The Pennsylvania State University

Home , Pennsylvania wine

The Graduate School

THE IMPACT OF WINE MATRIX COMPONENTS ON THE STATIC AND TEMPORAL

PERCEPTION OF SENSORY ATTRIBUTES IN A WHITE, HYBRID MODEL WINE

A Thesis in

Food Science

Andrew Poveromo

Submitted in Partial Fulfillment

of the Requirements

for the Degree of

Master of Science

August 2019

The thesis of Andrew Poveromo was reviewed and approved* by the following:

Helene Hopfer Assistant Professor of Food Science Thesis Advisor

Ryan Elias Associate Professor of Food Science

Michela Centinari Assistant Professor of Viticulture

Robert F. Roberts Professor of Food Science Head of the Department of Food Science

*Signatures are on file in the Graduate School

iii

ABSTRACT

Wine flavor is incredibly complex, in part, due to the complex chemical composition. The matrix consists of major components like water and ethanol, as well as minor and trace compounds such as glycerol, phenolics, proteins, minerals, organic acids, and volatiles. All of these components contribute to the overall flavor of the wine, whether directly or indirectly through interactions with other components.

Therefore, it is very important to understand the effects and interactions of matrix factors like ethanol, glycerol, phenolics, and proteins, to better understand wine flavor and aroma and to allow winemakers to have greater control over the sensory properties of their finished product through winemaking.

Currently, much of the literature on wine flavor explores the sensory properties of red, Vitis vinifera wines. This is interesting as white wine is not only important to the world’s wine industry, but it also behaves differently than red wine. Conclusions made about the flavor of red wine do not necessarily hold true in white wines. Much of the research done also involves Vitis vinifera species and not hybrid species. Hybrid grapes are very important to wine producing regions in the northern United States and

Canada as they tend to survive better in cooler and more humid climates and in areas where diseases are more prevalent. Hybrid grapes differ in their composition from Vitis vinifera grapes, in that they have a lower tannin concentration but a greater concentration of proteins and glycosylated anthocyanins. This difference in composition potentially leads to a difference in sensory properties. Therefore, it is important to understand how wine matrix components affect the flavor of hybrid wines. iv

There is also a lack of work using temporal evaluation methods to characterize wines. The sensory attributes of wine change over time and throughout the course of consumption. Traditional static sensory evaluation methods, such as descriptive analysis, may not capture sensory properties that change over time. It is necessary to address this lack of temporal research in white, hybrid wines to better understand the effect that the matrix components have on the temporal perception of sensory attributes.

In this study, we set out to characterize model white, hybrid wine samples using descriptive analysis (DA) and temporal check-all-that-apply (TCATA). The chemical data was also used in an attempt to predict the sensory attributes. In the model wine samples, pH was kept consistent while levels of ethanol, glycerol, phenolics, and protein were varied. An aroma fraction, modeled after a white, hybrid Traminette wine was added at a consistent concentration to each of the samples to evaluate the effects that each of the experimental wine components had on the static and dynamic perception of the sensory attributes along with the effects that each component had on the partitioning of volatile aroma compounds into the headspace of the samples.

A descriptive analysis panel, on 17 sensory attributes, found significant differences (p < 0.05) in pear and ethanol aroma, pear flavor, sweet, sour, and bitter taste, as well as astringent and warm/hot mouthfeel due to ethanol levels. Pear aroma and flavor, sweet taste, and sour taste decreased with increasing levels of ethanol, while ethanol aroma, bitter taste, astringent and warm/hot mouthfeel increased. There were significant increases for apple flavor, pear flavor, grape flavor, and sweet taste with increasing levels of glycerol, while protein was found to increase astringent v mouthfeel at higher concentrations. Phenolics did not have any significant main effects, however, all factors showed significant interaction effects, indicating indirect effects on sensory attributes.

To assess the dynamic changes, a TCATA panel assessed twelve model wine samples where protein was kept at a constant level. In this part of the study, eight sensory attributes from the DA (citrus flavor, apple flavor, pear flavor, sweet taste, sour taste, bitter taste, astringent mouthfeel, and warm/hot mouthfeel) were used by the panel, which found significant differences in all attributes at varying times during evaluation. Ethanol and glycerol generally had the largest effect on the sensory attributes. Many significant interaction effects were also observed. For flavor attributes, significant differences were typically seen within the first 30 seconds.

Attributes like bitter taste and warm/hot mouthfeel showed significant differences throughout the evaluation period. Comparing DA and TCATA results, similarities were observed for the first 30 seconds, with some attributes, like apple flavor, sweet taste, sour taste, and warm/hot mouthfeel matching up well, while others, like citrus flavor, bitter taste, and astringent mouthfeel, did not agree between DA and TCATA results.

Overall, the effects of ethanol, glycerol, protein, and phenolics were studied in both a static and dynamic manner. Significant main effects and interaction effects were found for all factors in both evaluation methods. This study provides a better understanding of how the matrix components interact and effect the sensory attributes in a white, hybrid model wine system. Once verified in a real wine system, these conclusions will be able to help winemakers better understand how the levels of the matrix components in their wines will affect their sensory properties. vi

TABLE OF CONTENTS

LIST OF FIGURES ...... ix

LIST OF TABLES ...... xi

ACKNOWLEDGEMENTS ...... xiii

1 Literature Review ...... 1

1.1 Flavor perception of wine ...... 1

1.2 Wine composition and its impact on sensory perception ...... 2

1.2.1 Trace matrix and aroma compounds ...... 4

1.2.2 Ethanol ...... 5

1.2.3 Glycerol ...... 7

1.2.4 Phenolics ...... 8

1.2.5 Protein ...... 10

1.2.6 Interactions ...... 12

1.3 Methods to characterize wine sensory properties ...... 15

1.4 Model Wine ...... 24

1.5 Hybrid Grapes ...... 25

1.6 Linking Instrumental to Sensory Data ...... 28

2 Descriptive Analysis and Chemical Analysis of Model Wine Samples ...... 32

2.1 Introduction ...... 32

2.2 Materials and Methods ...... 40

2.2.1 Determination of aroma fraction ...... 40 vii

2.2.2 Creation of model wines ...... 43

2.2.3 Sensory Evaluation with a trained Descriptive Analysis Panel ...... 44

2.2.4 Instrumental Evaluation of Model wines ...... 46

2.2.5 Statistical analysis of sensory and instrumental data ...... 47

2.3 Results ...... 48

2.3.1 Determination of the Traminette aroma fraction ...... 48

2.3.2 Descriptive Analysis revealed significant factor effects on sensory

perception ...... 48

2.3.3 Correlation between sensory and instrumental data ...... 64

2.4 Conclusion ...... 69

3 Temporal Evaluation of Model Wine Samples ...... 72

3.1 Introduction ...... 72

3.2 Materials and Methods ...... 80

3.2.1 Model Wines and Aroma Fraction ...... 80

3.2.2 Sensory Evaluation by Temporal Check-All-That-Apply (TCATA)...... 80

3.2.3 Statistical Analysis ...... 83

3.3 Results and Discussion ...... 85

3.3.1 Panelist Consistency and Panel Agreement ...... 85

3.3.2 Effect of varying ethanol, glycerol, and caffeic acid on temporal sensory

evaluation ...... 87

3.3.3 Visualization of temporal changes in sensory attributes due to the

experimental factors ...... 92

3.3.4 Comparing static DA to dynamic TCATA evaluation ...... 95 viii

3.4 Conclusion ...... 99

4 Conclusions and Future Work ...... 101

4.1 Significance ...... 101

4.2 Future Directions ...... 106

4.3 Overall Conclusions...... 107

References ...... 108

Appendix: Chapter 3 Supplementary Tables ...... 122

LIST OF FIGURES

Figure 1. Structure of ethanol. Structure is taken from PubChem.20 ...... 6

Figure 2. Structure of glycerol. Structure is taken from PubChem.20 ...... 7

Figure 3. Structures of phenolic compounds. (A) Structure of the simplest phenolic

compound, phenol. (B) Structure of a polyphenol, reservatrol. (C) Structure of

caffeic acid, tartrate ester, a hydroxycinnamic acid.10 (D) Structure of

epicatechin, a flavan-3-ol. (E) Structure of quercetin, a flavonol. (F) Structure of

malvidin-3-glucoside, an anthocyanin. Structures are taken from PubChem

unless otherwise cited.20 ...... 9

Figure 4. Ribbon diagram of wine protein F2/4JRU. Diagram is taken from

Marangon et al.32 ...... 11

Figure 5. Interaction plots for all significant two-way interactions. Significant

interactions do not imply significant differences in means between samples.

Interactions are plotted to visualize the effects they have on sensory attributes.

...... 59

Figure 6. (A) Score and (B) loadings plots for PC 1 and PC 2, and (C) Score and (D)

Loadings plot for PC1 and PC 3, using the DA results for the model wine

samples. In the score plot, individual wine samples are indicated as black dots,

while the different experimental factors are super-imposed as supplementary

factors (ethanol levels 10, 12, and 14 %(v/v) in red, protein levels low (0.1 g/L)

and high (0.5 g/L) in blue, glycerol levels low (5 g/L) and high (15 g/L) in black,

and phenolic levels low (20 mg/L) and high (60 mg/L) in green) ...... 63

Figure 7. Validation plots for each of the sensory attributes in the PLSR model. .... 66 x

Figure 8. (A) Score and (B) correlation plots for the PLSR model of our study. 10,

12, and 14 refer to the ethanol content. “C/c” refers to the caffeic acid

concentration, “G/g” refers to the glycerol concentration, and “B/b” refers to the

protein concentration. Capital letters indicate the high level and lowercase

letters indicate the low level. The score plot shows the samples and the

correlation plot shows how well the chemical composition predicts the sensory

attributes...... 69

Figure 9. The conversion of raw frequency data to histograms and then to TCATA

curves. Figure is taken from Castura et al.47 ...... 79

Figure 10. Interaction plot for the significant two-way interaction between ethanol

and glycerol on bitter taste at 135 seconds...... 90

Figure 11. Biplot of PCA trajectories for the model wine samples with samples color-

coded by ethanol content (shades of blue represent 10% EtOH; shades of

green represent 12% EtOH; shades of red represent 14% EtOH). Capital letters

in the sample name represent the high level of the factor and lowercase letters

represent the low level of the factor...... 93

Figure 12. Difference curves for samples 10 C G B and 14 C G B indicating

attributes that showed significant differences during evaluation...... 95

LIST OF TABLES

Table 1. Matrix factor effects on sensory attributes...... 15

Table 2. Aroma fraction composition...... 42

Table 3. List of major wine components and levels used in the model wines...... 43

Table 4. References used for training of sensory attributes. All standards were

prepared in 10% (v/v) ethanol solution, unless otherwise noted...... 46

Table 5. MANOVA table for significant factors on sensory attributes ...... 50

Table 6. F-values for the experimental factors and their interactions on the sensory

attributes of the model wine samples...... 54

Table 7. The effect of the wine matrix components on the sensory attributes of the

model wine samples...... 55

Table 8. MANOVA table for factor and interaction effects on the partitioning of

volatile compounds into the headspace of the model wine samples. Bold P-

values are significant at   0.05...... 65

Table 9. Panelist agreement indices for the three different sensory attributes (citrus

flavor, warm/hot mouthfeel, and sweet taste) and two ethanol levels (10 and

14% v/v), and repeatability indices...... 87

Appendix Table 1: ANOVA for citrus flavor showing F-values for the experimental

factors ethanol (E), caffeic acid (C), and glycerol (G), and their interactions.

Bolded F-values are significant (p < 0.05)...... 122

Appendix Table 2: ANOVA for the pear flavor attribute showing the F-values for the

experimental factors ethanol (E), caffeic acid (C), and glycerol (G), and their

interactions. Significant F-values are shown in bold font (p < 0.05)...... 123 xii

Appendix Table 3: ANOVA for the apple flavor attribute showing the F-values for the

experimental factors ethanol (E), caffeic acid (C), and glycerol (G), and their

interactions. Significant F-values are shown in bold font (p < 0.05) ...... 124

Appendix Table 4: ANOVA for the astringent mouthfeel attribute showing the F-

values for the experimental factors ethanol (E), caffeic acid (C), and glycerol (G),

and their interactions. Significant F-values are shown in bold font (p < 0.05).125

Appendix Table 5: ANOVA for the warm/hot mouthfeel attribute showing the F-values

for the experimental factors ethanol (E), caffeic acid (C), and glycerol (G), and

their interactions. Significant F-values are shown in bold font (p < 0.05). .. 126

Appendix Table 6: ANOVA for the bitter taste attribute showing the F-values for the

experimental factors ethanol (E), caffeic acid (C), and glycerol (G), and their

interactions. Significant F-values are shown in bold font (p < 0.05)...... 127

Appendix Table 7: ANOVA for the sour taste attribute showing the F-values for the

experimental factors ethanol (E), caffeic acid (C), and glycerol (G), and their

interactions. Significant F-values are shown in bold font (p < 0.05)...... 128

Appendix Table 8: ANOVA for the sweet taste attribute showing the F-values for the

experimental factors ethanol (E), caffeic acid (C), and glycerol (G), and their

interactions. Significant F-values are shown in bold font (p < 0.05)...... 129

Appendix Table 9: ANOVA for the “other” attribute showing the F-values for the

experimental factors ethanol (E), caffeic acid (C), and glycerol (G), and their

interactions. Significant F-values are shown in bold font (p < 0.05)...... 130

xiii

ACKNOWLEDGEMENTS

First and foremost, I’d like to thank my parents, Ellen and Paul Poveromo, for their love and support. The two of you are a big reason that I was able to make it through not only this program, but also the past 18 years of school. To my mother, I have always considered you to be one of my best friends and I always appreciate being able to talk to you whenever I need to. To my dad, you’ve always ensured that

I have everything I could ever need to be happy. You’re a big reason why I’d consider myself to be successful. Thanks for helping me to study and finish my homework, especially math and Spanish.

I’d also like to thank my advisor Dr. Helene Hopfer. You saw potential in me and invited me to join your group, and for that, I am ever grateful. You taught me how to blindly identify a wine as either red or white. You gave me the confidence I needed to do things like lead a DA panel or give a presentation on a difficult topic for an interview. I know I was probably not always the easiest to work with and there were probably times where you had to explain something to me five times. I remember times where I needed you to convince me that what I was doing was correct because I wasn’t sure it was. You have always made time for me and been available for when I need help with something. You made me feel like I belonged in this program when I wasn’t sure if I did. Above all, in addition to being my advisor, I consider you to be my friend.

I remember in 501 where Dr. Lambert said that your advisor cannot be your friend. I feel that you care about me more as a person than you do as a student. You’ve made my time here special and I will always be thankful to have worked with you. xiv

I’d like to thank the members, both past and present, of the sensory group for all of your help along the way. Dr. John Hayes, Dr. Alyssa Bakke, Tiffany, Modesto,

Jen, Allison, Marielle, Molly, Allie, Jess, Kate, Gloria, Scott, and Pat, you have all been a huge help to me during my time here. You guys are some of my best friends here.

I’ve enjoyed our lab meetings, both the productive and unproductive, cake-filled ones.

You guys have helped me with things as small as sending out a screener in

Compusense and as large as designing my project and analyzing results. With all of the fantastic work coming out of this group, I am proud to have been a part of it.

I’d also like to thank my friends, both here at Penn State and wherever else they may be. To my friends here at Penn State that I haven’t mentioned yet, Ben,

Kaleigh, Drew, Chrissy, Jared, Stephanie, Jose, Weslie, Lisa, Laura, Terianne and

Magda, thank you for making my time here memorable. I have really enjoyed my time here and my experience would not have been the same without you guys. To my guys back home, Kenny, Jay, Dooling, Conor, Henry, Pitman, and Brendan, you guys are some of my best friends and staying in touch with all of you has meant so much to me. I can’t wait until we’re all together again to grab a drink at the Onion or the Anchor!

To all of my Clemson friends, especially my roommates Rob, Nick, and Chase, I hope to see you guys soon and I’m so thankful for all of your support! Let’s go get a PBR at

TTT!

I’d like to thank my extended family as well. I’m very close with my grandparents and I can’t thank the two of them enough for their support over the years. Papa, the next round of golf is on me. Meme, let’s play a real game of ping pong where you don’t let me win. And make Papa a pizza too. The guy is 88 years old, he deserves it! I have xv too many aunts, uncles, and cousins to thank individually, but I want to highlight one.

I want to thank my cousin, Jimmy McLaughlin, for being so supportive of me. You’ve always been a role model to me. I know it’s difficult for you having a cousin who is younger, taller, better looking, and has more hair than you, but your love and kind words have meant so much to me over the years. I’m lucky to have a cousin like you.

I’d like Penn State University and the Food Science Department for providing me with this wonderful opportunity. I have been provided with all of the tools I need to be successful. This is a department that clearly cares about their graduate students and takes care of them. There are many people at this university, including my other committee members, Dr. Ryan Elias and Dr. Michela Centinari, who have helped me along my journey. I’m lucky to be a part of this special community.

I’d also like to thank my undergraduate university, Clemson University, for helping me get here. If you’ve ever spoken to me, you probably know how much I love

Clemson. Clemson University helped me to become the person I am today and I wouldn’t be in this position without Clemson. They say there is something in those hills. I don’t know who “they” are, but they’re right. There’s something special about those hills and the Clemson family that draws you and leaves a lasting impression, and for that, I am thankful. Go Tigers!

With that out of the way, let’s get into this thesis!

This work was supported by the USDA National Institute of Food and Agriculture Federal

Appropriations under project PEN04624 (Accession No. 1013412). The findings and conclusions of this work do not necessarily reflect the view of the funding agency. 1

1 Literature Review

1.1 Flavor perception of wine

Flavor perception is a combination of our sense of taste along with retronasal olfaction and chemesthetic sensations. These different sensory systems come together to form somewhat of a “sensory packaging center” in the brain.1 When we refer to the flavor of something, we are talking about the perceived sensations caused by the combination of basic tastes paired with retronasal olfaction and chemesthetic sensations. For example, when a person is sick and has difficulty smelling something because of a stuffy nose, their sense of taste is often also affected. This demonstrates that we need more than just our tongue to perceive and identify different flavors.

Chemesthetic sensations are chemical induced sensations that are not considered tastes or smells. These sensations are typically perceived by the trigeminal nerve endings in the mouth and nose, and include sensations like heat irritation from a spicy pepper, non-heat irritation from wasabi, cooling from menthol, and astringency,2 which will be talked about in greater detail later in this section.

Taste and olfaction are two very important aspects of sensory perception. The term “taste” is often misunderstood as many people confuse it with olfaction. Taste refers to the five basic tastes: sweet, salty, sour, bitter, and umami. In wine, the tastes sweet, sour, and bitter are usually perceived. There are two forms of olfaction: orthonasal and retronasal. Orthonasal olfaction is what we would typically think of as our sense of smell. Orthonasal olfaction allows us to perceive aromas directly through our nose. Retronasal olfaction occurs as volatiles are released from a food or 2 beverage while it is in mouth. These volatiles travel at the back of our mouth into the nasal cavity and bind there to some of our olfactory receptors just as they do in orthonasal olfaction.3 The perception of these volatiles is often what we may refer to as taste. In a wine, for example, we may say that it “tastes” like citrus. This is actually the retronasal perception of volatile compounds associated with citrus aroma. Our taste buds pick up the sweet, sour, or bitter tastes within the wines but the flavors and aromas are actually identified by olfactory receptors located in the olfactory epithelium.

Last, it is also important to understand mouthfeels, like astringency.

Astringency, the drying, puckering sensation perceived in the mouth, is caused by the binding of polyphenols with proline-rich proteins that are commonly found in saliva.4

Astringency is important to the overall sensory perception of wine as wines contain many polyphenols including caffeic acid, ferulic acid, and coumaric acid. Red wines are especially known for their astringent properties as they tend to have a higher concentration of polyphenols than white wines.

1.2 Wine composition and its impact on sensory perception

The flavor of wine, the combined perceptions of smell, taste, and mouthfeel sensations, is often described as being very complex. This is, in part, due to the wide variety of wine chemical components as well as the wide range of concentrations of these compounds. Water is the largest component in wine, accounting for about 86%, followed by ethanol, ranging from 7-17% (v/v) depending on wine style. However, it is not for this mildly acidic ethanolic solution that wine has been appreciated for

3 centuries; rather it is for the minor and trace compounds that affect the flavors, tastes, and mouthfeels that wine is appreciated.

In the following sections, each of these groups and their effects on sensory perception are discussed. Summarized effects of protein, ethanol, glycerol, and phenolics on sensory attributes in wine can be seen in Table 1. These conclusions come from many studies and each study’s overall conclusions on the different components are included in the table.

Besides non-volatile components, the volatile aroma fraction, although low in concentration in a wine, is what some might argue the most important one. This fraction, contains the compounds that activate the olfactory receptors in our nose, triggering both ortho- and retronasal aroma perception. Complex mixtures, such as wine, have been shown to have considerable mixture effects. This means that both non-volatile and volatile components of the wine matrix are affecting each other through interactions. Further, these interactions affect wine perception on multiple levels: (i) on a chemical level it has been shown that matrix components affect the partitioning of certain aroma compounds into the headspace, and thus, aroma perception,5 and (ii) on a perceptual level, it is known that taste sensations of sweet and bitter suppress each other, i.e., the sweetness of a substance is decreased by the presence of a bitter tasting component and vice versa.6 A better understanding of how individual wine compounds as well as their interactions contribute to the aroma and how they impact sensory perception is therefore required to ultimately provide winemakers with recommendations how to improve wine sensory quality.7 Below, I

4 summarize relevant literature on the topic of how wine composition has been shown to affect sensory perception.

1.2.1 Trace matrix and aroma compounds

There are many groups of volatile aroma compounds that are found in wine, such as ethyl and acetic esters, organic acids, alcohols, fatty acids, terpenes, phenols, and norisoprenoids, and they are known to elicit flavor and aroma. For example, 2- phenylethyl acetate is a commonly found compound, resulting from fermentation, that has a rose-like odor and has been shown to affect floral aroma and flavor perception.8,9 Understanding how these volatile aroma compounds behave and interact is vital to understanding the overall aroma of wine.

Organic acids, in particular the six acids that make up about 95% of the total acid in wine, (acetic acid, citric acid, malic acid, lactic acid, succinic acid, and tartaric acid) affect primarily sour taste perception.10 Malic acid comes from the grapes themselves. Some malic acid gets metabolized to lactic acid during fermentation.

Tartaric acid also comes from the grape but it does not get metabolized during fermentation. The remaining three acids, citric, succinic, and acetic acid, are all produced through alcoholic fermentation. Smaller, but still important, organic acids include butyric, decanoic, hexanoic, isovaleric, isobutyric, octanoic, and propanoic acid. These acids contribute to some of the rancid and cheese-like odors found in some wines.10

1.2.2 Ethanol

Ethanol is a simple alcohol (Figure 1) and the primary alcohol found in alcoholic beverages consumed by humans. Ethanol has been shown to affect the sensory perception of wine in multiple ways: (i) Ethanol activates TRPV1 receptors and thus affects the warm mouthfeel or heat of the wine.11 As ethanol levels increase, the warm or hot mouthfeel of a wine also increases,6,12–15 as a direct effect of the TRPV1 activation.11

(ii) Increasing the ethanol content of a wine has also been shown to lead to greater bitter taste.6,12,14–16 Ethanol activates a bitter taste receptor, TAS2R, thus increasing the perception of bitter taste.11 The increase in bitter taste can also lead to indirect effects on other sensory attributes, most likely through mixture suppression and enhancement. For example, increased levels of bitterness in wine caused by increasing levels of ethanol have been shown to lower attributes like sour taste14,17 and fruity aromas (suppression),14,18 while increasing attributes like chemical, spicy, and herbal aroma, and woody, chemical, and spicy flavor (enhancement).14,18 A study by King and Heymann in 2014 showed that increasing ethanol concentrations in a white wine led to increased overall aroma intensity ratings, likely due to the increases in the aroma attributes chemical, spicy, and herbal, as suggested earlier.13

(iii) Astringency is another important sensory attribute that can be affected by ethanol levels. Astringency and drying/rough mouthfeel have been shown to increase with increasing levels of ethanol, potentially due to ethanol dehydrating the oral epithelia, as suggested by Gawel et al. in 2013.6,15,19

Figure 1. Structure of ethanol. Structure is taken from PubChem.20

(iv) Ethanol also affects the partitioning of volatile aroma compounds into the headspace of a wine, thus potentially affecting aroma perception. A study by Robinson et al. studied how ethanol, glucose, catechin, glycerol, and proline affect partitioning of volatile compounds from the wine matrix into the headspace. Partitioning of volatile compounds from a matrix into the headspace above is required for orthonasal (and retronasal) olfaction, thus, understanding how this partitioning is affected by the wine matrix components is critical. All of the factors, along with the interactions between ethanol and glycerol, glycerol and catechin, and ethanol and glucose, had a significant effect on the partitioning of volatile compounds. Ethanol particularly decreased headspace concentration of polar compounds such as ethyl octanoate, 2-isobutyl-3- methoxypyrazine, ethyl benzoate, -damascenone, -ionone, and -ionone.

Focusing on the ethanol-glucose interaction, further experiments showed that ethanol decreased the concentration of all volatile compounds at higher levels, while glucose showed the opposite effect, increasing the concentration of volatiles in the headspace at higher levels.21 However, this study did not investigate whether the observed changes, measured instrumentally, could affect sensory perception of wines. It is, however, likely that varying the concentration of volatiles in the headspace will affect sensory perception, particularly, orthonasal olfaction.

For example, Serafini et al. found that increasing ethanol levels led to less tannin precipitation, indicating that ethanol reduces the interactions between tannins

7 and proteins, thus potentially affecting astringency perception. The authors speculate that these results could be due to the ethanol disrupting the hydrophobic interactions or modifying the random coil or collagen-like helices of the proteins.22

Overall, ethanol has been shown to both directly and indirectly affect sensory properties in wine. Besides its bitter taste and its warm and astringent mouthfeel, ethanol levels also impact volatile partitioning, aroma perception, and perceptual mixture effects with both suppression and enhancement of sensory attributes.

1.2.3 Glycerol

Glycerol, or propane-1,2,3-triol, is the most common sugar alcohol found in wine (Figure 2). Compared to ethanol’s effect on sensory properties, glycerol’s effect on sensory perception is smaller, with studies indicating that glycerol at levels typically found in wines (7-10 g/L) most likely affects sweet taste but not viscous mouthfeel, despite its viscous, syrup-like texture. Glycerol is thought to affect viscosity of wines once it reaches a level of about 25.8 g/L, well above the observed levels in white or red table wines.23

Figure 2. Structure of glycerol. Structure is taken from PubChem.20

Both a 1984 study by Noble and Bursick and a more recent one by Gawel and

Waters reported a difference threshold for glycerol of 5.2 g/L for sweet taste and about

26 g/L for a noticeable difference in viscosity. These results indicate that glycerol

8 primarily contributes to the sweet taste but not the viscous mouthfeel in table wines, where glycerol levels range from 1-10 g/L.

Contrary to these findings, Jones et al. found that increasing glycerol by 10 g/L, from 0 g/L to 10 g/L, led to significant differences in perceived viscosity in a model, white wine. There were significant interaction effects between ethanol, polysaccharides, and protein and between ethanol, volatiles, and protein present as well, suggesting that glycerol alone may not have been responsible for the differences in perceived viscosity. The same study also found that increasing the amount of glycerol also led to more overall flavor. “Overall flavor” was an attribute used by the trained panel to describe the overall intensity of the flavors in the samples.6 This could be an indirect effect of the increased sweetness introduced by higher levels of glycerol.

1.2.4 Phenolics

The diverse group of polyphenols make up another main wine matrix class that affects sensory perception. In general, phenolics are compounds with hydroxyl groups attached to aromatic rings. A simple phenol has one aromatic ring while polyphenols have multiple. The structure of a simple phenol is seen in Figure 3A. Polyphenols refers to compounds with multiple phenol rings within a structure. The structure of a polyphenol can be seen in Figure 3B. Major sources of phenolics in wine grapes are the grape skins and seeds, leading to significantly higher polyphenol levels in red wines compared to white wines. In white wines, phenolic concentration is typically around 200 mg/L compared to about 2000 mg/L in red wines. Despite red wines having a higher total phenolic concentration, phenolics in white wines still come from

9 the grape pulp, seeds, and skin. These phenolic compounds also include tannins.10

Tannins are a major polyphenol in wines and they are very important to some sensory attributes, like astringency. In white wines, the phenolics include hydroxycinnamates, flavanols, and flavan-3-ols, and these groups have been shown to affect the sensory perception of bitterness, astringency, as well as wine color. Structures of these types of compounds can be found in Figure 3. In white wines, oxidation of phenolic compounds like hydroxycinnamates leads to more brown colors as hydroxycinnamates oxidize quickly into quinones. The quinones react with flavanols to form the brown polymeric pigments.24 Tannins and catechins have been shown to be the phenolic compounds that contribute most to the bitterness of wines.25,26

A B C

D E F

Figure 3. Structures of phenolic compounds. (A) Structure of the simplest phenolic compound, phenol. (B) Structure of a polyphenol, reservatrol. (C) Structure of caffeic acid, tartrate ester, a hydroxycinnamic acid.10 (D) Structure of epicatechin, a flavan-3- ol. (E) Structure of quercetin, a flavonol. (F) Structure of malvidin-3-glucoside, an anthocyanin. Structures are taken from PubChem unless otherwise cited.20

Phenolic compounds have been shown to affect a couple of different sensory attributes that are important to wine. Astringency and bitterness have both been shown to increase with increasing levels of phenolic compounds, like tannins and catechins.14–16,27 The effects on these attributes are likely the result of phenolic compounds interacting with saliva after taking a sip of wine. Indirect effects of increased bitterness and astringency are lower sweetness and fruity flavors.14,27 While it may be expected that fruity flavors would decrease with more bitterness, Sun et al. showed that wines higher in phenolic content had more fruity notes. In this study though, there may have also been other factors present that led to an increase in fruity notes.28 Finally, phenolic compounds isolated from Riesling and Chardonnay wines, namely caftaric acid, grape reaction product, coumaric acid, tyrosol, catechin, quercetin, and dihydroquercetin rhamnoside, were also shown to increase perceived viscosity, and at low ethanol levels (11.4% v/v), an increase in hot mouthfeel was noted when these phenolics increased. A similar, yet more subtle, increase in hot mouthfeel was also found for ethanol levels of 12.6% (v/v). This effect was potentially smaller because of the increase in ethanol.15 This implies that the hot mouthfeel perception in white wines cannot be attributed to a single factor, such as pH, ethanol, or phenolic content, but rather that these components interact with each other and can both enhance as well as counteract each other’s actions.

1.2.5 Protein

Proteins in wine (Figure 4) originate from both grapes and also the yeast. In white wines, the concentration of protein is typically in the range of 30-275 mg/L.10 In white wines, protein haze is a primary concern with protein, this most work on proteins

11 in white wine has been focused on it. The denaturation of wine proteins leads to aggregation and a hazy appearance. This is undesirable as consumers expect white wines to be clear and not hazy.29 In general proteins are able to bind to tannins and other polyphenols within wine. High protein concentration in grapes has been show to lead to a lower tannin concentration in finished wines and this is used in winemaking to decrease the overall concentration of tannins and proteins in wine,10 but also affects sensory properties.

Proteins found in wine have no flavor themselves.30 Wine proteins’ main sensory effects come from their ability to bind to tannins in the wine.10 Proteins precipitating polyphenols lowered astringency and bitterness in wines as shown by

Fukui et al.31

Figure 4. Ribbon diagram of wine protein F2/4JRU. Diagram is taken from Marangon et al.32

The 2003 study by Fukui and Yokotsuka characterized proteins in both red and white wines. They found that white and red wines typically have a protein concentration between 30 and 100 mg/L. In addition to providing this information, they also gave some insight on how proteins present in wine could affect some sensory attributes such as decreasing astringency and bitterness caused by polyphenols by forming insoluble protein-tannin-polysaccharide complexes.31 There was no sensory research done on these wine samples.

Indirect sensory effects of proteins also include changes in aroma perception.

Attributes like estery, cheesey, and floral aromas increased with increasing levels of protein in a model white wine.6 Finally, similar to phenolic compounds, Sun et al. showed that wines with higher protein content had more fruity notes, but these observations may or may not be related.28

1.2.6 Interactions

In addition to the main effects that each of the discussed matrix components have on the sensory properties of wine, these components typically interact with each other, and subsequently, also affect sensory perception.

Jones et al., studied interactions between proteins and polysaccharides extracted from a Chardonnay wine, glycerol, ethanol, and volatiles and showed that they affect sensory perception. Several two- and three-way interactions showed significant effects on sensory attributes like floral aroma, bitter taste, and drying. Two- way interactions significantly affected estery aroma, metallic taste, and drying mouthfeel. For estery aroma, the interaction between polysaccharides and ethanol was found to reduce estery aroma as polysaccharides increased from 0 mg/L to 112

13 mg/L at the higher ethanol level (13% v/v) but this effect was not present when the ethanol level was low (11% v/v). Metallic taste ratings were affected by the interaction between ethanol and polysaccharides with increasing ethanol levels leading to increased metallic taste, but only in the absence of polysaccharides. When polysaccharides were present, ethanol had no effect on metallic taste. Last, drying mouthfeel was significantly affected by the interaction between ethanol and protein, where ratings were significantly higher for the high level of ethanol and protein than they were when both factors were low.

For overall aroma, overall flavor, and roughing mouthfeel, three-way interactions were found to significantly affect their intensities: overall aroma was significantly affected by the combined effects of ethanol and glycerol, with either protein or polysaccharides: Overall flavor was significantly affected by the interaction of ethanol, polysaccharides, and the volatile fraction, while roughness ratings were affected by the combined effects of glycerol, polysaccharides, and proteins; this was the only interaction that did not involve ethanol.6 The implications of this study are that major wine matrix components, such as ethanol, glycerol, polysaccharides, and protein, along with their interactions with each other as well as with minor aroma components greatly affect the sensory attributes of wine.

In the study done by Villamor et al. using a model red wine, the interactions between the experimental factors (ethanol, tannins, fructose) also lead to perceivable sensory differences. For example, the interaction between ethanol and tannins led to significant differences in caramel aroma and drying, although the difference in drying was mainly the result of the tannin concentration. Preliminary analysis of significant

14 interactions showed that they were dominated by main effects and therefore, were not discussed in the article.14

A study by Frost et al. investigated the effects and interactions of tannins, acidity, and ethanol on the taste and mouthfeel of red wine. This study used both static descriptive analysis (DA) and a temporal evaluation method, namely temporal dominance of sensation (TDS), to capture differences also over the period of wine consumption. Two significant interactions were found here. The interaction between acidity and tannin concentration significantly affected both sour taste and hot mouthfeel. For sour taste, samples with higher tannin concentration (~1200 mg/L) were perceived as more sour when acidity was high (pH=7). The interaction effect of acidity and tannin concentration was not discussed.27

Gawel et al. looked specifically at the mouthfeel effects caused by phenolics, and their interaction effects with pH and ethanol. Hydroxycinnamate phenolics, extracted from white Riesling and Chardonnay wine, were found to increase astringency at a higher pH level (3.3) but had no effect on astringency at the lower pH level of 3.0.15 This could be due to the direct effect that pH has on astringency, as shown by Obreque-Slier et al. where they demonstrated that wines with a pH of 3.5 were more astringent than wines with a pH of 7.0.33

Table 1. Matrix factor effects on sensory attributes Matrix Component Effect Ethanol Increase hot mouthfeel 6,12–15 Increase bitter taste 6,12,14–16 Decrease sour taste 14,17 Decrease fruity aromas 14,18 Increase chemical, spicy, and herbal aromas 14,18 Increase woody, chemical, and spicy flavor 14 Increase overall aroma 13 Increase astringency 6,15,19 Decrease concentrations of more polar compounds in the headspace of wine 21 Decrease tannin precipitation 22 Glycerol Increase sweet taste 23 No effect on perceived viscosity up to ~ 26 g/L 23,34 Increase overall flavor 6 Increase perceived viscosity 6 Phenolics Increase astringency 14,15,27 Increase bitter taste 14–16,27 Decrease sweet taste 27 Decrease fruity flavors 14 Increase fruity notes 28 Increase perceived viscosity 15 Increase hot mouthfeel 15 Protein Decrease astringency 31 Decrease bitter taste 31 Increase estery, cheesey, and floral aromas 6 Increase fruity notes 28

1.3 Methods to characterize wine sensory properties

Descriptive analysis (DA) is a commonly applied sensory method that provides a qualitative and quantitative sensory profile of a sample. In DA, a panel of participants is trained in several group sessions on sensory attributes and references relating to the sample of interest. Panelists are trained on basic taste and mouthfeel attributes along with aroma references that may help them describe the sample. They also

16 receive training on the scaling method that is used. The panel receives training as a group in order to ensure that every is using the scale in the same way. After training is deemed successful, the panel qualitatively and quantitatively assesses the sensory profile of the sample at a single point in time. These evaluations are done individually and are usually repeated two or three times.35 Descriptive analysis has been applied to wine in many different studies, including studies looking at the effects of matrix components on the sensory properties of wine.4,6,8,13,14,16,36

Although DA has been extensively and successfully used to characterize how different treatments affect wine sensory perception, it typically represents a measurement at a single point in time or a mental average. Thus, DA may miss temporal changes in complex and evolving samples, like wine. To better capture these dynamic changes during wine consumption, e.g., a single sip, temporal methods of characterization have been developed and applied to wine and other foods.37–41

One of the original methods of temporal evaluation is Time Intensity, or TI. In

TI, trained participants rate the intensity of a single attribute, for example, bitter taste, over a period of time. TI methods can be discrete or continuous. In a discrete TI method, participants are asked to rate the intensity at specific points in time, for example, every 10 seconds. In a continuous TI method, participants are constantly rating the intensity of an attribute. Continuous TI typically provides greater detail of the changes in an attribute over time compared to discrete TI. A drawback of both discrete and continuous TI is that usually one attribute is evaluated at a time. This means that for a full temporal characterization in a complex product, TI takes a very long time and is both labor- and cost-intensive.42 An example of TI for wine characterization involves

17 bitterness in white wines. This study found that TI was most useful when exact information regarding the temporal course of the bitterness was required but these intensities are often exaggerated since only one attribute is evaluated at a time.37

To overcome the drawbacks of TI, temporal dominance of sensation (TDS) was developed in the early 2000s and is now a common temporal method for wine evaluation. In TDS, trained or untrained participants are presented with a list of sensory attributes relevant to the sample. During evaluation, participants are asked to select the most “dominant” attribute. As this is a temporal evaluation method, participants change their selection of the dominant attribute throughout the evaluation period. In TDS, the most “dominant” attribute is not necessarily the most intense. The

“dominant” attribute is defined as the one that “catches the attention at a given time”.43

A limitation of TDS is that it only collects data for the most dominant attribute. This means that in a complex sample, like wine, small, yet important, changes in attributes are not necessarily captured.

Several studies compared TDA to either DA or TI. For example, Sokolowsky and Fischer compared DA to both TI and TDS, to characterize bitterness in commercial white wines. The DA panel found that glucose, fructose, glycerol, ethanol, and malic acid all significantly affected the bitter intensity of the wines, which increased with increasing levels of ethanol and glycerol, and decreased with increasing levels of glucose, fructose, and malic acid. Even though glycerol is typically regarded as a sweet taste contributor in wine, it was positively correlated to ethanol, which the authors thought would explain the positive correlation between glycerol and bitterness.37 Comparing the DA results to the two temporal methods, the authors

18 concluded that DA as a single point measurement was not able to fully explain differences in bitter taste. The TI method was found to give exact measurements of bitter intensity at any given time point. The TI results were significantly influenced by the impact of the ethanol. The study also suggests that the bitter intensity ratings could be exaggerated due to the focus on one attribute and resulting dumping effects.37

Dumping effects result from participant that is not able to rate an attribute that they are perceiving, thus “dumping” the attribute into the provided rating scale.44 Therefore, in this study, only asking about bitter taste intensity may have led to participants exaggerating the intensity by “dumping” other intensities in with it. Last, the TDS method prevents dumping as changes in the dominant attributes can be recorded, however, it does not provide intensity ratings as captured in DA or TI. Importantly,

TDS was able to show that the wines did not only differ in bitter taste intensity, as found by DA, but more interestingly, that the bitter taste duration and the time of dominance differed significantly between the wines. The authors concluded that not only bitter intensity, but also onset and persistence differed between the wines, thus, leading to bitterness differences. This study has shown that only the combination of static and dynamic evaluation was able to fully explain sensory differences in bitterness between the wines.37

In another study on red Merlot wines, Frost et al. compared DA to TDS to evaluate the effects of different maceration techniques on sensory perception. The DA panel was able to detect differences in red fruit aroma, pepper spice aroma, aldehydic aroma, alcohol aroma, bitter taste, hot mouthfeel, and astringent texture, with extended maceration generally leading to more red fruit aroma, aldehydic aroma,

19 bitter taste, hot mouthfeel, and astringent texture. Submerged cap without extended maceration led to less alcohol aroma and more pepper spice aroma.

For the TDS portion, panelists evaluated each wine for a total of 90 seconds, where wine was expectorated after 25 sec. This study found that DA better captured the initial astringency differences in the samples but that TDS captured the bitter persistence once the astringency died down towards the end of evaluation.38 This is of particular note as maceration treatments significantly affected the bitter taste persistence, with longer maceration leading to increased persistence. Similar to

Sokolowsky and Fischer, the dynamic method provided insight into sensory differences that can be linked to winemaking.

In a study on the effects of tannins, acidity, and ethanol in red wines, DA and

TDS were applied as well.27 As described in section 1.2, DA found that the factors had no effect on aroma but tannin and acidity levels did affect taste and mouthfeel attributes. As tannin concentration increased, sour taste, bitter taste, astringent mouthfeel, and drying mouthfeel also increased while sweet taste decreased. As the acidity increased, sour taste increased and bitter taste decreased. TDS, conducted for 90 seconds, revealed that sweet taste was the initial dominant sensation. Sour taste was most dominant just before expectoration, while bitter taste was dominant towards the end of evaluation. Astringency was perceived as the most dominant sensation throughout evaluation, likely caused by the increasing tannin levels. The authors were able to attribute the temporal differences to specific factors, for example, temporal differences in bitter taste were caused by acidity levels, while temporal differences in sour taste were correlated to tannin concentrations.27

The effect of ethanol content on the static and dynamic sensory perception of a Merlot and a Syrah wine was studied by Meillon et al.45 The wines differed only in ethanol content (10-14% v/v) as produced through reverse osmosis. Out of the 17 sensory attributes, the DA panel found significant differences in three sensory attributes (alcohol aroma, heat, and acid) for the Merlot wine, and 5 attributes

(undergrowth aroma, spice aroma, red fruit, sweet taste, and heat) for the Syrah.

Varying the ethanol content in the wines led to more alcohol aroma, blackcurrant aroma, heat, and texture.45 Using TDS to characterize the changes due to the dealcoholized treatments, results differing from those of the DA were captured. The

TDS method differentiated the different wines by twice the number of attributes compared to DA, including differences in dominance rates for fruity, heat, astringent, and bitter, throughout evaluation, acid and sweet taste at the beginning of evaluation, and woody, blackcurrant, and pungent flavors at the end of evaluation. TDS did not detect all the sensory differences that DA picked up, indicating that TDS is a complementary method and should not be used on its own to fully characterize a complex wine sample. This is likely due to the fact that participants are only allowed to select a single dominant attribute at any given time in TDS.45

A way to overcome the shortcomings of both TI and TDS is the latest development of descriptive temporal methods, temporal check-all-that-apply

(TCATA). The TCATA method uses the same principles of the check-all-that-apply

(CATA) method, but it extends the period of evaluation time. This turns the static CATA method into a dynamic method. In TCATA, the trained or untrained participants are given a list of attributes relevant to the sample. Studies have shown that list length

21 has no effect in CATA studies,46 but due to the more dynamic nature of TCATA, this list is typically limited to 10-15 attributes, similar to TDS. The participants select and update the attributes that apply to the sample continuously for an extended period of evaluation, e.g., a single sip or multiple sips. Participants are free to select or deselect as many attributes as they’d like. This is unlike TDS, where participants can only select one attribute at a time. Turning CATA into a temporal method allows researchers to track changes in the sensory attributes over time.47 It is important to remember that unlike DA, which provides intensity ratings of attributes, TCATA provides frequency data. While this is not the same data type, a study compared DA to check-all-that- apply (CATA), using Pinot Noir wines, and found that the results from the two methods were very comparable. While DA provides intensity data and CATA provides frequency data, this study showed that the two are often correlated. For example, the highest intensity attribute was red berries and the lowest intensity was bell pepper.

The most frequently cited attribute was fruity and the least frequent was bell pepper, showing that the two evaluation methods produced similar results even though DA provides intensity data and CATA provides frequency data.48 Although this correlation was not done with TCATA, it implies that CATA and DA lead to similar outcomes.

A drawback of TCATA is having to aggregate data over participants and time segments. For analysis of TCATA data, the citation proportions are compared across samples. The proportion of panelists that selected a specific attribute at a given time is analyzed. Aggregating across participants requires participant agreement.

Participant agreement can be checked to ensure data quality in TCATA.49 Data must also be aggregated over time segments to make conclusions on the temporal

22 evolution of attributes. For example, aggregating data into three time segments splits data into the beginning, the middle, and the end of evaluation, however binning of data requires some thoughts, to find the balance between large bins that are small enough to capture temporal changes.39 Typical bin times can range from 3 to 30 seconds.49

Due to TCATA being relatively novel, only a few studies on wine have been conducted. A study comparing DA to TCATA by McMahon et al. evaluated sparkling wines, differing in carbonation level from 0.0 to 7.5 g/L CO2. DA revealed that increasing CO2 in the sample led to increases in the mouthfeel attributes bite, burn, numbing, after-numbing, carbonation, prickly, pressure, foamy, tingly. CO2 level also increased sweet taste, vanilla aroma, and caramel flavor.39 For the TCATA portion, the same DA panel and two new assessors were first trained to use TCATA, and then evaluate the wines using 8 sensory attributes, selected from the attributes used in the

DA (bite/burn, numbing, carbonation/bubble pain, prickly/pressure, foamy, tingly, sour, and bitter). For evaluation, panelists took a sip of the sample, expectorated after ten seconds, and continued evaluation for up to 125 seconds. Overall, DA and TCATA results were found to be highly associated for the first 30 seconds of evaluation.

Compared to the DA data, TCATA was found to provide a more detailed profile of the different sparkling wines, as it captured temporal changes over time. Specifically,

TCATA better explained the dynamic nature of carbonation in sparkling wines, while

DA revealed that the intensities of the mouthfeel attributes were the main drivers of

39 variation between the different CO2 treatments. While this study was not only among the first to use TCATA with wine, but it also compared TCATA and DA, and found

23 similar to the DA-TDS comparison studies,27,37,38,45 the advantages of using both static and dynamic methods for a full sensory characterization of complex samples.

The effect of ethanol (10.5% and 15.5% v/v) on wine finish was studied by

Baker et al., using temporal check-all-that-apply (TCATA). For evaluation, a trained panel selected from a list of 10 sensory attributes (red fruit, dark fruit, spicy, green, earthy, sour, bitter, astringent, hot/ethanol burn, and body). Increased ethanol led to increased astringency at the beginning and end of evaluation, and increased hot/ethanol burn and bitter taste throughout the evaluation. Increasing the ethanol content decreased sour taste and red fruit at the middle and end of evaluation, but led to more dark fruit at the beginning of evaluation, and to less green flavor at the middle of evaluation. Finally, increased ethanol levels led to more spice flavor at the beginning and middle of evaluation, while earthy flavor was unaffected by ethanol content. Based on this study, ethanol affects both static and dynamic perception of wine.40

Last, comparing TDS to TCATA, using a variety of food products and both trained and untrained participants, Ares et al. concluded that TCATA provided higher product discrimination than TDS. This is likely due to the fact that TCATA allows for the identification of multiple attributes at once, instead of having only one dominant attribute at a time, like in TDS. This finding also implies that for samples where multiple sensory attributes are present and may be dominant at the same time, TCATA may be more appropriate than TDS. TDS however, may be more appropriate for studies that require the identification of attributes that catch the assessors’ attention at any given moment during evaluation.41

In this study, descriptive analysis (DA) was paired with temporal check-all-that- apply (TCATA) to provide a full sensory characterization of white, model wines that differed in ethanol, glycerol, protein, and phenolic concentration.

1.4 Model Wine

Model systems allow researchers to accurately and precisely control their factors of interest, while trying to stay as close to a real system as possible. A model wine typically contains water and alcohol. This water-alcohol solution is buffered at wine pH (3-4), and typically contains other minor wine components such as organic acids, sugars, proteins, phenolic compounds, glycerol, and aroma compounds. Model wines have been used to systematically study the interactions between wine matrix components, as it is not easy to control these factors in an accurate fashion through traditional winemaking procedures without changing other factors. In the following, model wine studies are summarized.

The previously mentioned study by Jones et al. utilized a model wine system that mimicked a white Chardonnay wine.6 The model wines consisted of water, malic acid (0.6 g/L), tartaric acid (enough to achieve pH 3.2), potassium hydrogen tartrate

(5 g/L), potassium metabisulfite (94 mg/L), ethanol (11 and 13% v/v), glycerol (0 and

10 g/L), carboxymethylcellulose (2 g/L), proteins reconstituted from wine (0 and

112 mg/L), and volatiles representative of those found in a traditional Chardonnay wine (70 and 130% v/v).6 As described in section 1.2, this study varied ethanol, glycerol, protein, polysaccharides, and volatiles and observed the changes in the aroma, flavor, taste, and mouthfeel properties.

A red, model wine system was utilized by Villamor et al. where ethanol (0, 8,

10, 12, 14, and 16% v/v), fructose (200 and 2000 mg/L), and tannins (500, 1000, and

1500 mg/L) were varied. In addition to ethanol, fructose, and tannins, tartaric acid and volatile flavor compounds reminiscent of a red wine were added at a constant level.14

Using this model system, the effects of ethanol, tannin, and fructose concentrations were studied, as described in section 1.2 in more detail.

In this study a model wine system, reminiscent of a white, hybrid Traminette wine was chosen to systematically evaluate the effects of major and minor wine components (ethanol, glycerol, protein, and phenolics) on static and dynamic sensory perception.

1.5 Hybrid Grapes

While most wine research is done using wines made from Vitis vinifera grapes, hybrid grapes remain largely important to the world’s wine industry. While more popular and widely used, Vitis vinifera grapes can be difficult to grow in cooler climates or in areas with high disease pressure, such as the Northeast of the United States.

Hybrid grapes tend to fare much better in these regions.50

A hybrid grape is one that is produced from a cross of two different Vitis species. For example, Concord grapes are interspecific hybrids of Vitis vinifera and

Vitis labrusca parents, where the non-Vitis vinifera parent provides cold hardiness or greater yield.10

Most of the research on hybrid wines is done on red hybrids. Hybrid wines typically have lower concentration of tannins compared to Vitis vinifera wines, which

26 causes hybrid wines to be less astringent, shorter in finish, and limits the aging potential of red hybrid wines.51 Red wines with low astringency are typically correlated with lower prices and consumer liking.36 The idea that hybrid grapes have a lower tannin concentration was tested in 2014 by Springer and Sacks. Using a protein precipitation method to determine the tannin concentrations of wines, they found that wines produced by hybrid grapes had at least a 4-fold lower concentration of tannins than wines produced from Vitis vinifera grapes. They attributed this difference to lower tannin levels in the hybrid grapes themselves.50 However, follow-up work showed that while hybrid grapes may have a lower tannin concentration, their higher protein content, in particular of cell-wall proline-rich proteins, results in greater precipitation of tannin-protein complexes at the beginning of fermentation, thus, leading to lower tannin concentrations in the finished red, hybrid wine.50,52,53

Another difference between red Vitis vinifera and hybrid wines comes in the glycosylated anthocyanin content, with the content being much higher in hybrid wines.54,55 The combination of low tannins and high diglucoside anthocyanins limits the formation of color components in hybrid wines.56 All of these are reasons why Vitis vinifera wines are typically regarded as being higher in quality than hybrid wines.

For white hybrid wines, very limited literature is available, particularly with regards to their sensory characteristics. Using Traminette, an interspecific hybrid of the Vitis vinifera Gewürztraminer and the French American hybrid Joannes Seyve

23.416 commonly grown in the Northeast of the United States and Canada, Skinkis et al. studied the effects of sunlight exposure on monoterpenes and resulting sensory characteristics. Sun-exposed clusters produced fruit with 30% more volatile terpenes

(i.e., linalool, cis-rose oxide, citronellol, geraniol, and -terpineol) compared to the heavily shaded fruit. The increase in volatile monoterpenes due to the increased sunlight exposure of the cluster was accompanied by significant increases in the aroma attributes of linalool, rose, and spice and a more intense color of the wines, while no changes in tastes or mouthfeels were observed.57

White hybrid Seyval blanc wines from Missouri were evaluated with DA by

Andrews et al. Seyval blanc wines from different regions in Missouri were compared and significant differences between different regions were observed for fruity, oxidized, and vegetative aromas, indicating that viticultural regions and age have a significant impact on the sensory properties of white wines produced from hybrid grapes.58

Another study on white, hybrid Chardonel wines by Mirarefi et al. was conducted to develop a sensory lexicon. In total, 23 attributes were used to describe the twelve wines, including lemon/lime, orange, clove, Jonagold apple, vanilla, unsalted butter, oak, and floral aromas, Granny Smith apple, pear, grape, sweet, sour, and bitter were used to describe flavors and tastes, and astringent and viscous mouthfeels. Chardonel wines also showed grapefruit and cheese aftertastes, leading to the creation of a Chardonel wine sensory attribute wheel.59 Overall, white hybrid wines show similar sensory attributes to white Vitis vinifera wines with fruity and floral aromas and flavors, sweet, sour, and bitter tastes, and warm/hot and astringent mouthfeels however, their sensory fingerprints differ from each other and show similar dependencies on winemaking and grape growing practices.

1.6 Linking Instrumental to Sensory Data

Numerous literature reports have attempted to predict sensory attributes using instrumental data, such as HS-SPME-GC-MS data. This could be done to potentially bypass sensory analysis when quick answers about sensory characteristics are required, however, this approach assumes that sensory perception is the result of direct effects of chemical compounds, an assumption that is not necessarily true.

Escudero et al. linked chemical measurements of wine aroma compounds in five commercial red wines to sensory analysis, using both gas chromatography- olfactometry (GC-O) and DA. Forty-five individual odorants, found by GC-O in the commercial wine samples that were associated with DA sensory attributes. They report that the phenolic character was significantly correlated to 12 of the identified odorants, implying that a non-volatile sensation could be significantly affected by volatile aroma compounds.60 Such cross-modal interactions have been demonstrated in other foods and beverages, for example, between sucrose/sweet taste and vanilla,61,62 and sucrose/sweet taste and fruity aromas (For example, isoamyl acetate).63 A sweet taste-inducing compound such as sucrose seems to be a key factor in the perceived intensity of fruit flavors. While sucrose is not typically present in wines, these cross-modal interactions described by Escudero et al. indicate that compounds that induce sweetness, like glycerol, may also increase the perception of fruit flavors. The authors also found that ethanol suppressed fruity notes however, this suppression was not necessarily the result of the loss of fruity esters, as previously suggested, but rather the addition of the strong ethanol aroma. Last, both norisoprenoids and dimethyl sulfide enhanced the fruity aroma and flavor of the wine,

29 indicating additional cross-modal interactions between aroma compounds and aroma perception as dimethyl sulfide by itself smells like rotten egg or cabbage and not fruity.60

Often, Partial Least Squares Regression (PLSR) is used to correlate sensory ratings to chemical measurements. PLSR is a statistical method where a linear regression model is used predict variables. PLSR is also well suited to deal with collinear data. For example, Aznar et al. correlated gas chromatography-flame ionization detector (GC-FID) and gas chromatography-mass spectrometry (GC-MS) results to wine expert ratings of sensory attributes. Satisfactory models (explaining more than 45% of the original variance) were found for 9 of the 18 sensory attributes in the study, including the attributes wood-vanillin-cinnamon, animal-leather-phenolic, toasted-coffee, old wood-reduction, vegetal-pepper, raisin-flowery, sweet-candy- cacao, fruity, and berry fruit, while the remaining 50% of attributes could not be satisfactorily modelled.64 This indicates that at least 50% of wine sensory attributes are the result of complex interactions between aroma and non-aroma compounds and/or perceptual mixture effects.

Similar findings were reported by González Álvarez et al. for Godello white wine. A panel of 7 participants rated the intensities of 12 descriptors, including apple, melon, apricot, floral, citrus, pineapple, tropical, and pear. The study found that apple, melon, tropical, and toasted aromas were explained by increasing amounts of octanoic acid, isovaleric acid, and isoamyl acetate. Apricot, floral, and pear were explained by increasing amounts of isoamyl acetate, ethyl butyrate, and ethyl hexanoate, and decreasing amounts of isovaleric acid and octanoic acid. Citrus and

30 herbaceous were explained by decreasing amounts of isovaleric acid and octanoic acid. Finally, pineapple was explained by increasing amounts of isoamyl acetate and ethyl hexanoate, and decreasing amounts of isovaleric acid, octanoic acid, and methionol.8

Another study by Vilanova et al. used GC-FID and DA for the characterization of Albariño white wines. The best prediction model for fruity and floral aromas was based on the ethyl esters ethyl butyrate and ethyl hexanoate, the acetates hexyl acetate and 2-phenylethyl acetate, and the monoterpenes linalool and nerol. In the

DA portion of the study, fruity (citrus, apple, tropical, ripe fruit) and floral (flowers and cut grass) aromas were also found to have the highest intensities,9 indicating that

PLSR may be more successful at predicting sensory attributes with a high intensity.

A study by Goldner et al. in 2009 used HS-SPME-GC-MS data to try to predict sensory attributes of Malbec wines. Their main results suggested that different compounds, such as hexanol, diethyl succinate, ethyl octanoate, and furfural were correlated to specific attributes, such as plum, raisin, cooked fruit, and floral, and that a single compound is rarely able to fully predict, or be responsible for, a sensory attribute. This approach seems to be the most promising, as psychophysics showed that human aroma perception is rather complex, and rarely caused by a single aroma compound.18 Additionally, as summarized in this chapter, the partitioning of aroma compounds is heavily affected by wine matrix components, thus, it is unlikely that differences in sensory attributes are the result of differing concentrations in volatile aroma compounds alone.

A different approach to link instrumental data to sensory DA data was taken by

Cozzolino et al.: They utilized near infrared spectroscopy (NIR) measurements of

Riesling and Chardonnay wines in combination with a chemometric model to predict sensory attributes. As NIR utilizes a full IR spectrum, the idea is that sensory aroma perception resulting from multiple components would be better modeled. Good correlations (>70%) were observed for estery, lemon, and honey aroma attributes but not for passionfruit aroma, sweet taste, and overall flavor intensity. This indicates that relationships between sensory data and NIR data exist and can be used to assess some of the sensory properties in white wines.65

2 Descriptive Analysis and Chemical Analysis of Model Wine

Samples

2.1 Introduction

The complexity of wine aroma results in part, from the large number of volatile aroma compounds present in ppt to ppb concentrations, the non-volatile matrix components, including glycerol, phenolic compounds, and proteins, and the interactions between these over 1000 different components. It is known that the volatiles interact both with each other as well as with other matrix and non-volatile components and that these interactions affect the overall aroma.7,66 For example,

Jones et al. showed that the interaction between volatiles and protein had a significant effect on the floral aroma in a model white wine with the addition of protein at the high level of volatiles leading to less floral aroma.6

From a chemical point of view, wine is about 86% is water, 10-14% is ethanol,

0.9-1.1% is glycerol, 0.25-0.45% are organic acids, 0.05-0.1% are tannins and phenolics, and about 0.5% are other compounds including volatile aroma compounds.

After water, ethanol is a very important component in wine. The ethanol concentration in white wine can range anywhere between 6% (v/v) and 17% (v/v), depending on wine style, growing and winemaking conditions, and cultivar. It is known that ethanol impacts headspace concentration of aroma compounds as well as sensory attributes. Previous studies on both red and white wines have shown that as ethanol increases, sensory attributes like woody aroma, bitter taste, and burning mouthfeel also increase, while attributes like fruity, floral, and caramel aroma and

33 flavor decrease.6,67 One could speculate that these sensory changes are likely due to a decrease in headspace concentration of the responsible aroma compounds caused by the increasing ethanol concentration, however, Robinson et al. found that changes in the headspace concentration, as assessed by instrumental headspace solid-phase microextraction (SPME) coupled to gas chromatography-mass spectrometry (GC-

MS), did not fully explain the observed sensory changes.21 This indicates potential non-chemical, perceptual effects, such as mixture suppression, playing a role as well.

Another large wine matrix component is glycerol, a simple polyalcohol produced by yeast during fermentation. Glycerol is normally found in a concentration between 5 and 20 g/L in table wine. This concentration range is below the concentration where glycerol might begin to directly impact perceived viscosity,23,34 but it has been shown to impact sweet taste perception and the partitioning of the volatile aroma compounds into the headspace of the wine. The same study21 that showed the effects of ethanol on headspace concentration of aroma volatiles also found that glycerol increased the concentration of volatiles in the headspace of wine in the absence of ethanol. However, at glycerol concentrations between 5 and 50 g/L and in the presence of ethanol (10-18% v/v), the effect of glycerol on the volatile partitioning was not significant.21 This indicates a small effect of glycerol on headspace partitioning. Glycerol however, has been previously shown to affect sensory perception, specifically, it has been shown to increase overall flavor perception in a model white wine.6 Glycerol has also been shown to increase sweetness in white wines. Noble and Bursick showed that wines with the addition of at least 5.2 g/L of

34 glycerol had more perceived sweet taste, indicating that glycerol is directly linked to sweet taste.23

The large and diverse group of wine phenolic compounds presents another important group of the wine matrix. While in red wines, winemakers typically attempt to increase polyphenol extraction, primarily from the grape skins and seeds, the situation is different for whites, where typically, the concentration of phenolic compounds is low. Primary phenolic compounds in white wines are catechins, tannins, and hydroxycinnamates which oxidize quickly to quinones.24 Examples of catechins include catechin and epicatechin, examples of tannins include procyanidin A1 and B1, and examples of hydroxycinnamates include caffeic acid, coumaric acid, and ferulic acid. In white wines, the source for phenolics is the grape pulp, as white winemaking uses minimal to no skin contact, and thus, little to no skin polyphenol extraction.

Phenolic compounds from the grape pulp, like hydroxycinnamates, were thought to mask fruity and fresh flavors and accelerate oxidation in the wine. In special cases of white winemaking however, higher phenolic concentrations are reached through prefermentative treatments, normally, skin maceration followed by must hyperoxygenation. This increase in phenolic content has been shown to increase palate fullness.68 Higher levels of phenolic compounds in white wines have also been shown to increase the perception of astringency, viscosity, hot mouthfeel, and bitterness.15 How phenolic compounds affect the sensory perception of wine is a rather complex question. The sensory effects have been shown to not only be driven by the phenolic content but also by pH and ethanol concentration of the wine. A pH- dependent effect of phenolic compounds has previously been shown to impact hot

35 mouthfeel and bitter taste in white wines.15 The effect of phenolics on hotness was greater in the low alcohol treatment (11.4% v/v), where an increase in hot mouthfeel was observed with increasing phenolic concentrations, while for the higher ethanol content (12.6% v/v), this effect was not seen. This suggests that ethanol may mask the effect of phenolic compounds on hotness perception at higher levels. Further, the same study also found that an increase in phenolic compounds increased bitter taste, where the effect on bitterness was dependent on the source of the phenolics. Phenolic compounds (caftaric acid, grape reaction product, coumaric acid, tyrosol, catechin, quercetin, and dihydroquercetin rhamnoside) extracted from a Gewürztraminer wine and added to Riesling and Chardonnay wines provided a significant increase in bitterness, while phenolic compounds extracted from a Riesling and added back to a

Riesling base wine did not significantly increase bitter taste. Aroma perception was not studied.15 These results indicate that polyphenols directly and indirectly affect the sensory perception of bitter taste, astringency, and hot mouthfeel, and that this effect is modulated by ethanol content as well as pH. Prior work with white wine, showed that sourness, not surprisingly, decreased with increasing pH, however, sour taste intensity was also affected by ethanol, with sour taste decreasing with increasing ethanol at the low pH.16

Another major wine matrix component that affects sensory perception is the protein fraction. Wine proteins typically vary between 30 and 100 mg/L in white wine.31

Wine proteins typically range between 21-32 kDa and are heat-unstable. This instability can lead to haziness in white wines. Two major classes of wine proteins are chitinases, glucanases, and other pathogenesis-related proteins, and thaumatin-like

36 proteins.69 Proteins in wine can form insoluble protein-tannin-polysaccharide complexes and thus decrease the availability of polyphenols to interact with taste and mouthfeel receptors. This, in turn, has been shown to decrease astringency and bitterness.31 While wine proteins may impact flavor, their effect is typically indirect, as they by themselves do not impart flavor at the concentrations found in wines.29,70

Jones et al. showed that proteins affect flavors, tastes, and mouthfeels in addition to affecting the perceived viscosity due to their interactions with ethanol and glycerol. An increase in protein concentration led to an increase in floral, fruit-estery, and cheesy aromas along with overall aroma intensity. These results however, only held true at the low volatile levels (70% v/v volatile reconstitution mixture). At the high volatile level

(130% v/v volatile reconstitution mixture), an increase in protein concentration had no effects on sensory attributes. For viscosity, wine proteins were found to interact with ethanol, polysaccharides, and volatiles. All of these interactions led to a decrease in perceived viscosity. The underlying mechanisms are not clear, although one might speculate that complexation and/or denaturation mechanisms may be involved.

Further, increased protein led to an increase in metallic taste, while the absence of protein led to a slight cooling effect and a greater suppression of bitterness.6

Overall, it is important to understand that a change in a single sensory attribute could be caused by any of a number of wine components, or factors, or the interaction between factors. Components themselves may not directly impact sensory properties but their presence may greatly affect sensory properties of a wine through various interactions, as described above. All wine matrix components described above have been shown to be manipulated via viticultural and/or winemaking techniques. For

37 example, ethanol concentration can be modulated via grape sugar content but also yeast strain selection; glycerol is affected by grape material as well as yeast strain, and phenolic content is affected by winemaking style, such as skin contact or oxygen levels during must, juice, and wine handling. Last, protein levels can be modulated by winemaking techniques such as fining or bentonite additions.

This project is specific to a white, hybrid wine. Previously, most work has been done to study the aromas and flavors of white wines but not necessarily white, hybrid wine. Hybrid grape varieties differ in composition from traditional Vitis vinifera grapes and also behave differently during fermentation. Occasionally, hybrid grapes require unique or additional processing steps which may very well impact the volatile and non- volatile components, and subsequently, the sensory properties of the wine.

Another reason that the lack of hybrid research is surprising is because of how important hybrid varieties are for the United States wine industry. Hybrid varieties grow well in the North East and Midwest climates of the United States. Wines produced in these regions from hybrid grapes are a key economic factor and have the potential to be of equal quality to wines produced from Vitis vinifera grapes. It is therefore of high importance to develop a systematic understanding of hybrid wine composition, the interactions between wine matrix components, and how these factors affect the sensory perception of hybrid wines.

The sensory properties of wine have traditionally been evaluated by sensory descriptive analysis (DA). In DA, a group of trained panelists qualitatively and quantitatively assesses the sensory attributes of a product at a single point in time.35

Many of the cited studies have utilized DA to evaluate the sensory profiles of their

38 wine samples. Similarly, in this study, we utilize DA to evaluate our model wine samples to study the effects of the wine matrix components on the flavor profile of the samples. We chose a model wine system as opposed to a real wine to control our factors of interest namely, ethanol, glycerol, protein, and phenolics. Using a factorial model, we are able to accurately control the levels of the matrix components. Previous studies have had success utilizing model wine systems to study, in a controlled fashion, how wine components affect perception.6,14 Despite this existing body of work there is a lack of research that studies interspecific hybrid wines and their unique aromas along with how volatile and non-volatile components interact with each other and affect sensory perception. Thus, our model wine samples are modeled after the interspecific hybrid variety, Traminette.

Traminette grapes are an interspecific cross of the hybrid Joannes Seyve

23.146 and the Vitis vinifera cultivar Gewürztraminer. They are grown in locations like

Pennsylvania, New York, Ohio, and the Carolinas in the United States. In

Pennsylvania, Traminette grapes are grown all across the state, but they grow especially well in the southeast and northcentral regions.71 In 2013, it was estimated that hybrid grapes accounted for about 13% of the total grapes grown in Pennsylvania, thus are of great economic importance.72

Systematic studies on the sensory properties of Traminette wines are scarce:

The effect of sunlight exposure on wines made from Traminette grapes by Skinkis et al. utilized DA to evaluate the aroma, flavor, taste, and mouthfeel properties of these wines. Terms used to describe the aroma were apple, pear, apricot, grapefruit, dried fruit, vanilla, spice, rose, linalool, grass, and earthy. The study found that increased

39 sunlight exposure led to increased rose, spice, and linalool aroma perception, attributed to changes in the terpene concentrations in the wines. No changes in taste or mouthfeel attributes were reported.57

The composition of Traminette grapes was compared to that of Gewürztraminer and Riesling grapes by Skinkis et al. to determine if Traminette was a quality white, hybrid winegrape for research. Using headspace-solid phase microextraction-gas chromatography-mass spectrometry (HS-SPME-GC-MS) the volatile components of pressed grape juice were analyzed. They found that Gewürztraminer and Traminette have a very similar monoterpene composition (2-carene, cis-rose oxide, linalool oxide, and perillyl alcohol) in their juices. However, Traminette grapes grown in the same vineyard with Gewürztraminer had double the concentration of monoterpenes.

Traminette and Riesling grapes were also found to be similar in terms of monoterpenes, but the Riesling grape juice profile was dominated by norisoprenoids, such as vitispirane and -damascenone.73 This study did not compare the composition of wines nor did it include any sensory work.

Using a model wine that mimics the aroma fraction of a Traminette hybrid wine, this study’s hypothesis that varying the major matrix components (ethanol, protein, glycerol, and phenolics) will differently affect the sensory attributes as assessed by a trained DA panel, with ethanol having the largest effects. It is also hypothesized that these matrix components will interact with each other, modulating sensory perception of the model wines. For example, it is expected that the fruity flavors and sweet taste will decrease with increasing ethanol levels and that bitter taste and astringency will

40 increase as ethanol increases, but also that these direct effects can be modulated by interactions.

2.2 Materials and Methods

2.2.1 Determination of aroma fraction

To create a model wine aroma fraction reminiscent of a Traminette wine, the aroma composition of a research Traminette wine was analyzed by HS-SPME-GC-

MS. The Traminette wine used was made in 2017 at Penn State University from grapes grown in Pennsylvania, using a standard winemaking protocol as described in the work by Vernarelli.74

For analysis, 2 mL of Traminette wine were added to 20 mL headspace vials

(Restek, Bellefonte, PA) along with 2 g of NaCl (Dot Scientific, Burton, MI; >99%) and

50 L of internal standards 2-octanol and d8-napthalene (both 90+%, Sigma-Aldrich,

Bellefonte, PA) (10 mg/L prepared in 100% methanol, HPLC grade, Fisher Scientific,

Hampton, NH). The samples were incubated by an auto sampler (Gerstel Robotic,

Linthicum, MD, USA) at 30C for 5 minutes before extracting with a 2 cm

DVB/CAR/PDMS SPME fiber (Supelco, Bellefonte, PA) for 30 minutes at 30C. An

Agilent 7890 GC 5977B MS (Santa Clara, CA) was used for analysis, where the GC oven was kept at 30C for 1 minute, ramped up to 250C at a rate of 10C/min, with a final hold at 250C for 5 minutes. The SPME fiber was desorbed for 10 minutes in the

GC inlet, held at 250C in splitless mode, with the purge valve opening at 1.2 minutes.

The MS was operated in scanning mode between 33-350 amu, collecting 8.1 spectra

41 per second. The GC column used was an Rtx-Wax 30m x 0.25mm x 0.25m column

(Restek). Analysis of the wine was done in analytical triplicate.

Collected GC-MS data was further analyzed by PARADISe software (ver. 3.1)75 for deconvolution and compound extraction. Compound concentration of identified volatiles are expressed as IS equivalents (Table 2).

A stock solution of the aroma fraction was created in 200 proof ethanol (Decon

Labs, Inc, King of Prussia, PA) and stored at 4C until use.

Table 2. Aroma fraction composition. Concentration Odor Compound Structure (g/L) Descriptor8,9 Floral, Linalool 165 lavender

Rose, Geraniol 15 geranium

Ethyl Acetate 50,000 Fruity

Ethyl Apple, 1250 Hexanoate banana

Fruity, Hexyl Acetate 790 cherry, pear

Ethyl Pineapple, 1700 Octanoate pear, apple

Acetic Acid 40,000 Vinegar

Ethyl 1600 Grape, apple Decanoate

Phenethyl 450 Rose Acetate

Beta- 415 Rose, honey Damascenone

Green, Hexanoic Acid 8100 geranium, vegetable Phenethyl 14,000 Rose Alcohol

Sweat, Octanoic Acid 9100 cheese

Ethyl 450 Fruity Butanoate

Isoamyl Banana, 3200 Acetate apple

All compounds listed were purchased from Sigma Aldrich and were at least 95% purity. All chemical structures were taken from PubChem.20

2.2.2 Creation of model wines

Twenty-four unique model wines were created for the initial descriptive analysis portion of this study. All model wines contained the same aroma fraction concentration as listed in Table 2, but varied in levels of ethanol, caffeic acid, glycerol, and bovine serum albumin (BSA) (Table 3). The levels of ethanol (Decon Labs Inc.) (10, 12, and

14% v/v) and glycerol (>99%, Sigma-Aldrich, St. Louis, MO) (5 and 15 g/L) are typical for Pennsylvania white wines. As for protein, white wine protein concentrations is typically around 0.1 g/L.31 BSA (>98%, Sigma-Aldrich) was selected as the model protein as it is widely used as a model protein, it is safe for consumption as it is found in food and beverages like beef and milk,76–78 and has been used as a model protein for wine research before.22 BSA is about double the size of typical wine proteins. BSA is about 66 kDa while wine proteins are normally around 30 kDa. The hydroxycinnamate caffeic acid (>98%, Sigma-Aldrich) was selected as the model phenolic compound as hydroxycinnamates are the largest group of polyphenols in white wines. It is found in many foods and beverages like wine along with other vegetables like celery, eggplant, and lettuce,79,80 and caffeic acid is typically found in a concentration of 2.4-18.8 mg/L in wines.81

Table 3. List of major wine components and levels used in the model wines. Component Levels Ethanol (%v/v) 10 12 14 Glycerol (g/L) 5 15 Phenolics (mg/L Caffeic acid) 20 60 Protein (g/L Bovine serum albumin) 0.1 0.5 Aroma fraction See Table 2 Titratable acidity (g/L Potassium 7.5 hydrogen tartrate) pH 3.5

In addition to the components listed in Table 3, potassium hydrogen tartrate

(99%, Spectrum Chemical MFG. Corp., Gardena, CA) was used to achieve a titratable acidity level of 7.5 g/L and a pH of 3.5 for all model wines.

On the day of processing, reverse osmosis water was combined with 200-proof ethanol and potassium hydrogen tartrate in 5-gallon buckets. These were aliquoted into smaller buckets where the remaining components, except for the aroma fraction, were added. Once mixed, model wines were measured into 500 mL glass canning jars (Ball, Broomfield, CO), and stored in a refrigerator (4C) until the day of sensory testing for a maximum of 6 weeks. One milliliter of the aroma fraction was added 5 hours prior to sensory evaluation (Training and evaluation sessions), to the 500 mL model wine samples.

2.2.3 Sensory Evaluation with a trained Descriptive Analysis Panel

A panel of 9 participants (8 females and 1 male, ages 23-65) was recruited for this study. This study was reviewed and approved by Penn State’s Internal Review

Board as STUDY00010095. Panelists were screened for availability and white wine consumption (at least 2-3 times per month) from a pool of participants that had participated in at least one DA panel prior to the study. Panelists were also screened for allergies, health concerns, and taste or smell defects.

Panelists received six, 90-minute training sessions over a period of 2 weeks.

In these sessions, panelists were provided with basic taste and mouthfeel references and aroma references (Table 4). For several aroma and flavor attributes, more than one reference standard was selected by the panel to illustrate the concept of the attribute, as multiple references may actually be more suitable to define sensory

45 attributes than single references for DA.82 Panelists developed a working lexicon to describe the aroma, flavor, taste, and mouthfeel attributes perceived in the model wines. The panelists were then trained on the terms that they had generate, and practiced using the line scale on paper ballots and on a computer. Panelists were deemed trained after successfully evaluating blind duplicates and rating them in a similar manner.

All 24 model wines were evaluated in sensory triplicate in 12, 30-minute evaluation sessions over four weeks. Evaluation was conducted in individual tasting booths at room temperature under white light, where each panelist evaluated six model wine samples per session, presented in a Williams Latin square design to control for carry-over effects. Panelists had a forced 60 second break between each sample, during which they were encouraged to cleanse their palate with DI water and an unsalted soda cracker (Mondeléz Global LLC, East Hanover, NJ). All sensory data were collected using Compusense Cloud software (Academic Consortium, Guelph,

ONT).

For evaluation, panelists received 20 mL of each model wine sample, poured into clear 7.5 oz. ISO tasting glasses. Glasses were labeled with a randomized three- digit blinding code and covered with a clean, plastic Petri dish. Samples were poured approximately 1 hour before evaluation to allow the headspace to equilibrate.

Table 4. References used for training of sensory attributes. All standards were prepared in 10% (v/v) ethanol solution, unless otherwise noted. Attribute Reference Reference Standard Name Apple Apple 1 green apple jolly rancher (city state) and 10.0 g fresh granny smith apple (Wegmans, State College, PA) in 30 mL solution Grape Grape 3 halved red grapes (Wegmans, State College, PA) in 10 mL white grape juice (Welch’s, Concord, MA) and 40 mL solution Floral Floral 1 1 drop of lily of the valley extract (Merlin Crystal Cave, Claremont, CA) in 25 mL solution. Next, 5 drops of the previous solution were then added to 20 mL solution Floral 2 1 drop of orange blossom extract (Indiana Botanic Gardens Inc., Hobart, IN) in 25 mL solution. Next, 10 drops of the previous solution were then added to 20 mL solution Floral 3 3 mL Floral 1 + 4 mL Floral 2 in 20 mL solution Pear Pear 1 30.0 g fresh green pear (Wegmans, State College, PA) in 20 mL solution Pear 2 5.0 g of apple and 20.0 g of pear in 20 mL solution Citrus Citrus 1 3 x 2 cm fresh orange peel (Wegmans, State College, PA) in 20 mL solution Citrus 2 3 x 2 cm fresh orange peel (Wegmans, State College, PA) + 3 x 2 cm fresh lemon peel (Wegmans, State College, PA) + 2 x 1 fresh lime peel (Wegmans, State College, PA) in 20 mL solution Sour Sour 1.5 g/L tartaric acid (>99.7%, Sigma-Aldrich) in water Sweet Sweet 30 g/L sucrose (Domino Foods Inc., Yonkers, NY) in water Bitter Bitter 0.8 g/L caffeine (Sigma-Aldrich) in water Astringent Astringent 1.5 g/L alum (McCormick, Hunt Valley, MD) in water Warm/Hot Hot 6% (v/v) ethanol (Decon Labs, Inc) solution Viscous Viscous 1.5 g/L carboxymethyl cellulose (Tic Gums, Belcamp, MD) in water

2.2.4 Instrumental Evaluation of Model wines

Instrumental analysis with HS-SPME-GC-MS was performed on the model wine samples in order to validate prior findings21 of nonvolatile components impacting the partitioning of volatile aroma compounds into the headspace. Model wines were analyzed using the same setup that was used for the aroma fraction determination with the following modifications: The split ratio was 10:1 and the split flow was

10 mL/min. The MS was operated in parallel scanning and selected ion monitoring

(SIM) modes, to quantify acetic acid (m/z of 43, 45, and 60; total dwell time of 210 ms) and beta-damascenone (m/z of 69, 121, 190; total dwell time of 210 ms). Calibration curves for each of the aroma compounds in the aroma fraction were created using the aroma fraction at 40%, 60%, 80%, 100%, 120%, 140%, and 160% of the aroma fraction concentration that was added to the model wines. Calibration standards were created in DI water similarly to Robinson et al. to assess the effects of ethanol on HS partitioning.21 All analyses were done by adding 1.96 mL of model wine and 0.04 mL of the aroma fraction to a 20 mL HS vial (Restek, Bellefonte, PA) together with 2 g

NaCl and the IS mix (50 L of 9.9 mg/L d8-napthalene and 13.7 mg/L 2-octanol). Vials were capped with a magnetic crimp cap (Restek), and analyzed in triplicate.

2.2.5 Statistical analysis of sensory and instrumental data

Statistical analysis was conducted using R Studio software (ver. 3.4.3).83

Significance was defined as p  0.05 for all analyses.

For the sensory data, a six-way multivariate analysis of variance (MANOVA) for the factor effects (panelist, replicate, ethanol, caffeic acid, glycerol, and protein) was first conducted to determine whether or not there were overall differences.

Subsequent univariate ANOVAs were then performed on each attribute, treating all factors (panelist, replicate, ethanol, caffeic acid, glycerol, and protein) as fixed effects.

Finally, principal component analysis (PCA) was conducted using the package

SensoMineR (ver. 1.23)84 in order to better understand which factors were driving the observed sensory differences in the model wines. Additional R packages used

48 included ggplot2 (ver. 2.2.1),85 plyr (ver. 1.8.4),86 dplyr (ver. 0.7.4),87 gridExtra (ver.

2.3),88 agricolae (ver. 1.2-8),89 and forcats (ver. 0.4.0).90

For the instrumental data, a four-way MANOVA was conducted for the factor effects (ethanol, caffeic acid, glycerol, and protein), followed by individual ANOVAs on the volatile aroma compound concentrations in the model wine samples to see which of the factors were responsible for differences in volatile partitioning. Partial least squares regression (PLSR) was used to link the instrumental and sensory data, similar to prior wine studies,8,9,64,65 with the R package pls (ver. 2.6).91

2.3 Results

2.3.1 Determination of the Traminette aroma fraction

Using HS-SPME-GC-MS, the aroma composition of the research Traminette wine was found to be in alignment with previously reported major aroma compounds that were used in a model wine experiment. Numerous alcohols, ethyl esters, acetates, volatile fatty acids, aldehydes, C6 compounds, terpenes, and volatile phenols were found. These compounds can come from a variety of areas but most are fermentation or grape derived.6 Based on these results and prior literature, the final aroma composition, to be used in all model wines, is listed in Table 2.

2.3.2 Descriptive Analysis revealed significant factor effects on sensory perception

A multivariate Analysis of Variance (MANOVA) was carried out initially on a model that included all four main factors, ethanol, glycerol, protein, and caffeic acid, and all possible interactions to determine whether the 24 different model wines differed significantly across all sensory attributes in any of the model wine parameters

(p0.05). Significant overall sensory differences were found due to two of the four factors: both ethanol and glycerol concentration affected sensory perception significantly, as tested by the Wilks’ test statistic (Table 5). No significant interaction effects were found in the MANOVA.

The significant ethanol effect is in alignment with previous studies that have shown that trained panelists are able to differentiate between wines with a minimum difference of 0.4% (v/v) ethanol.13 The model wines used in this experiment differed by 2% (v/v) alcohol, thus, the trained panel was expected to discriminate the wines based on ethanol content.

The second significant factor, glycerol, has also previously been shown to significantly affect sensory perception of model white wines.6 In the referenced study, increasing glycerol levels from 0 g/L to 10 g/L led to increased overall flavor and decreased bitter taste. Similarly, the glycerol levels in this experiment were set at either 5 g/L or 15 g/L, leading to expected significant effects on the sensory perception of the model wines.

Last, although varying the BSA and caffeic acid levels between 0.1-0.5 g/L, and

20-60 mg/L respectively, did not lead to significant overall differences, it must be noted that this does not necessarily mean that they do not affect sensory perception. The effects of BSA and caffeic acid were expected to be much smaller than the effects of glycerol and ethanol. Additionally, MANOVA tests for overall differences. This is done by comparing means across the factor levels. If factors show opposite effects at different levels (i.e., significant interaction effects), the overall differences in means would cancel each other out, leading to non-significant results. Secondly, if just one

50 or two sensory attributes would be affected by these factors, this would also not lead to a large enough signal for MANOVA to pick up.

Table 5. MANOVA table for significant factors on sensory attributes DF F-value P-value Ethanol 2 10.6 <2.2 x10-16 Glycerol 1 3.0 6.4 x10-5 Residuals 644

Following MANOVA, individual univariate analyses of variance (ANOVA) were carried out with an extended model that also included the panelist and sensory replicate, along with ethanol, caffeic acid, glycerol, and protein as main fixed effects factors. All possible interactions between the compositional factors (ethanol, caffeic acid, glycerol, and protein) were also included (Table 6).

The panelists factor had a significant effect on the rating of all of the sensory attributes. While this was a trained panel, this is still to be expected. Although the entire panel is in agreement, each assessor will perceive sensory sensations differently due to individual differences. Similarly, using the line scales differently will also lead to significant panelist effects.92 Lawless and Heymann suggest that there are limitations on the abilities of humans to be calibrated as instruments. People differ in sensitivities to odors and sensations and therefore produce different sensory results.92

For 4 out of the 17 attributes a significant sensory replicate effect was found

(Table 6); again, this is expected, most likely due to repeated exposure.92

Last, 3 of the 4 main compositional effects also showed significant effects on at least one sensory attribute (Table 6). As expected from the MANOVA, ethanol

51 showed the largest effects, significantly altering the perception of 8 attributes, namely pear and ethanol aroma, pear flavor, sweet, sour, and bitter taste, and astringent and warm/hot mouthfeel. Second, varying glycerol levels significantly affected 4 attributes, namely apple, pear, and grape flavor, as well as sweet taste. Last, varying protein levels significantly impacted astringent mouthfeel. Most of our significant interactions affected flavor and taste attributes. Significant two-way interactions were found for citrus aroma, citrus flavor, floral flavor, sweet taste, and astringent mouthfeel, while significant three-way interactions were found for grape aroma and flavor. Significant four-way interactions were found for apple flavor and sweet taste, likely due to cross- modal interactions and suppressing effects of the experimental factors. Factors that by themselves did not affect any sensory attributes were however, were involved in significant interactions, hinting at modulating and indirect sensory effects. For example, caffeic acid did not significantly affect any of the sensory attributes on its own. However, caffeic acid was involved in 7 of the 10 two, three, and four-way interactions, indicating the modulating effects of hydroxycinnamates in white wine perception.

Applying Tukey’s Honestly Significantly Difference (HSD) (Table 7), the following trends became apparent: for ethanol, increasing concentrations led to decreased intensity ratings for pear aroma, pear flavor, sweet taste, and sour taste.

In contrast, ethanol aroma, bitter taste, astringent mouthfeel, and warm/hot mouthfeel all increased with increasing ethanol levels. Looking at these results, one would expect ethanol aroma, bitter taste, and warm/hot mouthfeel to increase with increasing ethanol concentrations, as these attributes are direct measurements of ethanol

52 perception.14 The observed decrease in pear aroma and flavor at higher ethanol concentrations are in agreement with previous findings that suggest that ethanol causes a decrease in fruity flavors and aromas.60 The results for increased astringency ratings with higher ethanol concentrations are also in agreement with previous findings.6

The decrease in sweet and sour taste due to increases in ethanol, indicates an indirect mode of action, explained by the increasing bitterness that accompanies higher ethanol concentrations. Increased bitterness may suppress the perception of sweet and sour tastes through cross-modal suppression. Such cross-modal suppression effects have been demonstrated for ethanol by Fischer and Noble. They found that sour taste was suppressed as ethanol increased from 8% to 11% and

14% (v/v). Over the same range, bitterness also increased, suggesting that the bitterness, as a result of the ethanol, was suppressing the sourness and sweet taste.16

Varying the levels of glycerol led to significant effects on several sensory attributes: at higher glycerol levels, the perception of apple, pear, and grape flavor, as well as sweet taste also increased. Prior studies have shown that glycerol has a sweet taste,23 which could explain the increased sweetness and fruity perception. Cross- modal interaction effects between sweet taste and fruity aromas and flavors have been reported previously.63 It is possible that the increase in sweetness is driving the increase of fruity flavor perception and vice versa with increasing glycerol concentrations, because the difference between the glycerol levels of 10 g/L is above the difference threshold of 5.2 g/L for glycerol in wine.23

The last factor that significantly affected sensory perception of the model wines was BSA protein. As the BSA concentration increased from 0.1 to 0.5 g/L, the astringent mouthfeel also increased, which may be counter-intuitive. However, Jones et al. reported that the interaction between glycerol, protein, and polysaccharides had a significant effect on the perceived astringency of model white wines. When glycerol was not present, the model wines were rated as significantly more astringent, indicating that protein may increase astringency as long as glycerol levels are low.6

Although the protein levels used in this experiment did not vary by a lot, their effect on astringency was large enough to be perceived by our trained panel. A reason for this could be the reactivity of BSA. Wine proteins form complexes with phenolic compounds in wine, leading to lower astringency. The BSA in our model wine may not react with the caffeic acid in the same way that wine proteins react with wine phenolics, due to structural differences between BSA and typical wine proteins. For example,

BSA is about 66 kDa and wine proteins are about 30 kDa.10

Table 6. F-values for the experimental factors and their interactions on the sensory attributes of the model wine samples.

Aroma Flavor Taste Mouthfeel

Factor DF Apple Pear Grape Citrus Floral Ethanol Apple Pear Grape Citrus Floral Sweet Sour Bitter Astringent Warm/Hot Viscous Aroma Aroma Aroma Aroma Aroma Aroma Flavor Flavor Flavor Flavor Flavor Taste Taste Taste Mouthfeel Mouthfeel Mouthfeel Panelist Code 8 84.5 126.7 66.9 100.7 23.2 36.4 72.8 104.5 69.7 80.8 46.7 37.7 62.9 36.4 56.4 36.4 118.0

Rep 2 3.5 2.9 0.1 0.9 1.1 1.7 0.5 2.1 0.2 2.4 1.8 1.1 10.9 2.0 7.7 1.9 3.7

Ethanol (E) 2 0.8 5.3 0.5 1.3 0.7 39.4 2.8 6.3 0.8 2.7 0.6 9.3 8.9 65.2 41.1 166.7 0.2

Caffeic (C) 1 1.1 0.3 0.4 0.0 0.4 0.7 0.0 1.2 0.9 0.5 0.1 0.3 1.3 1.7 0.1 0.1 0.6

Glycerol (G) 1 0.1 0.1 0.1 0.0 0.0 0.4 6.0 11.4 10.9 0.0 0.2 47.7 0.2 0.4 0.4 0.2 0.1

Protein (P) 1 1.5 1.2 3.2 0.3 3.3 0.6 0.0 0.0 2.9 0.7 0.6 1.0 0.1 2.9 6.4 0.2 0.1

E*C 2 0.0 0.8 0.1 0.9 0.2 0.1 0.3 0.4 1.3 1.6 0.0 1.2 0.0 1.6 4.1 2.0 0.9

E*G 2 0.3 0.1 0.4 0.7 0.8 1.6 0.8 2.2 0.5 3.5 0.2 1.1 2.8 0.2 1.3 0.2 0.3

E*P 2 0.3 0.1 1.4 3.5 1.7 0.0 0.5 0.2 1.4 0.1 0.3 0.1 0.0 1.6 1.2 0.2 0.4

C*G 1 0.0 0.0 0.6 0.1 0.6 0.5 0.0 1.7 1.1 0.3 0.5 4.7 1.2 0.1 1.5 0.3 0.2

C*P 1 1.6 0.2 0.0 0.3 0.7 0.6 0.0 0.0 0.0 4.0 0.1 0.0 0.3 1.0 0.5 0.9 0.4

G*P 1 1.4 1.5 2.0 0.7 0.4 0.4 0.1 0.5 1.8 1.5 5.2 1.2 0.1 1.4 0.2 1.3 0.6

E*C*G 2 0.2 0.1 1.5 0.4 0.5 0.0 1.5 0.8 4.1 1.9 1.1 0.7 1.2 0.6 0.6 0.2 0.6

E*C*P 2 0.9 2.1 3.3 0.2 1.5 0.9 0.2 0.2 0.1 1.9 2.0 0.9 0.7 1.0 1.6 1.7 0.4

E*G*P 2 0.4 1.3 0.0 0.6 0.3 0.1 0.5 1.4 0.4 0.7 0.4 0.3 0.7 0.0 0.3 1.5 0.3

C*G*P 1 0.4 0.0 1.5 1.2 0.0 0.5 0.9 0.3 0.1 2.4 0.5 0.0 1.5 0.2 0.0 0.0 1.8

E*C*G*P 2 1.2 0.2 0.2 1.2 0.1 0.2 3.1 2.1 1.9 0.1 0.0 3.2 0.2 0.7 0.3 0.0 0.6

Residuals 614 208.7 201.7 149.0 83.1 195.9 273.5 195.5 195.5 145.4 89.4 107.3 124.6 119.9 128.6 160.0 217.0 31.9 (MSE)

Bolded values are significant at p <0.05 55

Table 6 also summarizes the significant two, three, and four-way interactions.

Interactions suggest that the experimental factors differ in their impact depending on what other factors are present or absent. In Figure 5, the six significant two-way interactions are plotted for easier interpretation. Significant interactions between factors occur when the combination of the factors leads to different outcomes depending on factor levels. Such a behavior is shown in the interaction plot as non- parallel lines. Overall, compositional interactions impacted aroma and flavor perception, namely, citrus aroma and flavor and floral flavor, while two interactions affected sweet taste and astringent perception.

Table 7. The effect of the wine matrix components on the sensory attributes of the model wine samples.

Ethanol Glycerol Protein

10% 12% 14% Low High Low High

Apple Aroma ------Pear Aroma 34.1 ab 36.8 a 32.4 b - - - -

Grape Aroma ------Citrus Aroma ------Floral Aroma ------Ethanol Aroma 31 c 36.2 b 45.0 a - - - - Apple Flavor - - - 32.8 b 35.5 a - -

Pear Flavor 30.9 a 30.7 a 26.7 b 27.5 b 31.3 a - - Grape Flavor - - - 22.3 b 25.4 a - - Citrus Flavor ------Floral Flavor ------Sweet Taste 25.2 a 22.9 ab 20.6 b 19.9 b 26.0 a - -

Sour Taste 32.3 a 29.4 b 28.0 b - - - - Bitter Taste 12.6 c 19.0 b 25.1 a - - - - Astringent Mouthfeel 21.1 c 26.5 b 32.2 a - - 25.4 b 27.9 a Warm/Hot Mouthfeel 26.3 c 40.3 b 52.2 a - - - - Viscous Mouthfeel ------Means with different letters for each factor indicate significant differences at p  0.05 using Tukey’s HSD 56

Figure 5A shows the combined effect of ethanol and protein on citrus aroma perception: at low levels of ethanol (10% v/v), the effect of high vs. low protein levels is not different, however, as ethanol content increases, the two protein concentrations affect citrus aroma perception to a different degree. At 12% (v/v) ethanol, citrus aroma intensity was higher at the high protein level, while the opposite was found at 14%(v/v), where citrus aroma was more intense at the low protein concentration. Using Tukey’s

HSD, the difference in means of the high and low protein levels at each of the three ethanol levels are not significant, however, the interaction effect indicates countering effects of the two factors for the citrus aroma. Prior studies have shown that ethanol tends to decrease fruity aromas,18,60 while protein may increase them,6 however, these studies did not test for interactions between protein and ethanol on fruity aromas. As these factors have been shown to have opposite effects, it is not surprising that there is an interaction effect present when both factors are involved in the wine matrix.

A similar interaction on citrus flavor was found between ethanol and glycerol concentration (Figure 5B): At low levels of ethanol (10% v/v), higher citrus flavor was perceived at the low glycerol level, but as ethanol levels increase to 12% and 14%, the perception of citrus flavor was highest at the high glycerol level. Similar to the previous interaction, there are no significant differences in the means of the samples at the three ethanol levels, as assessed by HSDs. The overall trend appears to be that citrus flavor decreased as ethanol increased, but this effect is modulated by protein concentration. Jones et al. showed that glycerol increased overall flavor while ethanol did not have an effect, and ethanol was shown to decrease citrus aroma.6 This

57 is similar to the previous interaction as glycerol and ethanol have previously demonstrated opposite effects.

Citrus flavor intensity was also affected significantly by an interaction of protein and caffeic acid (Figure 5C). At low levels of caffeic acid, citrus flavor perception was enhanced by a low level of protein. At high levels of caffeic acid however, we observed higher citrus flavor perception with a high level of protein. Similar to the first two interactions, there are no significant differences between the means at the two caffeic acid levels. These results are interesting as both protein and phenolics have been shown to increase fruity flavors.6,28 We would expect that the citrus flavor would increase with both protein and caffeic acid content.

Figure 5D shows the interaction between caffeic acid and glycerol on sweet taste. Using Tukey’s HSD test, there is a significant difference between the sample means at both levels of caffeic acid. Sweet taste is always greater at the high level of glycerol, however, the sheer presence of glycerol is not sufficient to counter the sweet- decreasing effects of caffeic acid, as seen at the low glycerol level. The difference in sweet taste between the two glycerol levels is greater in a high caffeic acid sample than it is in a low caffeic acid sample. This suggests that caffeic acid actually decreases sweetness but it is unable to overcome the increase in sweetness driven by glycerol when glycerol is present at high levels. Caffeic acid by itself has been shown to be perceived as bitter, thus, this interaction of glycerol and caffeic acid could be explained by sweet-bitter taste suppression.6

Figure 5E shows the interaction of glycerol and protein on floral flavor perception. At the low protein concentration (0.1 g/L), floral flavor was perceived at a

58 higher level in the high glycerol samples (15 g/L), but at the high protein concentration

(0.5 g/L) floral flavor was more pronounced in the low glycerol content samples (5 g/L).

There are not significant differences between the means of the samples at the two protein levels. The average floral flavor appears to be greater in high protein model wine samples (Table 6). Increased protein content has been previously shown to increase floral aroma, which, in turn, likely increases floral flavor. The same study also showed that glycerol increases overall flavor at higher levels.6 One would expect the floral flavor intensity of the high protein, high glycerol sample to be higher than it is.

Since it is not, this suggests that the interaction between protein and glycerol at high levels may actually decrease floral flavor, most likely due to a decrease in floral flavor with increasing protein.

Finally, Figure 5F shows the interaction between ethanol and caffeic acid on astringent mouthfeel. The first general trend is that as ethanol concentration increases, astringent mouthfeel also increases. At our lowest ethanol concentration, there really isn’t a difference between the caffeic acid levels but at the 12% ethanol level, the low caffeic acid samples (20 mg/L) are perceived as more astringent and at the 14% ethanol level, the high caffeic acid samples (60 mg/L) are perceived as more astringent. There are no significant differences between the means of the samples at the three ethanol levels. These results are consistent in that higher ethanol levels lead to more astringency while higher caffeic acid levels have no effect. One would have expected caffeic acid to also increase astringency, as shown by Gawel et al.15 It could be that the caffeic acid level used was not large enough to induce the same effect as in the study by Gawel. It could also be that caffeic acid is smaller than other phenolic

59 compounds typically found in wine. This could affect its ability to bind to salivary proteins and therefore, not necessarily increase astringency.

Figure 5. Interaction plots for all significant two-way interactions. Significant interactions do not imply significant differences in means between samples. Interactions are plotted to visualize the effects they have on sensory attributes.

In lieu of discussing the higher order interaction effects in rather complex univariate plots, a principal component analysis (PCA) is shown in Figure 6. About

35% of the total variance is explained by the first principal component, while an additional 16% is explained by PC 2, and another 12% explained by PC3, totaling to

63% of the total variance in the first 3 dimensions. Looking at the eigenvalues, a drop was observed after 3 dimensions, indicating that focusing on the first three dimensions of the PCA provides a sufficient capture of the underlying patterns.

In Figure 6A, we can see that ethanol concentration is the major driver of sample separation along PC 1, with the highest level of 14% (v/v) being positioned on the negative axis of PC 1, 12% (v/v) being located around the origin of PC 1, and the lowest concentration of 10% (v/v) being positioned on the positive axis of PC 1. The

“dimdesc” function in the SensoMineR package confirms that ethanol is driving the separation along PC 1. This means that 35% of the total variability observed in this experiment can be attributed to the ethanol level. To a smaller degree, the two different glycerol concentrations also separate along PC 1, indicating a similar effect on the sensory perception of the samples, but in an opposite direction to ethanol. While the low glycerol concentration of 5 g/L is loaded on the negative PC 1, the high glycerol level of 15 g/L is positively correlated to the positive PC 1. This means that ethanol and glycerol counteract each other to a certain degree, in that high glycerol-containing samples would counteract the effects of high ethanol levels to some degree. For example, this was shown for citrus flavor (see also Figure 5B).

From the “dimdesc” function, we see that none of our factors are significantly correlated with PC 2. Looking at Figure 6A, we do observe some small separation between the protein levels. In Figure 6A, PC 2 explains 16% of the total variance, and seems to be driven by changes in protein and, to a smaller degree, caffeic acid levels.

This means that about 16% of the total variance may be explained by the protein content. We see the high level of protein (0.5 g/L) aligned along the negative axis of

PC 2 and the low level of protein (0.1 g/L) aligned along the positive axis of PC 2.

Figure 6B shows how the sensory attributes are affected by the experimental factors in the first two PCA dimensions. While attributes affected by high ethanol levels are pointing towards the negative PC 1 axis on the left of the plot, attributes that are highest in low ethanol samples are pointing towards the positive PC 1 axis on the right.

Warm/hot mouthfeel, astringent mouthfeel, bitter taste, and ethanol aroma all show a very high negative correlation to PC 1. This means that they are higher in the highest ethanol level of 14% (v/v). Pear aroma, apple flavor, pear flavor, sweet taste, grape flavor, and sour taste are all highest in low ethanol wines, as they show a high positive correlation to PC 1. These results are in agreement with the results discussed in the

ANOVA section. Interestingly, all other sensory attributes provide a less clear correlation to ethanol content, as they show a close to 90-degree angle to the first dimension, indicating statistical independence. Along the second principal component, explaining 16% of the variance, viscous mouthfeel is loaded negatively on that axis, indicating that samples high in protein, caffeic acid, and glycerol are perceived as more viscous that their low-level counterparts, just not to a significant level. The sensory attributes apple aroma, grape aroma, pear aroma, citrus aroma, citrus flavor, and floral flavor all show a positive correlation to PC 2, and are rated higher in lower protein and lower caffeic acid model wine samples with medium ethanol levels. These complex interactions were not found in the ANOVAs but are important conclusions that can be derived from this PCA.

In Figure 6C, it becomes apparent that the third principal component, PC 3, is likely driven by glycerol content as the two glycerol levels are separated along the

PC 3 axis. This is confirmed by the “dimdesc” function. Samples with higher glycerol content (15 g/L) are positively correlated to PC 3 while samples with lower glycerol content (5 g/L) are located at the bottom of the plot, showing a negative correlation to

PC 3. About 12% of the variance is explained by PC 3, meaning that glycerol can account for about 12% of the total variation.

In Figure 6D, it is easy to understand which attributes are most affected by glycerol content, namely, floral flavor, pear flavor, grape flavor, and sweet taste, which all show a positive correlation to PC 3 and high glycerol samples, and sour taste, which shows a negative correlation to glycerol content, thus, is rated highest in low glycerol samples. Additionally, Figure 6D also shows a high correlation between various sensory attributes, such as sour taste and citrus flavor, apple and pear aroma and flavor, grape flavor, sweet taste, and viscous mouthfeel, which show a close to

90-degree angle (i.e., statistical independence) to ethanol-related attributes of ethanol aroma, bitter taste, astringency, and warm/hot mouthfeel.

Figure 6. (A) Score and (B) loadings plots for PC 1 and PC 2, and (C) Score and (D) Loadings plot for PC1 and PC 3, using the DA results for the model wine samples. In the score plot, individual wine samples are indicated as black dots, while the different experimental factors are super-imposed as supplementary factors (ethanol levels 10, 12, and 14 %(v/v) in red, protein levels low (0.1 g/L) and high (0.5 g/L) in blue, glycerol levels low (5 g/L) and high (15 g/L) in black, and phenolic levels low (20 mg/L) and high (60 mg/L) in green)

High correlations between corresponding aroma (i.e., orthonasally assessed) and flavor (i.e., retronasally assessed) attributes, such as pear aroma and flavor, have also been reported by Villamor et al., studying the effects of ethanol, tannins, and fructose concentrations in model red wines.14 Additionally, attributes like ethanol aroma and warm/hot mouthfeel, pear aroma and flavor, and citrus aroma and flavor

64 are all grouped close together with their matching pair. Using the “dimdesc” function in the SensoMineR package we can see the correlation values for the factors and the components. Ethanol aroma and warm/hot mouthfeel have a correlation of -0.82 and

-0.90 with PC 1, respectively. Pear aroma and flavor have significant correlations of

0.43 and 0.79 with PC 1, respectively. Citrus aroma and flavor have significant correlations of 0.41 and 0.72 with PC 2, respectively. This indicates that their intensities were correlated and rated very similarly.

2.3.3 Correlation between sensory and instrumental data

As a last step, the results from the sensory evaluation by the DA panel were related statistically to the instrumental headspace measurements. Based on prior studies, we hypothesized that as headspace concentrations of the different aroma compounds decrease, a similar effect would be seen in the intensity ratings of the aroma and flavor attributes. For example, increasing ethanol concentration was shown to decrease the headspace concentration of -damascenone, a compound that on its own has a rose or honey like odor.21 Similarly, in general, increasing the ethanol concentration has been shown to decrease fruity aromas and flavors, thus, we would expect that a decrease in headspace concentration of fruity-smelling aroma compounds would lead to a decrease in the fruity aroma perception. This assumption is limited in several ways. For example, the decrease in concentration may not be above the detectable threshold, meaning that a panelist would be unable to detect any differences.

In a first step, an initial MANOVA was carried out on the volatile compound concentrations using a four-way model with all possible interactions (Table 8). The

65 factors were ethanol, caffeic acid, glycerol, and protein. These results show that all experimental factors significantly affected the partitioning of the volatile compounds into the headspace of our model wine samples. In contrast to the DA results, more factors and interactions were found to significantly affect the headspace concentration, however, some similarities between the DA and the instrumental data were found. Ethanol and glycerol were responsible for significant differences in both the DA and instrumental data.

Table 8. MANOVA table for factor and interaction effects on the partitioning of volatile compounds into the headspace of the model wine samples. Bold P-values are significant at   0.05. DF F-value P-value Ethanol (E) 2 14.6 <2.2 x10-16 Caffeic Acid (C) 1 2.3 0.02 Glycerol (G) 1 2.5 0.01 Protein (P) 1 9.5 3.6 x10-8 E*C 2 3.6 8.6 x10-6 E*G 2 3.2 4.6 x10-5 E*P 2 3.9 1.7 x10-6 C*G 1 10.3 1.3 x10-8 C*P 1 3.8 6.7 x10-4 G*P 1 10.6 8.7 x10-9 E*C*G 2 4.8 5.6 x10-8 E*C*P 2 2.9 1.5 x10-4 E*G*P 2 2.8 2.2 x10-4 C*G*P 1 1.0 0.48 E*C*G*P 2 5.0 2.7 x10-8 Residuals 48

In a second step, the sensory and instrumental data were linked to each other using partial least squares regression (PLSR) to study how changes in headspace concentration might explain changes in the sensory perception of aroma and flavor

66 attributes. Although the primary goal of the PLSR was to study the link between volatile concentration and aroma perception, all sensory attributes, including all tastes and mouthfeels, were included as numerous aroma and flavor attributes showed strong correlations to taste and mouthfeel attributes (e.g., ethanol aroma and bitter taste).

This indicates that cross-modal interaction effects are taking place.

The PLSR model was created by using the concentrations of the aroma compounds along with the matrix factors as predictor variables for the sensory attributes. Model validation was done with a leave-one-out algorithm to determine the best model. Looking at the validation plots in Figure 7, one can see that the root mean square error of prediction (RMSEP) for most sensory attributes reached a minimum value for two model dimensions. Based on this result, a two-dimensional model seems to be the best prediction for our data set.

Figure 7. Validation plots for each of the sensory attributes in the PLSR model.

The sample score plot shown in Figure 8A displays all model wine samples in a PCA-like way: samples that are located close to each other share a similar volatile composition and sensory profile. Overall, the PLSR model explains a total of 70% of

67 the variance within our two data sets, with 53% explained by the first component and

17% explained by the second component. Similar to the sensory PCA, samples group based on ethanol content. The 10% ethanol samples are located to the left and bottom of the plot, the 12% ethanol samples located in the center-bottom right of the plot, and the 14% ethanol samples to the upper-right of the plot. Further, within an ethanol level, samples form sub-groups, indicating that the other experimental factors affect these samples as well.

The interpretation of Figure 8A is enhanced when looking at the correlation plot shown in Figure 8B, where both the predicting variables (i.e., compositional data) and the predicted, sensory attributes are shown in the same plot. As the name implies, variables that are close to each other show a high positive correlation, variables that span a 180-degree angle show a high negative correlation, and angles close to 90 degrees indicate statistical independence. None of the volatile compounds present in the aroma fraction of our model wines are located close to any of the sensory attributes. The exception to this trend is acetic acid, which is separated from all other volatiles, and close to the sensory attributes ethanol aroma, bitter taste, warm/hot mouthfeel, and astringent mouthfeel. At first, surprisingly, acetic acid shows a negative correlation to sour taste, however, one needs to keep in mind that the acetic acid concentration used in the aroma fraction is below the levels that are able to activate sour taste perception. Acetic acid, at 40 mg/L in our aroma fraction, has been described to be one of the major components of the aroma of white wines. Acetic acid is the major aroma compound in Gewürztraminer, one of the parent species of

Traminette. A study from Guth removed acetic acid from a Gewürztraminer model wine and found that similarity to a normal Gewürztraminer decreased.93

Further, the separation of acetic acid from all other volatile compounds would indicate that acetic acid is affected by the experimental factors differently, however, from this data, it is unclear what that effect might be.

The fact that all of the flavor and aroma sensory attributes, except for ethanol aroma and bitter taste, are located close to the center of the plot, indicates no direct and/or a non-linear relationship between the volatile concentrations and sensory perception. It also means that none of the aroma volatiles single-handedly drives any of the aroma and flavor sensory attributes. This is expected as, particularly for complex matrices such as wine, aroma and flavor perception is the result of complex mixture interactions, with suppressing, and enhancing effects. These results were demonstrated in a study done on bitters in a whiskey matrix. Volatile compounds showed both suppressing and enhancing effects on the aroma perception in the whiskey samples.94 Such non-linear behavior is therefore hard to model with PLSR, which assumes a linear relationship between predicting and predictor variables.

We can see in the upper-right corner of Figure 8B that ethanol is a good predictor of ethanol aroma, bitter taste, warm/hot mouthfeel, and astringent mouthfeel.

When we look at Figure 8A and B together, we also see that these sensory attributes are related to the higher ethanol samples. However, using headspace volatile concentration does not sufficiently predict sensory attributes.

Figure 8. (A) Score and (B) correlation plots for the PLSR model of our study. 10, 12, and 14 refer to the ethanol content. “C/c” refers to the caffeic acid concentration, “G/g” refers to the glycerol concentration, and “B/b” refers to the protein concentration. Capital letters indicate the high level and lowercase letters indicate the low level. The score plot shows the samples and the correlation plot shows how well the chemical composition predicts the sensory attributes.

2.4 Conclusion

In this study, four major wine components – ethanol, glycerol, protein, and phenolics, were systematically varied to create 24 different model wine samples to study their effect on the sensory attributes and volatile headspace concentrations. The used aroma fraction as well as the ranges of the experimental factors were selected to mimic a white, hybrid wine, thus, findings from this study can be related to hybrid wines, which are currently understudied when it comes to their sensory properties.

Findings from the experiments indicate that all experimental factors did indeed affect both sensory attributes (as assessed by DA) and volatile concentrations in the model wine samples. Varying ethanol between 10-14% (v/v) had the largest effect, leading to significant differences in 8 of the total 17 sensory attributes, and in the headspace concentrations of all volatile aroma compounds. Glycerol, and protein also significantly

70 impacted at least one sensory attribute, while the effect of the model phenolic compound, caffeic acid, was less pronounced.

While direct effects of all experimental factors were expected based on literature reports, numerous indirect effects as well as interactions between the four factors were also observed. For example, while varying ethanol content directly affected astringent mouthfeel, the intensity of this sensory attributes was significantly modulated by the presence of caffeic acid. Ethanol increased astringent mouthfeel but we also see that the interaction between ethanol and caffeic acid had a significant effect. Such examples of counteracting experimental factors provide a better understanding of real wine sensory perception and could elucidate mechanisms of how to modulate wine sensory properties through viticultural and enological practices.

Besides the expected results, such as increasing ethanol content leading to increasing ethanol aroma and warm/hot mouthfeel, some results were unexpected:

For example, what are the mechanisms that lead to significant differences in citrus flavor perception? As the aroma fraction was not varied, any observed differences are due to indirect effects. It could be that panelists associate sour taste with citrus flavor.

As sourness of the model wine samples differed, this could have been associated with a change in citrus taste, indicated by the correlation between sour taste and citrus flavor.

Similar to the sensory results, volatile headspace concentrations were significantly modulated by the all experimental factors, with additional significant interaction effects. Overall, instrumental changes were larger than sensory changes.

Our PLSR model demonstrated that it could model our data well, but that the volatile headspace concentrations did not predict the sensory data well.

Overall, the experiment provides insight on how major wine components interact with each other and with other, minor wine components, such as aroma compounds, and affect sensory perception of white, hybrid wines. While this study was conducted using a model system, it raises the question of whether and how these results translate into a real wine system, and should be validated for broader conclusions. The overall results of this study indicate that small changes in composition, such as changing ethanol, glycerol, protein, or phenolic content, within the levels observed in real wines, are able to significantly change multiple sensory attributes, including aroma, flavor, taste, and mouthfeels. Further, these changes seem to be the result of both direct and indirect actions, where the effect of one component may be counteracted by another, often minor, factor, often in an unexpected way.

3 Temporal Evaluation of Model Wine Samples

3.1 Introduction

As mentioned in the introduction of chapter 2, the aroma and flavor of wine is very complex. In addition, major matrix components like ethanol as well as non-volatile phenolic compounds and glycerol all have been shown to affect the overall aroma of wine. For example, ethanol has been shown to decrease citrus aroma and increase bitter taste in white wines.6 Phenolic compounds have been shown to increase astringency, viscosity, hot mouthfeel, and bitterness,15 and glycerol has been shown to increase the concentration of volatile compounds in the headspace of the wine in situations where ethanol is absent. This change in partitioning between the matrix and the headspace has been shown to affect aroma perception.21 The effects of these factors were summarized in Table 1. Further, these factors do not act isolated, but interact with each other as well as with aroma compounds, and these interactions also play a role in the perception of wine aroma.7 For example, C13-norisoprenoids and dimethyl sulfide have been shown to enhance fruity aromas while ethanol suppresses fruity aromas in red wine.60

The majority of literature that looked at the effect of wine matrix components on the sensory perception used descriptive analysis, where a trained sensory panel evaluates both qualitatively and quantitatively different sensory attributes as a single point measurement, that is, panelists make a single intensity rating for each attribute of a wine sample.

While this approach provides valuable information about sensory perception, wine matrix factors have also been shown to affect the dynamic perception of wine sensory attributes. Dynamic perception refers to the change in sensory attributes from moment to moment. Chewing, breathing, and saliva composition all have an impact on sensory perception. Wine flavor perception varies over time after the wine is exposed to the air.42 These changes over time are just as important to understand as any intensity differences in sensory attributes as temporal changes are part of the overall sensory experience.

Ethanol, for example, has been shown to affect both bitter taste intensity

(i.e., how high the intensity of the sensation is) as well as bitter taste duration (i.e., how long the sensation lasts). For white wines, higher ethanol levels led to both increased bitter intensity and duration.37

For red Syrah wines, higher levels of ethanol were found to increase the total duration of the wine finish, besides also increasing warmth, astringency, bitter taste, spice flavor, and dark fruit flavor. Wine finish was defined as the lingering tastes, flavors, and mouthfeels that participants experienced after swallowing of expectorating the wine.95 In contrast, lower alcohol wines were cited more frequently as sour, red fruit, and green. The low alcohol samples were described as higher in red fruit at the beginning, green in the middle, and sour at the end of the evaluation of a single sip.40

Last, a dynamic sensory method was used to evaluate the effects of carbonation levels in sparkling wines. Carbonation perception is an important quality attribute of sparkling wine, and it was found that lower carbonation (i.e., less CO2

74 present in the wine) typically resulted in lower perception for all sensory attributes, except for bitter and sour taste, which were unaffected by the increasing carbonation level throughout the entire evaluation period. For the highest level of carbonation, the duration of numbing and tingling sensations increased. Interestingly, the high carbonation treatment also led to an increase in sour taste perception between 45 and

60 seconds after evaluation began.39 This is interesting as sourness was initially higher in the lower carbonation treatments but flipped at the latter point in time, an observation that was only made due to the use of an temporal sensory evaluation method. Although no comparison to static DA was done, it could be assumed that this change in sour taste perception between the lowest and highest carbonation levels at a later point in time would have not been picked up by a static evaluation method, such as DA.

While it is known that wine matrix components impact the temporal perception of wine sensory attributes, the literature is very sparse, especially in the area of white, hybrid wines. The lack of work studying the temporal evaluation of white, hybrid wines is surprising, although it could be because hybrid wines are newer and have a more limited market compared to Vitis vinifera wines. That said, significant differences in wine perception have been described for the same wine when assessed statically with descriptive analysis vs. dynamically, for example by temporal dominance of sensation

(TDS) or temporal check-all-that-apply (TCATA).27,38 For example, previous studies on the effect of tannins, acidity, and ethanol in red wines not only found significant static differences by DA, but also that aroma, flavor, taste, and mouthfeel attributes changed to a different degree between the wines over time.27 In one study, tannin

75 concentration was varied in red Merlot wines either to a high (1157-1261 mg/L catechin equivalents) or low level (691-752 mg/L catechin equivalents). Acidity (pH of

5, 6, and 7) and ethanol levels (14%, 14.5%, and 15% v/v) were controlled by the additions of DI water and tartaric acid after fermentation. Notable results from this study were found for sour taste, bitter taste, and astringency. DA showed that sour taste increased with increasing levels of acidity, however, TDS revealed that tannin concentration was also responsible for differences in sour taste at the beginning of evaluation (within the first 15 seconds), where sourness was found to be higher in the higher tannin samples.

Also, as expected, bitter taste was found to be higher in high tannin samples but was found to be suppressed by higher acidity levels when assessed by the DA panel. Using the dynamic TDS method, bitter taste was found to not be the dominant attribute at the end of evaluation (~the final 15 seconds) for the high tannin samples, while bitter taste was the dominant attribute at the end of evaluation of the low tannin samples. This is interesting as it indicates that matrix components affect each other, with suppressing effects of bitterness and sourness. Last, for astringency, increased tannin concentration led to increased astringency in the DA, and TDS showed that this increase in astringency perception could be explained by astringency being perceived earlier in the higher tannin wines. This indicates that for astringency, the onset and duration of the sensation seems to drive the increased DA ratings.27

In the previous chapter, descriptive analysis (DA) was utilized to evaluate model wine samples, created to resemble a white, hybrid wine. However, while DA is able to provide a sensory fingerprint of different wines, this method has its limitations,

76 especially with products and sensory attributes that change during the duration of assessment. This was highlighted in a study involving commercial white wines that were reported to have different levels of bitter taste: The use of DA, where a single- point measurement of bitter taste intensity was recorded, left researchers unable to fully explain the differences in bitterness in the wines.37 In contrast, using a temporal method, namely TDS, the same group was able to show that the bitter taste persistence, an aspect that was not picked up by the DA panel, was able to explain the differences between the wines. Additionally, the researchers were able to link the persistence of bitter taste over the duration of a sip to not only ethanol, which was captured in DA, but also changes in fructose concentration. TDS revealed that fructose concentrations inversely affected the duration of bitter dominance more than ethanol, in that increasing levels of fructose led to a shorter duration of bitter dominance.37

These studies demonstrate the importance of combining static DA with a temporal method of sensory evaluation, particularly when trying to understand differences in complex, and changing products such as wine. In the recent past, two major temporal methods have been applied to dynamic assessment of wines, namely temporal dominance of sensation and temporal check-all-that-apply. In TDS, participants are given a list of sensory attributes relevant to the products at hand and are asked to select the most “dominant” attribute at that time. The most “dominant” attribute is the one that “catches the attention at a given time”, not necessarily the one that is the most intense.43 In an example provided by Pineau et al., the most intense attribute in a dairy sample may be “milk” or “diacetyl”. But, a less intense attribute, like

“fruity”, may appear. In this example, the panelist would select “fruity” when it appears as it now grabbing their attention, even though it isn’t the most intense attribute.43 TDS has previously been used to evaluate wine.37,45 For example, Sokolowsky and Fischer compared DA, time intensity (TI), and TDS to evaluate bitterness in white wine, and their findings suggest that a temporal method like TDS should be paired with a method like DA for precise measurements as DA will not capture temporal aspects and TDS will not capture intensity ratings.37 Meillon et al., found similar results to Sokolowsky and Fischer with TDS to study the effects of partially dealcoholized red wines on the sensory profiles. They suggest that while TDS cannot replace a complete sensory profile, it is easy to use and particular useful for products with a “weak-sensory- difference”.45 Finally, work by Frost et al. on red wines produced with different maceration practices, further confirmed that TDS is able to capture aspects of the sensory experience of wines that are not captured by DA or instrumental methods.38

While TDS as a temporal evaluation method has been shown to elucidate differences in dynamic vs. static wine aroma and flavor perception, the major limitation of this method is that it only provides information on the most dominant attributes. For a complex product like wine, focusing on the dominant attribute over time may not be enough as important aroma attributes, in addition to the dominant attribute, may need to be captured in order to understand their contribution to the overall flavor.41

Besides TDS, the second recently developed dynamic sensory method is temporal check-all-that-apply (TCATA). TCATA is an extension of the commonly used check-all-that-apply (CATA) method. In CATA, participants are provided with a list of attributes that are salient for a particular product, such as wine, and asked to select

78 any of the listed attribute that they perceive. The underlying thought is that the most important attributes will be picked at a higher frequency by a group of participants.

The result of CATA is frequency data, summarized in a contingency table, where one can see how often each attribute was selected. While intensity does not necessarily equate to frequency, Campo et al. showed that in Pinot Noir wines, attributes selected most often in CATA were also rated highest in intensity using classical DA.48

TCATA extends the CATA method as the amount of time given to participants increases in order to allow for temporal data collection. As an example of how the

TCATA method works, a participant starts the evaluation by tasting a wine sample and immediately checks all of the attributes that they perceive in the sample. As time goes on, the participant may notice the appearance of a new attribute and selects it or they may notice the disappearance of an attribute that was once present and deselects it. Data collection in this fashion results in present/absent data (i.e., 1 or 0) at each time point for a panelist that can be averaged over the panel to result in a histogram of attribute frequencies over the evaluation period. By applying statistical tests, differences between samples would be considered significant if they are above a certain threshold. These histograms can be further smoothed into frequency curves and compared across samples to highlight significant differences between samples over the temporal evaluation period. The process of converting frequency data into curves can be seen in Figure 9. TCATA has been applied to assess the effects of carbonation level in sparkling wines on the perception of sensory attributes such as bitter and sour taste, bite/burn, numbing, prickly, and tingly.39 It has also been used to study the impact of ethanol content on the sensory properties of Syrah wines.40

Figure 9. The conversion of raw frequency data to histograms and then to TCATA curves. Figure is taken from Castura et al.47

Together, these studies indicate that TCATA is an effective tool to study the temporal evolution of wine sensory perception. TCATA is similar to TDS, but may be more appropriate to identify smaller product differences, especially for complex products that may differ in more than one attribute. A study from Meyners and Castura showed that TCATA led to better product differentiation than TDS.49

In this chapter, the same model wine samples that have been characterized in a static fashion by DA (see Chapter 2), were evaluated by TCATA to assess the dynamic changes in sensory perception as a function of varying wine matrix components, such as ethanol, glycerol, and phenolics. The objective of this study was

80 to compare the static DA results to the results from a dynamic evaluation using

TCATA. Based on prior literature, we hypothesized that TCATA would identify different experimental factors than DA that are responsible for observed sensory differences.

Further, we hypothesized that ethanol content would lead to the largest changes in temporal sensory perception, both in duration and citation frequency.

3.2 Materials and Methods

3.2.1 Model Wines and Aroma Fraction

The same model wines used in the DA study were used for the TCATA; for details see Table 3. Model wines were prepared in the same way as described in section 2.2.2. Further, as the protein factor was found to not have a large effect on the static sensory perception of the model wines assessed by DA, and to reduce the overall number of samples, BSA protein levels were kept constant at 0.5 g/L (high protein level from DA) for all model wines, resulting in only 12 model wine samples evaluated by TCATA.

3.2.2 Sensory Evaluation by Temporal Check-All-That-Apply (TCATA)

A trained panel of 12 participants (8 females and 4 males, ages 23-65) was recruited for this study. Panelists were screened for availability and white wine consumption (min of 2-3 times/month). Panelists were also screened for allergies, health concerns, and taste or smell defects. People who received the initial screener had either previously participated on trained panels at the Sensory Evaluation Center at Penn State or were involved in grape and wine research. The study was reviewed by Penn State’s IRB (STUDY00010095).

Samples were prepared the same way as in the DA (see section 2.2.2), where on the day of testing, 1 mL of the aroma fraction was added to 500 mL model wine samples. Twenty milliliters of the model wines were poured approximately 1 hour before evaluation into clear ISO wine glasses and covered with a disposable, polystyrene Petri dish (60 x 15 mm, VWR International) to allow for headspace equilibration. Wine glasses were labeled with a randomized three-digit blinding code on the stem.

Overall, panelists received three, 90-minute training sessions. In these sessions, panelists were provided with basic taste and mouthfeel references and aroma references, as used in the DA (see Table 4). Based on the DA results, three flavors (Apple, pear, and citrus), three basic tastes (Sweet, sour, and bitter), and two mouthfeel (Astringent, warm/hot) attributes were included in the attribute list. An

“other” option was also provided to prevent potential “dumping” bias.44

Besides the training on the sensory attributes and references, panelists also received trained on the TCATA method through multiple methods. In order to first familiarize panelists with the TCATA method, they completed a guided practice example of TCATA, as provided by Compusense (Compusense Cloud, Academic

Consortium, Guelph, ONT). Panelists were instructed to imagine that they were tasting a cherry flavored beverage. After starting the timer, instructions popped up on the screen telling panelists what sensory attributes they were perceiving and telling them when to select or deselect them. Panelists were allowed to complete this exercise as many times as they wished.

In the next training session, a selection of music was played. The music used was “Somebody to Love” by Queen. This song was selected because it incorporated multiple instruments and vocals, many of which are present simultaneously. When the music began, panelists started their timer and selected the different components that they heard in the music. Panelists selected from musical elements like “lead vocals”,

“backup vocals”, “guitar”, “piano”, and “drums”. After familiarizing themselves with the method, panelists began practicing evaluating the model wine samples using TCATA.

Training success was assessed by (i) having panelists evaluate blind duplicate samples, and (ii) correctly identifying the references that they had been trained on.

Panelists were deemed sufficiently trained when they rated these blind duplicates in a similar manner and correctly identified all reference attributes. To assess that panelists were rating blind duplicates in a similar manner, the difference curves of the two duplicate samples were checked; as no significant differences between the duplicates were found, the panel was deemed trained as they were rating duplicates in a similar manner.

After training was complete, the 12 white, hybrid model wine samples were evaluated in three sensory replicates in 6, 60-minute evaluation sessions. Evaluation was conducted in individual booths at room temperature under white light. Each panelist evaluated six model wine samples in each individual evaluation session. A modified Williams-Latin square design was created so that panelists saw the same six wines in any given evaluation session.

In between individual samples, panelists had a forced 60 second break, during which they were encouraged to cleanse their palate with DI water and unsalted soda

83 cracker (Mondeléz Global LLC). Additionally, after the third sample (halfway through an evaluation session), panelists had a forced 10-minute break to avoid fatiguing.

During this break, panelists were encouraged to leave the test booth and walk around.

All sensory data were collected using Compusense Cloud software (Academic

Consortium, Guelph, ONT, Canada).

For the sample evaluation, panelists were first instructed to take a sip of water to cleanse their palate. Panelists were then instructed to take a sip of their model wine sample and start the onscreen timer. They were instructed to begin selecting and deselecting attributes immediately after starting the timer. After 10 seconds, they were instructed to spit out their sample. This 10 second time period was discussed during the training and panelists decided that they needed no more time than that. The evaluation of the sample continued for a total of three minutes. Panelists were not allowed to stop the timer early but they could deselect all attributes if they were no longer perceiving any of them.

3.2.3 Statistical Analysis

Statistical analysis was conducted using RStudio (ver. 1.1.383, Boston, MA) and R software (ver. 3.4.3), with the additional packages tempR (ver. 0.9.9.12)96 and ggplot2 (ver. 2.2.1).85 Significance was defined as p  0.05 for all analyses.

Initial analyses assessed panelist and panel agreement, in order to ensure that panelists agreed with each other, and with themselves for the replicate samples.

Panelist agreement with themselves was checked for all samples and attributes.

Panelist agreement with the rest of the panel was checked for selected attributes

(citrus flavor, sweet taste, and warm/hot mouthfeel) and samples, to cover all different

84 sensory modalities and include high, medium, and low intense attributes. Selected samples were 14% and 10% ethanol levels with low levels of caffeic acid (20 mg/L) and glycerol (5 g/L).

For panelist agreement, the similarity between one panelist and the rest of the panel was quantified by comparing a matrix of TCATA data for one panelist to a matrix of TCATA data for the rest of the panel, and expressed as an index between 0 and 1, with 1 indicating perfect similarity. Second, panelist consistency across the three sensory replicates was quantified as the similarity between repeated evaluations of a single panelist, and expressed as an index between 0 and 1, with 1 indicating perfect panelist consistency.

Data resulting from a TCATA study is binary, meaning that it is a series of zeros and ones for each attribute, sensory replicate, and panelist. Every second where an attribute is selected gets marked as a one. If the attribute is not selected, it gets marked as a zero. To assess the effects of our experimental factors – ethanol, glycerol, and caffeic acid levels, the TCATA data was treated as count data. This approach is similar to a prior study where proportions of selections were used for data analysis by calculating the proportion of panelists who selected an attribute at any given moment during evaluation.39 The authors used citation proportions for each sensory attribute for every 0.1 seconds, which was their selected time segmentation.

Similarly, to test which experimental factors (ethanol, glycerol, and caffeic acid) and interactions significantly affect the sensory attributes throughout the three-minute evaluation period, Analysis of Variance (ANOVA) was run for each of the 9 attributes, using 5-second time bins. These 5-second time segments should be small enough to

85 detect temporal changes during the 3-minute evaluation period, but allow for us to test for significant differences due to the experimental factors without being affected by slightly different selection times of the individual panelists, similar to Dinnella et al. and

Lepage et al.97,98

Last, a trajectory principal component analysis (PCA) was created to visualize the temporal changes of each sample during the 3-minute evaluation period. This type of PCA can be understood as the super-positioning of individual PCAs run at every time segment; the resulting lines, or trajectories, indicate how each sample changes throughout the evaluation period. This allows us to follow and better understand the sensory changes of each sample throughout time as well as enables us to compare samples to each other. It is a very powerful and elegant visualization method that presents literally several thousands of data points into one easy-to-understand figure.39

3.3 Results and Discussion

3.3.1 Panelist Consistency and Panel Agreement

The initial analysis of the TCATA data tested how in agreement the individual panelists were to the panel as a whole. To assess how well each panelist agreed with the panel, three different sensory attributes (citrus flavor, sweet taste, and warm mouthfeel) and two different ethanol levels (10% and 14%) were selected for this analysis. This allowed for the testing of the agreement for all three sensory modalities

(retronasal aroma, basic taste, mouthfeel) and the most extreme samples, i.e., the ones one would expect to significantly differ.

Second, the repeatability of each panelist was assessed to test whether they were consistent in their evaluations across the three sensory replicates.

The results for both analyses are summarized in Table 9. Overall, the panelists were in good agreement with the rest of the panel, with agreement indices in the range of 0.7 to 0.8, which can be considered to be very good, as values range from 0 to 1, with 0 indicating complete disagreement and 1 indicating complete agreement. For example, Panelists 4, 7, 8, and 10 had all but one of their indices in the range of 0.7 to 0.85. An exception to this trend can be seen for Panelist 1, who showed the least agreement with the rest of the panel (values between 0.16-0.60). As participants are human subjects, there is always variability between people and it is not unexpected that someone may stray from the group. However, as the majority of panelists were in high agreement, all data were kept for analysis, and no panelist was removed from the data.

Second, the repeatability index can take values between 0 and 1, where 0 is not repeatable and 1 is completely repeatable. As shown in Table 9 all panelists showed very high consistency across the three sensory replicates as repeatability indices ranged from 0.75 to 0.88.

Overall, the TCATA panel showed both very good panel agreement and repeatability across sensory replicates, indicating high quality TCATA data.

Table 9. Panelist agreement indices for the three different sensory attributes (citrus flavor, warm/hot mouthfeel, and sweet taste) and two ethanol levels (10 and 14% v/v), and repeatability indices.

Panelist 10% Citrus 14% Citrus 10% Warm/Hot 14% Warm/Hot 10% Sweet 14% Sweet Repeatability

Panelist 1 0.48 0.46 0.46 0.60 0.16 0.16 0.78 Panelist 2 0.73 0.78 0.54 0.62 0.79 0.80 0.85 Panelist 3 0.65 0.75 0.68 0.73 0.75 0.78 0.82 Panelist 4 0.75 0.76 0.65 0.75 0.79 0.82 0.88

Panelist 5 0.76 0.69 0.61 0.74 0.72 0.82 0.77 Panelist 6 0.74 0.79 0.64 0.58 0.79 0.82 0.87 Panelist 7 0.74 0.74 0.67 0.75 0.79 0.78 0.75 Panelist 8 0.71 0.77 0.65 0.72 0.77 0.80 0.86 Panelist 9 0.53 0.78 0.55 0.67 0.76 0.75 0.81

Panelist 10 0.73 0.79 0.61 0.75 0.8 0.76 0.86 Panelist 11 0.71 0.76 0.64 0.75 0.67 0.82 0.81 Panelist 12 0.76 0.75 0.65 0.75 0.67 0.57 0.79

3.3.2 Effect of varying ethanol, glycerol, and caffeic acid on temporal sensory

evaluation

The effects of varying levels of ethanol, caffeic acid, and glycerol on the sensory properties of the white, hybrid model wines were evaluated by Analysis of

Variance (ANOVA) applied to each sensory attribute and time bin.

The three experimental variables – ethanol, caffeic acid, and glycerol, were treated as fixed effects, and all possible interactions between the factors were included. Individual panelists nor the replicates were included in this model, as panelists were found to be in agreement with one another and consistent across the three replicates. The results of these ANOVAs can be found in Appendix Tables 1-9.

Looking at Appendix Tables 1-9, significant factors or interactions were typically found for a prolonged period of time (at least 10-15 sec). Take the analysis of apple flavor, as an example (Appendix Table 3): varying the ethanol level had a significant

88 effect on apple flavor for the first 15 seconds of evaluation, with higher ethanol levels leading to less apple flavor. In general, varying the ethanol content affected 8 of the 9 total attributes, with higher levels of ethanol increasing bitter taste, astringent mouthfeel, warm mouthfeel, and other taste while decreasing apple flavor, citrus flavor, sweet taste and sour taste. This held true throughout most of the evaluation period, with a higher number of significant differences due to varying ethanol levels found in the beginning portion of evaluation for most flavors and throughout the middle/end portion of evaluation for bitter taste and the mouthfeel attributes.

In very few instances did the factors and interactions cause a significant difference between samples for just a singular five second segment. Looking at

Appendix Table 3 again, we see that glycerol had a significant effect on apple flavor for the segment from 30 seconds through 34 seconds. More apple flavor was perceived in the high glycerol samples. However, because of the very brief nature of this significant effect, it is likely that the overall impact was small and possibly only detected by a couple of panelists.

The important finding from the ANOVAs is the ability to determine which factors and potential interactions between these factors significantly affect sensory attributes at a specific point during the 3-minute evaluation period. Compared to DA, where this is done at one point in time, the dynamic sensory method of TCATA provides the ability to look at these effects over a time period. For example, ethanol had a significant effect on warm/hot mouthfeel for the first 119 seconds (Appendix Table 5) as well as significantly affecting warm/hot mouthfeel from 155 seconds to 180 seconds. As one would expect, warm/hot mouthfeel increased with increasing levels

89 of ethanol, and the same result was found by DA. By using TCATA however, one is able to not only identify that there are significant differences well after the initial sip but also identify the factors and interactions that cause the significant differences. A great example of why this is important can be seen in Appendix Table 6. From 125 seconds to 164 seconds, significant differences in bitter taste were found between the samples, caused by the interaction between ethanol and glycerol, which is responsible for the differences from 125 seconds to 154 seconds, and that a different interaction, namely between caffeic acid and glycerol, is responsible for the differences in bitter taste from

155 seconds to 164 seconds. Figure 10 shows the interaction between ethanol and glycerol on bitter taste at 135 seconds. This is an interesting interaction as based on the DA results one would expect added glycerol to decrease bitterness and added ethanol to increase bitterness. This is indeed observed for the 12% ethanol level; however, this is found not to be true for either the 10% or 14% ethanol levels, which both show higher bitter taste levels than the 12% ethanol sample. This indicates that the glycerol may not be able to overcome the bitterness provided by the ethanol at the

14% levels, however, it is unclear what is happening at the 10% level. It is also important to note that two of the points in Figure 10 have a citation rate of 0. This means that none of our panelists selected the bitter attribute at this time segment.

Figure 10. Interaction plot for the significant two-way interaction between ethanol and glycerol on bitter taste at 135 seconds.

Similarly, for citrus flavor (Appendix Table 1), multiple factors and interactions were responsible for significant sample differences at different times. Ethanol was responsible for significant sample differences from 0 to 44 seconds, 115 to 144 seconds and 150 to 154 seconds, where higher ethanol levels decreased citrus flavor.

Additionally, glycerol was found to be responsible for significant sample differences from 15 to 19 seconds, where glycerol increased citrus flavor at higher levels. At the

15-19 second time segment, there was no significant interaction between ethanol and glycerol, but there was a significant interaction between ethanol and caffeic acid. This indicates that even though caffeic acid did not have a main effect on citrus flavor, it still had an indirect impact.

The interaction between ethanol and caffeic acid was responsible for significant differences from 15 to 24 seconds, while the interaction between caffeic acid and glycerol was responsible for significant differences from 5 to 9 seconds. With this example, it becomes apparent that different factors and interactions lead to significant sample differences, and that they may act at the same time or at different times. Of further note and interest is that these observed results for citrus flavor are all indirect modes of action, as citrus flavor is not directly linked to any of the experimental factors, like warm mouthfeel and ethanol are.

Overall, the sensory attributes seem to be more affected by the experimental factors and interactions towards the beginning of evaluation, roughly the first 45 seconds. Flavor attributes, such as citrus, apple, and pear, really only show significant differences within the first 45 seconds. The only exception to this is the decrease in citrus flavor caused by increasing ethanol levels from 115 to 144 seconds and from

150 to 154 seconds. Bitter taste and warm/hot mouthfeel were the only two attributes to show significant differences during the final two minutes of evaluation, caused by a multitude of factors and interactions, namely ethanol, glycerol, the interaction between ethanol and glycerol, and the interaction between caffeic acid and glycerol.

Generally, citrus flavor, apple flavor, sour taste, and warm/hot mouthfeel were most affected by varying ethanol levels. Pear flavor, sweet taste, and astringent mouthfeel were most affected by glycerol levels, and bitter taste was highly affected by both ethanol and glycerol along with their interaction.

3.3.3 Visualization of temporal changes in sensory attributes due to the

experimental factors

As a final analysis, the temporal changes due to varying the experimental factors were visualized with Principal Component Analysis (PCA) (Figure 11). In this

PCA, each of the 12 different samples are shown with their trajectories. As we follow the trajectory lines the changes in the sensory attributes are depicted, and it is possible to track the sensory evolution of each sample over the 180-second evaluation period.

Plotting the first two dimensions of the resulting PCA, close to 95% of the total variance is captured, with the first principal component (PC 1) explaining 85%, thus, capturing most of the variability.

Starting from the left-hand side of the PCA biplot (Figure 11), all samples start at the same location, i.e., the beginning of the evaluation where none of the attributes are selected. As the evaluation progresses, samples travel towards the right-hand side of the PCA biplot, and starting to separate based on the experimental factors: One would expect that samples higher in ethanol would be described as warmer and more bitter while samples higher in glycerol would be described as fruitier and sweeter.

Looking at Figure 11, higher ethanol samples (in shades of red) are always closer to the warm/hot and bitter attributes while the lower ethanol samples (in shades of blue) tend to be closer to fruity, sweet, and sour attributes, especially towards the beginning of evaluation. We can also see that the higher glycerol samples trended towards the fruity and sweet attributes more than the lower glycerol samples.

Figure 11. Biplot of PCA trajectories for the model wine samples with samples color- coded by ethanol content (shades of blue represent 10% EtOH; shades of green represent 12% EtOH; shades of red represent 14% EtOH). Capital letters in the sample name represent the high level of the factor and lowercase letters represent the low level of the factor.

Samples begin on the left side of the plot and trend towards the right before curling back towards where they began as the evaluation progresses. These trajectories show that for most of the samples, the panelists perceived sweetness, sourness, apple flavor, citrus flavor, and pear flavor fairly quickly (within the first 45 seconds) while attributes like astringency, bitter taste, and warm/hot mouthfeel were perceived later in evaluation, as the latter attributes are located more towards the right-hand side of the plot. Figure 11 also shows that bitter taste and warm/hot mouthfeel are located far apart from sour taste. As the high ethanol samples were described by bitter taste and warm/hot mouthfeel, it is likely that the bitter taste brought out by ethanol is suppressing the sour taste that would be expected in the samples.

Such suppressing effects have been reported in the past for white wines by Fischer and Noble. They demonstrated that sour taste was suppressed by the bitterness of increased ethanol in white wine, when assessed by static DA.16

To further explore the differences between samples and to gain a better understanding which of these trajectories depict significant sensory differences,

Figure 12 shows the so-called difference curve plot for two samples, varying only in ethanol content: 10 C G B and 14 C G B. These two samples are shown in the PCA plot as the dark blue (10 C G B) and the dark red line (14 C G B). A difference curve plot shows only those attributes which differed significantly between the two samples throughout the duration of evaluation. Attributes shown above the x-axis were higher in the 10 C G B sample and attributes beneath the x-axis were higher in the 14 C G B sample. Attributes like apple and citrus flavor, along with sweet taste, were significantly higher in the 10 C G B sample for about the first 40 seconds of evaluation.

In contrast, warm/hot mouthfeel was significantly higher in the 14 C G B sample for the first 110 seconds of evaluation and then again for about 15 seconds at the 150 second mark. These results are reflected in Figure 11 as well, as the dark blue line

(sample 10 C G B) is more closely associated with apple and citrus flavor and sweet taste, while the dark red line (sample 14 C G B) is closest to warm/hot mouthfeel. Of interest to note in Figure 12 is that sample 14 C G B showed significantly more pear flavor from about 40 to 65 seconds. While this may not be what one would expect, the same trend is captured in Figure 11, where the pear flavor attribute is located towards the middle of all of the trajectories. This means that there was not one specific group

95 of samples that was exclusively described by pear flavor and that most samples exhibited pear flavor at some point during evaluation.

Figure 12. Difference curves for samples 10 C G B and 14 C G B indicating attributes that showed significant differences during evaluation.

There are a couple of overall conclusions from Figure 11. Samples group mainly by ethanol content, which is also driving separation along PC 2. Attributes like apple flavor, citrus flavor, sweet taste, and sour taste were perceived more at the start of evaluation, while attributes like bitter taste, astringent mouthfeel, and warm/hot mouthfeel were perceived more at the end of evaluation. Bitter and sour tastes are far apart on PC 2. This distance likely suggests a suppressive interaction between the two tastes.

3.3.4 Comparing static DA to dynamic TCATA evaluation

Comparing the TCATA results to the DA results, some similarities are immediately apparent. For example, comparing the experimental factors that significantly affected astringent mouthfeel, in both evaluation methods, varying

96 ethanol and glycerol levels were found to significantly alter astringency perception. In general, increasing ethanol levels led to increase in astringency, while increasing glycerol led to less astringency. As DA intensity ratings were made immediately after taking a sip of the sample, this alignment with the TCATA results makes sense as in

TCATA ethanol and glycerol had a significant effect for the first 5-second time segment. Sometimes, like ethanol’s effect on astringent mouthfeel, these significant effects disappear after the first 5 seconds. Other times, like ethanol’s effect on warm/hot mouthfeel, these significant effects linger for up to 2 minutes.

As mentioned before, most of the significant differences, caused by the experimental factors, appear within the first 45 seconds of evaluation. These may or may not be captured with classical DA, where panelists usually rate their first impressions. Factors and interactions responsible for differences later on in evaluation, for example, past 60 seconds, are definitely not captured with classical DA and require a dynamic sensory method to be unveiled: More of the mouthfeel attributes, like warm/hot and astringent, showed significant differences later on in evaluation. These differences were typically caused by ethanol, glycerol, and the interaction between those two factors. This is not surprising as the mouthfeel attributes tend to linger after the taste and flavor attributes have disappeared. The flavors were also affected more in the first 40 seconds of evaluation, although citrus flavor did show some significant differences in the final minute of evaluation. Again, these differences were mainly caused by ethanol and glycerol. For citrus flavor, caffeic acid was involved in two significant interactions. For the basic tastes, most differences were found for bitter and sweet but not sour. Sweetness showed differences in the first two

97 minutes caused by ethanol and glycerol, while bitterness showed differences throughout evaluation, caused by ethanol, glycerol, the interaction between ethanol and glycerol, and the interaction between caffeic acid and glycerol.

The significant interactions from the two evaluation methods do not match up perfectly. For apple flavor, in DA, significant differences between samples were caused by glycerol. Although similar results were found with TCATA, initial differences were caused by ethanol, while significant differences caused by glycerol did not appear until 30 seconds into evaluation. Pear flavor was significantly affected by ethanol and glycerol in DA, while in TCATA, only glycerol caused significant sample differences, lasting for 30 seconds and starting at the 10 second mark. Citrus flavor is another example of where results between DA and TCATA did not match up perfectly.

In DA, differences between samples in citrus flavor were caused by the interaction between ethanol and glycerol. In TCATA, differences were seen for the first 44 seconds and then for about 35 seconds in the final minute. These differences were caused by ethanol, glycerol, the interaction between ethanol and caffeic acid, and the interaction between caffeic acid and glycerol. The taste data aligned better than the flavor data. Only ethanol was responsible for sour taste differences in both DA and

TCATA. The DA results for sweet taste had more significant interactions than the

TCATA results but ethanol and glycerol were both significant main effects. For bitter taste, only ethanol had a significant effect in DA. In addition to ethanol, the other significant factors in the TCATA results were glycerol, the interaction between ethanol and glycerol, and the interaction between caffeic acid and glycerol. The mouthfeel results are more like the taste results where pretty good agreement between the two

98 methods was found with only some slight differences. The warm/hot mouthfeel results match up really well with ethanol being the only significant factor for both the DA and the TCATA results. In the TCATA results, significant differences were seen for the first

2 minutes and then the final 25 seconds of evaluation. In the DA, astringency was significantly affected by ethanol and the interaction between ethanol and glycerol, while in TCATA, ethanol and glycerol, but not their interaction affected astringent mouthfeel. Additionally, significant interaction effects in the TCATA results were not always seen in the first 0 to 5 second time segment and appeared at later times during evaluation. For example, significant differences in pear flavor did not appear until the

10-14 second interval. This could mean that these interaction effects were not picked up by DA and required a temporal method to find them. This could also be the effect of panelists taking a couple of seconds to select their first attributes.

It is important to keep in mind that DA results are intensity ratings of flavors, tastes, and mouthfeel attributes, while in TCATA, panelists provide binary (present- not present) data that gets converted into frequencies of whether or not the attributes are present. This may explain why factors that have a significant effect on the sensory attributes in the DA were not found to differ significantly at the beginning of TCATA.

While previous studies have shown that intensity data is generally highly correlated with frequency data,48 this does not necessarily mean that they will always provide the exact same results and attribute significant differences to the same factors or interactions. Comparing DA to TCATA, the latter is a less refined method, as it allows only for 2 outcomes (present or absent), while a scale, as used in DA, allows for more nuanced evaluations.

Such an example of discrepancies between DA results and TCATA results can be seen for apple flavor. The DA results for apple flavor show that glycerol and the interaction between ethanol, caffeic acid, glycerol, and protein both had a significant effect on apple favor, with increasing apple flavor at lower ethanol and higher glycerol levels. However, in the TCATA results only ethanol showed a significant effect on apple flavor within the first 14 seconds of evaluation, and no other significant experimental factors or interactions were found at a latter point in time.

The study by McMahon et al. compared DA results to TCATA results. Their initial results showed a high correlation (RV=0.98) between DA and TCATA results.

For this analysis though, the TCATA data used was proportion of citations calculated across the entire evaluation period. This loses the details about when attributes were perceived. TCATA data was also binned into 15 second time segments and multiple factor analysis of mouthfeel attributes in the two data sets revealed that DA results were closely associated with the TCATA results in the first 30 seconds of evaluation.39

This is expected as DA is capturing intensities immediately after taking the sample in mouth and not 1 to 2 minutes later.

3.4 Conclusion

Overall, the findings of this study are the first to be reported for a model white, hybrid wine. Besides significant samples differences throughout the evaluation period of 3 minutes, these significant changes in the sensory attributes could also be attributed to specific experimental factors and interactions, due to the use of a factorial design.

100

Similar to other studies that compared results from DA and temporal evaluation, a complex product such as wine may be fully described by pairing a static sensory method, like DA, with a dynamic one, such as TCATA, to fully understand the evolving nature of perception of an ever-changing product. Descriptive analysis was able to create the sensory profile of a sample at a single point in time but by pairing it with a dynamic method, like TCATA, one is able to monitor attribute changes over time. It is also possible to then link these changes back to the responsible experimental factors, similar to what was done in this study. These temporal changes are important to fully understand a consumer’s sensory experience.

Future work following up on this study should validate these findings in a real white, hybrid wine, in order to fully understand the complex interactions between wine matrix components and aroma compounds. Only then could one provide suitable recommendations on matrix component levels to winemakers that are looking to improve the sensory experience and ultimately the overall quality of their wines.

101

4 Conclusions and Future Work

4.1 Significance

Results from the DA portion of this study indicate that all experimental factors, ethanol, glycerol, phenolics, and protein, have an effect on sensory perception as evaluated in a model white hybrid wine matrix. Among all factors, ethanol, which was varied between 10 and 14% (v/v), had the largest effect. Varying the ethanol levels led to changes in 8 sensory attributes in the static descriptive analysis (DA) and 8 attributes in the dynamic temporal check-all-that-apply (TCATA). Ethanol has been shown to be a sensory-active compound, inducing aroma (i.e., ethanol aroma), taste

(i.e., bitter), and hot/warm mouthfeel sensations, however, affected sensory attributes in this study also included pear aroma, pear flavor, sweet taste, and sour taste descriptors, which indicates that ethanol also had indirect effects on the sensory perception of wine. These indirect effects could be explained by chemical changes in headspace partitioning, as shown by Robinson et al.,21 as well as perceptual mixing effects, such as mixture suppression and enhancement.2 An example for the latter could be the observed effects in the study from Fischer and Noble where ethanol was found to suppress sour taste.16

The effects of ethanol were found for both modes of sensory evaluation (DA and TCATA), indicating that ethanol plays a major role in static and dynamic wine perception.

Glycerol, which was kept at either 5 or 15 g/L, and protein, modeled by varying bovine serum albumin (BSA) content between 0.1 and 0.5 g/L, showed moderate

102 effects as both factors affected a smaller number of sensory attributes (5 in DA and 7 in TCATA). The observed effects for sweet taste could be directly attributed to glycerol, as it has been reported to elicit a sweet taste sensation once the difference in levels reaches 5.2 g/L.23 It could be that adding glycerol masks the bitterness or sourness of the wine matrix, and that this masking effect inversely leads to increased sweetness. Such an effect was reported before in Jones et al. where glycerol was shown to suppress bitter taste.6

For the protein, BSA was chosen as a model compound. The selection was based on prior studies.22 Even though, compared to white wine proteins, BSA is larger

(average molecular weight of white wine proteins was reported to be around 21-

32 kDa vs. an average MW of 66 kDa for BSA), the use is justified as BSA was used as a model wine protein by Serafini et al. to study the effects of ethanol on tannin- protein interactions in red wine.22 The effects on the sensory perception by BSA were moderate as only one sensory attribute showed significant changes when BSA levels were changed in the DA study.

Last, the phenolic composition of the model white wine was modeled by caffeic acid, which was kept at either 20 or 60 mg/L. The major phenolic group in white wine are hydroxycinnamates, and caffeic acid is one of the major hydroxycinnamates reported in white wine.10 Caffeic acid did not have any significant main effects on any of the sensory attributes in the model wine samples. However, it was involved in seven interaction effects. This suggests that even though caffeic acid did not have any main effects on sensory attributes, it still contributed, at least indirectly, to the perception of the sensory attributes in the samples.

103

Besides the described main effects, several significant interaction effects were found as well. The presence of interaction effects shows that factors behave differently at different levels, i.e., the main effects of a factor depends on the levels of the other factors. Several two-way interactions between ethanol and the other factors were found in the DA results.

The instrumental measurements of changes in the headspace concentrations of the aroma fraction volatiles also showed significant effects by all four experimental factors. Similar to the sensory results, all interactions, except for the one between caffeic acid, glycerol, and protein, were significant. Ethanol had a significant effect on

13 of the 14 aroma compounds. Phenethyl alcohol was the only aroma compound unaffected by ethanol. Ethanol generally caused a decrease in the aroma compounds as it increased. The only two compounds that showed increases were acetic acid and ethyl decanoate. Glycerol had a significant effect on 5 of the 14 aroma compounds.

Ethyl hexanoate, isoamyl acetate, ethyl butanoate, and ethyl acetate all decreased as glycerol increased. Acetic acid increased with increasing levels of glycerol. Caffeic acid only had a significant effect on acetic acid. As caffeic acid increased, acetic acid increased. Finally, protein had an effect on 3 of the 14 aroma compounds. Phenethyl acetate, linalool, and -damascenone all decreased as protein increased.

In this work, two different modes of wine sensory evaluation were compared, namely static DA and dynamic TCATA. The same model wines were evaluated by both methods, allowing a direct comparison between the two methods.

104

TCATA showed less significant interactions than DA, suggesting that pairing a dynamic sensory method, like TCATA, with a static sensory method, like DA, would be needed to obtain a full understanding of sensory perception of an evolving sample.

Both methodologies detected significant differences in different sensory attributes, as indicated above, however, it seems that they complement each other, as slight differences were found. We speculated that TCATA results for the first 10-20 seconds of evaluation would match up well with the DA results, since DA ratings were done immediately after taking the sample in mouth. This was found to be true for the apple flavor, sweet taste, sour taste, and warm/hot mouthfeel, where similar conclusions about significant drivers were made.

However, for other attributes, like citrus flavor and bitter taste, the two methods did not compare as well. This is due to two different patterns, (i) for the earlier time of the TCATA evaluation, which should be most similar to the DA results, citrus flavor, for example, was significantly affected by the interactions between ethanol and glycerol and between caffeic acid and protein in the DA, while the major driver for differences in TCATA resulted from ethanol, glycerol, the interaction between ethanol and caffeic acid, and the interaction between glycerol and caffeic acid; (ii) significant changes in citrus flavor, bitter taste, sweet taste, sour taste, and warm/hot mouthfeel were found in TCATA for the later parts of evaluation (after 1 min). These changes are typically not captured by the DA, thus, the observed results in TCATA cannot be compared to DA.

Last, we attempted to correlate sensory attributes and the observed changes to chemical measurements. Partial least squares regression (PLSR) was used to

105 study whether changes in the headspace concertation of the aroma fraction, as measured instrumentally by HS-SPME-GC-MS, could be linked to the observed sensory changes. Besides the headspace concentration, the experimental factor concentrations were also included in the model as a positive control as one would expect that direct effects of those factors would be appropriately modeled.

As expected, ethanol content was able to predict ethanol aroma and warm/hot mouthfeel with high accuracy (RSMEP of about 2), however, none of the changes in volatile headspace concentration could accurately predict any of the sensory attributes. While such a result was expected for the taste and mouthfeel attributes, we would have expected that at least some of the sensory aroma descriptors would have been correlated to the headspace concentration of the aroma fraction volatiles. This is based on the fact that the same aroma fraction volatiles were the only source of aroma and flavor in the model wines; however, it is also known that aroma perception of complex products, such as wine is very rarely directly linked to a single or a few compounds. In most cases the resulting aroma and flavor perceptions are the result of perceptual mixture effects.21

In summary, all of our factors were found to affect, either directly or indirectly, the sensory attributes of our model wine samples in the DA portion of the study. The same can be said for our TCATA portion. Also, in the TCATA portion, we saw significant temporal differences that were not picked up by DA. Results for attributes like apple flavor and warm/hot mouthfeel matched up really well between the two data sets while results for attributes like citrus flavor and astringent mouthfeel did not match

106 up as well. The PLSR model that we used did fit our data well, but it was unable to predict the sensory attributes of our samples using our instrumental data.

4.2 Future Directions

The obtained results demonstrate that a model white hybrid wine system seems to be an appropriate way to study the effects of matrix components on aroma, flavor, taste, and mouthfeel perceptions. The results clearly indicate that wine matrix components, such as ethanol, protein, glycerol, and phenolics are able to significantly change both static and dynamic sensory perception, however, these findings should be validated in real wine systems.

As the real wine applications would have to control the specific wine matrix component of interest, commercial products could be used, or alternatively, research wines would have to be made, similar to the approach by Frost et al. Through different must treatments, they were able to produce red Cabernet Sauvignon wines that differed in tannin levels from ~700 to ~1200 mg/L total phenol content. After fermentation was complete, water and tartaric acid were added to the treatments to alter levels of ethanol between 14% and 15% (v/v) and acidity between pH 5 and 7.27

Similar measures could be taken to make, for example, hybrid, Traminette wines in- house to control matrix component concentrations to the highest degree possible.

Replicating this study in a real hybrid wine system would thus provide the wine industry with actionable recommendations on how to enhance or suppress specific sensory attributes in their wines, thus, improving overall wine quality and consumer appeal.

107

4.3 Overall Conclusions

In closing, wine matrix components and their interactions can have significant effects on both the static and the temporal perception of sensory attributes. Factors like ethanol and glycerol were shown to have larger effects, while factors like phenolics and protein had smaller effects. These factors also seem to have more of an effect on taste and mouthfeel attributes than aroma and flavor attributes. Most significant effects were observed within the first 35 seconds of evaluation, but changes in some sensory attributes were observed over two minutes after consumption. These temporal changes are not picked up by classical DA. Static and dynamic sensory methods should be paired with each other in order to get a complete sensory picture of an evolving sample. It is our hope that this work provides more details on the static and temporal effects that matrix composition has in a white, hybrid wine. Despite being important to the world’s wine industry, this area is largely understudied. We also hope that this work provides a better understanding of how to work with and analyze TCATA data, especially in studies pertaining to wine. Once this work is validated in real wine, this information could be incredibly valuable to winemakers producing white, hybrid wines. Winemakers could be able to produce wines with specific sensory profiles due to the manipulation of specific matrix factors as discussed in this study. For example, winemakers looking for more pear flavor in their wines may look to keep alcohol content lower and glycerol content higher. This, in turn, will hopefully lead to more profitability for the winemakers and more enjoyable experiences for the consumers.

108

References

(1) Shepherd, G. M. Smell Images and the Flavour System in the Human Brain.

Nature 2006, 444 (7117), 316–321. https://doi.org/10.1038/nature05405.

(2) Lawless, H. T.; Heymann, H. Physiological and Psychological Foundations of

Sensory Function. In Sensory Evaluation of Food; 2010; pp 19–56.

(3) Rozin, P. “Taste-Smell Confusions” and the Duality of the Olfactory Sense.

Percept. Psychophys. 1982, 31 (4), 397–401.

https://doi.org/10.3758/BF03202667.

(4) Goldner, M. C.; Zamora, M. C. Effect of Polyphenol Concentrations on

Astringency Perception and Its Correlation with Gelatin Index of Red Wine. J.

Sens. Stud. 2010, 25 (5), 761–777. https://doi.org/10.1111/j.1745-

459X.2010.00304.x.

(5) Robinson, A. L.; Adams, D. O.; Boss, P. K.; Heymann, H.; Solomon, P. S.;

Trengove, R. D. The Relationship between Sensory Attributes and Wine

Composition for Australian Cabernet Sauvignon Wines. Aust. J. Grape Wine

Res. 2011, 17 (3), 327–340. https://doi.org/10.1111/j.1755-

0238.2011.00155.x.

(6) Jones, P. R.; Gawel, R.; Francis, I. L.; Waters, E. J. The Influence of

Interactions between Major White Wine Components on the Aroma, Flavour

and Texture of Model White Wine. Food Qual. Prefer. 2008, 19 (6), 596–607.

https://doi.org/10.1016/j.foodqual.2008.03.005.

(7) Polášková, P.; Herszage, J.; Ebeler, S. E. Wine Flavor: Chemistry in a Glass.

Chem. Soc. Rev. 2008, 37 (11), 2478. https://doi.org/10.1039/b714455p.

109

(8) González Álvarez, M.; González-Barreiro, C.; Cancho-Grande, B.; Simal-

Gándara, J. Relationships between Godello White Wine Sensory Properties

and Its Aromatic Fingerprinting Obtained by GC-MS. Food Chem. 2011, 129

(3), 890–898. https://doi.org/10.1016/j.foodchem.2011.05.040.

(9) Vilanova, M.; Genisheva, Z.; Masa, A.; Oliveira, J. M. Correlation between

Volatile Composition and Sensory Properties in Spanish Albariño Wines.

Microchem. J. 2010, 95 (2), 240–246.

https://doi.org/10.1016/j.microc.2009.12.007.

(10) Waterhouse, A. L.; Sacks, G. L.; Jeffery, D. W. Understanding Wine

Chemistry; 2016. https://doi.org/10.1002/9781118730720.

(11) Allen, A. L.; Mcgeary, J. E.; Hayes, J. E. Polymorphisms in TRPV1 and

TAS2Rs Associate with Sensations from Sampled Ethanol. Alcohol. Clin. Exp.

Res. 2014, 38 (10), 2550–2560. https://doi.org/10.1111/acer.12527.

(12) King, E. S.; Dunn, R. L.; Heymann, H. The Influence of Alcohol on the Sensory

Perception of Red Wines. Food Qual. Prefer. 2013, 28 (1), 235–243.

https://doi.org/10.1016/j.foodqual.2012.08.013.

(13) King, E. S.; Heymann, H. The Effect of Reduced Alcohol on the Sensory

Profiles and Consumer Preferences of White Wine. J. Sens. Stud. 2014, 29

(1), 33–42. https://doi.org/10.1111/joss.12079.

(14) Villamor, R. R.; Evans, M. A.; Ross, C. F. Effects of Ethanol, Tannin, and

Fructose Concentrations on Sensory Properties of Model Red Wines. Am. J.

Enol. Vitic. 2013, 64 (3), 342–348. https://doi.org/10.5344/ajev.2013.12118.

(15) Gawel, R.; Van Sluyter, S. C.; Smith, P. A.; Waters, E. J. Effect of PH and

110

Alcohol on Perception of Phenolic Character in White Wine. Am. J. Enol. Vitic.

2013, 64 (4), 425–429. https://doi.org/10.5344/ajev.2013.13016.

(16) Fischer, U.; Noble, A. C. The Effect of Ethanol, Catechin Concentration, and

PH on Sourness and Bitterness of Wine. Am. J. Enol. Vitic. 1994, 45 (1), 6–10.

(17) Martin, S.; Pangborn, R. M. Taste Interaction of Ethyl Alcohol with Sweet,

Salty, Sour and Bitter Compounds. J. Sci. Food Agric. 1970, 21 (12), 653–655.

https://doi.org/10.1002/jsfa.2740211213.

(18) Goldner, M. C.; Zamora, M. C.; Lira, P. D. L.; Gianninoto, H.; Bandoni, A.

Effect of Ethanol Level in the Perception of Aroma Attributes and the Detection

of Volatile Compounds in Red Wine. J. Sens. Stud. 2009, 24 (2), 243–257.

https://doi.org/10.1111/j.1745-459X.2009.00208.x.

(19) Siebert, K. J.; Carrasco, A.; Lynn, P. Y. Formation of Protein-Polyphenol Haze

in Beverages. J. Agric. Food Chem. 1996, 44 (8), 1997–2005.

https://doi.org/10.1021/jf950716r.

(20) PubChem.

(21) Robinson, A. L.; Ebeler, S. E.; Heymann, H.; Boss, P. K.; Solomon, P. S.;

Trengove, R. D. Interactions between Wine Volatile Compounds and Grape

and Wine Matrix Components Influence Aroma Compound Headspace

Partitioning. J. Agric. Food Chem. 2009, 57 (21), 10313–10322.

https://doi.org/10.1021/jf902586n.

(22) Serafini, M.; Maiani, G.; Ferro-Luzzi, A. Effect of Ethanol on Red Wine Tannin

- Protein (BSA) Interactions. J. Agric. Food Chem. 1997, 45 (8), 3148–3151.

https://doi.org/10.1021/jf960864x.

111

(23) Noble, A. C.; Bursick, G. F. The Contribution of Glycerol to Perceived Viscosity

and Sweetness in White Wine. Am. J. Enol. Vitic. 1984, 35 (2), 110–112.

(24) Waterhouse, A. L. Wine Phenolics. Ann. N. Y. Acad. Sci. 2002, 957 (1), 21–

36. https://doi.org/10.1111/j.1749-6632.2002.tb02903.x.

(25) Arnold, R. A.; Noble, A. C.; Singleton, V. L. Bitterness and Astringency of

Phenolic Fractions in Wine. J. Agric. Food Chem. 1980, 28 (3), 675–678.

https://doi.org/10.1021/jf60229a026.

(26) Robichaud, J. L.; Noble, A. C. Astringency and Bitterness of Selected

Phenolics in Wine. J. Sci. Food Agric. 1990, 53 (3), 343–353.

https://doi.org/10.1002/jsfa.2740530307.

(27) Frost, S. C.; Harbertson, J. F.; Heymann, H. A Full Factorial Study on the

Effect of Tannins, Acidity, and Ethanol on the Temporal Perception of Taste

and Mouthfeel in Red Wine. Food Qual. Prefer. 2017, 62, 1–7.

https://doi.org/10.1016/j.foodqual.2017.05.010.

(28) Sun, S. Y.; Zhang, Q. F.; Liu, W. L.; Li, H. M.; Liu, Y. L.; Jiang, X. M.; Zhao, Y.

P. Influence of Maceration Techniques on the Chemical, Aromatic, Sensory

and Biogenic Amine Profiles of Cherry Wine. J. Inst. Brew. 2018, 124 (4), 477–

484. https://doi.org/10.1002/jib.524.

(29) Waters, E. J.; Alexander, G.; Muhlack, R.; Pocock, K. F.; Colby, C.; O’Neill, B.

K.; Høj, P. B.; Jones, P. Preventing Protein Haze in Bottled White Wine. Aust.

J. Grape Wine Res. 2005, 11 (2), 215–225. https://doi.org/10.1111/j.1755-

0238.2005.tb00289.x.

(30) Marchal, A.; Marullo, P.; Moine, V.; Dubourdieu, D. Influence of Yeast

112

Macromolecules on Sweetness in Dry Wines: Role of the Saccharomyces

Cerevisiae Protein Hsp12. J. Agric. Food Chem. 2011, 59 (5), 2004–2010.

https://doi.org/10.1021/jf103710x.

(31) Fukui, M.; Yokotsuka, K. Content and Origin of Protein in White and Red

Wines: Changes during Fermentation and Maturation. Am. J. Enol. Vitic. 2003,

3 (3), 178–188.

(32) Marangon, M.; Van Sluyter, S. C.; Waters, E. J.; Menz, R. I. Structure of Haze

Forming Proteins in White Wines: Vitis Vinifera Thaumatin-like Proteins. PLoS

One 2014, 9 (12), 1–22. https://doi.org/10.1371/journal.pone.0113757.

(33) Obreque-Slier, E.; Peña-Neira, Á.; López-Solís, R. Interactions of Enological

Tannins with the Protein Fraction of Saliva and Astringency Perception Are

Affected by PH. LWT - Food Sci. Technol. 2012, 45 (1), 88–93.

https://doi.org/10.1016/j.lwt.2011.07.028.

(34) Gawel, R.; Waters, E. J. The Effect of Glycerol on the Perceived Viscosity of

Dry White Table Wine. J. Wine Res. 2008, 19 (2), 109–114.

https://doi.org/10.1080/09571260802622191.

(35) Lawless, H. T.; Heymann, H. Descriptive Analysis. In Sensory Evaluation of

Food; 2010; pp 227–257. https://doi.org/10.1007/978-1-4419-6488-5_10.

(36) Fanzone, M.; Peña-Neira, A.; Gil, M.; Jofré, V.; Assof, M.; Zamora, F. Impact

of Phenolic and Polysaccharidic Composition on Commercial Value of

Argentinean Malbec and Cabernet Sauvignon Wines. Food Res. Int. 2012, 45

(1), 402–414. https://doi.org/10.1016/j.foodres.2011.11.010.

(37) Sokolowsky, M.; Fischer, U. Evaluation of Bitterness in White Wine Applying

113

Descriptive Analysis, Time-Intensity Analysis, and Temporal Dominance of

Sensations Analysis. Anal. Chim. Acta 2012, 732, 46–52.

https://doi.org/10.1016/j.aca.2011.12.024.

(38) Frost, S. C.; Blackman, J. W.; Ebeler, S. E.; Heymann, H. Analysis of

Temporal Dominance of Sensation Data Using Correspondence Analysis on

Merlot Wine with Differing Maceration and Cap Management Regimes. Food

Qual. Prefer. 2018, 64, 245–252.

https://doi.org/10.1016/j.foodqual.2016.11.011.

(39) McMahon, K. M.; Culver, C.; Castura, J. C.; Ross, C. F. Perception of

Carbonation in Sparkling Wines Using Descriptive Analysis (DA) and

Temporal Check-All-That-Apply (TCATA). Food Qual. Prefer. 2017, 59, 14–26.

https://doi.org/10.1016/j.foodqual.2017.01.017.

(40) Baker, A. K.; Castura, J. C.; Ross, C. F. Temporal Check-All-That-Apply

Characterization of Syrah Wine. J. Food Sci. 2016, 81 (6), S1521–S1529.

https://doi.org/10.1111/1750-3841.13328.

(41) Ares, G.; Jaeger, S. R.; Antúnez, L.; Vidal, L.; Giménez, A.; Coste, B.; Picallo,

A.; Castura, J. C. Comparison of TCATA and TDS for Dynamic Sensory

Characterization of Food Products. Food Res. Int. 2015, 78 (2015), 148–158.

https://doi.org/10.1016/j.foodres.2015.10.023.

(42) Lawless, H. T.; Heymann, H. Time-Intensity Methods. In Sensory Evaluation of

Food; 2010; pp 179–201.

(43) Pineau, N.; Schlich, P.; Cordelle, S.; Mathonnière, C.; Issanchou, S.; Imbert,

A.; Rogeaux, M.; Etiévant, P.; Köster, E. Temporal Dominance of Sensations:

114

Construction of the TDS Curves and Comparison with Time-Intensity. Food

Qual. Prefer. 2009, 20 (6), 450–455.

https://doi.org/10.1016/j.foodqual.2009.04.005.

(44) Lawless, H. T.; Heymann, H. Conetext Effects and Biases in Sensory

Judgement. In Sensory Evaluation of Food; 2010; pp 203–225.

(45) Meillon, S.; Urbano, C.; Schlich, P. Contribution of the Temporal Dominance of

Sensations (TDS) Method to the Sensory Description of Subtle Differences in

Partially Dealcoholized Red Wines. Food Qual. Prefer. 2009, 20 (7), 490–499.

https://doi.org/10.1016/j.foodqual.2009.04.006.

(46) Ares, G.; Antúnez, L.; Giménez, A.; Jaeger, S. R. List Length Has Little Impact

on Consumers’ Visual Attention to CATA Questions. Food Qual. Prefer. 2015,

42, 100–109. https://doi.org/10.1016/j.foodqual.2015.01.015.

(47) Castura, J. C.; Antúnez, L.; Giménez, A.; Ares, G. Temporal Check-All-That-

Apply (TCATA): A Novel Dynamic Method for Characterizing Products. Food

Qual. Prefer. 2016, 47, 79–90. https://doi.org/10.1016/j.foodqual.2015.06.017.

(48) Campo, E.; Ballester, J.; Langlois, J.; Dacremont, C.; Valentin, D. Comparison

of Conventional Descriptive Analysis and a Citation Frequency-Based

Descriptive Method for Odor Profiling: An Application to Burgundy Pinot Noir

Wines. Food Qual. Prefer. 2010, 21 (1), 44–55.

https://doi.org/10.1016/j.foodqual.2009.08.001.

(49) Meyners, M.; Castura, J. C. The Analysis of Temporal Check-All-That-Apply

(TCATA) Data. Food Qual. Prefer. 2018, 67, 67–76.

https://doi.org/10.1016/j.foodqual.2017.02.003.

115

(50) Springer, L. F.; Sacks, G. L. Protein-Precipitable Tannin in Wines from Vitis

Vinifera and Interspecific Hybrid Grapes (Vitis Ssp.): Differences in

Concentration, Extractability, and Cell Wall Binding. J. Agric. Food Chem.

2014, 62 (30), 7515–7523. https://doi.org/10.1021/jf5023274.

(51) Spada, P. Tannin Additions in Red Hybrid Wines.

(52) Springer, L. F.; Chen, L. A.; Stahlecker, A. C.; Cousins, P.; Sacks, G. L.

Relationship of Soluble Grape-Derived Proteins to Condensed Tannin

Extractability during Red Wine Fermentation. J. Agric. Food Chem. 2016, 64

(43), 8191–8199. https://doi.org/10.1021/acs.jafc.6b02891.

(53) Nicolle, P.; Marcotte, C.; Angers, P.; Pedneault, K. Pomace Limits Tannin

Retention in Frontenac Wines. Food Chem. 2019, 277 (March 2018), 438–

447. https://doi.org/10.1016/j.foodchem.2018.10.116.

(54) Burns, J.; Mullen, W.; Landrault, N.; Teissedre, P. L.; Lean, M. E. J.; Crozier,

A. Variations in the Profile and Content of Anthocyanins in Wines Made from

Cabernet Sauvignon and Hybrid Grapes. J. Agric. Food Chem. 2002, 50 (14),

4096–4102. https://doi.org/10.1021/jf011233s.

(55) Favretto, D.; Flamini, R. Application of Electrospray Ionization Mass

Spectrometry to the Study of Grape Anthocyanins. Am. J. Enol. Vitic. 2000, 51

(1), 55–64.

(56) Manns, D. C.; Coquard Lenerz, C. T. M.; Mansfield, A. K. Impact of

Processing Parameters on the Phenolic Profile of Wines Produced from

Hybrid Red Grapes Maréchal Foch, Corot Noir, and Marquette. J. Food Sci.

2013, 78 (5), 696–702. https://doi.org/10.1111/1750-3841.12108.

116

(57) Skinkis, P. A.; Bordelon, B. P.; Butz, E. M. Effects of Sunlight Exposure on

Berry and Wine Monoterpenes and Sensory Characteristics of Traminette.

Am. J. Enol. Vitic. 2010, 61 (2), 147–156.

(58) Andrews, J. T.; Heymann, H.; Ellersieck, M. Sensory and Chemical Analyses

of Missouri Seyval Blanc Wines. Am. J. Enol. Vitic. 1990, 41 (2), 116–120.

(59) Mirarefi, S.; Menke, S. D.; Lee, S. Y. Sensory Profiling of Chardonel Wine by

Descriptive Analysis. J. Food Sci. 2010, 69 (6), S211–S217.

https://doi.org/10.1111/j.1365-2621.2004.tb11007.x.

(60) Escudero, A.; Campo, E.; Fariña, L.; Cacho, J.; Ferreira, V. Analytical

Characterization of the Aroma of Five Premium Red Wines. Insights into the

Role of Odor Families and the Concept of Fruitiness of Wines. J. Agric. Food

Chem. 2007, 55 (11), 4501–4510. https://doi.org/10.1021/jf0636418.

(61) Wang, G.; Hayes, J.; Ziegler, G.; Roberts, R.; Hopfer, H. Dose-Response

Relationships for Vanilla Flavor and Sucrose in Skim Milk: Evidence of

Synergy. Beverages 2018, 4 (4), 73.

https://doi.org/10.3390/beverages4040073.

(62) Wang, G.; Bakke, A. J.; Hayes, J. E.; Hopfer, H. Demonstrating Cross-Modal

Enhancement in a Real Food with a Modified ABX Test. Food Qual. Prefer.

2019. https://doi.org/10.1016/j.foodqual.2019.05.007.

(63) Hort, J.; Hollowood, T. Controlled Continuous Flow Delivery System for

Investigating Taste−Aroma Interactions. J. Agric. Food Chem. 2004, 52 (15),

4834–4843. https://doi.org/10.1021/JF049681Y.

(64) Aznar, M.; López, R.; Cacho, J.; Ferreira, V. Prediction of Aged Red Wine

117

Aroma Properties from Aroma Chemical Composition. Partial Least Squares

Regression Models. J. Agric. Food Chem. 2003, 51 (9), 2700–2707.

https://doi.org/10.1021/jf026115z.

(65) Cozzolino, D.; Smyth, H. E.; Lattey, K. A.; Cynkar, W.; Janik, L.; Dambergs, R.

G.; Francis, I. L.; Gishen, M. Relationship between Sensory Analysis and near

Infrared Spectroscopy in Australian Riesling and Chardonnay Wines. Anal.

Chim. Acta 2005, 539 (1–2), 341–348.

https://doi.org/10.1016/j.aca.2005.03.019.

(66) The Key Chemicals in Red Wine – Colour, Flavour, and Potential Health

Benefits.

(67) Villamor, R. R.; Ross, C. F. Wine Matrix Compounds Affect Perception of Wine

Aromas. Annu. Rev. Food Sci. Technol. 2013, 4 (1), 1–20.

https://doi.org/10.1146/annurev-food-030212-182707.

(68) Cejudo-Bastante, M. J.; Castro-Vázquez, L.; Hermosín-Gutiérrez, I.; Pérez-

Coello, M. S. Combined Effects of Prefermentative Skin Maceration and

Oxygen Addition of Must on Color-Related Phenolics, Volatile Composition,

and Sensory Characteristics of Airén White Wine. J. Agric. Food Chem. 2011,

59 (22), 12171–12182. https://doi.org/10.1021/jf202679y.

(69) Waters, E. J.; Shirley, N. J.; Williams, P. J. Nuisance Proteins of Wine Are

Grape Pathogenesis-Related Proteins. J. Agric. Food Chem. 1996, 44 (1), 3–

5. https://doi.org/10.1021/jf9505584.

(70) Bayly, F. C.; Berg, H. W. Grape and Wine Proteins of White Wine Varietals.

Am. J. Enol. Vitic. 1967, 18 (9), 18–32.

118

(71) Pennsylvania Wine School: What is a Traminette?

(72) Dombrosky, J.; Gajanan, S. Pennsylvania Wine Industry - An Assessment;

2013.

(73) Skinkis, P. A.; Bordelon, B. P.; Wood, K. V. Comparison of Monoterpene

Constituents in Traminette, Gewürztraminer, and Riesling Winegrapes. Am. J.

Enol. Vitic. 2008, 59 (4), 440–445.

(74) Vernarelli, L. A. Novel Vinification Techniques to Improve Pennsylvania Wine

Quality, The Pennsylvania State University, 2018.

(75) Johnsen, L. G.; Skou, P. B.; Khakimov, B.; Bro, R. Gas Chromatography –

Mass Spectrometry Data Processing Made Easy. J. Chromatogr. A 2017,

1503, 57–64. https://doi.org/10.1016/j.chroma.2017.04.052.

(76) Harmon, R. J.; Schanbacher, F. L.; Ferguson, L. C.; Smith, K. L.

Concentration of Lactoferrin in Milk of Normal Lactating Cows and Changes

Occurring during Mastitis. Am. J. Vet. Res. 1975, 36 (7), 1001—1007.

(77) Fiocchi, A.; Restani, P.; Riva, E.; Qualizza, R.; Bruni, P.; Restelli, A. R.; Galli,

C. L. Meat Allergy: I–Specific IgE to BSA and OSA in Atopic, Beef Sensitive

Children. J. Am. Coll. Nutr. 1995, 14 (3), 239–244.

https://doi.org/10.1080/07315724.1995.10718502.

(78) Lieske, B.; Jantz, A.; Finke, B. An Improved Analytical Approach for the

Determination of Bovine Serum Albumin in Milk. Lait 2005, 85 (3), 237–248.

https://doi.org/10.1051/lait:2005018.

(79) Huang, Z.; Wang, B.; Eaves, D. H.; Shikany, J. M.; Pace, R. D. Phenolic

Compound Profile of Selected Vegetables Frequently Consumed by African

119

Americans in the Southeast United States. Food Chem. 2007, 103 (4), 1395–

1402. https://doi.org/10.1016/j.foodchem.2006.10.077.

(80) Mattila, P.; Hellström, J. Phenolic Acids in Potatoes, Vegetables, and Some of

Their Products. J. Food Compos. Anal. 2007, 20 (3–4), 152–160.

https://doi.org/10.1016/j.jfca.2006.05.007.

(81) Rothwell, J. A.; Perez-Jimenez, J.; Neveu, V.; Medina-Remón, A.; M’Hiri, N.;

García-Lobato, P.; Manach, C.; Knox, C.; Eisner, R.; Wishart, D. S.; et al.

Phenol-Explorer 3.0: A Major Update of the Phenol-Explorer Database to

Incorporate Data on the Effects of Food Processing on Polyphenol Content.

Database 2013, 2013, 1–9. https://doi.org/10.1093/database/bat070.

(82) Ishii, R.; O’Mahony, M. Use of Multiple Standards to Define Sensory

Characteristics for Descriptive Analysis: Aspects of Concept Formation. J.

Food Sci. 1991, 56 (3), 838–842. https://doi.org/10.1111/j.1365-

2621.1991.tb05395.x.

(83) RStudio Team. RStudio: Integrated Development Environment for R. RStudio,

Inc: Boston, MA 2016.

(84) Husson, F.; Le, S.; Cadoret, M. SensoMineR: Sensory Data Analysis. 2017.

(85) Wickham, H. Ggplot2: Elegant Graphics for Data Analysis; Springer-Verlag

New York, 2009.

(86) Wickham, H. The Split-Apply-Combine Strategy for Data Analysis. J. Stat.

Softw. 2015, 40 (1). https://doi.org/10.18637/jss.v040.i01.

(87) Wickham, H.; Francois, R.; Henry, L.; Müller, K. Dplyr: A Grammar of Data

Manipulation. 2017.

120

(88) Augie, B. GridExtra: Miscellaneous Functions for “Grid” Graphics. 2017.

(89) de Mendiburu, F. Agricolae: Statistical Procedures for Agricultural Research.

2017.

(90) Wickham, H. Forcats: Tools for Working with Categorical Variables (Factors).

2019.

(91) Mevik, B.-H.; Wehrens, R.; Liland, K. H. Pls: Partial Least Squares and

Principal Component Regression. 2016.

(92) Lawless, H. T.; Heymann, H. Scaling. In Sensory Evaluation of Food; 2010; pp

149–177.

(93) Guth, H. Quantitation and Sensory Studies of Character Impact Odorants of

Different White Wine Varieties. J. Agric. Food Chem. 1997, 45 (8), 3027–3032.

https://doi.org/10.1021/jf970280a.

(94) Johnson, A. J.; Hopfer, H.; Heymann, H.; Ebeler, S. E. Aroma Perception and

Chemistry of Bitters in Whiskey Matrices: Modeling the Old-Fashioned.

Chemosens. Percept. 2017, 10 (4), 135–148. https://doi.org/10.1007/s12078-

017-9229-3.

(95) Jackson, R. S. Glossary. In Wine Tasting: A Professional Handbook; Elsevier:

San Francisco, CA, 2017; pp 401–403. https://doi.org/10.1016/b978-0-12-

801813-2.00020-3.

(96) Castura, J. C. TempR: Temporal Sensory Data Analysis. 2017.

(97) Dinnella, C.; Masi, C.; Naes, T.; Monteleone, E. A New Approach in TDS Data

Analysis: A Case Study on Sweetened Coffee. Food Qual. Prefer. 2013, 30

(1), 33–46. https://doi.org/10.1016/j.foodqual.2013.04.006.

121

(98) Lepage, M.; Neville, T.; Rytz, A.; Schlich, P.; Martin, N.; Pineau, N. Panel

Performance for Temporal Dominance of Sensations. Food Qual. Prefer.

2014, 38, 24–29. https://doi.org/10.1016/j.foodqual.2014.05.002.

122

Appendix: Chapter 3 Supplementary Tables

Appendix Table 1: ANOVA for citrus flavor showing F-values for the experimental factors ethanol (E), caffeic acid (C), and glycerol (G), and their interactions. Bolded F- values are significant (p < 0.05). Factor Ethanol (E) Caffeic Acid (C) Glycerol (G) E:C E:G C:G E:C:G Residuals (MSE) DF 2 1 1 2 2 1 2 420 0-4 Sec 3.97 0.01 0.01 0.48 1.70 2.83 2.03 0.30 5-9 Sec 4.56 0.11 2.13 2.13 2.35 5.91 0.37 3.17 10-14 Sec 4.61 0.44 2.90 1.87 1.26 1.55 0.49 5.37 15-19 Sec 4.43 1.19 4.86 3.11 0.72 0.27 0.51 5.46 20- 24 Sec 4.63 0.27 2.94 5.15 0.54 0.27 0.54 5.42 25-29 Sec 4.08 0.91 1.55 2.89 0.37 0.01 0.32 5.37 30-34 Sec 5.37 0.54 0.48 0.86 0.23 0.03 0.01 5.60 35-39 Sec 4.20 0.14 0.05 0.66 0.04 0.01 0.03 5.82 40-44 Sec 3.03 0.43 0.07 0.51 0.06 0.48 0.69 5.92 45-49 Sec 2.56 0.30 0.80 0.33 0.02 0.30 0.39 6.09 50-54 Sec 1.78 0.47 0.70 0.18 0.26 0.37 0.26 6.09 55-59 Sec 0.60 0.16 0.64 0.24 1.03 0.36 0.15 5.78 60-64 Sec 0.66 0.04 1.15 0.10 2.50 0.02 0.07 5.45 65-69 Sec 0.43 0.07 0.09 0.08 2.42 0.03 0.02 5.08 70-74 Sec 0.29 0.02 0.18 0.18 1.17 0.00 0.04 4.69 75-79 Sec 0.40 0.03 0.17 0.52 1.43 0.02 0.22 4.30 80-84 Sec 0.61 0.01 0.08 0.29 1.11 0.12 0.33 3.94 85-89 Sec 0.44 0.03 0.11 0.20 0.62 0.47 0.42 3.62 90-94 Sec 0.50 0.20 0.25 0.25 0.57 0.43 0.35 3.36 95-99 Sec 0.85 0.00 0.44 0.00 0.35 0.20 0.29 3.00 100-104 Sec 0.76 0.00 0.45 0.00 0.06 0.14 0.22 2.75 105-109 Sec 1.81 0.17 1.53 0.09 0.06 0.03 0.15 2.30 110-114 Sec 2.81 0.37 0.89 0.56 0.43 0.22 0.04 2.03 115-119 Sec 3.58 0.36 0.55 0.76 0.07 0.45 0.08 1.86 120-124 Sec 4.30 0.41 0.41 0.28 0.01 1.00 0.21 1.82 125-129 Sec 4.37 0.50 0.61 0.04 0.04 0.73 0.31 1.68 130-134 Sec 4.16 0.21 0.48 0.19 0.06 0.48 0.14 1.56 135-139 Sec 3.68 0.03 1.72 0.09 0.05 0.03 0.19 1.38 140-144 Sec 3.57 0.01 2.28 0.09 0.04 0.03 0.32 1.31 145-149 Sec 3.01 0.01 2.94 0.33 0.06 0.20 0.76 1.14 150-154 Sec 3.20 0.19 1.99 0.41 0.22 0.19 1.54 0.98 155-159 Sec 2.24 0.07 1.25 0.44 0.53 0.00 2.82 0.82 160-164 Sec 2.99 0.28 1.24 0.28 0.31 0.28 2.55 0.67 165-169 Sec 2.27 0.57 1.28 0.33 0.57 0.06 1.72 0.59 170-174 Sec 1.63 0.41 1.63 0.41 0.41 0.00 1.22 0.57 175-180 Sec 1.35 0.25 2.22 0.58 0.25 0.03 1.02 0.76

123

Appendix Table 2: ANOVA for the pear flavor attribute showing the F-values for the experimental factors ethanol (E), caffeic acid (C), and glycerol (G), and their interactions. Significant F-values are shown in bold font (p < 0.05). Factor Ethanol (E) Caffeic Acid (C) Glycerol (G) E:C E:G C:G E:C:G Residuals (MSE) DF 2 1 1 2 2 1 2 420 0-4 Sec 0.42 0.09 3.38 0.24 0.20 0.04 0.35 0.89 5-9 Sec 0.89 1.08 3.01 1.09 0.03 0.33 1.35 4.33 10-14 Sec 0.62 0.45 4.51 2.67 0.66 0.15 1.07 5.66 15-19 Sec 0.35 1.47 8.03 2.63 0.65 0.20 0.92 5.65 20- 24 Sec 1.26 0.56 14.30 1.03 0.60 0.34 1.84 5.36 25-29 Sec 1.20 0.34 13.47 0.43 0.52 0.01 1.42 5.32 30-34 Sec 1.59 0.03 8.90 1.07 0.33 0.00 1.53 5.47 35-39 Sec 1.80 0.16 4.58 2.01 0.49 0.00 2.22 5.67 40-44 Sec 1.12 0.38 2.96 1.89 0.38 0.05 2.27 5.91 45-49 Sec 0.51 1.02 2.82 1.35 1.17 0.72 1.36 5.93 50-54 Sec 0.75 0.64 2.50 0.94 0.94 1.42 0.71 6.07 55-59 Sec 1.45 0.10 2.08 0.13 1.61 1.11 0.73 6.10 60-64 Sec 2.02 0.20 2.13 0.08 1.46 0.99 0.82 6.11 65-69 Sec 1.00 0.01 2.22 0.00 0.72 0.12 0.83 6.04 70-74 Sec 1.10 0.07 2.69 0.04 0.54 0.06 0.51 6.07 75-79 Sec 0.71 0.54 2.22 0.19 0.96 0.25 0.42 5.87 80-84 Sec 0.45 0.23 1.66 0.09 1.11 1.01 0.25 5.73 85-89 Sec 0.61 0.32 1.22 0.19 0.80 2.19 0.23 5.34 90-94 Sec 0.83 0.76 1.47 0.09 0.60 1.09 0.46 5.12 95-99 Sec 1.02 0.58 1.76 0.16 0.44 0.34 0.39 4.90 100-104 Sec 2.18 0.05 2.00 0.48 0.21 0.44 0.11 4.74 105-109 Sec 1.80 0.12 1.68 0.43 0.04 0.63 0.03 4.49 110-114 Sec 1.69 0.59 2.03 0.24 0.13 1.01 0.03 4.25 115-119 Sec 1.36 1.09 2.02 0.16 0.46 0.14 0.05 4.13 120-124 Sec 1.29 1.09 1.77 0.14 0.46 0.02 0.16 4.10 125-129 Sec 1.69 1.51 1.76 0.08 0.72 0.09 0.24 3.84 130-134 Sec 1.19 1.74 1.48 0.12 0.64 0.02 0.07 3.60 135-139 Sec 0.98 2.24 0.96 0.10 0.43 0.13 0.13 3.48 140-144 Sec 1.15 2.06 0.41 0.19 0.46 0.34 0.15 3.28 145-149 Sec 0.61 1.21 0.37 0.30 0.51 0.68 0.17 3.06 150-154 Sec 0.37 1.14 0.11 0.11 0.35 0.46 0.12 2.93 155-159 Sec 0.32 1.09 0.13 0.06 0.27 0.42 0.06 2.92 160-164 Sec 0.18 0.89 0.43 0.02 0.40 0.51 0.04 2.82 165-169 Sec 0.12 0.82 0.38 0.00 0.63 0.14 0.07 2.71 170-174 Sec 0.12 0.57 0.33 0.01 0.50 0.20 0.11 2.55 175-180 Sec 0.21 0.37 0.37 0.02 0.44 0.37 0.16 3.63

124

Appendix Table 3: ANOVA for the apple flavor attribute showing the F-values for the experimental factors ethanol (E), caffeic acid (C), and glycerol (G), and their interactions. Significant F-values are shown in bold font (p < 0.05). Factor Ethanol (E) Caffeic Acid (C) Glycerol (G) E:C E:G C:G E:C:G Residuals (MSE) DF 2 1 1 2 2 1 2 420 0-4 Sec 5.77 2.40 0.27 1.51 0.33 0.00 0.76 0.56 5-9 Sec 4.92 1.64 1.76 2.15 0.24 0.02 0.28 4.11 10-14 Sec 3.04 0.67 1.04 0.57 0.16 2.04 0.14 5.57 15-19 Sec 2.85 0.85 0.41 0.43 0.55 0.41 0.75 5.74 20- 24 Sec 1.62 0.01 0.65 0.12 0.44 0.01 0.29 5.74 25-29 Sec 0.84 0.10 2.44 0.12 0.78 0.03 0.33 5.77 30-34 Sec 0.33 0.08 3.88 0.28 1.17 0.04 0.61 5.74 35-39 Sec 0.28 0.05 2.32 0.15 0.60 0.01 0.30 6.07 40-44 Sec 0.15 0.29 1.01 0.01 0.62 0.01 0.19 6.19 45-49 Sec 0.39 0.93 0.14 0.08 0.77 0.01 0.04 5.98 50-54 Sec 0.48 0.61 0.00 0.37 0.97 0.01 0.26 6.06 55-59 Sec 0.55 1.56 0.21 0.43 1.73 0.14 0.39 5.88 60-64 Sec 0.01 0.74 0.39 0.12 0.89 0.21 0.20 5.75 65-69 Sec 0.14 0.00 0.54 0.07 0.35 0.43 0.36 5.54 70-74 Sec 0.64 0.00 1.56 0.16 0.36 0.21 0.40 5.33 75-79 Sec 1.36 0.18 0.99 0.65 0.41 0.11 0.19 4.76 80-84 Sec 0.72 0.27 0.64 0.31 0.36 0.01 0.05 4.46 85-89 Sec 0.09 0.08 0.57 0.81 0.57 0.18 0.17 4.18 90-94 Sec 0.04 0.20 0.79 0.96 0.20 0.20 0.03 3.81 95-99 Sec 0.06 0.17 1.16 0.42 0.04 0.04 0.37 3.52 100-104 Sec 0.08 0.47 1.35 0.13 0.13 0.40 0.42 3.32 105-109 Sec 0.13 0.57 1.96 0.09 0.37 0.57 0.40 3.20 110-114 Sec 0.21 0.01 2.03 0.19 0.16 0.28 0.56 2.97 115-119 Sec 0.24 0.00 0.64 0.24 0.02 0.21 0.55 2.85 120-124 Sec 0.08 0.00 0.59 0.20 0.08 0.17 0.35 2.66 125-129 Sec 0.15 0.08 1.27 0.53 0.04 0.16 0.90 2.50 130-134 Sec 0.08 0.06 1.40 0.35 0.06 0.31 1.13 2.38 135-139 Sec 0.03 0.02 1.03 0.30 0.05 0.15 1.04 2.30 140-144 Sec 0.08 0.34 1.51 0.23 0.15 0.10 1.13 2.21 145-149 Sec 0.03 0.27 1.37 0.22 0.18 0.04 0.88 2.19 150-154 Sec 0.07 0.07 0.86 0.18 0.31 0.00 0.78 2.10 155-159 Sec 0.08 0.03 0.69 0.19 0.36 0.03 0.69 2.09 160-164 Sec 0.08 0.03 0.69 0.19 0.36 0.03 0.69 2.09 165-169 Sec 0.07 0.02 0.64 0.20 0.32 0.04 0.73 2.08 170-174 Sec 0.04 0.00 0.80 0.31 0.21 0.08 1.05 1.96 175-180 Sec 0.06 0.00 1.10 0.19 0.19 0.05 0.84 2.72

125

Appendix Table 4: ANOVA for the astringent mouthfeel attribute showing the F-values for the experimental factors ethanol (E), caffeic acid (C), and glycerol (G), and their interactions. Significant F-values are shown in bold font (p < 0.05). Factor Ethanol (E) Caffeic Acid (C) Glycerol (G) E:C E:G C:G E:C:G Residuals (MSE) DF 2 1 1 2 2 1 2 420 0-4 Sec 4.76 0.43 4.97 0.02 1.67 0.02 0.22 0.13 5-9 Sec 1.61 0.81 2.07 0.70 0.18 0.18 0.62 2.58 10-14 Sec 1.33 1.11 0.94 1.25 0.09 0.39 0.52 5.00 15-19 Sec 1.31 0.04 3.67 0.76 0.18 0.23 0.36 5.81 20- 24 Sec 0.64 0.05 5.95 0.01 0.55 0.09 0.19 5.89 25-29 Sec 1.86 0.02 4.71 0.08 0.52 0.15 0.10 5.62 30-34 Sec 0.89 0.00 3.43 0.11 0.36 0.03 0.39 5.71 35-39 Sec 1.40 0.01 7.22 0.02 0.08 0.00 1.07 5.75 40-44 Sec 1.81 0.01 4.29 0.11 0.79 0.08 1.73 5.84 45-49 Sec 1.05 0.02 2.19 0.35 0.59 0.02 1.71 5.95 50-54 Sec 1.17 0.20 3.00 0.08 0.03 0.01 1.24 6.12 55-59 Sec 1.12 0.39 2.54 0.01 0.07 0.60 0.64 6.14 60-64 Sec 0.45 0.41 2.10 0.01 0.07 0.41 0.08 6.20 65-69 Sec 0.34 0.00 1.57 0.21 0.02 0.23 0.00 6.23 70-74 Sec 0.65 0.03 0.46 0.11 0.09 0.17 0.00 6.19 75-79 Sec 0.74 0.18 0.01 0.01 0.13 0.01 0.12 6.16 80-84 Sec 0.37 0.37 0.32 0.00 0.17 0.03 0.29 6.09 85-89 Sec 0.38 0.50 0.82 0.01 0.23 0.35 0.23 5.97 90-94 Sec 0.06 0.45 0.63 0.04 0.21 0.16 0.16 5.88 95-99 Sec 0.12 0.41 0.58 0.07 0.12 0.52 0.13 5.72 100-104 Sec 0.27 0.24 0.24 0.64 0.13 0.13 0.22 5.65 105-109 Sec 0.44 0.01 0.25 0.95 0.34 0.25 0.53 5.43 110-114 Sec 0.27 0.02 0.32 0.99 0.34 0.16 0.52 5.30 115-119 Sec 0.41 0.04 0.37 0.25 0.45 0.05 0.62 5.30 120-124 Sec 0.35 0.06 0.51 0.21 0.69 0.11 1.00 5.20 125-129 Sec 0.25 0.13 0.16 0.16 1.03 0.16 1.03 5.09 130-134 Sec 0.25 0.55 0.37 0.17 0.94 0.49 1.04 4.86 135-139 Sec 0.54 1.25 0.29 0.19 1.47 0.58 1.12 4.63 140-144 Sec 0.26 1.48 0.03 0.46 1.78 0.58 0.80 4.38 145-149 Sec 0.05 1.89 0.01 0.34 0.38 0.10 1.08 3.98 150-154 Sec 0.03 1.27 0.26 0.08 0.33 0.09 1.35 3.54 155-159 Sec 0.05 0.89 0.18 0.27 0.52 0.13 1.26 3.37 160-164 Sec 0.14 0.86 0.12 0.09 0.38 0.02 0.98 3.29 165-169 Sec 0.37 0.56 0.02 0.02 0.73 0.17 0.57 3.03 170-174 Sec 0.30 0.14 0.10 0.02 0.66 0.37 0.60 2.79 175-180 Sec 0.17 0.22 0.38 0.08 0.29 0.51 0.39 3.83

126

Appendix Table 5: ANOVA for the warm/hot mouthfeel attribute showing the F-values for the experimental factors ethanol (E), caffeic acid (C), and glycerol (G), and their interactions. Significant F-values are shown in bold font (p < 0.05). Factor Ethanol (E) Caffeic Acid (C) Glycerol (G) E:C E:G C:G E:C:G Residuals (MSE) DF 2 1 1 2 2 1 2 420 0-4 Sec 5.24 0.06 2.52 0.06 1.47 3.69 1.10 0.33 5-9 Sec 8.15 0.29 1.40 0.77 1.47 0.97 0.48 3.81 10-14 Sec 14.02 0.04 0.33 1.34 0.55 0.67 0.01 5.52 15-19 Sec 13.39 0.45 0.21 0.63 0.12 0.04 0.69 4.89 20- 24 Sec 12.04 1.48 0.45 0.13 0.48 0.67 1.26 3.75 25-29 Sec 16.00 1.19 0.21 0.02 0.45 0.95 0.53 2.82 30-34 Sec 19.07 2.40 0.02 0.14 0.10 0.07 0.90 2.51 35-39 Sec 18.34 1.34 0.13 0.07 0.54 1.21 0.69 2.49 40-44 Sec 19.19 0.29 1.76 0.28 0.99 0.36 0.58 2.54 45-49 Sec 22.33 0.32 2.34 0.05 0.87 0.46 0.87 2.89 50-54 Sec 19.65 0.02 2.95 0.10 0.86 0.71 0.30 3.11 55-59 Sec 19.24 0.13 2.82 0.05 1.19 0.40 0.16 3.36 60-64 Sec 19.02 0.25 2.54 0.15 0.72 0.16 0.09 3.73 65-69 Sec 18.36 0.71 2.69 0.08 0.50 0.49 0.24 4.22 70-74 Sec 20.25 0.25 3.29 0.08 0.09 0.00 0.46 4.50 75-79 Sec 19.34 0.47 3.04 0.02 0.05 0.04 0.95 4.75 80-84 Sec 17.85 0.43 2.78 0.77 0.01 0.00 0.94 5.19 85-89 Sec 12.77 0.05 1.65 0.34 0.08 0.22 1.18 5.56 90-94 Sec 9.63 0.07 0.61 0.07 0.13 0.14 0.35 5.78 95-99 Sec 8.75 0.15 0.00 0.03 0.04 0.13 0.47 5.99 100-104 Sec 9.87 0.20 0.05 0.14 0.04 0.59 0.28 5.98 105-109 Sec 8.63 0.14 0.05 0.45 0.38 1.65 0.27 5.91 110-114 Sec 4.90 0.08 0.00 0.81 0.50 0.77 0.48 5.84 115-119 Sec 4.31 0.00 0.01 1.19 0.16 0.22 0.82 5.64 120-124 Sec 2.50 0.14 0.00 1.07 0.20 0.08 1.05 5.47 125-129 Sec 1.93 0.02 0.33 0.44 0.04 0.10 0.79 5.15 130-134 Sec 2.18 0.00 0.84 0.01 0.25 0.28 0.98 4.85 135-139 Sec 2.38 0.03 1.34 0.05 0.25 0.64 0.95 4.67 140-144 Sec 1.95 0.19 0.89 0.06 0.32 0.58 1.30 4.35 145-149 Sec 2.39 0.23 0.04 0.01 0.43 0.73 0.56 4.09 150-154 Sec 2.93 0.12 0.00 0.28 0.68 0.86 0.30 3.88 155-159 Sec 3.18 0.02 0.11 0.22 0.72 0.54 0.12 3.61 160-164 Sec 4.60 0.38 0.06 0.15 0.50 0.61 0.52 3.19 165-169 Sec 5.51 0.67 0.00 0.19 0.19 0.77 0.56 2.70 170-174 Sec 4.32 0.30 0.01 0.20 0.07 1.44 0.58 2.21 175-180 Sec 4.42 0.05 0.33 0.03 0.30 1.20 0.51 2.78

127

Appendix Table 6: ANOVA for the bitter taste attribute showing the F-values for the experimental factors ethanol (E), caffeic acid (C), and glycerol (G), and their interactions. Significant F-values are shown in bold font (p < 0.05). Factor Ethanol (E) Caffeic Acid (C) Glycerol (G) E:C E:G C:G E:C:G Residuals (MSE) DF 2 1 1 2 2 1 2 420 0-4 Sec 2.66 0.03 4.90 0.66 1.14 0.72 0.35 0.32 5-9 Sec 7.80 0.01 10.41 0.44 0.41 0.93 0.61 2.89 10-14 Sec 6.57 0.08 11.93 0.15 0.33 0.02 0.78 4.78 15-19 Sec 9.03 0.10 14.68 0.69 0.69 0.24 0.58 5.17 20- 24 Sec 9.34 0.04 6.10 0.26 0.34 0.01 0.64 5.46 25-29 Sec 10.21 0.14 4.29 0.20 0.26 0.56 0.86 5.40 30-34 Sec 11.12 0.01 7.07 0.75 0.59 1.27 0.48 5.12 35-39 Sec 11.99 0.32 6.97 1.00 1.46 1.57 0.40 4.95 40-44 Sec 13.84 0.36 3.08 1.90 2.09 2.25 0.32 5.05 45-49 Sec 10.11 0.05 1.76 1.20 1.02 1.44 0.98 5.22 50-54 Sec 6.32 0.80 2.69 1.11 1.22 0.95 1.08 5.37 55-59 Sec 7.71 0.22 4.84 0.76 0.78 1.47 1.91 5.48 60-64 Sec 9.63 0.18 4.71 0.68 0.40 1.07 0.82 5.63 65-69 Sec 7.75 0.44 3.81 0.44 0.43 0.00 1.49 5.72 70-74 Sec 7.16 2.05 3.34 0.81 0.34 0.16 0.84 5.87 75-79 Sec 6.29 2.12 4.22 1.11 0.58 0.00 0.42 5.81 80-84 Sec 7.17 1.89 5.12 0.32 0.71 0.37 0.54 5.67 85-89 Sec 5.36 1.28 4.86 0.78 0.36 1.10 0.97 5.46 90-94 Sec 5.60 1.83 1.61 0.58 0.37 0.22 1.14 5.18 95-99 Sec 5.22 0.95 1.23 0.40 1.09 0.13 1.39 4.72 100-104 Sec 2.21 1.84 3.18 0.67 2.14 0.69 1.33 4.09 105-109 Sec 1.14 0.25 4.50 0.23 1.40 1.93 0.14 3.37 110-114 Sec 0.75 0.30 2.33 0.15 1.55 1.13 1.05 2.79 115-119 Sec 0.87 0.62 0.84 0.24 2.27 0.07 1.18 2.15 120-124 Sec 1.45 1.33 0.21 0.11 2.45 0.05 0.99 1.57 125-129 Sec 2.67 1.18 0.02 0.25 3.38 0.09 0.53 1.23 130-134 Sec 2.12 0.33 0.91 0.06 5.50 0.23 0.25 1.01 135-139 Sec 1.59 0.88 1.34 0.40 5.80 0.15 0.55 0.76 140-144 Sec 0.89 1.18 1.18 0.31 4.24 0.01 0.89 0.63 145-149 Sec 1.08 1.15 1.15 0.32 4.62 0.25 0.16 0.45 150-154 Sec 1.10 1.47 1.91 1.26 3.51 2.99 0.50 0.31 155-159 Sec 0.83 1.86 1.86 2.49 2.49 5.18 0.83 0.28 160-164 Sec 1.07 1.50 1.50 2.84 2.84 5.03 1.07 0.22 165-169 Sec 1.14 0.14 0.38 1.96 2.66 2.57 0.74 0.15 170-174 Sec 0.50 2.00 0.00 0.50 1.50 0.00 1.50 0.12 175-180 Sec 0.50 2.00 0.00 0.50 1.50 0.00 1.50 0.17

128

Appendix Table 7: ANOVA for the sour taste attribute showing the F-values for the experimental factors ethanol (E), caffeic acid (C), and glycerol (G), and their interactions. Significant F-values are shown in bold font (p < 0.05). Factor Ethanol (E) Caffeic Acid (C) Glycerol (G) E:C E:G C:G E:C:G Residuals (MSE) DF 2 1 1 2 2 1 2 420 0-4 Sec 1.71 0.47 0.00 0.31 2.02 0.35 0.09 1.10 5-9 Sec 2.56 2.17 0.21 0.11 0.74 1.57 0.15 4.78 10-14 Sec 3.49 0.98 0.98 0.09 0.05 0.03 0.07 5.70 15-19 Sec 1.28 2.36 0.74 0.29 1.00 0.03 0.44 4.54 20- 24 Sec 0.39 1.31 0.40 0.64 1.02 1.08 0.12 3.59 25-29 Sec 0.17 1.43 0.01 0.41 0.59 1.10 0.02 3.72 30-34 Sec 0.23 0.06 0.06 0.36 0.60 0.96 0.10 4.89 35-39 Sec 0.52 0.24 1.13 0.74 0.54 0.28 0.17 5.53 40-44 Sec 0.41 0.90 0.62 0.70 0.32 0.02 0.45 5.95 45-49 Sec 0.21 1.19 0.01 0.96 0.18 0.34 0.77 6.09 50-54 Sec 0.44 0.63 0.24 1.25 0.74 0.02 0.89 6.14 55-59 Sec 0.47 0.34 0.00 1.38 0.36 0.08 0.22 6.04 60-64 Sec 0.70 0.03 0.01 1.77 0.26 0.09 0.37 5.73 65-69 Sec 0.80 0.15 0.12 0.83 0.03 0.01 0.67 5.42 70-74 Sec 0.45 0.00 0.03 0.47 0.03 0.41 0.41 5.06 75-79 Sec 0.07 0.04 0.08 1.04 0.11 0.00 0.14 4.61 80-84 Sec 0.14 0.13 0.13 1.45 0.05 0.00 0.04 4.16 85-89 Sec 0.25 0.25 0.25 0.45 0.58 0.64 0.06 3.72 90-94 Sec 0.77 0.34 0.47 0.21 0.75 0.47 0.21 3.30 95-99 Sec 0.86 0.05 0.11 0.24 0.67 0.11 0.42 3.03 100-104 Sec 1.28 0.08 0.01 0.39 0.42 0.12 0.54 2.85 105-109 Sec 1.97 0.00 0.16 0.12 0.09 0.00 0.06 2.42 110-114 Sec 3.48 0.02 0.07 0.88 0.82 0.00 0.28 2.14 115-119 Sec 2.23 0.45 0.10 0.74 1.25 0.21 0.32 1.87 120-124 Sec 1.22 0.23 0.03 0.74 0.92 0.30 0.30 1.73 125-129 Sec 0.96 0.00 0.14 1.38 0.46 0.57 0.46 1.63 130-134 Sec 1.06 0.05 0.21 0.97 0.46 0.48 0.46 1.55 135-139 Sec 0.93 0.23 0.01 1.08 0.32 0.78 0.27 1.44 140-144 Sec 0.26 0.06 0.03 1.58 0.26 0.34 0.41 1.35 145-149 Sec 0.05 0.05 0.05 1.71 0.05 0.42 0.42 1.25 150-154 Sec 0.04 0.07 0.52 2.29 0.04 0.40 0.53 1.13 155-159 Sec 0.05 0.43 1.07 2.25 0.04 0.14 0.57 1.05 160-164 Sec 0.12 0.77 0.77 1.59 0.09 0.15 0.90 0.97 165-169 Sec 0.12 1.27 0.49 0.68 0.38 0.00 1.23 0.80 170-174 Sec 0.07 0.80 0.61 0.28 0.28 0.05 1.12 0.74 175-180 Sec 0.08 0.70 0.70 0.23 0.23 0.08 1.01 1.07

129

Appendix Table 8: ANOVA for the sweet taste attribute showing the F-values for the experimental factors ethanol (E), caffeic acid (C), and glycerol (G), and their interactions. Significant F-values are shown in bold font (p < 0.05). Factor Ethanol (E) Caffeic Acid (C) Glycerol (G) E:C E:G C:G E:C:G Residuals (MSE) DF 2 1 1 2 2 1 2 420 0-4 Sec 1.21 0.53 6.18 0.69 0.06 1.54 0.46 0.86 5-9 Sec 3.85 1.13 17.75 0.02 0.62 0.28 0.50 4.71 10-14 Sec 5.63 0.53 15.92 0.09 0.72 0.04 0.17 5.70 15-19 Sec 6.26 1.17 22.41 0.49 0.47 0.24 0.22 5.14 20- 24 Sec 3.49 0.05 18.30 0.25 0.09 0.51 0.05 5.53 25-29 Sec 4.30 0.28 12.32 1.26 0.56 0.04 0.40 5.56 30-34 Sec 2.35 0.09 9.43 0.82 0.13 0.34 0.46 5.75 35-39 Sec 1.04 0.04 8.46 0.14 0.41 0.64 0.30 5.83 40-44 Sec 1.11 0.04 8.85 0.37 0.87 0.78 0.34 5.73 45-49 Sec 1.30 0.02 11.81 0.67 0.97 0.67 0.80 5.53 50-54 Sec 0.96 0.00 12.67 0.36 0.96 1.51 1.93 5.16 55-59 Sec 0.94 0.02 10.97 0.33 0.69 2.14 1.90 5.00 60-64 Sec 1.11 0.04 10.11 0.48 0.36 2.78 1.67 4.68 65-69 Sec 0.85 0.00 9.29 0.28 0.21 1.09 1.40 4.47 70-74 Sec 0.63 0.03 9.09 0.45 0.18 0.92 0.83 4.44 75-79 Sec 0.42 0.05 8.22 0.34 0.10 0.77 0.57 4.33 80-84 Sec 0.50 0.01 5.39 0.28 0.10 0.42 0.50 4.04 85-89 Sec 0.38 0.00 4.56 0.35 0.79 0.00 0.35 3.84 90-94 Sec 0.26 0.00 7.17 0.18 1.32 0.23 0.29 3.56 95-99 Sec 0.46 0.28 5.71 0.09 1.49 0.05 0.17 3.28 100-104 Sec 0.26 0.23 5.00 0.21 1.00 0.10 0.21 3.27 105-109 Sec 0.27 0.04 4.40 0.10 0.59 0.17 0.24 3.12 110-114 Sec 0.18 0.11 3.02 0.31 0.79 0.38 0.20 2.95 115-119 Sec 0.59 0.01 2.73 0.30 0.64 0.03 0.32 2.65 120-124 Sec 0.70 0.29 1.29 0.35 0.57 0.13 0.02 2.59 125-129 Sec 0.39 0.52 1.18 0.34 0.39 0.18 0.05 2.55 130-134 Sec 0.12 0.71 1.34 0.14 0.33 0.28 0.15 2.36 135-139 Sec 0.20 0.15 1.05 0.04 0.20 0.15 0.04 2.25 140-144 Sec 0.11 0.00 0.52 0.10 0.11 0.00 0.10 2.14 145-149 Sec 0.05 0.03 0.25 0.16 0.05 0.03 0.16 2.09 150-154 Sec 0.00 0.03 0.25 0.11 0.00 0.03 0.11 2.11 155-159 Sec 0.01 0.00 0.09 0.06 0.01 0.00 0.06 2.03 160-164 Sec 0.03 0.03 0.03 0.03 0.03 0.03 0.03 2.01 165-169 Sec 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.96 170-174 Sec 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.96 175-180 Sec 0.00 0.00 0.00 0.00 0.00 0.00 0.00 2.83

130

Appendix Table 9: ANOVA for the “other” attribute showing the F-values for the experimental factors ethanol (E), caffeic acid (C), and glycerol (G), and their interactions. Significant F-values are shown in bold font (p < 0.05). Factor Ethanol (E) Caffeic (C) Glycerol (G) E:C E:G C:G E:C:G Residuals (MSE) DF 2 1 1 2 2 1 2 420 0-4 Sec 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.02 5-9 Sec 1.33 0.11 4.29 0.57 1.33 0.11 0.57 0.19 10-14 Sec 1.40 0.01 3.16 1.40 0.26 0.58 0.26 0.32 15-19 Sec 2.72 0.02 2.24 1.45 0.17 0.00 0.82 0.41 20- 24 Sec 0.93 0.24 0.74 1.15 0.46 0.00 0.03 0.62 25-29 Sec 1.63 0.20 1.00 1.63 0.03 1.00 0.03 0.75 30-34 Sec 2.29 0.01 0.96 1.78 0.06 0.96 0.33 0.78 35-39 Sec 2.88 0.03 0.17 1.18 0.04 1.22 0.43 0.69 40-44 Sec 3.07 0.00 0.00 1.02 0.00 1.37 0.34 0.68 45-49 Sec 2.08 0.04 0.19 0.86 0.19 2.08 0.55 0.59 50-54 Sec 1.63 0.00 0.41 1.22 0.41 1.63 0.41 0.57 55-59 Sec 0.87 0.27 0.02 0.82 0.32 1.67 0.62 0.55 60-64 Sec 0.52 0.66 0.07 0.50 0.52 2.23 0.68 0.50 65-69 Sec 0.19 1.50 0.05 0.22 0.48 2.28 0.76 0.45 70-74 Sec 0.41 1.81 0.02 0.02 0.41 2.71 0.58 0.41 75-79 Sec 0.82 3.28 0.06 0.27 0.39 1.79 0.08 0.37 80-84 Sec 1.51 2.59 0.01 1.30 0.44 0.87 0.22 0.32 85-89 Sec 1.42 1.82 0.20 1.82 0.20 1.82 0.61 0.29 90-94 Sec 1.39 3.68 0.62 1.39 0.24 0.62 0.24 0.18 95-99 Sec 1.00 3.00 0.33 1.00 0.33 0.33 0.33 0.17 100-104 Sec 1.00 3.00 0.33 1.00 0.33 0.33 0.33 0.17 105-109 Sec 0.81 1.38 1.38 0.81 0.81 1.38 0.81 0.06 110-114 Sec 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.06 115-119 Sec 1.69 1.69 0.31 1.69 0.31 0.31 0.31 0.07 120-124 Sec 2.00 2.00 0.00 2.00 0.00 0.00 0.00 0.12 125-129 Sec 2.00 2.00 0.00 2.00 0.00 0.00 0.00 0.12 130-134 Sec 2.00 2.00 0.00 2.00 0.00 0.00 0.00 0.12 135-139 Sec 2.00 2.00 0.00 2.00 0.00 0.00 0.00 0.12 140-144 Sec 1.27 1.07 0.27 2.07 0.27 0.07 0.07 0.14 145-149 Sec 1.00 0.62 0.62 0.62 1.76 0.11 0.11 0.18 150-154 Sec 1.00 0.47 0.47 0.47 2.05 0.21 0.21 0.18 155-159 Sec 1.00 0.89 1.33 0.89 1.00 0.01 0.01 0.21 160-164 Sec 0.64 0.82 2.27 0.82 0.64 0.82 0.82 0.03 165-169 Sec 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 170-174 Sec 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 175-180 Sec 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00