I Flavor Chemistry and Gastronomy

Home , Chemistry (Smash), Harold McGee, Wine chemistry

Flavor Chemistry and Gastronomy: Sensory, Instrumental, and Multivariate Approaches

ARIELLE JURCHAK JOHNSON B.S. (New York University) 2009

DISSERTATION

Submitted in partial satisfaction of the requirements for the degree of

DOCTOR OF PHILOSOPHY

Agricultural and Environmental Chemistry

in the

OFFICE OF GRADUATE STUDIES

of the

UNIVERSITY OF CALIFORNIA DAVIS

Approved:

______Susan E. Ebeler, Chair

______Hildegarde Heymann

______Roger B. Boulton

Committee in Charge

2014

i Acknowledgements

I would be remiss not to acknowledge the many people who have made this work possible, more or less chronologically:

My parents, Martha Jurchak and Larry Johnson, for encouraging and pushing my intellectual, scientifc, and creative development from an early age; my brother and sister, Eban and Adriana Johnson; my Grandmother, Marie Jurchak, for planting the seed of bon vivantism and supplying the younger me with plenty of cookbooks to read; to my de-facto in-laws Alice Meadows and Pete, Grace, Jack and Joe Pold for welcoming me into their family. From the early days of what would turn out to be my career, Kent Kirshenbaum, for letting me stand in the back of the room at the frst Experimental Cuisine Collective meeting and giving me the space and support to start developing ideas about how to do scientifc research on culinary questions; and Dave Arnold for taking my half-developed ideas seriously and giving me a lot of challenging questions about smell and chemistry to fgure out. At UC Davis, my advisor, Susan Ebeler, for providing the tools and space to develop as an analytical chemist and learn the hard work of a writing, lecturing scientist, and putting up with my scientifc wanderings; Hildegarde Heymann, for teaching me everything I know about sensory science and experimental design, cultivating my ideas about how science should be dealing with questions of favor, chemistry, and cuisine, and freely ofering profound words of encouragement when I needed them. Carolyn Doyle, for invaluable instrumentation help. My lab and ofce mates, for their support and friendship, Larry Lerno, Helene Hopfer, Ellie King, Maya Hood White and especially Anna Hjelmeland, a stalwart friend both in and outside of work. Roger Boulton, for introducing me to the technology of distilled beverages and being a valuable and enjoyably challenging presence on both my qualifying exam and dissertation committees. Alyson Mitchell and Doug Adams for their presence on my qualifying exam committee and help with academic issues and conference travel. Harold McGee, frst as a role model, then as a QE committee member, and now as a dear friend and colleague.

ii Related to the University of Copenhagen, Line Holler Mielby for being an early scientifc role model and encouraging me to go to Denmark, Michael Bom Frøst, for enabling my coming to the Nordic Food Lab and KU, my sensory panel at KU, and Wender Bredie, Belinda Nielsen, and Charlotte Dandanell of the sensory science group. My collaborators, all of which I thank primarily for giving me interesting and challenging questions (gastronomic and otherwise) to think about-Rene Redzepi for founding the Nordic Food Lab and welcoming me into the noma/mad family; David Chang for inviting me to explain olive phenolics Harvard; Rachel Dutton and Ben Wolfe; Josh Evans, Guillemette Barthouil, and Mark Emil Hermansen with their help making it happen at the Nordic Food Lab; Dan Felder; Rosio Sanchez for giving me interesting things about the psychological, cultural, and hedonic aspects of dessert to think about; Lars Williams, for taking my ideas and expertise seriously, teaching me how to throw a punch, and fundamentally changing the way I think about the production and development of knowledge and ideas.

Paramount among the thanked is Tom Pold, the most encouraging, supportive, intellectually challenging, and wonderful signifcant other I think could possibly exist. Tank you for your patience, your vigilance, your encouragement, your love. I truly could never have done this without you.

iii Contents Abstract 1 Chapter 1: Introduction and Literature Review 3 What is Flavor? 3 Smell and Chemistry 4 Analyzing Flavor 7 Analytical Flavor Chemistry 7 Descriptive Analysis and Sensometrics 11 Flavor and Multivariate Statistics 13 Principal Component Analysis 13 Partial Least Squares Regression 15 Multivariate Statistical Techniques on Non-Continuous or Nontraditional Data 16 Applied Flavor Chemistry: Science, Food, Culture 17 Flavor and chemistry of wine and other beverages 17 Cuisine Research and Development 19 References 27 Chapter 2: Perceptual Characterization and Analysis of Aroma Mixtures using In-Instrument Gas Chromatography Recombination-Olfactometry1 39 Abstract 39 Introduction 40 Materials and Methods 42 Instrument: 42 Sampling and Chromatographic Conditions 43 Sensory Conditions 44 Data Analysis 45 Results and Discussion 45 Conclusions 51 References 52 Acknowledgments 54 Supplementary Information: 55 Chapter 3: Volatile and Sensory Profling of Cocktail Bitters 59 Introduction 59 Materials & Methods 64 Samples 64 Chemical Analysis 64 Sensory Analysis 65 Statistical Analysis 67 Results & Discussion 67 Sensory Analysis 67 Gas Chromatography-Mass Spectrometry 70 Flavor Chemistry of Bitters 77 Conclusions 85 References 87 Chapter 4: GC-Recomposition-Olfactometry (GRO) and multivariate study of three terpenoid compounds in the aroma profle of Angostura bitters 96 iv Introduction 96 Materials and Methods 99 Bitters: 99 Sample Preparation and Extraction: 99 Instrument and Conditions: 100 Sensory Conditions: 101 Statistical Analysis: 101 Results and Discussion: 102 References 110 Chapter 5: Aroma Perception and Chemistry of Bitters in Whiskey Matrices: Modeling the Old- Fashioned Cocktail 113 Introduction 113 Materials and Methods 114 Whiskey: 114 Bitters: 114 Model Old-Fashioned: 114 Sensory Analysis: 116 GC-MS: 117 Analysis of Variance: 118 Means and Signifcant Diferences: 118 Principal Component Analysis (PCA): 118 Partial Least Squares Regression (PLS): 122 Results and Discussion: 122 References 132 Chapter 6: Sensory Attributes and Flavor Chemistry of Acetic Fermentations with Novel Plant Ingredients 136 Introduction 136 Materials and Methods 137 Acetifcation: 137 Juice Vinegars: 138 Tea Vinegars: 138 Wine Vinegars: 139 Sensory Analysis: 139 Volatile Analysis: 140 Chromatographic Conditions: 140 Compound Identifcation and Relative Quantifcation: 142 Capillary Electrophoresis: 142 Statistical Analysis: 142 Results and Discussion 143 Sensory Analysis: 143 Chemical Analysis: 146 PLS 152 References 159 Chapter 7: Correlating Labeled Sorting Sensory Analysis and Volatile Analysis of Malt Vinegars with Novel Ingredients 165 v Introduction: 165 Materials and Methods: 167 Vinegar Production 167 a. Base Beer Fermentation: 168 b. Acetifcation and favoring: 169 Sensory Analysis: 169 GC-MS: 169 Statistical Analysis: 170 a. Sorting: 170 b. Labels: 171 c. Volatiles: 171 d. Multiple Factor Analysis (MFA): 171 Results and Discussion: 171 References 186 Conclusions 191

vi List of Tables and Figures Figure 2.1 Conceptual schematic of the In-Instrument Gas Chromatograph Recombination Olfactometer (GRO) instrument. Volatiles are extracted onto a solid phase (via solid-phase microextraction or SPME) from the headspace of a food, beverage, or other sample, in this case, lavender fowers, and initially they are separated conventionally on an analytical capillary GC column. In-line with the GC column, a pneumatic Deans Switch followed by a cold trap allows the experimenter to build a mixture of these separated volatiles that is held until the cryotrap is rapidly heated, releasing the mixture for a subject to smell at the olfactory port and evaluate. 41 Figure 2.2 Schematic of (a) standard GC-MS; (b) GC-MS with splitter at end of column for olfactometry; and (c) In-instrument Gas chromatograph- Recombination Olfactometer or GRO with Deans switch, splitter, cryogenic trap and olfactory port. Abbreviations: i-inlet; c-column; d-detector; o-oven; olf-olfactometry port; sp-splitter; sw-Deans switch 1; w-waste; cr-cryogenic trap; and cb-switch 2 on control box. 42 Table 2.1 Experimental GC-O conditions and aroma descriptors for mixtures of volatiles from the lavender chromatograms. 43 Figure 2.3 Top aroma descriptors for mixtures of sections of the lavender chromatogram by cut time and chromatogram composition. Abbreviations correspond to Experimental Conditions described in Table 1. As chemical complexity and number of components per mixture approaches the makeup of the whole chromatogram (W) mixture, there is evidence of perceptual additivity as increasing cross-utilization of terms from simpler mixtures, masking as reduced use of dominant terms for simpler (P1-P6) mixtures, and synergistic efects as new complex or composite terms like “fresh lavender” become important. 45 Figure 2.4 Correspondence Analysis of (A) lavender volatile mixtures; and (B) lavender volatile mixture descriptors. Abbreviations for mixtures correspond to those in Table 1. Terms generated by the panelists to describe the perceived odor of from each Experimental Condition described in Table 1 were tabulated by frequency of use and used for the Correspondence Analysis. 30.57% of variance explained by dimension 1 (x), 22.84% of variance explained by dimension 2 (y). 46 Figure 2.5 Te rated representativeness of the aroma of samples W, O1-O3, and P1-P6 as compared by panelists to the aroma of whole fowering lavender. Letters a, b, c refer to the mixture’s Signifcant Diference from each other- if two samples do not share a letter, they are signifcantly diferent. Samples P1, P5, and P6 are signifcantly less representative of the aroma of fowering lavender than sample W, which incorporates all the volatiles in fowering lavender. 48 Supplementary Figure S2.1 Te chromatogram of mixture O2. Compounds eluting between 16 and 25 minutes were vented to waste by the Deans Switch and were consequently excluded from the smelled mixture and not sent to the mass spectrometer. 55 Supplementary Figure S2.2 Te chromatogram of mixture P5. Compounds eluting between 0 and 25 minutes and 32 and 40 minutes were vented to waste by the Deans Switch and were consequently excluded from the smelled mixture and not sent to the mass spectrometer. 55 Supplementary Figure S2.3 Alternate views of correspondence analysis (fgure 4) incorporating the frst 3 dimensions of variation. 30.57% of variance explained by dimension 1 (x), 22.84% of variance explained by dimension 2 (y), 14.03% of variance explained by dimension 3 (z). 56

vii Supplementary Table S2.4 Tentative identifcation of lavender volatile compounds. Volatiles were identifed by matching their mass spectra to the NIST 05 Mass Spectral Library (National Institute of Standards and Technology, Gaithersberg, MD) and to chemical standards, as noted. Te table is divided by cut time for perceptual mixtures P1-P6. 57 Supplementary Table S2.4 continued 58 Table 3.1: Ingredients used in historical and contemporary recipes for bitters, listed by taxonomic name and literature source 61 Table 3.2 Bitters samples used in the study, with historical sources and precedents, and style noted. 63 Table 3.3 Sensory terms and references used in the descriptive analysis on bitters. 66 Table 3.4 Mean values of sensory qualities for each sample 68 Figure 3.1 Principal Component Analytsis (PCA) of bitters. PC 1 (X-axis) explains 40.8% of variance in sensory data; PC 2 (Y-axis) explains 30.2% of variance. Sensory descriptors are in red italicized text. Samples are in bold capital letters, coded by style: citrus in orange, aromatic in purple, tiki in blue, mole in brown, New Orleans-style in dark red, and celery in green. 69 Table 3.5 Compounds identifed by GC-MS in samples of bitters, by name, CAS number, Retention Time (RT), and Calculated (CRI) C8-C20 Retention Index (RI). ni=not identifed 71 Table 3.6 Volatile components reported as literature headspace abundances (% of total peak area) in ingredients used for historical and literature recipes for bitters. Ten most abundant compounds reported where available. T=Tomas 1862; P=Parsons et al. 2011; W=Wondrich, 2007; O=Orange bitters (Regan 2003); J=Jerry Tomas’ Own Decanter Bitters, (Tomas 1862); S=Stoughton’s bitters (Wondrich 2007); B=Boker’s bitters, (Wondrich 2007). 76 Figure 3.2 Plots of Partial Least Squares Regression (PLS) analysis of bitters volatile composition and sensory qualities by descriptive analysis. Sample names from Table 3.2; Sensory Attributes from Table 3.3; Compound Identifcations in Table 3.5. 78 3.2A: Positions of samples 78 3.2B: Biplot of sensory descriptors (red) and volatiles (blue) 78 3.2C: Exploded view of compounds 78 Supplementary Table S3.1 Headspace concentrations of volatiles in bitters, in ug/L 2-undecanone equivalents. ni=not identifed, nd=not detected 91 Figure 4.1 Chromatograms of (top) sample A, the control, uncut sample; and (bottom) sample E, with linalool, terpinyl acetate, and caryophyllene excluded from the reconstitution 99 Table 4.1 Aroma properties of GRO reconsitution mixtures listed by volatiles excluded and calcuated odor activity values (OAV) of excluded compounds. “intensity” is average overall aroma intensity rated by three panelits from 1-10 for each mixture. Aroma descriptors expressed as overall counts for each descriptor for each sample across three panelists and four replicates. 103 Table 4.2 Diferences in aroma qualities of samples with volatiles excluded compared to the control sample. Decreases in descriptor count for experimental conditions highlighted in yellow, increases in descriptor count highlighted in magenta. 104 Figure 4.2 Correspondece analysis biplot of samples (blue circles, blu text) and descriptor counts (red triangles, black text). Dimension 1 (x) Variance explained=37.2%, Dimension 2 (y) variance explained=27.6% 108 Figure 4.3 Multiple Factor Analysis (MFA) individual factor map comparing panelists’ individual map positions with consensus map positions 109

viii Supplementary Table S4.1 PLS1 analysis positions in biplot of compounds and Angostura sample compared to specifc descriptors (positions of descriptors is appoximately 0.2 along each PC), from bitters descriptive analysis data, chapter 3. 112 Table 5.1 Whiskeys and bitters used to prepare the 16 samples in the study 114 Table 5.2 Descriptors used in the sensory analysis and their aroma references. Signifcance by ANOVA denoted by superscripts: b= signifcant by bitters; w=signifcant by whiskey; o, b*w =signifcant bitters x whiskey interaction. References were presented as described, in opaque, lidded, black glasses. 115 Table 5.3 Mean values of descriptors rated in model Old-Fashioneds calculated as signifcant by ANOVA, by sample, bitters, and whiskey. Means by bitters (or whiskey) were only calculated when attribute intensities for model Old-Fashioneds difered signifcantly between the model Old-Fashioneds by bitters (or whiskey) in the 4-way ANOVA. Letters next to mean values are results of Tukey’s HSD test; attribute intensities with diferent letters are signifcantly diferent from each other for the samples studied. 119 Figure 5.1 Principal Component Analysis of model old fashioneds descriptive analysis data. Samples are in bold, all caps; descriptors are in red italic. Top: PC1(31% Variance explained, X-Axis) and PC2 (23% variance explained, Y-Axis); Bottom :PC1 (X-axis) and PC3 (14% Variance explained, y-axis) 121 Figure 5.2 Plots of mean intensities of oak aroma (A) and cola aroma (B), showing signifcant interaction efect between bitters and whiskey. 123 Table 5.4: Compounds identifed in samples, identifed by retention time (RT), Mass Spectra, and Calculated (CRI) and literature C8-C20 retention indices (RI) 124 Figure 5.3 Partial Least Squares Regression Analysis (PLS) plots of model old-fashioned volatile and sensory profles. Variance Explained PC 1: 40% X, 21% Y; PC 2: 21% X, 19% Y; PC 3: 13% X, 14% Y 126 5.3A: Positions of samples, PC1 vs PC2 126 5.3B: Positions of samples, PC2 vs PC3 126 5.3C: PC 1 vs PC 2 plot of sensory descriptors 126 5.3D: PC 1 vs PC 2 plot of headspace volatiles 126 5.3E: PC 2 vs PC 3 plot of sensory descriptors 126 5.3F: PC 2 vs PC 3 plot of headspace volatiles 126 Supplementry Table S5.1 Headspace concentrations in samples (listed by type of bitters and type of whiskey) in ug/L 2-Undecanone equivalents. ni=not identifed; nd=not detected. 134 Table 6.1 Samples used in the study, their substrates, and source of alcohol for acetifcation 138 Table 6.2 Descriptors used in the sensory anaysis, their references, and their signifcance for product by pseudo-mixed model ANOVA 141 Table 6.3 Mean intensities for sensory descriptors for each sample. Signifcantly diferent values determined by a Tukey’s Honest Signifcant Diference test have diferent letter groupings. 144 Figure 6.1 Principal Component Analysis (PCA) biplot of sensory descriptive analysis data. PC1 (X) explains 50.6% of variance, PC2 (Y) explains 22.9% of variance. Samples (scores) are in bold black text, descriptors (loadings) are in red text. 145 Table 6.4 Volatiles identifed in vinegar samples, by chemical name, Chemical Abstracts Service number (CAS), retention time (RT), C8-C20 retention index (CRI), and literature

ix retention index. ni= not identifed. 147 Table 6.5 Organic acids in g/L determined by capillary electrophoresis 149 Figure 6.2 Partial Least Squares Regression (PLS) analysis of sensory and chemical data on vinegar samples. 150 6.2A: Positions of samples 150 6.2B: Biplot of descriptors and compounds 150 6.2C: exploded view of positions of compounds 150 Supplementary Table S6.1 Peak area of volatiles in vinegar samples determined by GC-MS, normalized to 2-undecanone. ni= not identifed. nd=not detected. 163 Table 7.1: Samples used in the study, their base compositions, botanials added, and codes 167 Figure 7.1: Consensus plots from DISTATIS analysis of sorted groups from sensory analysis by 16 panelists on 16 vinegars, with bootstrapped confdence ellipses. 172 7.1A: Dimensions 1(X-axis) and 2(Y-axis) 172 7.1B: Dimensions 1(X-axis) and 3 (Y-axis) 172 Figure 7.2 Correspondence analysis biplot of descriptors used for malt vinegar samples. Dimension 1 Explains 30% of variance, dimension 2 explains 19% of variance. Descriptors are represented by red triangles and samples by blue squares. 174 Table 7.2 Volatiles identifed in malt vinegar samples, their retention times (RT), C8-C20 Kovats retention index (CRI) and literature retention indices (RI), ni=not identifed, tentative identifcations included where possible with unidentifed peaks 176 Figure 7.3 Principal Component Analysis (PCA) score plot of malt vinegars based on volatile composition. PC 1 explains 45% of variance, PC2 explains 21% of variance, PC3 explains 8.5% of variance. 178 7.3A: PCs 2 and 3. 178 7.3B (inset): PCs 1 and 2, included to show overall distribution of scores along PC 1 dominated by one sample. 178 Figure 7.4 Multiple Factor Analysis (MFA) of Sorting, Label, and Volatile data for 16 malt vinegars. 180 7.4A Individual Factor Map shows compromise positions of samples in the consensus space with positional disagreements plotted by dataset. 180 7.4B Factor Map for the Contingency Table shows the positions of samples and labeled sensory descriptors in the consensus space. 180 Supplementary Table S7.1 Relative quantifcation of volatiles in malt vinegar samples, in μg/L 2-undecanone equivalents. ni= not identifed. nd=not detected. 188

x Arielle Jurchak Johnson June 2014 Agricultual and Environmental Chemistry

Flavor Chemistry and Gastronomy: Sensory, Instrumental, and Multivariate Approaches Abstract

One of the subjects in which the currently co-evolving interests in a scientifc understanding of food in a gastronomic context and applying scientifc knowledge and empiricism to culinary development is the development, chemistry, and perception of favor.

Flavor is a complex phenomenon involving chemistry, biology, psychology, memory, and mixing- dependent perceptual efects.

Methods and analytical approaches to characterize the relationship between chemical composition and perceived favor must address this complexity, and the efects of synergy, masking, and gestalt perception in order to produce useful data and knowledge. In this dissertation, these methods include in-instrument reconstitution and omission experiments and multivariate statistical analysis of sensory and volatile data.

Te frst section of this dissertation focuses on directly evaluating mixing-dependent aroma perception efects using a newly developed technique called Gas Chromatography- Recomposition-Olfactometry (GRO) and lavender (Lavandula angustifola) as a model system. Using a GC-MS ftted with with a pneumatic fow switch, cryotrap, and olfactory port, selectively reconstituted mixtures of volatiles from a sample can be prepared and evaluated, omitting one or more compounds and describing the resulting sensory efects.

1 Te second section is a three-part analysis of the favor chemistry of aromatic cocktail bitters, a historically important product made from the extraction of aromatic plants in alcohol with recent resurgence in craf cocktails. Sixteen commercial bitters are analyzed using volatile and sensory profling and multivariate statistics including Principal Component Analysis (PCA) and Partial Least Squares Regression (PLS). Te mixing-related sensory roles of three terpenoid compounds in Angostura bitters are analyzed using GRO. Finally, as bitters are commonly used as a component in mixed drinks, the efect of a whiskey matrix on the favor chemistry of bitters is analyzed using model systems of “Old-Fashioned” cocktails.

Te third section is an analysis of favor and chemistry of vinegars produced by applying acetic fermentation to novel substrates, in the context of culinary research and development. In the frst part of this section, traditional descriptive analysis of vinegars produced from celery juice, strawberry wine, pine needles, and other substrates is correlated to volatile profles using PLS, as in the second section on bitters. Te second part of the section is an analysis of malt-based vinegars with diferent plant and fermented ingredients. A labeled sorting sensory analysis, a rapid method which has previously shown good comparability to descriptive analysis, is analyzed with DISTATIS and correlated to volatiles using Multiple Factor Analysis (MFA) and Correspondence Analysis (CA)

2 Chapter 1: Introduction and Literature Review

Tis dissertation is primarily concerned with two ideas: the complex nature of favor and its relationship to chemistry, and the favor chemistry of gastronomically interesting products and processes. Tese ideas direct the investigation of aromatic bitters, a product with a long cultural history and key use in craf cocktails, and applications of acetic fermentation to novel ingredients in the high-end restaurant Research and Development (R&D) kitchen.

Methodologically, this dissertation uses volatile and sensory profling and multivariate statistics to determine correlative relationships between perceived favor and chemical composition, and it develops new instrumental methods to causatively investigate specifc mixing-dependent sensory-molecular interactions.

Topically, aromatic bitters and acetic fermentation are chosen based on recent interest from restaurants, chefs, bartenders, and the media, and a lack of information on their favor chemistry as they are currently used.

Tis literature review will cover, as background, the psychology, chemistry, and analysis of favor; the history of gastronomy as a research subject in the sciences, and the developing demand for research, science, and technology in cuisine.

What is Flavor?

Flavor is a perceptual construct that primarily combines taste, retronasal olfaction, and oral somatosensation (Small et al 2007) with accessory information from the visual and auditory senses (Verhagen and Engelen 2006). Tis dissertation will focus primarily on aroma as a driver of favor and favor complexity, and a primary source of sensory diferences among samples.

It is important to note that while a food, beverage, or other type of sample may contain 3 molecules that impart favor, these cannot be said to ‘contain’ favor. Rather, these molecules have a bioactivity that stimulates receptors in the sensory system, and favor is the perception that arises from the processing of this stimulus (Auvray and Spence 2008).

Smell and Chemistry

Aroma1 is the perception of molecules that bind with olfactory receptors. Tese receptors are located, in humans, in the olfactory epithelium, an area of tissue located at the top of the nasal cavity. To reach these receptors, a molecule needs to be small (generally under 300 atomic mass units) and non-polar enough to be volatile, i.e. to have signifcant vapor pressure or partition into the gas phase (Rossiter 1996). As a shorthand, these ‘volatile molecules’ will be referred to throughout as ‘volatiles’. Humans possess around 500 types of olfactory receptors, (Axel 2005, Glusman et al 2001, Zozulya et al 2001, Young et al 2002) which enable us to detect tens of thousands, if not more, of diferent molecules. Te olfactory system is able to detect and diferentiate a larger number of volatiles than it has receptors through a process called combinatorial coding (Malnic et al 1999) in which each volatile interacts with and activates a number of receptors to various afnities (Floriano et al 2000). Likewise, each receptor can be activated by several diferent molecules. Each molecule will produce a relatively unique ‘code’ of receptors and levels of activation, and the processing of this set of signals is interpreted in the brain as an odor. When mixtures of volatile compounds are considered, which take on unique odor qualities of their own as discussed below, humans are estimated to be able to discriminate 1 trillion diferent odors (Bushdid et al 2014).

Te ultimate drivers of the molecular feature-perceived odor relationship for a given molecule are not fully understood, however, we do know that features such as alkyl chain length (de Melho Castano Amboni et al 2001); rigidity arising from double bonds, rings and bridged rings (Hong and Corey 2006), aromatic rings (Rossiter 1996), heteroatom functional groups such as alcohols, carbonyl groups, aldehydes, ketones, organic acids, esters, and lactones (Laing et al 1 “Aroma,” “odor,” and “smell” will be considered synonymous for the purposes of this dissertation. Additionally, unless taste or other stimuli are referenced directly in context, “favor” will also refer to aroma. 4 2003, Sanz et al 2008); halides, sulfur, and nitrogen; as well as stereoactivity (Bentley 2006) all have impacts on the perceived aroma quality of a given molecule. Besides aroma quality, these factors may also afect odor threshold (Edwards et al 1991, Mihara et al 1988) (the amount or concentration of a particular molecule required to impart a detectible odor), odor pleasantness, and odor complexity (Kermen et al 2011).

Te foods and food products we use for nourishment and culinary enjoyment typically contain numerous molecules that contribute to aroma. Plant species (lavender [An et al 2001], pine [Domrachev et al 2011], elderfower [Kaack et al 2006]), product that have undergone thermal or other physical or chemical processing (chocolate [Afoakwa et al 2008], cofee [Ribeiro et al 2009]), products arising from the microbial processing of plant or animal species (wine [Ebeler and Torngate 2009], beer), and dishes, ingredients, or other food products created by combining several of these together (whiskey, gin [Pryde et al. 2011, Sanchez 2011]) have dozens or even hundreds of individual volatile components . Essentially, all foods are mixtures2, and so understanding food favor means understanding not only constituent aromas, but also how aroma functions in multicomponent mixtures.

In multicomponent mixtures, there is a complex relationship between chemical composition3 and perceived aroma (Wilson & Stevenson 2006). Understanding the perceived aroma of a mixture in relation to its chemical makeup requires investigating perceptual efects that arise from mixing, such as additivity (Berglund et al 1993), masking (Cain 1975), and emergent or synergistic efects. Perception of the characteristic aroma of a mixture is stored by the brain as a fairly unitary encoded pattern called a ‘smell image’ or gestalt (Livermore and Laing 1998), related to both the original receptor signals stimulated by a sample’s component volatiles and processing-related interactive efects (additivity, masking, synergy, etc). In other words, once a mixture of volatiles becomes familiar as the aroma of a particular entity or object (e.g. the

2 Due to chemical complexity introduced by biochemical processes from plants and animals themselves as well as microorganisms, oxidation, reduction, heat related and other reactions yielding multiple types of molecules. 3 Which can be determined either by making a synthetic model mixture out of pure compounds, or analyzing the volatile composition of an existing or naturally-occurring mixture by Gas Chromatography-Mass Spectrometry [GC-MS]. 5 characteristic mixture of volatiles that makes up chocolate aroma), that object’s aroma is encoded and processed as a discrete entity rather than the conglomerate of the aromas of its component volatiles (Livermore and Laing 1998).To understand the scope of perceptual efects on aroma mixtures, it is illustrative to consider both “top-down” and “bottom-up” approaches. From the top down, the characteristic favor of many products (for example, wine, chocolate, or lavender) is not traceable to any one particular volatile. Rather, it arises from the characteristic mixture of volatiles that typifes that product. From the bottom up, model mixtures with as few as three components have been found to have aroma qualities not attributable to any of the aroma qualities of their three individual component volatiles (Le Berre et al 2008). Terefore, in the context of favor, mixing efects are both widespread, playing a role in many products, and pervasive, coming into play even for simple mixtures.

To expand on the concepts outlined above, the ultimate aroma of a mixture as it relates to chemical makeup may be impacted by additivity, whereby components that are undetectable individually become detectable when smelled together (Ryan et al 2008); by masking, where one component will suppress the aroma or intensity of another (Pineau et al 2008, Preston et al 2008, Hein et al 2009); and by synergy or other emergent qualities, where the perceived aroma (as a gestalt) of the mixture has characteristics not found in any of its components (Le Berre et al 2008). An aroma image or gestalt is encoded and stored in the brain through a combination of analytic processing—where components can be distinguished and do not interact—and synthetic processing—where information about components is lost to the unique and blended qualities of the mixture (Laing and Francis 1989). A mixture of as few as three volatile compounds can have aroma qualities that do not exist in any of the individual compounds when smelled on their own, providing evidence that aroma synergy and synthetic processing require very little in terms of chemical complexity to be signifcant (Le Berre et al 2008). However, a study of mixtures made up of one to fve compounds found that, while mixing impairs the ability of human test subjects to identify the components, these subjects were nonetheless able to name some (up to three) of the components involved (Laing and Francis 1989). A related example of gestalt processing is 6 Livermore and Laing’s 1998 study of multicomponent mixtures of familiar aromas—including chocolate, lavender, and honey—that are themselves mixtures of many odor-active compounds. Mixtures of these familiar smells (for example, a 5-component mixture of chocolate + honey + cheese + lavender + strawberry4) are perceived in very similar ways as mixtures of individual compounds (Livermore and Laing 1998). In both cases, subjects were able to identify some of the odorants present, a maximum of approximately three to four odors in an eight-component mixture. Mixtures of complex aromas (chocolate, lavender, honey), each processed as a gestalt, are perceived through a combination of analytic processing—as some component gestalts can be identifed—and synthetic processing—as some get lost to the aroma background or to interactive perceptions (Jinks and Laing 1999, Barkat et al 2011). Te association of a mixture of signals into a single gestalt is dependent on experience and learned associations (Auvray and Spence 2008).

Analyzing Flavor

Analytical Flavor Chemistry

Picking the right tools for analyzing favor, and designing experiments that can actually answer the questions we want to ask about the relationship between perception and chemistry depends on understanding as fully as possible the nature of the interconnected systems that contribute to favor. As we know that volatile favor compounds interact in complex ways to determine aroma, we need methods that address mixing and interactions in order to fully understand the relationships between sensory and instrumental data on perceived favor and the molecules that cause it. Analytical favor chemistry necessarily deals with two questions and therefore tool sets – the chemical nature of the product, and the perceived favor of the product.

Identifying the chemical components potentially contributing to the aroma of a sample requires several steps. Generally, it is necessary to separate the volatile components from the non- 4 Which is processed perceptually as a 2-component mixture (of two odors) because of gestalt processing, but is chemically speaking a highly multicomponent mixture of volatiles from chocolate, honey, cheese, lavender, and strawberry 7 volatile components or matrix while preserving their relative composition through some form of extraction. Next, chromatographic separation of this mixture of volatiles is performed, and the separated components are analyzed qualitatively, semi-quantitatively, or quantitatively by a detector.

Extraction methods used for favor analysis include static headspace, dynamic headspace/ purge and trap (Kanavouras et al 2005, Ferraces-Casais et al 2013), solvent extraction (Caldeira et a 2007), solid-phase extraction (SPE) (Lopez et al 2002), simultaneous distillation-extraction (SDE) (Lee and Ahn 2009), solvent-assisted favor evaporation (SAFE) (Willner et al 2013, Engel et al 1999), and sorptive methods such as solid-phase microextraction (SPME) and stir bar sorptive extraction (SBSE or Twister) (Fan et al 2011, Bicchi et al 2002). Headspace-Solid-Phase- Microextraction or HS-SPME, in addition to being solventless and having a low margin for error (Wardencki et al 2004), has been shown to create extracts that are particularly representative of the aromas of the samples on which the extraction has been performed (Poinot et al 2004, Aceña et al 2010), and this method has been employed extensively in the literature and in this dissertation.

Gas chromatography (GC) has been available since the 1950’s as a method for separating mixtures of volatiles for individual detection (Ebeler and Torngate 2009). In this method, a fused silica capillary with internal diameter typically between 0.15-0.5mm and length typically between 15-60 meters has a stationary phase bonded or coated to its inner walls in a layer usually less than one micron thick. Volatilized analytes are introduced into the capillary or ‘column’ in a stream of gas, typically helium, hydrogen, or nitrogen, that also acts as the mobile phase. Analytes are separated by their afnity to the stationary phase, where those analystes that experience more, or stronger, interactions with the stationary phase are retained longer and therefore elute from the column later, and those with fewer or weaker interactions are less retained by the stationary phase and elute earlier. Te chosen column stationary phase chemistry can vary and is usually chosen to best separate the analytes of interest; dimethylpolysiloxane is a very common stationary phase

8 for aroma work, either in its pure form or 5% substituted with phenyl groups, as is polyethylene glycol and these phases were used for the work covered in this dissertation.

Once the volatiles are separated by the column, they can be detected individually. Detectors commonly coupled to GC include fame ionization detection (FID), heteroatom- specifc detectors such as the Nitrogen-Phosphorus detector (NPD) or sulfur chemiluminescence detector (SCD), or various types of mass spectrometer (MS)—single-quadrupole (employed in this dissertation) as well as tandem methods such as triple-quadrupole (QQQ) and quadrupole- time-of-fight (Q-TOF). GC-MS has the beneft of being applicable to both quantitative and semi-quantitative analysis, as well as providing structural information about each analyte through comparison to the mass spectra of lab-run standards or those deposited in libraries such as the NIST mass spectral database.

HS-SPME-GC-MS allows for the determination of the volatile profle of a sample—a chromatogram showing the volatiles it contains separated in time and with relative concentrations refected by peak area that can be tabulated for further analysis. Te practice of using a human volunteer as an “odor detector” is referred to as GC-Olfactometry or GC-O (Grosch 2001). In this method, part or all of the stream of volatilized compounds eluting of the column are diverted from the instrumental detector (e.g. MS or FID) to a port or cone from which the subject snifs over the course of the chromatographic run and records their perceptions of the aroma quality, and sometimes intensity, as it changes over time. By comparing these perceptions in time to the elution times (as determined instrumentally) of the volatiles, the individual odor contributions of each volatile in a sample can be analyzed.

Commonly, a favor-chemical study of a product seeks to quantify the impact or contribution of each volatile compound in a sample to the sample’s overall aroma. One benchmark for the importance of an individual volatile compound in a sample is the odor activity value (OAV), defned (Patton and Josephson 1957) as the concentration of the compound divided by its experimentally determined odor detection threshold concentration in a model matrix or

9 solution. Any compound found in concentrations “above threshold” will have OAV>1, and any compounds “below threshold” will have OAV<1. A second benchmark for volatile importance uses GC-O in a methods such as Charm (Acree and Barnard 1984)5 or Aroma Extract Dilution Analysis (AEDA) (Grosch 1993); in the latter, an extract is successively diluted and olfactometry is performed on the GC-separated eluant of each diluted extract. Each compound or peak is then assigned a “dilution factor” value, the number of N-fold dilutions required to suppress its detectability by olfactometry. Te more dilutions required, the higher the dilution factor for an odor compound and the more important that compound is considered to the aroma of the sample.

OAV and/or dilution factor in these types of analytical studies are ofen used as a criterion for building reconstitution and omission experiments. In reconstitution and omission experiments, a model matrix is used to create a reconstitution of the volatiles in the sample. Each volatile is quantifed in the original sample, and dosed at this measured concentration in the model from a stock of purchased or synthesized pure compound. Commonly, a sensory- activity cutof is used to eliminate compound from the model reconstitution; for example, only compounds with OAV or dilution factor above a certain value may be included. Preparing the same reconstitution several times, each time omitting a diferent compound, allows for comparison of the omission models with the full reconstitution. Te underlying logic is that if the mixture omitting compound ‘x’ is detectably diferent from the full reconstitution, then compound ‘x’ is important for the aroma of the original sample. Tere are several limitations with this design – it requires every compound in a sample to be above the limit of quantitation for the instrumental system in use, to be available as a pure standard, and for the matrix used for threshold determinations and reconstitution to model the sample well. Furthermore, compounds in concentrations below their sensory detection thresholds may have a signifcant role in aroma perception through interactive efects, and examination of more than one sample, or any interactive efects within a sample, may extend the experimental design to a ponderous

5 Charm refers to an aroma quality and intensity at a given retention index 10 size6. Te existing instrumental and disciplinary toolset for analyzing favor chemistry causatively is undesirably reductive, given its inability to adequately address known mixing and interactive efects, and is hindered by the artifcial constraints of the OAV, thresholds, dilution factor, and model reconstitutions (Ryan et al 2008). Flexible, facile methods for producing representative reconstitutions will help bridge this gap between theory and practice in favor chemistry, and one such method is discussed in chapter 2.

Olfactometry-based approaches consider favor chemistry by investigating the aroma roles of compounds on an individual basis, ofen on one sample or a small set of samples. For analyzing perceived favor in a class or larger set of samples (for example, Malbec wines from diferent regions, or commercially available gins) and its relationship to chemistry, an approach of sensory profling followed by multivariate statistical analysis, ofen in tandem with data on chemical composition, is frequently used.

Descriptive Analysis and Sensometrics

Sensory descriptive analysis is a technique for generating quantitative data on the sensory characteristics of foods or other samples (Lawless and Heymann 2010). Since sensory qualities7 cannot be directly measured instrumentally, systematic evaluation of these attributes by human volunteers yielding numerical data allows for analysis that would otherwise be impossible.

Descriptive analysis (DA) was frst developed in the late 1940’s as a technique called

“Flavor Profle®” (Cairncross and Sjostrom 1950) and has evolved across a number of formats, proprietary and otherwise, over the decades since (Stone et al 1974, Brandt et al 1963, Civille and Lyon 1996, Powers 1988, Murray et al 2001). Generically speaking (Amerine et al 1965, Heymann et al 1993, Stone and Sidel 1974), DA uses a panel of judges to quantitatively rate the intensity

6 For example, in a sample with 10 volatile compounds with OAV>1, an experiment examining only single-compound contributions would require the preparation of 11 reconstitutions (full model

+ 10 full-minus-1 models); examining 2-component interactions would require 10C8 = 45 additional reconstitutions. A reconstitution-based study of 3 such products in a category with 2-component interactions being considered would require 3(11+45)=168 reconstitutions. 7 For example aroma (generally) or astringency, pear-like smell, or sourness (to give just a few specifc examples) 11 of one or more sensory attributes in a product or set of products. Tese panelists are trained with (ideally) physical references for each attribute they are asked to rate, so that they agree on the meaning for each term used to describe an attribute and this meaning is grounded in a real object (Giboreau et al 2007). Where physical references are not available, specifc defnitions are agreed-upon by all panelists for each attribute (Lawless and Heymann 2010). For example, using the verbal concept of “a damp basement” to defne “mustiness” when it was not practical to have panelists visit a damp basement each time they were asked to rate mustiness (Wood et al 1995).

It is typical in a descriptive analysis to have the panelists generate the terms that they, as a group, feel are important for characterizing and distinguishing the samples they are to analyze. Tis is usually accomplished through balloting, or by consensus, where panelists in groups smell, taste, or otherwise sensorially interact with the samples a few at a time, and generate sensory terms for each, which are collected by a panel leader. Te panel leader will then make references for each term, which the panelists will test and either accept or tell the panel leader how or why the reference doesn’t match the sensory attribute. Tis is repeated over several sessions, with the panel agreeing to eliminate redundant or unimportant attributes, until a refned list of attributes and agreed-upon references for each have been generated.

During DA, panelists typically rate attributes by selecting a point along an unstructured line scale with labeled ends (e.g. “very low” to “very high” or “not present” to “most intense”); this has an advantage over rating with integers (for example 0-9 or 1-10) because it results in truly continuous data that can be analyzed with a variety of statistical techniques. One goal of DA is to determine which attributes are most important for describing how samples in a set difer from one another. For this purpose, a univariate Analysis of Variance (ANOVA) on each descriptor will identify those descriptors which difer signifcantly in intensity between the samples (by comparing variance that can be identifed as coming from samples to variance which derives from random error). Since the sensory attributes of foods, beverages, and other similar products tend to be complex, DA ofen results in data on many attributes whose explication and analysis

12 call for the use of multivariate statistical techniques. Whereas univariate methods can describe the relationship between one dependent variable and (sometimes) several independent variables, multivariate methods can simultaneously examine multiple dependent variables (e.g. sensory characteristics) and the relationships among them.

Flavor and Multivariate Statistics

Principal Component Analysis

Principal Component Analysis (PCA) is a method for simplifying a dataset with minimal information loss, wherein multiple dependent variables (e.g. sensory descriptors) each describe multiple samples or objects. PCA identifes the main interrelationships among samples, among dependent variables, and among samples and dependent variables. Formalized in 1933, it is possibly the oldest multivariate data analysis technique, and is used in one form or another in most scientifc disciplines (Abdi and Williams 2010). In the context of favor research, PCA is usually performed on sensory data, but it can also be used to examine latent relationships among other variables, such as volatile profles (Wang et al 2009).

Descriptive analysis performed on a set of X samples with N descriptors results in a dataset that can be thought of as an N-dimensional cloud of data—if every descriptor is an axis, and the intensity of each descriptor has been rated in each samples, then each of these ratings can be thought of as describing a distance along this axis, and each product8 has a position defned by its coordinates on each of the N axes. In a descriptive analysis experiment with ratings on thirty sensory qualities (which is fairly typical), this representation of the sensory qualities of each sample will be a thirty-dimensional “cloud” with each sample as a point with a defned position on each of thirty axes, each of which are orthogonal to the others.

A Principal Component Analysis (PCA) is a dimensionality-reducing statistical technique

8 Either each product as rated by one panelist during one replicate, or the mean value of all of these ratings, of which there is (# of panelists)*(# of replicates) for each product 13 that identifes latent correlations (“Principal Components”) between axes in a multivariate dataset, and uses these to reduce the dimensionality of the dataset while preserving as much variance as possible. Tis can aid interpretation of the descriptive analysis data “cloud” mentioned above by spatially describing relationships between many products and descriptors in two or three dimensions instead of thirty. Te graphical output of a principal component analysis on a sensory dataset is a space where the relative positions of samples and descriptors conveys information about their interrelationships. Tis is usually presented as a two-dimensional biplot (though higher dimensions can be plotted and are ofen useful to examine), showing the positions of scores (samples) and loadings (descriptors) in relation to the frst two principal components.

Loadings are typically plotted as vectors, and the greater the magnitude of a vector, the more variance there is in the dataset for that descriptor. A loading with a very long vector can be interpreted as having a greater importance for determining diferences in the dataset. Te smaller the angle of a loading vector to a principal component, the more that principal component refects that particular descriptor. Between multiple loadings, the closer the angle between their loadings is to zero, the more highly correlated they are to each other; a 90-degree angle between loadings means those descriptors are uncorrelated in relation to the plotted principal components, and an angle close to 180 degrees refects a negative correlation between descriptors.

Te positions of samples in the biplot refects how close or diferent each sample is from a sensory ‘average’ score, by its overall distance from the center, and how much of its sensory qualities are explained by each PC, i.e., by its x- and y- coordinates. Samples with scores plot that are close to each other are more similar, overall, than samples with scores plot that are far apart from each other. In addition, samples that plot closer to a particular loading vector are more highly correlated to that sensory quality than samples plotting away from that vector. In this way, a PCA allows for interpretations of latent correlations or similarities between samples and their descriptors overall, as well as between descriptors, between samples, and between specifc descriptors and samples in a holistic, multivariate context. Te validity of this interpretation

14 depends on the amount of variance explained by the retained PCs, and if a fairly low amount of variance is explained by the frst two PCs, or if the third or higher PCs explain a similar amount of variance as the frst or second PCs, it may be necessary to retain, replot, or at least address diferences in interpretation introduced by these PCs (Heymann and Lawless 2010).

Partial Least Squares Regression

Partial Least Squares (PLS, also called Projection onto Latent Spaces) is a form of data analysis that allows for the simultaneous modeling and comparison of relationships among multiple independent and multiple dependent variables. PLS borrows features from both PCA and multiple linear regression with a goal of predicting as much of the dependent dataset as possible given a dataset of independent variables, based on their common structure (Abdi 2010).

As a variance-based form of structural equation modeling, PLS focuses on capturing as much variance in multiple dependent variables that can be explained by variance in multiple independent variables (Heinlein and Kaplan, 2004). PLS has been shown to work robustly on data with high multicolinearity of dependent variables (Cassel et al 1999). Both of these make PLS a useful analytical tool for studying the relationships between (dependent) sensory characteristics and the (independent) instrumentally-measurable chemical and physical properties that provide the stimuli for those sensory characteristics. PLS has been used to predict sensory characteristics of Chardonnay and aged red wines from their volatile profles (Noble and Ebeler 2002, Lee and Noble 2006, Aznar et al 2003), predict the geographic origin of wines from selected chemical parameters (Capron et al 2007), predict the sensory qualities of cooked salmon from the sensory qualities of raw salmon (Rodbotten et al 2009), and to diferentiate between “stuck” or sluggish wine fermentations and fermentations proceeding normally using headspace volatiles (Malherbe et al 2009). In a favor-chemical study, the goal is to understand the molecular drivers of favor, so chemical diferences not correlated to sensory diferences are not useful for analysis. Multicolinearity in sensory data could come from certain diferent favor attributes arising in parallel from the same process, and from individual compounds contributing to more than one aroma quality. 15 It may be useful to distinguish between two approaches to using chemical data in a multivariate analysis. In a chemometric approach to analyzing, say, GC-MS data, one would seek to retain as much chemical variance as possible in distinguishing among samples. PLS maximizes the variance explained in the dependent (Y) rather than the independent (X) dataset, so with sensory (Y) and chemical (X) data, less variance in the chemical dataset may be retained than in a simple PCA of the chemical dataset, but the variance that is retained explains more of the variance in the sensory dataset.9

Multivariate Statistical Techniques on Non-Continuous or Nontraditional Data

In some cases, useful data can be collected from sensory experiments that don’t follow a typical descriptive analysis procedure, requiring the use of multivariate statistical analysis techniques beyond the PCA and the PLS. Verbal methods (Valentin et al 2012) such as check- all-that-apply (CATA) and free choice (FCP) profles have each panelist describe samples by the presence or absence (rather than intensity) of attributes, which are either from a master list (CATA) or are panelist-generated (FCP). Using citation frequency as a measure of the importance of sensory descriptors rather than variations in intensity necessitates analytical methods which deal with categorical rather than continuous data, such as a correspondence analysis. Similarity- based sensory methods such as the sorting task (Chollet et al 2011, Lelievre et al 2008) ask panelists to group samples based on their similarity to each other and don’t necessarily use a descriptive component. Methods such as DISTATIS (Abdi et al 2005, 2007) can then be used to compare latent similarities in individual distance matrices to generate a consensus map that can be interpreted much like a PCA plot of scores. Rapid profling data from FCP, CATA, Sorting, Projective Mapping and other types of sensory analysis can be co-analyzed with descriptive or instrumental data using a multiple factor analysis (MFA) (Chollet et al 2011, Abdi 2003), allowing the latent similarities in several data sets to be visualized and interpreted simultaneously. Specifc details are described more fully in pertinent following chapters.

9 So, the ultimate question is, are we trying to model chemical variance or use chemistry to explain sensory variance? In the case of this dissertation it is generally the latter. 16 Applied Flavor Chemistry: Science, Food, Culture

Flavor and chemistry of wine and other beverages

Studies on the relationship between chemistry, especially volatile profles, and sensory perception of plants and plant products are ofen undertaken or published under the guise or purview of the favor and fragrance industries, with some intersection with plant sciences, entomology, and other felds. Agricultural and food products that have undergone some kind of human-directed processing and that have economic importance are also common subjects for sensory and/or favor chemical analysis; for example cheddar cheeses, sherry vinegars, wines from diverse varietals and regions, aged buckwheat vinegar (Aili et al 2011), huitlacoche (Lizarraga- guerra et al 1997), qu (rice wine) (Mo et al 2010), molded-surface ripened cheeses (Martin et al 2001), and others. Tese favor chemistry studies of foods, in characterizing an economically or culturally important product in terms of its sensory and chemical qualities, provide data in a form that can be communicated objectively10 and publicly11; that can be used for quality and process optimization; and that enables further generation of knowledge through comparison to other analyses on similar or related products.

Sensory and volatile analyses have been applied extensively to a wide range of beverages, including wine, beer, whisk(e)y—Scotch, Bourbon, etc.—tequila, rum, mezcal, gin, and others. Despite the long history of alcoholic beverages as part of human culture, many of these products have at best been investigated only preliminarily. Artisanal and traditional distilled beverages, and alcoholic beverages incorporating plants, such as gin, bitters, amaro, chartreuse, or absinthe, have long histories, diverse formulations, highly developed cultural signifcances, and complex favors (Tonutti et al 2010), but their favor chemistries have not been examined in-depth.

Wine is probably the most well-studied of any alcoholic beverage, and has been analyzed, extensively for all of the outlined purposes. Early wine favor chemistry studies (in the 19th

10 In the sense that described sensory qualities are tied to real references, and chemical profles are instrumentally measurable qualities 11 In the sense that published data, even if paywalled, is not secret 17 to mid-20th century) (Ebeler and Torngate 2009) focused on characterizing major chemical components—alcohol, tannin, organic acids, sugars—as well as categorizing and attempting to prevent common chemical and microbial defects (Polaskova et al 2008). As analytical methods have become more sensitive and sophisticated, and computing power has increased, emphasis has shifed from defects to characterizing the fundamentals of wine favor chemistry and perception (Hjelmeland et al 2013, Hein et al 2009, Heymann et al 2013, King et al 2013), and the favor chemistry of high-quality wines. Tis includes identifying the many hundreds to thousands of volatile and non-volatile compounds present in wines that afect their favors, tracing the relationship between these compounds and grape, microbial, or oak metabolism and chemical or physical interactions, and characterizing their sensory interactions which result in perceived favor.

Sensory and favor-chemical studies of wine have been followed by the development of research that has explicitly addressed wine not just as a complex assemblage of intersecting chemical, biological, and sensory qualities, but as a cultural entity. Considering wine in isolation from all other factors ignores the fact that perception of favor during consumption is, in everyday life, an experience mediated by culture, and that culturally, wine is frequently consumed as an accompaniment to food. Terefore, a scientifc understanding of the sensory qualities of wine is incomplete until it includes an understanding of how these sensory qualities, and changes in these qualities, are related to the various culturally-important ways and contexts in which wine is experienced by humans.

One example of this is the sensory analysis of wine and food pairing. Tis practice is subject to extensive attention in the popular press, is the primary task of sommeliers, and is a defning component of wine economics, however, until recently it received very little attention from sensory descriptive analysis research. Preference for one of six diferent wines paired with a chicken dish and a wild boar dish was evaluated in a 1997 Swedish study involving 220 subjects (Nilsson et al 1997), and the frst quantitative study focusing on perceptual changes induced by

18 food-wine interactions focused on Chardonnay wines and Hollandaise sauces. Wine and cheese pairing has been studied with white wine and blue mold cheese (Nygren et al 2002), red wine and eight diferent cheeses of diverse styles (Madrigal-Galan and Heymann 2006), and a mix of red and white wines with diferent cheeses (Harrington and Hammond 2006). A 2010 study of specifc pairings of ultra-premium wines ($25-$80 bottle price) and artisanal cheeses, with and without additional food accompaniments (such as chutney or nuts) had panelists rate “match level” of favor intensity and body, but did not look at specifc favor attributes (Harrington et al 2010).

Food, is a signifcant intersection between physical qualities, sensation, and culture. An academic understanding of this continues to develop, as does a parallel appreciation by the crafspeople who produce these foods. Besides wine pairings, sensory and hedonic qualities of other food combinations have received attention, including pork and vegetable accompaniments (Aaskyn et al 2010), olive oil with other foods (Cerretani et al 2007), salmon and sauces (Paulsen et al 2012, 2013), chocolate and beverages (Donadini et al 2012), and beer and cheese (Donadini et al 2013). An analysis of beef stock production with wine as an ingredient found that the use of chemically and sensorially diferent wines correlated to chemically and sensorially diferent stocks and reductions, though how diferences in wines led to these diferences in stocks was not clear (Snitkjaer et al 2011). Tis latter study intersects wine chemistry with the nascent academic discipline of “molecular gastronomy”, discussed further below.

Cuisine Research and Development

Beyond wine, explorations of food as an intersection of physical and chemical properties, human perception, and culture, have occurred in a number of guises and forms. Food science and technology is ofen defned by a research focus benefting the interests and interrogating the issues facing the food processing industry (see, for example, the homepages of the food science departments at UC Davis or Perdue University). Interest in the science behind cuisine or “the art 19 of cookery” has developed parallel to these concerns. In 1794 Sir Benjamin Tompson, Count Rumford wrote (Tompson, 1794):

Te advantage that would result from an application of the late brilliant discoveries in philosophical chemistry and other branches of natural philosophy and mechanics to the improvement of the art of cookery are so evident that I cannot help fattering myself that we shall soon see some enlightened and liberal-minded person of the profession to take up the matter in earnest and give it a thoroughly scientifc investigation. In what art or science could improvements be made that would more powerfully contribute to increase the comforts and enjoyments of mankind?

Tis early proposal of turning science towards the study of food as a product of enjoyment did not take root much further until one hundred and seventy-fve years later, when in a presentation to the Royal Society entitled “Te Physicist in the Kitchen”, the physicist Nicholas Kurti echoed this concept, remarking “I think it is a sad refection on our civilization that while we can and do measure the temperature in the atmosphere of Venus we do not know what goes on inside our soufés” (McGee 1999). Te year afer the publication of his seminal On Food and Cooking (McGee 1984, 2004), food science writer Harold McGee argued for “considering cookery as a discipline with a scientifc base” as “the perspective ofered by science enriches the experiences of cooking and eating” (McGee 1986). McGee and Kurti, along with the cookbook writer Elizabeth Cawdry Tomas and physical chemist Herve Tis, went on to organize a series of seminars at the Ettore Majorana Centre for Scientifc Culture in Erice, Sicily on “Molecular and Physical Gastronomy,” later renamed “Molecular Gastronomy,” starting in 1992 (Arnold 2006). Te latter term was also used in several books by Tis (Tis and DeBevoise 2013, Tis 2013) and became the colloquial umbrella name for research focusing on cooking or cuisine (Mielby et al 2010, Barham et al 2010, Risbo et al 2013, Snitkjaer 2010, Snitkjaer 2011). Academic “molecular gastronomy”

20 has examined the chemical and sensory aspects of beef stock cooking (both with (Snitkjaer 2011) and without (Snitkjaer 2010) the addition of red wine), as well as the efects of expectation and surprise in diners’ perceptions of during an ultra-high-end, avant-garde meal (Mielby 2010).

Te evolution of molecular gastronomy as a research discipline has run parallel to the development of modernist and experimental approaches to cooking and ideas about food at infuential restaurants. In a study of how priming by waitstaf afected diner’s perceptions of an haute cuisine meal molecular gastronomy is referred to as “the scientifc study of deliciousness” as well as a “culinary trend”, highlighting some controversy about the diferences between popular food trends or movements and their relationships to science (Mielby et al. 2010). Tis was highlighted in a 2006 “Statement on the new cookery” signed by McGee along with chefs Ferran Adria, Heston Blumenthal, Tomas Keller, and others- all of whom had gained prominence by way of the San Pellegrino World’s 50 Best Restaurants list and signifcant media attention as part of the experimental culinary avant-garde. Te Statement argued that “the term ‘molecular gastronomy’ does not describe our cooking, or any style of cooking” (Adria et al 2006). Te “New Cookery” that the statement references, also called “Modernist Cuisine”, is widely agreed to have its roots in Adria’s restaurant, elBulli, beginning in the late 1980’s (Myhrvold 2011). Chefs identifed or labeled as “modernist” today are heavily represented among what are widely considered to be the best restaurants in the world (Edelstein 2013, Smith 2013). While they have extremely diverse styles, generally speaking “Modernist Cuisine” describes a culinary modernism, which rejects culinary rules in favor of creatively re-defning cuisine and cooking; experimentation to develop new techniques, technology and ideas; new uses for ingredients and entirely new ingredients; and the perfection of recipes by way of borrowing from, collaborating with, and performing scientifc research to better understand food functionality and enable development, the informed rejection of tradition, and new culinary possibilities (Myhrvold et al 2011, Blumenthal 2006, Keller 2008, Atala 2013, Aduriz 2012, Smith 2013, Kamozawa and Talbot 2010). Te same statement by Adria et al that criticized the use of the term “molecular gastronomy” to describe cuisine went on to say that “a spirit of collaboration and sharing is 21 essential to true progress […] to explore the full potential of food and cooking, we collaborate with scientists, from food chemists to psychologists” (Adria et al 2006).

R&D activity has played an increasingly important role at the restaurants of these and similarly avant garde chefs, both internally and via collaboration with scientists working in industry and academia, with innovation becoming a marker and driver of quality in high-end restaurants (Tan 2013). Blumenthal’s restaurant Te Fat Duck has had a dedicated experimental kitchen (separate from the service kitchen which prepares food for diners) since 2004 (Jonny Lake, Fat Duck head of R&D, personal communication, Mar 26 2014) , and elBulli opened a physically separate R&D facility called El Taller in 1997. El Taller was active for six months of each year (with the restaurant operating for the other six) until the restaurant closed, permanently, in 2011, to become a dedicated research foundation devoted to food (Andrews 2011). Te Basque restaurant Mugaritz, headed by chef Andoni Luiz Aduriz, has a formal relationship and full R&D kitchen within the food science research center Azti-Tecnalia in Bilbao, Spain (Aduriz 2012). Similarly, the Danish chef Rene Redzepi, of restaurant Noma in Copenhagen, founded the Nordic Food Lab in 2008 (Mouritsen 2013) and also maintains a dedicated test kitchen (Redzepi 2013), and the Momofuku group of restaurants, centered in New York City, established a culinary lab12 in 2010.

Te introductory notes of papers on favor chemistry frequently make mention of the economic and/or cultural importance of the foodstuf being analyzed. Besides having an important place within the restaurant world, experimental haute cuisine, with an emphasis on empiricism and R&D, is increasingly an important part of the public consciousness. Chef Rene Redzepi has appeared on the cover of the international edition of Time magazine twice in the last two years (March 26, 2012 and November 18, 2013). On the latter, he appeared with chefs David

12 It should be noted for clarity that these facilities aren’t strictly laboratories in the academic sense of the term, functioning more like development kitchens with a heavy research component. However, many of them disseminate research fndings through talks, lectures, and peer-reviewed publications, and many have actively collaborated with scientists and other academics, as addressed below. 22 Chang and Alex Atala, who have both published peer-reviewed work on applications of botany, microbiology, and chemistry to cuisine, as have Aduriz (Arboleya et al 2008) and Blumenthal (Oruna-Concha et al 2007). In addition to formal papers, many of these chefs have been active in, and received considerable attention for, engaging in research independent of their restaurants (Redzepi’s Nordic Food Lab and Alex Atala’s ATA group), sharing of results via Internet platforms (Williams 2011) and other media (Blumenthal 2002), and public dissemination working alongside academics (Chang et al 2011, 2012, Chang 2011, 2012).

Research and development in contemporary experimental cuisine has had a focus on culinary uses for new, native, very old, or underutilized plant and animal ingredients; including seaweeds, insects, and foraged and indigenous plants unique to a region, such as Scandinavia or the Amazon (Blumenthal 2013, Mouritsen et al 2012, Williams and Hermansen 2012, Redzepi 2010, Redzepi 2013, Nilsson 2012, Atala 2012, 2013). Novel techniques and new applications for existing techniques are also of great interest; from adapting food-industry hydrocolloids for fne-dining applications like gels, fuid gels, foams/espumas, and “spherifcation” (Myhrvold et al 2011, Achatz and Kokonas 2008, Adria et al 2006), to interest in craf cocktails and other alcoholic beverages (Adria et al 2006, Conigliaro 2012), to current interest in traditional chemical processes such as nixtamalization and microbial processing by lactic acid bacteria, yeasts, acetic bacteria, and molds and fungi such as Aspergillus oryzae (Chang et al 2014, Chang 2010, Katz 2012, Nilsson 2012, Felder et al 2012).

Te development and implementation of many of these processes and products for the kitchen has frequently been directed by the goals of creating, capturing, and controlling favor. A theoretical, experimental, or technical approach to favor explicitly has also been the focus of a number of recent cookbooks (Kunz 2008, Page 2008, Segnit 2012, Afel and Patterson 2004). Some observations and theories on favor originating in the culinary world have become the basis for peer-reviewed work, refecting the potential for further collaboration on questions of favor. In one such example, the observation by a chef13 of more intense umami favor in the pulp and seeds

13 Heston Blumenthal of the Fat Duck 23 of tomatoes than the fesh led to a collaboration with favor chemists at the University of Reading that measured higher levels, on average, of the umami-tasting compounds glutamic acid, aspartic acid, and several 5’-ribonucleotides in the pulp than the fesh of several diferent tomato varieties (Oruna-Concha et al 2007). In another case, questions about why ingredients pair together well in terms of favor led to a creative theory, developed by the same chef working with favorists, which stated that new favor pairings of ingredients not commonly used together in one dish could be identifed by comparing their volatile profles, and that shared supra-threshold volatiles would allow for a harmonious pairing (Blumenthal 2002, 2008). Some investigation of the mechanisms underlying paired favors by a sensory science team found that the rated pleasantness of binary odor mixtures is not solely determined by the rated pleasantness of individual components, but did not measure overlapping compounds (Møller et al 2011).

Other work on favor combinations has been more methodologically fawed. A data- mining experiment of recipes collected from three internet repositories concluded that North American recipes tended to pair ingredients that shared favor compounds, while East Asian recipes did not tend to pair ingredients that share favor compounds (Ahn et al 2011). However, the authors do not describe any experimental design or controls on bias for collecting the 56,000 recipes in the dataset, and state that the largeness of the dataset allows for factors relating to artistic creativity, ingredients (such as egg) included for structural rather than favor properties, and favors arising from cooking method rather than measurable in raw ingredients to be systematically fltered out, though they do not describe a method for this fltering. Te authors describe the most “authentic” ingredients in North American cuisine (“authenticity” defned as used more ofen in North American recipes in the dataset than other region’s recipes) to be milk, butter, vanilla, egg, molasses, and wheat, which would seem to suggest that North American cuisine is defned almost entirely by dairy and desserts.

Issues with false positives and with conclusions drawn from analysis of large datasets without taking into account current knowledge from small, well-designed experiments in the feld

24 have been noted in other disciplines such as evolutionary biology, genomics, and medicine (Graur et al 2013, Alberts 2012, PLoS Medicine Editors 2005); Ahn et al may be doing something similar in drawing conclusions about the prevalent recipes and ingredients that make up the essence of a culture’s cuisine, favor chemistry, and cooking, without citing references from the felds of food cultural anthropology, favor chemistry, or gastronomy in any manner more in-depth than passing. Given that as of April 2014 it has been cited 30 times, and an analysis by Altmetric puts it in the top 5% of all articles in terms of social and traditional media attention, the high interest in favor and chemistry as it relates to cooking and cuisine is an opening, even a demand, for more rigorously designed favor chemistry studies of gastronomically interesting products and ideas.

Snitkjaer et al. (2010 and 2011) provide one example of how this can be done: both papers are singular in their combined use of established experimental methods and explicit gastronomic context. Beyond this this, the application of sensory and chemical profling, shown to provide valuable insight for so many other food products, has been applied infrequently to these new and resurgent products and processes. Several high-profle calls (Dufresne 2012, Barber 2012, Myhrvold et al 2011), for greater chef-scientist involvement reveal a latent demand for science that addresses cuisine and culinary questions explicitly, and research characterizing the favor chemistry of ingredients and culinary processes builds on both this demand and the nascent body of scientifc work addressing food in a culinary context, under the heading of “molecular gastronomy” or otherwise.

With the above as context, the research covered in this dissertation is particularly interested in characterizing the favor chemistry of products of cultural and gastronomic interest, particularly those in current use that have not received much scientifc attention. In particular, I will focus on aromatic cocktail bitters and uses of acetic fermentation to produce novel vinegars.

Bitters, highly concentrated alcoholic extractions of favorful plant materials, have become popular as an ingredient for delivering favors in craf cocktails (Sandham 2012, Soole 2013), with many well-respected cocktail bars stocking dozens or more diferent styles for specifc

25 favor profles (Parsons 2011). Bitters were in widespread use in the 19th and early 20th centuries (Wondrich 2007, Haigh 2009) and have recently experienced a resurgence, with homemade and house-made recipes becoming internationally distributed brands (Regan 2003, Parsons 2011). Teir use refects a growing interest in application for favor extraction in a culinary environment (Conigliaro 2013, Liu 2013). Despite this interest, there is essentially no data available on their favor or favor chemistry.

Transformation of alcoholic liquids into those containing acetic acid (vinegars) by aerobic bacteria has been a process used in various cuisines for thousands of years (Mazza et al 2009). More recently, the process of acetic “fermentation”14 has become a popular tool in the kitchens of a number of high-end and experimental restaurants (Porcelli 2013, Redzepi 2013, Nilsson 2012, Reade 2012) for preserving seasonal favors, transforming the favor of ingredients not widely made into vinegar (such as mushroom juice or spruce needles), and creating culinary sources of acidity in locations where citrus fruits are not grown. Te favor chemistry of traditional vinegars, such sherry vinegar, Balsamic vinegar, and Shanxi buckwheat vinegar, have been characterized (Callejon et al 2008, Cirlini et al 2011, Aili et al 2010), but the products of this new interest in making novel vinegars in the restaurant world have not until now been studied. In collaboration with the research and development staf at the Nordic Food Lab, novel vinegars produced using rapid, forced-air acetifcation methods and passive acetifcation are analyzed.

Methods designed to address the complex relationship between chemical makeup and

perceived favor are also used and developed, including multivariate statistical techniques and in- instrument gas chromatography methods for recomposition olfactometry experiments.

14 Biochemically speaking, not technically fermentation, as transformation of ethyl alcohol into acetic acid is an aerobic process; however, from this point forward I will use the common convention of describing this process as acetic fermentation, without the use of quotation marks 26 References Aaslyng, M., & Frøst, M. (2010). Te efect of the combination of salty, bitter, and sour accompaniment on the favor and juiciness of pork patties. Journal of Sensory Studies, 25(4), 536–548. doi:10.1111/j.1745-459X.2010.00285.x Abdi, H. (2003). Multivariate Analysis. In M. Lewis-Beck, A. Bryman, & T. Futing (Eds.), Encyclopedia of Social Sciences Research Methods (pp. 1–4). Tousand Oaks, CA: Sage. Abdi, H., Toole, A. J. O., Valentin, D., & Edelman, B. (2005). DISTATIS : Te Analysis of Multiple Distance Matrices. In C. Schmid, S. Soatto, & C. Tomasi (Eds.), Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (p. 42). Los Alamitos, CA: IEEE Computer Society. Abdi, H., Valentin, D., Chollet, S., & Chrea, C. (2007). Analyzing assessors and products in sorting tasks: DISTATIS, theory and applications. Food Quality and Preference, 18(4), 627– 640. doi:10.1016/j.foodqual.2006.09.003 Abdi, H., & Williams, L. J. (2010). Principal Component Analysis. Wiley Interdisciplinary Reviews: Computational Statistics, 2(4), 433–459. Abend, L. (2012, March 26). Locavore Hero. Time Magazine, p. Cover. Aceña, L., Vera, L., Guasch, J., Busto, O., & Mestres, M. (2010). Comparative study of two extraction techniques to obtain representative aroma extracts for being analysed by gas chromatography-olfactometry: application to roasted pistachio aroma. Journal of chromatography. A, 1217(49), 7781–7787. doi:10.1016/j.chroma.2010.10.030 Achatz, G., & Kokonas, N. (2008). Alinea. Ten Speed Press. Retrieved from http://books.google. com/books?id=y59tuJs4v5QC Acree, T. E., & Barnard, J. (1984). A Procedure for the Sensory Analysis of Gas Chromatographic Efuents. Food Chemistry, 14, 273–286. Adria, F, Soler, J., & Adria, A. (2005). El Bulli: 1998-2002. New York, NY: HarperCollins. Retrieved from http://books.google.com/books?id=ofvwGwAACAAJ Adria, F, Soler, J., & Adria, A. (2006a). El Bulli 2003-2004. New York, NY: HarperCollins. Retrieved from http://books.google.com/books?id=MXarnQEACAAJ Adria, F, Soler, J., & Adria, A. (2006b). El Bulli 1994-1997. New York, NY: HarperCollins. Retrieved from http://books.google.com/books?id=6PCrdGUUyz4C Adrià, F., Soler, J., & Adrià, A. (2006). El bulli: 2005. RBA Libros, Barcelona. Retrieved from http://books.google.com/books?id=5ihfAAAACAAJ Adria, F., Blumenthal, H., Keller, T., & McGee, H. (2006, December 9). Statement on the “new cookery.” Te Observer. Retrieved from http://www.theguardian.com/uk/2006/dec/10/ foodanddrink.obsfoodmonthly Aduriz, A. L. (2012). Mugaritz: A Natural Science of Cooking. London: Phaidon Press. Retrieved from http://books.google.com/books?id=Uvf-ygAACAAJ

27 Afoakwa, E. O., Paterson, A., Fowler, M., & Ryan, A. (2008). Flavor formation and character in cocoa and chocolate: a critical review. Critical reviews in food science and nutrition, 48(9), 840–57. doi:10.1080/10408390701719272 Afel, M., & Patterson, D. (2004). Aroma. New York, NY: Artisan. Retrieved from http://books. google.com/books?id=msEMk17upAMC Ahn, Y.-Y., Ahnert, S. E., Bagrow, J. P., & Barabási, A.-L. (2011). Flavor network and the principles of food pairing. Scientifc reports, 1, 196. doi:10.1038/srep00196 Aili, W., Huanlu, S., & Zaigui, L. (2012). Key aroma compounds in Shanxi aged tartary buckwheat vinegar and changes during its thermal processing. Flavour and Fragrance Journal, (January 2011), 47–53. doi:10.1002/f.2079 Alberts, B. (2012). Te End of “ Small Science ”? Science, 337(September), 1583. Amerine, M. A., Pangborn, R. M., Roessler, E. B., & others. (1965). Principles of sensory evaluation of food. Principles of sensory evaluation of food. New York, NY: Academic Press, Inc.,. An, M., Hai, T., & Hatfeld, P. (2001). On-site feld sampling and analysis of fragrance from living lavender (Lavandula angustifolia L.) fowers by solid-phase microextraction coupled to gas chromatography and ion-trap mass spectrometry. Journal of chromatography. A, 917(1-2), 245–50. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/11403476 Andrews, C. (2011). Ferran: Te Inside Story of El Bulli and the Man Who Reinvented Food. New York: Penguin Group US. Retrieved from http://books.google.com/books?id=k- KAn8krJyUC Arboleya, J.-C., Olabarrieta, I., Luis-Aduriz, A., Lasa, D., Vergara, J., Sanmartín, E., … Martínez de Marañón, I. (2008). From the Chef’s Mind to the Dish: How Scientifc Approaches Facilitate the Creative Process. Food Biophysics, 3(2), 261–268. doi:10.1007/s11483-008- 9078-3 Arnold, D. (2006, June). What About Tis. Food Arts, 31–33. Atala, A. (2013). D.O.M.: Rediscovering Brazilian Ingredients. London: Phaidon Press. Retrieved from http://books.google.com/books?id=vNlWmQEACAAJ Atala, Alex. (2012). A new ingredient: Te introduction of priprioca in gastronomy. International Journal of Gastronomy and Food Science, 1(1), 61–63. doi:10.1016/j.ijgfs.2011.11.001 Authors, T. Pl. M. (2005). Why bigger is not yet better: the problems with huge datasets. PLoS medicine, 2(2), e55. doi:10.1371/journal.pmed.0020055 Auvray, M., & Spence, C. (2008). Te multisensory perception of favor. Consciousness and cognition, 17(3), 1016–31. doi:10.1016/j.concog.2007.06.005 Axel, R. (2005). Scents and sensibility: a molecular logic of olfactory perception (Nobel lecture). Angewandte Chemie (International ed. in English), 44(38), 6110–27. doi:10.1002/ anie.200501726 Aznar, M., López, R., Cacho, J., & Ferreira, V. (2003). Prediction of aged red wine aroma properties from aroma chemical composition. Partial least squares regression models. Journal of agricultural and food chemistry, 51(9), 2700–7. doi:10.1021/jf026115z 28 Barber, D. (2012). Te Taste of Wheat. Mad Symposium, July 2 2012, Copenhagen. Retrieved October 04, 2014, from http://www.madfood.co/dan-barber-2/ Barham, P., Skibsted, L. H., Bredie, W. L. P., Frøst, M. B., Møller, P., Risbo, J., … Mortensen, L. M. (2010). Molecular gastronomy: a new emerging scientifc discipline. Chemical reviews, 110(4), 2313–65. doi:10.1021/cr900105w Barkat, S., Le Berre, E., Coureaud, G., Sicard, G., & Tomas-Danguin, T. (2011). Perceptual Blending in Odor Mixtures Depends on the Nature of Odorants and Human Olfactory Expertise. Chemical senses, (1994), 159–166. doi:10.1093/chemse/bjr086 Bentley, R. (2006). Te nose as a stereochemist. Enantiomers and odor. Chemical reviews, 106(9), 4099–4112. doi:10.1021/cr050049t Berglund, B., & Olsson, M. J. (1993). Odor-intensity interaction in binary and ternary mixtures. Perception & psychophysics, 53(5), 475–82. Retrieved from http://www.ncbi.nlm.nih.gov/ pubmed/8332416 Bicchi, C., Iori, C., Rubiolo, P., & Sandra, P. (2002). Headspace Sorptive Extraction ( HSSE ), Stir Bar Sorptive Extraction ( SBSE ), and Solid Phase Microextraction ( SPME ) Applied to the Analysis of Roasted Arabica Cofee and Cofee Brew. Journal of agricultural and food chemistry, 50(3), 449–459. Blumenthal, H. (2006). Heston Blumenthal: In Search of Perfection: Reinventing Kitchen Classics. New York, NY: Bloomsbury USA. Retrieved from http://books.google.com/ books?id=GcdmngEACAAJ Blumenthal, H. (2013). Historic Heston. New York, NY: Bloomsbury USA. Retrieved from http:// books.google.com/books?id=bQqcAQAAQBAJ Blumenthal, H, Lane, T., & Sewell, A. (2010). Heston’s Fantastical Feasts. New York, NY: Bloomsbury USA. Retrieved from http://books.google.com/books?id=xQ2IRAAACAAJ Blumenthal, Heston. (2002, May 3). Weird but wonderful. Te Guardian. London. Retrieved from http://www.guardian.co.uk/lifeandstyle/2002/may/04/foodanddrink.shopping Bushdid, C., Magnasco, M. O., Vosshall, L. B., & Keller, a. (2014). Humans Can Discriminate More than 1 Trillion Olfactory Stimuli. Science, 343(6177), 1370–1372. doi:10.1126/ science.1249168 Cairncross, S. E., & Sjostrom, L. B. (1950). Flavor profles: a new approach to favor problems. In M. Gakula (Ed.), Descriptive Sensory Analysis in Practice. Trumbull, CT: Food & Nutrition Press. Caldeira, M., Rodrigues, F., Perestrelo, R., Marques, J. C., & Câmara, J. S. (2007). Comparison of two extraction methods for evaluation of volatile constituents patterns in commercial whiskeys Elucidation of the main odour-active compounds. Talanta, 74(1), 78–90. doi:10.1016/j.talanta.2007.05.029 Callejón, R. M., Morales, M. L., Ferreira, A. C. S., & Troncoso, A. M. (2008). Defning the typical aroma of sherry vinegar: sensory and chemical approach. Journal of agricultural and food chemistry, 56(17), 8086–8095. doi:10.1021/jf800903n

29 Capron, X., Smeyersverbeke, J., & Massart, D. (2007). Multivariate determination of the geographical origin of wines from four diferent countries. Food Chemistry, 101(4), 1585– 1597. doi:10.1016/j.foodchem.2006.04.019 Cerretani, L., Biasini, G., Bonoli-Carbognin, M., & Bendini, A. (2007). Harmony of virgin olive oil and food pairing: a methodological proposal. Journal of Sensory Studies, 22(4), 403–416. doi:10.1111/j.1745-459X.2007.00115.x Chang, D., & Meehan, P. (2010). Momofuku. New York, NY: Crown Publishing Group. Retrieved from http://books.google.com/books?id=ovwudzKRRKQC Chang, D., Meehan, P., & Ying, C. (2014). Lucky Peach Issue 10: Street Food Issue. New York, NY: Lucky Peach. Retrieved from http://books.google.com/books?id=pWk3nwEACAAJ Chollet, S., Lelièvre, M., Abdi, H., & Valentin, D. (2011). Sort and beer: Everything you wanted to know about the sorting task but did not dare to ask. Food Quality and Preference, 22(6), 507–520. doi:10.1016/j.foodqual.2011.02.004 Cirlini, M., Caligiani, a., Palla, L., & Palla, G. (2011). HS-SPME/GC–MS and chemometrics for the classifcation of Balsamic Vinegars of Modena of diferent maturation and ageing. Food Chemistry, 124(4), 1678–1683. doi:10.1016/j.foodchem.2010.07.065 Civille, G. V., & Lyon, B. G. (1996). Aroma and favor lexicon for sensory evaluation: terms, defnitions, references, and examples. West Conshohocken, PA: ASTM. Conigliaro, T. (2013). Te Cocktail Lab: Unraveling the Mysteries of Flavor and Aroma in Drink, With Recipes. Ten Speed Press. Retrieved from http://books.google.com/ books?id=vF3RYlosi80C Conigliaro, Tony. (2012). Q&A: the science of cocktails. Flavour, 1(1), 19. doi:10.1186/2044-7248- 1-19 De Mello Castanho Amboni, R. D., da Silva Junkes, B., Yunes, R. a, & Heinzen, V. E. (2000). Quantitative structure-odor relationships of aliphatic esters using topological indices. Journal of agricultural and food chemistry, 48(8), 3517–21. Retrieved from http://www.ncbi. nlm.nih.gov/pubmed/10956142 Domrachev, D. V., Karpova, E. V., Goroshkevich, S. N., & Tkachev, a. V. (2012). Comparative analysis of volatiles from needles of fve-needle pines of northern and eastern Eurasia. Russian Journal of Bioorganic Chemistry, 38(7), 780–789. doi:10.1134/S1068162012070059 Donadini, G., Fumi, M. D., & Lambri, M. (2012). Te hedonic response to chocolate and beverage pairing: A preliminary study. Food Research International, 48(2), 703–711. doi:http://dx.doi. org/10.1016/j.foodres.2012.06.009 Donadini, G., Fumi, M. D., & Lambri, M. (2013). A preliminary study investigating consumer preference for cheese and beer pairings. Food Quality and Preference, 30(2), 217–228. doi:http://dx.doi.org/10.1016/j.foodqual.2013.05.012 Dufresne, W. (2012). An Appetite for Knowledge. Mad Symposium, July 2 2012, Copenhagen. Retrieved October 04, 2014, from http://www.madfood.co/wylie-dufresne/

30 Ebeler, S. E., & Torngate, J. H. (2009). Wine chemistry and favor: looking into the crystal glass. Journal of agricultural and food chemistry, 57(18), 8098–8108. doi:10.1021/jf9000555 Edelstein, S. (2013). Food Science: An Ecological Approach. Sudbury, MA: Jones & Bartlett Learning. Retrieved from http://books.google.com/books?id=aDI5Bmn7TRQC Edwards, P. a., Anker, L. S., & Jurs, P. C. (1991). Quantitative structure-property relationship studies of the odor threshold of odor active compounds. Chemical Senses, 16(5), 447–465. doi:10.1093/chemse/16.5.447 Fan, W., Xu, Y., & Han, Y. (2011). Quantifcation of Volatile Compounds in Chinese Ciders by Stir Bar Sorptive Extraction (SBSE) and Gas Chromatography-Mass Spectrometry (GC-MS). Journal of the Institute of Brewing, 117(1), 61–66. doi:10.1002/j.2050-0416.2011.tb00444.x Felder, D., Burns, D., & Chang, D. (2012). Defning microbial terroir: Te use of native fungi for the study of traditional fermentative processes. International Journal of Gastronomy and Food Science, 1(1), 64–69. doi:10.1016/j.ijgfs.2011.11.003 Ferraces-Casais, P., Lage-Yusty, M. A., Rodríguez-Bernaldo de Quirós, A., & López-Hernández, J. (2013). Rapid identifcation of volatile compounds in fresh seaweed. Talanta, 115, 798–800. doi:10.1016/j.talanta.2013.06.049 Floriano, W. B., Vaidehi, N., Goddard, W. a, Singer, M. S., & Shepherd, G. M. (2000). Molecular mechanisms underlying diferential odor responses of a mouse olfactory receptor. Proceedings of the National Academy of Sciences of the United States of America, 97(20), 10712–6. Retrieved from http://www.pubmedcentral.nih.gov/articlerender. fcgi?artid=27088&tool=pmcentrez&rendertype=abstract Giboreau, A., Dacremont, C., Egorof, C., Guerrand, S., Urdapilleta, I., Candel, D., & Dubois, D. (2007). Defning sensory descriptors: Towards writing guidelines based on terminology. Food quality and preference, 18(2), 265–274. Graur, D., Zheng, Y., Price, N., Azevedo, R. B. R., Zufall, R. a, & Elhaik, E. (2013). On the immortality of television sets: “function” in the human genome according to the evolution- free gospel of ENCODE. Genome biology and evolution, 1–43. doi:10.1093/gbe/evt028 Grosch, W. (2001). Evaluation of the key odorants of foods by dilution experiments, aroma models and omission. Chemical senses, 26, 533–545. Retrieved from http://www.ncbi.nlm. nih.gov/pubmed/11418500 Grosch, Werner. (1993). Detection of potent odorants in foods by aroma extract dilution analysis. Trends in Food Science & Technology, 4, 68–73. Haenlein, M., & Kaplan, A. M. (2004). A Beginner’s Guide to Partial Least Squares Analysis. Understanding Statistics, 3(4), 283–297. doi:10.1207/s15328031us0304_4 Haigh, T. (2009). Vintage Spirits and Forgotten Cocktails: From the Alamagoozlum to the Zombie 100 Rediscovered Recipes and the Stories Behind Tem. Beverly, MA: Quarry Books. Retrieved from http://books.google.com/books?id=sCR7wWhM7IQC

31 Harrington, R. J., & Hammond, R. (2006). Te Direct Efects of Wine and Cheese Characteristics on Perceived Match. Journal of Foodservice Business Research, 8(4), 37–54. doi:10.1300/ J369v08n04 Harrington, R. J., McCarthy, M., & Gozzi, M. (2010). Perceived Match of Wine and Cheese and the Impact of Additional Food Elements: A Preliminary Study. Journal of Foodservice Business Research, 13(4), 311–330. doi:10.1080/15378020.2010.524541 Hein, K., Ebeler, S. E., & Heymann, H. (2009). Perception of Fruity and Vegetative Aromas in Red Wine. Journal of Sensory Studies, 24(3), 441–455. doi:10.1111/j.1745-459X.2009.00220.x Heymann, H., Hopfer, H., & Bershaw, D. (2013). An Exploration of the Perception of Minerality in White Wines by Projective Mapping and Descriptive Analysis. Journal of Sensory Studies, 29(1), 1–13. doi:10.1111/joss.12076 Hjelmeland, A. K., King, E. S., Ebeler, S. E., & Heymann, H. (2012). Characterizing the Chemical and Sensory Profles of United States Cabernet Sauvignon Wines and Blends. American Journal of Enology and Viticulture, 64(2), 169–179. doi:10.5344/ajev.2012.12107 Hong, S., & Corey, E. J. (2006). Enantioselective syntheses of georgyone, arborone, and structural relatives. Relevance to the molecular-level understanding of olfaction. Journal of the American Chemical Society, 128(4), 1346–52. doi:10.1021/ja057483x Jinks, A., & Laing, D. G. (1999). A limit in the processing of components in odour mixtures. Perception, 28(3), 395–404. doi:10.1068/p2898 Kaack, K., Christensen, L. P., Hughes, M., & Eder, R. (2006). Relationship between sensory quality and volatile compounds of elderfower (Sambucus nigra L.) extracts. European Food Research and Technology, 223(1), 57–70. doi:10.1007/s00217-005-0122-y Kamozawa, A., & Talbot, H. A. (2010). Ideas in Food: Great Recipes and Why Tey Work. New York, NY: Crown Publishing Group. Retrieved from http://books.google.com/ books?id=GKgyvKxI4DcC Kanavouras, A., Kiritsakis, A., & Hernandez, R. J. (2005). Comparative study on volatile analysis of extra virgin olive oil by dynamic headspace and solid phase micro-extraction. Food Chemistry, 90, 69–79. doi:10.1016/j.foodchem.2004.03.025 Katz, S. E., & Pollan, M. (2012). Te Art of Fermentation: An In-Depth Exploration of Essential Concepts and Processes from around the World. White River Junction, VT: Chelsea Green Publishing. Retrieved from http://books.google.com/books?id=-zmLa205d0QC King, E. S., Dunn, R. L., & Heymann, H. (2013). Te infuence of alcohol on the sensory perception of red wines. Food Quality and Preference, 28(1), 235–243. doi:10.1016/j. foodqual.2012.08.013 Laing, D. G., & Francis, G. W. (1989). Te capacity of humans to identify odors in mixtures. Physiology & behavior, 46, 809–814. Retrieved from http://www.ncbi.nlm.nih.gov/ pubmed/2628992

32 Laing, D. G., Legha, P. K., Jinks, a L., & Hutchinson, I. (2003). Relationship between molecular structure, concentration and odor qualities of oxygenated aliphatic molecules. Chemical senses, 28(1), 57–69. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/12502524 Lawless, H. T., & Heymann, H. (2010a). Data Relationships and Multivariate Applications. In Sensory Evaluation of Food (pp. 433–449). Berlin: Springer Science+Business Media. doi:10.1007/978-1-4419-6488-5 Lawless, H. T., & Heymann, H. (2010b). Descriptive Analysis. In Sensory Evaluation of Food (pp. 227–257). Berlin: Springer Science+Business Media. doi:10.1007/978-1-4419-6488-5 Le Berre, E, Béno, N., Ishii, A., Chabanet, C., Etiévant, P., & Tomas-Danguin, T. (2008). Just noticeable diferences in component concentrations modify the odor quality of a blending mixture. Chemical senses, 33(4), 389–395. doi:10.1093/chemse/bjn006 Le Berre, Elodie, Tomas-Danguin, T., Béno, N., Coureaud, G., Etiévant, P., & Prescott, J. (2008). Perceptual processing strategy and exposure infuence the perception of odor mixtures. Chemical senses, 33(2), 193–199. doi:10.1093/chemse/bjm080 Lee, S, & Ahn, B. (2009). Comparison of volatile components in fermented soybean pastes using simultaneous distillation and extraction (SDE) with sensory characterisation. Food Chemistry, 114(2), 600–609. doi:10.1016/j.foodchem.2008.09.091 Lee, Seung-joo, & Noble, A. C. (2006). Use of Partial Least Squares Regression and Multidimensional Scaling on Aroma Models of California Chardonnay Wines. American Journal of Enology and Viticulture, 57(3), 363–370. Lelièvre, M., Chollet, S., Abdi, H., & Valentin, D. (2008). What is the validity of the sorting task for describing beers? A study using trained and untrained assessors. Food Quality and Preference, 19(8), 697–703. doi:10.1016/j.foodqual.2008.05.001 Liu, K. (2013). Craf Cocktails at Home: Ofeat Techniques, Contemporary Crowd-Pleasers, and Classics Hacked with Science. Seattle, WA: Amazon Digitla Services. Retrieved from http:// books.google.com/books?id=OOsX44gYoo8C Livermore, a, & Laing, D. G. (1996). Infuence of training and experience on the perception of multicomponent odor mixtures. Journal of experimental psychology. Human perception and performance, 22(2), 267–77. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/8934843 Livermore, A., & Laing, D. G. (1998). Te infuence of chemical complexity on the perception of multicomponent odor mixtures. Perception & Psychophysics, 60(4), 650–661. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/9628996 Lizarraga-Guerra, R., Guth, H., & Lopez, M. G. (1997). Identifcation of the Most Potent Odorants in Huitlacoche ( Ustilago maydis ) and Austern Pilzen ( Pleurotus sp .) by Aroma Extract Dilution Analysis and Static Head-Space Samples. Journal of agricultural and food chemistry, 45, 1329–1332. Lopez, R., Aznar, M., Cacho, J., & Ferreira, V. (2002). Determination of minor and trace volatile compounds in wine by solid-phase extraction and gas chromatography with mass spectrometric detection. Journal of Chromatography A, 966, 167–177.

33 Madrigal-galan, B., & Heymann, H. (2006). Sensory Efects of Consuming Cheese Prior to Evaluating Red Wine Flavor. American Journal of Enology and Viticulture, 1, 12–22. Malherbe, S., Watts, V., Nieuwoudt, H. H., Bauer, F. F., & du Toit, M. (2009). Analysis of volatile profles of fermenting grape must by headspace solid-phase dynamic extraction coupled with gas chromatography-mass spectrometry (HS-SPDE GC-MS): novel application to investigate problem fermentations. Journal of agricultural and food chemistry, 57(12), 5161– 6. doi:10.1021/jf900532v Malnic, B., Hirono, J., Sato, T., & Buck, L. B. (1999). Combinatorial receptor codes for odors. Cell, 96(5), 713–23. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/10089886 Martin, N., Berger, C., Du, C. Le, & Spinnler, H. E. (2001). Aroma Compound Production in Cheese Curd by Coculturing with Selected Yeast and Bacteria, 2125–2135. Mazza, S., & Murooka, Y. (2009). Vinegars Trough the Ages. In L. Solieri & P. Giudici (Eds.), Vinegars of the World (pp. 18–39). Springer Milan. McGee, H. (1999). Taking stock of new favours. Nature, 400(6739), 17–8. doi:10.1038/21775 Mielby, L. H., & Frøst, M. B. (2010). Expectations and surprise in a molecular gastronomic meal. Food Quality and Preference, 21(2), 213–224. doi:10.1016/j.foodqual.2009.09.005 Mo, X., Xu, Y., & Fan, W. (2010). Characterization of aroma compounds in Chinese rice wine Qu by solvent-assisted favor evaporation and headspace solid-phase microextraction. Journal of agricultural and food chemistry, 58(4), 2462–9. doi:10.1021/jf903631w Møller, P., Hartvig, D., & Bredie, W. (2011). Determinants of the pleasantness of odor mixtures. Chemical Senses, 36, E71–E72. doi:10.1093/chemse/bjq126 Mouritsen, O G, Johansen, M., & Mouritsen, J. D. (2013). Seaweeds: Edible, Available, and Sustainable. Chicago, IL: University of Chicago Press. Retrieved from http://books.google. com/books?id=sLDb13RjmHcC Mouritsen, Ole G, Williams, L., Bjerregaard, R., & Duelund, L. (2012). Seaweeds for umami favour in the New Nordic Cuisine. Flavour, 1(1), 4. doi:10.1186/2044-7248-1-4 Murray, J. M., & Delahunty, C. M. (2000). Selection of standards to reference terms in a cheddar- type cheese favor language. Journal of Sensory Studies, 15(2), 179–199. Myhrvold, N. (2011). Te Art in Gastronomy: A Modernist Perspective. Gastronomica, 11(1), 13–23. Nilsson, M. (2012). Fäviken. London: Phaidon Press. Retrieved from http://books.google.com/ books?id=JvKxuAAACAAJ Noble, A. C., & Ebeler, S. E. (2002). Use of multivariate statistics in understanding wine favor. Food Reviews International, 18(1), 1–20. doi:10.1081/FRI-120003414 Nygren, I T, Gustafsson, I., & Johansson, L. (2002). Editorial review Perceived favour changes in white wine afer tasting blue mould cheese. Food Service Technology, 2, 163–171.

34 Nygren, I.T., Gustafsson, I., Haglund, Å., Johansson, L., & Noble, a C. (n.d.). Flavor changes produced by wine and food interactions: Chardonnay wine and hollandaise sauce. Journal of Sensory Studies, 16(2001), 461–470. Nygren, Ingemar T, Gustafsson, I., & Johansson, L. (2003). Perceived favour changes in blue mould cheese afer tasting. Food Service Technology, 143–150. Oruna-Concha, M.-J., Methven, L., Blumenthal, H., Young, C., & Mottram, D. S. (2007). Diferences in glutamic acid and 5’-ribonucleotide contents between fesh and pulp of tomatoes and the relationship with umami taste. Journal of agricultural and food chemistry, 55(14), 5776–5780. doi:10.1021/jf070791p Page, K., & Dornenburg, A. (2008). Te Flavor Bible: Te Essential Guide to Culinary Creativity, Based on the Wisdom of America’s Most Imaginative Chefs. New York, NY: Little, Brown and Co. Parsons, B. T., & Anderson, E. (2011). Bitters: A Spirited History of a Classic Cure-All, with Cocktails, Recipes, and Formulas. Berkeley, CA: Ten Speed Press. Retrieved from http:// books.google.com/books?id=OPlRDjfGnloC Patton, S., & Josephson, D. (1957). A method for determining signifcance of volatile favor compounds in foods. Journal of Food Science, 22(3), 316–318. Paulsen, M. T., Næs, T., Ueland, Ø., Rukke, E.-O., & Hersleth, M. (2013). Preference mapping of salmon–sauce combinations: Te infuence of temporal properties. Food Quality and Preference, 27(2), 120–127. doi:http://dx.doi.org/10.1016/j.foodqual.2012.09.010 Paulsen, M. T., Ueland, Ø., Nilsen, A. N., Öström, Å., & Hersleth, M. (2012). Sensory perception of salmon and culinary sauces – An interdisciplinary approach. Food Quality and Preference, 23(2), 99–109. doi:http://dx.doi.org/10.1016/j.foodqual.2011.09.004 Pineau, B., Barbe, J.-C., Van Leeuwen, C., & Dubourdieu, D. (2007). Which impact for beta- damascenone on red wines aroma? Journal of agricultural and food chemistry, 55(10), 4103– 4108. doi:10.1021/jf070120r Pineau, B., Barbe, J.-C., Van Leeuwen, C., & Dubourdieu, D. (2009). Examples of perceptive interactions involved in specifc “red-” and “black-berry” aromas in red wines. Journal of agricultural and food chemistry, 57(9), 3702–8. doi:10.1021/jf803325v Poinot, P., Grua-Priol, J., Arvisenet, G., Rannou, C., Semenou, M., Le Bail, A., & Prost, C. (2007). Optimisation of HS-SPME to study representativeness of partially baked bread odorant extracts. Food Research International, 40(9), 1170–1184. doi:10.1016/j.foodres.2007.06.011 Polásková, P., Herszage, J., & Ebeler, S. E. (2008). Wine favor: chemistry in a glass. Chemical Society reviews, 37(11), 2478–89. doi:10.1039/b714455p Porcelli, A. (2013). Cook it Raw. London: Phaidon Press. Retrieved from http://books.google.com/ books?id=iaFmngEACAAJ Powers, J. J. (1988). Current practices and application of descriptive methods. In J. Piggott (Ed.), Sensory analysis of foods (pp. 187–266). London: Elsevier Applied Science.

35 Preston, L. D., Block, D. E., Heymann, H., Soleas, G., Noble, A. C., & Ebeler, S. E. (2008). Defning Vegetal Aromas in Cabernet Sauvignon Using Sensory and Chemical Evaluations. American Journal of Enology and Viticulture, 59(2), 137–145. Pryde, J., Conner, J., Jack, F., Lancaster, M., Meek, L., Paterson, R. (2011). Sensory and Chemical Analysis of “Shackleton’s” Mackinlay Scotch Whisky. Journal of the Institute of Brewing, 117(2), 156–165. Redzepi, R. (2010). Noma: Time and Place in Nordic Cuisine. London: Phaidon Press. Retrieved from http://books.google.com/books?id=z3jyRQAACAAJ Redzepi, R. (2013). A Work in Progress. London: Phaidon Press. Retrieved from http://books. google.com/books?id=njlOnwEACAAJ Regan, G. (2003). Te Joy of Mixology. New York, NY: Crown Publishing Group. Retrieved from http://books.google.com/books?id=7TlhtrpXa-MC Ribeiro, J. S., Augusto, F., Salva, T. J. G., Tomaziello, R. A., & Ferreira, M. M. C. (2009). Prediction of sensory properties of Brazilian Arabica roasted cofees by headspace solid phase microextraction-gas chromatography and partial least squares. Analytica chimica acta, 634(2), 172–9. doi:10.1016/j.aca.2008.12.028 Rødbotten, M., Lea, P., & Ueland, Ø. (2009). Quality of raw salmon fllet as a predictor of cooked salmon quality. Food Quality and Preference, 20(1), 13–23. doi:10.1016/j. foodqual.2008.06.004 Rossiter, K. J. (1996). Structure − Odor Relationships. Chemical Reviews, 96(8), 3201–3240. Ryan, D., Prenzler, P. D., Saliba, A. J., & Scollary, G. R. (2008). Te signifcance of low impact odorants in global odour perception. Trends in Food Science & Technology, 19(7), 383–389. doi:10.1016/j.tifs.2008.01.007 Sanchez, J. V. S. G. (2011). Comparison of Descriptive Analysis and Projective Mapping Techniques in the Aroma Evaluation of the Distilled Spirits , Gin and Tequila. Master’s Tesis, University of California, Davis. 87 pp. Sandham, T. (2012). World’s Best Cocktails: 500 Signature Drinks from the World’s Best Bars and Bartenders. Minneapolis, MN: Fair Winds Press. Retrieved from http://books.google.com/ books?id=bQQNAAAAQBAJ Sanz, G., Tomas-Danguin, T., Hamdani, E. H., Le Poupon, C., Briand, L., Pernollet, J.-C., … Tromelin, A. (2008). Relationships between molecular structure and perceived odor quality of ligands for a human olfactory receptor. Chemical senses, 33(7), 639–53. doi:10.1093/ chemse/bjn032 Schoeller, M. (2013, November 18). Gods of Food. Time Magazine, p. Cover. Segnit, N. (2010). Te Flavor Tesaurus: A Compendium of Pairings, Recipes and Ideas for the Creative Cook. New York: Bloomsbury. Small, D. M., Bender, G., Veldhuizen, M. G., Rudenga, K., Nachtigal, D., & Felsted, J. (2007). Te role of the human orbitofrontal cortex in taste and favor processing. Annals of the New York Academy of Sciences, 1121, 136–51. doi:10.1196/annals.1401.002 36 Smith, A., & Kraig, B. (2013). Te Oxford Encyclopedia of Food and Drink in America. New York: Oxford University Press. Retrieved from http://books.google.com/ books?id=DOJMAgAAQBAJ Snitkjær, P., Frøst, M. B., Skibsted, L. H., & Risbo, J. (2010). Flavour development during beef stock reduction. Food Chemistry, 122(3), 645–655. doi:10.1016/j.foodchem.2010.03.025 Snitkjær, P., Risbo, J., Skibsted, L. H., Ebeler, S., Heymann, H., Harmon, K., & Frøst, M. B. (2011). Beef stock reduction with red wine – Efects of preparation method and wine characteristics. Food Chemistry, 126(1), 183–196. doi:10.1016/j.foodchem.2010.10.096 Soole, S., & Caudle, N. (2013). Cocktail Culture: Recipes & Techniques from Behind the Bar. Victoria, BC: TouchWood Editions. Retrieved from http://books.google.com/ books?id=g7H51fXdqV8C Stephenson, T. (2013). Te Curious Bartender. London: Ryland Peters & Small. Retrieved from http://books.google.com/books?id=jbq-lwEACAAJ Stone, H., Sidel, J., Oliver, S., Woolsey, A., & Singleton, R. C. (1974). Sensory evaluation by quantitative descriptive analysis. In M. Gacula (Ed.), Descriptive Sensory Analysis in Practice (pp. 23–34). Trumbull, CT: Food & Nutrition Press. Szczesniak, A. S., Brandt, M., & Friedman, H. (1963). Development of standard rating scales for mechanical parameters of texture and correlation between the objective and the sensory methods of texture evaluation. Journal of Food Science, 28(4), 397–403. Tan, V. (2013, January). Intentional Ambiguity. PhD Dissertation, Harvard University, Cambridge, MA. 163 pp. Tis, H, & DeBevoise, M. (2013). Molecular Gastronomy: Exploring the Science of Flavor. New York, NY: Columbia University Press. Retrieved from http://books.google.com/ books?id=QVXKZ1-z-eIC Tis, Hervé. (2013). Building a Meal: From Molecular Gastronomy to Culinary Constructivism. New York, NY: Columbia University Press. Retrieved from http://books.google.com/ books?id=qRc2AAAAQBAJ Tompson, B. (1802). On the Construction of Kitchen Fireplaces and Kitchen Utensils: Together with Remarks and Observations Relating to the Various Processes of Cookery; and Proposals for Improving that Most Useful Art. London: T. Cadell and W. Davies. Retrieved from http:// books.google.com/books?id=DylhMgEACAAJ Tonutti, I., & Liddle, P. (2010). Aromatic plants in alcoholic beverages. A review. Flavour and Fragrance Journal, 25(5), 341–350. doi:10.1002/f.2001 Valentin, D., Chollet, S., Lelièvre, M., & Abdi, H. (2012). Quick and dirty but still pretty good: a review of new descriptive methods in food science. International Journal of Food Science & Technology, 47(8), 1563–1578. doi:10.1111/j.1365-2621.2012.03022.x Verhagen, J. V, & Engelen, L. (2006). Te neurocognitive bases of human multimodal food perception: sensory integration. Neuroscience and biobehavioral reviews, 30(5), 613–50. doi:10.1016/j.neubiorev.2005.11.003

37 Wang, Y., Yang, C., Li, S., Yang, L., Wang, Y., Zhao, J., & Jiang, Q. (2009). Volatile characteristics of 50 peaches and nectarines evaluated by HP–SPME with GC–MS. Food Chemistry, 116(1), 356–364. doi:10.1016/j.foodchem.2009.02.004 Wardencki, W., Michulec, M., & Curylo, J. (2004). A review of theoretical and practical aspects of solid-phase microextraction in food analysis. International Journal of Food Science and Technology, 39(7), 703–717. doi:10.1111/j.1365-2621.2004.00839.x Williams, L. (2011). Fish Sauce. Nordic Food Lab Blog. Retrieved March 11, 2013, from http:// nordicfoodlab.org/blog/2011/10/fsh-sauce Williams, L., & Hermansen, M. E. (2012). Delineating the Edible and Inedible. Mad Symposium, July 2 2012, Copenhagen. Retrieved October 04, 2014, from http://www.madfood.co/ nordic-food-lab-2/ Willner, B., Granvogl, M., & Schieberle, P. (2013). Characterization of the Key Aroma Compounds in Bartlett Pear Brandies by Means of the Sensomics Concept. Journal of agricultural and food chemistry, 61, 9583–9593. doi:10.1021/jf403024t Wilson, D. A., & Stevenson, R. J. (2006). Learning to smell: olfactory perception from neurobiology to behavior. Baltimore, MD: Johns Hopkins University Press. Wondrich, D. (2007). Imbibe!: From Absinthe Cocktail to Whiskey Smash, a Salute in Stories and Drinks to “Professor” Jerry Tomas, Pioneer of the American Bar. New York, NY: Perigee Books. Retrieved from http://books.google.com/books?id=IV3s-NUYnfEC Wong, J. W., Webster, M. G., Halverson, C. A., Hengel, M. J., Ngim, K. K., & Ebeler, S. E. (2003). Multiresidue Pesticide Analysis in Wines by Solid-Phase Extraction and Capillary Gas Chromatography − Mass Spectrometric Detection with Selective Ion Monitoring. Journal of Agricultural and Food Chemistry, 51(5), 1148–1161. Woods, V. (1995). Efect of Geographical Origin and Extraction Method on the Sensory Characteristics of Vanilla Essences. M.S. Tesis, University of Missouri, Columbia, MO. 77 pp.

38 Chapter 2: Perceptual Characterization and Analysis of Aroma Mixtures using In-Instrument Gas Chromatography Recombination- Olfactometry1

Abstract Tis paper describes the design of a new instrumental technique, Gas Chromatography Recomposition-Olfactometry (GC-R), that adapts the reconstitution technique used in favor chemistry studies by extracting volatiles from a sample by headspace solid-phase microextraction (SPME), separating the extract on a capillary GC column, and recombining individual compounds selectively as they elute of of the column into a mixture for sensory analysis (Figure 2.1). Using the chromatogram of a mixture as a map, the GC-R instrument allows the operator to “cut apart" and recombine the components of the mixture at will, selecting compounds, peaks, or sections based on retention time to include or exclude in a reconstitution for sensory analysis. Se- lective recombination is accomplished with the installation of a Deans Switch directly in-line with the column, which directs compounds either to waste or to a cryotrap at the operator's discretion. Tis enables the creation of, for example, aroma reconstitutions incorporating all of the volatiles in a sample, including instrumentally undetectable compounds as well those present at concentrations below sensory thresholds, thus correcting for the “reconstitution discrepancy" sometimes noted in favor chemistry studies. Using only fowering lavender (Lavandula angustifola ‘Hidcote Blue’) as a source for volatiles, we used the instrument to build mixtures of subsets of lavender volatiles in-instrument and characterized their aroma qualities with a sensory panel. We showed evidence of additive, masking, and synergistic efects in these mixtures and of “lavender' aroma character as an emergent property of specifc mixtures. Tis was accomplished without the need 1 Note: Tis was published as Johnson, A. J., Hirson, G. D., & Ebeler, S. E. (2012). Perceptual Char- acterization and Analysis of Aroma Mixtures Using Gas Chromatography Recomposition-Olfactometry. (E. M. C. Skoulakis, Ed.)PLoS ONE, 7(8), e42693. doi:10.1371/journal.pone.0042693 and appears here as the accepted text (including abstract)and fgures. 39 for chemical standards, reductive aroma models, or calculation of Odor Activity Values, and is broadly applicable to any aroma or favor.

Introduction Aroma plays a dominant role in the multisensory perception of favor. It is itself a construct perceived in response to stimulation of the olfactory system by volatile chemicals and mixtures thereof, with mixtures being commonly encountered in everyday life in the form of food, wine, plants, perfume, etc. While our understanding of the neurobiological and psychological mechanisms that translate volatiles into aroma perceptions has advanced signifcantly in recent years (Buck 2004, Axel 2004), analytical approaches for characterizing the perception of these aroma mixtures are still limited. Te relationship between chemical composition of a mixture of volatiles and its perceived aroma or favor is complex and difcult to predict on the basis of chemical data or simple sensory data alone. Analytical chemistry approaches for characterizing aromas or favors typically rely on separation-based chromatographic methods that quantify the aroma strength of individual compounds in a mixture, refected as either the concentration present in the mixture divided by a measured sensory threshold concentration (Odor Activity Value, OAV) (Patton and Josephson 1957, Guadagni et al. 1966) or the number of N-fold dilutions required to suppress detectability of a compound when analyzed by gas chromatography with a human subject acting as an olfactory detector (GC-Olfactometry or GC-O; CHARM; or Aroma Extract Dilution Analysis) (Acree and Barnard 1984, Grosch 1993, Grosch 2001). Reconstitution and omission experiments evaluate the role of specifc compounds in the perceived aroma of a mixture, whereby a blend of compounds hypothesized to be detectable in a food, beverage, or other sample by OAV is mixed from chemical standards, and compared to similar mixtures prepared by omitting one of these compounds at a time (Grosch 2001). If a diference is detectable in the “whole” mix versus a

“whole-minus-one-compound” mix, that particular compound is considered important to the aroma of the sample.

40 Figure 2.1 Conceptual schematic of the In-Instrument Gas Chromatograph Recombination Olfactometer (GRO) instrument. Volatiles are extracted onto a solid phase (via solid-phase microextraction or SPME) from the headspace of a food, beverage, or other sample, in this case, lavender fowers, and initially they are separated conventionally on an analytical capillary GC column. In-line with the GC column, a pneumatic Deans Switch followed by a cold trap allows the experimenter to build a mixture of these separated volatiles that is held until the cryotrap is rapidly heated, releasing the mixture for a subject to smell at the olfactory port and evaluate.

Knowledge from other disciplines studying aroma, such as sensory psychophysics, cognitive psychology, and molecular neurobiology, suggests limitations of these methodologies. Chromatographic techniques only assess the aroma quality of individual compounds, rather than mixtures of compounds. However, the aroma of a mixture is frequently perceptually distinct from that of its individual components (Wilson and Stevenson 2006, Laing and Francis 1989) and may have qualities not found in any of these components (Le Berre et al. 2008). Te mixing-dependent nature of aroma quality is evidenced by the relative lack of aroma impact compounds, or those compounds that are singularly responsible for the overall aroma impression of a food or beverage. On the other hand, omission experiments rely on an assumption that all sensorially important compounds have been correctly identifed and quantifed and that any compound occurring at a concentration below its putative sensory threshold is not important to the overall aroma. Recently published results suggest that this is not the case (Pineau et al. 2007). Despite having identical concentration profles of supra-threshold odorants, the aroma of a reconstitution sometimes still smells diferent from the original mixture (Steinhaus et al. 2009) a phenomenon referred to as “reconstitution discrepancy” (Bult et al. 2001). Some recent omission experiments have included sub-threshold components in the reconstitution (Bult et al. 2001), but this is not a universal practice, and can greatly complicate and enlarge the experimental design.

41 Figure 2.2 Schematic of (a) standard GC-MS; (b) GC-MS with splitter at end of column for olfactometry; and (c) In-instrument Gas chromatograph- Recombination Olfactometer or GRO with Deans switch, splitter, cryogenic trap and olfactory port. Abbreviations: i-inlet; c-column; d-detector; o-oven; olf-olfactometry port; sp-splitter; sw-Deans switch 1; w-waste; cr-cryogenic trap; and cb-switch 2 on control box.

We propose here a novel platform for the analytical characterization of aroma and favor perception that incorporates and merges aspects of the previously described techniques and knowledge from other related disciplines. We describe a series of non-reductive, in-instrument recombination and omission experiments using a Gas Chromatograph modifed with a switch and then a cold trap in-line between the capillary column and the chemical and olfactory detectors to characterize the aroma of lavender (Lavandula angustifola ‘Hidcote Blue’). Te volatile chemical composition of lavender, a potently aromatic herb with numerous culinary, cosmetic, and fragrance uses, has previously been characterized (An et al. 2001), but there are no lavender impact compounds currently identifed. Tis suggests that “lavender” aroma character arises from the perception of a mixture of volatiles rather than a single molecule, making this an ideal mixture for evaluation of perceptual interactions using our in-instrument gas chromatography recombination-olfactometry (GRO) approach.

Materials and Methods Instrument: Te GRO Gas Chromatograph is shown in Figure 2.2C. An Agilent model 6890 gas chromatograph/5972 mass spectral detector (GC-MSD) was modifed with the addition of a Deans switch apparatus (Agilent Technologies, Santa Clara, CA), an auxiliary pressure controller (EPC, Agilent) to control fow through the Deans switch, a splitter (Gerstel), a cryotrap (Micro Cryo- 42 Table 2.1 Experimental GC-O conditions and aroma descriptors for mixtures of volatiles from the lavender chromatograms.

Experimental Abbreviation Chromatogram Top Descriptors Condition Sections Included in Mixture Whole W 0-40 minutes Floral, citrus, dried lavender, fresh lavender, mint, wood Omission 1 O1 16-40 minutes Citrus, fresh lavender, dusty, foral, grassy/green, mint, pine, rotten Omission 2 O2 0-16+25-40 Citrus, haylike, foral, pine, root beer minutes Omission 3 O3 0-25 minutes Citrus, grassy/green, mint, wood, soapy Perceptual P1 0-11 minutes grassy/green, wood Interaction 1 Perceptual P2 11-16 minutes Floral, wood Interaction 2 Perceptual P3 16-20.5 minutes Citrus, foral, soapy Interaction 3 Perceptual P4 20.5-25 minutes Dusty, rotten, wet dirt Interaction 4 Perceptual P5 25-32 minutes Black pepper, haylike, citrus, foral, grassy/green Interaction 5 Perceptual P6 32-40 minutes Citrus, smoke Interaction 6 Lavender Reference Not separated; Citrus, foral, fresh lavender, mint, wood, hay, dried Flowers lavender, grassy/green Reference whole lavender fowers trap and model 971 controller, Scientifc Instrument Services, Ringoes NJ) and an olfactometry port (ODP-2, Gerstel, Linthicum, MD). Deactivated fused silica was used for all transfer lines. Te transfer line from the Deans switch to the splitter was 4 m. Te dimensions of the transfer line from the splitter to the MSD was 1 m x 0.15 mm; the dimensions of the transfer line from the splitter to the olfactory port was 1 m x 0.25 mm resulting in a 1.86:1 split ratio between the olfactory port and MSD.

Sampling and Chromatographic Conditions Lavender (Lavandula angustifola ‘Hidcote Blue’) fowers ( 0.50 g) were weighed and placed in a 20 mL amber glass headspace vial and sealed with a crimp cap with a PTFE-faced silicone septum (Supelco, St. Louis, MO). A Solid Phase Microextraction fber (2 cm length, 50/30 um divinylbenzene/carboxen/polydimethylsiloxane coating, Supelco) was used for extraction. 43 Te fber was exposed to the headspace of the vial for 30 minutes at room temperature, then withdrawn and immediately desorbed in the GC inlet. Chromatographic conditions were adapted from (An et al. 2001). Separation was performed using a 30m x 25mm i.d. x 0.25 um flm thickness DB-5MS column (J&W, Folsom, CA). Inlet was maintained at 240 °C in splitless mode. Helium was used as the carrier gas and was held at constant pressure at 15.5 psi. Te auxiliary pressure controller was maintained at 3.4 psi. Te SPME assembly was introduced manually into the inlet and allowed to desorb for a total of 10 minutes. Te oven was held at 60 °C for 3 minutes, then ramped to 150 °C at a rate of 3 °C /min, then ramped to 325 at a rate of 30 °C /min and held for 1 min for a total runtime of 40 minutes. Te olfactory port transfer line was maintained at 100 °C and the MSD transfer line was maintained at 260 °C. Afer a 0.5 min solvent delay, the mass spectrometer scanned from 50-230 amu. With the Deans switch set in the “of” position, the fow is directed to the splitter, MSD, cold trap, and ODP. When set to the “on” position, the fow is directed to waste. Te switch is programmed in the “runtime” tab of the Enhanced Chemstation Sofware (Hewlett Packard, version B.01.00) to direct the fow over the course of the runtime as desired by the operator.

Sensory Conditions Based on retention time, the Deans Switch sends specifc packets of volatiles to the cryotrap. Here we used one of 10 programs (W, O1-O3, P1-P6; see Figure 2.3, Table 2.1) where at the conclusion of the separation run, the cold trap was heated and the mixture was snifed and described by a sensory panelist. Te W condition, analogous to a full aroma reconstitute, contains all the volatiles of lavender, with conditions O1-O3 and P1-P6 omitting groups of these volatiles for descriptive comparison to the aroma of the W sample and to lavender fowers. Tree panelists (Females, ages 28-45 with previous sensory experience) smelled each of the 10 mixtures in triplicate and generated terms to describe the perceived odor. Before smelling each mixture, each panelist frst smelled and described a standard of lavender fowers, picked at the same time as the fowers used for SPME sampling, and also rated how well the sample mixture represented the aroma of the standard on a scale of 0-10. 44 Data Analysis Te terms used to describe the 10 mixtures were tabulated by frequency of use. Te descriptors used most ofen for each mixture, in a mixture-by-descriptor data matrix, was analyzed with a correspondence analysis to identify latent trends in similarity and diference in the multidimensional set. A three-way Analysis of Variance (ANOVA) with all two-way interactions was performed with rated representativeness of each mixture compared to a fresh lavender standard as the response factor and panelist, mixture, and replicate as main efects. A Tukey’s Honest Signifcant Diference multiple comparisons test (HSD) was performed on the representativeness ratings. Te

R statistical computing package was used for all statistical analyses (http://www.r-project.org/).

Results and Discussion We modifed a GC-MS to allow for the in-instrument preparation of volatile mixtures containing precise sections from a chromatogram, up to and including the entire volatile fraction and allowing for aroma characterization of the aroma of one or a few of the volatiles in a complex

Figure 2.3 Top aroma descriptors for mixtures of sections of the lavender chromatogram by cut time and chromatogram composition. Abbreviations correspond to Experimental Conditions described in Table 1. As chemical complexity and number of components per mixture approaches the makeup of the whole chromatogram (W) mixture, there is evidence of perceptual additivity as increasing cross- utilization of terms from simpler mixtures, masking as reduced use of dominant terms for simpler (P1- P6) mixtures, and synergistic efects as new complex or composite terms like “fresh lavender” become important.

45 mixture (Figure 2.1). Compounds were introduced into the inlet of the modifed GC-MS and separated on the analytical column. At the end of the column, the fow of carrier gas and analytes encountered a frst switch, a commercially available Deans switch, that was set to direct the fow either towards the splitter or towards waste (here waste was vented to the oven). Te splitter subsequently split the fow to both a mass spectrometer (MS) detector and to an olfactory port. Along the transfer line to the olfactory port was a trap controlled by a second switch at the control box; the switch allowed the trap to be cooled with liquid carbon dioxide or heated so that the eluant was either held within the trap (i.e., cryotrapped) or released to the olfactory port. By programming the switches to cryotrap or exclude selected peaks or peak regions (Table 2.1) two types of experiments were performed. In perceptual interaction experiments, all of the chromatogram except for a small section of peaks was cut away, and the section of interest was assessed at the olfactory port as a mixture. In omission experiments, small groups of peaks (or individual peaks) were cut away and the rest of the compounds in the chromatogram were smelled as a mixture. See supporting fgures S2.1 and S2.2 for examples of these chromatograms.

Figure 2.4 Correspondence Analysis of (A) lavender volatile mixtures; and (B) lavender volatile mixture descriptors. Abbreviations for mixtures correspond to those in Table 1. Terms generated by the panelists to describe the perceived odor of from each Experimental Condition described in Table 1 were tabulated by frequency of use and used for the Correspondence Analysis. 30.57% of variance explained by dimension 1 (x), 22.84% of variance explained by dimension 2 (y).

46 Using our new approach, ten aroma mixtures (Table 2.1, Figure 2.3) were created in-instrument directly from the headspace-extracted volatiles of fowering lavender. “Fresh Lavender” and “Dried Lavender” were both predominant descriptors for the “Whole Volatile” recombination mixture W. Of the more chemically complex omission mixtures O1-O3, only O1, which incorporated the section of volatiles eluting from 16-40 min of the lavender chromatogram and omitted volatiles eluting between 0-16 min, was described as having “fresh lavender” properties. O1 overlapped with O2 from 25-40 min and with O3 from 16-25 minutes and incorporated the perceptual mixtures P3-P6, however, none of these other omission or perceptual mixtures had fresh or dried lavender among their commonly used descriptors. Tis suggests that there are two subsets of compounds, the frst eluting between 16-25 min and the other eluting between 25-40 min, that are each necessary for the perception of “lavender character” but are not alone sufcient for inducing this perception without some mixing with compounds in the other elution group. Tese results also suggest that “lavender character” is an emergent perceptual property arising from the mixing of these volatiles or some subset thereof. We performed a Correspondence Analysis on the descriptors-by-mixtures data matrix to compare dimensionally-reduced latent trends in the sensory profles of the mixtures to the diferences evident in top descriptors for each mixture (Figure 2.4). Correspondence Analysis separates dissimilar categories in space; mixtures and sensory descriptors spaced closely together share more similarities than those spaced further apart. Tis plot shows that, generally, removing more volatiles results in greater dissimilarity between a given mixture and the all-volatiles- included mixture W. Te relatively tight clustering of W and omission mixtures O1-O3 in the Correspondence Analysis refects the sensory similarity of these mixtures; perceptual mixtures P2 and P3 also cluster nearby, refecting some of the overlapping characteristics of these mixtures (Figure 2.4). Te location of mixture W in the center of the main cluster in the Correspondence

Analysis, suggests its aroma was perceived, in part, as a sensory average of some of the less- complex mixtures. However, a truly averaged perceptual character would be in the center of

47 the plot; the fact that mixture W is ofset from the geometric center implies that the mixing- dependent interactive efects of the lavender volatiles perceived in mixture W play a noticeable role in afecting its overall aroma character. Mixture W shares many similar descriptors (Table 2.1) with O1-O3 and P2 and P3, but all of these except O1 lack a dominant lavender character. Mixtures P1 and P5 are close to the central cluster but are approximately equi-distant in space from mixture W. Tis refects some of the similarities in the descriptors that P1 and P5 share with mixture W, but also refects the domination of the aromas of these mixtures by either a unique character (“black pepper”) in the case of P5, or the relative simplicity of the aroma in the case of P1 (Figure 2.4a). Te comparative distancing of mixtures P4 and P6 from the other mixtures refects the relative uniqueness of their aroma descriptors. Locations of descriptors suggest that along the frst (x) dimension of Figure 2.4b, there is a distinction between fresher, more “sweet” and fower-associated terms on the right side and earthier, heavier aroma terms on the lef. Borrowing more qualitative terms from the tradition of perfumery (which at its essence is the craf of observing and optimizing the perceptual efects of mixing volatiles), we observe a rough Figure 2.5 Te rated representativeness of the aroma progression, from lef to right along the of samples W, O1-O3, and P1-P6 as compared by panelists to the aroma of whole fowering lavender. x-axis, of base, middle, and top-note (Afel Letters a, b, c refer to the mixture’s Signifcant Diference from each other- if two samples do 2004) related terms. Along the second (y) not share a letter, they are signifcantly diferent. Samples P1, P5, and P6 are signifcantly less dimension the separation is dominated by the representative of the aroma of fowering lavender than sample W, which incorporates all the volatiles marked diference of P4 and P6 from each in fowering lavender. other and from the rest of the mixtures, and correspondingly by their unique descriptors “wet dirt” and “smoky” in fgure 4b. Generally, the terms on the other arm of the y-dimension tend to be shared by multiple mixtures, or refect more composite aroma characteristics.

48 While sample P1 appears to be the closest to the central or average sample in this set, it is clearly separated from the cluster centered around mixture W along the third (z) dimension (Figure S3 supporting information). Te third dimension also further separates mixture P5 from the central W-associated cluster and increases the distinction between “grassy/green”- “woody” descriptors on one side and “dried lavender”-“black pepper” descriptors on the other. Importantly, the Correspondence Analysis, while unable to describe absolute diferences, provides valuable information not only on the sources of variation in the complex sensory data but also on the interrelationships of the mixtures and their sensory properties. Te method used to create an extract of volatile compounds can alter the perceived aroma of that extract and failure to obtain a representative sample can lead to unreliable conclusions about the composition of the aroma active components (Abbott et al. 1993, Etievant et al. 1993, van Ruth et al. 2004, Plutowska and Wardencki 2008, Aceña et al. 2010). While many extraction methods have been employed in order to produce an aroma extract (Plutowska et al. 2008, Aceña et al 2010, San-Juan et al. 2010), the creation of a representative aroma can be very difcult for complex matrices (Aceña et al. 2010, Pérez-Silva et al 2006), and the sensory representativeness of this extract is not always evaluated. Here, the aroma of the SPME extracts of lavender corresponded closely to the original product (Table 2.1). Similar representative aroma samples have been obtained using SPME to sample “baked potato” aroma (Poinot et al. 2007). Importantly, the GRO approach provides a rapid, easy, and efective tool to assess the representativeness of an extract regardless of the extraction method employed, such as in cases where SPME coatings may not be able to produce an appropriate extract (Ferreira et al. 2002). Since the SPME extraction produced an aroma mixture representative of lavender, it was possible to perform omission and interaction experiments based on a starting point nearly identical to the intact lavender sample, eliminating “reconstitution discrepancy” (Bult et al. 2001). Comparing the aroma of the GRO mixtures in this study to the aroma of whole lavender fowers, panelists found that mixtures P1, P5, and P6 were signifcantly less representative (Figure 2.5) of the aroma of the whole fowers than mixtures W, O1-O3 and P2-P4. Tese samples also tended to have either fewer commonly used descriptors or descriptors not found for other mixtures (such as “black pepper” for P5 and “smoke” for P6; Table 2.1). 49 In this experimental design, mixtures of compounds were omitted to assess the resulting aroma. Cut times were chosen to include chemically similar compounds in the same mixture, for example, monoterpene acetate esters in mixture P5 and sesquiterpenes in mixture P6. However, the omitted compounds/fractions in a theoretical GRO experiment need not be contiguous. It is possible, for example, to remove every other chromatographic peak, to remove only the 3rd and 17th peak, etc. while trapping and evaluating the remaining components. Te apparatus could additionally be used to perform single omission experiments, where compounds are omitted one at a time to screen for potential impact odorants, or perceptual interaction experiments where only 2 or 3 peaks are included in the mixture. Te fexibility in the compounds that can be removed and assessed is only limited by the rapid switching time of the Deans switch. By using a Mass Spectrometric detector, compounds in the sample can be identifed (Table S2.4, Supporting Information) however, an obvious advantage of performing an omission experiment in this manner is that the compounds need not be identifable or available to perform the experiment. Reconstitution experiments ofen require the experimenter to perform lengthy and labor-intensive syntheses to prepare a component for the reconstitution model (Steinhaus et al. 2009) only to fnd that the component can be omitted with no change in the overall aroma of the solution. Furthermore, there is always some fraction of the total compounds identifed that are not included in the reconstitution because they are deemed to have a concentration too low to have an efect on the overall aroma. However, compounds with low odor activity values ofen still have a considerable efect on the overall aroma of the mixture (Pineau et al. 2007, Escudero et al. 2004, Ryan et al. 2008). With this instrument there is no simplifed reconstitute - the omission experiment is performed on the entire sample. While compounds with low OAVs may be important to the aroma of the mixtures, the opposite case can also occur, and the sensitivity of the human nose is frequently orders of magnitude greater than an instrumental detector. As a result, the nose may detect an aroma where there is no peak on a chromatogram (Etievant et al. 1993). Particularly as compared to reconstitution studies, this is another distinct advantage of the GRO approach since even compounds not detected by the detector (MS, FID) will be included in the aroma sample as it is

50 assessed by a subject at the olfactometry port. Traditionally, full separation of volatile compounds on the chromatographic column is necessary in order to meaningfully describe the aroma character of the eluant by GC-O since it simplifes the recognition task for the assessor (San-Juan et al. 2010). However, it is more ofen the case that a complex mixture of aroma compounds is responsible for the overall aroma of a food or beverage. In addition, a mixture of two or more odorants can frequently lead to an aroma that is not similar to any of its individual components (Le Berre et al. 2008a, b). Using an GRO technique, any of these interactions can readily be investigated; and all that is necessary to characterize any type of aroma interaction is a sample of the food, beverage, fower, etc. of interest. Compounds detectable by GC-O but not GC-MS, compounds below putative aroma thresholds, compounds at levels that cannot be quantifed, and compounds not commercially available or easily synthesized can all be perceptually analyzed if they are found in one or more aromatic samples available to the researcher.

Conclusions Te perception of aroma and favor has ofen been approached as a problem of many individual parts, with chemistry, neurobiology, sensory science, psychology, and other disciplines focused on answering questions about some aspect of the relationship between stimulus (a fower, a glass of wine, a plate of food), response (perceived favor, liking or disliking, intake and satiety), or the pathway between the two (genetics, receptor binding, transduction, translation to cortical neurons). Tis has yielded a great deal of information about those individual parts, but not a well- developed understanding of how they work together for complex, everyday stimuli and activities like eating and drinking. Te need for a holistic approach to address this has been identifed previously (Shepherd 2006), i.e., a praxis which would bring together knowledge and research techniques from these diverse, ofen isolated, but orthogonally-related scientifc felds, and would include expertise or information from applied, non-analytical felds with a well-developed shared intuition about the nature of aroma and favor in practice, such as cuisine and perfumery. While the described approach of in-instrument gas chromatography recombination-olfactometry has its

51 roots in a traditional coupling of analytical chemistry and sensory science, it is highly informed by this multidisciplinary understanding of aroma and favor and allows for the analysis of previously uncharacterized emergent perceptual properties of complex mixture interaction efects in everyday smell and favor situations.

References

Abbott, N., Etihvant, P., Langlois, D., Lesschaeve, I., Issanchou, S., Recherche, L. De, … Cedex, D. (1993). Evaluation of the Representativeness of the Odor of Beer Extracts Prior to Analysis by GC Eluate Snifng. Journal of agricultural and food chemistry, 41, 777–780. Aceña, L., Vera, L., Guasch, J., Busto, O., & Mestres, M. (2010). Comparative study of two extraction techniques to obtain representative aroma extracts for being analysed by gas chromatography-olfactometry: application to roasted pistachio aroma. Journal of chromatography. A, 1217(49), 7781–7787. doi:10.1016/j.chroma.2010.10.030 Acree, T. E., & Barnard, J. (1984). A Procedure for the Sensory Analysis of Gas Chromatographic Efuents. Food Chemistry, 14, 273–286. Afel, M. (2004). Essence and Alchemy: A Natural History of Perfume. New York: Gibbs Smith. An, M., Hai, T., & Hatfeld, P. (2001). On-site feld sampling and analysis of fragrance from living lavender (Lavandula angustifolia L.) fowers by solid-phase microextraction coupled to gas chromatography and ion-trap mass spectrometry. Journal of chromatography. A, 917(1-2), 245–50. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/11403476 Axel, R. (2004). Scents and Sensibility: A Molecular Logic. In Les Prix Nobel. Te Nobel Prizes 2004 (pp. 234–256). Buck, L. B. (2004). Unraveling the sense of smell. In Les Prix Nobel. Te Nobel Prizes 2004 (pp. 267–283). Bult, J. H., Schiferstein, H. N., Roozen, J. P., Voragen, A. G., & Kroeze, J. H. (2001). Te infuence of olfactory concept on the probability of detecting sub- and peri-threshold components in a mixture of odorants. Chemical senses, 26(5), 459–469. Retrieved from http://www.ncbi.nlm. nih.gov/pubmed/11418491 Escudero, A., Gogorza, B., Melus, M. A., Ortin, N., Cacho, J., & Ferreira, V. (2004). Characterization of the aroma of a wine from maccabeo. Key role played by compounds with low odor activity values. Journal of agricultural and food chemistry, 52(11), 3516–24. doi:10.1021/jf035341l Etiévant, P. X., Moio, L., Guichard, E., Langlois, D., Leschaeve, I., & Schlich, P. (1993). Aroma extract dilution analysis (AEDA) and the representativeness of the odour of food extracts. In H. Maarse & D. G. Van Der Heij (Eds.), Trends in favour research, volume 35 of 52 Developments in food science (pp. 179–190). New York: Elsevier. Ferreira, V., Lopez, R., & Aznar, M. (2002). Olfactometry and aroma extract dilution analysis of wines. In J. F. Jackson & H. F. Linskens (Eds.), Analysis of Taste and Aroma Volume 21 (pp. 89–122). Heidelberg: Springer. Grosch, W. (2001). Evaluation of the key odorants of foods by dilution experiments, aroma models and omission. Chemical senses, 26, 533–545. Retrieved from http://www.ncbi.nlm. nih.gov/pubmed/11418500 Grosch, Werner. (1993). Detection of potent odorants in foods by aroma extract dilution analysis. Trends in Food Science & Technology, 4, 68–73. Guadagni, D. G., Buttery, R. G., & Harris, J. (1966). Odour intensities of hop oil components. J. Sci. Food Agric., 17(1), 142–144. Laing, D. G., & Francis, G. W. (1989). Te capacity of humans to identify odors in mixtures. Physiology & behavior, 46, 809–814. Retrieved from http://www.ncbi.nlm.nih.gov/ pubmed/2628992 Le Berre, E, Béno, N., Ishii, A., Chabanet, C., Etiévant, P., & Tomas-Danguin, T. (2008). Just noticeable diferences in component concentrations modify the odor quality of a blending mixture. Chemical senses, 33(4), 389–395. doi:10.1093/chemse/bjn006 Le Berre, Elodie, Tomas-Danguin, T., Béno, N., Coureaud, G., Etiévant, P., & Prescott, J. (2008). Perceptual processing strategy and exposure infuence the perception of odor mixtures. Chemical senses, 33(2), 193–199. doi:10.1093/chemse/bjm080 Patton, S., & Josephson, D. (1957). A method for determining signifcance of volatile favor compounds in foods. Journal of Food Science, 22(3), 316–318. Pérez-Silva, a., Odoux, E., Brat, P., Ribeyre, F., Rodriguez-Jimenes, G., Robles-Olvera, V., … Günata, Z. (2006). GC–MS and GC–olfactometry analysis of aroma compounds in a representative organic aroma extract from cured vanilla (Vanilla planifolia G. Jackson) beans. Food Chemistry, 99(4), 728–735. doi:10.1016/j.foodchem.2005.08.050 Pineau, B., Barbe, J.-C., Van Leeuwen, C., & Dubourdieu, D. (2007). Which impact for beta- damascenone on red wines aroma? Journal of agricultural and food chemistry, 55(10), 4103– 4108. doi:10.1021/jf070120r Plutowska, B., & Wardencki, W. (2008). Application of gas chromatography–olfactometry (GC– O) in analysis and quality assessment of alcoholic beverages – A review. Food Chemistry, 107(1), 449–463. doi:10.1016/j.foodchem.2007.08.058 Poinot, P., Grua-Priol, J., Arvisenet, G., Rannou, C., Semenou, M., Le Bail, A., & Prost, C. (2007). Optimisation of HS-SPME to study representativeness of partially baked bread odorant extracts. Food Research International, 40(9), 1170–1184. doi:10.1016/j.foodres.2007.06.011 Ryan, D., Prenzler, P. D., Saliba, A. J., & Scollary, G. R. (2008). Te signifcance of low impact odorants in global odour perception. Trends in Food Science & Technology, 19(7), 383–389. doi:10.1016/j.tifs.2008.01.007

53 San-Juan, F., Pet’ka, J., Cacho, J., Ferreira, V., & Escudero, A. (2010). Producing headspace extracts for the gas chromatography–olfactometric evaluation of wine aroma. Food Chemistry, 123(1), 188–195. doi:10.1016/j.foodchem.2010.03.129 Shepherd, G. M. (2006). Smell images and the favour system in the human brain. Nature, 444(7117), 316–21. doi:10.1038/nature05405 Steinhaus, M., Sinuco, D., Polster, J., Osorio, C., & Schieberle, P. (2009). Characterization of the key aroma compounds in pink guava (Psidium guajava L.) by means of aroma re- engineering experiments and omission tests. Journal of agricultural and food chemistry, 57(7), 2882–2888. doi:10.1021/jf803728n Van Ruth, S. M., Geary, M. D., Buhr, K., & Delahunty, C. M. (2004). Representative sampling of volatile favor compounds: Te model mouth combines with gas chromatography and direct mass spectrometry. In K. D. Deibler & J. Delwiche (Eds.), Handbook of Flavor Characterization: Sensory Analysis, Chemistry, and Physiology (pp. 303–311). New York: CRC Press. Wilson, D. A., & Stevenson, R. J. (2006). Learning to smell: olfactory perception from neurobiology to behavior. Baltimore, MD: Johns Hopkins University Press.

Acknowledgments

We thank Jim McCurry and Agilent Technologies for the auxiliary pressure controller and John Torngate for helpful discussions in early stages of this project. We thank the sensory judges who participated in this study.

54 Supplementary Information:

Supplementary Figure S2.1 Te chromatogram of mixture O2. Compounds eluting between 16 and 25 minutes were vented to waste by the Deans Switch and were consequently excluded from the smelled mixture and not sent to the mass spectrometer.

Supplementary Figure S2.2 Te chromatogram of mixture P5. Compounds eluting between 0 and 25 minutes and 32 and 40 minutes were vented to waste by the Deans Switch and were consequently excluded from the smelled mixture and not sent to the mass spectrometer.

55 Supplementary Figure S2.3 Alternate views of correspondence analysis (fgure 4) incorporating the frst 3 dimensions of variation. 30.57% of variance explained by dimension 1 (x), 22.84% of variance explained by dimension 2 (y), 14.03% of variance explained by dimension 3 (z).

56 Supplementary Table S2.4 Tentative identifcation of lavender volatile compounds. Volatiles were identifed by matching their mass spectra to the NIST 05 Mass Spectral Library (National Institute of Standards and Technology, Gaithersberg, MD) and to chemical standards, as noted. Te table is divided by cut time for perceptual mixtures P1-P6. Retention Time Retention Identif- Perceptual Mixture Compound CAS Number Index cation Group 7.23 3-hexen-1-ol 544-12-7 893 standard P1 7.64 1-hexanol 111-27-3 900 standard P1 9.82 alpha-thujene 2867-05-2 940 library P1 10.15 alpha-pinene 80-56-8 946 standard P1 10.79 camphene 79-92-5 958 standard P1 11.83 sabinene 3387-41-5 978 library P2 12.00 beta-pinene 127-91-3 982 standard P2 12.45 3-octanone 106-68-3 990 standard P2 12.57 myrcene 123-35-3 993 standard P2 12.72 not identifed 996 P2 13.15 alpha-phellandrene 99-83-2 1004 library P2 13.50 hexyl acetate 142-92-7 1011 standard P2 13.23 not identifed 1006 P2 13.65 3-carene 13466-78-9 1014 standard P2 14.52 not identifed1 1032 P2 14.67 eucalyptol 470-82-6 1035 standard P2 15.41 trans-beta-ocimene 3779-61-1 1050 library P2 15.79 cis-beta-ocimene 3338-55-4 1058 library P2 16.06 gamma-terpinene 99-85-4 1064 library P3 16.44 beta-terpineol 138-87-4 1072 library P3 17.34 terpinolene 586-62-9 1091 standard P3 18.90 linalool 78-70-6 1125 standard P3 18.62 octen-1-ol acetate 32717-31-0 1119 standard P3 19.04 3-octylacetate 4864-61-3 1128 library P3 19.64 allo-ocimene 7216-56-0 1142 library P3 20.21 lavandulol 507-70-0 1155 library P3 20.43 camphor 464-49-3 1160 standard P3 22.08 terpinen-4-ol 562-74-3 1199 standard P4 22.66 hexyl butyrate 2639-63-6 1212 library P4 23.20 alpha-terpineol 98-55-5 1225 standard P4 24.34 bornyl formate 7492-41-3 1253 library P4 26.53 linalyl acetate 115-95-7 1309 library P5 27.28 isobornyl acetate 125-12-2 1328 library P5 27.42 lavandulyl acetate 25905-14-0 1332 library P5 27.71 geranyl acetate 105-87-3 1340 library P5 30.63 neryl acetate 141-12-8 1420 library P5 31.67 Not identifed 2 1449 P5 57 Supplementary Table S2.4 continued 32.85 alpha-bergamotene 17699-05-7 1484 library P6 33.65 alpha-santalene 512-61-8 1507 library P6 33.78 beta-caryophyllene 87-44-5 1511 standard P6 34.48 beta-farnesene 77129-48-7 1532 standard P6 34.99 alpha-bisabolene 17627-44-0 1547 library P6 35.16 germacrene D 37839-63-7 1553 library P6 35.73 alpha-amorphene 483-75-0 1570 library P6 1possible terpene 2possible acetate ester

58 Chapter 3: Volatile and Sensory Profling of Cocktail Bitters

Introduction Extracting plant matter into alcohol is an ancient process, see, for example, the Hippocratic wine of the Greeks (Tonutti et al. 2010). As distilled liquor became more widely available, it was put to use in making plant extractions, mostly for medicinal purposes. Stoughton’s Great Cordial Elixir, a distilled-alcohol based herbal extraction bittered with gentian root became available commercially in 1690. While this was a patent medicine marketed for its medicinal properties it is the closest ancestor of what we today know as bitters. While the Elixir could be taken straight, it was ofen diluted into wine to make “instant” Purl-royal, a popular drink resembling vermouth, and was also ofen subsequently mixed with straight or burnt brandy (brandy with sugar added and reduced in alcohol by igniting it). Adding Stoughton’s Elixir to a dram of brandy yielded a “bitter draught” that was to be administered medicinally; recreational mixing soon followed (Wondrich 2007). Two brands of bitters on the market today, Angostura and Peychaud’s, have been in production since the nineteenth century; the Angostura company was founded in 1824 and Peychaud’s in 1837 (Parsons 2012). Angostura is ofen considered a “default” bitters in a cocktail recipe when a specifc type is not called for, and Peychaud’s bitters have a famously anise-heavy aroma and bright red color, and are used in the so-called signature cocktail of New Orleans, the Sazerac (Simmons 2011). Several named types of bitters are called for in Jerry Tomas’ Te Bon Vivant’s Companion or How to Mix Drinks, the frst book on cocktails (Grimes 2002), frst published in 1862. Tese named bitters include Boker’s (or Bogart’s), Stoughton’s, and “Jerry Tomas’ Own Decanter Bitters”, with recipes provided for the bitters themselves. While the commercial production of many popular nineteenth-century bitters, such as Boker’s, Stoughton’s, and Abbot’s, ceased during or relatively soon afer Prohibition (Parsons 2012), Boker’s and Jerry Tomas’ Own Decanter Bitters are both currently commercially available, having been re- 59 created from historical recipes (Table 1). Boker’s, Angostura, and Decanter bitters are commonly informally categorized as “aromatic” styles, with spice favors such as cinnamon, cloves, and cardamom (Clarke 2010). Peychaud’s and other brands inspired by it are ofen noted as either a subtype of aromatic bitters or a separate style, due to anise favors that are more intense than in other aromatic bitters (Clarke 2010, Parsons 2012, Bovis 2012, Sandham 2012). A third important “classic” style is citrus bitters, especially orange bitters (Clarke 2010, Parsons 2012); Regan’s Orange Bitters (Regan 2002), a popular contemporary commercial brand (Parsons 2012), were based on a recipe from another historical manual, Charles H Baker’s Te Gentleman’s Companion: Being an exotic drinking book; or, Around the world with jigger, beaker and fask, published in 1939. Baker also called for the use of celery bitters in the recipe for a cocktail called the Fourth Regiment; celery bitters is another style that had died out commercially until relatively recently (Baker 1939). Along with commercial reintroduction of defunct historical styles of bitters since the early 2000’s, there has been an introduction of many new styles and types of bitters, driven in part by a rise in bartenders developing their own bitters in-house (Sandham 2012, Parsons 2012, Clarke 2011). While these “new bitters” have a range of ingredients and favors with nearly indefnable boundaries, a number of recently invented styles have gained prominence. “Xocolotl Mole Bitters”, produced by the company Bittermen’s starting in 2007 based on the chocolate, chile, and cinnamon favor profle of Mole Poblano, has been widely praised in print and used widely at craf cocktail bars (Parsons 2012, Clarek 2012). A similar “new classic” trend is more heavily-spiced bitters intended for Tiki drinks, which ofen use spice-heavy ingredients such as falernum (and almond and clove syrup) and pimento dram (an allspice liqueur) in conjunction with robust Jamaican or Agricole rums. With these somewhat loose categories as a guide, at least two commercially available examples of each style (aromatic, citrus, New Orleans-style, tiki, mole, celery) were included in the samples for this study (Table 3.2).

Table 3.1 shows ingredients commonly used in making historical and contemporary bitters. Specifc historical recipes for Boker’s, Stoughton’s, Orange, and Jerry Tomas’ Own

60 Table 3.1: Ingredients used in historical and contemporary recipes for bitters, listed by taxonomic name and literature source

Component Source1 Taxonomic name Component Source1 Taxonomic name orange T,W,O,J,B Citrus sinensis black walnut leaf P Juglans nigra Elettaria cardamom T,P,W,O,B cardamomum burdock P Arctium lappa quassia T,P,W,O,B Quassia amara chirayata P Swertia chirayita Syzygium clove T,P,W,O,J aromaticum dandelion P Taraxacum ofcinale gentian T,P,W,O,S Gentiana spp. devil’s club P Oplopanax horridus calamus T,P,W,B Acorus calamus fennel P Foeniculum vulgare allspice T,P,W,J Pimenta dioica fringe tree P Chionanthus spp Coriandrum coriander T,P,O sativum horehound P Marrubium vulgare Cinnamomum cinnamon P,W,J verum hyssop P Hyssopus ofcinalis raisins T,W,J Vitis vinifera juniper P Juniperus spp snake root T,W,J several species lavender P Lavandula angustifolia Cinnamomum cassia P,W,S verum lemongrass P Cymbopogon citratus Matricaria chamomile T,W,S chamomilla licorice P Glycyrrhiza glabra catechu W,B Senegalia catechu milk thistle P Silybum marianum caraway P,O Carum carvi peppermint P Mentha x piperita cinchona P,O Cinchona spp rose hip P Rosa spp lemon T,W,J Citrus limon sarsparilla P Smilax regelii bitter orange W,S Citrus aurantium sassafras P Sassafras spp calumba W,S Jateorhiza palmata schizandra P Schisandra chinensis angelica T,P Angelica sylvestris star anise P Illicium verum Aframomom grains of paradise T,P melegueta wild cherry wood P Prunus avium hops T,P Humulus lupulus aloe T Aloe spp nutmeg T,P Myristica fragrans centaurium T Centaurium erythraea orris T,P Iris spp galanga root T Alpinia spp Artemisia wormwood T,P absinthium ginger T Zingiber ofcinale anise T,P Pimpinella anisum myrrh T Commiphora myrrha arnica P Arnica spp polypody T Polypodium spp barberry P Berberis vulgaris safron T Crocus sativus birch leaf P Betula spp 1 T=Tomas 1862; P=Parsons et al. 2011; W=Wondrich, 2007; O=Orange bitters (Regan 2003); J=Jerry Tomas’ Own Decanter Bitters, (Tomas 1862); S=Stoughton’s bitters (Wondrich 2007); B=Boker’s bitters, (Wondrich 2007).

61 Decanter Bitters are included; most contemporary makers of bitters do not publicize their formulas, so ingredients listed in contemporary recipes are included. In some cases there is quite a bit of overlap; for example, many aromatic bitters recipes include citrus peel, and the included orange bitters recipe uses several spices. Te ultimate favor profle of any of these is therefore likely more dependent on proportions of ingredients than use of specifc ingredients. Te earliest written description of the cocktail comes from an 1806 edition of the Hudson, New York newspaper Te Balance and Columbian Repository: “Cock Tail, then, is a stimulating liquor, composed of spirits of any kind, sugar, water and bitters” (Sampson et al. 1806). While spirits had been imbibed sweetened and diluted with water, in a drink called the “sling”, for some time before, this was the frst time that the cocktail, as a separate drink and including bitters as an ingredient, was codifed in writing (Wondrich 2007). Two conclusions can be made from this: the cocktail is almost as old as America itself, and bitters are its defning component, setting it apart from other spirit-based drinks. In cocktail-making (or mixology, which curiously enough is not a recent neologism but rather a term introduced facetiously in the Knickerbocker Magazine in 1856 [Wondrich 2007]) bitters are usually used at a maximum concentration of 1-2% of the total volume of the drink. Bitters are probably most familiar in drinks like the Old-Fashioned (Whiskey, sugar, and bitters) and the Manhattan (Whiskey, sweet vermouth, and bitters), though they were considered an essential component in the Martini until the middle of the twentieth century, and in current practice are added to a wide variety of similarly spirit-forward drinks as well as citrus-lightened sours. Indeed, while cocktail bitters are formulated to be concentrated and bitter to the point that they are considered non-potable unless heavily diluted, there is a growing corpus of drinks recipes that makes use of, for example, two ounces of angostura bitters as their most abundant ingredient (Baker 1939). But setting aside these extreme cases, bitters are more generally used to add aroma complexity to a cocktail, to complement and contrast the favors already present in the component liquors, and, by selecting diferent styles, to subtly alter the favor of the same base cocktail without changing its essence.

62 Table 3.2 Bitters samples used in the study, with historical sources and precedents, and style noted.

Name Brand Type Abbreviation Code

Boker’s Bittersa Dr. Adam Elmegirab Aromatic BOKERS A1

Angostura Bittersb Angostura Aromatic ANGOSTURA A2

Jerry Tomas’ Own Bitter Truth Aromatic JTDECANTER A3 Decanter Bittersc

WHISKEY BARREL- Whiskey Barrel-Aged Bitters Fee Brothers Aromatic A4 AGED Regan’s Orange Bitters Bufalo Trace Citrus REGAN’S ORANGE C1 Number 6d

Hopped Grapefruit Bitters Bittermen’s Citrus HOP-GRAPEFRUIT C2

Grapefruit Bitters Scrappy’s Citrus SCRAPPY GRAPEFRUIT C3

Orange Bitters Scrappy’s Citrus SCRAPPY ORANGE C4

Xocolatl Mole Bitters Bittermen’s Mole XOCOLOTL MOLE M1

Mole Bitters Bitter Truth Mole BT-MOLE M2

‘Elamakule Tiki Bitters Bittermen’s Tiki ELAMAKULE-TIKI T1

Jamaica Bitters Bittercube Tiki JAMAICA T2

Creole Bitters Bitter Truth New Orleans BT-CREOLE NO1

Peychaud’s Bittersb Peychaud New Orleans PEYCHAUD NO2

Orchard St Celery Bittermen’s Celery BMCELERY C1

Celery Bitters Scrappy’s Celery SCRAPPYCELERY C2

a=based on historical recipe for now-defunct Boker’s brand (Parsons et al. 2012). b=19th-century brand (Parsons et al. 2012). c=based on historical recipe from Te Bon Vivants Companion or How to Mix Drinks (Wondrich 2007). d=based on historical recipe from Te Gentleman’s Companion: Being an Exotic Drinking Book or Around the World with Jigger, Beaker and Flask (Baker 1939, Regan 2003).

63 Despite the long timeline of bitters, and the cultural role they have played both historically and in their current resurgence, they have not been the subject of any published study of their chemical composition or sensory properties. Terefore, the objectives of this study are to describe, map, and analyze the favor chemistry of the most common styles of bitters currently available (16 commercial bitters refecting six common categories [see Table 3.2] are used, as it would be unfeasible to capture every available product) using volatile profling via Gas Chromatography-Mass Spectrometry, sensory descriptive analysis with trained panelists, and multivariate statistical analysis to reveal product-descriptor and sensory-chemical correlations.

Materials & Methods Samples 16 bitters (Table 3.2) were purchased from Astor Wines & Spirits (New York, NY), Cask (San Francisco, CA), Amor y Amargo (New York, NY), and Union Square Liquors (New York, NY). Chemical Analysis A 200 μL aliquot of bitters was pipetted into 10 mL of water in 20 mL amber glass headspace vials (Agilent Technologies, Santa Clara, CA) capped with magnetic, PTFE-lined silicone septa headspace caps. 2-undecanone was used as an internal standard at 50 μg/L (Sigma-Aldrich, St Louis, MO). A conditioned, 2-cm long PDMS-DVB-Carboxen SPME fber (Supelco, Bellefonte, PA) was introduced into the headspace of the vial for 40 minutes at 25ºC with rotational shaking at 250 RPM. A Gerstel MPS2 autosampler performed the extraction and the injection (Gerstel, Linthicium, MD). Te fber was removed from the headspace of the vial and immediately introduced into the inlet of an Agilent model 6890 GC-single quadrupole-MS (Agilent Technologies) with a DB-WAX column (30 meters long, 0.25 mm ID, 0.25 μm flm thickness) (J&W Scientifc, Folsom, CA). Te inlet was held at 250ºC with a 10:1 split. Te carrier gas was helium, with a 1 mL/minute constant fow rate. Te starting oven temperature was 40ºC, held for 3 minutes, followed by a 2ºC/minute ramp until 180ºCwas reached, then the ramp was

64 increased to 30ºC/minute until 250ºC was reached, and held for 3 minutes. Te total runtime was 47 minutes. Te mass spectrometer had a 1.5-minute solvent delay and was run in scan mode with Electron Impact Ionization at 70eV, from m/z 40 to m/z 300. Te samples were analyzed in triplicate. Kovats retention indices were calculated using C8-C20 hydrocarbon mixture (Sigma- Aldrich) retention times for identical instrumental conditions. Peak identifcations were made by matching the background-subtracted average mass spectrum across half peak height for each peak to the NIST 05 mass spectral database, followed by verifcation by retention index and pure standards where available. Following identifcation, GC peaks were manually integrated and converted into headspace concentration in μg/L 2-undecanone equivalents by dividing the analyte peak area by the 2-undecanone peak area.

Sensory Analysis A descriptive analysis procedure was used to profle the sensory characteristics of the bitters. A group of 14 panelists, 10 Male, 4 Female, ages 21-35 were recruited from a pool of students and postdoctoral scholars in the department of Viticulture and Enology at the University of California, Davis. Over four training sessions, the panelists met in groups, smelled the bitters blind, and generated, discussed, and pooled descriptors by consensus until a fnal list of 30 terms was agreed upon. Samples were presented as 400 uL bitters in 20 mL deionized water in opaque black wineglasses. In the frst training session, four of the bitters were smelled and discussed; in the second, third, and fourth sessions, six bitters were smelled and discussed, so that each bitters was smelled at least once during the training. References (Table 3.3) were made for each descriptor, and these were smelled and refned over the second, third, and fourth sessions. Over two additional sessions, the descriptors and references were fxed and the panelists analyzed the intensity of each descriptor for each bitters in a training exercise in sensory booths. Te descriptive analysis proper was performed in triplicate, with each panelist smelling each reference, then rating the intensity of the aroma of each reference in each bitters over six sessions on an unstructured 9-cm line scale from “low intensity” to “high intensity.” Te panelists were

65 Table 3.3 Sensory terms and references used in the descriptive analysis on bitters.

Descriptor Reference aroma intensity overall intensity cardamom 4 crushed green cardamom pods grapefruit 2 cm * 8cm strip fresh grapefruit peel, oils manually expressed into glass frst molasses 10 mL molasses chocolate 10g shaved dark chocolate (Valhrona) celery seed 2g celery seeds, crushed cola 20 mL cola soapy 1 g unscented ivory soap root beer 20 mL root beer (Virgil’s) orange candy 5 orange jelly beans, halved (Jelly Belly) green 2 g each fresh cilantro leaf, fennel, and cucumber, brunoised tea 2 g black English Breakast tea leaves (Peet’s) brown sugar 5 g brown sugar (C&H) lime peel 1 cm * 5 cm strip fresh lime peel, oils expressed manually into glass frst black pepper 6 black peppercorns, lightly crushed alfalfa hay 1g alfalfa juniper 3 dried juniper berries, crushed mint 2 peppermint and 2 spearmint leaves (fresh) ginger 2cm * 2cm * 3mm piece of fresh ginger, minced orange peel 2 cm * 8cm strip fresh orange peel, oils manually expressed into glass frst earthy 5 g freshly dug soil with 2 mL water dried fruit 10 golden raisins, 2 dried apricots, 2 dried cherries (Trader Joe’s) anise 1 star anise pod cinnamon 2 g cinnamon powder wood 2g medium toasted oak chips (Evoak) clove 3 cloves nutmeg 1 g shaved nutmeg pod caraway 1 g caraway seeds, lightly crushed vanilla 2 cm length vanilla pod + 500 uL vanilla extract (Nieman-Massey) chile 1 g dried ancho chile, chopped

66 presented with eight samples per session, in lidded, opaque, black wineglasses under red light with random 3-digit codes as labels in a Williams Latin Square presentation design. Descriptive analysis was performed using FIZZ (Biosystemes, Couternon, France).

Statistical Analysis Te sensory data was subjected to a 3-way analysis of variance (ANOVA) with 2-way interactions for all 30 descriptors in the R statistical package. Te main efects were product, judge, and replicate. For descriptors with a signifcant Judge*Product interaction, a pseudo- mixed model (with Mean Square of Judge*Product replacing Mean Square of Error in the F-value calculation for Product efect) was used. Products were considered signifcantly diferent in a given aroma when p<0.05. For signifcant descriptors, mean values for each descriptor-product pair were calculated over all judges and replicates and a principal component analysis (PCA) was performed on the mean data in R. Mean ratings for each descriptor and normalized mean relative headspace concentrations of each volatile in each sample were analyzed with Partial Least Squares Regression (PLS) in Unscrambler, with the sensory data being set as the dependent variable to the independent-variable headspace volatile dataset. For purposes of data interpretation, aroma descriptors referenced for individual compounds were taken from the Perfavory and Flavornet websites (Luebke 2014, Acree and Arn 2004).

Results & Discussion

Sensory Analysis Te panel agreed upon 30 aroma descriptors for the bitters, listed in table 3.3 with their references. Of these, all except nutmeg were signifcantly diferent among samples by the ANOVA (p<0.05). Many of the descriptors, such as clove, cinnamon, and celery seed refected ingredients commonly used in bitters. Others, such as cola, soapy, root beer, and earthy, refected non- ingredient aromas that may arise from perceptual blending of the mixtures.

Mean values of sensory qualities for each sample each for sensory of qualities values Mean Table 3.4 Table 68 Figure 3.1 Principal Component Analytsis (PCA) of bitters. PC 1 (X-axis) explains 40.8% of variance in sensory data; PC 2 (Y-axis) explains 30.2% of variance. Sensory descriptors are in red italicized text. Samples are in bold capital letters, coded by style: citrus in orange, aromatic in purple, tiki in blue, mole in brown, New Orleans-style in dark red, and celery in green.

Principal Component Analysis (Figure 3.1) was performed to describe latent interrelationships between the bitters samples and aroma descriptors. Together the frst two dimensions account for 71% of the sensory variance in the bitters. In the PCA the bitters are separated into roughly 3 “lobes” or groups; PC 1 (the x-axis) accounts for 40.8% of the variance and represents a continuum from orange/citrus aromas on the lef to green/celery aromas on the right with spice and other aromas in the middle. PC 2 (the y-axis) accounts for 30.2% of the variance and represents a continuum from botanical notes from citrus and plants on the bottom to chocolate, cinnamon, and other spice aromas at the top. Te bitters were separated fairly well by category; the citrus bitters (orange-colored text) grouped together as did (separately) the celery (green-colored text) and New Orleans-style bitters (maroon-colored text). Te aromatic (purple text), tiki, (light blue text) and mole-style (brown text) bitters grouped close to each other, with 69 some overlap, with, for example, the cinnamon-heavy mole bitters grouping close to the similarly cinnamon-forward ‘Elamakule Tiki and Whiskey-Barrel Aged Bitters. Angostura and especially Boker’s (purple text) bitters both plotted very close to the center of the plot, suggesting that as a style they are close to a kind of “average” style or are blended such that no one aroma descriptor dominates their favor.

Gas Chromatography-Mass Spectrometry 148 compounds were found across the set of bitters, with a minimum of 19 compounds, a maximum of 78 compounds, and a mean of 46 compounds per sample of bitters. Te compounds identifed included aldehydes; simple aromatic compounds; esters; terpenes and sesquiterpenes, and derivatives thereof (alcohols, acetates, esters, etc); and phenylpropenes. A list of peaks can be found in table 3.5, with their headspace concentrations (in 2-undecanone μg/L equivalents) in table S3.1. Twenty-three of these compounds were detected in only one sample out of the sixteen, while in aggregate each compound was detected in an average of fve of the samples (SD=3.96). Te most common compounds were: camphene, beta-phellandrene, nonanal, alpha- p-dimethylstyrene, octyl acetate, ethyl nonanoate, ethyl benzoate, geranyl acetate, and safrole, present in nine samples; alpha-thujene, neryl acetate, nerolidol, and eugenol, present in ten samples; caryophyllene, myristicin, and an unidentifed compound (“ni.d”), present in eleven samples; anethole and methyleugenol, present in twelve samples; gamma-terpinene, octanal, ethyl octanoate, alpha-terpinyl acetate, and an unidentifed compound (“ni.j”), present in thirteen samples; p-cymene and terpinolene, present in fourteen samples; limonene, eucalyptol, linalool, and bornyl acetate, present in ffeen samples; and decanal, present in all sixteen samples. A number of these highly shared compounds vary in concentration across the dataset by several orders of magnitude. For example, decanal was detected in all samples, but there is a 448-fold diference in headspace concentration between the samples with the least (Peychaud’s) and most (Scrappy’s Orange) decanal. Limonene, detected in ffeen out of the sixteen samples, shows a similarly high 598-fold diference in headspace concentration, between ‘Elamakule Tiki

70 Table 3.5 Compounds identifed by GC-MS in samples of bitters, by name, CAS number, Retention Time (RT), and Calculated (CRI) C8-C20 Retention Index (RI). ni=not identifed Literature RI # namea CAS RT CRI Pherobase Flavornet other B1 2-methylbutanal s 96-17-3 3.12 831 864 912 B2 3-methylbutanal s 590-86-3 3.18 903 912 910 B3 ethyl propanoate s 105-37-3 3.85 920 950 951 B4 ethylisobutyrate s 97-62-1 4.03 940 955-972 955 B5 alpha-thujene 2867-05-2 5.52 1014 1038 1021 B6 alpha-pinene s 80-56-8 5.54 1014 1027-1034 1032 B7 2-butanol s 78-92-2 5.65 1018 1022 B8 toluene s 108-88-3 5.76 1021 1042 B9 ethyl butanoate s 105-54-4 5.86 1024 1022-1057 1028 B10 ehtyl 2-methylbutyrate s 7452-79-1 6.41 1040 1056-1069 1050 B11 camphene s 79-92-5 6.63 1047 1077 1075 B12 ethyl isovalerate s 108-64-5 7.15 1063 1053-1082 1060 B13 hexanal s 66-25-1 7.30 1067 1067-1093 1084 B14 isobutanol s 78-83-1 8.00 1088 1099 B15 Alpha-thujene 2867-05-2 8.32 1097 1038 1059 B16 ni.a 8.44 1101 B17 sabinene 3387-41-5 8.69 1106 1123 B18 isoamyl acetate s 123-92-2 9.00 1112 1118-1147 1117 B19 ethyl pentanoate s 539-82-2 9.53 1123 1120-1170 1133 B20 3-carene s 13466-78-9 9.87 1130 1148 1148 B21 Beta-pinene s 127-91-3 10.59 1144 1113-1124 1116 B22 beta-myrcene s 123-35-3 10.72 1147 1161-1187 1145 B23 sabinene 3387-41-5 10.72 1147 1178 1178 B24 heptanal s 111-71-7 11.85 1170 1197 1174 B25 limonene s 138-86-3 12.40 1181 1198-1234 1201 B26 beta-phellandrene s 555-10-2 12.53 1184 1241 1209 B27 eucalyptol s 470-82-6 12.57 1184 1214-1224 B28 isoamyl alcohol s 123-51-3 12.88 1191 1169-1247 1205 B29 ethyl hexanoate s 123-66-0 13.89 1220 1224-1270 1220 B30 beta-trans-ocimene 3779-61-1 13.90 1221 1242 1242 B31 gamma-terpinene 99-85-4 14.16 1226 1262-1265 1238 B32 ni.b 14.32 1230 1555 B33 styrene (cinnamene) 100-42-5 14.41 1232 1273 1241 B34 cis-beta-ocimene 3338-55-4 14.59 1236 1225-1245 1245 B35 o-cymene 527-84-4 15.14 1248 1267 1261 B36 hexyl acetate s 142-92-7 15.61 1259 1264 1270 B37 terpinolene 586-62-9 15.78 1262 1275-1297 1284 B38 octanal s 124-13-0 16.20 1272 1300-1307 1280 B39 1-octen-3-ol s 3391-86-4 16.59 1280 1305-1323 1285 B40 ni.c 16.72 1283 B41 E-2-Heptenal 18829-55-5 17.67 1301 1336 B42 Ethyl-E-3-hexenoate 26553-46-8 17.88 1309 1301 B43 ni.d 17.98 1312 B44 ethyl heptanoate s 106-30-9 18.10 1315 B45 6-Methyl-5-heptene-2-one s 110-93-0 18.29 1319 1319 B46 ni.e 19.36 1343 B47 ni.f 20.06 1359 a= mass spectrum matched by NIST>80% b=Lee et al. 2005; c=Jones et al. 2011; d=Nakata et al. 2013; s= matched to authentic standard 71 Table 3.5 Continued: Compounds identifed by GC-MS in samples of bitters, by name, CAS number, Retention Time (RT), and Calculated (CRI) C8-C20 Retention Index (RI). ni=not identifed Literature RI # namea CAS RT CRI Pherobase Flavornet other B48 heptyl acetate s 112-06-1 20.13 1360 1370 B49 ni.g 20.28 1364 B50 fenchone 1195-79-5 20.60 1371 1402-1410 B51 ni.h 20.64 1372 1560 B52 nonanal s 124-19-6 20.83 1376 1402-1415 1385 B53 ni.i 21.07 1392 B54 Benzene, pentyl- 538-68-1 21.84 1400 1433 B55 alpha,p-dimethylstyrene 1195-32-0 22.39 1412 1414 B56 ethyl octanoate s 106-32-1 22.81 1422 1422-1446 1436 B57 ni.j 23.11 1429 B58 linalool oxide s 5989-33-3 23.24 1432 1423 B59 p-methone 89-80-5 23.50 1438 1440 B60 fenchyl acetate 13851-11-1 23.86 1447 1443 B61 methyl nonanoate s 1731-84-6 23.88 1448 1487 B62 trans-sabinene-hydrate 17699-16-0 24.09 1453 1459 B63 ni.k 24.25 1456 B64 octyl acetate s 112-14-1 24.48 1462 1478 B65 menthone s 14073-97-3 24.62 1465 1454-1478 B66 copaene 3856-25-5 24.85 1471 1488 B67 ni.l 24.90 1472 B68 Alpha-cubebene 17699-14-8 25.15 1478 1463-1480 B69 decanal s 112-31-2 25.32 1482 1447-1510 B70 camphor s 76-22-2 25.40 1484 1498 B71 methyl nonanoate s 1731-84-6 25.42 1484 1487 B72 benzaldehyde s 100-52-7 25.59 1488 1525 1495 B73 ni.m 26.00 1498 1753 B74 ni.n 26.60 1513 B75 ethyl nonanoate s 123-29-5 27.00 1523 1528 B76 linalool s 78-70-6 27.50 1535 1484-1570 B77 methyl decanoate s 110-42-9 27.68 1540 1590 B78 isomenthyl acetate 20777-45-1 27.82 1543 1597b B79 linalyl acetate s 115-95-7 27.96 1547 1569 B80 1-octanol 72-69-5 27.97 1547 1557-1566 1553 B81 bornyl acetate 76-49-3 28.34 1556 1580 B82 nonyl acetate 143-13-5 28.62 1563 1585 B83 caryophyllene s 13877-93-5 28.90 1570 1608-1618 B84 Alpha-bergamotene 13474-59-4 28.97 1572 1586c B85 undecanal s 112-44-7 29.51 1585 1624 B86 lavandulyl acetate 20777-39-3 30.02 1598 1597 B87 Acetophenone s 98-86-2 30.41 1608 1645 B88 methyl 4-decenoate 7367-83-1 30.63 1614 1622 B89 decanal diethyl acetal 34764-02-8 30.79 1618 B90 menthol s 89-78-1 30.98 1623 1626 B91 ethyl decanoate s 110-38-3 31.04 1625 1630 1655 B92 (-)-trans-Pinocarvyl acetate 33045-02-2 31.24 1630 1638 B93 ni.o 31.39 1634 B94 ethyl benzoate s 93-89-0 31.42 1635 1648 a= mass spectrum matched by NIST>80% b=Lee et al. 2005; c=Jones et al. 2011; d=Nakata et al. 2013; s= matched to authentic standard

72 Table 3.5 Continued: Compounds identifed by GC-MS in samples of bitters, by name, CAS number, Retention Time (RT), and Calculated (CRI) C8-C20 Retention Index (RI). ni=not identifed

Literature RI # namea CAS RT CRI Pherobase Flavornet other B95 estragole 140-67-0 31.61 1640 1655 B96 humulene 6753-98-6 31.62 1640 1680 1663 B97 citronellyl acetate 150-84-5 31.78 1645 1607-1663 1607 B98 E-beta-farnesene s 18794-84-8 31.95 1649 1658-1674 B99 Alpha-himchalene 3853-83-6 32.28 1658 1649 B100 isolongifolan-8-ol 1139-08-8 32.36 1660 B101 alpha-terpinyl acetate 80-26-2 32.80 1672 1700 B102 Gamma-muurolene 30021-74-0 32.80 1672 1684 B103 Beta-eudesmene 17066-67-0 32.83 1672 1711 B104 decyl acetate 112-17-4 32.90 1674 1691 B105 p-menth-1-en-8-ol s 98-55-5 32.95 1682 1669-1720 1688 B106 gamma-selinene 515-17-3 32.98 1686 1724 1711 B107 Eremophilene 10219-75-7 33.50 1690 1744d B108 dodecanal s 112-54-9 33.53 1691 1700-1722 B109 alpha-muurolene 31983-22-9 33.83 1699 1727 1714 B110 carvone s 99-49-0 33.96 1703 1715 1720 B111 Alpha-selinene 473-13-2 34.01 1704 1724 B112 nerol acetate 141-12-8 34.09 1706 1728 B113 ni.p 34.81 1726 1808 B114 beta-cadinene 523-47-7 35.00 1731 1749-1752 B115 geranyl acetate s 105-87-3 35.22 1737 1711-1760 B116 ni.q 35.68 1750 B117 perilla aldehyde 2111-75-3 35.70 1750 1797 1765 B118 citronellol s 106-22-9 35.79 1753 1737-1786 B119 Alpha-curcumene 644-30-4 35.84 1771 1777 1773 B120 nerol s 106-25-2 37.99 1780 1753-1770 1770 B121 anethole s 104-46-1 37.25 1793 1808 B122 calamene 1406-50-4 37.55 1801 1826-1837 B123 safrole s 94-59-7 38.75 1836 1863b B124 Alpha-calacorene 21391-99-1 40.62 1889 1906-1916 B125 caryophyllene oxide s 1139-30-6 42.52 1969 1999 B126 perilla alcohol 536-59-4 42.73 1979 2003b B127 methyleugenol s 93-15-2 42.94 1988 2007b B128 safrole s 94-59-7 43.05 1994 B129 cinnamaldehyde 104-55-2 43.11 1996 2017 B130 nerolidol 7212-44-4 43.44 1998 1961-2054 B131 cubenol 21284-22-0 43.58 >2000 1993 B132 ethyl tetradecanoate s 124-06-1 43.62 >2000 2042 B133 Elemol 8024-27-9 43.83 >2000 2089 B134 cinnamyl acetate 103-54-8 44.25 >2000 2104 B135 eugenol s 97-53-0 44.37 >2000 2141-2192 2141 B136 eudesmol 473-15-4 44.47 >2000 2182 B137 Tau-cadinol 5937-11-1 44.53 >2000 2165 B138 tau-muurolol 19912-62-0 44.57 >2000 2178 B139 carvacrol 499-75-2 44.67 >2000 2206b B140 elemicin 487-11-6 44.85 >2000 2167b a= mass spectrum matched by NIST>80% b=Lee et al. 2005; c=Jones et al. 2011; d=Nakata et al. 2013; s= matched to authentic standard 73 Table 3.5 Continued: Compounds identifed by GC-MS in samples of bitters, by name, CAS number, Retention Time (RT), and Calculated (CRI) C8-C20 Retention Index (RI). ni=not identifed

Literature RI # namea CAS RT CRI Pherobase Flavornet other B141 Beta-eudesmol 473-15-4 44.92 >2000 2248 B142 EugenolAcetate 93-28-7 44.93 >2000 B143 myristicin 607-91-0 45.06 >2000 2257b B144 isoeugenol s 97-54-1 45.47 >2000 2365-2367 B145 ni.r 46.11 >2000 B146 Apiol 523-80-8 46.69 >2000 B147 nootkatone s 4674-50-4 46.84 >2000 2573 B148 benzyl benzoate 120-51-4 47.42 >2000 2071 a= mass spectrum matched by NIST>80% b=Lee et al. 2005; c=Jones et al. 2011; d=Nakata et al. 2013; s= matched to authentic standard

74 (least) and Scrappy’s Celery (most), as does anethole, with a 634-fold diference in headspace concentration between Bitter Truth’s Creole bitters and Bittermen’s Celery. As is the case for many other alcoholic beverages that incorporate multiple plant products (for example, gin, absinthe, or chartreuse), the exact formulations of commercially sold bitters are generally kept entirely or partially secret by the companies that make them (Tonutti et al. 2011). Tis means that the volatile profles characterized in the current work cannot be compared directly to volatile profles of the plant products used in each sample, because the complete lists of plant products and their proportions used are not made publicly available. However, while exact formulations are not available, several historical recipes that have inspired current formulations are available. Table 3.1 lists 57 botanicals commonly used in bitters, with use in specifc recipes or other sources listed. Boker’s Bitters, Jerry Tomas’ Own Decanter Bitters, and Regan’s Orange Bitters are all historical versions of samples used in this study (A1, A3, and C1 in Table 3.2, respectively) and Stoughton’s Bitters is a 19th-century version of Stoughton’s Great Cordial Elixir, the patent medicine frst marketed in 1690 that was a precursor to cocktail bitters (see introduction and Wondrich 2007 for more information). Table 3.6 contains reported volatile compositions of a subset of fourteen of these ingredients, each used explicitly in at least one bitters recipe, with the ten most abundant volatiles (where available, listed as reported percentages of total peak area) listed with their relative abundances. Ofen, herbal components are designated separately as “bittering” and “favoring” agents (Parsons 2011), with “bitter” components being less frequently analyzed for their volatiles. However, in the case of gentian and cinchona, common as bittering agents, volatile profling has been performed (Chialva et al. 1985, 1986) which shows that these “bittering” agents also produce a number of compounds which may contribute to the aroma of bitters which use them in their recipe. A number of the compounds detected in a majority of the samples—decanal, limonene, nerol, eugenol, eucalyptol, linalool, gamma-terpinene, caryophyllene, cymene, and anethole—are major components of one or several of these named and commonly used plant ingredients.

75 omas 1862; P=Parsons et al. 2011; W=Wondrich, 2007; O=Orange bitters bitters 2007; O=Orange 2011; W=Wondrich, et al. 1862; P=Parsons omas T T= omas 1862); S=Stoughton’s bitters (Wondrich 2007); B=Boker’s bitters, (Wondrich 2007). (Wondrich bitters, 2007); B=Boker’s (Wondrich bitters 1862); S=Stoughton’s omas T omas’ Own Decanter Bitters, ( Bitters, Own Decanter omas’ T Volatile components reported as literature headspace abundances (% of total peak area) in ingredients used for historical and literature recipes recipes literature and historical used for ingredients in area) peak total (% of abundances headspace as literature reported components Volatile Table 3.6 Table available. where reported compounds abundant most Ten bitters. for 2003); J=Jerry(Regan 76 In some cases, a compound was present in a few samples at relatively similar concentrations—for example, alpha-curcumene was detected in four samples, within a 1.5- fold concentration range, which could mean that the material(s) used in the recipes for these samples that contained alpha-curcumene were simply not used for the other samples in which the compound was not detected. Conversely, the presence of limonene in almost every sample, but in a concentration range covering nearly three orders of magnitude, could refect several diferent paths to the presence of limonene in bitters, which are generalizable for other compounds present in multiple ingredients. Limonene is a component of many ingredients commonly used for bitters, for example sweet orange, cardamom, gentian, caraway, cinchona, coriander, lemon, and bitter orange, and so the total concentration of limonene in, for example, orange bitters, will come from the additive contributions of the limonene extracted from multiple sources, in this case, orange peel, cardamom, gentian, caraway, cinchona, and coriander. A high concentration of limonene in one sample could be from, for example, the use of one material (e.g. citrus peel) that is very high in limonene, or from the use of several materials each with an intermediate concentration of limonene. Contributions of one compound from several botanical sources may lead to concentration- and mixing-dependent sensory attributes that are not present in any of the raw materials. Chemical-sensory relationships are discussed further, below.

Flavor Chemistry of Bitters Samples made with multiple aromatic plant components, such as bitters, have complex chemical compositions with many overlapping compounds among samples as well as many “orphan” compounds present in only one or a few samples. Tey also have complex aromas, with multiple, difering sensory characteristics and many compounds contributing to these sensory characteristics. While some of these sensory characteristics may have strong correlations to one particular compound (and vice versa), it is likely that perceptual interactions involving multiple compounds and multiple aromas play a role in the overall favor of these samples.

77 Figure 3.2 Plots of Partial Least Squares Regression (PLS) analysis of bitters volatile composition and sensory qualities by descriptive analysis. Sample names from Table 3.2; Sensory Attributes from Table 3.3; Compound Identifcations in Table 3.5. 3.2A: Positions of samples 3.2B: Biplot of sensory descriptors (red) and volatiles (blue) 3.2C: Exploded view of compounds

B 78 C

79 Given that many independent variables (volatile molecules) are interacting to produce many dependent variables (aromas), multivariate regression is an ideal tool to visualize and begin to understand these complex, interacting relationships. To model how diferences in sensory characteristics are produced by variations in volatile composition, a Partial Least Squares regression (PLS) was performed on the sensory and volatile datasets on bitters (Figure 3.2). Te PLS identifes variance in the volatile (independent-variable) dataset and uses it to explain as much as possible of the variance it identifes in the sensory (dependent-variable) dataset. In this way, two types of variance are explained: the independent-variable variance, and the dependent- variable variance. In this particular regression analysis, the frst two principal components explained 23% of the independent, chemical variables (13% in PC1 and 10% in PC2), but this 23% of variance was able to explain 60% of the variance in the sensory dataset (36% in PC1, 24% in PC2) (Figure 3.2). Tis means that the relationship between chemistry and aroma in these samples of bitters is complex enough that, when chemical data is taken into account, two Principal Components capture about 10% less of the variance in the sensory dataset than in the sensory-only PCA, which explained 70% of the variance in the frst two PCs. Conversely, this also means that a relatively small amount of the variance in chemistry (23%) is able to explain a large portion of the sensory variance, and explain it spatially in such a way that mimics the spatial explanation provided by the sensory-only . Te general shape of the sensory PCA is preserved in the PLS (Figure 3.2B); with PC1, the x-axis, defned by the contrast from citrus aromas on the lef to celery seed on the right, with herbal and spicy aromas in the middle; and with PC2, the y-axis defned by the progression from more citrus- and green-type aromas on the bottom of the biplot and spicy and chocolate aromas on the top of the biplot. In terms of the placement of the bitters samples themselves, the 3-lobed grouping of the plot is somewhat preserved (Figure 3.2A), with (working counterclockwise from approximately the lower lef quadrant) citrus, celery-New Orleans, and aromatic-tiki-mole in the same fairly distinct areas as in the PCA. However, there is more overlap between these groups in

80 the PLS – the Regan’s Orange and Angostura positions create overlap between the citrus group and the aromatic group, and there is more overall intercalation between the tiki, aromatic, and mole groups. Te distribution of volatiles in the plot (fgure 3.2C) mimics the 3-lobed shape of the samples and the sensory descriptors, with some compounds plotting between groupings of samples. Limonene, for example, is present at a high concentration in both citrus and celery bitters (table S3.1), and is located between these two groups of samples in the plot. Te sample set contains volatiles from several classes of compounds, with aliphatic aldehydes, phenylpropenes, terpenoids, and sesquiterpenoids being the four most abundant of these. Generally, there is not a strong association between any chemical class overall and a particular sensory characteristic, sample, or area of the plot. Tis means that, within this dataset, there aren’t “terpenic” or “aldehydic” favors so much as trends, associations, and diferences within each chemical classes correlating to sensory diferences in samples. In some cases the spatial relationships between sensory descriptors and compounds are directly relatable to aroma characteristics of the compounds in isolation. For example, the (aliphatic, non-terpenic) aldehydes separate into roughly three groups, mimicking the partitioning of the dataset as a whole into three groups. Along the citrus-related “arm,” with the descriptors lime peel, orange peel, grapefruit, and orange candy, dodecanal, octanal, decanal, and nonanal, are grouped together. All of these compounds, when isolated have some citrusy (as well as fatty) aroma characteristics (Acree and Arn 2004). Conversely hexanal and heptanal, group with the green, alfalfa-hay, and celery seed descriptors, suggesting that their “greener” (Acree and Arn 2004, Luebke 2014) aroma characteristics are more emphasized in these samples. Octanal and nonanal are also described as having partially green aromas, but their citrus characteristics are emphasized in these samples. Overall, the C6-C12 aldehydes appear to generally decrease in green qualities and increase in citrus qualities in these bitters as chain length increases. 2- and 3-Methylbutanal, which are described as having some chocolatey and nutty characteristics (Acree and Arn 2004, Luebke 2014) group with the aroma descriptors that include chocolate.

81 A similarly well-defned trend can be seen for the phenylpropenes, a class of aromatic compounds with a conserved allylbenzene structure and various other functional groups. Unlike the aldehydes, which separate into three groups, the phenylpropenes map more like a continuum along, roughly, low PC1-high PC2 to high PC1-low PC2. Te low-PC1/high PC2 phenylpropenes are eugenol derivatives, have similar clove-spicy type aromas, and overlap with the aromatic-tiki- mole group of bitters, and the spice-cola-chocolate group of aroma descriptors. Further along PC1, anethole, estragole, chavicol, safrole, isosafrole, and myristicin group with the New Orleans bitters and the far right edge of the aromatic/mole bitters, as well as the anise, nutmeg, woody, caraway, green, and alfalfa-hay aroma descriptors. While these compounds are not as chemically similar overall as the eugenol group, they on their own tend to have woody-anisic characteristics with some spiciness, which lines up conceptually with their proximate aroma descriptors in the PLS. Finally, apiol, which has an herbal-parsley aroma (Perfavory) and is found in dill, parsley, and celery, plots on its own with an extremely strong correlation to the celery seed descriptor. With the esters, terpenes, and sesquiterpenes, the relationships between compounds and aromas are more complicated to explain than for aldehydes and phenylpropenes. Tis may arise from a number of factors – for one, the esters, terpenes, and sesquiterpenes encompass more compounds per group than the other compound classes, and so each class encompasses more points on the plot. Tis means that while the shape of the chemical data – terpenes, sesquiterpenes, and esters alike – mimics the shape of the sensory product/attribute data, there are enough compounds plotted and enough diversity in their aroma, even between fairly proximal compounds, that trends are not immediately visible. Another factor contributing to the complexity of the terpene, sesquiterpene, and ester PLS data may have root in the odor complexity, conceptual similarity of aroma, and shared structural features of the compounds within these groups. While the relationship between odorant structure and odor quality is complex and poorly understood, recent research has shown that molecular complexity of odorants tends to correlate with odor complexity – specifcally, odorants with greater molecular complexity tend, by both experts and naïve panelists, to require more terms

82 for their full description compared to odorants with lesser molecular complexity (Kermen et al. 2011). Some of the correlations between terpenic compounds and aroma descriptors appear straightforward – the association of the strongly orange-smelling limonene and grapefruit- smelling nootkatone with the citrus descriptors, for example. Non-substituted terpenes tend to plot towards the negative side of PC2, though there are a few on the positive end of PC2, while oxygenated terpenes (alcohols, aldehydes, ketones, and oxides) appear scattered over the whole area of the PLS plot, with a small area in the cinnamon-cola dominated low PC1-high PC2 quadrant containing mostly these oxygenated terpenes and not other terpenes. Finally, the terpene acetates appear in all three primary lobes of the PLS – citronellyl-, geranyl-and alpha-terpinyl acetate with the citrus group, bornyl acetate with the spicy group, and pinocarvyl acetate with the celery-green group. A similar trend, or rather, lack thereof, is evident with the sesquiterpinic compounds, with both unsubstituted and oxygenated sesquiterpenes appearing along all lobes of the PLS. Considering common literature aroma descriptors for individual pure compounds, similarities and overlaps are common within chemical classes. It is common for terpenic compounds to have aroma characteristics that include some subset of woody, citrus, foral, herbal, spicy, and green; for example, citronellyl acetate has foral, rosy, green, fatty, woody, tropical fruit, aldehydic, and citrus characteristics; linalool has citrus, orange, green, woody, aldehydic, foral, terpy, and waxy characteristics; and beta-pinene has fresh, green, piney, woody, haylike, terpy, minty, spicy, and resinous characteristics (All descriptors from Luebke, 2014). In this particular dataset, however, these compounds are each associated with three diferent, distant areas of the PLS, despite their aroma similarities as pure standards—citronellyl acetate with the citrus bitters and orange candy descriptor; linalool with the higher-variance aromatic bitters and the cola and cinnamon descriptors; and beta-pinene with the lower-variance aromatic and citrus bitters and the aroma descriptors ginger, cardamom, juniper, and soapy. Being built from modifcations of basic 5-carbon isoprene units, diferent terpenic compounds will ofen share structural

83 similarities with each other, which may be the source of the shared sensory characteristics these compounds have when isolated. Tat these compounds contribute diferently to diferent sensory qualities, or have diferent aspects of their own sensory qualities emphasized, may have to do with concentration efects or synergistic or masking efects. Tese make straightforward sensory contributions more difcult to predict, and especially so in samples whose analysis is novel and therefore have fewer previous studies to be compared to. Pinpointing the direct relationship between volatiles which have similar and overlapping but not identical smells and the characteristic notes of the mixture remains a challenge. For example, why does the citrus aroma of citronellyl acetate lead to its correlation to citrus-like aromas in these samples, but not linalool, which alone also has citrusy aroma qualities but is not associated with citrus aromas in this study? Novel sensory or psychophysical approaches may be necessary for determining which specifc qualities of molecules are emphasized in a particular mixture, and how mixing afects the aroma contribution of a particular molecule in diferent situations. Chida et al. used cross-matching tests and correspondence analysis to determine which citrus-related compounds were more representative of lemon, orange, and sudachi fruits (Chida et al 2006); in some cases, the most representative compounds for each fruit were only present in that fruit, such as citral in lemons, in other cases, compounds which sensorially were particularly representative for one type of fruit were present in all three fruits, such as alpha-pinene for sudachi aroma, or linalool for orange. Again, characteristic qualities of these plants or products may have a signifcant contribution from mixing efects, as well (Francis and Newton 2005, Johnson et al 2013). Generally, in complex mixtures such as these bitters and others of culinary interest, understanding how multiple compounds contribute to the overall aroma perception of the mixture, and which of these efects dominate across multiple samples is of greater import and value than only identifying the specifc aromas of individual compounds, as would be more precisely elucidated with traditional GC-Olfactometry. Since the PLS reveals correlative rather than causative relationships, the correlative data discussed above for terpenic compounds could be showing that one or more of these compounds are highly associated with a particular

84 descriptor or descriptors just by virtue of being present in a sample that is perceived to be high in the quality described by that descriptor, without necessarily acting as an impact compound for that descriptor. Conversely, the PLS may reveal real mixing-dependent perceptual efect where certain aspects of the multifaceted aroma of any one compound may be emphasized above the others, which could be masked depending on the other aroma qualities, or other types of compounds (aldehydes, phenylpropenoids), present. One way to examine this question is to perform GC-Recomposition Olfactometry (GRO, Johnson et al. 2012) on bitters samples where the relationship between one or more compounds and descriptors appears strong but ambiguous. In this technique, the same extraction and separation steps occur as outlined in the methods section of this paper, but the separated sample is selectively recomposed peak-by-peak on a cold trap within the instrument, with one more more peaks excluded from the recomposition as chosen by the operator. By evaluating the recomposition with compounds omitted in comparison to the aroma of no omitted compounds, it is possible to assess the contribution of the omitted compounds to the mixture, specifcally. Based on correlations identifed by the PLS analysis in the current work, in the next chapter experiments in the sensory efects of omitting three compounds—linalool, alpha-terpinyl acetate, and caryophyllene—from a recomposition of the volatiles of one of the samples used in this study are detailed in greater depth.

Conclusions Tis study is the frst attempt to characterize the favor chemistry of cocktail bitters, an historically and gastronomically important product with complex volatile chemistry. 16 bitters samples of six diferent styles were found to have a well-diferentiated range of 29 signifcantly diferent aroma characteristics by sensory analysis, and a total of 148 volatiles across many classes of natural products, some varying by up to three orders of magnitude across the set. Additionally, multivariate statistical analysis was able to predict 60% of the variance in the sensory characteristics in two dimensions with 23% of the variance in volatiles. In a creative context, the chemical and sensory dataset could have interdisciplinary benefts as well, providing new

85 information about a product class widely used by chefs and bartenders that could complement their professional, intuitive knowledge of favor and guide suggestions for further research questions. Additionally, the PCA and PLS, showing latent correlations both positive and negative, reveal unexplored favor combinations– a bitters high in negatively correlated aromas, such as green and ginger or chocolate and grapefruit; or high in an aroma character with low variance, such as black pepper, could be developed as a creative exercise directly inspired by this dataset. As discussed above, there are several unresolved questions relating to aroma and chemistry that impact both the present study specifcally and broader understanding of favor more theoretically and generally. One of these is why and how volatile molecules have several diferent aroma qualities (Chastrette et al. 1988), and the structure-perception relationships and mechanisms by which this phenomenon occurs (Hann et al. 2001, Hendrickson et al. 1987, Zarzo 2011, Haddad et al. 2010, Kermen et al. 2011). How these efects integrate and interact to determine the perceived aroma of complex mixtures is a further question with great import to real-world favor perception (Teixeira et al. 2010, Jinks and Laing 2001, Bushdid et al. 2014). Understanding these properties, both in a mechanistic context and in the context of what humans choose to eat for pleasure, will require continued use and further development of instrumental and statistical tools, as well as more comprehensive analysis of products of gastronomic interest which haven’t been addressed in the favor chemistry literature. Further investigative work on the contributions of specifc compounds in bitters to the aroma of the overall mixture, detailed in the next chapter, shows that these compound-mixture-aroma relationships are somewhat unpredictable, even taking into account descriptive properties of the mixture and the compounds. As alcoholic extractions of mixtures of aromatic plants, bitters are similar to gin, absinthe, chartreuse, or vermouth within the category of alcoholic beverages. All of these products are gastronomically important, chemically and sensorially complex, and are afected by the volatile compositions of many species. Tis may present unique challenges compared to strategies employed for more commonly analyzed spirits such as brandy or whiskey, which have volatile compositions and favors tied to interactions between grapes or grain and yeast, and oak wood

86 barrels and heat, rather than up to several dozen botanical species. Like bitters, these herbal- alcholic products’ favor chemistry has not been extensively analyzed. Gin has had some sensory and volatile profling performed (Sanchez 2011, Riu-Amatell et al. 2008), and absinthe has been the subject of some chemical analysis that has mostly focused on quantifying concentrations of thujone and a few other components from wormwood (Lachenmeier et al. 2007, 2008). Vermouth, chartreuse, and other blended-aromatic-plant-based alcohol such as Campari have not been the subject of any published volatile or sensory studies that the authors have identifed. Characterizing the favor and volatile spaces and interrelationships of commercial versions of these products can guide further investigation into favor development during extraction and distillation phases of production, and experiments characterizing their contributions to craf cocktails and other culinary situations.

References

Acree, T., & Arn, H. (2004). Flavornet. Retrieved December 05, 2014, from http://www.favornet. org/ Baker, C. H. (1939). Te Gentleman’s Companion:Being an exotic drinking book; or, Around the world with jigger, beaker and fask. New York: Derrydale Press. Retrieved from http://books. google.com/books?id=a3l1KQEACAAJ Bovis, N. (2012). Edible Cocktails: From Garden to Glass - Seasonal Cocktails with a Fresh Twist (p. 224). New York: F+W Media. Bushdid, C., Magnasco, M. O., Vosshall, L. B., & Keller, A. (2014). Humans Can Discriminate More than 1 Trillion Olfactory Stimuli. Science, 343(6177), 1370–1372. doi:10.1126/ science.1249168 Chastrette, M., Elmouafek, A., & Sauvegrain, P. (1988). A multidimensional statistical study of similarities between 74 notes used in perfumery. Chemical Senses, 13(2), 295–305. doi:10.1093/chemse/13.2.295 Chialva, F., Frattini, C., Martelli, A., Martini, L., Ivlas, R., Pessione, S. A., Es, Z. (1986). Unusual essential oils with aromatic properties III . Volatile components of gentian roots. Zeitschrif fuer Lebensmittel-Untersuchung und -Forschung, 182, 212–214.

87 Chialva, F., Frattini, C., Martelli, A., Nmr, L., Farmaceutica, I. C., & Rafaello, C. (1985). Unusual Essential Oils with Aromatic Properties II . Volatile Components of Cinchona Bark. Zeitschrif für Lebensmittel-Untersuchung und -Forschung, 181, 32–34. Chida, M., Yamashita, K., Izumiya, Y., Watanabe, K., & Tamura, H. (2006). Aroma Impact Compounds in Tree Citrus Oils : Cross-matching Test and Correspondence Analysis Approach. Journal of Food Science S, 71(1), S54–S58. Clarke, P. (2010, September). Bittersweet Symphony. Imbibe Magazine, 46–53. Deterre, S., Rega, B., Delarue, J., Decloux, M., Lebrun, M., & Giampaoli, P. (2012). Identifcation of key aroma compounds from bitter orange ( Citrus aurantium L .) products : essential oil and macerate – distillate extract. Flavour and Fragrance Journal, 27, 77–88. doi:10.1002/ f.2087 Embong, M. B., Hadziyev, D., & Molnar, S. (1977). Essential oils from spices grown in Alberta. Anise oil (Pimpinella anisum). Canadian journal of plant science, 57(3), 681–688. Ferhat, M. a, Meklati, B. Y., Smadja, J., & Chemat, F. (2006). An improved microwave Clevenger apparatus for distillation of essential oils from orange peel. Journal of Chromatography A, 1112, 121–126. doi:10.1016/j.chroma.2005.12.030 Ferhat, M. A., Meklati, B. Y., & Chemat, F. (2007). Comparison of diferent isolation methods of essential oil from Citrus fruits : cold pressing , hydrodistillation and microwave “ dry ” distillation. Flavour and Fragrance Journal, 22, 494–504. doi:10.1002/f Francis, I. L., & Newton, J. L. (2005). Determining wine aroma from compositional data. Australian Journal of Grape and Wine Research, 11(2), 114–126. doi:10.1111/j.1755-0238.2005.tb00283.x Gopalakrishnan, N., & Narayanan, C. S. (1991). Supercritical carbon dioxide extraction of cardamom. Journal of Agricultural and Food Chemistry, 39(11), 1976–1978. doi:10.1021/ jf00011a018 Grimes, W. (2002). Straight Up Or On the Rocks: Te Story of the American Cocktail (p. 208). New York: Macmillan. Grosso, C., Ferraro, V., Figueiredo, A. C., Barroso, J. G., Coelho, J. A., & Palavra, A. M. (2008). Supercritical carbon dioxide extraction of volatile oil from Italian coriander seeds. Food Chemistry, 111, 197–203. doi:10.1016/j.foodchem.2008.03.031 Guan, W., Li, S., Yan, R., Tang, S., & Quan, C. (2007). Comparison of essential oils of clove buds extracted with supercritical carbon dioxide and other three traditional extraction methods. Food Chemistry, 101(4), 1558–1564. doi:10.1016/j.foodchem.2006.04.009 Haddad, R., Weiss, T., Khan, R., Nadler, B., Mandairon, N., Bensaf, M., … Sobel, N. (2010). Global features of neural activity in the olfactory system form a parallel code that predicts olfactory behavior and perception. Journal of Neuroscience, 30(27), 9017–9026. Haigh, T. (2009). Vintage Spirits and Forgotten Cocktails: From the Alamagoozlum to the Zombie 100 Rediscovered Recipes and the Stories Behind Tem. Beverly, MA: Quarry Books. Retrieved from http://books.google.com/books?id=sCR7wWhM7IQC

88 Hann, M. M., Leach, A. R., & Harper, G. (2001). Molecular complexity and its impact on the probability of fnding leads for drug discovery. J ChemInf Comput Sc, (41), 856–64. Hendrickson, J. B., Huang, P., & Toczko, A. G. (1987). Molecular complexity: a simplifed formula adapted to individual atoms. J. Chem. Inf. Model., (27), 63–67. Jinks, a, & Laing, D. G. (2001). Te analysis of odor mixtures by humans: evidence for a confgurational process. Physiology & behavior, 72, 51–63. Retrieved from http://www.ncbi. nlm.nih.gov/pubmed/11239981 Johnson, A. J., Hirson, G. D., & Ebeler, S. E. (2012). Perceptual Characterization and Analysis of Aroma Mixtures Using Gas Chromatography Recomposition-Olfactometry. (E. M. C. Skoulakis, Ed.)PLoS ONE, 7(8), e42693. doi:10.1371/journal.pone.0042693 Kermen, F., Chakirian, A., Sezille, C., Joussain, P., Le Gof, G., Ziessel, A., Bensaf, M. (2011). Molecular complexity determines the number of olfactory notes and the pleasantness of smells. Scientifc reports, 1(206), 1–5. doi:10.1038/srep00206 Lachenmeier, D. W. (2007). Assessing the authenticity of absinthe using sensory evaluation and HPTLC analysis of the bitter principle absinthin. Food Research International, 40(1), 167– 175. doi:10.1016/j.foodres.2006.09.002 Lachenmeier, D. W., Nathan-Maister, D., Breaux, T. a, Sohnius, E.-M., Schoeberl, K., & Kuballa, T. (2008). Chemical composition of vintage preban absinthe with special reference to thujone, fenchone, pinocamphone, methanol, copper, and antimony concentrations. Journal of agricultural and food chemistry, 56(9), 3073–81. doi:10.1021/jf703568f Luebke, W. (2014a). Beta-Pinene. Perfavory. Retrieved from http://www.thegoodscentscompany. com/data/rw1344431.html Luebke, W. (2014b). Perfavory. Retrieved December 05, 2014, from http://www.perfavory.com/ Luebke, W. (2014c). Citronellyl Acetate. Perfavory. Retrieved from http://www. thegoodscentscompany.com/data/rw1012172.html Luebke, W. (2014d). Linalool. Perfavory. Retrieved from http://www.perfavory.com/docs/ doc1007872.html Padmakumari, K., Sasidharan, I., & Sreekumar, M. (2011). Composition and antioxidant activity of essential oil of pimento (Pimenta dioica (L) Merr.) from Jamaica. Natural product research, 25(2), 152–160. Parsons, B. T., & Anderson, E. (2011). Bitters: A Spirited History of a Classic Cure-All, with Cocktails, Recipes, and Formulas. Berkeley, CA: Ten Speed Press. Retrieved from http:// books.google.com/books?id=OPlRDjfGnloC Raina, V. K., Srivastava, S. K., & Syamasunder, K. V. (2003). Essential oil composition of Acorus calamus L. from the lower region of the Himalayas. Flavour and Fragrance Journal, 18, 18– 20. doi:10.1002/f.1136 Regan, G. (2003). Te Joy of Mixology. New York, NY: Crown Publishing Group. Retrieved from http://books.google.com/books?id=7TlhtrpXa-MC

89 Reverchon, E., & Senatore, F. (1994). Supercritical Carbon Dioxide Extraction of Chamomile Essential Oil and Its Analysis. Journal of agricultural and food chemistry, 42, 154–158. Riu-Aumatell, M., Vichi, S., Mora-Pons, M., López-Tamames, E., & Buxaderas, S. (2008). Sensory Characterization of Dry Gins with Diferent Volatile Profles. Journal of Food Science, 73(6), S286–S293. doi:10.1111/j.1750-3841.2008.00820.x Sampson, E., Chittenden, G., & Croswell, H. (1806). Te Balance and Columbian Repository. Hudson, NY: Sampson, Chittenden & Croswell. Retrieved from http://books.google.com/ books?id=M9MRAAAAYAAJ Sanchez, J. V. S. G. (2011). Comparison of Descriptive Analysis and Projective Mapping Techniques in the Aroma Evaluation of the Distilled Spirits , Gin and Tequila. Master’s Tesis, University of California, Davis. 87 pp. Sandham, T. (2012). World’s Best Cocktails: 500 Signature Drinks from the World’s Best Bars and Bartenders. Minneapolis, MN: Fair Winds Press. Retrieved from http://books.google.com/ books?id=bQQNAAAAQBAJ Simmons, M. (2011). DIY Cocktails: A Simple Guide to Creating Your Own Signature Drinks (p. 240). New York: F+W Media. Singh, G., Maurya, S., DeLampasona, M. P., & Catalan, C. a N. (2007). A comparison of chemical, antioxidant and antimicrobial studies of cinnamon leaf and bark volatile oils, oleoresins and their constituents. Food and chemical toxicology : an international journal published for the British Industrial Biological Research Association, 45(9), 1650–1661. doi:10.1016/j. fct.2007.02.031 Teixeira, M. A., Rodriguez, O., & Rodrigues, A. E. (2010). Perfumery Radar : A Predictive Tool for Perfume Family Classifcation. Industrial Engineering Chemical Research, 49(22), 11764– 11777. Tonutti, I., & Liddle, P. (2010). Aromatic plants in alcoholic beverages. A review. Flavour and Fragrance Journal, 25(5), 341–350. doi:10.1002/f.2001 Wondrich, D. (2007). Imbibe!: From Absinthe Cocktail to Whiskey Smash, a Salute in Stories and Drinks to “Professor” Jerry Tomas, Pioneer of the American Bar. New York, NY: Perigee Books. Retrieved from http://books.google.com/books?id=IV3s-NUYnfEC Wondrich, D. (2010). Punch: Te Delights (and Dangers) of the Flowing Bowl (p. 320). New York: Penguin. Yin, Y., Zarghami, N., & Heinz, D. (1970). Efects of pH and temperature on volatile constituents of caraway. Journal of Food Science, 35, 531–533. Zarzo, M. (2011). Hedonic Judgments of Chemical Compounds Are Correlated with Molecular Size. Sensors, 11(15), 3667–3686.

ed, nd=not nd=not ed,

Headspace concentrations of volatiles in bitters, in ug/L 2-undecanone equivalents. ni=not identi ni=not equivalents. 2-undecanone ug/L in bitters, in volatiles of concentrations Headspace

Supplementary Table S3.1 Table Supplementary detected 91 continued Supplementary Table S3.1 Table Supplementary 92 continued Supplementary Table S3.1 Table Supplementary 93 continued Supplementary Table S3.1 Table Supplementary 94 continued Supplementary Table S3.1 Table Supplementary 95 Chapter 4: GC-Recomposition-Olfactometry (GRO) and multivariate study of three terpenoid compounds in the aroma profle of Angostura bitters

Introduction Aromas encountered in everyday life are almost always the result of multicomponent mixtures of volatiles. Te relationship between the perceived aroma of a mixture of volatiles, its chemical composition, and the aromas of its components is complex and afected by several difcult-to- predict factors. In some cases, the aroma of a mixture is not equivalent to the summation of the aroma qualities of its components (Wilson and Stevenson 2006, Laing and Francis 1989); and mixtures with as few as three components have been found to have aroma qualities not found in any of these individual components (Le Berre et al. 2008). One common method for estimating the relative contribution of volatiles in a sample to its perceived aroma is a calculation called an Odor Activity Value (OAV), the measured concentration of a volatile present in a sample divided by its measured sensory detection threshold in a similar matrix (Patton and Josephson 1957, Guadagni et al. 1966). A second method, which assesses the volatiles in a sample directly, uses GC-Olfactometry or CHARM with successive dilutions of a sample (Aroma Extract Dilution Analysis or AEDA), where a human subject evaluates the aroma and intensity of each peak in a GC chromatogram, and a “dilution factor” required to suppress the detectability of each component is calculated (Acree and Barnard 1984, Grosch 1993, Grosch 2001).

96 Both of these separative methods seek to evaluate the potency and quality of a particular volatile at the concentration at which it is found in a particular sample, but cannot address mixing efects. In many cases, these efects are investigated via reconstitution and omission experiments. Having quantifed all the volatiles in a sample based on GC-MS or GC coupled to another type of detector, a model reconstitution is prepared using chemical standards containing those volatiles (at their in-sample concentrations) hypothesized to contribute to the sample’s aroma, as determined by a certain cutof OAV or dilution factor. To evaluate the relative contribution of each of these volatiles, several omission mixtures, each of which excludes one compound, are prepared and their aroma is calculated in comparison to the “whole” reconstitution mixture (Grosch 2001). Tere are several drawbacks to this approach. OAV and dilution factor may not be able to accurately predict whether a compound is above or below its detection threshold, truly “sub-threshold” volatiles may contribute sensory impact when mixed with other volatiles, and instrumental limits of detection may be too high to truly quantitate every compound with a sensory impact. “Sub-threshold” compounds, which have been quantifed below their putative aroma threshold, are typically excluded; though it has been found that these may play a role in the aroma perception of a mixture (Pineau et al. 2007, Ryan et al 2008). Despite containing volatiles calculated to have a sensory impact at their in-sample concentrations, reconstitutions have sometimes been found to difer in aroma from the samples they are supposed to model (Steinhaus et al. 2009), refecting a “reconstitution discrepancy” (Bult et al 2001). While some studies include components calculated to be below their detection threshold in reconstitutions (Bult et al. 2001) this is not necessarily a widespread practice. Practically speaking, the time and expense of quantitation, sourcing standards, and reconstitution and omission mixture preparation can be limiting. In Johnson et al. 2012 (see chapter 2), the technique of and instrumentation required for In-Instrument Gas Chromatography-Recomposition-Olfactometry (GRO) was developed and used to characterize interactive efects of mixing in the production of lavender aroma. Tis

97 allowed for the production of reconstitutions in-instrument from volatiles extracted directly from a sample of lavender, with omissions selected by means of a fow switch in real-time during the course of the chromatographic separation. Lavender-like aroma does not have a character impact compound associated with it, and it was found that lavender was sometimes used as a descriptor for the mixtures containing a larger subset of lavender volatiles, but not for smaller and less complex subset mixtures of these volatiles. Chapter 3 characterized the favor-chemical space of commercial aromatic cocktail bitters, and a PLS2 regression was performed to relate volatile profling data to sensory profling data on the bitters used in the experiment. Te PLS (fgure 3.4) modeled 60% of the sensory variance in the frst two PCs (in a similar spatial confguration as the PCA on the sensory data only, which modeled 70% of that variance in the frst two PCs) with 23% of the variance in volatile composition; there were clear associations between volatiles and sensory qualities suggested by the overall similarities in spatial confguration of the two datasets in the PLS, and by proximity of sensory descriptors to individual compounds and groupings of compounds. Some sensory- chemical relationships appeared fairly straightforward—clove aroma spatially associated with eugenol and its derivatives, which are found in cloves and are ofen described as “clove-like” (cinnamon aroma and cinnamaldehyde is another such familiar or expected relationship)—but the number of compounds (such as terpenoids) present in the dataset without a familiar or straightforward relationship to a single aroma quality lef unanswered many questions about how aroma in these mixtures, and or these compounds, is translated from chemistry. In addition, the presence of complex aroma descriptors such as cola, chocolate, and ginger, which aren’t usually associated with an impact compound, means that some interactive complexity is probably at play in the aroma of bitters. In the current work, specifc compound-aroma and mixture-aroma relationships of three volatile molecules associated with several aroma descriptors were investigated using GRO reconstitution and omission experiments.

98 Figure 4.1 Chromatograms of (top) sample A, the control, uncut sample; and (bottom) sample E, with linalool, terpinyl acetate, and caryophyllene excluded from the reconstitution

Materials and Methods Bitters: Angostura bitters were purchased commercially.

Sample Preparation and Extraction: 2.5 mL of bitters were diluted with 7.5 mL of deionized water in a 20 mL amber glass headspace vial and sealed with a crimp cap with a PTFE-faced silicone septum (Supelco, St.

99 Louis, MO). Te sample was shaken at 500 rpm for one minute, afer which a Solid Phase Microextraction fber (2 cm length, 50/30 um divinylbenzene/carboxen/polydimethylsiloxane coating, Supelco) was immediately used for extraction. Te fber was exposed to the headspace of the vial for 30 minutes at room temperature, then withdrawn and immediately desorbed in the GRO inlet. Instrument and Conditions: An Agilent model 6890 gas chromatograph/5972 mass spectral detector (GC-MSD) modifed with Deans Switch, auxiliary pressure controller, cryotrap, and olfactometry port (as described in Johnson et al. 2012) was used, with the modifcation of an Agilent 3-way splitter (instead of a Gerstel 4-way splitter) splitting the eluant of volatiles between the mass spectrometer and olfactory port. Tere was a 3:1 split between the MSD and olfacotry port, with a 0.75 m long, 0.1 mm i.d. restrictor leading from the splitter to the MSD, and a 1.0 m, 0.18 mm i.d. restrictor from the splitter to the olfactory port. Separation was performed using a 30 m X 50 μm i.d. X 0.32 μm flm thickness DB-5MS column (J&W, Folsom, CA). Te inlet was maintained at 270ºC in splitless mode. Helium was used as the carrier gas and was held at a constant pressure of 12.6 psi. Te auxiliary pressure controller was maintained at 3.4 psi. Te SPME assembly was introduced manually into the inlet and allowed to desorb for a total of 10 minutes. Te oven was held at 60ºC for 3 minutes, then ramped to 150ºC at a rate of 3ºC/min, then ramped to 325ºC at a rate of 30ºC/min and held for 5 min for a total runtime of 49 minutes. Both the olfactory port transfer line and the MSD transfer line were maintained at 300ºC. Afer a 4 minute solvent delay, the mass spectrometer scanned from m/z 50–300. Te eluant from the GC column, minus any retention time segments cut by the Deans Switch, were collected in a cryotrap (Micro Cryo-trap and model 971 controller, Scientifc Instrument Services, Ringoes NJ) using liquid carbon dioxide. Te switch was programmed in the ‘‘runtime’’ tab of the Enhanced Chemstation Sofware (Hewlett Packard, version B.01.00) to direct the fow over the course of the runtime as desired by the operator.

100 Sensory Conditions: Tree volatile compounds—linalool, alpha-terpinyl acetate, and trans-caryophyllene— were chosen based on criteria of functional group and chemical class diversity, minimal co-elution with other compounds in the GC-MS chromatogram, and correlation to sensory characteristics (see “Statistics,” below, and supplementary table S4.1) to be the basis of recomposition mixtures for sensory evaluation. Literature orthonasal detection thresholds in water for caryophyllene and linalool (Guadagni et al. 1966) were used to calculate putative odor activity values, based on headspace oncentrations determined in chapter 3. Five recomposition samples were produced (see Table 4.1) by cutting sections of the chromatogram in real time to waste based on retention time. Sample A incorporated all volatiles with no cuts; sample B incorporated all headspace volatiles in Angostura bitters except linalool; sample C incorporated all volatiles except alpha-terpinyl acetate; sample D incorporated all volatiles except caryophyllene; and sample E incorporated all volatiles except linalool, alpha-terpinyl acetate, and caryophyllene. Tree panelists (2 female, ages 26-33) smelled each recomposition mixture at the olfactory port, rated the aroma intensity of the sample from 1-10, and indicated on a list of aroma descriptors generated by the panel in chapter 3 all descriptors that applied to the sample. Four replicates were performed per panelist. Statistical Analysis: Preliminary identifcation of candidate compounds for omission was performed via PLS1 analyses (Supplementary table S4.1) of the descriptive analysis data and volatile profling data from chapter 3, performed using the whole volatile dataset as the x-variable and each aroma descriptor as a y-variable in Unscrambler. Using the data generated by the three panelists in the current study, the mean aroma intensity for each sample was calculated, and the total number of times each attribute was identifed for each sample was calculated, reported as frequency counts. Correspondence analysis (CA) on the aggregated, check-all-that-apply datasets and Multiple Factor Analysis (MFA) comparing each panelists’ datasets were performed using the “ca” and “FactoMineR”

101 packages for the R statistical program (R Foundation for StatisticalComputing, Vienna, Austria), respectively. Results and Discussion: Based on orthonasal detection thresholds for reported in Guadagni et al. (1966), and relative headspace concentration reported in Table S3.1, the Odor Activity Values of linalool and caryophyllene in Angostura bitters were calculated as 7 and 1.3, respectively. Table 4.1 shows average rated overall aroma intensity for each mixture, and the number of times each descriptor was applied to each mixture reported as frequency counts. Table 4.2 shows these diferences highlighted as a heatmap; diferences in frequency counts close to zero are highlighted in green, increases in frequency counts for a particular descriptor are proportionally bluer, and decreases in frequency counts for a given descriptor are proportionally yellower. For each descriptor in each experimental sample, the number of frequency counts relative to the number of frequency counts for the control sample (sample A, nothing cut, every peak cryotrapped) refects the sensory role that individual compound plays in the aroma of the control mixture. For example, no change in frequency counts for a descriptor when a compound is cut shows that that compound does not impact that aroma (e.g., efect of linalool removal on clove aroma of the recomposition mixture). A decrease in frequency count upon the cutting of a compound shows that that compound contributes to that aroma with larger decreases indicating a greater efect (e.g., decreases in cola aroma upon removal of linalool and alpha terpinyl acetate). An increase in frequency counts upon cutting shows that a compound masks or otherwise dampens or modulates that attribute (e.g., increase in ginger aroma upon removal of linalool from the recomposition mixture). Linalool contributes most to the aromas of cinnamon, root beer, and black pepper in the reconstitution mixtures, refected in the decrease in frequency counts for each of these descriptors when linalool is cut. It makes a small contribution to cardamom, anise, nutmeg, grapefruit, and juniper aroma qualities in the mixture. Finally, linalool masks vanilla, mint, wood, and soapy aroma attributes, which increase in frequency counts when linalool is cut from the mixture.

102 Table 4.1 Aroma properties of GRO reconsitution mixtures listed by volatiles excluded and calcuated odor activity values (OAV) of excluded compounds. “intensity” is average overall aroma intensity rated by three panelits from 1-10 for each mixture. Aroma descriptors expressed as overall counts for each descriptor for each sample across three panelists and four replicates. Sheet3 mixture A B C D E Linalool, Alpha Alpha terpinyl terpinyl acetate, volatiles excluded none Linalool acetate Caryophyllene Caryophyllene overall intensity 5.5 4.4 4.6 5.1 4.8 cola 8 7 4 6 4 ginger 3 4 1 0 3 orange peel 7 6 8 9 6 cardamom 3 1 0 1 4 anise 5 3 4 4 5 clove 8 8 3 10 7 orange candy 6 6 7 5 7 cinnamon 7 3 3 3 4 lime peel 6 6 6 4 2 tea 1 0 0 1 2 vanilla 3 6 3 6 1 nutmeg 5 3 1 3 4 root beer 5 1 2 5 1 dried fruit 1 0 1 0 0 wood 4 6 5 2 5 brown sugar 3 3 4 3 3 molasses 2 1 0 0 0 Black pepper 5 2 3 0 2 grapefruit 2 0 2 0 2 caraway 1 0 1 0 0 juniper 3 1 0 0 2 earthy 1 1 2 4 3 alfalfa/hay 0 0 0 0 0 chile 0 0 0 0 1 celery seed 1 0 0 0 1 mint 4 7 5 7 7 chocolate 1 1 1 2 1 soapy 2 4 2 3 1 green 1 2 2 2 3

103

Page 1 Table 4.2 Diferences in aroma qualities of samples with volatiles excluded compared to the control sample. Decreases in descriptor count for experimental conditions highlighted in yellow, increases in descriptor count highlighted in magenta. descriptors and differences in counts by mix Difference in count from mix A Mixture: B C D E Linalool Alpha Alpha terpinyl terpinyl acetate CUT: Linalool acetate Caryophyllene Caryophyllene overall aroma intensity -1.1 -0.9 -0.4 -0.8 cola -1 -4 -2 -4 ginger 1 -2 -3 0 Orange peel -1 1 2 -1 cardamom -2 -3 -2 1 anise -2 -1 -1 0 clove 0 -5 2 -1 Orange candy 0 1 -1 1 cinnamon -4 -4 -4 -3 Lime peel 0 0 -2 -4 tea -1 -1 0 1 vanilla 3 0 3 -2 nutmeg -2 -4 -2 -1 Root beer -4 -3 0 -4 Dried fruit -1 0 -1 -1 wood 2 1 -2 1 Brown sugar 0 1 0 0 molasses -1 -2 -2 -2 Black pepper -3 -2 -5 -3 grapefruit -2 0 -2 0 caraway -1 0 -1 -1 juniper -2 -3 -3 -1 earthy 0 1 3 2 Alfalfa/hay 0 0 0 0 chile 0 0 0 1 Celery seed -1 -1 -1 0 mint 3 1 3 3 chocolate 0 0 1 0 soapy 2 0 1 -1 green 1 1 1 2

104

Page 1 Alpha-terpinyl acetate contributes most to the aromas of cola, cardamom, clove, cinnamon, nutmeg, root beer, and juniper, with small contributions to ginger, molasses, and black pepper aromas in the reconstitution mixtures. It does not appear to contribute any signifcant masking efects. Caryophyllene contributes most to aromas qualities of ginger, cinnamon, black pepper, and juniper in the reconstitution mixtures, with smaller contributions to cola, cardamom, lime peel, nutmeg, wood, and grapefruit aromas. Caryophyllene masks aromas of vanilla, earthy, mint, orange peel, and clove. Frequency count diferences between the control sample and sample E, which omitted all three test compounds reveal mixing-dependent perceptual efects, where the contribution of once compound to a descriptor is afected by the contributions of the other two, or where omitting all three compounds has an efect that isn’t predictable from examing the individual efects of cutting single compounds. In sample E, cola, clove, cinnamon, root beer, molasses, mint and black pepper aroma qualities all appear to show similar changes when all three compounds are simultaneously cut as when compared to the aroma qualities in aggregate for mixtures B, C, and D, when only one of the compounds is cut. For each of these descriptors, the change in frequency counts for mixture E compared to the control mirrors the sample (B, C, or D) with the largest change in frequency counts, or refects an averaging where cutting one compound increased counts and cutting another decreased them. Lime Peel aroma appears to have some synergistic efects associated with it, as it has no change in frequency counts afer cutting either linalool or alpha-terpinyl acetate, and only a small decrease in counts when caryophyllene is cut; however, there is a marked decrease in frequency counts when all three compounds are cut. Tis suggests that while none of the three compounds has a strong lime peel quality on their own, this aroma characteristic depends in part on the impact of having all three compounds present together in the mixture.

For other aroma qualities—ginger, cardamom, and vanilla— cutting all three compounds in sample E leads to an opposite efect compared to the samples where only a single compound

105 is cut. For ginger, cutting either alpha-terpinyl acetate and caryophyllene leads to a decrease in frequency counts, but cutting all three compounds from the reconstitution mixture leads to no change in the number of counts for experimental sample E as for the control sample A. Te same is true for cardamom, for which all three compounds appear to contribute in isolation (cutting any one of them leads to a small decrease in counts), but cutting all three at once leads to an essentially unchanged descriptor count compared to the control. In the case of vanilla, cutting either caryophyllene or linalool increases the descriptor count, but cutting all three compounds slightly decreases the descriptor count; it is unclear why this would be the case for this odor quality in particular.. Supplementary Table S4.1 summarizes the correlation between the three compounds in this study and selected aroma descriptors for Angostura bitters, as determined by PLS1 analyses of descriptive analysis profling data. For the aforementioned descriptors in the descriptive analysis study, Angostura was found to be correlated to black pepper aroma (and was in the more black pepper-intense grouping of samples in the DA as determined by ANOVA and a Tukey Honest Signifcant Diference (HSD) test) while caryophyllene was not well-correlated to black pepper aroma. Tis lack of correlation contrasts with the results of the current study, which found that cutting caryophyllene markedly decreased descriptor counts for black pepper. Te Angostura bitters sample was negatively correlated to earthy aroma in the dominant frst principal component of the PLS1 on the descriptive analysis data, as was caryophyllene; Angostura was also rated low in earthy aroma compared to other samples. In this experiment, cutting caryophyllene led to an increase in descriptor counts for earthy. In both cases performing GRO on a sample allowed for a rapid, causative determination of the relationship between a particular compound and particular odor qualities, which were examined correlatively in the previous study. Figure 4.2 shows a biplot of the correspondence analysis performed on the frequency data summed over all panelists and replicates, which is useful for comparing diferences when all sensory descriptors are taken into account. In this biplot, mixtures are represented as circles and descriptors as triangles. Positional proximity of a sample to descriptors refects correlations

106 between those descriptors and that sample. Te recomposition mixtures evaluated here have each omitted specifc compounds; therefore the descriptors which are negatively correlated to a sample are the ones to which the compound that has been cut in that sample are positively correlated. For example, mixture D, which omits caryophyllene, plots closely to chocolate, green, and earthy, which suggests that caryophyllene is negatively correlated to those aromas, and caryophyllene itself would plot at a point directly refected through the origin, close to ginger, black pepper, wood, and cinnamon aroma qualities. Tis is also refected in Table 4.2, where omitting caryophyllene (as discussed above) leads to a slight increase in the number of frequency counts for green and chocolate, a slightly larger increase in frequency counts for earthy aroma, and a decrease in the number of counts for cinnamon, black pepper, ginger, juniper, and wood aromas. Spatially, the biplot suggests that in a multivariate context, mixtures B and C are most similar to each other, and each is fairly diferent from mixtures A, D, and E. Tis suggests that omitting linalool and alpha-terpinyl acetate has similar efects overall on the aroma profle, while omitting caryophyllene produces a more diferent efect. And importantly these results indicate that omitting all three compounds either unmasks or reduces aromas diferently than omitting any single compound. Aroma descriptors such as cola, soapy, vanilla, lime peel, brown sugar, orange peel, and orange candy appear to be the most dependent on more than one compound, as they plot directly opposite mixture E, which excludes all three compounds under investigation. Te position of sample D, which omits caryophyllene, puts the position of caryophyllene at a point refected directly through the origin of the plot—essentially, very close to mixture A, the uncut control sample. Tis suggests that, of the three compounds studies, caryophyllene is the most directly responsible for the aromas associated with the whole sample. Figure 4.3 shows a consensus map of samples made by comparing panelists’ data separately, with disagreement over consensus positions denoted by colored vectors for each panelist. Generally there is good agreement about the relative similarities and diferences of each mixture as determined by each panelist, with mixtures A and D each in an isolated quadrant and

107 Figure 4.2 Correspondece analysis biplot of samples (blue circles, blu text) and descriptor counts (red triangles, black text). Dimension 1 (x) Variance explained=37.2%, Dimension 2 (y) variance explained=27.6% Dimension 2: 27.6% variance explained 2: 27.6% variance Dimension

Dimension 1: 37.2% variance explained more spatial similarity for mixtures B, C, and E. Tis plot shows a similar separation of samples A, D, and E from each other as was evident from the correspondence analysis, however, mixtures B and C are somewhat closer and therefore more similar to mixture E than denoted by the correspondence analysis. It should be noted that positions in the Correspondence Analysis are refective of summing the incidence of 12 possible hits (given three panelists and four replicates) for any given mixture-descriptor pair, while each panelist-derived position in the Multiple Factor

Analysis (MFA) consensus map is derived from a maximum of 4 hits. While MFA is a useful check for panelist agreement, the correspondence analysis may be more information-rich as a basis for comparisons among samples. 108 Figure 4.3 Multiple Factor Analysis (MFA) individual factor map comparing panelists’ individual map positions with consensus map positions

GRO was developed with a view to simplifying the elucidation of the role of both individual volatiles to complex aromas, and interactions between volatiles. Interestingly, the results of the current work show that even in experiments focusing on the roles of a limited number of compounds, the sensory results of investigating each are still quite complex. Diferential contributions of compounds to the intensity of some aromas and not others, as well as evidence for masking of other sensory qualities once specifc compounds are removed from the reconstitution, are evident. Negative spatial correlation in the correspondence analysis between

the sample (E) with multiple compounds removed and a number of aromas that do not have an impact compound (such as cola, root beer, and soapy) refects an association between the lack of these compounds and the lack of those aromas, or more simply, an association between all three

109 compounds together and those aromas. Tat these compounds appear to enhance some aromas and suppress others suggests a complex relationship to perception, mediated through mixing, which has also been described elsewhere.(Pineau et al. 2007, Jinks and Laing 2001, Goyert et al. 2007, Lytra et al. 2012, Laing and Francis 1989, Livermore and Laing 1998). Both linalool and caryophyllene have fairly low putative OAVs in this sample, 7 in the case of linalool and 1.3 in the case of caryophyllene, and while caryophyllene in a traditional reconstitution and omission study would therefore be considered to be barely detectible, removing it by GRO tends to have a larger efect on descriptor counts than removing either linalool or alpha-terpinyl acetate. Again, it is important to emphasize that these sensory efects could be analyzed directly, from a SPME extract of the sample of Angostura bitters, without a quantitation step, without determining odor thresholds, and without performing time-intensive aroma extract dilution analysis. Te simplicity of performing GRO could be used to investigate more compound-level sensory interaction efects in complex samples, and allows for greater detail about diferences in multiple descriptors than a simple forced choice diference test reveals. Tis suggests that further development and application of methods, statistical, instrumental, or otherwise, to elucidate interactive and synergistic sources for such aromas in these and other samples (some of which may already be well-characterized by current methods) may reveal relationships that we have not currently theorized.

References

Acree, T. E., & Barnard, J. (1984). A Procedure for the Sensory Analysis of Gas Food Chemistry, 14, 273–286. Bult, J. H., Schifferstein, H. N., Roozen, J. P., Voragen, A. G., & Kroeze, J. H. (2001). The components in a mixture of odorants. Chemical senses, 26(5), 459–469. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/11418491 Goyert, H. F., Frank, M. E., Gent, J. F., & Hettinger, T. P. (2007). Characteristic component odors emerge from mixtures after selective adaptation. Brain research bulletin, 72(1), 1–9. doi:10.1016/j.brainresbull.2006.12.010

110 Grosch, W. (2001). Evaluation of the key odorants of foods by dilution experiments, aroma models and omission. Chemical senses, 26, 533–545. Retrieved from http://www.ncbi.nlm. nih.gov/pubmed/11418500 Grosch, Werner. (1993). Detection of potent odorants in foods by aroma extract dilution analysis. Trends in Food Science & Technology, 4, 68–73. Guadagni, D. G., Buttery, R. G., & Harris, J. (1966). Odour intensities of hop oil components. J. Sci. Food Agric., 17(1), 142–144. Jinks, a, & Laing, D. G. (2001). The analysis of odor mixtures by humans: evidence for a Physiology & behavior, 72, 51–63. Retrieved from http://www. ncbi.nlm.nih.gov/pubmed/11239981 Johnson, A. J., Hirson, G. D., & Ebeler, S. E. (2012). Perceptual Characterization and Analysis of Aroma Mixtures Using Gas Chromatography Recomposition-Olfactometry. (E. M. C. Skoulakis, Ed.)PLoS ONE, 7(8), e42693. doi:10.1371/journal.pone.0042693 Laing, D. G., & Francis, G. W. (1989). The capacity of humans to identify odors in mixtures. Physiology & behavior, 46, 809–814. Retrieved from http://www.ncbi.nlm.nih.gov/ pubmed/2628992 Le Berre, E., Béno, N., Ishii, A., Chabanet, C., Etiévant, P., & Thomas-Danguin, T. (2008). Just noticeable differences in component concentrations modify the odor quality of a blending mixture. Chemical senses, 33(4), 389–395. doi:10.1093/chemse/bjn006 of multicomponent odor mixtures. Perception & Psychophysics, 60(4), 650–661. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/9628996 Lytra, G., Tempere, S., Revel, G. De, & Barbe, J.-C. (2012). Impact of Perceptive Interactions on Red Wine Fruity Aroma. Journal of Agricultural and Food Chemistry, 60(50), 12260– 12269. doi:10.1021/jf302918q compounds in foods. Journal of Food Science, 22(3), 316–318. Pineau, B., Barbe, J.-C., Van Leeuwen, C., & Dubourdieu, D. (2007). Which impact for beta- damascenone on red wines aroma? Journal of agricultural and food chemistry, 55(10), 4103–4108. doi:10.1021/jf070120r odorants in global odour perception. Trends in Food Science & Technology, 19(7), 383–389. doi:10.1016/j.tifs.2008.01.007 Steinhaus, M., Sinuco, D., Polster, J., Osorio, C., & Schieberle, P. (2009). Characterization of the key aroma compounds in pink guava (Psidium guajava L.) by means of aroma re- engineering experiments and omission tests. Journal of agricultural and food chemistry, 57(7), 2882–2888. doi:10.1021/jf803728n Wilson, D. A., & Stevenson, R. J. (2006). Learning to smell: olfactory perception from neurobiology to behavior. Baltimore, MD: Johns Hopkins University Press.

111 Supplementary Table S4.1 PLS1 analysis positions in biplot of compounds and Angostura sample compared to specifc descriptors (positions of descriptors is appoximately 0.2 along each PC), from bitters descriptive analysis data, chapter 3.

112 Chapter 5: Aroma Perception and Chemistry of Bitters in Whiskey Matrices: Modeling the Old- Fashioned Cocktail

Introduction Herbs and other aromatic plants are popular ingredients in strongly-favored alcoholic beverages (Tonutti et al 2010), and the combination of these and other products into cocktails is a foundational aspect of mixology (Regan 2002, Haigh 2009). One of the primary uses of aromatic bitters is their addition to mixed drinks to accent favors and increase aromatic complexity (Parsons 2011, Clarke 2010). What we call an “Old-Fashioned” cocktail today is the simplest and oldest style of cocktail (Grimes 2002, Simonson 2014), and in its most essential form is whiskey (usually bourbon or rye), water, bitters, and a little sugar, combined and served over ice.1 Tis closely mirrors the earliest print defnition of what was then called simply, “Cocktail”: “Cock Tail, then, is a stimulating liquor, composed of spirits of any kind, sugar, water and bitters” (Sampson et al 1806). As drinks-mixing became more elaborate through the 19th century, this drink had “Old-Fashioned” appended to its name by the 1890’s (Wondrich 2007). American whiskey has historically been made out of mixtures of corn, rye, wheat, and/or barley, with two predominant styles being rye whiskey (containing at least 51% rye) and bourbon whiskey (containing at least 51% corn); rye tends to have a “spicier” quality than bourbon, and both are used for cocktails such as the Old-Fashioned (Stewart 2013, Gruber 2007). Having previously characterized the favor chemistry of sixteen commercial bitters, this experiment was designed to explore the interaction efect of bitters in a whiskey matrix, representing the simplest type of cocktail. Four typical whiskies (two bourbons and two ryes) and four common types of bitters were combined factorially into all sixteen possible pairs of whiskey and bitters, which were then subjected to sensory analysis by aroma, and volatile analysis. 1 Some cocktails called “Old-Fashioneds” are permutations containing fruit, seltzer, or soda, but for the purposes of this experiment we’re concerned with the simplest possible defnition of the cocktail.

113 Table 5.1 Whiskeys and bitters used to prepare the 16 samples in the study Product Code Name Producer Description B1 Elijah Craig Bourbon Heaven Hill Distilleries Bourbon B2 Evan Williams Bourbon Heaven Hill Distilleries Bourbon R1 Rittenhouse Rye Heaven Hill Distilleries Rye R2 Old Overholt Rye Jim Beam Rye

A Angostura Bitters Angostura Ltd Aromatic Bitters M Xocolotl Mole Bitters Bittermen’s Mole Bitters NO Peychaud’s Bitters Sazerac Company New Orleans Bitters O Regan’s Orange Bitters Bufalo Trace Orange Bitters

Materials and Methods Whiskey: Two bourbons (B1 and B2) and two rye whiskeys (R1 and R2) were purchased commercially (Table 5.1). Tese representative samples were chosen for being relatively typical lower-price, “basic” (B2 and R2, $12-$15 for 750 mL) and typical “premium” (B1 and R1, $24-$28 for 750 mL) whiskeys commonly used in making Old-Fashioneds in a bar context.

Bitters: Four bitters, identifed as those most commonly utilized in Old-Fashioneds and other whiskey-based cocktails, were purchased commercially (Table 5.1).“A” was a typical aromatic bitters, “NO” was an anise-heavy, New Orleans-style bitters, “O” was an orange bitters, and “M” was a mole-style bitters, a new variety incorporating the chocolate, chili, and spice favors of Mexican mole poblano.

Model Old-Fashioned: An old-fashioned cocktail is typically made by stirring room temperature whiskey, sugar, and bitters over ice. Tis melts the ice, which chills and dilutes the cocktail. To estimate dilution, an Old-Fashioned was made by stirring 5 g of spring water (Crystal Geyser, Calistoga, CA), 50 114 g of whiskey, and 3 dashes of bitters from a standard bitters bottle with 100g of cracked ice for 60 seconds in a chilled (4 ºC) mixing glass, straining, and measuring the mass of the resulting cocktail and the lefover ice separately. Measurements were repeated in triplicate and showed that the fnal mixture was approximately 50% water and 50% whiskey. Tis meant that diluting the spirits by half with water, as had previously been reported in descriptive analyses of gin and

Table 5.2 Descriptors used in the sensory analysis and their aroma references. Signifcance by ANOVA denoted by superscripts: b= signifcant by bitters; w=signifcant by whiskey; o, b*w =signifcant bitters x whiskey interaction. References were presented as described, in opaque, lidded, black glasses. Descriptor b w b*w Reference Herbalb * 3-cm sprig of thyme and oregano, 2 spearmint leaves (fresh) Cardamomb * 3 green cardamom pods, cracked Pencil shavingsb * 1 cm of yellow dixon ticonderoga pencil, shaved manually in pencil sharpener banana * 30 2-cm pieces alfalfa hay Dried fruit 2 unsulfered dried apricots, 10 thompson seedless raisins, 3 dried cherries Coriander 10 coriander seeds, cracked Gingerb * 1- cm thick slice of ginger, cut into sticks Cofeeb * 15 cofee beans, Peet’s cofee blend 101 Black pepperbw * * 8 black peppercorns Oakwo * * 5 grams oak shavings, medium toast Coconutw * 10 grams unrefned coconut oil Caramelw * 15 mL caramel dessert sauce, Nestle Nutmegb * ½ nutmeg seed, shaved Cloveb * 10 whole dried cloves Orangeb * 2 cm x 8 cm piece orange albedo Vanillabw * * 10 ml Nieman-Massey vanilla extract + 10 mL water Anisebw * * 1 star anise pod cinnamonb * 1 tsp powdered cinnamon caraway 1 tsp caraway seeds bananaw * 2 1-cm thick slices ripe banana chocolate * 1 cm cube dark Brix chocolate + 1 tsp Ghiradelli cocoa powder earthybw * * 10 g freshly dug soil + 5 mL water soap 1 cm x 1 cm x 2 cm piece dove unscented soap, shaved colabo * * 25 mL Coca-Cola smoky 2 tsp Lapsang Souchong smoked tea, Peet’s cofee vinegar 15 mL apple cider vinegar, Spectrum Organic 115 tequila for purposes of panelist safety (Sanchez 2011), would provide an accurate model of the alcohol content of a stirred old fashioned. To control for dilution and temperature efects, samples used for analysis were diluted to this measured level with water, but served at room temperature rather than chilled or over ice.

Sensory Analysis: A descriptive analysis procedure was used to profle the sensory characteristics of the 16 model Old Fashioned cocktails. 14 volunteer panelists, (4 females; aged 21 to 43) were recruited from a pool of students and postdoctoral scholars from the department of Viticulture and Enology and the surrounding community at the University of California, Davis. Samples were made by diluting whiskeys to 20% abv with spring water. 15 mL of each diluted whiskey sample was dispensed into a lidded opaque black tulip-shaped ISO wineglass and 200 μL bitters were added, mimicking the composition and dilution of the experimental model old-fashioned, above. Samples were served at room temperature (25 ºC). Over four training sessions, panelists met in groups and smelled the samples blind, and then generated, discussed, and refned descriptors by consensus until an agreed-upon list of 26 terms was determined. In the frst training session, four of the sixteen model Old-Fashioneds, each made with B2 and one of the four bitters, were smelled and discussed. Over the next three training sessions, the sixteen samples were presented in a random order, with four in session two and six each in sessions three and four, so that all samples were smelled at least once during training. References were made for each descriptor and these were smelled and discussed by the panelists, and changed and adjusted if necessary, over the second, third, and fourth sessions. During two additional sessions, the references fxed in the fourth training session were smelled by each panelist, and the intensity of each descriptor was rated in each sample with each panelist rating eight samples in a random order during each session. Te descriptive analysis was performed in triplicate over six sessions. Each panelist smelled all 26 references before each analysis session, then rated the intensity of each descriptor on an unstructured 9-cm line scale anchored by “low intensity” and “high intensity” in each sample, which were presented 8 per session, in black glasses as described above for training, 116 under red light in a Williams Latin Square presentation design. Descriptive analysis data were captured using FIZZ (Biosystemes, Couternon, France)

GC-MS: Model Old-Fashioneds were prepared for volatile analysis in the same proportions used for sensory analysis. A mixture of 10 mL whiskey diluted to 10% ethanol with deionized water, 130 μL of bitters, 50 μg/L 2-undecanone (Sigma-Aldrich, St Louis, MO) as an internal standard, and 3g of sodium chloride to improve volatile partitioning into headspace and increase analysis sensitivity was added into 20 mL amber glass headspace vials (Supelco, Bellefonte, PA)which were then capped with magnetic, PTFE-lined silicone septa headspace caps (Supelco). Extraction protocol was adapted from Hjelmeland et al. 2013. Samples were warmed to 40ºC and agitated at 500 rpm for 5 minutes directly before extraction. A conditioned, 2-cm long 50/30 um-thick PDMS-DVB-Carboxen SPME fber (Supelco) was introduced into the headspace of the vial for 45 minutes at 40ºC with rotational shaking at 250 RPM. A Gerstel MPS2 autosampler (Gertsel, Linthicum, MD) performed the extraction and the injection. Te fber was removed from the headspace of the vial and immediately introduced into the inlet of an Agilent model 6890 GC- single quadrupole-MS (Agilent Technologies) with a DB-Wax column (30 meters long, 0.25 mm ID, 0.25 μm flm thickness, J&W Scientifc, Folsom, CA). Te inlet was held at 250ºC with a 10:1 split and had a 0.75mm i.d. SPME inlet liner installed (Agilent Technologies). Te carrier gas was Helium, at a constant fow rate of 1 mL/minute. Te starting oven temperature was 40ºC, held for 3 minutes, followed by a 2ºC/minute ramp until 180ºCwas reached, then the ramp was increased to 30ºC/minute until 250ºC was reached, and held for 3 minutes. Te total runtime was 47 minutes. Te transfer line to the mass spectrometer was held at 250ºC , the source temperature was 230ºC, and the quadrupole temperature was 150ºC. Te mass spectrometer had a 1.5-minute solvent delay and was run in scan mode with Electron Impact ionization at 70eV, from m/z 40 to m/z 300. Te samples were analyzed in triplicate in random order. Kovats retention indices were calculated using C8-C20 hydrocarbon mixture (Sigma-Aldrich) retention times for identical

117 instrumental conditions. Peak identifcations (Table 5.4) were made by matching the background- subtracted average mass spectrum across half peak height for each peak to the NIST 05 mass spectral database, followed by verifcation by retention index and pure standards where available. Following identifcation, GC peaks were manually integrated and converted into headspace concentration in μg/L 2-undecanone equivalents by dividing the analyte peak area by the 2-undecanone peak area.

Analysis of Variance: Te sensory data were subjected to a 3-way Analysis of Variance (ANOVA) treating every model Old Fashioned as a separate product with main efects (panelist, replication, product) and all 2-way interactions for all 25 descriptors using the R statistical package. A second, 4-way analysis of variance (ANOVA) with main efects (panelist, replication, whiskey, and bitters) and all 2-way interactions was performed to evaluate interactive efects between whiskeys and bitters. For descriptors with a signifcant Judge*Product interaction, a pseudo-mixed model (with Mean Square of Judge*Product replacing Mean Square of Error in the F-value calculation for Product efect) was used. Descriptors were considered signifcantly diferent between samples when the p<0.05 by the ANOVA.

Means and Signifcant Diferences: For descriptors where samples were found to difer signifcantly by ANOVA, a Tukey Honest Signifcant Diference test (HSD) was performed using the ‘HSD.test’ function in the Agricolae package for the R statistical package.

Principal Component Analysis (PCA): Mean values for aroma descriptors found to difer signifcantly by ANOVA were calculated for each of the sixteen model Old-Fashioned samples over judges and replicates. A principal component analysis on these mean values was performed in R.

118 a b b b abc abc abc c ab abc bc bc ab bc bc bc a abc bc abc 2.5 1.9 1.7 1.5 2.2 1.8 1.9 1.1 2.5 1.8 1.6 1.7 2.5 1.7 1.6 1.5 3.0 2.2 1.7 1.8 nutmeg a ab ab b a a a a a a a a a a a a a a a a 2.5 2.1 2.1 1.9 2.6 2.3 2.6 2.2 2.6 1.9 2.1 1.9 2.1 2.1 2.2 2.1 1.9 2.1 2.1 1.7 caramel cantly between the the between cantly f a ab b b a a a a a a a a a a a a a a a a 1.7 1.4 1.3 1.2 1.8 1.3 1.9 1.8 1.4 1.5 1.3 1.2 1.3 1.2 1.4 1.2 1.5 1.0 1.4 1.1 coconut ered signi ered f a a ab b ab ab ab a a ab ab ab ab a ab ab ab ab b ab oak 3.5 3.5 3.2 2.9 3.2 3.3 3.6 3.7 3.8 3.3 3.5 3.4 2.8 3.8 3.1 3.1 2.9 3.1 2.6 2.9 ab ab b a ab bc bc abc abc abc c abc abc c c abc a abc abc abc a b b b 1.9 1.8 1.6 2.2 2.7 1.6 1.6 1.7 1.7 2.0 1.5 2.0 1.9 1.6 1.4 1.7 2.8 2.1 2.0 1.8 2.3 1.8 1.8 1.6 cant by ANOVA, by sample, bitters, and whiskey. whiskey. and bitters, sample, by ANOVA, by cant f black pepper a a a a a a a a a a a a a a a a b a b ab 1.2 1.5 1.0 1.4 1.1 1.5 1.0 1.2 0.9 1.2 1.2 1.3 1.0 1.6 1.0 1.4 1.1 1.4 1.1 1.3 coffee a a a a a a a a a a a a a a a a a bc ab c 2.0 1.6 2.0 1.6 2.1 1.6 1.7 1.6 2.2 1.7 2.5 1.5 2.4 2.0 2.3 1.8 2.2 1.7 2.1 1.6 ginger abc abc c abc abc abc bc abc abc abc c a abc abc c ab a a b a hay 2.4 1.9 1.5 2.5 2.4 2.0 1.6 2.3 2.3 2.6 1.4 2.9 2.2 2.2 1.6 2.9 2.3 2.2 1.5 2.6 erent from each other for the samples studied. samples the for other each from erent f a a a a a a a a a a a a a a a a ab ab b a 3.1 3.3 2.5 3.3 2.8 2.9 2.9 3.3 3.0 2.8 2.9 3.4 2.9 2.7 2.7 3.4 2.9 2.9 2.8 3.4 cantly di cantly f pencil shavings ab ab ab ab ab b ab ab ab ab a b ab ab ab ab ab bc a c 2.9 2.4 3.0 2.4 2.9 1.8 2.9 2.2 2.7 2.7 3.5 2.0 2.5 2.1 3.0 2.2 2.7 2.2 3.1 2.2 cardamom erent letters are signi are letters erent a c abc abc abc bc abc abc ab abc abc abc ab bc abc abc a b a a f 3.6 1.9 2.7 2.9 2.8 2.2 2.8 2.7 3.3 2.6 3.0 2.7 3.2 2.2 3.0 2.9 3.2 2.2 2.9 2.8 herbal Mean values of descriptors rated in model Old-Fashioneds calculated as signi calculated Old-Fashioneds model in rated descriptors of values Mean B1-M B1-O B1-NO B2-A B2-M B2-O B2-NO R2-A R2-M R2-O R2-NO R1-A R1-M R1-O R1-NO B1 B2 R2 R1 B1-A Mixture A M O NO Table 5.3 Table di Old-Fashioneds model for intensities attribute when calculated only were whiskey) (or bitters by Means attribute test; HSD Tukey’s of results are values mean to next Letters ANOVA. 4-way the in whiskey) (or bitters by Old-Fashioneds model di with intensities 119 erent erent f cantly di cantly f ab bc a c ab ab ab b ab b ab b ab b ab b b b a b cola 2.4 2.0 2.7 1.9 2.7 2.3 2.3 1.9 2.4 1.8 2.6 2.1 2.8 2.1 2.5 2.0 1.9 1.8 3.4 1.7 erent letters are signi are letters erent f b b a a c a bc ab d bcd bcd cd cd abcd cd abcd abc abc bcd abc bcd a abc ab earthy 1.3 1.5 2.0 2.3 1.4 2.1 1.6 2.0 0.8 1.6 1.5 1.3 1.2 1.8 1.2 1.9 2.1 2.1 1.6 2.2 1.6 2.9 2.2 2.6 b a b b bc ab bc bc bc ab c bc c abc bc bc bc a bc bc chocolate 1.8 3.1 1.9 2.1 2.1 3.0 2.0 2.3 1.8 2.9 1.6 2.2 1.5 2.5 2.2 2.0 1.9 3.8 1.7 1.9 a ab b ab a a a a a a a a a a a a a a a a 2.2 1.8 1.6 1.9 1.9 2.4 2.2 2.2 1.7 2.0 1.8 1.9 1.5 1.8 1.5 1.8 1.8 1.7 2.0 2.0 banana cantly between the model Old-Fashioneds by bitters (or whiskey) in the 4-way 4-way the in whiskey) (or bitters by Old-Fashioneds model the between cantly bc c c c abc abc c c ab abc abc c a abc c c a b b b f 2.0 2.0 1.7 1.9 2.8 2.3 1.8 1.8 3.2 2.2 2.2 1.7 3.3 2.3 1.9 1.8 2.8 2.2 1.9 1.8 cant by ANOVA, by sample, bitters, and whiskey. Means by bitters (or whiskey) whiskey) (or bitters by Means whiskey. and bitters, sample, by ANOVA, by cant f cinnamon ered signi ered f b b b a ab b b ab b b b ab ab b b ab ab ab b a ab b b a anise 1.3 1.3 1.4 1.9 1.5 1.1 1.1 1.5 1.2 1.2 1.1 1.6 1.8 1.1 1.3 1.5 2.1 1.5 1.4 2.6 1.6 1.2 1.2 1.8 a ab b b ab a ab ab ab ab ab ab ab ab b ab ab ab ab b a a a a vanilla 3.1 2.9 2.6 2.5 3.0 3.5 3.0 2.9 2.9 3.4 2.7 2.6 2.8 2.5 2.4 2.6 2.7 2.7 2.5 2.3 2.8 3.0 2.6 2.6 a a a a a a a a a a a a a a a a b b a b 2.2 1.9 2.7 2.2 2.5 2.2 3.0 2.0 2.1 1.9 2.7 1.9 2.2 1.9 2.8 1.7 2.3 2.0 2.8 2.0 Mean values of descriptors signi descriptors of values Mean orange ab ab b ab ab ab ab ab a ab ab ab ab ab ab ab a b b b clove 2.5 2.1 1.8 2.0 2.8 2.3 2.3 2.1 3.1 2.1 2.2 2.1 2.8 2.3 2.2 2.0 2.8 2.2 2.1 2.1 mixture B1-M B1-O B1-NO B2-A B2-M B2-O B2-NO R2-A R2-M R2-O R2-NO R1-A R1-M R1-O R1-NO B1 B2 R2 R1 B1-A A M O NO Table 5.3 (continued) 5.3 (continued) Table di intensities attribute when calculated only were di with intensities attribute test; HSD Tukey’s of results are values mean to next Letters ANOVA. studied. samples the for other each from 120 Figure 5.1 Principal Component Analysis of model old fashioneds descriptive analysis data. Samples are in bold, all caps; descriptors are in red italic. Top: PC1(31% Variance explained, X-Axis) and PC2 (23% variance explained, Y-Axis); Bottom :PC1 (X-axis) and PC3 (14% Variance explained, y-axis)

121 Partial Least Squares Regression (PLS): Partial Least Squares Regression was used to compare correlations between sensory qualities and headspace volatiles of model Old-Fashioneds. Mean peak areas of volatiles for each sample (normalized to peak area of 2-undecanone in each sample) were standardized by dividing by their standard deviation. Aroma intensity ratings were assigned a weighting of 1.0 (i.e. not standardized).

Results and Discussion: Te sensory panel agreed upon 25 aroma descriptors which are summarized along with their references in table 2. Te PCA plot (Figure 5.1) of signifcant descriptors explained 31% of the variance in Principal Component (PC) 1, 23% of the variance in PC 2, and 14% of the variance in PC 3. PC 1 separates the samples incorporating Aromatic (A) and Orange (O) bitters, on the lef, from those containing mole-style (M) or New Orleans-style (NO) bitters, on the right. Pencil shavings, cofee, and chocolate aromas contributed strongly to PC 1 (their loadings vectors have less than a 45 degree angle with PC 1) in the positive, Mole-New Orleans direction, and orange, cola, cardamom, ginger, and herbal aromas contributed strongly to PC1 in the opposite, Orange- Aromatic direction. Along PC 2, clusters separated by PC 1 are sorted diferently depending on bitters type: on the lef side, PC 2 separates samples containing Aromatic bitters, in the positive direction, from those containing Orange bitters, in the negative direction. Within both Aromatic and Orange bitters, the samples cluster by whiskey type- B1 with B2, and R1 with R2. In both of the clusters by bitters type, the ryes plot higher on PC 2 than the bourbons, meaning that within each cluster the ryes are more infuenced by the high-PC2 attributes such as clove, black pepper, cinnamon, nutmeg, and anise than the bourbons in the same cluster are. Within a cluster, bourbons plot lower on PC 2, so are more infuenced by the low-PC 2 attributes like oak, vanilla, banana, coconut, and caramel. On the other (right) side of the plot, PC 2 separates the premium

122 Figure 5.2 Plots of mean intensities of oak aroma (A) and cola aroma (B), showing signifcant interaction efect between bitters and whiskey.

3.8 bw$whiskey bw$bitters ElijahCraig Orange EvanWilliams 3.6 Angostura OldOverholt Mole

Rittenhouse 3.0 Peychaud 3.4 3.2 2.5 mean of bw$oak mean of bw$cola 3.0 2.8 2.0 2.6

Angostura Mole Orange Peychaud ElijahCraigEvanWilliams OldOverholt Rittenhouse bw$bitters bw$whiskey A B

rye (R1) and the premium bourbon (B2) at its most extreme ends, with the “basic” rye (R2) and bourbon (B2) clustered together close to the middle of this PC. Rye is ofen considered to be a ‘spicier’ whiskey than Bourbon (Maclean 2008 Hellmich 2010, Stewart 2013, Buglass 2011), and this is recapitulated in the tendency in this dataset for rye-containing mixtures to associate more highly with spice terms such as cinnamon and nutmeg. PC 3, on the lef (aromatic-orange) end of PC 1, shows some overlaps between bitters types that were separated by PC 2; however on the right-hand side of PC 1, the samples are separated a little more clearly by type of bitters- on the top end of PC 3 (and strongly correlated to chocolate aroma), the samples containing mole bitters; on the bottom end of PC 3, the samples containing New Orleans-style bitters (correlated to anise, pencil shavings, hay, and earthy). Interestingly, the basic rye-mole bitters sample plots closer along PC 3 to some of the samples containing New Orleans-style bitters than the rest of the samples

with mole bitters; this is the sample that is the least intense along the chocolate vector in PC 2 as well.

123 Table 5.4: Compounds identifed in samples, identifed by retention time (RT), Mass Spectra, and Calculated (CRI) and literature C8-C20 retention indices (RI)

Literature RI # namea CAS RT CRI Pherobase Flavornet Other OF1 Alpha-pinene s 80-56-8 4.91 1008 1027-1034 1032 OF2 Ethyl butanoate s 105-54-4 5.34 1021 1022-1057 1028 OF3 Camphene* s 79-92-5 6.06 1042 1077 1075 OF4 Hexanal* s 66-25-1 6.67 1060 1067-1093 1084 OF5 Isobutyl alcohol s 110-19-0 7.42 1083 1005-1007 1015 OF6 Isoamyl acetate s 123-92-2 8.28 1107 1118-1147 1117 OF7 ni.a 9.06 1125 1262-1265 1238 OF8 Beta Pinene s 127-91-3 9.68 1139 1113-1124 1116 OF9 alpha Phellandrene s 99-83-2 9.74 1141 1205 1166 OF10 Myrcene s 125-35-3 9.92 1145 1176 OF11 sabinene s 127-91-3 9.92 1145 1113-1124 1116 OF12 alpha Terpinene 99-86-5 10.35 1155 1178 1178 OF13 Limonene s 138-86-3 11.22 1175 1198-1234 1201 OF14 beta phellandrene s 555-10-2 11.66 1185 1241 1209 OF15 Eucalyptol s 470-82-6 11.77 1188 1214-1224 OF16 Isoamyl alcohol s 123-51-3 12.21 1198 1169-1247 1205 OF17 Ethyl caproate s 123-66-0 13.12 1219 1224-1270 1220 OF18 Cymene 527-84-4 14.27 1246 1267 1261 OF19 Terpinolene 586-62-9 14.88 1260 1275-1297 1284 OF20 ni.b 17.11 1311 OF21 Ethyl heptanoate s 106-30-9 17.60 1321 OF22 p-menth-2-en-1-ol 29803-82-5 19.79 1365 OF23 Alpha-p-dimethylstyrene 1195-32-0 21.47 1399 1414 OF24 Ethyl caprylate s 106-32-1 22.11 1413 1422-1446 1436 OF25 Isoamyl caproate 2198-61-0 23.03 1435 1450b OF26 Camphor s 76-22-2 24.47 1468 1498 OF27 ni.d 25.22 1486 OF28 Ethyl nonanoate s 123-29-5 26.22 1509 1528 OF29 Linalool s 78-70-6 26.62 1519 1484-1570 1537 OF30 Caryophyllene s 87-44-5 28.00 1553 1608-1618 OF31 Ethyl caprate s 110-38-3 30.40 1615 1630 1655 OF32 Estragole 140-67-0 30.64 1622 1655 OF33 Ethyl trans-4-decenoate 7367-88-6 31.06 1636 1694 OF34 Isoamyl caprylate 2035-99-6 31.12 1638 1674b OF35 alpha Terpineol acetate 80-26-2 31.92 1663 1687-1700 OF36 Carvone s 99-49-0 33.00 1697 1715 1720 OF37 Geranyl acetate s 105-87-3 34.29 1735 1711-1760 OF38 ni.e 34.55 1743 1789 OF39 Isobutyl decanoate 30673-38-2 34.66 1746 1755c OF40 cis-4-Decen-1-ol 57074-37-0 35.74 1778 OF41 Phenethyl acetate s 103-45-7 35.90 1782 1803 1829 OF42 anethole s 104-46-1 36.27 1793 1808 OF43 2-Tridecanone s 593-08-8 36.35 1796 OF44 Ethyl dodecanoate 54982-83-1 37.69 1841 1822 OF45 Isopentyl decanoate 2306-91-4 38.33 1863 1871b OF46 Benzeneethanol s 60-12-8 39.15 1891 1873-1940 OF47 Whiskey lactone s 60-12-8 40.61 1976 1977 a=mass specta matched to NIST >80%; b=Tao et al 2008; C=Maztekin 2014; D=Lee et al 2005; F=Jones et al 2011; s=matched to authentic standard

124 Table 5.4 (continued) Compounds identifed in samples, identifed by retention time (RT), Mass Spectra, and C8-C20 retention indices (RI) Literature RI # namea CAS RT CRI Pherobase Flavornet Other OF48 methyleugenol s 93-15-2 42.44 >2000 2099d, 2007e OF49 Cinnamaldehyde 104-55-2 42.66 >2000 2017 OF50 Nerolidol 7212-44-4 43.04 >2000 1961-2054 OF51 Ethyl myristate s 124-06-1 43.21 >2000 OF52 Eugenol s 97-53-0 44.06 >2000 2142-2192 OF53 Gamma Eudesmol 209-71-8 44.20 >2000 2182 2121 OF54 Guaiol 489-86-1 44.25 >2000 2077 2104f OF55 Elemicin 487-11-6 44.54 >2000 2167b OF56 Beta Eudesmol 473-15-4 44.61 >2000 2246 OF57 Myristicin 607-91-0 44.72 >2000 2257b OF58 Ethyl hexadecanoate 628-97-7 44.83 >2000 2229 OF59 ni.f 45.22 >2000 a=mass specta matched to NIST >80%; b=Tao et al 2008; C=Maztekin 2014; D=Lee et al 2005; F=Jones et al 2011; s=matched to authentic standard

As explained by the separations plotted in the PCA, the strongest driver of perceived favor diferences in these model Old-Fashioneds- the main source of separation along PC 1– is type of bitters. PC 1 separates samples containing orange or aromatic bitters from those containing New Orleans-style or mole bitters regardless of whiskey type. For samples containing New Orleans-style or mole bitters, type of whiskey is a stronger separator along PC 2 than type of bitters, and the efect is more pronounced for the premium bourbon and rye than for the basic bourbon and rye, which were more similar to each other when mixed with these bitters. Two of the sensory descriptors, oak and cola were found to have a signifcant whiskey- by-bitters interaction efect. Tis means that the mixing of one or more of the bitters with one

or more of the whiskeys caused that sensory quality to either be heightened or dampened, on average, compared to other pairings of the same whiskey with other bitters, or the same bitters with other whiskeys (Figure 5.2). Oak aroma difered signifcantly among the whiskeys, with both B1 and B2 rated signifcantly higher than R1, with R2 not difering signifcantly from either R1 or B1 and B2. In the whiskey plus Mole bitters mixtures, this efects was dampened, with both R1 and R2 rated higher than average for oak, and B1 and B2 rated lower than average.

125 Figure 5.3 Partial Least Squares Regression Analysis (PLS) plots of model old-fashioned volatile and sensory profles. Variance Explained PC 1: 40% X, 21% Y; PC 2: 21% X, 19% Y; PC 3: 13% X, 14% Y 5.3A: Positions of samples, PC1 vs PC2 5.3B: Positions of samples, PC2 vs PC3 5.3C: PC 1 vs PC 2 plot of sensory descriptors 5.3D: PC 1 vs PC 2 plot of headspace volatiles 5.3E: PC 2 vs PC 3 plot of sensory descriptors 5.3F: PC 2 vs PC 3 plot of headspace volatiles PC 2: 21% X, 19% Y variance exp. PC 2: 21% X, 19% Y variance

A PC 1: 40% X, 21% Y variance exp. PC 3: 13% X, 14% Y variance exp. PC 3: 13% X, 14% Y variance

B PC 2: 21% X, 19% Y variance exp. 126 Figure 5.3 continued 5.3C: PC 1 vs PC 2 plot of sensory descriptors 5.3D: PC 1 vs PC 2 plot of headspace volatiles PC 2: 21% X, 19% Y variance exp. PC 2: 21% X, 19% Y variance

C PC 1: 40% X, 21% Y variance exp. PC 2: 21% X, 19% Y variance exp. PC 2: 21% X, 19% Y variance

D PC 1: 40% X, 21% Y variance exp.

127 Figure 5.3 continued. 5.3E: PC 2 vs PC 3 plot of sensory descriptors 5.3F: PC 2 vs PC 3 plot of headspace volatiles PC 3: 13% X, 14% Y variance exp. PC 3: 13% X, 14% Y variance

E PC 2: 21% X, 19% Y variance exp. PC 3: 13% X, 14% Y variance exp. PC 3: 13% X, 14% Y variance

F PC 2: 21% X, 19% Y variance exp. 128 Cola-like favor difered signifcantly among the bitters, with Orange and New Orleans- style bitters rated highest and lowest, respectively, and Aromatic bitters difering signifcantly from New Orleans-style and Mole difering signifcantly from Orange. Mixing R1 with Orange bitters amplifed this mixture’s cola-like character compared with other bitters, which tended to have low or lower-than-average ratings for cola aroma when mixed with R1. A Partial Least Squares Regression (PLS, Figure 5.3) was performed on standardized signifcant sensory descriptors and volatile compounds (listed in table 5.4, with headspace concentrations given in supplementary table S5.1). Te frst three PCs explained 40%, 21%, and 20% of the variance in the volatile data, and 21%, 19%, and 14% of the variance in the sensory data. PC1 primarily separated the samples into groups of those mixtures containing Aromatic bitters and those containing other types of bitters, with PC 2 separating the latter group into clusters by type of bitters used. In the (sensory data only) PCA, some spatial groupings were more dependent on type of whiskey than type of bitters. By contrast, taking into account chemical diferences between the samples in the PLS, the type of bitters used is the primary driver of spatial separation and grouping. While this efect dominates overall separation along PC 1 and PC 2, within each group of Old-Fashioneds (separated by type of bitters), the B1 (premium bourbon) sample plots highest along PC 2 compared to the other samples, and the R1(premium rye) sample plots lowest, with B2 and R2 somewhere in the middle. Tis mirrors the tendency in the PCA for R1 and B1 mixtures to plot furthest away from each other within mixtures containing the same type of bitters, suggesting that, when mixed with any given type of bitters, latent favor diferences between bourbon and rye are expressed most obviously when comparing more premium whiskeys. Te spatial position of the R1-containing mixtures in both the PCA and the PLS ties them to the descriptors anise, hay, and pencil shavings; for B1, the same efect is true for the vanilla, coconut, and caramel descriptors. Much of the separation in the PLS, as noted above, derives from diferences in the

Aromatic bitters compared to the other types of bitters. Many of the compounds contributing strongly to the separation in the PLS (noted by their position further out along one or more of

129 the axes of the plot) are terpenoids highly associated with the Old-Fashioneds containing the Aromatic type bitters, and to nutmeg, cinnamon, black pepper, clove, and herbal aromas. Across all types of whiskeys, the Aromatic bitters was rated highest for each of these descriptors, and signifcantly higher than at least one other type of bitters. Te compounds associated most with these aromas were elemicin, caryophyllene, geranyl acetate, alpha-p-dimethylstyrene, beta- eudesmol, camphor, gamma-terpinene, eugenol, camphene, limonene, myrcene, alpha terpinene, cymene, sabinene, and alpha phellandrene, as well as several unidentifed compounds. PC 2 primarily separates orange, cardamom, and cola and their associated volatiles at one extreme from pencil shavings, hay, and anise and their associated volatiles at the other. Additionally, descriptors that difered signifcantly between the whiskeys but not the bitters all plot exclusively in the upper half of PC 2, as do nearly all of the non-terpenic esters, which are ofen associated with yeast fermentation. A number of compounds and aromas load strongly onto PC 2 but not to PC 1 – this may be because they describe relationships shared between samples that are separated by PC 1. Most dominant among these, plotting positively along PC 2, are orange, cardamom, and cola aromas and their associated volatiles eucalyptol, linalool, alpha- terpinyl acetate, terpineol, and alpha-pinene. Tese descriptors difered signifcantly in intensity between the types of bitters, with the Aromatic and Orange bitters rated most highly in cardamom and cola, and orange rated signifcantly higher in Orange bitters than all three of the other types. Te Aromatic bitters were not signifcantly diferent in orange aroma intensity from the Mole and New Orleans-style bitters. Conversely, the pencil shavings, anise, and hay aromas plot negatively along PC2, and are associated with a cluster of phenylpropenoid compounds – estragole, myristicin, anethole, and methyleugenol , as well as 2-tridecanone which is also associated with cofee and earthy aromas in the plot. In isolation, estragole and anethole are both described as having sweet and anise-like aromas; myristicin is described as spicy and woody; and 2-tridecanone as waxy, dairy, herbal, and earthy (Luebke 2014 a, b, c, d).

130 Mixing bitters and whiskeys into Old-Fashioned cocktails results in identifable diferences in favor arising from both the bitters and the whiskey used for the cocktail. In other words, this type of mixing does not mask diferences between either ingredient, and in fact, more expensive whiskeys are more signifcantly diferent upon mixing into an old-fashioned than lower priced whiskeys. Tis suggests that commonly held wisdom that more expensive, or more carefully crafed spirits should not be mixed because their favor will be lost is not necessarily true, and that their characteristics continue to come through in careful mixology. From a holistic standpoint, both the sensory data and the sensory data analyzed in tandem with volatile data suggest that the diferences in Old-Fashioned type cocktails are driven more strongly by the type of bitters used than by the type of whiskey used, though this depends on the type of bitters used, with the PCA suggesting that Aromatic and Orange bitters have a greater efect of separating whiskeys from each other, while Mole and New Orleans-style bitters express more similarities by whiskey type than by bitters type. While positions in the PCA plot suggest that aromatic bitters emphasize the spicy qualities of rye whiskeys and other types of bitters emphasize the sofer, oakier qualities of bourbons, it should be emphasized that this relationship was not found to be statistically signifcant. In favor-chemical terms, while the bitters type was a stronger overall spatial separator of samples, within each cluster of samples grouped by type of bitters, a conserved spatial pattern separating bourbon- and rye- based samples along PC 2 is evident. A number of aroma descriptors generated in the present study—cardamom, hay, ginger, nutmeg, clove, orange, vanilla, anise, cinnamon, chocolate, earthy, black pepper, dried fruit, and cola—were also (independently) generated in the previous work focusing on profling bitters without the addition of whiskey. Te terms pencil shavings, coriander, cofee, oak, coconut, caramel, caraway, banana, smoky, and vinegar were unique to this experiment. Of the terms shared with the bitters-only dataset, black pepper, vanilla, anise, and earthy difered signifcantly in intensity between both the whiskeys and the bitters when made into model Old-Fashioneds. Te aromas oak and cola showed signifcant interactive efects between bitters and whiskeys, with Mole

131 bitters signifcantly dampening oak aroma in the bourbons compared to the rye whiskeys, and cola-like character was, synergystically, signifcantly higher in the premium rye (R1)-Orange bitters mixture than the other mixtures. Te presence of interactive sensory efects suggests further questions of interest about the inherent sensory complexity of cocktail-making; if sensory qualities in even simple cocktails only exist upon mixing and for specifc combinations, further complexities and unique interactions could be envisioned for more complex mixtures, such as ones incorporating vermouth (aromatized, fortifed wine to which herbs have been added), fresh citrus, potable bitters such as campari or cynar, or several distilled spirits, each of which might vary in proportion or composition from bartender to bartender based on their experience and intuition.

References

Buglass, A. J. (2011). Handbook of Alcoholic Beverages: Technical, Analytical and Nutritional Aspects. Hoboken, NJ: Wiley. Retrieved from http://books.google.com/ books?id=gNc34oNpg0AC Clarke, P. (2010, September). Bittersweet Symphony. Imbibe Magazine, 46–53. Grimes, W. (2002). Straight Up Or On the Rocks: The Story of the American Cocktail (p. 208). New York: Macmillan. Gruber, M. C. (2007). Whiskey. In A. F. Smith (Ed.), The Oxford Companion to Food and Drink in America (pp. 620–621). New York: Oxford University Press. Haigh, T. (2009). Vintage Spirits and Forgotten Cocktails: From the Alamagoozlum to the Zombie 100 Rediscovered Recipes and the Stories Behind Them. Beverly, MA: Quarry Books. Retrieved from http://books.google.com/books?id=sCR7wWhM7IQC Hellmich, M. (2010). The Ultimate Bar Book: The Comprehensive Guide to Over 1,000 Cocktails. San Francisco, CA: Chronicle Books. Retrieved from http://books.google.com/ books?id=q43NAVAcIAkC Hjelmeland, A. K., King, E. S., Ebeler, S. E., & Heymann, H. (2012). Characterizing the American Journal of Enology and Viticulture, 64(2), 169–179. doi:10.5344/ajev.2012.12107 Luebke, W. (2014a). Estragole. com/docs/doc1013251.html Luebke, W. (2014b). 2-tridecanone. . Retrieved May 01, 2014, from http://www. 132 Luebke, W. (2014c). Myristicin. . Retrieved May 01, 2014, from http://www. Luebke, W. (2014d). Anethole. com/docs/doc1001151.html MacLean, C. (2008). Eyewitness Companions: Whiskey. London: Dorling Kindersley. Retrieved from http://books.google.com/books?id=vifQ1RKo3v4C Parsons, B. T., & Anderson, E. (2011). Bitters: A Spirited History of a Classic Cure-All, with Cocktails, Recipes, and Formulas. Berkeley, CA: Ten Speed Press. Retrieved from http:// books.google.com/books?id=OPlRDjfGnloC Regan, G. (2003). The Joy of Mixology. New York, NY: Crown Publishing Group. Retrieved from http://books.google.com/books?id=7TlhtrpXa-MC Sampson, E., Chittenden, G., & Croswell, H. (1806). The Balance and Columbian Repository. Hudson, NY: Sampson, Chittenden & Croswell. Retrieved from http://books.google.com/ books?id=M9MRAAAAYAAJ Sanchez, J. V. S. G. (2011). Comparison of Descriptive Analysis and Projective Mapping Techniques in the Aroma Evaluation of the Distilled Spirits , Gin and Tequila. Master’s Thesis, University of California, Davis. 87pp. Simonson, R., & Krieger, D. (2014). The Old-Fashioned: The Story of the World’s First Classic Cocktail, with Recipes and Lore. Berkeley, CA: Ten Speed Press. Retrieved from http:// books.google.com/books?id=NqocAgAAQBAJ Stewart, A. (2013). The Drunken Botanist. New York: Algonquin Books. Retrieved from http:// books.google.com/books?id=SKDzb3BKEWEC Tonutti, I., & Liddle, P. (2010). Aromatic plants in alcoholic beverages: A review. Flavour and Fragrance Journal, 25(5), 341–350. doi:10.1002/ffj.2001 Wondrich, D. (2007). Imbibe!: From Absinthe Cocktail to Whiskey Smash, a Salute in Stories and Drinks to “Professor” Jerry Thomas, Pioneer of the American Bar. New York, NY: Perigee Books. Retrieved from http://books.google.com/books?id=IV3s-NUYnfEC

133 21 33 16 32 63 43 12 13 23 1.9 0.4 2.3 2.7 2.4 2.5 3.1 4.4 4.9 1.5 6.9 2.7 2.8 3.6 A nd nd nd nd nd nd nd 221 297 427 949 B2 42 88 27 56 79 45 14 71 4.4 1.2 5.8 5.7 6.5 1.8 669 M nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd B2 1302 3043 42 81 34 84 80 46 57 87 4.6 1.2 6.8 8.0 2.1 6.9 8.4 6.3 682 O nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd B2 1321 3573 34 57 49 33 56 2.8 1.1 5.3 7.2 2.3 509 839 108 nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd B2 NO 2309 16 19 15 35 54 25 29 14 13 28 1.7 0.8 0.3 0.2 2.1 2.3 2.0 2.2 2.7 4.6 5.0 6.9 3.5 3.6 3.4 4.2 A nd nd nd nd 200 226 541 882 R2 45 53 28 60 84 10 49 17 1.9 0.9 7.4 4.5 1.4 615 100 M nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd R2 1589 3064 48 61 48 94 12 50 66 2.2 7.7 3.6 9.3 1.4 111 104 679 nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd O R2 1611 3656 29 36 64 31 91 70 1.2 1.0 6.6 1.7 419 nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd R2 NO 1095 2384 15 22 16 37 54 45 13 16 16 2.4 1.0 0.3 1.7 1.9 2.1 2.0 1.4 4.2 3.7 2.2 5.9 3.7 3.3 2.8 A nd nd nd nd nd nd 206 277 507 724 R1 bitters 54 67 64 67 98 10 43 18 49 whiskey 3.0 1.6 2.6 7.2 8.5 4.3 926 M nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd R1 1590 2510 49 63 93 12 13 50 78 53 1.6 3.5 1.9 8.0 2.1 2.2 113 122 918 O nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd R1 1768 2800 11 25 40 60 27 25 1.8 0.7 0.9 6.7 1.8 487 971 108 nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd R1 NO 1474 49 65 41 43 61 88 3.9 6.9 8.1 6.8 9.2 1.7 9.1 1.8 8.1 106 107 762 109 O nd nd nd nd nd nd nd nd nd nd nd nd nd B1 1819 3126 30 43 89 29 56 4.0 2.1 1.5 1.0 0.9 5.9 1.9 5.5 494 103 nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd B1 1158 NO 1952 62 76 24 70 37 17 1.4 1.7 1.2 5.3 8.8 1.7 1.7 6.2 2.5 884 139 M nd nd nd nd nd nd nd nd nd nd nd nd nd nd nd B1 1639 2529 11 11 21 10 49 12 88 10 63 38 2.1 1.7 0.3 5.0 2.1 3.3 3.9 2.6 8.9 7.4 2.5 5.2 9.0 4.9 297 181 101 226 A nd nd nd nd B1 1325 7889 4.91 6.06 6.67 8.28 9.92 7.42 9.92 9.06 9.68 5.34 9.74 11.77 11.22 11.66 17.11 22.11 10.35 26.22 26.62 30.40 12.21 13.12 14.27 14.88 17.60 19.79 21.47 23.03 25.22 28.00 30.64 31.12 31.06 24.47 RT ed; nd=not detected. nd=not ed; Headspace concentrations in samples (listed by type of bitters and type of whiskey) in ug/L 2-Undecanone 2-Undecanone ug/L in whiskey) type of and bitters type of by (listed samples in concentrations Headspace f ni.b ni.d ni.a Name Linalool Cymene Myrcene Hexanal* sabinene Camphor Estragole Limonene Eucalyptol Terpinolene Camphene* Beta Pinene Ethyl caprate Alpha-pinene Ethyl caproate Caryophyllene Ethyl caprylate Isobutyl alcohol Isoamyl alcohol Ethyl butanoate Isoamyl acetate alpha Terpinene alpha Ethyl nonanoate Ethyl heptanoate Isoamyl caproate Isoamyl caprylate beta phellandrene p-menth-2-en-1-ol alpha Phellandrene Ethyl trans-4-decenoate Alpha-p-dimethylstyrene # OF1 OF3 OF4 OF6 OF7 OF8 OF5 OF2 OF9 OF11 OF15 OF10 OF12 OF13 OF14 OF16 OF17 OF18 OF19 OF20 OF21 OF22 OF23 OF24 OF25 OF27 OF28 OF29 OF30 OF31 OF32 OF34 OF33 OF26 Supplementry Table S5.1 Table Supplementry identi ni=not equivalents. 134 - 11 11 12 24 98 90 56 15 13 53 81 60 23 2.2 6.1 6.1 3.4 7.1 6.5 7.3 6.4 A nd nd nd 321 159 B2 54 14 60 16 24 25 14 72 32 56 4.3 6.1 2.5 3.4 0.6 2.1 165 630 269 M nd nd nd nd nd nd nd B2 11 44 15 50 29 15 21 14 28 32 67 70 0.8 4.2 180 878 347 103 O nd nd nd nd nd nd nd nd B2 11 29 13 12 15 10 16 65 19 62 21 36 6.5 3.0 5.4 4.3 0.4 2.6 594 616 246 580 167 nd nd nd B2 NO 59 13 12 38 12 21 68 85 23 12 17 3.3 3.7 3.0 4.0 8.2 8.2 5.4 5.0 A nd nd nd nd 212 138 176 R2 55 12 18 14 60 62 23 10 47 41 58 2.4 3.0 9.3 1.9 118 179 471 314 M nd nd nd nd nd nd nd R2 11 48 15 25 18 25 53 65 21 10 39 42 65 79 9.2 5.9 205 664 394 nd nd nd nd nd nd nd O R2 13 29 16 14 34 17 21 18 23 26 32 6.4 8.1 6.9 7.9 2.6 113 592 595 245 512 170 nd nd nd nd R2 NO 58 81 37 13 22 12 5.1 4.4 3.6 7.3 3.6 7.1 7.6 1.8 5.9 5.6 6.4 0.4 5.9 A nd nd nd nd 185 133 171 R1 bitters 78 51 15 21 13 17 59 29 58 10 74 20 whiskey 9.3 2.3 3.3 2.0 336 289 M nd nd nd nd nd nd nd nd R1 9 11 12 44 10 15 62 35 51 80 44 14 7.9 7.5 4.8 383 393 O nd nd nd nd nd nd nd nd nd R1 11 12 59 29 18 15 41 5.7 6.9 3.3 1.1 2.2 3.4 6.1 1.0 236 576 299 577 171 nd nd nd nd nd nd R1 NO 11 11 10 23 41 42 57 51 71 27 2.3 1.4 1.4 356 281 477 O nd nd nd nd nd nd nd nd nd B1 1013 21 13 27 24 29 31 15 17 18 0.8 7.0 5.1 2.4 9.0 1.9 605 128 605 166 221 204 nd nd nd nd nd B1 NO 26 47 61 73 26 32 10 19 2.3 0.8 1.8 1.5 3.6 119 566 304 329 M nd nd nd nd nd nd nd nd nd B1 11 12 17 16 58 24 73 71 41 81 6.3 5.2 3.0 7.3 2.4 4.2 6.9 306 161 502 A nd nd nd nd nd B1 4363 ed; nd=not detected. nd=not ed; f 39.15 40.61 44.61 31.92 34.66 35.74 35.90 37.69 38.33 44.20 44.83 33.00 34.29 34.55 36.27 36.35 42.44 42.66 43.04 43.21 44.06 44.25 44.54 44.72 45.22 45.32 RT (continued) Headspace concentrations in samples (listed by type of bitters and type of whiskey) in ug/L 2-Unde ug/L in whiskey) type of and bitters type of by (listed samples in concentrations Headspace (continued) ni=not identi ni=not ni.f ni.e ni.g Guaiol Name Eugenol Carvone Elemicin anethole Nerolidol Myristicin 2-Tridecanone methyleugenol Ethyl myristate Beta Eudesmol Geranyl acetate Benzeneethanol Whiskey lactone cis-4-Decen-1-ol Cinnamaldehyde Phenethyl acetate Ethyl dodecanoate Gamma Eudesmol Isobutyl decanoate Isopentyl decanoate Ethyl hexadecanoate alpha Terpineol acetate Terpineol alpha # OF39 OF40 OF44 OF35 OF36 OF37 OF38 OF41 OF42 OF43 OF45 OF46 OF47 OF48 OF49 OF50 OF51 OF52 OF53 OF54 OF55 OF56 OF57 OF58 OF59 OF60 Supplementry Table S5.1 Table Supplementry equivalents. canone 135 Chapter 6: Sensory Attributes and Flavor Chemistry of Acetic Fermentations with Novel Plant Ingredients

Introduction Vinegar is an acidic liquid resulting from the process of oxidation of ethanol into acetic acid by various strains of acetic bacteria (sometimes called acetic “fermentation”) (Perlman et al. 1977). Acetifcation has been used for centuries to make Aceto Balsamico and other grape-based vinegars, as well as vinegars from substrates such as fruit, cereals, palm, and whey (Giudici et al. 2009, Ou et al. 2009, Liu et al. 2004, Igbinadolor et al. 2009, Gonzalez and Vuyst 2009). Vinegar- making is documented in ancient cultures such as the Egyptians, Sumerians, Babylonians, and the Zhou Dynasty, and is mentioned in medical and religious texts dating back to antiquity. It was in part the study of vinegar, that led Antoine Lavoisier to give oxygen its name, from the Greek term for “Acid Former” (Mazza and Murooka 2009). Te culinary uses of vinegar include pickling and other forms of preservation, as a condiment, as a component of other condiments or sauces, in beverages such as switchel or shrub, and as a fnishing component to a dish (Smith 2013). It is a source of sour favors as an alternative to, or in cuisines which lack ready access to, citrus, as well as a source of favors retained from the substrate from which it is made.

Tere has been an increased interest in recent years, among chefs at the high levels of gastronomy as well as home cooks, in studying the processes behind typically commercial or artisanal ingredients and applying them in the kitchen to new or atypical substrates to yield new ingredients and favors. Tis has been paralleled by a similar rising interest in developing culinary uses for underutilized local, wild, and waste products, some of which may be used “as-is” in a dish and some of which beneft from some kind of processing, including microbial fermentations (Nilsson 2012, Redzepi 2010 and 2013, Katz and Pollan 2012, Williams 2012, Mouritsen et al.

136 2012, Felder et al. 2012, Atala 2012 and 2013). In many cases the results of this culinary research and development are both a specifc, highly local product, as well as knowledge that is broadly applicable for other ingredients and by cooks in any kitchen in the world. A more formal analysis and communication of the volatile chemistry and resulting favor qualities of these novel ingredients and processes, especially as used in a culinary setting, has not been performed, despite expressed interest on the part of chefs and consumers. Novel products resulting from these interests have been put on menus as dishes or components of dishes, and described in peer-reviewed literature (Felder et al. 2012, Atala 2012, Mouritsen et al. 2012), cookbooks (Atala 2013, Nilsson 2012, Redzepi 2013), as well as more rapidly and informally in digital formats (Williams 2011). An accelerated technique for making vinegar by adding alcohol and an unpasteurized vinegar starter to novel substrates and acetifying with continuous addition of atomized air has been described thusly (Reade 2012). Tis relatively inexpensive process, adapted from forced-air methods used in industrial vinegar production (Garcia-Garcia et al. 2009), has been used for the production of vinegars out of various seasonal and atypical ingredients. In the present study the main purpose was to describe the sensory, volatile, and organic acid profles of several novel acetic fermentations carried out by this technique; to use multivariate statistical techniques to model their relationships, describe the efects of this accelerated acetifcation process on the chemistry and favor of diverse ingredients; and to use this data to inspire further culinary development in acetic fermentation.

Materials and Methods Acetifcation: Forced aeration was used to acetify eight substrates (Table 6.1), which included celery juice, asparagus juice, rhubarb juice, strawberry wine, elderfower wine, spruce needles extracted into water, and licorice root extracted into water. Te substrates were chosen based on seasonal availability in Copenhagen, Denmark and previous experimentation carried out at the Nordic Food Lab; preparation of the substrates is described below. Acetifcation was accomplished by

137 adding ethanol at 6% v/v to each liquid, except where noted below; adding 20% v/v to this of unpasteurized apple vinegar (Meyers, Copenhagen, Denmark) except in the case of pine vinegar, noted below, to inoculate with acetic bacteria and provide a reduced-pH, environment; and aerating, covered with nylon mesh but not sealed, with an aquarium pump rated at 1.5 L/min ftted with a 4-cm airstone for 5 days. Juice Vinegars: Celery (Apium graveolens), green asparagus (Asparagus ofcinalis), and rhubarb (Rheum rhabarbarum) were purchased commercially and each processed while fresh in a centrifugal juicer. To the celery and asparagus juices, grain neutral spirits, 95% alcohol by volume, was added to reach 6% (v/v) ethanol. Two separate rhubarb vinegars were produced; one with 8% ethanol (v/v) (Rhubarb 1) added, and one with 6% ethanol (v/v juice) and the rhubarb solids by-product from the juicing (equivalent to 50% of the weight of the juice) along with 1% of Celluclast enzyme (Novozymes, Bagsværd, Denmark) added (Rhubarb 2). Rhubarb 2 was strained through a fne- mesh strainer following acetifcation. All of the juices were processed into vinegar, as above. Tea Vinegars: Commercially purchased dry licorice roots (Glycyrrhiza glabra) and fresh Norway spruce needles (Picia abies) foraged in Copenhagen, Denmark were each pulverized in a thermomix blender (Vorwerk, Wuppertal, Germany ). Each were separately mixed with water at a rate of

Table 6.1 Samples used in the study, their substrates, and source of alcohol for acetifcation

Sample Name Substrate Alcohol Source Celery Juiced Celery grain alcohol Asparagus Juiced Asparagus grain alcohol Rhubarb 1 Juiced Rhubarb grain alcohol Rhubarb 2 Juiced Rhubarb + Rhubarb Solids grain alcohol Licorice Licorice roots extracted into water at 60°C grain alcohol Pine Spruce needles extracted into water at 60°C grain alcohol Strawberry Juiced Strawberries yeast fermentation Elderfower Elderfowers extracted into acidifed sugar syrup yeast fermentation

138 15% plant material by mass, sealed in plastic bags using a chamber vacuum sealer, and cooked in a water bath held at 60°C for 2 hours to extract into a “tea.” Solids were removed by straining through a fne mesh chinois and the “tea” was acetifed, as above. Wine Vinegars: Wines were produced from strawberries and elderfowers. Strawberry wine was made from commercially purchased Danish strawberries (Fragaria x ananassa var ‘Dania’) by frst juicing the fruit, which yielded juice measuring 7 Brix. Juice was reduced in a pan over an induction stove at a slow boil (approximately 95°C) to ¼ of its initial volume and mixed with enough fresh juice to yield a mixture with sugar measured at 13 °Brix using a refractometer. 4 L of this partially reduced juice mixture was fermented with 2 g yeast (Saccharomyces cervisiae var. bayanus, Vinoferm Bioferm Champ, Brouwland, Beverlo, Belgium), sealed with an airlock in a 5-L plastic container and fermented for 10 days, until carbon dioxide production had ceased, with a fnal refractometer reading of 4 °Brix. Elderfower wine was made by gathering wild Elderfowers (Sambucus nigra), dissolving sucrose into fltered water to reach 130g/L and adding 5g/L of citric acid, then steeping 24 heads of fowers in 3.75 L of the sugar-citric acid solution for 48 hours at room temperature, afer which the fowers were strained out through a fne mesh chinois. 2 g Champagne yeast (Saccharomyces cervisiae var. bayanus Vinoferm Bioferm Champ, Brouwland, Beverlo, Belgium) was added and the mixture was fermented, sealed with an airlock,

until CO2 production had ceased, 8 days, yielding a fnal measurement by refractometer of 7 ° Brix. Both wines were transferred by pouring into a secondary container, leaving lees behind, then acetifed using the process outlined above. Sensory Analysis: Te vinegar samples were profled using a descriptive analysis method, whereby the intensity of panel-generated terms are evaluated for each sample. Ten panelists, two male, eight female, ages 20-70, and screened for specifc anosmias, were recruited through the University of Copenhagen’s Sensory Science group. Over three days, trainings were conducted, where the panelists tasted fve of the eight vinegar samples and discussed the favor attributes they felt were

139 important for each. In the second and third sessions, panelists were presented with references for the terms they had previously agreed on as being important, and narrowed down the list of possible descriptors by discussion and consensus to the fnal list, Table 6.2. References were presented in opaque black glasses. No matrix (e.g. white vinegar) was used for the references as panelists were fatigued by the pungency. During the third training session the panelists were trained in the use of the FIZZ computer terminal program (Biosystemes, Couternon, France), which was used to collect their intensity ratings for each descriptor in each sample. Te panelists performed the descriptive analysis in triplicate over three days, rating the intensity of all descriptors for each sample on a 14-cm ungradiated line scale ranging from “very low” to ”very high”. Samples were presented in randomized order determined by a Williams Latin Square design, in opaque black glasses marked by 3-digit codes. Te panelists were asked to smell each vinegar, rate the aroma terms, then taste the sample and rate the taste terms, and fnally to expectorate the sample. Panelists were provided with crackers, cucumber slices, and water to rinse and cleanse their palates between samples. Volatile Analysis: Te volatiles in each vinegar were profled using Headspace-Solid Phase Microextraction- Gas Chromatography-Mass Spectrometry (HS-SPME-GC-MS). Extraction: In triplicate, 10 mL of each vinegar was placed in 20 mL headspace vials (Supelco, Bellefonte, PA), along with 25 μg/L of 2-undecanone (Sigma-Aldrich, St Louis, MO) as an internal standard, and capped with magnetic caps with a 2-mm thick PTFE-faced silicone septum (Supelco). Each vial was agitated at 500 rpm for 5 minutes and then a 2-cm long, 50/30 μm-thick Polydimethylsiloxane/ Divinylbenzene/Carbowax-coated fber (Supelco) was inserted into the headspace and volatiles were extracted for 15 minutes at 30°C, afer which they were directly introduced into the inlet of the GC. Chromatographic Conditions:

Te volatile-loaded SPME fber was inserted into the inlet of an Agilent model 5975 Gas Chromatograph-Single Quadrupole Mass Spectrometer (Agilent Technologies, Santa Clara, CA),

140 Table 6.2 Descriptors used in the sensory anaysis, their references, and their signifcance for product by pseudo-mixed model ANOVA

Aroma Term Reference Signifcance Red berry 3 fresh raspberries and 10 thawed frozen redcurrants P<0.05 strawberry 2 fresh strawberries, halved P<0.05 Acetic acid 20 mL of distilled white vinegar P<0.05 Rotten fruit 3 2-cm cubes each of fresh apple and pear, sprinkled with P<0.05 yeast and lef on the counter overnight to partially ferment chemical 10 mL nail polish remover P<0.05 apple 2 slices fresh green apple P<0.05 licorice 3 black licorice candies P<0.05 yeast 1 fresh yeast cube P<0.05 wine 20 mL of red wine P<0.05 Tropical fruit 20 mL tropical fruit juice + 2 2-cm cubes fresh mango ns rhubarb 20 mL fresh rhubarb juice P<0.05 celery 5 1-cm cubes celery root, microwaved for 30 sec P<0.05 earthy 20 mL of freshly dug soil P<0.05 Green vegetable 5 1-cm strips of cucumber skin P<0.05 citrus 1 wedge orange and ½ wedge grapefruit ns pine Approximately 50 fresh pine needles P<0.05 Blue cheese 2 cm cube gorgonzola cheese P<0.05 Taste/mouth- feel Term Reference Signifcance sour 2 g/L citric acid in water P<0.05 bitter 1g/L cafeine in water P<0.05 astringent 1 g/L alum in water P<0.05 sweet 20 g/L sucrose in water P<0.05 salt 1 g/L salt in water P<0.05 umami 10 g/L monosodium glutamate in water P<0.05 which was held at 250°C in splitless mode for all injections and had a 0.75 mm I.D. SPME inlet liner (Agilent Technologies). Te fber was desorbed for 10 minutes. Te GC oven was held at an initial temperature of 40°C for 5 minutes, then the temperature was increased at a rate of 2°C per minute to 150 °C, followed by a second temperature increase at a rate of 30°C per minute to 240°C; the oven was held at the fnal temperature for 5 minutes. A 30-meter long, 250-μm

141 internal diameter DB-Wax column with a flm thickness of 0.25 μm (J&W Scientifc, Folsom, CA) was used at constant fow of 1.0 mL/minute with helium as the mobile phase. Te split vent was opened at 2 min into the run with a split fow of 9.5 mL/ minute. Te separated volatiles were directed via a transfer line at 240 °C to the MSD, run in EI mode at 70 eV scanning, afer a 1.5 minute solvent delay, from 40.0-300.0 m/z with a source temperature of 230°C and a Quadrupole temperature of 150°C. Compound Identifcation and Relative Quantifcation: For each sample, volatiles were identifed by comparing the background-subtracted mass spectrum of each peak from the sample chromatograms to the NIST 05 Mass Spectral Database and comparing calculated retention indices and retention times to published retention index data and authentic standards, when available. Retention Indices were calculated from series of C8-C20 alkanes (Sigma-Aldrich) injected under identical analysis conditions. Each chromatographic peak was manually integrated using Chemstation version E.01.01.335 sofware (Agilent Technologies) and all peaks were quantifed relative to 2-undecanone by normalizing to 2-undecanone peak area on that chromatogram. Capillary Electrophoresis: Organic acids were analyzed on an Agilent model 7100 capillary electrophoresis system (Agilent Technologies) with a 72-cm fused-silica capillary and organic acid bufer at 20°C, using the Agilent PN 5063-6510 organic acids analysis kit. A mixture of sodium salts of malic, citric, lactic, succinic, tartaric, and oxalic acids (Agilent Technologies and Fisher Scientifc, Fair Lawn, NJ) were used to create 5-point calibration curves (data not shown). Vinegar samples were diluted 1:10, 1:100, and 1:1000, and the dilutions which fell within the linear range of the curve were used for quantifcation. Statistical Analysis: A 3-way Analysis of Variance (ANOVA) with two-way interactions was performed on the sensory descriptive analysis data with products, judges, and replicates as the main efects and each descriptive term as the response variable. Where necessary, a pseudo-mixed

142 model was employed. Using terms found to be signifcant by ANOVA, a Principal Component Analysis (PCA) was performed on the sensory data, and a Partial Least Squares (PLS) regression analysis was performed using the GC and CE data as independent variables and sensory data as dependent variables. To account for variations in scale usage, for example between peak areas in the GC data and ratings between 0 and 15 in the sensory data, as well as the indirect relationship between relative analyte concentration and sensory impact for volatile compounds, the data was standardized by weighting to the standard deviation of each attribute. Te ANOVA and PCA was performed in the R statistical package (R Foundation for Statistical Computing, Vienna, Austria) and the PLS was performed in Unscrambler (CAMO sofware Inc, Woodbridge, NJ, USA)

Results and Discussion Sensory Analysis: Using a pseudo-mixed model, all terms except tropical fruit and citrus difered signifcantly between the samples at p<0.05 (Table 6.2, 6.3). Principal Component Analysis (PCA) was performed to explore the relationships among samples and attributes. Te frst and second Principal Components (PCs) of the PCA (Figure 6.1) explained 50.65% and 22.96% of the total variance in the sensory descriptive data, respectively. Te third PC described 11.59% of the total variance. PC 1 generally separates samples made from “fruity” substrates from those made with vegetable material (rhubarb is, botanically speaking, a vegetable rather than a fruit, but its strong red berry aroma makes it more fruit-like than vegetable-like in this context). Te second PC is dominated by a bitter/sour/unsweet-sweet continuum, separating the rhubarb-based samples from the sweeter strawberry, and the pine, celery, and asparagus vinegars from the sweeter licorice and elderfower. Some of the high-variance loadings (those with longer vectors) apply mostly to a specifc substrate- strawberry aroma is rated signifcantly higher for the strawberry sample than for any others, with the rhubarb samples rated second-highest but signifcantly lower. Similarly, rhubarb was rated signifcantly higher for the rhubarb samples than for any others, and second-highest though signifcantly lower for the strawberry sample. Red Berry aroma

143 Table 6.3 Mean intensities for sensory descriptors for each sample. Signifcantly diferent values determined by a Tukey’s Honest Signifcant DiferenceSheet4 test have diferent letter groupings.

Vinegar Sample Attribute Asparagus Celery Elderflower Licorice Pine Rhubarb 1 Rhubarb 2 Strawberry Red berry 0.1 d 0.2 cd 1.3 c 0.1 cd 0.2 cd 8.0 a 7.8 a 5.6 b strawberry 0.1 d 0.1 d 1.8 c 0.2 d 0.1 d 4.8 b 4.2 b 11.1 a Acetic acid 6.1 b 2.2 d 6.5 ab 7.8 a 4.1 c 1.5 d 1.5 d 2.1 d rotten fruit 6.0 a 1.7 d 4.7 ab 3.7 bc 2.2 d 2.8 cd 2.6 cd 4.9 ab chemical 7.0 b 3.1 cd 8.5 ab 9.8 a 5.6 b 1.5 d 1.1 d 1.7 cd apple 1.6 bcd 1.6 cd 4.5 a 2.6 b 2.2 bc 1.4 cd 1.3 cd 0.7 d licorice 0.4 cd 1.1 bcd 1.6 ab 2.5 a 1.5 abc 0.3 d 0.3 d 0.4 d yeast 3.9 ab 1.8 de 3.0 bcd 2.2 cde 1.4 e 3.0 bc 1.8 de 4.3 a wine 1.5 e 1.4 e 4.0 bc 2.8 cd 2.4 de 5.0 b 4.2 b 7.4 a Tropical fruit 1.7 a 1.6 a 2.4 a 2.7 a 0.9 a 1.6 a 1.4 a 2.1 a rhubarb 0.5 c 0.5 c 0.6 c 0.5 c 0.4 c 9.8 a 9.8 a 4.0 b celery 2.1 b 12.4 a 0.5 cd 0.3 d 1.2 c 0.3 d 0.2 d 0.1 d earthy 4.0 b 4.5 ab 2.4 c 2.5 c 5.4 a 1.4 cd 1.4 cd 1.0 d Green vegetable 2.1 b 3.8 a 1.7 b 1.6 bc 2.2 b 0.7 cd 0.7 cd 0.4 d citrus 2.2 a 2.9 a 2.8 a 3.1 a 3.3 a 2.8 a 2.5 a 1.4 a pine 3.6 b 3.8 b 3.0 b 3.3 b 11.3 a 0.8 c 0.2 c 0.6 c Blue cheese 6.7 a 0.7 de 1.0 bc 0.8 bcd 0.7 bcd 0.1 cd 0.1 d 1.0 b sour 9.6 cd 8.7 de 9.3 cd 10.4 bc 8.7 de 12.3 a 11.9 ab 7.4 e bitter 6.9 bc 8.3 ab 4.1 d 6.0 c 8.9 a 9.2 a 7.8 ab 5.7 cd astringent 7.7 cde 7.0 de 6.4 e 9.5 bc 8.7 cd 12.5 a 11.0 ab 7.0 de sweet 3.4 b 2.8 b 8.7 a 9.2 a 3.1 b 2.2 b 2.2 b 8.1 a salt 9.4 a 8.7 a 3.3 c 5.9 b 5.3 b 3.6 c 3.4 c 1.8 d umami 8.8 a 8.3 a 3.1 c 4.5 b 4.5 b 2.1 c 2.1 c 1.9 c follows the same pattern as rhubarb aroma, but describes a raspberry/redcurrant-like quality rather than pure rhubarb. Celery aroma is rated signifcantly higher in the celery sample than any other sample, with a weaker expression in the asparagus and pine samples. It was decided by the panelists that the descriptor celery was better represented by a slightly cooked celery root, rather than fresh celery, while the juice-based samples were made with uncooked vegetables. Te acetifcation process, for these samples, appears to create or bring out a “cooked” type aroma for these ingredients. Green vegetable aroma, the standard for which was raw cucumber skin, was rated signifcantly higher for the celery sample than for any other, though lower within the celery sample than the celery descriptor (signifcance not calculated), and second-highest for pine, asparagus, elderfower, and licorice vinegars (not signifcantly diferent from each other).

Two aroma descriptors—chemical and acetic acid—suggest a similarity in processing-derived aromas in the water-based samples (pine and licorice were extracted into water as an initial

144 Page 1 Figure 6.1 Principal Component Analysis (PCA) biplot of sensory descriptive analysis data. PC1 (X) explains 50.6% of variance, PC2 (Y) explains 22.9% of variance. Samples (scores) are in bold black text, descriptors (loadings) are in red text.

step, and elderfower into a water and sucrose mixture) and the asparagus sample. For both the chemical and acetic acid descriptors, the licorice and elderfower samples were rated highest (and not signifcantly diferent from each other, table 3) and the asparagus and pine samples were rated second-highest, and signifcantly higher than the other four samples. Te sweet descriptor separates the samples into two groups- those likely containing sugar that had not been fully converted to alcohol by the yeast (the elderfower and strawberry samples both had a °Brix level above zero when acetifcation began) or non-saccharide sweet compounds (licorice root contains glycyrrhizin) were signifcantly sweeter than the rest of the samples, whose substrates were not processed to concentrate sugars to begin with.

145 Chemical Analysis: A total of 104 peaks were found in the non-targeted GC-MS profle of the vinegars, with between 19 and 54 compounds per vinegar sample (Table 6.4). Nine compounds were detected in all 8 vinegars which suggests that they are likely related to the acetic fermentation itself: acetaldehyde, ethyl acetate, ethanol, isoamyl acetate, acetic acid, ethyl hexanoate, benzaldehyde, ethyl decanoate, and ethyl cinnamate. 56 of the observed compounds were present in only one vinegar sample, suggesting that they likely originated in the starting materials rather than a fermentation step. Interestingly, the capillary electrophoresis (table 6.5) showed that many of the samples had a low concentration of acetic acid, compared to levels of 4-5% or higher common in commercial vinegars. Tis was especially true for the rhubarb samples, which had little to no measurable acetic acid. Since rhubarb juice is very acidic to start with, it is difcult to gauge by taste the extent to which acetic fermentation occurs in this substrate, something kitchen practitioners (who do not typically perform capillary electrophoresis) should be aware of. Alongside acetic acid and the endogenous malic, citric, and oxalic acids, a number of the samples contained lactic acid, suggesting that lactic bacteria may have co-fermented the mixtures alongside the acetic bacteria that were added intentionally. It is becoming increasingly clear that microbial ecology, rather than the presence of any one species, is essential for understanding the functional and metabolomic properties of fermented systems. Co-fermentation is a familiar phenomenon in wine, where Saccharomyces cerevisiae, Brettanomyces sp., and malolactic bacteria such as Oenococcus oenii may ferment alongside each other in the must (Fleet 2003, Versari et al. 1999). Co-fermentation also takes place in cheeses (especially those with mold species, such as blue cheeses or bloomy-rinded cheeses such as Camembert), sourdough breads, kimchi and other lacto-fermented vegetables, and soy sauce (Wei et al. 2013, Tanaka et al. 2012, Bokulich et al. 2014, Cho et al. 2014, Hansen 2004).

146 Table 6.4 Volatiles identifed in vinegar samples, by chemical name, Chemical Abstracts Service number (CAS), retention time (RT), C8-C20 retention index (CRI), and literature retention index. ni= not identifed. Literature Retention Index # namea CAS RT CRI pherobase flavornet other

V1 acetaldehyde s 75-07-0 1.82 >800 500

V2 Dimethyl Sulfide s 75-18-3 1.96 >800 844 716

V3 octane s 11-65-9 2.20 800 800

V4 methyl acetate s 79-20-9 2.34 815 856-864

V5 ethyl acetate s 141-78-6 2.88 877 885-898 907

V6 3-methylbutanal s 590-86-3 3.22 908 912 910

V7 ethyl propanoate s 105-37-3 3.92 944 950 951

V8 propyl acetate s 109-60-4 4.25 961

V9 pentanal s 110-62-3 4.29 963 935 935 V10 ni.a 4.44 971

V11 isobutyl acetate s 110-19-0 5.20 1005 1005-1007 1015

V12 Alpha-pinene s 80-56-8 5.41 1011 1027-1034 1032

V13 2-butanol s 78-92-2 5.79 1021 1022 V14 E 2-butenal 1191-99-7 5.83 1023 1023-1034

V15 ethyl butanoate s 105-54-4 5.93 1025 1022-1057 1028

V16 1-propanol s 71-23-8 6.12 1030 1038-1045 1037

V17 ethyl 2-methylbutyrate s 7452-79-1 6.50 1041 1056-1069 1050

V18 camphene s 79-92-5 6.84 1050 1077 1075

V19 ethyl isovalerate s 108-64-5 7.07 1056 1053-1082 1060

V20 hexanal s 66-25-1 7.44 1066 1067-1093 1084

V21 Beta-pinene s 127-91-3 8.05 1082 1113-1124 1116

V22 isobutanol s 78-83-1 8.54 1095 1085-1125 1099

V23 isoamyl acetate s 123-92-2 9.24 1110 1118-1147 1117

V24 ethyl valerate s 539-82-2 9.85 1122 1120-1170 1133

V25 3-carene s 13466-78-9 9.99 1125 1148 1148

V26 Alpha-phellandrene s 99-83-2 10.87 1142 1205

V27 Beta-myrcene s 123-35-3 11.18 1148 1161-1187 V28 Alpha-terpinene 586-62-9 11.67 1157 1178 1178

V29 heptanal s 111-71-7 12.28 1169 1197 1174

V30 methyl hexanoate s 106-70-7 12.51 1174 1188

V31 limonene s 138-86-3 12.78 1179 1198-1234 1201

V32 E-2-Hexenal s 6728-26-3 13.92 1201 1201 1192-1220

V33 isoamyl alcohol s 123-51-3 14.21 1206 1205 V34 2-pentylfuran 3777-69-3 14.88 1217 1240

V35 ethyl hexanoate s 123-66-0 15.30 1224 1224-1270 1220 V36 Gamma-terpinene 99-85-4 15.53 1228 1262-1265 1238 V37 M-cymene 535-77-3 16.84 1250 1267 V38 ni.b 17.21 1256

V39 hexyl acetate s 142-92-7 17.66 1264 1264 1270 V40 acetoin 513-86-0 17.68 1264 1272-1295 1287 a= mass spectrum matched to NIST >80%; b=Tao et a. 2006; c=Gyawali et a., 2009; d=Kaack et .l, 2006; s=matched to authentic standard

147 Table 6.4 (continued)

Literature Retention Index # namea CAS RT CRI pherobase flavornet other

V41 octanal s 124-13-0 18.35 1275 1300-1207 1280

V42 1-octen-3-one s 4312-99-6 19.09 1287 1305-1323 1285 V43 ethyl-3z-3-hexenoate 64187-83-3 19.31 1291 1269

V44 cis-3-hexenyl acetate s 3681-71-8 20.17 1305 1308 1327 V45 cis-2-penten-1-ol 1567-95-0 20.73 1314 1321

V46 ethyl heptanoate s 106-30-9 21.38 1325

V47 ethyl lactate s 97-64-3 21.73 1330 1353 1358 V48 isobutyl hexanoate 105-79-3 22.62 1345

V49 hexanol s 111-27-3 22.96 1350 1351-1392 1360

V50 3-hexen-1-ol s 928-96-1 23.37 1357 1378-1407

V51 nonanal s 124-19-6 24.77 1380 1402-1415 1385 V52 pentyl benzene 538-68-1 25.67 1394 V53 acetic acid 64-19-7 27.45 1423 1434-1477 1450

V54 ethyl octanoate s 106-32-1 27.68 1427 1422-1446 1436 V55 isopentyl hexanoate 626-77-7 29.16 1452 1450b

V56 benzaldehyde s 100-52-7 31.49 1490 1525 1495

V57 ethyl nonanoate s 123-29-5 33.77 1529 1528 V58 Junipene 475-20-7 34.31 1538 1595c

V59 linalool s 78-70-6 34.63 1544 1484-1570 1534d

V60 1-octanol 72-69-5 35.29 1555 1557-1566 1553

V61 bornyl acetate s 5655-61-8 35.52 1559 1580

V62 isobutyric acid s 79-31-2 35.44 1557 1584-1588 1563

V63 caryophyllene s 87-44-5 36.20 1571 1608-1618 2,5-Dimethyl-4-methoxy- V64 3(2H)-furanone 4077-47-8 36.18 1570 1584 1580 V65 camphene hydrate 465-31-6 36.68 1579 1517 V66 hydroxylinalool 68042-45-5 38.02 1602 1580d

V67 butanoic acid s 107-92-6 38.81 1616 1628-1650 1619

V68 acetophenone s 98-86-2 38.82 1616 1645 V69 hotrienol 29957-43-5 39.11 1622 1586 1623

V70 ethyl decanoate s 110-38-3 39.68 1632 1630 1655 V71 Alpha-humulene 6753-98-6 40.18 1641 1680 1663 V72 isoborneol 124-76-5 40.63 1649 1660

V73 isovaleric acid s 503-74-2 41.14 1658 1660-1691 1665 V74 ni.c 41.71 1668 V75 D-verbenone 80-57-9 41.96 1673 1733 V76 ethyl-9-decenoate 67233-91-4 42.46 1682 1694

V77 Borneol s 464-45-9 42.57 1684 1642 1677

V78 Alpha-terpineol s 98-55-5 42.58 1684 1669-1720 1688 V79 B-eudesmene 17066-67-0 42.89 1689 a= mass spectrum matched to NIST >80%; b=Tao et a. 2006; c=Gyawali et a., 2009; d=Kaack et .l, 2006; s=matched to authentic standard

148 Table 6.4 (continued) Literature Retention Index # namea CAS RT CRI pherobase flavornet other V80 Alpha-selinene 473-13-2 43.22 1695 1724 1711 V81 piperitone 89-81-6 43.23 1696 1739 V82 Alpha-amorphene 483-75-0 43.59 1702 1691 V83 benzyl acetate 140-11-4 43.59 1702

V84 linalool oxide s 39028-58-5 44.78 1724 1731-1747 V85 Beta-cadinene 24406-05-1 45.26 1733 V86 epiglobulol 88728-58-9 45.55 1739 V87 Alpha-curcumene 644-30-4 46.28 1752 1777 1773

V88 phenethyl acetate s 103-45-7 48.22 1788 1803 1829

V89 Trans-carveol s 1197-07-5 49.83 1819 1876 1839

V90 hexanoic acid s 142-62-1 50.57 1834 1863 1829 V91 ethyl dodecanoate 54982-83-1 50.71 1837 1822

V92 benzyl alcohol s 100-51-6 51.38 1850 1837 1865

V93 ethyl dihydrocinnamate s 2021-28-5 51.77 1857 1897 1906 V94 dihydrocarveol 38049-26-2 52.15 1865 1713-1941

V95 phenylethyl alcohol s 60-12-8 53.12 1884 1905-1940 1829 V96 ni.f 53.70 1895 V97 4-methoxy-3-methyl-phenol 14786-82-4 55.15 >2000

V98 Gamma-nonalactone s 104-61-0 58.40 >2000 2042

V99 ethyl tetradecanoate s 124-06-1 60.62 >2000 2029 2042

V100 ethyl cinnamate s 103-36-6 61.63 >2000 2139

V101 eugenol s 97-53-0 62.10 >2000 2141-2192 2141 V102 ni.d 62.50 >2000 V103 Alpha-cadinol 481-34-5 62.80 >2000 2211 2191 V104 ni.e 65.54 >2000 a= mass spectrum matched to NIST >80%; b=Tao et a. 2006; c=Gyawali et a., 2009; d=Kaack et .l, 2006; s=matched to authentic standard

Table 6.5 Organic acids in g/L determined by capillary electrophoresis

149 Figure 6.2 Partial Least Squares Regression (PLS) analysis of sensory and chemical data on vinegar samples. 6.2A: Positions of samples 6.2B: Biplot of descriptors and compounds 6.2C: exploded view of positions of compounds

150 C

151 PLS Te PCA shows that, when examined on its own, the favor profles of these samples are both refective of the original favors of the starting materials employed in their production, as well as those resulting from the biochemical transformations brought about during fermentation. As we know from the analysis of other foods, especially those which are botanically-derived or microbially altered (both of which are true for all of these samples), the relationship between chemical composition and favor is not necessarily straightforward (Johnson et al, 2012). While there is sometimes a direct relationship between the presence of an “impact compound” and the presence of a particular aroma, it is more ofen the case that perceived aroma results from the efects of mixing many aroma-active compounds at a variety of concentrations, leading to additive, subtractive, and synergistic relationships between chemistry and favor. A PLS (fgure 6.2) was performed to relate chemical composition of the samples to their sensory attributes. Interpretation of the PLS is guided by recognizing that PLS reveals correlative relationships rather than causative ones. Te PLS explained 23% of the chemical and 34% of the sensory variance in the frst PC, and 21% of the chemical and 19% of the sensory variance in the second PC. Comparing the placement of the scores between the PCA and the PLS, the models show some of the same trends: the strawberry and rhubarb samples cluster in one quadrant, as do the elderfower and licorice samples. Te pine, celery, and asparagus samples are more separated in the PLS than they are in the PCA; in both analyses, they are not separated very much by PC1, but are separated more by PC2 in the PLS than by PC2 in the PCA. Tis may mean that the chemical composition of the asparagus sample is a greater source of diferences from the pine and celery samples than its sensory profle. Similar to the PCA, PC1 of the PLS appears to describe a continuum going from fruity, sweet, and yeast-fermentation-related terms on the lef, to vegetative and earthy terms on the right; PC2 describes fresher, fruit- and vegetable- related terms on its lower end, and yeasty, lipolytic terms on its high end.

152 Te PLS analysis models some of the straightforward relationships we would expect to see- oxalic acid, produced out of all the starting materials only by rhubarb, is highly associated with the rhubarb vinegars and generally with sour taste and pH is highly negatively correlated with sour taste (since as pH decreases, the concentration of acid increases) A number of volatile compounds were detected in a majority of the samples. Ethyl pentanoate, hexanol, ethyl octanoate, and isovaleric acid were detected in fve samples; hexanal, ethyl heptanoate and ethyl nonanoate in six samples; isoamyl alcohol, octanal, and phenylethyl alcohol in seven samples; and acetaldehyde, ethyl acetate, isoamyl acetate, ethyl hexanoate, acetic acid, benzaldehyde, ethyl decanoate, and ethyl cinnamate in all eight samples. Ethyl acetate, ethyl decanoate, ethyl hexanoate, and ethyl cinnamate are very close to the center of the PLS plot, suggesting that they contribute little to the separation of the samples via sensory diferences. On the other hand, several of these common compounds occupy the high-variance edges of the plot, such as isoamyl alcohol, ethyl nonanoate, phenylethyl alcohol, and isoamyl acetate, suggesting that some of the sensory diferences in the samples arise from diferential concentrations or contributions of these compounds, rather than their presence or absence (as would be suggested by an uncommon compound plotting at the high-variance edges of the PLS). As observed from its greater relative distance from other samples in the PLS, and location near the periphery, one of the most diferent samples with regard to chemistry and sensory qualities is asparagus vinegar, which is also highly associated with blue cheese, acetic acid, and chemical aromas. Tese are not necessarily associated with the aroma of fresh asparagus, so the presence and separation of this vinegar and its favors may refect something unique in

the transformation of asparagus, specifcally, into vinegar1. Some of the volatile components associated with these descriptors in the PLS have been described in the literature as having sensory qualities congruent with this hypothesis—dimethyl sulfde has cabbage-, sulfur-, and gasoline-like qualities, acetaldehyde is pungent and ether-like, hexanoic acid has sweaty qualities, and octane has alkane and gasoline-like qualities (Flavornet 2014). Some PLS-correlated

1 Unfortunately there is not much information in the literature of the fatty acid 153 compounds like 2-pentylfuran, pentanal and octanal have some qualities that match these correlations and some that don’t: 2-pentylfuran has green bean, waxy, musty, and fruity aromas (Perfavory 2014), pentanal has pungent, malty, and almond characteristics, and octanal has fatty, soapy, lemony, and green qualities (Flavornet 2014). Blue cheese, acetic acid, and chemical are also associated in the PLS with a number of compounds with more typically fruity (isoamyl acetate, isobutyl acetate, ethyl octanoate, hexyl acetate), foral (benzyl alcohol), minty/spicy (verbenone), or coconut/peach qualities (gamma-nonalactone). Te aromas of these compounds may be masked by the more obviously chemical or blue cheese-like volatile components of the sample, or alternatively, while in relatively high concentrations compared to other samples, may contribute to a base aroma that is less important for distinguishing it from the other samples. Both of the yeast-fermented vinegars (strawberry and elderfower) plot towards the lef side of the frst PC, and cluster near the yeast and wine sensory terms that appear to arise from this process. Te strawberry vinegar is more highly associated with wine, and the fermentation- related alcohols and esters isobutyl and isoamyl alcohol as well as ethyl nonanoate, propanoate, decanoate, and dodecanoate. Elderfower vinegar, on the other hand, is more associated with the yeast term and the compounds acetoin, isovaleric acid, ethyl isovalerate, isobutyric acid, and ethyl acetate, as well as with the likely foral-derived (Kaack et al. 2006) compounds benzaldehyde, linalool, and cis-linalool oxide. Benzaldehyde is also a fermentation byproduct, but is present in much higher levels in the elderfower vinegar than the other samples. Te linalool, hydroxylinalool, linalool oxide, hotrienol, and benzaldehyde present in the vinegar have been previously identifed in elderfower (Kaack et al. 2006). Both the strawberry and elderfower vinegars express compounds and aromas that seem to be distinctly related to the yeast fermentation that they underwent, suggesting that this additional fermentation step adds distinct compounds and aromas to the resulting vinegar, which may be desirable for inclusion in vinegar production. However, the PLS also shows that this fermentation step is expressed diferently, both in terms of chemical and sensory results, in the fnished product. Te strawberry wine was fermented from partially reduced strawberry juice, which contains a variety of non-volatiles such

154 as anthocyanins, acids, and glycosides (Kalt et al. 1999, Ubeda et al. 2012) which may be afected diferently in terms of post-fermentation chemical and aroma profles. Tis is in contrast to the wine made from the elderfower extraction. While the licorice vinegar plots somewhat close to the elderfower vinegar sample on the PLS, it is also located close to the low-variance center of the plot, where few compounds or sensory attributes dominate; with 18 headspace compounds it is the least complex of all the samples in terms of volatile. Red wine vinegar, traditional balsamic vinegar, sherry vinegar, qu or koji-based solid state vinegars, and vinegars from less common substrates such as strawberries, onions, and banana have previously been investigated for their chemical and sensory properties. A number of these studies identifed compounds with sensory activity that were common to a majority of the samples in the current study. Tesfaye et al. (2009) found acetaldehyde, ethyl acetate, n-propanol, isoamyl acetate, and phenethyl alcohol in sherry vinegars, and Aceña et al. (2011) calculated high odor activity values (which suggests potential sensory importance) in sherry vinegar for isovaleric acid, found in fve of the present samples, isoamyl alcohol, found in seven of the present samples, and isoamyl acetate, found in all of the present samples. Te latter study also calculated high odor activity values for ethyl isovalerate and isobutyric acid, which were detected in both of the yeast-fermented samples in this study. Similarly, Charles et al. (2000) found isoamyl acetate, benzaldehyde, phenethyl acetate, and phenethyl alcohol in red wine vinegars. A study of strawberry vinegar found a number of compounds in common with the strawberry vinegar in the current study, namely ethyl propanoate, ethyl isovalerate, ethyl-2-methylbutyrate, ethyl octanoate, and ethyl dodecanoate; methyl hexanoate, isoamyl alcohol, 2-hexenal, linalool, isobutyric acid, benzaldehyde, butyric acid, isovaleric acid, and phenethyl alcohol (Ubeda et al. 2012). Previous sensory analysis of vinegars has taken a number of diferent approaches. Gas Chromatography-Olfactometry (GC-O) has been performed on several styles of vinegars, and identifed potent aroma compounds and their individual aroma qualities (Ubeda et al. 2012,

Callejon et al. 2008). In one study of sherry vinegar, diference testing was able to distinguish the vinegars by their aging time in oak-barrels, from 0, 6, 12, 18, and 24 months (Tesfaye et al.

155 2000). Descriptive analysis on the samples used seven terms to describe the samples (aroma intensity, richness in aroma, ethyl acetate, woody odour, wine character, pungent sensation, and general impression). While “Richness” and “general impression” are hard to generalize, ethyl acetate, wine character, and pungent sensation were also important sensory characteristics (as the terms chemical, wine, and acetic acid) for the samples in the current study. A descriptive analysis on Aceto Balsamico Tradizionale (traditional balsamic vinegar), made from the acetifcation of heat-reduced grape must and aged for a minimum of twelve years, used twenty sensory terms to describe the samples. Many of these were fruit- or cooking-related terms (e.g. cooked apple, tamarind, caramel, cherry jam) and almost all were signifcantly diferent among the samples, refecting the use of both substrate- and processing-related terms in our work (Zeppa et al. 2013). In another study, sherry, red wine, white wine, apple, spirit, honey, and balsamic vinegar were used to develop a thirteen-term list of attributes for wine vinegar, which consisted of ethyl acetate, alcohol, pungent sensation, medicinal, winy character, raisin, woody, citrus, apple, coconut, red fruit, vanilla, clove, sweet aroma, rancid, bitter almond, leather, bacteria, cheese, and sawdust. Tis shows some similarity with the terms generated by the panel in the current study, especially red berry, blue cheese, citrus, chemical, and wine. While the use of diferent plant varieties in the current study will change the sensory and chemical profle of the samples, compared to vinegars made from grapes and wine, this suggests that a number of alcohols, esters, aldehydes and organic acids are common to many acetic fermentations, as are fruity, winey, chemical, and cheese-like sensory qualities. US and EU standards of identity for commercial vinegars require acetic acid concentrations of 4% or higher (Solieri et al. 2009); while this would disqualify some of the samples in this study from that designation, their culinary value, and the value of acetifcation as a tool in the restaurant kitchen-should not be overlooked. Elucidating the favor chemistry of products and processes such as these, which arise from research and experimentation that emphasizes a strong creative component in pursuit of capturing or creating favors can inform choices made in the R&D process, and bridge the gap between kitchen-developed intuitive

156 knowledge of favor and an instrumental-driven one. Vinegars produced in-house, using this process and with novel ingredients, are in use at a number of restaurants, and explicating the process and its favor chemistry can help devise high-quality uses for by-products, such as favorful components of vegetables that are lef over from their primary use. While purging and oxidation of volatiles through bubbling large amounts of air through the substrate likely has an afect on sensory qualities and volatile composition on top of the microbially-introduced changes, the data we collected suggests that some sensory and volatile characteristics of the substrate are preserved by the process. In some cases the acetifcation process appears to bring out diferent favors, such as cooked celery qualities or blue cheese aromas that might not be expected from the raw ingredients. Tere has recently been an increased interest in exploring the application of science and technology methods and practices to research and development in cuisine. New ingredients-such as pripicoa in Amazonian haute cuisine (Atala 2012), seaweeds for Nordic cuisine (Mouritsen et al. 2012), and culinary uses for foraged sea-fennel (Renna and Gonnella 2012)-have been communicated in peer-reviewed literature. In the case of Nordic seaweeds, dashi (a Japanese-style broth traditionally made from

Saccharina japonica) and other recipes incorporating wild and foraged seaweeds were developed, and free amino acids were profled, giving a fuller picture of the umami favor imparted by each species tested. Other research has focused on more traditional processes, such as that for producing Katsuobushi, a smoked, dried, Aspergillus sp-molded tuna loin product used for seasoning and soup-making in Japanese gastronomy, and applying its production process to novel substrates such as beef and pork (Felder et al. 2012). Tis study included a genomic profle of the microorganisms that had colonized the meat bushi as part of the production process. While the product under study had undergone microbial and chemical transformation leading to new and altered favors, further sensory and chemical investigations of the product (such as those performed in the current work) could help to elucidate these favors and how they were produced.

157 Some studies on culinary R&D include an informal sensory component, ofen in the form of tasting notes (such as in Perry’s experimentation on dry-aging beef and Aduriz et al.’s development of clay-coated potatoes) (Perry 2012, Aduriz et al. 2012). Garcia-Segovia et al. (2012) used paired comparison tests with 83 consumers as a measure of liking in terms of color, aroma, texture, taste, and appearance in the development of an apple dessert (Garcia-Segovia et al. 2012). Sous-vide, cook-vide, and moist heat were compared in tandem with three diferent cooking times, and the most preferred method (sous-vide) and cooking time (120 min) were used to devise a “New Tarte Tatin.” In the current study, preference or consumer sensory aspects were not examined, with the focus placed instead on sensory profles, volatiles, and multivariate relationships between these components, an approach that has not previously been undertaken as part of the culinary development process. Te development of further uses for favor chemistry and sensory analysis in the culinary R&D process can be envisioned. While the approach described here clearly provided useful information in characterizing products and understanding efects of ingredients and processing on sensory and chemical profles, it became evident that a more streamlined sensory analysis procedure was desirable for further and more frequent use in the development process. Restaurant development kitchens are ofen small, fast-moving operations, working with relatively small-sized test batches of new products or dishes. Te development process itself, too, generally proceeds quickly, with empirical decisions made on-the-fy. Of course, these qualities are not unique to research and development occurring within restaurants; industrial food R&D and winemaking are two other areas that beneft from sensory profling but where doing full descriptive analysis may be disadvantageous. Less intensive sensory methods, in any of these situations, could aid the transfer of experimental knowledge to production techniques, and help maximize the ability to replicate empirical or accidental successes. Rapid sensory profling methods, which bypass an extensive training and analysis period and have less intensive sample requirements (since panelists receive samples of a product during one or more analysis sessions, rather than several analysis sessions and several training sessions), have great potential for the

158 R&D process. Some examples of these sensory methods, such as projective mapping and sorting, have shown good comparability to traditional descriptive analysis (Preston et al. 2008, Sanchez 2011, Collins 2012, Valentin et al. 2012) and the ability to use only free nonproprietary tools such as R to acquire/analyze data is also attractive for smaller operations. Further development of the vinegar process-- addressing some of the questions raised by the current work on substrate favor, yeast fermentation contribution to favor, and slower fermentation speed via passive aeration-- and incorporating rapid sensory profling is addressed in the next chapter.

References

Aceña, L., Vera, L., Guasch, J., Busto, O., & Mestres, M. (2011). Chemical characterization of commercial Sherry vinegar aroma by headspace solid-phase microextraction and gas chromatography-olfactometry. Journal of agricultural and food chemistry, 59(8), 4062–4070. doi:10.1021/jf104763u Aduriz, A., Vergara, J., Lasa, D., Oliva, O., & Perisé, R. (2012). Culinary trompe-l’oeil: A new concept in coating. International Journal of Gastronomy and Food Science, 1(1), 70–77. doi:10.1016/j.ijgfs.2011.11.008 Atala, A. (2013). D.O.M.: Rediscovering Brazilian Ingredients. London: Phaidon Press. Retrieved from http://books.google.com/books?id=vNlWmQEACAAJ Atala, Alex. (2012). A new ingredient: Te introduction of priprioca in gastronomy. International Journal of Gastronomy and Food Science, 1(1), 61–63. doi:10.1016/j.ijgfs.2011.11.001 Bokulich, N. A., Torngate, J. H., Richardson, P. M., & Mills, D. A. (2014). Microbial biogeography of wine grapes is conditioned by cultivar, vintage, and climate. Proceedings of the National Academy of Sciences of the United States of America, 111(1), E139–E148. doi:10.1073/pnas.1317377110 Callejón, R. M., Morales, M. L., Ferreira, A. C. S., & Troncoso, A. M. (2008). Defning the typical aroma of sherry vinegar: sensory and chemical approach. Journal of agricultural and food chemistry, 56(17), 8086–8095. doi:10.1021/jf800903n Charles, M., Martin, B., Ginies, C., Etievant, P., Coste, G., & Guichard, E. (2000). Potent aroma compounds of two red wine vinegars. Journal of agricultural and food chemistry, 48(1), 70– 77. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/10637054 Cho, K. M., Math, R. K., Islam, S. M. A., Lim, W. J., Hong, S. Y., Kim, J. M., … Yun, H. D. (2009). Novel multiplex PCR for the detection of lactic acid bacteria during kimchi fermentation. Molecular and cellular probes, 23(2), 90–94. doi:10.1016/j.mcp.2008.12.006

159 Connor, H., & Allgeier, R. (1977). Vinegar: Its history and development. In D. Perlman & A. Perlman (Eds.), Advances in Applied Microbiology (pp. 81–133). London: Academic Press. Retrieved from http://books.google.com/books?id=_jvQV-v2ij0C Felder, D., Burns, D., & Chang, D. (2012). Defning microbial terroir: Te use of native fungi for the study of traditional fermentative processes. International Journal of Gastronomy and Food Science, 1(1), 64–69. doi:10.1016/j.ijgfs.2011.11.003 Fleet, G. (2003). Yeast interactions and wine favour. International Journal of Food Microbiology, 86(1-2), 11–22. doi:10.1016/S0168-1605(03)00245-9 García-García, I., Santos-Dueñas, I. M., Jiménez-Ot, C., Jiménez-Hornero, J. E., & Bonilla- Venceslada, J. L. (2009). Vinegar Engineering. In Lisa Solieri & P. Giudici (Eds.), Vinegars of the World (pp. 97–120). Milan: Springer. García-Segovia, P., Barreto-Palacios, V., Iborra-Bernad, C., Andrés-Bello, A., González- Carrascosa, R., Bretón, J., & Martínez-Monzó, J. (2012). Improvement of a culinary recipe by applying sensory analysis: Design of the New Tarte Tatin. International Journal of Gastronomy and Food Science, 1(1), 54–60. doi:10.1016/j.ijgfs.2011.11.011 González, Á., & Vuyst, L. De. (2009). Vinegars from Tropical Africa. In Lisa Solieri & P. Giudici (Eds.), Vinegars of the World (pp. 209–221). Milan: Springer. Gullo, M., & Giudici, P. (2009). Acetic Acid Bacteria Taxonomy from Early Descriptions to Molecular Techniques. In Lisa Solieri & P. Giudici (Eds.), Vinegars of the World (pp. 41–60). Milan: Springer. Hansen, A. S. (2004). Sourdough Bread. In Y. Hui, L. Meunier-Goddik, A. S. Hansen, J. Josephson, W.-K. Nip, P. S. Stanfeld, & F. Toldra (Eds.), Handbook of Food and Beverage Fermentation Technology (pp. 729–756). Boca Raton, FL: CRC Press. Hjelmeland, A. K., King, E. S., Ebeler, S. E., & Heymann, H. (2012). Characterizing the Chemical and Sensory Profles of United States Cabernet Sauvignon Wines and Blends. American Journal of Enology and Viticulture, 64(2), 169–179. doi:10.5344/ajev.2012.12107 Igbinadolor, R. O. (2009). Other Tropical Fruit Vinegars. In Lisa Solieri & P. Giudici (Eds.), Vinegars of the World (pp. 261–271). Milan: Springer. Johnson, A. J., Hirson, G. D., & Ebeler, S. E. (2012). Perceptual Characterization and Analysis of Aroma Mixtures Using Gas Chromatography Recomposition-Olfactometry. (E. M. C. Skoulakis, Ed.)PLoS ONE, 7(8), e42693. doi:10.1371/journal.pone.0042693 Kaack, K., Christensen, L. P., Hughes, M., & Eder, R. (2006). Relationship between sensory quality and volatile compounds of elderfower (Sambucus nigra L.) extracts. European Food Research and Technology, 223(1), 57–70. doi:10.1007/s00217-005-0122-y Kalt, W., Forney, C. F., Martin, a, & Prior, R. L. (1999). Antioxidant capacity, vitamin C, phenolics, and anthocyanins afer fresh storage of small fruits. Journal of agricultural and food chemistry, 47(11), 4638–44. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/10552863 Katz, S. E., & Pollan, M. (2012). Te Art of Fermentation: An In-Depth Exploration of Essential Concepts and Processes from around the World. White River Junction, VT: Chelsea Green Publishing. Retrieved from http://books.google.com/books?id=-zmLa205d0QC

160 Liu, D., Zhu, Y., Beefink, R., Ooijkaas, L., Rinzema, A., Chen, J., & Tramper, J. (2004). Chinese Vinegar and its Solid-State Fermentation Process. Food Reviews International, 20(4), 407– 424. doi:10.1081/LFRI-200033460 Mazza, S., & Murooka, Y. (2009). Vinegars Trough the Ages. In Lisa Solieri & P. Giudici (Eds.), Vinegars of the World (pp. 18–39). Springer Milan. Mouritsen, O. G., Williams, L., Bjerregaard, R., & Duelund, L. (2012). Seaweeds for umami favour in the New Nordic Cuisine. Flavour, 1(1), 4. doi:10.1186/2044-7248-1-4 Nilsson, M. (2012). Fäviken. London: Phaidon Press. Retrieved from http://books.google.com/ books?id=JvKxuAAACAAJ Ou, A. S., & Chang, R. (2009). Taiwan Fruit Vinegar. In Lisa Solieri & P. Giudici (Eds.), Vinegars of the World (pp. 224–242). Milan: Springer. Perry, N. (2012). Dry aging beef. International Journal of Gastronomy and Food Science, 1(1), 78–80. doi:10.1016/j.ijgfs.2011.11.005 Preston, L. D., Block, D. E., Heymann, H., Soleas, G., Noble, A. C., & Ebeler, S. E. (2008). Defning Vegetal Aromas in Cabernet Sauvignon Using Sensory and Chemical Evaluations. American Journal of Enology and Viticulture, 59(2), 137–145. Reade, B. (2012). Vinegar: From the Orleans Method to Food Lab Experiments. Nordic Food Lab Blog. Retrieved March 11, 2013, from http://nordicfoodlab.org/blog/2012/03/vinegar-from- the-orleans-method-to-food-lab-experiments Redzepi, R. (2010). Noma: Time and Place in Nordic Cuisine. London: Phaidon Press. Retrieved from http://books.google.com/books?id=z3jyRQAACAAJ Redzepi, R. (2013). A Work in Progress. London: Phaidon Press. Retrieved from http://books. google.com/books?id=njlOnwEACAAJ Renna, M., & Gonnella, M. (2012). Te use of the sea fennel as a new spice-colorant in culinary preparations. International Journal of Gastronomy and Food Science, 1(2), 111–115. doi:10.1016/j.ijgfs.2013.06.004 Sanchez, J. V. S. G. (2011). Comparison of Descriptive Analysis and Projective Mapping Techniques in the Aroma Evaluation of the Distilled Spirits , Gin and Tequila. University of California, Davis. Smith, A., & Kraig, B. (2013). Te Oxford Encyclopedia of Food and Drink in America. New York: Oxford University Press. Retrieved from http://books.google.com/ books?id=DOJMAgAAQBAJ Solieri, L, & Giudici, P. (2008). Yeasts associated to Traditional Balsamic Vinegar: ecological and technological features. International journal of food microbiology, 125(1), 36–45. doi:10.1016/j.ijfoodmicro.2007.06.022 Solieri, Lisa, Giudici, P., & Overview, G. (2009). Vinegars of the World. Tanaka, Y., Watanabe, J., & Mogi, Y. (2012). Monitoring of the microbial communities involved in the soy sauce manufacturing process by PCR-denaturing gradient gel electrophoresis. Food microbiology, 31, 100–106. doi:10.1016/j.fm.2012.02.005 161 Tesfaye, W., Morales, M. L., Callejón, R. M., Cerezo, A. B., González, a. G., García-Parrilla, M. C., & Troncoso, a. M. (2010). Descriptive Sensory Analysis of Wine Vinegar: Tasting Procedure and Reliability of New Attributes. Journal of Sensory Studies, 25(2), 216–230. doi:10.1111/ j.1745-459X.2009.00253.x Tesfaye, W., Morales, M. L., Garcia-Parrilla, M. C., & Troncoso, a. M. (2009). Improvement of Wine Vinegar Elaboration and Quality Analysis: Instrumental and Human Sensory Evaluation. Food Reviews International, 25(2), 142–156. doi:10.1080/87559120802682748 Ubeda, C., Callejón, R. M., Troncoso, A. M., Moreno-rojas, J. M., Peña, F., & Morales, M. L. (2012). Characterization of odour active compounds in strawberry vinegars, (May), 313– 321. doi:10.1002/f.3103 Valentin, D., Chollet, S., Lelièvre, M., & Abdi, H. (2012). Quick and dirty but still pretty good: a review of new descriptive methods in food science. International Journal of Food Science & Technology, 47(8), 1563–1578. doi:10.1111/j.1365-2621.2012.03022.x Versari, A., Parpinello, G. P., Cattaneo, M., San, V., Bologna, G., Brecce, V., & Ancona, B. (1999). Leuconostoc oenos and malolactic fermentation in wine : a review, (May), 447–455. Wei, C.-L., Chao, S.-H., Tsai, W.-B., Lee, P.-S., Tsau, N.-H., Chen, J.-S., … Tsai, Y.-C. (2013). Analysis of bacterial diversity during the fermentation of inyu, a high-temperature fermented soy sauce, using nested PCR-denaturing gradient gel electrophoresis and the plate count method. Food microbiology, 33(2), 252–61. doi:10.1016/j.fm.2012.10.001 Williams, L. (2011). Serious Ferment. Nordic Food Lab Blog. Retrieved March 11, 2013, from http://nordicfoodlab.org/blog/2011/05/serious-ferment Williams, L. (2012). Q&A: Te Nordic food lab. Flavour, 1(1), 11. doi:10.1186/2044-7248-1-11 Zeppa, G., Zoccoli, M. G., Nasi, E., Masini, G., Meglioli, G., & Zappino, M. (2013). Descriptive sensory analysis of Aceto Balsamico Tradizionale di Modena DOP and Aceto Balsamico Tradizionale di Reggio Emilia DOP, (May). doi:10.1002/jsfa.6219

162 Supplementary Table S6.1 Peak area of volatiles in vinegar samples determined by GC-MS, normalized to 2-undecanone. ni= not identifed. nd=not detected.

163 Supplementary Table S6.1 (continued)

164 Chapter 7: Correlating Labeled Sorting Sensory Analysis and Volatile Analysis of Malt Vinegars with Novel Ingredients

Introduction: In this chapter, we continue the investigation of novel acetifed products that started with fruit and vegetable vinegars in chapter 6. For further sensory characterization work, it was desirable to use a technique that could be performed on-site at a culinary R&D lab without sensory facilities, would not take a great deal of time for training, and could be performed using relatively small amounts of sample. While the context for this experiment was a culinary research and development environment, these concerns could infuence choice of sensory methods in a variety of similar contexts, such as other early-stage research and production environments where multiple small batches or trial runs are produced quickly, and time is at a premium. Panelists for the analysis were to be recruited from the research and culinary staf at an independent culinary lab and afliated restaurant; in previous experiments using experts, other professionals, or people not available for long training and analysis periods as panelists, various rapid sensory techniques have been employed. Tese include fash profling, Napping or perceptual mapping, and sorting; rather than using term generation and shared vocabularies as in Descriptive Analysis, these rapid techniques ask panelists to assign products into groups, or arrange them in a space, based on their similarities and diferences (see Valentin et al. 2012). Napping has previously been used to characterize the sensory qualities of gins, tequilas (Sanchez 2011), and Chardonnay wines aged in diferent oak barrels (Collins 2012), and sorting to characterize the sensory qualities of beers (Lelievre et al. 2008, Chollet et al. 2011), olive oil (Santosa et al. 2010), Cabernet Sauvignon wines (Preston et al. 2008), and fragrance materials (Lawless et al. 1990). Rapid techniques have repeatedly been shown previously to produce similar confgurations or maps as descriptive analysis on the same samples when analyzed through multivariate statistical techniques (Dehlholm et al. 2012, Sanchez 2011, Collins 2012, Faye et al. 2004, Preston et al. 2008). 165 Te samples in the previous chapter were acetifed by forcing air through a sample mixture of substrate, alcohol, and raw vinegar to provide an inoculation of acetic bacteria. Tis mimics industrial processes for vinegar production, which use turbines or other methods for maximizing the surface area in contact with oxygen of the liquid to be acetifed (Garcia-garcia et al. 2009). Tis allowed for rapid development and testing of vinegars from nontraditional substrates such as celery juice and pine needle “tea”. In this experiment, barley malt was used as a substrate for vinegars that were acetifed through passive aeration, similar to the Orleans method of vinegar production (Webb 2000), and incorporated novel botanical ingredients. Some of the botanicals, such as pine needles and licorice root, were used in the previous study, and some, such as Aspergillus oryzae-inoculated, saccharifed, and roasted barley1 had not been previously tested in acetic fermentation, but had novel caramel, cofee, and chocolate-like favors (Evans 2013). Barley malt is made by sprouting, germinating, and kilning barley grains, a process which activates amylase enzymes. Te basis for brewing beer involves a mashing process where malt is steeped in water at a controlled temperature to create wort. During steeping, the amylases hydrolyze starch molecules into smaller saccharide molecules, some of which (especially maltose, a disaccharide of glucose) can be converted by yeast into alcohol (Gupta et al. 2010). Malt vinegars are traditional to many cultures which brew and drink beer, and they are an industrially produced food product in the UK, the US, and parts of Europe (Grierson 2009). Products of lipid oxidation, such as pentanal and hexanal; Strecker degradation products such as 2- and 3-methylbutanal; and Maillard reaction products such as furfural and furfural alcohol have all been found to contribute to the aroma of wort (De Schutter et al. 2008). While malt vinegar has not been widely analyzed for its volatile composition, ethanol, propanol, isobutanol, pentanol, isoamyl alcohol, isobutyl acetate, butyl acetate, pentyl acetate, acetic acid, propionic acid, butyric acid, acetoin, and acetaldehyde are produced during the alcoholic and acetic fermentations that transform wort into malt vinegar (Jones 1969). 1 Based on roasting barley koji or qu, uses for which in the modern restaurant were also outlined in Felder et al. 2012 and Evans 2013

166 In this experiment, a labeled sorting task with sixteen panelists is used to characterize the sensory qualities and similarities in a set of sixteen malt vinegars with diferent botanicals and amounts of roasted koji added. Te sorting data are then correlated using Multiple Factor Analysis (MFA) to the headspace volatiles in each sample in order to better understand relationships between composition and aroma attributes.

Materials and Methods: Vinegar Production Flavored malt vinegars were produced in two stages. In the frst, malt was extracted into water and fermented with yeast to provide the base beers (Table 7.1). In the second, botanicals (Table 7.1, “Additions”) were added to the beer along with unpasteurized apple vinegar, and the

Table 7.1: Samples used in the study, their base compositions, botanials added, and codes

167 mixture was stored at 30 °C and allowed to passively acetify for three months, with a cloth cover to allow for airfow. Tese two steps are described further below. a. Base Beer Fermentation: Tree sources of malt were used for the beer production: Maris Otter Pale Barley Malt (Maltbazaren, Rødovre, Denmark), Cara Rye Malt (Maltbazaren, Rødovre, Denmark), and roasted barley koji made from pearled barley (Hordeum vulgare). Barley koji was produced by soaking pearled barley in water overnight, then steaming for ninety minutes, cooling to 30°C, and inocu- lating with 0.1% by weight of Aspergillus oryzae spores (GEM cultures, Lakewood, Washington, USA). Te inoculated barley was poured to a depth of 4 cm into a metal tray, then incubated in a humidifed chamber held at 30°C covered with a damp kitchen towel. Te barley was stirred and turned by hand every 12 hours for 36 hours, at which point the individual grains had bound together with a white mycelium and become sweet-tasting. Te barley koji was separated into individual grains by hand, then roasted in an oven at 180°C for 45 minutes, stirring every 8 minutes, until dark brown. Beers were produced using a standard pale ale (“Pale Ale”) recipe that was modifed as specifed to produce fve diferent base beers (Table 7.1, “Mash Bill”). An unhopped ‘Pale Ale’ -style wort was produced from Maris Otter malt that was mashed with water at a 3:1 water:malt ratio (w/w) at 65 °C for one hour, then lautered by fltering of the resulting wort and passing it through the remaining bed of grain twice to extract as much sugar as possible. Te fltered mixture was then diluted with water to a specifc gravity of 1.06, measured with a hydrometer.. “Amber Ale” was produced using the same protocol with 80% pale ale malt and 20% cara rye malt in the mash bill, “Light Koji Ale” with 90% pale ale malt and 10% roasted koji, “Koji Ale” with 80% pale ale malt and 20% roasted koji, and “Dark Koji Ale” with 70% pale ale malt and 30% roasted koji. Burton Ale yeast (WLP023, White Labs San Diego, California, USA) was added to each dilute wort, at a pitching rate recommended by manufacturers. Beers were fermented at 25°C for 14 days.

168 b. Acetifcation and favoring: 20% by volume of unpasteurized apple vinegar (Meyer’s, Copenhagen, Denmark) was added to the beer, which was subsequently placed in plastic containers, covered with muslin cloths, and allowed to passively acetify for three months at 30°C, then bottled in amber glass bottles. At the time of acetifcation, botanicals (see “additions”, table 7.1) were added at 0.5% w/v to “Pale Ale” and ‘Koji ale”. Te botanicals added to “Pale Ale” included Summit hops, East Kent Goldings hops, Hallertauer hops, Cascade hops, and Saaz hops, for a total of fve samples with diferent hops (Table 1) (Maltbazaren, Rødovre, Denmark). Botanicals added to “Koji ale” were foraged Douglas fr (Pseudotsuga menziesii) needles, foraged pine (Picea abies) needles, juniper (Juniperus communis) berries, juniper wood, licorice root (Glycyrrhiza glabra), and kelp (Saccharina japonica), yielding six diferent samples (Table 7.1). Sensory Analysis: A labeled sorting task with untrained panelists, adapted from Chollet et al. 2011, Preston et al. 2008, and Lelievre et al. 2008 was used to analyze the favors of the resulting vinegars. 10 mL of each vinegar were poured into plastic, lidded cups and labeled with a randomly generated, 3-digit code. 16 volunteers (6 female, 10 male, ages 21-38), recruited from the cooks, waiters, and interns of restaurant Noma and the Nordic Food Lab, were used as panelists. Each panelist was instructed to taste and smell each vinegar, and sort all vinegars into groups (making at least 2, but no more than 15 of these groups) as they saw ft, based on their perception of the similarities and diferences in their sensory character. Panelists were also asked to provide descriptors for the favor of the vinegars as explanation for the groups chosen. Panelists were compensated for their time with cofee and cake. GC-MS: A 10 mL aliquot of each vinegar samples was pipetted into a 20 mL amber glass headspace vial (Supelco, Bellefonte, PA), along with 25 μg/L of 2-undecanone (Sigma-Aldrich, St Louis,

MO) as an internal standard, and capped with magnetic caps with a 2-mm thick PTFE-faced silicone septum (Supelco). Each vial was agitated at 500 rpm for 5 minutes and then a 2-cm

169 long, 50/30 μm-thick Polydimethylsiloxane/ Divinylbenzene/Carbowax-coated fber (Supelco) was inserted into the headspace and volatiles were extracted for 40 minutes at 25°C, afer which they were directly introduced into the inlet of the GC. A Gerstel MPS2 autosampler (Gerstel, Linthicum, Maryland) performed the extraction and the injection. Te fber was removed from the headspace of the vial and immediately introduced into the inlet of an Agilent model 6890 GC-single quadrupole-MS with a DB-Wax column (30 meters long, 0.25 mm ID, 0.25 μm flm thickness) (J&W Scientifc, Folsom, CA). Te inlet was held at 250°C with a 10:1 split. Te carrier gas was helium, at a constant fow rate of 1 mL/minute. Te starting oven temperature was 40°C, held for 3 minutes, followed by a 2°C/minute ramp until 180°C was reached, then the ramp was increased to 30°C/minute until 250°C was reached, and held for 3 minutes. Te total runtime was 47 minutes. Te MSD transfer line was held at 250°C. Te mass spectrometer had a 1.5-minute solvent delay and was run in scan mode with Electron Impact Ionization at 70eV, from m/z 40 to m/z 300 and a source temperature of 230°C . Te samples were analyzed in triplicate. Peak identifcations were made by matching background-subtracted average mass spectrum across half peak height to the NIST 05 mass spectral database, followed by verifcation using Kovats retention indices calculated from C8-C20 alkane retention times (Sigma-Aldrich) and pure standards where available. Following identifcation, GC peaks were manually integrated and relative concentrations were calulated by normalizing to the 2-undecanone peak area.

Statistical Analysis: a. Sorting: Te panelists’ sorting data was compiled by assigning, for each panelist, a number to each group of vinegars, and then compiling this vinegar-by-group-number into a table where each panelists’ group assignments are represented in one column. Te ‘DistatisR’ package for the R statistical program was used to analyze the latent similarities in the samples based on this sorting data, which was converted into a “vinegars x groups” matrix. Each panelist’s set of vinegar groups was then converted into a distance matrix, resulting in a set of 16 distance matrices describing the

170 diferences in the samples. Tese were run through the distatis routine to generate a compromise matrix and plot of the samples, along with a bootstrapping procedure to produce confdence ellipses for their positions on the plot. Positions of each sample in the frst fve dimensions of the distatis model were extracted for further multivariate analysis b. Labels: Te descriptors used by panelists were compiled into a frequency table. Descriptors not used more than once for any sample were removed from the dataset, and a correspondence analysis was performed on the remaining frequency counts using the ‘ca’ package for the R statistical program. c. Volatiles: Peak areas normalized to internal standard peak area (relative quantifcations) were standardized by dividing by standard deviation. A principal component analysis was performed on these values using the R statistical package. d. Multiple Factor Analysis (MFA): A MFA was performed using the FactoMineR package for the R statistical program. Tree datasets– positions of each sample in the frst fve dimensions calculated by the Distatis analysis of the sorting data, peak areas of volatiles normalized to the internal standard, and the descriptor counts contingency table—were used as groups in the MFA, with the frst two sets treated as quantitative data and the third treated as frequency data.

Results and Discussion: Vinegars based on worts in the style of Pale Ale, Amber Ale, and Pale Ale including several diferent percentages of roasted barley koji were produced. Koji, the product of growing the mold Aspergillus oryzae on a grain or legume substrate, has been explored recently as a food ingredient,

171 Figure 7.1: Consensus plots from DISTATISDistatis-Bootstrap analysis of sorted groups from sensory analysis by 16 panelists on 16 vinegars, with bootstrapped confdence ellipses. 7.1A: Dimensions 1(X-axis) and 2(Y-axis) 7.1B: Dimensions 1(X-axis) and 3 (Y-axis)

Variance explained: Dimension 1 46%, Dimension 2 13%, Dimension 3 7%

dark.koji

koji.licoricekoji amber.cascade light.koji pale.cascadepale.summitpale pale.hallerpale.saazpale.ekg koji.douglas

koji.juniperkoji.kelp Distatis-Bootstrapkoji.pine koji.jwood Dimension 2: 13% variance explained 2: 13% variance Dimension

7.1A Dimension 1: 46% variance explained

koji.douglaskoji.pine light.kojiamber.cascadekoji.jwood pale.hallerpale koji.licoricekoji pale.saazpale.ekgpale.summit pale.cascade koji.kelp

dark.koji

koji.juniper Dimension 3: 7% variance explained 3: 7% variance Dimension

7.1B Dimension 2: 13% variance explained 172 in large part for its proteolytic functional properties, to produce traditional and nontraditional misos and in similar products where enzymatic proteolysis is desirable (Zhu and Tramper 2013, Felder at al 2012). In an in-house pilot study it was observed that the roasted koji had novel caramel, cofee, and chocolate-like favors (Evans 2013). We were interested in determining if the roasted koji could be further used to create novel malt vinegars. Terefore, a beer made with this koji was used as a substrate for several vinegars, keeping constant the ratios of beer to starter vinegar but varying the botanicals added. Botanicals were also incorporated into the malt vinegars to further enhance the complexity and potential for development of novel favor mixtures. Te Distatis compromise plot (Figure 7.1) shows that the panelists tended to aggregate samples into groups primarily based on the presence of grains in the beer base besides Maris Otter, For example the Pale Ale vinegars were more ofen grouped with each other on the lef side of the frst dimension, and were more separate from the other vinegars containing rye malt or roasted koji in their bases. Tis separation is primarily in the frst dimension of the plot, which explains 46% of the variance in the dataset. Along the second dimension of this plot, explaining 13% of the variance, the non-pale vinegars are separated by the amount of roasted barley koji they contain, and the botanicals used to favor them. Te dark koji vinegar, which contained the largest amount of roasted barley koji, was generally identifed as being most diferent from the other vinegars, i.e., by its position highest up the y-axis of this plot. From the large size of its identifying circle, though, we see that there was some disagreement about the diferentness of this sample among the panelists. Te koji-juniper wood vinegar defnes the other extreme of this axis. Te 95% confdence ellipses derived from bootstrapping show the overall overlaps and diferences among the samples—the pale ale vinegars overlap highly with each other, with an overlap between two of these pale ale vinegars and the ‘light koji’ vinegar (containing the least amount of roasted koji). Te ‘dark koji’ vinegar, plotted highest on the second axis, has a confdence ellipse that almost touches the ‘koji’ and ‘koji-licorice root’ ellipses, which overlap with the ‘koji-douglas fr’ and ‘amber’ vinegars, and very slightly with the ‘light koji’ vinegar. A lower-

173 Figure 7.2 Correspondence analysis biplot of descriptors used for malt vinegar samples. Dimension 1 Explains 30% of variance, dimension 2 explains 19% of variance. Descriptors are represented by red triangles and samples by blue squares.

juniper koji.jwood 1.5 wood pine 1.0

koji.junb

smoky 0.5 Dimension 2: 19% variance explained 2: 19% variance Dimension yeast honeypale.ale.ekgsour pale.ale.sum koji.douglas wine pale.ale koji.pineberry pale.ale.caspale.ale.halpale.ale.saaz bitter koji underripe.fruit

0.0 sweet floral oak apple malt fruit red.fruit strong mild earthy harsh soft koji.lickoji.kelp amber dark roastedlight -0.5 light.kojimolasses dark.koji -1.0

coffee

-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 Dimension 1: 30% variance explained right cluster of samples seem to be grouped together (beyond their shared use of a mid-level dose of roasted koji) by botanical ingredients- these are the ‘koji- juniper berry,’ ‘koji-kelp,’ ‘koji-pine,’ and ‘koji-juniper wood’ vinegars. Of this cluster, the juniper berry and juniper wood vinegars overlap with both the kelp and pine vinegars but not with each other. Te Douglas fr and pine vinegars show overlap with each other.

174 In the third dimension, which explains 7% of the variance, the separations noted above are compressed, except concerning the ‘koji-juniper berry’ vinegar. Tis sample, despite having a large confdence ellipse and variation in its compromise position (refecting disagreement among panelists as to which group it should be assigned) does not overlap with any other samples. Te correspondence analysis plot of the descriptors used to label the samples in the sorting exercise refects some similar trends (Figure 7.2). Te frst axis separates the ‘pale’ vinegars from the ‘koji’ vinegars, and the second dimension separates these koji-containing samples from each other based on botanicals added and amount of koji added. Besides positional information about similarities between each sample, the correspondence analysis adds a second level of information about the sensory drivers for these diferences. Te frst dimension, which explains 30% of the total variance, is dominated by wine, honey, sour, strong, and harsh descriptors on its lef side (Pale Ale), and descriptors that appear to relate to botanicals and additional malts (pine, roasted, etc) on its right end. Te second dimension explains 19% of the variance and, again, tends to separate some of the more botanical-related favors from the roasted-koji ones, with juniper and cofee at its extremes. In the same direction as cofee, but not as far along the axis, molasses, light, roasted, dark, sof, red fruit, malt, oak, and earthy are responsible for separating the samples, and are most associated spatially with the ‘light koji’, rye malt, ‘koji-licorice’, ‘koji-kelp’, and ‘dark koji’ vinegars, the latter of which plots at the most extreme end of this axis. Te samples separated from this group by the second axis—’koji’, ‘koji- douglas fr’, ‘koji-pine’, ‘koji-juniper berry’, and ‘koji-juniper wood’—are associated with the descriptors juniper, wood, pine, smoky, yeast, and berry, with juniper, wood, and pine contributing the most variance to this axis. Te ‘koji’ sample—the vinegar made with a mid-level dose of roasted koji, and both acetifed as-is and with added botanical ingredients—plots essentially right at the zero-point of this axis; that is to say, the point of lowest variance. Tis ‘koji’ sample has pronounced diferences in the dimension-1 sensory attributes that separate the pale ale vinegars from it and the other vinegars containing additional grains. However, within the latter group the ‘koji’ sample serves as a kind of sensory average or neutral point along the second axis of variation.

175 Table 7.2 Volatiles identifed in malt vinegar samples, their retention times (RT), C8-C20 Kovats retention index (CRI) and literature retention indices (RI), ni=not identifed, tentative identifcations included where possible with unidentifed peaks Literature RI # namea CAS RT CRI Pherobase Flavornet

MV1 dimethyl sulfide s 75-18-3 1.988 <800 716

MV2 methyl acetate s 79-20-9 2.459 827 856-864 MV3 2-methylfuran 534-22-5 2.874 872 877

MV4 ethyl acetate s 141-78-6 3.068 893 885-898 907

MV5 pentanal s 110-62-3 3.493 917 935 935

MV6 ethyl propanoate s 105-37-3 4.294 955 950 951

MV7 ethyl isobutyrate s 97-62-1 4.475 964 955

MV8 diacetyl s 431-03-8 4.873 983 970 MV9 Beta-terpinene 99-84-3 5.346 1003

MV10 Alpha-pinene s 80-56-8 5.814 1016 1027-1034 1032 MV11 Alpha-fenchene 471-84-1 6.994 1049 1054

MV12 camphene s 79-92-5 7.232 1055 1077 1075

MV13 ethyl isovalerate s 108-64-5 7.7 1068 1053-1082 1060 MV14 hexanal s 66-25-1 8.179 1082 1067-1093 1084

MV15 isobutanol s 78-83-1 8.84 1100 1099

MV16 isoamyl acetate s 123-92-2 9.875 1123 1118-1147 1117

MV17 3-carene s 13466-78-9 9.908 1124 1148 1148 MV18 ni.a 586-63-0 10.587 1140 1331 Alpha- MV19 phellandrene s 99-83-2 10.881 1146 1205

MV20 Beta-pinene s 127-91-3 11.357 1157 1113-1124 1116 MV21 isocineole 470-67-7 11.634 1163 1176 MV22 Alpha-terpinene 586-62-9 11.828 1168 1178 1178

MV23 limonene s 138-86-3 12.031 1172 1198-1234 1201

MV24 eucalyptol s 470-82-6 13.193 1199 1214 Beta- MV25 phellandrene s 555-10-2 13.251 1200 1209-1241

MV26 isoamyl alcohol s 123-51-3 13.749 1211 1205

MV27 ethyl hexanoate s 123-66-0 14.865 1236 1224-1270 1220 MV28 crithmene 99-85-4 15.117 1241 MV29 styrene 100-42-5 15.798 1256 1261 1241

MV30 cymene s 99-87-6 16.298 1267 1270 MV31 terpinolene 586-62-9 16.797 1278 1275-1297

MV32 acetoin s 513-86-0 17.115 1285 1272-1295 1287 Alpha-p- MV33 dimethylstyrene 1195-32-0 23.112 1421 1414

MV34 ethyl octanoate s 106-32-1 23.826 1438 1422-1446 1436

MV35 nonanal s 124-19-6 23.913 1440 1416 1385 MV36 ni.b 127-41-3 24.124 1445 1809 MV37 acetic acid 64-19-7 24.116 1445 1450

MV38 furfural s 98-01-1 25.13 1468 1455 Alpha- MV39 campholenal 4501-58-0 25.894 1487 MV40 acetylfuran 1192-62-7 26.73 1507 1490 a=Mass Spectrum Matched to NIST 05 Database >75%, s=mass spectrum and RT matched to standard

176 Table 7.2 (continued)

Literature RI # namea CAS RT CRI Pherobase Flavornet

MV41 benzaldehyde s 100-52-7 27.325 1521 1525 1495

MV42 ethyl nonanoate s 123-29-5 28.046 1539 1528

MV43 propionic acid s 79-09-4 28.428 1549 1523

MV44 linalool s 78-70-6 28.632 1554 1484-1570 1537 MV45 Alpha-cedrene 469-61-4 28.647 1554 1528

MV46 bornyl acetate s 5655-61-8 29.509 1576 1580

MV47 isobutyric acid s 79-31-2 29.607 1578 1563 MV48 fenchol 1632-73-1 29.868 1585 1574 camphene MV49 hydrate 465-31-6 30.24 1594 1517 MV50 4-terpineol 138-87-4 30.528 1601 1616

MV51 ethyl decanoate s 110-38-3 32.059 1641 1630 1655

MV52 isovaleric acid s 503-74-2 33.5 1679 1660-1691 1665 Ni- MV53 sesquiterpene-a 33.817 1687

MV54 Alpha-terpineol s 98-55-5 34.203 1697 1669-1720 1688

MV55 borneol s 464-45-9 34.259 1699 1642 1677 Ni- sesquiterpene- MV56 eudesmene? 34.607 1708 Ni- MV57 sesquiterpene-b 34.82 1714 Ni- sesquiterpene- MV58 muurolene? 34.987 1718 MV59 Delta-cadinene 483-76-1 36.216 1752 Ni- sesquiterpene- MV60 delta-selinene? 36.574 1762 ethyl MV61 phenacetate 101-97-3 37.517 1788 MV62 cuparene 16982-00-6 38.439 1809 1831 phenethyl MV63 acetate s 103-45-7 38.54 1811 1803 1829

MV64 hexanoic acid s 142-62-1 39.86 1835 1847-1872 phenethyl MV65 alcohol s 60-12-8 41.829 1872 1905-1940 1829 Gamma- MV66 nonalactone s 104-61-0 43.633 >2000 2042

MV67 octanoic acid s 124-07-2 43.975 >2000 2083 MV68 Alpha-cedrol 77-53-2 44.384 >2000 2100 MV69 widdrol 6892-80-4 44.62 >2000 Ni- sesquiterpenol- MV70 cadinol? 44.75 >2000

MV71 eugenol s 97-53-0 44.757 >2000 2141-2192 2141 Ni- sesquiterpenol- MV72 muurolol? 44.851 >2000 MV73 Delta-cadinol 36564-42-8 44.922 >2000 2167 MV74 decanoic acid 334-48-5 45.333 >2000 2361 a=Mass Spectrum Matched to NIST 05 Database >75%, s=mass spectrum and RT matched to standard 177 Figure 7.3 Principal Component Analysis (PCA) score plot of malt vinegars based on volatile composition. PC 1 explains 45% of variance, PC2 explains 21% of variance, PC3 explains 8.5% of variance. 7.3A: PCs 2 and 3. 7.3B (inset): PCs 1 and 2, included to show overall distribution of scores along PC 1 dominated by one sample.

PCA of vinegar volatiles, PC 2 and PC3

dark koji 4

koji licorice koji kelp

koji pine koji

amber PC3: 8.5% variance explained PC3: 8.5% variance 0 koji juniper

lightpale koji saazpale pale ekg

PC3 pale hallertauer pale cascade

pale summit

koji juniperwood

0 4 6 8 PC 2: 21% variance explained A PCA of vinegar volatiles

koji juniperwood

Te analysis of headspace volatiles by SPME-GC- 6 koji kelp

MS found 75 compounds in the set of samples, with 22— 4

koji pine including acetic acid and other organic acids, diacetyl,

koji juniper acetoin, furfural, and some alcohols and esters—common to 0

koji koji licorice dark koji amber

all the vinegars (Table 7.2). Many of these, especially organic palepale ekg

pale saaz palepale hallertauer cascade pale summit light koji acids, alcohols, acetoin, and acetate esters, appear to be 0 B fermentation-derived compounds (Jones 1969), and furfural is a Maillard reaction product was previously identifed in beer wort (de Schutter et al. 2008). Eight samples had only these 22 compounds detectible in the headspace. Te ‘koji-juniper 178 berry’ vinegar had the most compounds in its headspace, with 54 including the 22 common to all the samples. Te additional volatiles in this sample included several terpenoids. A number of these, including alpha-pinene, alpha-fenchene, camphene, delta-3-carene, alpha-phellandrene, beta-pinene, alpha-terpinene, limonene, cymene, terpinolene, linalool, bornyl acetate, 4-terpineol, alpha-terpineol, and delta-cadinene were previously identifed in juniper berries using HS-SPME (Vichi et al. 2007). Te juniper wood vinegar also contained a number of terpenoids, but these generally difered from those in the juniper berry sample; these included more sesquiterpenoids, such as alpha-cedrene, cuparene, alpha-cedrol, and widdrol, which are commonly produced by many Juniperus species (Adams 1987). A principal component analysis performed on the normalized volatile data was dominated in the frst dimension, explaining 45% of the variance, not by the separation between the pale ale and koji-containing samples as had been previously seen in the sorting and labeling data, but rather by a separation between the koji-juniper berry vinegar and all the other samples, which had almost no separation by the frst PC (Figure 7.3B). Te analysis was plotted again to compare the second and third PCs (explaining 21% and 8.5% of the variance, respectively) and to examine variances in the data beyond the frst, single-sample-dominated one (Figure 7.3A). In this plot of volatiles and samples the koji-juniper berry vinegar appears close to the low-variance center of the plot (Figure 7.3). PC2, plotted in this way with PC3, has a similar separation of pale ale- from not-pale vinegars as in the frst dimension of variance in the Distatis and Correspondence Analyses. In this case though, the ‘light koji’ vinegar is more similar chemically to the pale ale vinegars than it is sensorially and plots very close to them along PC 2. At its other end, PC2 is dominated by the diferences between the koji-kelp, koji-juniper wood, and koji-douglas fr vinegars and the remaining samples. PC 3, which explains 8% of the variance in volatile chemistry, has similar extremes as the second dimension of the Correspondence Analysis, with the dark koji and koji-juniper wood vinegars at its opposite ends.

A multiple factor analysis (MFA) was performed for more direct comparison of the relative overlaps in the three datasets (Figure 7.4). Since the Distatis analysis translated

179 Figure 7.4 Multiple Factor Analysis (MFA) of Sorting, Label, and Volatile data for 16 malt vinegars. 7.4A Individual Factor Map shows compromise positions of samples in the consensus space with positional disagreements plotted by dataset. 7.4B Factor Map for the Contingency Table shows the positions of samples and labeled sensory descriptors in the consensus space.

Individual factor map

sorting volatiles descriptors 4

koji.juniper 2

pale.ekg pale.cascade pale.saaz pale.haller pale

Dim 2 (19.68%) pale.summit

0 light.koji koji.pine koji.kelp amber.cascade koji.jwood koji.licorice koji koji.douglas -2 dark.koji

-2 0 2 4 6

Dim 1 (23.34%) A Factor map for the contingency table(s) 4 descriptors

koji.juniper 3 2

pale.ekg pale.cascade pale.haller pale.saaz 1 pale pale.summit juniper wine honey strong sour berry harsh floral underripe.fruit applesweetfruit pine 0 mildbitterred.fruitwood

Dim 2 (19.68%) light.koji yeast smoky soft koji.pine molasses dark oakkoji.kelp amber.cascade maltroasted light

-1 coffee earthy koji.jwood koji.licorice koji koji.douglas -2 dark.koji -3

-2 0 2 4

Dim 1 (23.34%) B 180 Figure 7.4 cont’d. 7.4C Correlation circle shows correlations between the frst fve dimensions of the DISTATIS sorting analysis and the volatiles in the samples. 7.4D Groups representation shows agreement between datasets plotted as proximity. 7.4E Partial axes shows correlations of the highest dimensions of each dataset plotted on the consensus space. Correlation circle

sorting volatiles 1.0

acetic.acid crithmenecymeneterpinoleneX3.careneisoterpinoleneAlpha.fenchenecampheneAlpha.phellandreneBeta.pineneisocineoleAlpha.terpinenelimoneneBeta.phellandreneAlpha.p.dimethylstyreneAlpha.campholenalcamphene.hydrateX4.terpineolNi.sesquiterpene.aAlpha.terpineolborneolNi.sesquiterpene.eudesmeneNi.sesquiterpene.bNi.sesquiterpene.muurolenecadineneNi.sesquiterpene.delta.selineneNi.sesquiterpenol.cadinolNi.sesquiterpenol.muurolollinaloolbornyl.acetatefencholBeta.terpineneAlpha.pinene Alpha.iononefurfural decanoic.acid 0.5 benzaldehydeacetoin phenethyl.alcoholphenethyl.acetate dimethyl.sulfide Gamma.nonalactone Factor.5 diacetyl acetylfuran

0.0 nonanal octanoic.acid Factor.4 isobutanolethyl.propanoateAlpha.cedrolwiddrolDelta.cadinolcupareneAlpha.cedrene ethyl.isobutyrate Dim 2 (19.68%) eucalyptolhexanal methyl.acetate styrene eugenol ethyl.acetate propionic.acidisoamyl.acetateisoamyl.alcohol hexanoic.acidethyl.phenacetateethyl.nonanoateethyl.isovalerateethyl.decanoateethyl.octanoateethyl.hexanoateethanol -0.5 Factor.3 Factor.2pentanal isovaleric.acidX2.methylfuran isobutyric.acid Factor.1 -1.0

-1.0 -0.5 0.0 0.5 1.0

Dim 1 (23.34%) C Groups representation Partial axes

sorting 1.0 volatiles

sorting 1.0 descriptors

0.8 Dim1.sorting Dim1.volatiles

0.5 Dim3.volatiles

0.6 Dim4.volatiles Dim3.descriptors descriptors Dim5.sorting Dim3.sorting

volatiles 0.0 Dim4.descriptors Dim 2 (19.68%) 0.4 Dim 2 (19.68%) Dim5.descriptors Dim5.volatiles Dim2.descriptors Dim4.sorting Dim2.volatiles -0.5 Dim1.descriptors Dim2.sorting 0.2 -1.0 0.0

0.0 0.2 0.4 0.6 0.8 1.0 -1.0 -0.5 0.0 0.5 1.0

Dim 1 (23.34%) Dim 1 (23.34%) D E 181 similarities in groupings by panelist into a multidimensional consensus space, it was possible to extract the positions of each sample in the frst fve dimensions of this space. Te positions were then treated as quantitative (as opposed to categorical or frequency) variables and used as a measure of the samples’ sensory characteristics. Tese positions were compared to the normalized, standardized peak areas of volatiles from the GC-MS analysis, also specifed as quantitative variables; and to the frequency data from the contingency table of labels generated during the sorting task, which were also used for the correspondence analysis. Te MFA produced a consensus space for the three sets of data that explained 23.3% of the variance in its frst dimension, 19.7% of the variance in its second dimension, and 15.1% of the variance in its third dimension (Figure 7.4). An individual factor map (fgure 7.4A) with variations in the consensus positions of products as predicted by the individual datasets shows the familiar positioning of the pale ale vinegars separated in the frst dimension from the darker and koji-based vinegars, with the second dimension describing a separation from the negative to the positive ends of the axis of the dark koji to the koji with added botanicals (specifcally juniper), respectively. Te koji-juniper berry, koji-juniper wood, light koji, and pale ale-summit hops samples display the most positional disagreement. Tis is especially true for the koji-juniper berry, which is more similar to the other koji samples according to the descriptor data, and more dissimilar according to the sorting and volatiles data. Conversely, the sorting data for the koji- juniper wood vinegar mostly agrees with its consensus sorting position, but is more similar to the other koji samples according to the volatiles data, and less similar according to the descriptors used for it. A plot of the “groups representation” in the frst two dimensions of this shared space (fgure 7.4D), which describes relative diferences among the three datasets plotted as indivdual points, shows the volatiles and the descriptors plot more closely together than either does to sorting, which shares a similar position in the frst dimension but contributes more to the variation in the second dimension. Te triplot of partial axes (fgure 7.4E) compares the main dimensions of variation in the individual datasets in relation to each other and the consensus

182 space. Tis explains some of the sources of the variations in fgure 7.4D; the frst dimension of the volatiles is poorly correlated with both the frst dimensions of the sorting and descriptors, which are negatively correlated with each other. However, the frst dimension of the descriptors and second dimension of the volatiles, the frst dimension of the sorting and the third dimension of the volatiles, and the second dimensions of both the descriptors and the sorting are very well- correlated to each other (Figure 7.4E). Tis means that many of the latent sources of similarities and diferences are the same for the three datasets, but are emphasized in diferent orders of importance by the volatiles, descriptors, and sorting. In other words, shared sources of variation exist in all three datasets, but contribute diferent rankings of variance to each. Tese shared sources of variance, but with diferent emphases, are also evident from inspecting the volatiles PCA in comparison to the Distatis consensus map (Figure 7.1) and the correspondence analysis space (Figure 7.2)—the frst PC describes a marked separation between the koji-juniper berry vinegar and all the other samples that is not refected in the other datasets, while the second and third PCs model a space with separations that are more similar to the frst- and second- dimensional separations in the multivariate analyses of the other two datasets. Possibly the most useful aspect of the MFA is its ability to begin to explain some of the favor-chemical relations in these samples, which in more traditional chemical-sensory experiments would be examined by performing a Partial Least Squares regression (PLS) on the continuous variables from a descriptive analysis and volatiles dataset. Here the descriptors are frequency data derived from a contingency table, which a PLS is not appropriate to analyze, but a MFA is able to examine these diferent data types together. Te wine, honey, strong, foral, and sour favors (Figure 7.4B) that explain the most variance in the pale ale vinegars are related to higher amounts of (likely fermentation-related) acetic acid, decanoic acid, phenethyl acetate, phenethyl alcohol, and dimethyl sulfde, as well as furfural, alpha-ionone, and benzaldehyde (Figure 7.4C). Te yeast, molasses, and underripe fruit aromas that separate the light koji and amber ale vinegars in the frst dimension from the pale ale vinegars are correlated to nonanal and octanoic acid. Te cofee, smoky, wood, dark, oak, roasted, malt, light, and earthy favors are correlated to the

183 koji-based samples in the lower-lef quadrant and to a number of ethyl esters, isovaleric acid,

isobutyric acid, hexanoic acid, isobutanol, alpha-cedrene, widdrol2, cadinol, styrene, eugenol, and 2-methylfuran. Finally, the juniper, pine, and berry favors plotting in the upper-right quadrant of the MFA factor map correlate very strongly to the koji-juniper berry vinegar, moderately to gamma-nonalactone and acetylfuran, and strongly to a large number of terpenoids found only in this sample. Many of these—alpha-pinene, alpha-fenchene, camphene, limonene, beta- phellandrene, terpinolene, delta-3-carene, and linalool—have been previously identifed in juniper berries (Vichi et al. 2007). Interestingly, the hops added to some samples appear to have extracted poorly, as there are no detectible amounts of terpenes or other hop-related compounds in these samples, and no obviously “hoppy” sensory labels. Several limitations on performing favor research on acetic fermentations in pilot scale and research and development environments were identifed in previous experiments. For example trial batch sizes of products in the R&D operations that contributed to this study are ofen in the range of 1-2 kg, if not much less, with production of larger amounts impractical due to a combination of the expense of ingredients, the time and labor needed to process large amounts of these ingredients, and equipment capacity. Tis means that a full descriptive analysis panel, with multiple replicates and training sessions adding up to large sample demand may tax the resources of the average research and development operation. As a result, interesting samples for which sensory or favor-chemical information is of interest may not be available in large enough quantities to realistically use for a Descriptive Analysis (DA). In addition, the time needed for training and utilizing a panel— the time per session, number of sessions, and number of days or weeks that the panel takes to perform—may be unrealistic in the restaurant R&D kitchen or other resource- and labor-intensive creative and production environments. Conversely, these operations will ofen employ a large number of personnel who are curious, highly motivated potential panelists, who have had a signifcant amount of on-the-job palate training, but who are only available briefy or sporadically and therefore ill-suited for formal DA.

2 Widdrol is a sesquiterpene alcohol found in the wood of several species of the Juniperus genus, see Kwon et al. 2010 184 Tese factors all point to rapid profling methods as a potentially useful tool, with acknowledgment of their drawbacks. In the case of this experiment, potential panelists with previous exposure to perceptual mapping (or ‘napping’) expressed dismay at the conceptual difculty of having had to defne a space describing the diferences between and among samples, which had taken them up to two hours to perform in a previous experiment unrelated to this study, and were more willing to volunteer for the sorting when told it would likely be faster to perform. Both perceptual mapping and sorting produce similar information about overall sources of sensory variations in sample sets compared to each other and compared to descriptive analysis. However, they are not as precise at identifying smaller diferences among samples that more time- and training- intensive methods are able to distinguish (Valentin et al. 2012, Chollet et al. 2011, Preston et al. 2008). While the use of sorting alone in this study, rather than a comparison of sorting and descriptive analysis, was motivated by multiple previous studies showing good agreement between sorting and other rapid methods and descriptive analysis (Dehlholm et al. 2012, Sanchez 2011, Collins 2012, Faye et al. 2004, Preston et al. 2008), more of these validating studies are needed to fully evaluate the application of this methods to a wider variety of foods and consumer products. In this case, the conceptual process of separating samples could have been driven by larger sensory diferences in samples (especially between botanical ingredients), and fewer diferences or poorer separations might be found in a set where samples are more similar to each other. In the present research, the sensory analysis was performed in one day, onsite in an R&D kitchen workspace, with no panelist taking more than forty minutes to complete the task. Multivariate analyses were able to correlate this sensory data with a GC-MS derived volatile dataset. As far as the author has been able to ascertain, this is the frst application of Multiple Factor Analysis to correlate GC-MS profles with rapid sensory analysis results. Diferences in the resulting MFA consensus space, and diferences evident upon visual inspection of the datasets, were generally dominated by one or two samples. Both datasets were typifed by an underlying structure of diferences between malt vinegar samples based on pale ale malt from those

185 containing a darker rye malt or percentages of roasted koji, as well as diferences between those samples containing roasted koji and various botanical ingredients.

References

Abdi, H., Valentin, D., Chollet, S., & Chrea, C. (2007). Analyzing assessors and products in sorting tasks: DISTATIS, theory and applications. Food Quality and Preference, 18(4), 627– 640. doi:10.1016/j.foodqual.2006.09.003 Adams, R. P. (1987). Investigation of Juniperus Species of the United States for New Sources of Cedarwood Oil. Economic Botany, 41(1), 48–54. Chollet, S., Lelièvre, M., Abdi, H., & Valentin, D. (2011). Sort and beer: Everything you wanted to know about the sorting task but did not dare to ask. Food Quality and Preference, 22(6), 507–520. doi:10.1016/j.foodqual.2011.02.004 De Schutter, D. P., Saison, D., Delvaux, F., Derdelinckx, G., Rock, J.-M., Neven, H., & Delvaux, F. R. (2008). Optimisation of wort volatile analysis by headspace solid-phase microextraction in combination with gas chromatography and mass spectrometry. Journal of chromatography A, 1179, 75–80. doi:10.1016/j.chroma.2007.11.103 Dehlholm, C., Brockhof, P. B., Meinert, L., Aaslyng, M. D., & Bredie, W. L. P. (2012). Rapid descriptive sensory methods – Comparison of Free Multiple Sorting, Partial Napping, Napping, Flash Profling and Conventional Profling. Food Quality and Preference, 26(2), 267–277. doi:10.1016/j.foodqual.2012.02.012 Faye, P., Brémaud, D., Durand Daubin, M., Courcoux, P., Giboreau, A., & Nicod, H. (2004). Perceptive free sorting and verbalization tasks with naive subjects: an alternative to descriptive mappings. Food Quality and Preference, 15(7-8), 781–791. doi:10.1016/j. foodqual.2004.04.009 Felder, D., Burns, D., & Chang, D. (2012). Defning microbial terroir: Te use of native fungi for the study of traditional fermentative processes. International Journal of Gastronomy and Food Science, 1(1), 64–69. doi:10.1016/j.ijgfs.2011.11.003 Grierson, B. (2009). Malt and Distilled Malt Vinegar. In L. Solieri & P. Giudici (Eds.), Vinegars of the World (pp. 135–143). Milan: Springer. Gupta, M., Abu-ghannam, N., & Gallaghar, E. (2010). Barley for Brewing : Characteristic Changes during Malting, Brewing and Applications of its By-Products. Comprehensive Reviews in Food Science and Food Safety, 9, 318–328. Kwon, H.-J., Lee, E.-W., Hong, Y.-K., Yun, H.-J., & Kim, B.-W. (2010). Widdrol from Juniperus chinensis induces apoptosis in human colon adenocarcinoma HT29 Cells. Biotechnology and Bioprocess Engineering, 15(1), 167–172. doi:10.1007/s12257-009-0154-4 Lawless, H. T., & Clatter, S. (1990). Consistency of multidimensional scaling models derived from odor sorting. Journal of Sensory Studies, 5, 217–230.

186 Lelièvre, M., Chollet, S., Abdi, H., & Valentin, D. (2008). What is the validity of the sorting task for describing beers? A study using trained and untrained assessors. Food Quality and Preference, 19(8), 697–703. doi:10.1016/j.foodqual.2008.05.001 Preston, L. D., Block, D. E., Heymann, H., Soleas, G., Noble, A. C., & Ebeler, S. E. (2008). Defning Vegetal Aromas in Cabernet Sauvignon Using Sensory and Chemical Evaluations. American Journal of Enology and Viticulture, 59(2), 137–145. Sanchez, J. V. S. G. (2011). Comparison of Descriptive Analysis and Projective Mapping Techniques in the Aroma Evaluation of the Distilled Spirits , Gin and Tequila. University of California, Davis. Santosa, M., Abdi, H., & Guinard, J.-X. (2010). A modifed sorting task to investigate consumer perceptions of extra virgin olive oils. Food Quality and Preference, 21(7), 881–892. doi:10.1016/j.foodqual.2010.05.011 Valentin, D., Chollet, S., Lelièvre, M., & Abdi, H. (2012). Quick and dirty but still pretty good: a review of new descriptive methods in food science. International Journal of Food Science & Technology, 47(8), 1563–1578. doi:10.1111/j.1365-2621.2012.03022.x Vichi, S., Riu-Aumatell, M., Mora-Pons, M., Guadayol, J. M., Buxaderas, S., & López-Tamames, E. (2007). HS-SPME coupled to GC/MS for quality control of Juniperus communis L. berries used for gin aromatization. Food Chemistry, 105(4), 1748–1754. doi:10.1016/j. foodchem.2007.03.026 Webb, A. D. (2000). Vinegar. In Kirk-Othmer Encyclopedia of Chemical Technology. John Wiley & Sons, Inc. doi:10.1002/0471238961.2209140523050202.a01.pub2 Zhu, Y., & Tramper, J. (2013). Koji--where East meets West in fermentation. Biotechnology advances, 31(8), 1448–1457. doi:10.1016/j.biotechadv.2013.07.001

187

ni= not not ni=

g/L 2-undecanone equivalents. equivalents. g/L 2-undecanone

cation of volatiles in malt vinegar samples, in in samples, vinegar malt in volatiles of cation

Relative quanti Relative

ed. nd=not detected. nd=not ed. f Supplementary Table S7.1 Table Supplementary identi 188

/L 2-undecanone equivalents /L 2-undecanone

cation of volatiles in malt vinegar samples, in in samples, vinegar malt in volatiles of cation

continued. Relative quanti Relative continued.

Supplementary Table S7.1 Table Supplementary 189

/L 2-undecanone equivalents /L 2-undecanone

cation of volatiles in malt vinegar samples, in in samples, vinegar malt in volatiles of cation

continued. Relative quanti Relative continued.

Supplementary Table S7.1 Table Supplementary 190 Conclusions

Flavor chemistry as a practice has matured since the mid-20th century, producing a body of data that delves deeply into aroma qualities and detection thresholds for a huge variety of volatiles, and that covers the composition and perceived aroma of a wide variety of samples of plants, beverages, and other processed and unprocessed food products. Volatile profling by GC- MS, calculation of putative Odor Activity Values (OAV), GC-Olfactometry, model reconstitution and omission experiments, Sensory Descriptive Analysis (DA), and multivariate statistical techniques such as Principal Component Analysis (PCA) and Partial Least Squares regression (PLS) have enabled and directed the production of this body of knowledge. Arguments for a scientifc study of the culinary aspects of food date back to at least the late eighteenth century. Gastronomy has been enthusiastically and encyclopedically analyzed and compiled according to a humanistic or social-science perspective, including the writings of Apicius and Brillat-Savarin and modern compendiums such as the Oxford Companion to Food. Te existing scientifc knowledge of food as it can be applied to cuisine received a similar treatment in Harold McGee’s seminal 1984 work On Food and Cooking, updated and revised in 2004, but a concerted cuisine-focused experimental discipline in the sciences has developed only sporadically. Demand for this discipline, and specifcally an aspect focusing rigorously on favor chemistry, is evident from several corners, however, from the growth of empirical approaches to restaurant culinary development; interest in the “science of gastronomy” broadly and on favor more specifcally, including work on favor from those outside the discipline; and a growing body of published work, some related to culinary development, some characterizing explicitly culinary processes with favor chemistry. Tis dissertation looked to existing methods for analyzing favor and chemistry, and has sought to apply them to novel and overlooked systems and adapt them to several conceptual and experimental challenges. Besides the aforementioned interest in gastronomic subject matter, this has also included a focus on methodology. Recent fndings on the relationship between perceived

191 aroma and composition in multicomponent mixtures argue for a sea change in the way these relationships are evaluated in real-life samples. Tis would include a focus on accounting for and evaluating mixture-dependent perceptual efects, and development and use of techniques to make these experiments facile to perform, even given the limitations of instrumental limits of detection, matrix efects, and other issues. In chapter two, In-instrument Gas Chromatography-Recomposition-Olfactometry (GRO) was developed to explicitly address both of these issues. Traditionally, reconstitution and omission experiments are performed on systems undergoing favor chemistry analysis, and these experiments involve quantitating profles of volatiles in a sample, estimating the sensory impact of each volatile compound in isolation, then building model mixtures—one to represent the sample, and several each omitting a potentially sensorially important volatile—for sensory evaluation, ofen by diference testing. GRO allows for the preparation of reconstitution mixtures inside a GC-MS with an olfactory port, cryotrap, and pneumatic fow switch to include or omit selected volatiles, directly from a SPME extraction of a sample’s headspace. Tis means that reconstitution and omission sensory experiments can be performed without quantitation, without chemical standards, and without calculating dilution factors or OAV. Tis approach was used to characterize sensory diferences of mixtures of lavender volatiles, fnding that “lavender-like” character was an emergent property of specifc mixtures. In chapter four, applying GRO to the aroma of Angostura bitters, a check-all-that-apply sensory procedure was able to show which sensory diferences ocurred upon specifc omissions, and to what magnitude; it also suggested that a low-abundance, low-OAV compound contributed most to the aroma of the sample of the three tested, which might have been excluded or overlooked had OAV been used as a reconstitution criterion. Volatile and sensory descriptive analysis profling were used in chapters three, fve, and six in conjunction with multivariate statistical analysis to determine correlations between headspace composition and sensory characteristics in samples of commercial bitters (chapter three), model Old-Fashioned cocktails (chapter fve), and acetic fermentations of novel substrates (chapter six).

192 Each of these subjects were chosen based on historical and current interest, and lack of scientifc attention and data. In chapter three, sixteen samples of aromatic cocktail bitters were found to difer mainly in citrus, celery, and spice characteristics; since bitters are typically made up of alcoholic extractions of several aromatic plant ingredients, each of which may share volatile compounds, many of these sensory diferences arise from volatiles from several of these ingredients. Te PLS analysis of this data showed that the volatile data correlated well overall to the general distributions of samples, but that the ultimate sensory impact of any given volatile when compared to its aroma in isolation was difcult to predict in some cases, especially for terpenoid compounds, but that many primary notes of aldehydes and phenylpropenoids were more directly related to their sensory correlations in the sample. Tis chapter also provided guidance for selecting specifc compounds to characterize directly in terms of their sensory roles using GRO in chapter four. In chapter fve, four of these bitters samples were evaluated in diferent whiskey matrices, modeling the Old-Fashioned, one of the oldest and most fundamental styles of cocktail. Tis chapter found that diferences in samples were generally driven by diferences in bitters more than diferences in whiskeys, though the presence of alcohol dampened the headspace concentrations of many volatiles. Te data in this chapter also suggested that the favor diferences in more premium bourbon and rye whiskeys remain emphasized upon mixing with bitters, more so than with more inexpensive bourbons and ryes. Chapters six and seven were developed based on an interest in the restaurant world on novel fermentation processes; these experiments dealt explicitly with acetic fermentation used to create vinegars with novel ingredients. In chapter six, sensory descriptive analysis and volatile profling were performed on vinegars made using an aerated, rapid acetifcation process on substrates such as celery juice, asparagus juice, pine needle tea, and elderfower wine. Tis chapter found that sensory relationships among the samples were determined partially by diferences in substrate favor, and partially to similarities induced by fermentation process, some of which had previously been characterized in more traditional vinegars. Te work in this chapter also

193 suggested the potential of more rapid sensory profling as a useful tool for work related to culinary R&D (and other similar) environments, where speed and economic use of resources are paramount. Tis led to the use of a labeled sorting sensory task and Multiple Factor Analysis (MFA) of this and volatile profling data on a relatively un-designed experiment on malt-based vinegars with novel favors such as juniper wood, roasted koji, and kelp in chapter seven. Tis chapter showed that volatile data and rapid sensory data could be correlated using DISTATIS and MFA to explain sensory diferences driven by winelike and sour, roasted, and juniper and pinelike qualities that were the main sources of separation in the sensory data. Te results of this dissertation suggests several areas for future work. First, the application of GRO and other sensory, instrumental, and statistical methods can be applied towards more specifc molecular- and descriptor- level characterizations of sensory interaction efects in aroma mixture perception. While we now know the aromas of many individual compounds, their putative importance in many samples, and correlations between their presence and several aroma qualities, a full understanding of the exact sensory roles in terms of additive contributions, masking, gestalt characteristics, and synergistic and emergent qualities of specifc compounds remains an elusive and at times dauntingly complex task. Facile methods for evaluating these efects may help lower the barrier to experiments designed to understand them. Second, favor chemistry as it relates to cuisine ofers an opportunity for truly interdisciplinary research, between chemists, sensory scientists, psychologists, neurobiologists, and the crafspeople working directly with favors—chefs, bartenders, and perfumers, integrating their skills and perspectives in a way that has not yet become typical. Developing approaches for data collection, experimental design, analysis, and aggregation using input from these disciplines could lead to avenues for research not yet theorized or identifed by academics currently working in these felds, and a new molecular of understanding of favor in terms of process and product that could enable faster, more sophisticated, or entirely new creative methods and approaches to culinary research and development.

194