Abstract
Three monovarietal wines (Cabernet Sauvignon, Merlot, and Cabernet franc) were used to produce 11 two-wine blends and four three-wine blends to study the changes in the sensory perception of the blends using descriptive analysis. In addition, chemical analyses (solid-phase microextraction and gas chromatography–mass spectrometry, polyphenol composition, titratable and volatile acidity, and pH) were performed to evaluate whether the sensory perception of the wine blends was matched by chemical parameters. A comparison of all data sets was performed with multifactor analysis and revealed that the overall results of both sensory and chemical analyses were very similar to those obtained from the sensory analysis alone, but none of the chemical analyses provided the same results as descriptive analysis.
One of the oldest winemaking techniques—blending of wines—is a commonly used but not well-studied area in winemaking (Amerine et al. 1980). It is generally thought that wine blending increases the quality of the wine, regardless of the definition of quality. Wines are blended for several reasons, including increasing complexity, maintaining a consistent character and quality, and developing winery-typical wines. The most common wine blends are mixtures of different varieties (e.g., wines from Bordeaux); however, wine blends can also be produced by mixing wines from different vineyard blocks, vintages, or regions. Another blending technique is co-winemaking, which is the comaceration and co-fermentation of different grape varieties (Lorenzo et al. 2005, 2008, Garcia-Marino et al. 2010, Gomez Garcia-Carpintero et al. 2010). The latter authors performed a descriptive analysis (DA) on co-winemaking blends made with three red Spanish varieties cultivated in the La Mancha region in Spain (Cencibel, Bobal, and Moravia Agria, producing 50-50 and 33-33-33 blends) (Gomez Garcia-Carpintero et al. 2010). All blends were ranked higher in the evaluated sensory attributes than the monovarietals and the three-wine blend was the most complex, showing some synergistic effects in the measured sensory attributes.
For wine blends produced by mixing completely fermented wines (coupage), most studies have been conducted on changes in color and polyphenol composition with some sensory analyses. One study examined Tempranillo–Graciano blends using binary mixtures and monitored the color and anthocyanin changes immediately after, 24 hours after, and 100 days after the mixing process (Escudero-Gilete and et al. 2010). In the blends, darker and higher color values and higher color stability were found with increasing ratios of Graciano. In all blends, a rather constant mixing effect on the color was found independent of the blend ratios. Other studies examined blends of Tempranillo with 10% or 25% of Graciano or Cabernet Sauvignon after storage for 4, 5, 9, 16.5, and 23 months in the bottle (Monagas et al. 2006a, 2006b). Color differences were detectable by eye for Tempranillo with 10% of either Graciano or Cabernet Sauvignon. A sensory trial was conducted for visual, smell, taste, and taste-smell attributes for the 23-month aged samples, with higher overall ratings for the blends compared to the base wines. Adding wine made from Graciano grapes improved especially the color ratings, while the addition of Cabernet Sauvignon resulted in better color, aromatic complexity, and flavor.
The effect of wine blending on perceived complexity was studied using 34 pairs of similar wines to make 50-50 blends of each pair, which were then scored for quality by a sensory panel in two sessions (Singleton and Ough 1962). All 34 blends were rated higher than the low-scoring wine of the pair on its own. In seven cases the blends scored higher than the highest scored monovarietal base wine of that blend. According to the authors, the higher scoring resulted in some cases from an increase in perceived complexity.
A recent study optimized wine blends based on overall consumer liking scores using three red wine varieties (Cabernet Sauvignon, Merlot, and Zinfandel) (Dooley et al. 2011). The three monovarietal base wines and seven wine blends (three two-wine blends) were both evaluated by a DA panel and a consumer panel. Based on the liking scores, three optimized wine blends were then formulated based on consumer segmentation and subsequently reassessed by consumers.
Despite the studies noted above, there is no thorough sensory evaluation of wine blends in comparison to their respective base wines. Our goal was to add to the knowledge about the sensory effects of wine blending. Three monovarietal base wines—Cabernet Sauvignon (S), Merlot (M), and Cabernet franc (F)—were used for the production of two- and three-wine blends with varying ratios to study the underlying phenomena of changed sensory perception of blends. While chemical components such as ethanol or acids usually follow linear relationships in mixtures, we were especially interested in possible nonaveraging effects on the sensory perception, as aroma and flavor compounds have been shown to follow nonlinear behavior in mixtures (Ferrier and Block 2001, Lawless and Heymann 2010).
Previous studies on wine blends showed that polyphenols and color are affected by the blending process (Escudero-Gilete et al. 2010, Monagas et al. 2006a, 2006b, 2007). In the present work, changes in anthocyanins and total phenols were measured spectrometrically using the Harbertson-Adams and Folin-Ciocalteu assays, respectively (Harbertson et al. 2003, Singleton and Rossi 1965, Zoecklein et al. 1995). We also expected changes in the volatile pattern created by the wine-blending process, as previous studies showed that volatile compounds in wine, including aroma compounds, are dramatically influenced by changes in matrix components such as alcohol content (Robinson et al. 2009), and thus included headspace solid-phase microextraction gas chromatography–mass spectrometry (HS-SPME-GC–MS) measurements. HS-SPME coupled to GC–MS is extensively used to study volatile compounds in wines, including esters, acetates, alcohols, lactones, acids, terpenes, and aldehydes (Howard et al. 2005, Robinson et al. 2011, Lee et al. 2011, Polaskova et al. 2008, Francioli et al. 2003) as well as taints such as 2,4,6-trichloroanisole (Fischer and Fischer 1997).
Materials and Methods
Materials.
Three monovarietal wines (all vintage 2009) were used for the blending experiments: Cabernet Sauvignon (S), Merlot (M), and Cabernet franc (F). The Cabernet Sauvignon and the Merlot grapes were 100% from the Napa Valley AVA while the Cabernet franc grapes were 100% from the Alexander Valley AVA and were donated to UC Davis by two Californian wineries. All three base wines were dry (residual sugar: S = 0.02%; M = 0.01%; F = 0.07%) and alcohol contents ranged from 14.4% to 15.3% (S = 14.8%, M = 15.3%, F = 14.4%), as determined by ebulliometer (Ough and Amerine 1988, Zoecklein et al. 1995). The wines had been aged in oak barrels for ~9 months.
The monovarietal base wines were transported to the UC Davis winery in stainless-steel kegs (Cabernet Sauvignon and Merlot 208.2 L, Cabernet franc 113.6 L. Since the densities of the three wines were essentially the same (S = 0.9877 g/cm3; M = 0.9866 g/cm3; F = 0.9863 g/cm3), the wines were blended gravimetrically to avoid as much oxygen pick-up as possible. The wine blending was performed one week after the wines arrived at the winery. During that period wines were stored in the kegs at 10°C. Approximately 15.2 ± 0.05 L of each wine blend was produced. Wines were immediately bottled in dark green glass bottles with roll-on tamper-evident screwcaps (Saran Tin liner) and stored at 10°C until further use. pH levels ranged between 3.73 and 3.63 (S = 3.73, M = 3.63, F = 3.77). Dissolved oxygen levels of the base wines were measured with a conventional dipping probe (Orion 5 Star Plus, Thermo Scientific, Beverly, MA) prior to bottling in the kegs and were well below 1 mg/L (S = 0.27 mg/L, M = 0.62 mg/L, F = 0.45 mg/L). All produced blends were measured again after the blending prior to bottling and were <1 mg/L; the highest value was for the final blend (18) at 0.93 mg/L.
In total, 18 wines were analyzed, which included three monovarietal base wines, 11 two-wine blends, four three-wine blends with varying ratios of each base wine (Table 1). Wine 18 was blended by a winemaker after tasting trials so that one sample reflected a winemaker’s perspective on wine blending. The number of wine blends was limited to 15 and the more common Cabernet Sauvignon- and Merlot-based blends were included.
Sensory analysis.
A generic descriptive analysis (DA) was conducted on the 18 wines approximately three months after bottling using the methodology described in Lawless and Heymann (2010). The sensory panel consisted of 14 people (four females, 10 males, ages 24 to 35 years) who were all UC Davis students from the Food Science and Technology or Viticulture and Enology departments. All panelists had previous experience with DA. Panelists completed six one-hour training sessions over a period of two weeks, where they were exposed to subsets of the wines and asked to generate attributes to describe differences among the wines. Each wine was blindly presented twice during these training sessions. Aroma, taste, and mouthfeel attributes were generated and reference standards for these attributes were provided to anchor and obtain consensus on the attributes. In total, 23 attributes and their reference standards were chosen by the panel and used in the evaluation (Table 2).
All judges completed nine wine evaluations, tasting six wines in each session. Wines were served in a randomized William Latin Square block design to control for possible carry-over effects. The wines were evaluated in individual temperature-controlled tasting booths with positive air flow (Lawless and Heymann 2010). Water and unsalted crackers were provided for palate cleansing. Prior to each assessment, session panelists smelled the aroma standards to refresh their memories. To decrease fatigue, there was a 1-min break between each wine and an additional 3-min break between the third and fourth wine. During each break, panelists chewed on a cracker and then rinsed their mouths with water to clean their palates. The judges rated the aroma, taste, and mouthfeel attributes on a 10-cm unstructured linescale anchored from “low” to “high” provided by FIZZ sensory analysis software (ver. 2.54A; Biosystèmes, Couternon, France). The wines (25 mL aliquots) were presented in clear pear-shaped ISO glasses (ISO 3591:1977) labeled with three-digit random number codes under red light to prevent color bias. All judges rated each wine in triplicate during the three-week evaluation. Judges did not know how many samples or replicates they were evaluating.
HS-SPME-GC-MS analysis.
The volatile patterns of the 18 wines were measured by headspace solid-phase microextraction coupled to gas chromatography–mass spectroscopy (HS-SPME-GC–MS) using a MPS autosampler (Gerstel, Baltimore, MD) attached to an Agilent 7890 GC with an 5973N MS (Agilent, Santa Clara, CA). Extraction of the volatiles was optimized carrying out tests with two different SPME fibers as in previous studies (Robinson et al. 2009, Chapman et al. 2004) for wine volatiles (100 μm PDMS and 2 cm StableFlex DVB/Car/PDMS 50/30 μm; Supelco, St. Louis, MO) as well as varying extraction times (15 and 30 min) and temperatures (30 and 40°C), NaCl additions (0.5 and 3 g), and sample volume (2, 5, and 10 mL). Blank runs after sample runs using a 10-min fiber desorption time resulted in no sample carry-over. The optimized method facilitated a volatile extraction for 30 min at 40°C with a 2 cm DVB/Car/PDMS 50/30 μm SPME fiber (Supelco) after 5 min thermostatting followed by desorption for 10 min in the 250°C hot GC inlet equipped with a 0.7 mm i.d. SPME inlet liner (Supelco) in splitless mode after 1 min. Separation of the volatiles was performed on an Agilent DB-WAX capillary GC column (30 m × 0.25 mm × 0.25 μm), applying an oven program starting at 40°C, held for 5 min, and ramped with 5°C/min to 240°C, with another hold for 10 min. A constant, ultrapure helium carrier flow of 1 mL/min was used. The MS transfer line was held at 240°C. A 2-min solvent delay was used to protect the MS from excess ethanol. The MS was operated in scan mode from 50 to 350 m/z using an electron ionization of 70 eV. All samples and standards were measured five times. Areas of 60 selected compounds were obtained in extracted ion mode using the base peak or characteristic m/z for functional classes (Agilent MSD Chemstation software; ver. E.02.00.493) and analyzed for statistical significant differences among the wine blends.
Compound identification was performed by retention time and mass spectrum matching with authentic reference standards when available; otherwise, compounds were tentatively identified by the mass spectrum provided by NIST 05 mass spectral library (ver. 2005; Scientific Instrument Services, Ringoes, NJ). All used reference compounds (min. purity 80%) were from Aldrich (St. Louis, MO) with the exception of heptyl acetate (Pfalz & Bauer, Waterbury, CT), 1-dodecanol and hexanoic acid (Acros, Pittsburgh, PA), 1-undecanol, 2-methyl-1-butanol, and 2-phenyl ethyl alcohol (Sigma, St. Louis, MO), 1-octanol (Fisher Scientific, Pittsburgh, PA), 2-methyl-1-propanol (Fluka, St. Louis, MO), 2-methyl butanoic acid (TCI America, Portland, OR), and linalool (Alfa Aesar, Ward Hill, MA).
Polyphenol analyses.
Polyphenol analyses were performed using two assays. The modified Adams-Harbertson assay was used (Harbertson et al. 2003) to determine anthocyanins expressed as malvidin-3-glycoside equivalents and total Fe-reactive phenolics (IRP) as catechin equivalents. Due to resuspension issues, tannins could not be measured with this assay, and total phenols expressed as gallic acid equivalents were determined using a small-scale Folin-Ciocalteu method (Singleton and Rossi 1965, Zoecklein et al. 1995). For both methods, each sample was measured in triplicate.
Basic winery lab parameters.
Titratable acidity (TA) determined as tartaric acid equivalents and pH and volatile acidity (VA) expressed as acetic acid equivalents were determined in duplicate (Amerine et al. 1980).
Statistical analyses.
All statistical analyses were performed in R (ver. 2.11.1; R Foundation for Statistical Computing, Vienna, Austria). Missing values in the DA (that is, a judge missed a session) were imputed with the arithmetic means of the other replicates, as R excludes the whole observation if values are missing. That was the case for 24 (four judges missing one session of six wines each) out of a total of 756 observations (18 wines × 14 judges × 3 replicates).
All data sets were statistically tested for significance using multivariate analysis of variance (MANOVA) for the overall main effect wine. If significant, an univariate analysis of variance (ANOVA) for a two-way fixed effect model (wine, replication) in the case of all chemical analyses and a three-way fixed effect model with all two-way interactions for the DA (wine, replication, judge) was performed. Fisher’s least significant differences (LSD) were calculated to express significant differences between the wines for each sensory descriptor or chemical value.
Canonical variate analysis (CVA) (also known as linear discriminant analysis; LDA) was performed on all data sets to obtain the product space. In contrast to a principal component analysis (PCA), a CVA takes replicate measurements into account and an averaging across judges and replications for a PCA is not needed. By implication, a CVA is able to show differences among samples which in a PCA might be lost because of the averaging. In addition, 95% confidence intervals (CI) can be added to the CVA product plot. The CI definition of Chatfield and Collins (1980) was used, with the 95% CI indicated by circles around the sample means with a radius of 2 divided by the square root of number of values for the sample mean calculation. The addition of the CI to the CVA provides significance testing, as circles which do not touch each other in any significant dimension are significantly different from each other. The CVA calculation was performed using the candisc package (http://CRAN.R-project.org/package=candisc). To plot the CVA the plotcc package (G. Hirson, 2010, unpublished package) was used. Multifactor analysis (MFA) was used to compare the different data sets and to see how similar the product spaces were to each other. A MFA can be explained as a PCA of PCAs to integrate different data sets with different variables, but all describing the same objects, as outlined elsewhere (Abdi and Valentin 2007, Escofier and Pagès 1994) and integrated into R in the SensoMineR package (http://CRAN.R-project.org/package=SensoMineR). Scaling differences among the data sets are equalized by scaling to unit variance. A useful tool for judging the “goodness” of the provided MFA solution is given by the RV coefficient, a multivariate version of a Pearson’s coefficient, where RV values >0.75 indicate a strong relationship.
In this study, we were especially interested in whether the blending of wines would change the sensory characteristics and chemical values in a different way than the linear average. For this, we calculated for each variable and each blend the theoretical linear average for the specific blend based on the percentages of each monovarietal base wine and compared these values to the actual overall mean value determined by the descriptive panel or the chemical method. We defined three effects: (1) an averaging effect, when for a particular variable the blend’s mean was close to the simple average of the monovarietal base wines; (2) a suppressing effect, when the variable mean for a given blend was lower than the theoretical average and higher than the Fisher LSD value for that variable; and (3) an amplifying effect, for blends that showed variable means that were higher than the theoretical average value.
Results and Discussion
Overview of descriptive analysis results.
The MANOVA for wine showed significant differences among the samples at p < 0.05. Therefore, performing univariate ANOVA for individual attributes was secure, as the MANOVA “protects” against type I errors (Lawless and Heymann 2010). Fourteen of the 23 sensory attributes showed significant differences among the wines at p < 0.05 in the ANOVA using all three main effects and all two-way interactions in a fixed-effect model. For significant wine (W) as well as judge*wine (JxW) or wine*replication (WxR) interactions, a pseudo-mixed model according to Gay (1998) was used to test whether the significant wine effect was due to the interaction or not. That was the case for the attributes fresh veggie, floral, medicinal, soy sauce, earthy, sweet, and astringent, which all remained significant at p < 0.05 with the exception of fresh veggie. The attribute soy sauce showed both significant JxW and WxR interactions, but the pseudo-mixed model was only applied on the JxW interaction (see Supplemental Table 1).
Fisher’s LSD and overall mean scores for all sensory attributes were calculated (Table 3). The three base wines were different from each other: the Cabernet Sauvignon (S) was mainly dominated by high values in oak, medicinal, soy sauce, and astringency, the Merlot (M) showed high ratings in chemical and hot mouthfeel, and the Cabernet franc (F) showed the lowest values for astringency and sourness and the highest values for sweetness.
A CVA was calculated and Bartlett’s test (Chatfield and Collins 1980) was applied to determine the number of significant canonical variates (CV). The first two CVA dimensions were significant at p < 0.05, and the angles between the two dimensions were 90°. In addition, a significant drop and a knee in the scree plot (canonical dimensions vs. eigenvalues) were observed after the second dimension. Within the first two dimensions, 52.5% of the total variance ratio could be explained, with the first dimension accounting for 38.7% and the second dimension for an additional 13.8% (Figure 1). Of particular interest was whether the product positions in the CVA plot would reflect the blending: that is, whether 50-50 blends were located between the two monovarietal base wines. As expected, all three base wines were significantly different from each other and located in three different quadrants representing the extremes of the product space. The Cabernet Sauvignon was mainly described by high ratings in medicinal, soy sauce, sulfur, and astringent and low ratings in fruit, sherry, and black pepper attributes. The Merlot was highly positively correlated with floral, hot, chemical, and sweet characteristics and negatively correlated with the oak and earthy attributes. The Cabernet franc was dominated by fruit, sherry, and black pepper aromas and the nonsignificant canned veggie and lactic attributes. The 50-50 blends (SF11, SM11, and MF11) and SMF111 were all located around the center. Interestingly, both blends made with the Cabernet Sauvignon base wine were more distant from the Cabernet Sauvignon base wine than from the other monovarietal base wines (either Merlot for the SM11 or Cabernet franc for the SF11), while the 50-50 blend of Merlot and Cabernet franc (MF11) was nearly exactly located between the two base wines. All 80-20 and 90-10 blends were very close to their respective major component, although the Merlot-based blends (MS91, MS82, MF91, MF82, MSF811) were closer to the M than the Cabernet Sauvignon-based blends (SF91, SF82, SM91, SM82, SMF811) were to Cabernet Sauvignon. It seems that due to the intense characteristics of the Cabernet Sauvignon, especially in medicinal and soy sauce, blending this wine with the Merlot or the Cabernet franc decreased the perception of these attributes. With the exception of the winemaker’s blend and the 50-50 blend with Merlot (MF11), all wines were significantly different from the Cabernet franc.
Blending effects determined by descriptive analysis.
Averaging, suppressing, and amplifying effects in the wine blends were examined. Most of the wines did not show clear effects and could not be categorized in one of the above categories; however, examples for all three above described effects could be found, with the averaging effect the most common one, present in all attributes except floral. Each wine blend showed at least one averaging effect in at least one of the significant attributes. The highest number of averaging effects for 6 of the 14 significant attributes was found for the Merlot-based blend with 20% Cabernet franc (MF82), which was close to the theoretical average value (less than 0.06 difference) for the attributes fruit, oak, earthy, sulfur, sherry and astringent. For five attributes (medicinal, soy sauce, earthy, sherry, and astringent), the 50-50 Merlot–Cabernet franc blend (MF11) showed an averaging effect as well.
Suppressing effects were also observed for six attributes: for medicinal in the winemaker’s blend, soy sauce in all Cabernet Sauvignon–Cabernet franc blends (SF11, SF82, SF91), chemical in the Merlot–Cabernet franc blend (MF11), sherry in the Cabernet Sauvignon-based blend with 20% Cabernet franc (SF82), black pepper in the Merlot-based blend with 10% Cabernet franc (MF91), and perceived astringency in the Cabernet Sauvignon-based blend with 10% Merlot (SM91). An amplifying effect was found in only three wines, all Cabernet Sauvignon-based blends, and only for three attributes. For the earthy and sulfur attributes, the Cabernet Sauvignon with 20% and 10% Merlot, respectively, had higher overall means than the theoretical average would suggest. The 80-20 Cabernet Sauvignon-based–Cabernet franc blend had an amplified black pepper character.
These findings suggest that complex mixtures such as wine blends do not follow any predictable relationship in the sensory perception. Most frequently, a linear averaging occurs. However, especially for attributes that are rated high in the monovarietal base wines, a suppressing or a masking effect occurs, as can be seen for the perceived astringency. The opposite was also observed for the sulfur and black pepper characteristics, where the blends, which showed an amplified effect, were more intense than the monovarietal base wines.
Blending wines that lacked overly expressed single sensory attributes appeared to produce blends that were less differentiable from the monovarietal base wines. This was found when blending Merlot and Cabernet franc, where both the 80-20 and the 50-50 blends had the highest number of averaged sensory attributes.
On the other hand, blending an “uneven” wine with a few very strong sensory attributes, as was the case with the Cabernet Sauvignon, led to more differentiated blends with suppressing or amplifying effects. A suppressing effect was observed for the soy sauce characteristic, which was significantly reduced in the 90-10 Cabernet Sauvignon–Cabernet franc blend (SF91), or for the sulfur attribute in the 80-20 Cabernet Sauvignon–Merlot blend (SM82). These effects could be why other studies on wine blends found the highest quality scores for blends rather than their respective base wines (Singleton and Ough 1962, Monagas et al. 2007, Gomez Garcia-Carpintero et al. 2010). Singleton and Ough (1962) speculated that sensational differences follow geometric rather than linear functions of stimuli and, therefore, flavors perceived strongly in one base wine are not perceived that much weaker when mixed with a base wine lacking that strong flavor in a blend. However, this less strong flavor might add positively to the perceived complexity without overpowering the wine.
Some intense attributes present in one monovarietal wine were masked to a different extent, if the other monovarietal wine in the blend was rated high in some “contrary” attribute. For example, the fruity Merlot seemed to have less masking properties than the sweet Cabernet franc, as 10% of the latter produced a blend that was significant differently from the Cabernet Sauvignon monovarietal base wine, which was high in soy sauce characteristics, while the Merlot blend was not.
Volatile patterns determined by HS-SPME-GC-MS.
The 60 selected compounds identified by GC-MS (Supplemental Table 2) were integrated and their peak areas were submitted to a two-way main effect ANOVA. Initially, a MANOVA for wine was performed and showed a significant effect at p < 0.05. In the ANOVA all 60 compounds were significantly different among the wines at p < 0.05. For some compounds, a significant replication effect was found (Supplemental Table 3). Fisher’s LSD values were calculated to show differences among the wines based on the volatile analysis. The overall means for all wines together with the LSD and groups are given in Supplemental Table 2.
For the 10 compounds extracted with the m/z 56 (mostly primary alcohols), the highest overall means were found for all 90-10 or 80-20 Cabernet Sauvignon blends (SF82, SF92, SM82, SM91). For the m/z 60 compounds (organic acids), the highest mean was again found for 80-20 and 90-10 Cabernet Sauvignon blends (SF82, SF91, SM82), with the exception of compounds mz60_c5 and _c6 (octanoic and decanoic acid, respectively), for which the highest mean was found in the winemaker’s blend (55% Cabernet franc, 30% Cabernet Sauvignon, and 15% Merlot).
Calculating the CVA for the volatiles, the first 15 dimensions were significant at p < 0.05 using the Bartlett’s test (Chatfield and Collins 1980). The angles between the first three dimensions were 90°. A dramatic drop and a knee in the scree plot (canonical dimensions vs. eigenvalues) were observed after the third dimension, justifying the use of only the first three dimensions. In addition, 84.8% of the total variance ratio could be explained within the first three dimensions. CV1 explained 46.9% and CV2 30.4% of the total variance ratio.
The two-dimensional CVA product plots for the first three dimensions are shown (Figure 2). The first two dimensions (explained variance 77.3%) show well-separated sample groups (Figure 2A). The Merlot and Merlot-based blends are located in the lower left quadrant, while both monovarietal Cabernet Sauvignon and Cabernet franc are positioned with their 50-50-blend (SF11) in the lower right. Two Cabernet Sauvignon–Merlot blends are located at the top, with SM82 in the right quadrant and SM11 in the left. The remaining Cabernet Sauvignon–Merlot blend (SM91) is located closer to the Cabernet Sauvignon along the positive CV1 axis, showing a high similarity with the Cabernet Sauvignon-based three-wine blend SMF811 located right next to it. The remaining wines (Blend, MF11, SMF111, SF91, SF82, MF91, and MF82) are located around the center, with the Merlot-based blends closer to the monovarietal Merlot.
The product plot on the canonical variates CV2 and CV3 (explained variance ratio of 37.9%) show three major groups of wines distinguished along the CV2 axis (Figure 2B). The three monovarietal wines (S, M, F) are located on the left, two Cabernet Sauvignon–Merlot blends (SM82, SM11) on the right, and all remaining wines between those two groups. Along CV3, the wines are separated based on their Cabernet franc percentage: the winemaker’s blend with 55% Cabernet franc is positively loaded and the Cabernet Sauvignon-based–Merlot blend (SM91) is negatively loaded.
For CV1 and CV3 (explained variance ratio of 54.3%), the Cabernet Sauvignon is located in the lower right quadrant with the Cabernet Sauvignon-based blends SM91 and SMF811 (Figure 2C). A group consisting of the monovarietal Merlot and all Merlot-based blends is in the lower left quadrant, while the monovarietal Cabernet franc and all Cabernet franc blends are grouped around it.
The volatile compounds with the highest CVA loadings on the first three CV dimensions are summarized (Table 4). These compounds are the most discriminating among the wine samples. On CV1, methyl-γ-octalactone had the highest absolute loading, followed by various esters such as diethyl succinate, ethyl octanoate, ethyl-2-hydroxy hexanoate, ethyl-3-methyl butanoate, and 2-phenylethyl ester. The CV1 axis spanned from the negatively loaded methyl-γ-octalactone to the positively loaded diethyl succinate and ethyl octanoate. On CV2 the same compounds were loaded highest with the addition of 1-hexanol: diethyl succinate was loaded highest followed by ethyl-2-hydroxy hexanoate, 2-phenylethyl ester, ethyl-3-methyl butanoate, and 1-hexanol. Diethyl succinate was negatively loaded on CV2, while ethyl-2-hydroxy hexanoate and 2-phenylethyl ester were loaded positively. CV3 was highest loaded negatively with ethyl octanoate and positively with methyl octanoate.
Combining the loadings with the wines, the Merlot and the Merlot-based blends showed a high correlation with methyl-γ-octalactone, ethyl-2-hydroxy hexanoate, ethyl butanoate, and ethyl decanoate, while the Cabernet franc, Cabernet Sauvignon, and their 50-50 blend (SF11) were highly correlated with diethyl succinate, ethyl hexanoate, octanoic acid, and 1-hexanol. The 50-50 Cabernet Sauvignon–Merlot blend (SM11) was highly correlated with ethyl-2-hydroxy hexanoate and δ-nonalactone, while the 80-20 Cabernet Sauvignon–Merlot blend (SM82) had a high correlation with 2-phenylethyl ester. The Cabernet Sauvignon-based blends SM91 and SMF811 also stood out: these wines were located along the positive CV1 mainly because of their high correlation with hexanoic acid, ethyl-2-methyl butanoate, and phenylmethanol. Two Cabernet Sauvignon-based blends (SMF811 and SM91) were positively correlated with hexanoic acid, ethyl hexanoate, and diethyl succinate. The winemaker’s blend was mostly characterized by high loadings of methyl octanoate.
On CV3 only a few compounds stood out and could be correlated to the wines, such as hexanoic acid and ethyl-3-methyl butanoate, which showed a high correlation to SM11. The group of Cabernet Sauvignon-based blends (SMF111, SF91, and SF82) correlated positively with methyl octanoate, while ethyl butanoate correlated with Cabernet franc and SF11.
Blending effects determined by HS-SPME-GC-MS.
Similar to the sensory results, most wines did not show clear blending effects; in cases where they did, however, an amplifying effect was the most common, followed by suppressing effects. Simple linear averaging was not observed. With the exception of the aliphatic alcohols, all listed compounds (Table 4) were more often amplified than suppressed in the wines. The three-wine blend SMF111 followed by the winemaker’s blend and the 90-10 Cabernet Sauvignon–Cabernet franc blend (SF91) had the most amplified compounds: 30, 26, and 26, respectively. Other blends with amplifying effects had either 80 or 90% Cabernet Sauvignon (SF82, SF91, SM82, SM91).
In contrast, suppressing effects could only be observed for Merlot-based blends (MF11, MF82, MF91, MS82, MSF811) with the exception of the winemaker’s blend for one compound. The wine with the highest number of suppressed compounds (six) was the 50-50 Merlot–Cabernet franc blend (MF11) followed by the Merlot-based three-wine blend MSF811 (five) and the 90-10 Merlot–Cabernet franc blend MF91 (five). According to these results, Cabernet Sauvignon-based blends seemed to have higher concentrations of various esters, organic acids, and terpenes than the linear average would suggest, while the Merlot-based blends had suppressed concentrations of aliphatic alcohols.
General overview of the polyphenol pattern.
Anthocyanins, total Fe-reactive phenolics (IRP), and total phenols were determined for all 18 wines in triplicate. For all three analytes, a MANOVA for the main effect wine was significant at p < 0.05. Subsequent ANOVAs using a two-way fixed effect model with wine and replication at p < 0.05 revealed significant wine effects at p < 0.05 (Supplemental Table 4). The highest overall mean for anthocyanins was for the 80-20 Cabernet Sauvignon–Cabernet franc blend (SF82), which was also significantly different from all other wines (Table 5). The highest mean for IRP was for the 90-10 Cabernet Sauvignon–Merlot blend (SM91), which was also significantly different from all other wines. For total phenols as measured by Folin-Ciocalteu, the highest mean was for the Cabernet Sauvignon-based three-wine blend SMF811, which was significantly different from all other wines with the exception of Cabernet Sauvignon, Merlot, the Cabernet Sauvignon-based blends SF91, SM82, and SM91, and the 50-50 Merlot blend SM11.
According to Bartlett’s test (Chatfield and Collins 1980), the first two dimensions were significant at p < 0.05 in the CVA. The angles between the first two dimensions were 90°. Within these dimensions, 98.0% of the total variance ratio could be explained with anthocyanin values as the greatest discriminating variable; anthocyanins nearly exclusively separated the wines along the second dimension (Figure 3). The wines were clearly separated along the first dimension by Cabernet franc volume, with wines that contained at least 33% Cabernet franc located on the positive first dimension.
The blending of Cabernet Sauvignon and Merlot seemed to shift the anthocyanin pattern so that the blends showed a higher correlation with the anthocyanins than the monovarietal base wines. The base wines were more correlated to TP concentration and IRP (see Figure 3). It is not surprising that IRP and total phenols had a high correlation, as the Folin-Ciocalteu reagent reacts with mono- and dihydroxylated phenolics (Harbertson and Spayd 2006) and the FeCl in the IRP determination reacts with all phenolics, with the exception of anthocyanins and monohydroxylated phenolics (Harbertson et al. 2004); thus, similar trends should be obtained with both assays that were used.
Blending effects determined by polyphenol analyses.
For both anthocyanin and IRP levels, some amplifying effects could be found in eight blends. In both assays, the greatest deviation from the linear average was ~2.5 times the LSD value, which, for anthocyanins, was the 80:20 Cabernet Sauvignon–Cabernet franc blend (SF82), and for IRP, was the winemaker’s blend. In addition, averaging effects were found for anthocyanins for three Merlot-based blends (MF82, MS82, and MSF811). No effects were observed for the total phenols determined with the Folin-Ciocalteu method. Similar to previous studies, the highest mean values in all three polyphenol measurements were in wine blends rather than the monovarietals (Monagas et al 2006a, 2006b, Singleton and Ough 1962).
General overview of basic winery parameters.
Titratable acidity (TA), volatile acidity (VA), and pH were determined for all 18 wines in duplicate and a MANOVA for the main effect wine was significant at p < 0.05 for all three parameters. ANOVAs using a two-way fixed effect model with wine and replication at p < 0.05 found significant wine effects at p < 0.05 (Supplemental Table 4). The highest overall TA mean was for the Merlot and the lowest for the Cabernet franc; both were significantly different from all other wines (Table 5). The highest VA levels were in Cabernet Sauvignon, the winemaker’s blend, Cabernet franc, and the 80-20 blend of the two Cabernets (SF82), while the lowest levels were in Merlot, and the Cabernet Sauvignon blends with at least 50% Merlot. The highest pH mean was in the Cabernet franc, which was significantly different from all other wines, and the lowest in Merlot and all Merlot-based blends.
According to Bartlett’s test, all three dimensions in the CVA were significant at p < 0.05 (Chatfield and Collins 1980) and 97.2% of the total variance ratio was explained within the first two dimensions (Figure 4). pH explained most of the first dimension, while TA and VA explained the variance along CV2. Along CV1, the Cabernet franc showed the highest positive correlation with pH, while all Merlot-based blends were negatively correlated with pH. Both VA and TA differentiated among the samples along CV2. The monovarietal Cabernet Sauvignon (S), the winemaker’s blend, and the 80-20 Cabernet Sauvignon–Cabernet franc blend (SF82) were positively correlated, while all other wines were negatively correlated to TA and VA. However, the differentiation in the CVA biplot wass mainly due to pH, as the first dimension explained ~90% of the total variance ratio.
Blending effects determined by basic winery parameters.
Similar to Singleton and Ough (1962) and in contrast to the sensory, volatile, and polyphenols results, the highest overall means for titratable acidity, volatile acidity, and pH were in the monovarietal base wines, supporting the theory that these simple wine analysis parameters follow a linear relationship. For all three analytes, mainly suppressing and some averaging effects were observed: in all cases Cabernet Sauvignon-based blends had the greatest deviation from the linear average (up to 4 times the LSD for the pH values). The winemaker’s blend and the Merlot blends (MF11, MF82, MF91, MSF811, MS82, MS91, SM11) followed the linear average for volatile acidity and pH.
Comparison of data sets.
The first three dimensions of the individual CVAs were used in a multifactor analysis (MFA) to obtain an accurate comparison of all data sets. The MFA performed on the first three canonical variates of each data set explained 65.8% of the total variance, with 27.8% in the first dimension and 23.1% in the second dimension. Each data set showed a good fit in the overall MFA solution, with a minimum RV coefficient of 0.71 for the acidity data set (TApHVA) and a maximum RV coefficient of 0.77 for the GC-MS data, closely followed by the DA data (0.76) (Table 6). The latter two data sets also showed the highest correlation with each other (RV = 0.50), followed by the correlation of GC-MS with polyphenol data (RV = 0.40). The lowest correlation was between polyphenols and TApHVA.
A partial product plot, where the MFA consensus position of each wine is grouped with the position of that wine in each data set, clearly shows how well the MFA solution represents the individual data sets. The closer the partial position is to the MFA position, the more representative the MFA solution is for the particular wine. Good agreement among all data sets and the MFA solution was obtained for the Cabernet Sauvignon, the winemaker’s blend, the 50-50 Cabernet Sauvignon–Cabernet franc blend (SF11), the 90-10 Merlot–Cabernet Sauvignon blend (MS91), and the Merlot-based three-wine blend (MSF811). The greatest deviation of the MFA consensus position was for the Merlot, the Cabernet franc, and the 50-50 Cabernet Sauvignon–Merlot blend (SM11) (graph not shown).
In the MFA product plot, one group was formed by the Cabernet franc, the winemaker’s blend, the 50-50 Merlot–Cabernet franc blend (MF11), the 80-20 Cabernet Sauvignon–Cabernet franc blend (SF82), and the 33-33-33 blend of the base wines (SMF111) (Figure 5). The Merlot and all Merlot-based blends were located in the lower left quadrant. The top two quadrants contained the Cabernet Sauvignon and the Cabernet Sauvignon-based blends with Merlot (SM11, SM82, SM91, SMF811) on the left and the Cabernet Sauvignon-based blends with Cabernet franc (SF91, SF11) on the right. The 50-50 blends were located closely between the respective monovarietal wines, with the exception of SF11, which was closer to the Cabernet franc than to the Cabernet Sauvignon, and the three-wine blend SMF111, which was closer to the Cabernet franc and the Merlot than to the Cabernet Sauvignon. All blends with at least 80% Merlot showed more similarity with the monovarietal Merlot than did the blends with at least 80% Cabernet Sauvignon and the monovarietal Cabernet Sauvignon. In comparing the MFA product plot with the individual CVA product plots of all data sets, the highest similarity was with the sensory product plot.
Conclusions
Blending of wines leads to changes in both sensory and chemical characteristics that are beyond simple averaging effects. Overall, blending wines that had a rather “smooth” flavor profile without any spikes in sensory attributes seemed to produce blends that were less differentiable from the monovarietal base wines. Conversely, blending of an “uneven” wine with few sensory attributes driving the whole wine, as with the Cabernet Sauvignon, seemed to lead to more different blends where suppressing or amplifying effects occurred. In addition, some intense attributes arising from one monovarietal wine were masked to a different extent if the other monovarietal wine(s) in the blend were rated high in some contrary attribute.
Several sensory descriptors as well as chemical measurements of volatiles and polyphenols exhibited the highest overall mean values in the wine blends. These results should be confirmed with different wines, as we assume that different base wines would perform differently and a database of many blending studies would be needed to make global conclusions.
Acknowledgments
All wines were donated to UC Davis by Franciscan Estate (Cabernet Sauvignon and Merlot) and Clos du Bois (Cabernet franc). The authors thank Jerry Lohr for providing postdoctoral financial assistance.
All analyses were performed at the Dept. of Viticulture and Enology, UC Davis. The authors thank all panelists for participating in the sensory evaluations, Charles Brenneman for making the wine blends, Wender L.P. Bredie for helpful discussions, Greg Hirson for the initial project idea, and Duncan Temple Lang for help with R graphics.
Footnotes
-
Supplemental data is freely available with the online version of this article at www.ajevonline.org.
- Received October 2011.
- Revision received January 2012.
- Accepted March 2012.
- Published online December 1969
- ©2012 by the American Society for Enology and Viticulture