributes) are shown in black.Abbreviations for chemistry measures are as follows: TA, titratable acidity; VA, volatile acidity (as acetic acid); Malic, malic acid; Total Anth, total anthocyanins; WCD, winecolour density, SO2 adjusted; SO2 Res Pig, SO2 resistant pigments; DMS, dimethyl sulfide; IBMP, isobutyl methoxypyrazine; Me thioAc, methyl thioacetate; EA, ethyl acetate;EP, ethyl propanoate; EB, ethyl butanoate; EH, ethyl hexanoate; 2-MBA, 2-methylbutyl acetate; 3-MBA, 3-methylbutyl acetate; E-2MP, ethyl-2 methyl propanoate; 2-MB, 2-methyl butanol; Z-3-H-1-ol, Z-3 hexen-1-ol.

K. Bindon et al. / Food Chemistry 154 (2014) 90–101 95

positive association with wine alcohol content and glycerol concen-tration, as well as polysaccharide-derived mannose (derived pri-marily from mannoprotein); and a strong negative relationshipwith malic acid and grape-derived polysaccharide. Hotness percep-tion was associated strongly with wine alcohol content, but was alsopositively associated with the concentration of glycerol, butanoland ethyl acetate.

3.3. Consumer acceptance

A single replicate of wine from each of the harvest dates was se-lected based on examination of the sensory descriptive data(Table S1 and S2). Those individual replicates that had higher rat-ings for the sewage/drain attribute were excluded, on the basis thatthis attribute may have been related to sulphur-derived off-flavourthrough fermentation variation rather than a true treatment effect.

Thus, one replicate from each treatment was excluded on this basis,except for H1 and H4 where two replicates were excluded. For Har-vests 2, 3 and 5, there were negligible differences between theremaining duplicates in any attribute, and accordingly one winereplicate was selected randomly. Subsequently, the wines weretasted by consumers, who rated their overall liking of each sampleon a standard nine point hedonic scale. Means for consumer likingratings are shown in Fig. 4A, where it is apparent that the likingscore increased from H1 to H2, then from H2 to H3. There was nosignificant change in consumer liking for the wines for H4 andH5, suggesting that H3 may represent an optimal harvest point inorder to combine high consumer acceptance with a lower alcoholconcentration in wine. The wines were generally well-liked (asindicated by scores of 6 or greater), which was somewhat surprisingconsidering that they had not undergone malolactic fermentation,nor had any oak treatment.

Alcohol

pH

TA

VA

Glycerol

Malic

DMS

IBMPMe thioAcTannin

Total Anth

WCD

SO2 Res Pig

mdp%Skin EA

EP

E2-MP

EB

E3-MB

EH

2-MBA

3-MBA

Butanol

Galacturonic

Rhamnose

Fucose

ArabinoseXylose

Mannose

Hexanol

Z-3-H-1-ol

Overall fruitDark fruit

Red fruit

Vanilla

Fresh Green

Acidity

Astringency

Viscosity

Hotness

Bitter

Fruit Aftertaste2-MB

A

B

Fig. 3. Scores plot and X and Y loadings plot from PLS regression for palate terms using chemical compositional data. (A) Scores plot for wines from five harvest dates; (B) Xand Y loadings for PLS2 regression. X loadings (chemical variables) are shown in grey, Y loadings (sensory attributes) are shown in black. Abbreviations for chemistrymeasures are as follows: TA, titratable acidity; VA, volatile acidity (as acetic acid); Malic, malic acid; Total Anth, total anthocyanins; WCD, wine colour density, SO2 adjusted;SO2 Res Pig, SO2 resistant pigments;% skin, percent skin tannin (of total tannin); mdp, mean degree of polymerisation (tannin); DMS, dimethyl sulfide; IBMP, isobutylmethoxypyrazine; Me thioAc, methyl thioacetate; EA, ethyl acetate; EP, ethyl propanoate; EB, ethyl butanoate; EH, ethyl hexanoate; 2-MBA, 2-methylbutyl acetate; 3-MBA, 3-methylbutyl acetate; E-2MP, ethyl-2 methyl propanoate; E3-MB, ethyl 3-methyl butanoate; 2-MB, 2-methyl butanoate; Z-3-H-1-ol, Z-3 hexen-1-ol.

96 K. Bindon et al. / Food Chemistry 154 (2014) 90–101

Three significantly different clusters were found within the con-sumer sample. The consumer liking data for each cluster and thetotal sample tested was related to the sensory data by internalpreference mapping (Fig. 4B) using the mean liking scores of eachcluster and of the total sample. It was apparent that the total sam-ple liking score was positively related to attributes such as purplecolour, dark fruit, vanilla, fruit aftertaste and viscosity, and inverselyrelated to acidity as well as the red fruit and fresh green attributes.The attributes cooked vegetable, sewage, earthy and astringencywere generally negatively-related to consumer liking of the winesfor the total sample. Cluster 1 (Fig. 4B), representing 44% of theconsumers, more closely reflected the total sample average thanthe other clusters, having a high degree of liking for wines H2,H3 and H4. The principal positive attribute associated with this

cluster was overall fruit, and the primary negative driver was acid-ity. Other attributes correlated to the liking scores of Cluster 1 werevanilla and purple colour, while cooked vegetable, acidity and sew-age/drain attributes were negatively correlated with liking. Cluster2, representing 33% of the consumer sample, disliked the winesfrom the first two harvest dates, and preferred wines H3, H4 andH5. Strong liking drivers for the second cluster were: overall fruit,fruit aftertaste, dark fruit and hotness, as well as bitterness, whilenegative attributes were the red fruit, fresh green attributes (botharoma and palate). The smallest group, Cluster 3, representing23% of consumers, showed less selectivity in their preferencechoice, having relatively high liking scores for H1 and H2 andnotably lower scores for H4, with positive attributes being acidity,red fruit and fresh green.

H1

H2

H3

H4

H5

Liking

Cluster 1 (44%)

Cluster 2 (33%)

Cluster 3 (23%)

PC

2 (2

4%)

PC1 (60%)

A

B

Fig. 4. Consumer response to wines from five harvest dates. (A) Consumer liking mean scores. (B) Principal component analysis biplot of the mean liking scores of threeidentified clusters and the total sample shown as vectors. The proportion of consumers in each cluster is shown in parentheses, and the five wines are shown as open symbols.The sensory attributes from the sensory descriptive analysis are also indicated, as well as certain wine compositional attributes, being supplementary variables. PC, principalcomponent; G+F, glucose and fructose; TA, titratable acidity; VA, volatile acidity.

K. Bindon et al. / Food Chemistry 154 (2014) 90–101 97

As commonly observed with segmentation analysis based onconsumers’ tasting preferences, socio-demographic variables (Sup-porting information S5) did not define the clusters very well, withthe exception that Cluster 1 consumers generally had higher in-comes, and Cluster 2 had a slightly lower proportion of consumerswith a University degree (data not shown). On the other hand,Cluster 3 had a somewhat higher percentage of consumers whohad a University degree, few had been drinking wine for longerthan 20 years, and a slightly larger proportion had been drinkingwine for less than 2 years (data not shown).

An additional L-PLS regression was performed to illustrate asso-ciations between socio-demographics, liking and sensory attributes(Supporting information S6), with individual consumer responses

modelled. Consumers with high variety-seeking patterns of choicefor food showed a preference toward H4, and consumers with high-er wine involvement scores preferred wines higher in vanilla, fruitaftertaste, and somewhat higher alcohol. Consumers with higherrisk-averseness to food preferred the more astringent, H5 wine.

4. Discussion

4.1. The effect of volatile compounds on aroma and flavour

Previously-published results on this sample set havedemonstrated significant changes in wine non-volatile compound

98 K. Bindon et al. / Food Chemistry 154 (2014) 90–101

concentration with grape maturity, and also an increase in yeastvolatile metabolites with the elevation of must sugar concentration(Bindon et al., 2013a, 2013b). Potential modulation of aroma can beinferred from changes in the ratio between concentration and sen-sory detection threshold of odourants, i.e. odour activity value(OAV) (Cacho & Ferreira, 2010; Escudero et al., 2007; Robinsonet al., 2009). In the present study many of the esters known to havea significant odour impact (Escudero et al., 2007; Sumby, Grbin, &Jiranek, 2010) were present at concentrations well above theirdetection threshold, with OAVs (based on sensory thresholds listedin Francis and Newton (2005)) the esters included in the PLS mod-els ranged from 2 (ethyl 2-methyl propanoate) to 102 (ethyl hexa-noate), with the exception of ethyl propanoate which had amaximal OAV of 0.2. The increase in total esters in wines contain-ing higher ethanol concentrations may simply be the result ofhigher yeast metabolic production, whereas the decline in otherimpact odorants, such as IBMP and C6 alcohols, reflects a changein grape-derived compounds.

The results of the present study showed that for both aroma andflavour the attributes fresh green, dark fruit and red fruit showedsubstantial changes with harvest date, being strongly correlatedto one another, and were defined by similar chemical variables,with a notable difference in the effect of esters. The dark fruit aro-ma attribute appeared to be driven primarily by increases in theconcentration of esters and also linked with higher dimethyl sul-fide in wines harvested at later dates, together with a lower con-centration of IMBP and C6-alcohols. The red fruit and fresh greenaroma attributes were inversely related to these compounds. Thepalate attributes were linked to similar compounds, but the effectof the esters was less marked, being negatively associated withdark fruit (palate) and positively associated with the red fruit (pal-ate) attribute. A further point of interest was that the C6 ester hexylacetate was found to decline in the wines with advancing maturity(Bindon et al., 2013a, 2013b) but was not found to be a significantcontributor to the PLS model developed for chemical and sensorydata for either aroma or palate, and was well below its reported ar-oma threshold. Other studies have highlighted the possible corre-lation of hexyl acetate to ‘fruity’ notes in wine (Forde et al.,2011) but the current dataset suggests a less significant role.

These observations highlight the potential role of the interac-tion of ‘green’ volatiles (IMBP and C6-alcohols) with increasing ordecreasing ester concentrations as an important consideration indefining red fruit and dark fruit aroma and flavour. As mentioned,the developed PLS models indicate associations between variables,but do not demonstrate causative relationships, and as such inter-pretation of the results should be made within the context of thecurrent literature. With this in mind, specific ethyl esters havebeen defined as responsible for conferring red fruit versus blackfruit (or dark fruit) aromas when observed across a number of redwine varieties (Pineau et al., 2009). In the present study there wereno distinct changes in the proportions of the different ethyl estersaccording to wine harvest date (Bindon et al., 2013a, 2013b). It isnoteworthy, however, that of all the fatty acid ethyl esters mea-sured in the current sample set, ethyl propanoate was the moststrongly correlated with dark fruit aroma, increasing 70% in thewine series from the earliest to the latest harvest. Since this wasobserved to be an important ethyl ester in defining black fruit char-acters (Pineau et al., 2009) it provides some confirmation that thiscompound may be a significant contributor to the dark fruit aromaand flavour attribute. However, it was below the threshold concen-trations noted in that study (Pineau et al., 2009). Therefore, tointerpret the results of the present study, ethyl propanoate mostlikely contributes to this sensory attribute in concert with otheresters.

A further noteworthy finding was the strong relationship be-tween dark fruit aroma and flavour and dimethyl sulfide. Dimethyl

sulfide was below its reported detection threshold in all the winesstudied, but work on the interactive effects between classes ofwine volatiles (Escudero et al., 2007) has revealed that this com-pound, which has a negative, cooked tomato or alliaceous odour athigh concentrations, may play a role in enhancing the fruity attri-butes of wines at lower concentrations, such as those observed inthe current study. A similar observation was made for the nori-soprenoids b-damascenone and b-ionone, which at low concentra-tions can enhance the fruity attributes of esters, but at higherconcentrations can contribute to a raisin note in wine. However,in the current study, b-damascenone concentration in wines,although present at an OAV of above 200, increased only margin-ally by harvest date, and b-ionone was only detected in H1 (OAVof 6) (Bindon et al., 2013a, 2013b). As such, these two C13-norisopr-enoids did not contribute significantly to the PLS models. In termsof interactive effects on aroma and flavour, it is of relevance to notethat increasing alcohol content in wines with advancing harvestdate may decrease the general volatility of aroma compounds inthe wine headspace, and may have dampened aroma and flavoursomewhat, potentially contributing to increased perception of pun-gency in the wines from later harvest dates (Robinson et al., 2009).

4.2. Non-volatiles and wine palate attributes

The results of the current study have shown that higher antho-cyanin concentration, wine colour density and SO2-resistant pig-ments in wine were associated with increased opacity (colourintensity) and purple colour. These attributes were also positivelycorrelated with overall tannin concentration, tannin mean degreeof polymerisation (mDP), and the percentage of tannin derivedfrom the grape skin (% prodelphinidin). Tannin forms complexeswith anthocyanin, resulting in a significant contribution to stablecoloured pigments (absorbance at 520 nm) at wine pH, the pH atwhich anthocyanins are weakly coloured. This phenomenon wasdemonstrated in our earlier work (Bindon et al., 2013a) with winesfrom later harvest dates having a higher proportion of 520 nm-absorbing polymeric material, assumed to be complexed with tan-nin. This may partly explain the relationship between tannin com-position and SO2-resistant pigments shown here, but it isimportant to highlight that increased wine tannin concentrationwas found only for H5, compared to the other treatments (Bindonet al., 2013a). Considering this, wines containing a higher anthocy-anin concentration could form SO2-resistant pigments via multiplepathways, irrespective of tannin concentration. Therefore, the cor-relation of tannin compositional features with the attributes opac-ity and purple colour is likely to be an indirect relationship.

Increasing concentrations of tannin generally results in an in-crease of the perception of astringency (Mercurio & Smith, 2008).With some variability observed between treatment replicates forthe perception of astringency, this attribute was scored signifi-cantly higher only in the wines from the final harvest date, andwas related to tannin concentration, the proportion of skin tannin(% skin) in the wines, and tannin mDP. As discussed previously,wine tannin concentration was similar for the H1–H4 treatments(Bindon et al., 2013a). Therefore, in interpreting the PLS model re-sults, the observation that astringency was positively associatedwith total tannin concentration, skin tannin concentration (% skin),and tannin mDP, most likely reflects compositional differences be-tween H5 and the other harvest treatments. On the other hand, animportant consideration was that wine titratable acidity and pHwere related to astringency, as well as being strongly associatedwith the acidity attribute. The wines had not been through malo-lactic fermentation and the titratable acidity level was accordinglyhigh, and the corresponding pH relatively low (Bindon et al.,2013a). It is known that at wine pH values at or below 3.5, a low-ering of pH may increase the perception of astringency, (Fontoin,

K. Bindon et al. / Food Chemistry 154 (2014) 90–101 99

Saucier, Teissedre, & Glories, 2008). Therefore, for the wine treat-ments containing similar tannin concentration (H1–H4), it appearsthat small fluctuations in titratable acidity, and the correspondingshift in pH were important in defining astringency perception.These fluctuations in titratable acidity and pH most likely reflectthe additions of tartaric acid and its subsequent stabilisation dur-ing the winemaking process, and may not necessarily reflect winecompositional differences related to harvest date. However, it isnoteworthy that wine malic acid concentration decreased withharvest date (Bindon et al., 2013a) and had a relatively high regres-sion coefficient in the PLS model for the acidity attribute, but notfor astringency (Supporting information S4). Therefore, in the winesmade from earlier harvest points, in particular H1, it is likely thatthe malic acid:titratable acidity ratio was important in increasingthe perception of acidity (Table 2, Fig. 3).

For bitterness, there was a similar association with tannin con-centration to that shown for astringency, but a significant role ofpolysaccharides was also apparent (Fig. 3, Supporting informationS4). Work done using model studies has indicated a negligible roleof tannin in eliciting bitterness (Vidal et al., 2004b), but has demon-strated the suppression of bitterness by wine proteoglycans (man-noproteins combined with arabinogalactan proteins). In thecurrent study, a negative correlation was found between bitternessand the molar percentage of polysaccharide-derived rhamnose andarabinose in wine, which suggests that a suppression effect mayhave been exerted by grape-derived polysaccharides. On the otherhand, mannose (as mol%) which is indicative of enhanced winemannoprotein content, was positively correlated with bitterness.Should polysaccharides play a role in the suppression or enhance-ment of bitterness, the current results indicate that the compositionof the proteoglycan fraction may be important in defining this re-sponse. To date, the authors were not aware of any study whichhas explored the effect of specific wine proteoglycan classes on bit-terness. Surprisingly, while some work has shown a positive asso-ciation between wine alcohol content and bitterness (Fontoinet al., 2008) the PLS model regression coefficients indicated ithad a negligible role in the current study (Supporting informationS4).

Amongst the mouthfeel attributes identified, hotness, as ex-pected, was simpler to define. Hotness was the most well-modelled(R2 0.98, predicted versus measured), with a very strong associa-tion with wine alcohol content, and a contribution of butanoland ethyl acetate. A study on Californian Cabernet Sauvignonwines produced from grapes between 20 and 30 �Brix over threevintages showed a similar correlation between hotness and winealcohol content, even when changes in the other sensory proper-ties of the wines were minor (Heymann et al., 2013). Additionally,a further treatment superimposed on grape ripeness (Heymannet al., 2013) was the manipulation of wine alcohol by chaptalisa-tion (sugar addition) or water addition to grape musts prior to fer-mentation; or the addition of water or alcohol to finished wine. Itwas found that increasing wine alcohol content consistently raisedthe perception of hotness and also of viscosity, but had minor orinconsistent effects on other wine sensory properties. The modelpresented in the current study showed similar increases in viscosityin wines produced at later harvest points, but the correlation withwine alcohol was less significant than for the hotness attribute. Itmight be expected that increasing glycerol concentration in winesproduced from later harvest dates could enhance viscosity, how-ever we note that this has been shown to exert no effect in whitewines (Gawel & Godden, 2008). On the other hand, a higherproportion of mannoprotein may be expected to contribute to en-hanced viscosity (Vidal et al., 2004b). However, total wine polysac-charide concentration decreased slightly from H1 to H5 (Bindonet al., 2013a). Therefore, should a causative effect of polysaccharideon viscosity exist, it would most likely be due to polysaccharide

compositional changes, namely increased polysaccharide molecu-lar mass, rather than a simple concentration effect. It is thereforelikely that the viscosity attribute results from an additive effect ofalcohol, glycerol, and mannoprotein.

It is evident that for almost all the sensory attributes, complexinteractive effects of the multiple volatile and non-volatile com-pounds exist, making the connection of compositional changes tospecific sensory attributes difficult. In a study such as this withfruit taken from a single vineyard studied over multiple harvestpoints, there are multiple, co-correlating variables. Nevertheless,some strong relationships between compositional and sensorydata were observed, many of which give greater weight to conclu-sions arising from previous model studies, suggesting causative ef-fects rather than simple correlations.

4.3. Implications of consumer perception of wine quality relating toharvest date and wine alcohol content

A recent publication has shown a strong relationship betweenwine colour, tannin and polymeric pigments with enhanced per-ception of overall wine ‘quality’, as scored by an expert sensory pa-nel (Ristic, Bindon, Francis, Herderich, & Iland, 2010). Similarly, astudy on commercial red wines showed that winemaker’s qualityscores related strongly to wines with higher alcohol and astrin-gency (Lattey, Bramley, & Francis, 2010). In the last decade, thesearch for wines with increased colour and enhanced palate full-ness has driven the alcohol content in commercial wines to highervalues. A key question in the present study was whether delayingharvest results in increased consumer acceptance.

The results of the consumer study showed a clear outcome inwhich wines with lower alcohol concentration, 12% v/v to 13% v/v, were the least preferred, with liking scores increased for wineswith intermediate alcohol levels, after which no further increasesin preference was observed. Although the current study representsa narrow sample set in a single vineyard and season, thus limitinggeneralisation, it is noteworthy that the wines made from fruit inthe current study reached an ‘optimal ripeness’ point in terms ofconsumer preference, before wine compositional factors currentlythought to be positive for wine ‘quality’ (such as colour, tannin andesters) were at a maximum. Importantly, wine alcohol content it-self may contribute significantly to this trend in preference, withhigher concentrations of alcohol moderating otherwise positivewine characteristics, and eliciting hotness. Extrapolating the resultsto other red grape varieties and growing regions is problematic,especially in warmer, drier regions which may have a compressedripening phase, and where grape and wine colour levels are low(Cozzolino et al., 2010). Nevertheless for a given variety, grape‘optimal ripeness’ (as it relates to the corresponding wine sensoryproperties), may not necessarily track with measurable increasesin grape sugar and colour.

Aside from the goal of optimising perceived wine quality forconsumers, the global perspective on alcohol overuse and itsresulting negative social and health effects has caused mountingpressure for beverage industries to reduce alcohol levels, togetherwith changes in tax legislation around alcohol (Roerecke & Rehm,2012). While moderate wine consumption is generally perceivedto have health benefits (Stockley, Teissedre, Boban, Di Lorenzo,& Restani, 2012), the wine industry is not excluded from thisinternational trend toward reducing alcohol consumption. In lightof the current results, earlier harvest presents a relatively simplesolution for the wine industry to address questions of increasedwine alcohol, especially when it is observed that delaying harvestproduced no gain in terms of overall consumer acceptance. Con-sidering the average liking of the total consumer sample, the attri-butes dark fruit, overall fruit flavour, vanilla and viscosity, whichincreased with grape maturity, were positively related to overall

100 K. Bindon et al. / Food Chemistry 154 (2014) 90–101

liking, while acidity, fresh green and red fruit were negative driversof consumer liking. In addition, when dividing consumers intosegments, it is clear that for two of the clusters, wine attributesthat increased with grape maturity (opacity, purple colour, darkfruit, fruit aftertaste, pungency and hotness) were preferred, whilewines with higher levels of acidity, red fruit and fresh green char-acteristics related to earlier-harvest wines were not well accepted.Even so, within these clusters, H5 was not preferred over H4. Thepreference of a small group of consumers for wines producedfrom earlier harvest points, having with lower hotness, lower darkfruit and astringency/bitterness, who were more likely to be new towine, shows that years of wine consumption experience can be astrong predictor of red wine preference, as indicated in earlierstudies (Lattey et al., 2010; Williamson, Robichaud, & Francis,2012).

5. Conclusions

The current study has provided a clear picture of changes in keysensory attributes of Cabernet Sauvignon wines produced fromgrapes harvested at different stages of ripeness. A decrease in green(vegetative) attributes was found in wines made from later harvestdates, with a clear shift from red fruit to dark fruit attributes. Thecorrelation of the red fruit character with high IBMP concentrationand higher levels of C6-volatiles was of interest, as well as a link be-tween dimethyl sulfide (albeit below sensory detection threshold)with the dark berry attribute. The suggested influence of grape-de-rived polysaccharides in reducing bitterness is worthy of furtherstudy. As expected, purple colour and viscosity increased in winesproduced from later-picked grapes. This occurred in conjunctionwith greater wine hotness, pungency and bitterness, which mayhave contributed to decreases in the acceptance of wines by con-sumers. The current study, together with previously published data(Bindon et al., 2013a), confirmed a good correlation exists betweenanalytical measures of grape colour with wine colour density; andthat these measures relate closely to the perceived wine colour.However, based on the consumer study, measures of phenolicmaturity did not necessarily track with a targeted ‘optimal ripe-ness’ for wine quality in commercial wine production, as definedby consumer preference. While analytical measures of grape phen-olics, such as grape colour and tannin concentration, are importantas indicators of wine composition and style (Ristic et al., 2010),delaying harvest to achieve higher phenolic concentrations maysimply result in hotter, more pungent wines.

Acknowledgements

The authors would like to thank Pernod Ricard Australia (Orlan-do Wines) for the donation of grape samples. We particularly thankDr. Mike McCarthy of the South Australian Research and Develop-ment Institute (SARDI) for access to field trial data. The ARC Centreof Excellence in Plant Cell Walls (University of Melbourne) isacknowledged for the use of their facilities and for technical advicein polysaccharide analysis. Mark Stevens (Sensory Insights, NSW)is acknowledged for the consumer test administration. We wouldlike to thank AWRI staff for their contributions to winemakingand sample analysis: Gemma West, Belinda Bramley, Stella Kas-sara, Fiona Brooks, Dimitra Capone, Natoiya Lloyd, Richard Gawel,the AWRI sensory panellists, the AWRI Commercial Services labo-ratory and the AWRI Trace Laboratory. The Australian WineResearch Institute, a member of the Wine Innovation Cluster inAdelaide, is supported by Australian grape growers and winemak-ers through their investment body, the Grape and Wine Researchand Development Corporation, with matching funds from theAustralian Government.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, inthe online version, at http://dx.doi.org/10.1016/j.foodchem.2013.12.099.

References

Bindon, K., Varela, C., Kennedy, J., Holt, H., & Herderich, M. (2013a). Relationshipsbetween harvest time and wine composition in Vitis vinifera L. cv. CabernetSauvignon 1. Grape and wine chemistry. Food Chemistry, 138, 1696–1705.

Bindon, K., Varela, C., Kennedy, J., Holt, H., & Herderich, M. (2013b). Erratum torelationships between harvest time and wine composition in Vitis vinifera L. cv.Cabernet Sauvignon 1. Grape and wine chemistry (vol 138, pg 1696, 2013). FoodChemistry, 141, 147.

Cacho, J., & Ferreira, V. (2010). The aroma of wine. In Y. H. Hui, F. Chen, & L. M. Nollet(Eds.), Handbook of fruit and vegetable flavors (pp. 303–317). John Wiley andSons.

Cadot, Y., Caille, S., Samson, A., Barbeau, G., & Cheynier, V. (2012). Sensoryrepresentation of typicality of Cabernet franc wines related to phenoliccomposition: Impact of ripening stage and maceration time. Analytica ChimicaActa, 732, 91–99.

Cozzolino, D., Cynkar, W. U., Dambergs, R. G., Gishen, M., & Smith, P. (2010). Grape(Vitis vinifera) compositional data spanning ten successive vintages in thecontext of abiotic growing parameters. Agriculture Ecosystems & Environment,139, 565–570.

de Boubee, D. R., Cumsille, A. M., Pons, M., & Dubourdieu, D. (2002). Location of 2-methoxy-3-isobutylpyrazine in Cabernet Sauvignon grape bunches and itsextractability during vinification. American Journal of Enology and Viticulture, 53,1–5.

de Boubee, D. R., Van Leeuwen, C., & Dubourdieu, D. (2000). Organoleptic impact of2-methoxy-3-isobutylpyrazine on red Bordeaux and Loire wines. Effect ofenvironmental conditions on concentrations in grapes during ripening. Journalof Agricultural and Food Chemistry, 48, 4830–4834.

Escudero, A., Campo, E., Fariña, L., Cacho, J., & Ferreira, V. (2007). Analyticalcharacterization of the aroma of five premium red wines. Insights into the roleof odor families and the concept of fruitiness of wines. Journal of Agricultural andFood Chemistry, 55, 4501–4510.

Ferreira, V., Lopez, R., & Cacho, J. F. (2000). Quantitative determination of theodorants of young red wines from different grape varieties. Journal of the Scienceof Food and Agriculture, 80, 1659–1667.

Fontoin, H., Saucier, C., Teissedre, P.-L., & Glories, Y. (2008). Effect of pH, ethanol andacidity on astringency and bitterness of grape seed tannin oligomers in modelwine solution. Food Quality and Preference, 19, 286–291.

Forde, C. G., Cox, A., Williams, E. R., & Boss, P. K. (2011). Associations between thesensory attributes and volatile composition of Cabernet Sauvignon wines andthe volatile composition of the grapes used for their production. Journal ofAgricultural and Food Chemistry, 59, 2573–2583.

Francis, I. L., & Newton, J. L. (2005). Determining wine aroma from compositionaldata. Australian Journal of Grape and Wine Research, 11, 114–126.

Gawel, R., & Godden, P. W. (2008). Evaluation of the consistency of wine qualityassessments from expert wine tasters. Australian Journal of Grape and WineResearch, 14, 1–8.

Hein, K., Ebeler, S. E., & Heymann, H. (2009). Perception of fruity and vegetativearomas in red wine. Journal of Sensory Studies, 24, 441–455.

Herraiz, T., Herraiz, M., Reglero, G., Martinalvarez, P. J., & Cabezudo, M. D. (1990).Changes in the composition of alcohols and aldehydes of C-6 chain-lengthduring the alcoholic fermentation of grape must. Journal of Agricultural and FoodChemistry, 38, 969–972.

Heymann, H., LiCalzi, M., Conversano, M. R., Bauer, A., Skogerson, K., & Matthews, M.(2013). Effects of extended grape ripening with or without must and winealcohol manipulations on Cabernet Sauvignon wine sensory characteristics.South African Journal of Enology and Viticulture, 34, 86–99.

Heymann, H., & Noble, A. C. (1987). Descriptive analysis of commercial Cabernet-Sauvignon wines from California. American Journal of Enology and Viticulture, 38,41–44.

Joslin, W. S., & Ough, C. S. (1978). Cause and fate of certain C6 compounds formedenzymatically in macerated grape leaves during harvest and wine fermentation.American Journal of Enology and Viticulture, 29, 11–17.

Kalua, C. M., & Boss, P. K. (2009). Evolution of volatile compounds during thedevelopment of Cabernet Sauvignon grapes (Vitis vinifera L.). Journal ofAgricultural and Food Chemistry, 57, 3818–3830.

Kalua, C. M., & Boss, P. K. (2010). Comparison of major volatile compounds fromRiesling and Cabernet Sauvignon grapes (Vitis vinifera L.) from fruitset toharvest. Australian Journal of Grape and Wine Research, 16, 337–348.

King, E. S., Osidacz, P., Curtin, C., Bastian, S. E. P., & Francis, I. L. (2011). Assessingdesirable levels of sensory properties in Sauvignon Blanc wines – consumerpreferences and contribution of key aroma compounds. Australian Journal ofGrape and Wine Research, 17, 169–180.

Lattey, K. A., Bramley, B. R., & Francis, I. L. (2010). Consumer acceptability, sensoryproperties and expert quality judgements of Australian Cabernet Sauvignon andShiraz wines. Australian Journal of Grape and Wine Research, 16, 189–202.

Mercurio, M. D., & Smith, P. A. (2008). Tannin quantification in red grapes and wine:Comparison of polysaccharide- and protein-based tannin precipitation

K. Bindon et al. / Food Chemistry 154 (2014) 90–101 101

techniques and their ability to model wine astringency. Journal of Agriculturaland Food Chemistry, 56, 5528–5537.

Pineau, B., Barbe, J.-C., Van Leeuwen, C., & Dubourdieu, D. (2009). Examples ofperceptive interactions involved in specific ‘‘Red-’’ and ‘‘Black-berry’’ Aromas inRed wines. Journal of Agricultural and Food Chemistry, 57, 3702–3708.

Preston, L. D., Block, D. E., Heymann, H., Soleas, G., Noble, A. C., & Ebeler, S. E. (2008).Defining vegetal aromas in Cabernet Sauvignon using sensory and chemicalevaluations. American Journal of Enology and Viticulture, 59, 137–145.

Ristic, R., Bindon, K., Francis, L. I., Herderich, M. J., & Iland, P. G. (2010). Flavonoidsand C-13-norisoprenoids in Vitis vinifera L. cv. Shiraz: relationships betweengrape and wine composition, wine colour and wine sensory properties.Australian Journal of Grape and Wine Research, 16, 369–388.

Robinson, A. L., Ebeler, S. E., Heymann, H., Boss, P. K., Solomon, P. S., & Trengove, R.D. (2009). Interactions between wine volatile compounds and grape and winematrix components influence aroma compound headspace partitioning. Journalof Agricultural and Food Chemistry, 57, 10313–10322.

Roerecke, M., & Rehm, J. (2012). Alcohol intake revisited: Risks and benefits. CurrentAtherosclerosis Reports, 14, 556–562.

Roufet, M., Bayonove, C. L., & Cordonnier, R. E. (1986). Changes in fatty-acids fromgrape lipidic fractions during crushing exposed to air. American Journal ofEnology and Viticulture, 37, 202–205.

Ryona, I., Pan, B. S., & Sacks, G. L. (2009). Rapid measurement of 3-alkyl-2-methoxypyrazine content of winegrapes to predict levels in resultant wines.Journal of Agricultural and Food Chemistry, 57, 8250–8257.

Sala, C., Busto, O., Guasch, J., & Zamora, F. (2005). Contents of 3-alkyl-2-methoxypyrazines in musts and wines from Vitis vinifera variety CabernetSauvignon: Influence of irrigation and plantation density. Journal of the Scienceof Food and Agriculture, 85, 1131–1136.

San-Juan, F., Ferreira, V., Cacho, J., & Escudero, A. (2011). Quality and aromaticsensory descriptors (mainly fresh and dry fruit character) of Spanish red winescan be predicted from their aroma-active chemical composition. Journal ofAgricultural and Food Chemistry, 59, 7916–7924.

Scheiner, J. J., Vanden Heuvel, J. E., Pan, B., & Sacks, G. L. (2012). Modeling impacts ofviticultural and environmental factors on 3-isobutyl-2-methoxypyrazine incabernet franc grapes. American Journal of Enology and Viticulture, 63, 94–105.

Stockley, C., Teissedre, P.-L., Boban, M., Di Lorenzo, C., & Restani, P. (2012).Bioavailability of wine-derived phenolic compounds in humans: A review. Food& Function, 3, 995–1007.

Sumby, K. M., Grbin, P. R., & Jiranek, V. (2010). Microbial modulation of aromaticesters in wine: Current knowledge and future prospects. Food Chemistry, 121,1–16.

Vidal, S., Courcoux, P., Francis, L., Kwiatkowski, M., Gawel, R., Williams, P., et al.(2004a). Use of an experimental design approach for evaluation of key winecomponents on mouth-feel perception. Food Quality and Preference, 15,209–217.

Vidal, S., Francis, L., Williams, P., Kwiatkowski, M., Gawel, R., Cheynier, W., et al.(2004b). The mouth-feel properties of polysaccharides and anthocyanins in awine like medium. Food Chemistry, 85, 519–525.

Webb, L. B., Whetton, P. H., Bhend, J., Darbyshire, R., Briggs, P. R., & Barlow, E. W. R.(2012). Earlier wine-grape ripening driven by climatic warming and drying andmanagement practices. Nature Climate Change, 2, 259–264.

Williamson, P. O., Robichaud, J., & Francis, I. L. (2012). Comparison of Chinese andAustralian consumers’ liking responses for red wines. Australian Journal of Grapeand Wine Research, 18, 256–267.