DEPLOYING CONSUMER-DRIVEN STRATEGIES IN THE BREEDING OF

LEAFY BRASSICA OLERACEA L. GENOTYPES

A Dissertation

Presented to the Faculty of the Graduate School of Cornell University In Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy

by

Hannah Rae Swegarden August 2020

© 2020 Hannah Rae Swegarden

ii

DEPLOYING CONSUMER-DRIVEN STRATEGIES IN THE BREEDING OF LEAFY BRASSICA OLERACEA L. GENOTYPES

Hannah Rae Swegarden, Ph.D. Cornell University 2020

Strategically targeting diversity within Brassica oleracea L. could expand existing market classes and products for changing food systems. Consumer acceptance and sensory perceptions of leafy Brassica cultivars have received minimal attention in developing strategic breeding approaches, and few sensory or consumer resources exist to inform the development of new cultivars. We deployed a series of consumer methods to generate a multi-faceted resource for breeding leafy Brassica vegetables with improved quality and potential to innovate produce markets.

Qualitative Multivariate Analysis (QMA) identified underlying consumer values associated with kale consumption and highlighted consumer adherence to iconic kale and collard market classes. Quantitative approaches, including a trained descriptive panel and large consumer acceptance study, enabled the identification of

21 sensory attributes (primarily within the texture category) that varied among commercial cultivars and breeding materials. External preference mapping and agglomerative hierarchical clustering (AHC) identified four consumer clusters grouped closely with curly and Tuscan kale genotypes.

In-situ sensory testing of collard genotypes in Upstate New York and Western

Kenya underscored key considerations in conducting cross-cultural sensory studies.

Flash Profiling (FP) with untrained consumer panels elucidated descriptive attributes

iii

in the consumer lexicon, while the number of attributes generated by regional panels did not closely correlate with familiarity or ability to differentiate products. Regional collard preferences were identified by a significant (p < 0.05) country-by-cultivar interaction for consumer acceptance.

Eight morphologically distinct B. oleracea inbred lines were used as parents to develop a half-diallel mating design, which was subject to genotyping-by-sequencing

(GBS), preliminary nutritional analyses, and an online consumer acceptance survey.

After accounting for a strong correlation between liking and familiarity, a small heritable (h2 = 0.37) genetic component of consumer liking was identified. Genetic characterization of breeding materials confirmed their potential to add diversity to commercial market classes.

Partiality to familiar kale and collard genotypes was a common theme throughout these studies. Continued interdisciplinary work is required to understand barriers to mitigate familiarity and introduce successful cultivars. This research serves as an important resource to leafy Brassica breeding and provides several models to integrate consumer research into breeding programs working with diverse or new crop types.

iv

BIOGRAPHICAL SKETCH

Hannah Swegarden was raised in the Midwest and graduated in 2006 from

John Marshall High School in Rochester, Minnesota. She pursued a B.S. in Biology at the University of Wisconsin – Eau Claire and subsequently entered a Floral Design program upon graduation. In 2012, she continued a lifelong quest to blend science and artistic endeavors by pursuing a M.S. in Horticulture at the University of Minnesota –

Twin Cities with Dr. Thomas Michaels (Horticulture) and Dr. Craig Sheaffer

(Agronomy). Her work at the University of Minnesota focused on the breeding and evaluation of heirloom dry beans for organic farming operations in the Midwest.

Following graduation, Hannah continued to a Ph.D. program in the Horticulture

Section at Cornell University. She joined the vegetable breeding laboratory of Dr.

Phillip Griffiths (Cornell – AgriTech) in 2015. Her research in the Griffiths lab focused on breeding consumer quality traits in leafy Brassica vegetables and implementing methodology to integrate consumer feedback into the breeding program.

Working in a diversified vegetable breeding program opened opportunities to explore fresh market vegetable development, production, and markets and more generally, allowed her to think about how she can creatively contribute to evolving food systems.

In 2019, she married her husband, Roberto, and they eagerly welcome many more years of laughing and exploring together.

v

ACKNOWLEDGMENTS

Thank you to my committee members, Dr. Phillip D. Griffiths, Dr. Steve

Reiners, and Dr. Olga Padilla-Zakour for their continued patience and feedback throughout the implementation of this dissertation research. In providing me the freedom to think broadly, I’ve explored facets of interdisciplinary research I may not have otherwise.

Thank you to our collaborators at the Cornell Sensory Center, Dr. Robin

Dando and Alina Stelick, who provided me with the energy and foundational knowledge on which much of this dissertation was built. To the numerous individuals who sat with me and helped make sense of what I could not – Dr. Lynn Johnson, Dr.

Zachary Stansell, Dr. Nicholas Santantonio, Dr. Charles Wasonga, and Dr. Carl Sams

– thank you for your patience and creative interpretations. I am grateful to have worked with diverse partners throughout my degree, including

GrowNYC/Greenmarkets, Cornell Cooperative Extension, Advantage Crops Ltd., field staff at the Homer C. Thompson Vegetable Research Farm, and grower collaborators in the upstate NY region. These experiences broadened my capacity as a horticulturist and created for impactful memories.

Finally, I’d like to express my sincerest gratitude to my family and friends, namely Roberto, Mom, Dad, and Ben, for their continued support and encouragement.

You all have been my sounding board through thick and thin, but your unconditional support has given me the confidence to persevere.

vi

TABLE OF CONTENTS BIOGRAPHICAL SKETCH...... v ACKNOWLEDGMENTS...... vi LIST OF FIGURES ...... IX LIST OF TABLES ...... XII LIST OF ABBREVIATIONS ...... XIV CHAPTER 1. LITERATURE REVIEW ...... 1 DEFINING FRESH VEGETABLE QUALITY FOR CHANGING MARKETS AND CONSUMERS ... 1 PLANT BREEDING THROUGH THE LENS OF PRODUCT DEVELOPMENT ...... 5 CONSUMER-DEFINED QUALITY IN THE CONTEXT OF A PLANT BREEDING PROGRAM ..... 7 BOTANY AND BREEDING OF BRASSICA OLERACEA L. VEGETABLES ...... 13 Domestication and taxonomy ...... 13 Botany and mating systems...... 16 BREEDING FOR QUALITY TRAITS IN COLE CROPS ...... 18 Pigment profiles ...... 19 Flavor constituents...... 24 Leaf morphology ...... 27 TRANSLATING SENSORY SCIENCE TO A LEAFY BRASSICA BREEDING PROGRAM ...... 30 REFERENCES ...... 34 CHAPTER 2. BRIDGING SENSORY EVALUATION AND CONSUMER RESEARCH FOR STRATEGIC LEAFY BRASSICA (BRASSICA OLERACEA) IMPROVEMENT ...... 47 ABSTRACT ...... 47 INTRODUCTION...... 48 MATERIALS AND METHODS ...... 53 Qualitative Multivariate Analysis (QMA) ...... 56 Descriptive Analysis (DA) ...... 58 Consumer Central Location Test (CLT) ...... 60 Statistical procedures and multivariate analysis...... 64 RESULTS AND DISCUSSION ...... 66 HUT journal feedback and value diagramming ...... 66 Trained panel assessments ...... 70 Hedonic CLT and consumer clustering ...... 75 Perceptual mapping and external preference maps ...... 77 Integrative observations ...... 80 Application to future breeding efforts ...... 82 CONCLUSION ...... 84 REFERENCES ...... 86 CHAPTER 3. FLASH-PROFILING COLLARD (BRASSICA OLERACEA VAR. VIRIDIS) MARKETS TO IDENTIFY REGION-SPECIFIC PREFERENCES . 92 ABSTRACT ...... 92

vii

INTRODUCTION...... 93 MATERIALS AND METHODS ...... 99 Plant materials and production ...... 99 Human subject assurance ...... 102 Consumer acceptance testing ...... 102 Rapid sensory analysis: Flash Profiling (FP) ...... 103 Statistical procedures and multivariate analysis...... 105 RESULTS ...... 107 Regional consumer acceptance ...... 107 Lexicon development and variation by region ...... 110 Collard product differentiation ...... 112 DISCUSSION ...... 115 CONCLUSION ...... 123 REFERENCES ...... 125 CHAPTER 4. DEVELOPMENT OF A LEAFY BRASSICA DIALLEL TO ASSESS GENETIC ARCHITECTURE OF CONSUMER LIKING ...... 130 ABSTRACT ...... 130 INTRODUCTION...... 131 MATERIALS AND METHODS ...... 139 Plant materials and field phenotyping ...... 139 Genotyping-by-Sequencing (GBS) ...... 142 Online consumer acceptance survey...... 143 Statistical procedures and mixed model analysis ...... 146 RESULTS ...... 150 Genetic characterization of diallel materials ...... 150 Parental effects on overall liking of leafy genotypes ...... 151 Role of familiarity in assessment of liking ...... 155 Associated morphological and nutritional traits...... 157 DISCUSSION ...... 160 CONCLUSION ...... 168 REFERENCES ...... 169 CHAPTER 5. FINAL CONCLUSIONS AND FUTURE DIRECTIONS...... 177 KEY FINDINGS FROM SENSORY AND CONSUMER RESEARCH IN LEAFY BRASSICAS .... 177 PRAGMATISM AND CREATIVITY IN DESIGNING SENSORY AND CONSUMER RESEARCH METHODS FOR VEGETABLE BREEDING PROGRAMS ...... 183 BALANCING STAKEHOLDER EXPECTATIONS IN BREEDING FOR QUALITY ...... 185 REFERENCES ...... 188 APPENDIX: CHAPTER TWO ...... I APPENDIX: CHAPTER THREE ...... XVI APPENDIX: CHAPTER FOUR ...... XXII

viii

LIST OF FIGURES

Figure 2.1. Qualitative Multivariate Analysis (QMA) method process, as adapted from Lopetcharat and Beckley (2012). Six leafy Brassica samples representing the kale product space were subject to home-use tests (HUT) by n = 14 consumers. In- home journals, photos, and observations provided a foundation for discussion during a facilitated three hour focus group, which included lexicon development and perceptual mapping exercises...... 57

Figure 2.2. Kale sensory wheel depicting lexicon developed through Qualitative Multivariate Analysis (QMA) and descriptive analysis. Descriptors (n=77) in the outermost ring of the wheel may exist within the kale product space, although not all were vetted through descriptive panel evaluations. Those descriptors protruding (solid color) from the outermost ring were observed to be significantly (P _ 0.05) different between kale samples subject to descriptive evaluation and may serve as foundational descriptors in future kale sensory studies...... 61

Figure 2.3. Graphical representation of product separation in both qualitative and quantitative studies. Products closers together in the visual space provided are understood to be more similar to one another than those spaced further apart. (A) First two PCA dimensions (axes) explaining 61.24% of the variation among 10 kale samples (blue) subject to descriptive analysis. Estimated means of 15 sensory attributes (orange) within aroma, flavor, aftertaste, and texture modalities, all of which were significantly different among kale samples, were used to compile a correlation matrix. Attribute contributions to the first two PCA dimensions illustrate a slight separation of texture (mechanical)/flavor and textural (surface) attributes between Dimensions 1 and 2, respectively. (B) Qualitative compilation of group consensus perceptual mapping exercise from the QMA facilitated focus group. Separation of samples in the QMA product set is represented along axes of “Texture” and “Flavor Intensity,” both of which were identified as primary sensory distinguishers among samples. Orange ellipses depict range of sample sensory space identified between two consumer groups performing mapping exercises. (C) External consumer preference mapping obtained by PLS analysis of descriptive and AHC consumer hedonic data. Axes represent the first (F1) and second (F2) dimensions of PCA performed on descriptive data. Four consumer clusters (orange) were identified through AHC of consumer hedonic scores for “Overall Liking” of 10 kale samples (white). A contour plot (olive) has been overlaid to illustrate the percentage of cluster exhibiting an above-average preference for samples in a particular region of the sensory map...... 74

Figure 3.1. Demographic information for farmers, educators, and consumers who participated in consumer acceptance testing of n=6 collard genotypes in Western Kenya (May ‘19) and Upstate New York (Sept. ‘19). Demographics highlight

ix

major discrepancies in product familiarity between the two regions...... 108

Figure 3.2. Estimated mean liking (9pt hedonic scale) of each collard genotype subject to consumer acceptance testing in New York and Western Kenya (Homa Bay). ‘Top Bunch’ was only sampled in New York evaluations. Bars = SE...... 109

Figure 3.3. Cultivar coordinates for each Flash Profiling participant configuration (n = 12-15) in Rodi Kopany, Kenya (Homa Bay) and Geneva, NY (New York) (A, C) and New York, USA (B, D) after Generalized Procrustes Analysis (GPA) transformations and principal components analysis (PCA). Cultivar coordinates are summarized with 90% CI ellipses to highlight product separation. Participant- provided sensory attributes correlated (r > 0.6) with each dimension are noted along their respective axis for raw (A, B) and cooked (C, D) product preparations. External preference mapping was performed for cooked product preparation in Homa Bay (C) using PREFMAP in XLSTAT. A contour plot (gray) has been overlaid to illustrate the percentage of consumer clusters exhibiting an above-average preference for samples in a particular region of the sensory map. Dark gray indicates 80-100% of consumer clusters exhibiting above average preference in that region; additional preferences are shades in increments of 20%...... 113

Figure 4.1. Genomic descriptions of Brassica oleracea cultivars, wild types, and diallel breeding materials. A and B) Principal component analysis (PCA) of the SNP marker matrix (32,366 SNPs) were developed from all sequenced materials (A) and full diallel mating design (B). B) Parental inbred (colored) are illustrated in triplicate alongside reciprocal hybrids (gray). (C) Linkage disequilibrium (LD) decay plot across the genome. Marker correlations were placed in bins of 1000bp for enhanced resolution. D) Inbreeding coefficients of all sequenced materials, as calculated using the observed number of heterozygous markers in each genotype...... 145

Figure 4.2. Diallel mating design with n = 8 inbred leafy B. oleracea parents. Parents are pictured along the axes. Background of hybrid progeny are shaded in orange from least liked (5.25; light orange) to most liked (7.08; dark orange) on a 9pt. hedonic scale by participants in the online consumer survey. Gray background was not included in the survey due to low seed supply...... 152

Figure 4.3. Visual assessment of parental effects on overall consumer liking of genotypes (n = 35) in the online consumer survey. Genotypes are vertically ordered from least (bottom) to most (top) liked (raw mean data). A) Genotypes colored by the four most predominant colors (HEX) identified using k-means clustering (n = 5 clusters; background color removed). percent color composition of four k-means clustering (K = 5; removed black background cluster). B) Genotypes colored according to their parental cluster (K = 7) representation identified using FastStructure analysis. Elongated and green collard parental

x

types grouped to a single cluster...... 154

Figure 4.4. Broad (H2) and narrow (h2) sense heritability of consumer liking calculated using variance components from genotype-informed mixed models with and without the fixed effect of familiarity. In each model, heritability was calculated on an experimental (n = 113.2; green) and individual participant (n = 1; yellow) basis to illustrate the effect of sample size in heritability equations...... 155

Figure 4.5. Relationships between liking and familiarity of B. oleracea genotypes (n = 35; gray) included in the online consumer survey. A) Correlation between mean liking and familiarity among genotypes, as evaluated by a Spearman Rank correlation test. B) Predicted mean change in genotype BLUPs between models with and without the fixed effect of familiarity. Green line represents an x = y line with a slope of one and an intercept of zero; ranks changed are noted by genotypes above or below this line...... 156

Figure 4.6. Principal components analysis (PCA) of leaf morphological traits (A), glucosinolates (B), and carotenoids (C) among n = 36 inbred and hybrid lines developed from a half diallel B. oleracea mating design. Genotypes (green) are labeled with a two-digit identifier denoting the female (#) designators from each parental inbred it was derived. Factor loadings (yellow) indicate the influence of each variable (red) on the component...... 158

xi

LIST OF TABLES

Table 2.1. Kale germplasm included in Qualitative Multivariate Analysis (QMA), descriptive analysis (DA), and consumer central location test (CLT) studies. All genotypes represent breeding lines or cultivars within Brassica oleracea, with the exception of 'White Russian' (Brassica napus)...... 54

Table 2.2. Attribute lexicon developed by trained kale sensory panel (n=9) over 17, one-hour training sessions in accordance with descriptive analysis. Reference products and attribute definitions were adjusted after panel discussion and agreement...... 62

Table 2.3. Values associated with the purchasing and consumption of kale. Values were described by a focus group of consumers (n=14) after home-use testing (HUT) of six kale genotypes. Values are separated according to higher-level and sensory descriptors and listed in order of relative importance...... 69

Table 2.4. Estimated means and groupings of 44 sensory attributes in kale genotypes (n=15) evaluated by a trained descriptive panel (n=9). Attribute intensity was assessed on an unstructured numeric line scale (0 = “none” to 100 = “extremely”). ‘Consumer Liking’ was evaluated using a 9-pt. hedonic scale by n=90 consumers in a central location test (CLT). Tukey's HSD was performed on nested mixed model estimate means; means with the same grouping letter are not significantly different (P ≤ 0.05)...... 71

Table 3.1. Hybrid collard germplasm used in consumer acceptance and Flash Profiling evaluations in Western Kenya (May ‘19) and Upstate New York (Sept. ‘19). . 101

Table 3.2. Site information for acceptance evaluations of hybrid collard germplasm in Western Kenya (May ‘19) and Upstate New York (Sept. ‘19)...... 103

Table 3.3. Summary of Flash Profiling descriptive analysis with untrained panelists in Rodi Kopany, Kenya (Homa Bay County) and Geneva, NY (New York). Panelist descriptors were subject to GPA transformations and a PCA was produced to highlight product separation...... 104

Table 4.1. Inbred parental genotypes (Brassica oleracea) used in the development of a diallel mating design in early 2018. Each genotype was selfed to at least the S4 generation and evaluated for uniformity/stability. Inbreeding coefficients of each parent were calculated from SNP marker data of three biological replicates. ... 140

Table 4.2. Random effect variance components from participant liking models. Models were performed with (Liking ~ Familiarity) and without (Liking ~ 1) the fixed effect of familiarity. Random effect variances due to general combining

xii

2 2 2 ability (𝜎GCA ), specific combining ability (𝜎SCA ), participant (𝜎p ), and random 2 2 error (𝜎 ) are reported. Percentage of total random variance (% 𝜎total ) provides an indication of the effect’s contribution to the overall random effect variance...... 147

Table 4.3. Mean participant liking (1-9) and familiarity (1-5) of each hybrid genotype and parental inbred in the half-diallel design. Best unbiased linear predictors (BLUPs) for participant liking were calculated from liking models performed with (Liking ~ Familiarity) and without (Liking ~ 1) the fixed effect of familiarity...... 149

xiii

LIST OF ABBREVIATIONS

GPA Generalized Procrustes Analysis FP Flash Profiling CLT Central Location Test DA Descriptive Analysis QMA Qualitative Multivariate Analysis HUT Home-Use-Tests SNP Single Nucleotide Polymorphism PCA Principal Component Analysis ANOVA Analysis of Variance GCA General Combining Ability SCA Specific Combining Ability REML Restricted Maximum Likelihood RCBD Randomized Complete Block Design GLS Glucosinolates AHC Agglomerative Hierarchical Clustering

xiv

Chapter 1

LITERATURE REVIEW

Defining fresh vegetable quality for changing markets and consumers

Vegetables play a key role in healthy diets and comprise an important part of food and agricultural systems. Vegetables represent approximately 5% ($19.6 billion) of total agricultural farmgate sales in the U.S. and are essential to maintaining a nutritionally balanced diet (USDA-NASS, 2019). Consumption of vegetables in the

U.S. remains well below the daily recommended levels; only 9.3% of adults meet the recommended intake of vegetables (2-3 cups per day) (HHS & USDA, 2015). The vegetable sector has seen declines in shelf-stable and processed products, but sales of fresh vegetables in the U.S. have seen continued growth since 2014. The fresh vegetable segment of the vegetable market is expected to surpass $41 billion in total sales by 2024 (Roberts, 2019).

Horticultural crops, including fruits, vegetables, mushrooms, and other high- value plant products, are often perishable in nature and require transportation from the site of production to markets/end-users (Acquaah, 2009). Horticultural plant products feed into fresh produce markets, minimally processed food items (e.g. bagged salads and frozen foods), and shelf-stable processed products (e.g. canned goods, pastas, chips, etc.). Crops are harvested at their peak ripeness or at an appropriate developmental stage so when a product reaches the end-user it is of acceptable quality for fresh eating or processing. Products can see rapid deterioration in quality without

1

adequate post-harvest management and transportation to market (Kader, 2002).

Freshness, price, and flavor are now the most important consumer-defined quality factors when purchasing fresh vegetables in the United States (Roberts, 2019).

Diverse consumer diets and expectations for quality are a challenge to horticultural supply chain complexity and resilience.

Understood definitions of quality vary among disciplines and are often specific to users along the supply chain; there is no perfect definition of quality because quality is relative (Jongen, 2000). Quality control is further complicated by variation in environment, region, and production methods. Fresh market products often undergo minimal processing or alteration, but end-users and stakeholders along the supply chain maintain very different expectations of quality. Product quality is most commonly defined by extrinsic and intrinsic parameters (Jongen, 2000; Steenkamp,

1990). Extrinsic product parameters refer to quality cues that are not innately part of a product but affect the perception of quality. Production methods, brand name, packaging, and product guarantees and fall within the category of extrinsic quality parameters. In contrast, intrinsic quality refers to traits which remain intact and unaltered in the physical product. Collective sensory characteristics, including appearance, flavor (taste and aroma), texture, and nutrition, qualify as intrinsic product traits (Barrett et al., 2010). These intrinsic qualities may not always be visible (e.g. nutrition, flavor, texture), but alteration of these products changes the inherent nature of the product.

Vegetable crops are most commonly defined by intrinsic product properties for players in current supply chains, as these quality parameters are objective and

2

measurable during pre- and post-harvest quality evaluation. As the capacity to transport and distribute fresh produce increased in the mid 1900’s, the fresh produce industry saw a marked increase in the breeding and development of intrinsic supply- chain traits, such as uniformity of shape, size, firmness, and lack of perishability. A notable dichotomy in fresh market produce now exists between breeding for

“supermarket” or retail quality, where a small number of plant phenotypes are appropriate for large distribution and retail markets, and developing “niche” markets, where non-standard, unique plant phenotypes cater to smaller, often higher-value direct-to-consumer markets.

This dichotomy may also be viewed as a production-oriented vs. consumer- oriented disparity, exemplified by the commercial tomato industry. Recent breeding for shelf-life traits and selection for genes implicated in slow ripening resulted in increased production intensity, higher yields, and reduced waste across the supply chain, but it also reduced the consumer quality experience (Folta and Klee, 2016;

Lukyanenko, 1991). The revival of heirloom tomato cultivars among niche and culinary markets has expanded the diversity of new market classes of tomatoes and prompted researchers to re-evaluate the potential of these types in modern plant breeding programs (Rodríguez-Burruezo et al., 2005; Tieman et al., 2017). Similar approaches to introgress flavor and consumer-quality traits have since been adopted in the breeding of other vegetable crops such as melon (The Packer, 2013), Brussels sprouts (Doorn, 1999), and lettuce (Chadwick et al., 2016). Coupled with advancements in supply chain management logistics, these breeding endeavors have opened new opportunities to supply consumers with fresher, higher-quality perishable

3

produce items without the high levels of waste previously associated with these products.

These trends in breeding priorities are emblematic and predictive of a larger shift away from a producer-oriented model of developing new crop cultivars to a consumer-oriented model (Folta and Klee, 2016). In a consumer-oriented model, quality traits are judged relative to the impact on consumer purchase intent, represented by both the perceived product quality (the sum of both extrinsic and intrinsic parameters) and acceptance (Jongen, 2000). Ultimately, the viability of a product depends on translating consumer expectations for quality to underlying product characteristics. Understanding changing supply chain, market, and consumer- constructed definitions of quality is critical to addressing systemic and generational changes occurring in the fresh produce sector. A sustained interest in health and wellness, demand for plant-based product alternatives, reduction of artificial additives, and expectations for both quality and convenience have altered the interface between consumers and the fresh-market sector (Roberts, 2019). Connectivity with local food systems has also brought demand for local products to supermarkets (Low et al., 2015) and led to a boon in direct-to-consumer, including community supported agriculture

(CSA) and farmers’ markets (USDA-NASS, 2015). The local food movement has reached urban consumers, leading to an increase in urban farming (Eigenbrod and

Gruda, 2015), meal-delivery services (AmazonFresh®, Blue Apron ®, HelloFresh ®), door-step delivery (Doordash®, Grubhub®, etc.), and ventures into vertical agriculture (Despommier, 2013). The demand for “all-season” produce in these local food systems has spawned the expansion of on-farm season extension practices,

4

including high-tunnels (Janke et al., 2017) and controlled environment agriculture

(CEA) production systems. In aggregate, these changes have - and will continue - to alter opportunities for the agricultural sector to develop quality products for fresh produce markets.

Plant breeding through the lens of product development

The discipline of plant breeding seeks to advance the scientific understanding of genetics, domestication traits, environmental interactions, and populations with the purpose of plant improvement for human benefit (Bernardo, 2010; Fehr, 1987). The expectations of a plant breeder are not easily defined and vary considerably based on the crop species and market demands. The demands of horticultural crop improvement can be fluid in the face of a changing environment, hostile social and political climates, shifts in popular trends, socioeconomic criteria, and rapidly developing food systems. Plant breeding programs focused on cultivar development can be characterized by a common progression from trait or concept identification to cultivar release and adoption; intermediary steps involve germplasm evaluation, generation of variable plant populations, selection/early testing of genotypes, and advanced field trials in different environments. The complexity of the breeding pipeline changes dramatically depending on crop species, regional adaptation, reproductive systems, genetic architecture of desired trait, and the incorporation of genetic tools (Gepts,

2002). The assumption that all plant breeding programs fit within an idealized plant breeding pipeline is an oversimplification, especially for crops whose markets are based on product quality and marketable value. Regardless, plant breeders have a unique opportunity to address quality traits at the onset of plant development and set a

5

precedent for improved product quality in downstream supply channels. The ultimate success of breeding efforts manifests in an adopted finished cultivar or variety with the potential to alter downstream activities within horticultural systems.

In practice, plant breeding pipelines closely mimic traditional food product development pipelines in that they rely on an iterative and integrated approach to identify valuable material. A traditional product development pipeline is often characterized by first identifying company objectives and perceived needs of the market. The pipeline concludes by revisiting those objectives and market assumptions, but only after further idea generation, product screening, scaling, and product launch

(Earle, 2001). As with plant breeding, product developers are in a constant flux among stages of the pipeline, engaging in continuous qualitative and quantitative review of products and markets. This iterative process allows for flexibility and accommodation as developers progress through the pipeline and reassess product positioning and feasibility relative to different market sectors. Product launch, the final step in the production, is viewed as not only the culmination of a multidisciplinary effort, but also as an opportunity to further test markets and consumers in hopes of continued product refinement (Fuller, 2011). Consumer and market-oriented philosophies drive many modern product development pipelines, while upstream breeding efforts can drive the introduction of novel or unfamiliar products and traits.

A plant breeding pipeline and a traditional product development pipeline, while similar, cannot be fully equated, especially for fresh market products where quality is based largely on marketable value. The nature of plant product development differs from traditional product development pipelines in that it can take years to

6

properly identify and develop stable genetics that are regionally adapted for production. Plant breeding is an ongoing, long-term endeavor. Natural plant products are unpredictable, and their performance is complicated by genotypic and environmental interactions. Development of clonally propagated fruits, such as apples and grapes, can take over 25 years from initial population development to cultivar release. Annual seed crops, even in programs capable of multiple plant generations in a year, are often confined to a 5-15 year timeline from initial population development to release. This is in stark contrast to many product development pipelines in the food and technology industries, which have been refined for swift and efficient adjustments.

The timescale of a breeding project is not easily responsive to rapid market fluctuations and catering breeding objectives to reflect market projections five, ten, or twenty years into the future is highly challenging and requires insight into changing food systems. Finally, plant breeders are also responsible for maintaining quality expectations of nearly all the supply chain stakeholders, including growers, processors, distributors, retailers, and consumers. A plant breeder is taxed with not only recognizing stakeholder expectations, but also incorporating these into their selection criteria via formal or information selection indices. As a result, many breeding pipelines often involve introgressing a single trait into a genotype of known supply chain acceptability.

Consumer-defined quality in the context of a plant breeding program

Many players across food supply chains are often removed from the day-to-day activity of a plant breeder, but communication across members of a food system is considered a necessary component of successful cultivar development programs.

7

Partnerships and interactions between plant breeders and supply chain stakeholders depends on crop type, breeder motivation, and available resources. Informal or formal approaches define the framework by which a plant breeder collects and integrates quality information into their program (Pecher and Oppen, 2000). Informal approaches engage with market and consumer information through on-farm trials, field demonstrations, agricultural conferences, public seminars, conversations with agricultural professionals, and interactions with groups across supply chains. The expectation within these settings is that information flows bi-directionally, but there is minimal capacity to understand the utility of these interactions and their weight in the breeder’s personal selection index. Formal approaches to understanding quality information are typically standardized and provide more objective trait information which can feed into a selection index. Formal approaches employ consumer surveys

(e.g. online surveys or hedonic evaluation) or consumer panels (e.g. focus groups or trained sensory panels) to elucidate preferred quality traits or a relative ranking of preference among samples. The combination of informal and formal approaches generates a holistic perspective of quality according to stakeholders along the supply chain. Formal approaches, however, allow for targeted stakeholder feedback, enable the recognition of market opportunities, and can be quantitatively integrated into selection indices.

Literature connecting consumers to fresh-market produce is commonly conducted within the disciplines of food science, post-harvest technology, or applied economics research. Consumer quality feedback relevant to plant breeding is commonly acquired through formal sensory analysis. The field of sensory science,

8

also commonly referred to as sensory analysis or sensory evaluation, is rooted in psychology and psychophysics and aims to quantify human perception and response to external stimuli (Lawless and Heymann, 2010a; Stone et al., 2012). Since its inception in the 1950s, sensory analysis has been cultivated within industry settings as a tool to generate consumer insights that inform product development, quality control, and marketing efforts (Moskowitz et al., 2004a). The majority of sensory research falls into either analytic or affective testing categories. Analytical measures describe whether certain products are perceptively different from one another (i.e. discrimination testing) and how or what product attribute(s) contribute to the product perceptions (i.e. descriptive analyses). Both methods typically employ small, trained panels of experts to assess controlled product samples (Lawless and Heymann,

2010b). Affective testing, in contrast, is more commonly used to assess acceptance and preference (i.e. hedonic perceptions) among a larger population of untrained panelists, often product consumers (Lawless and Heymann, 2010c, 2010d).

The flexibility and capacity of sensory analysis to cater to both analytic or affective research questions provides an invaluable resource, especially to fresh market crop breeders who seek to better understand quality perception of breeding materials and finished cultivars. The complexity of flavor profiles in raw agricultural products are difficult to phenotype using instrumentation, and their perception is further complicated by an interaction with texture and appearance-related traits. The combination of analytic and affective sensory analysis enables the description of plant materials according to sensory characteristics and quantitative measures of consumer perception. Amyotte et al. (2017) noted that “human sensory perception removes the

9

ambiguity of modeling perceived fruit quality as a function of analytical traits and directly selects the quality attributes themselves.” Integrating consumer quality perception in the upstream product development pipeline could influence the ultimate success and value of new cultivars.

Sensory analysis has been performed extensively within fruits and vegetables.

Numerous studies have outlined flavor profiles through sensory analysis and reviews have been duly compiled (Bayarri and Costell, 2010; Ong and Liu, 2010). A large majority of sensory studies performed with raw agricultural products explore quality characteristics among commercially available cultivars or focus on product quality as it pertains to post-harvest demands. Sensory studies are typically deployed to correlate sensory perception with objective measures, develop lexicon and product profiles for the food industry, or evaluate finished products with consumers. Fruit breeders have successfully integrated sensory analysis into plant breeding pipelines; many sensory studies in plant breeding have been conducted within fruit crop breeding programs: strawberries (Colquhoun et al., 2012; Lado et al., 2010; Wang et al., 2017), blueberries

(Gilbert et al., 2015), peaches (Olmstead et al., 2015), and apples (Amyotte et al.,

2017; Hampson et al., 2000). Similar efforts could be more effectively deployed in vegetable crops, particularly highly perishable vegetable crops such as leafy greens.

Consumer sensory studies conducted within the context of a fresh market vegetable breeding program, with the intent of informing breeding decisions, appear less frequently. Tomato, albeit a botanically defined fruit, is perhaps the most well- studied vegetable in this context (Causse et al., 2003; Klee and Tieman, 2013; Rocha et al., 2013; Tieman et al., 2017). Studies within vegetable breeding programs include

10

carrots (Simon et al., 1981), sweet corn (Azanza et al., 1996), sweet pepper (Eggink et al., 2012), and lettuce (Chadwick et al., 2016; Park et al., 2009). Current ambiguity in sensory literature makes it difficult to assess the role of sensory analysis in developing new vegetable cultivars or draw a clear connection between sensory analysis and breeding decisions. In addition, the practice of using sensory analysis to determine the utility of new or unique traits is poorly defined among plant breeders.

Descriptive sensory analysis is particularly useful at the onset of developing quality standards or flavor-related breeding priorities; descriptive methods are commonly employed to assess targeted sensory traits or to define the sensory landscape of new products in breeding programs. In controlled testing environments, researchers have the capacity to conduct refined sensory methodology, including descriptive analyses such as the Spectrum Method® (Muñoz and Civille, 1992),

Texture Profile Method (Brandt et al., 1963; Muñoz et al., 1992), or Quantitative

Descriptive Analysis™ [QDA™; (Stone, 1992; Stone et al., 1974)]. These methods require attribute lexicon-development, extensive panelist training, and standardized sampling procedures before product evaluations (Lawless and Heymann, 2010b). They have demonstrated replicable discriminatory power and can be correlated with instrumental measurements. Formal descriptive analysis is still used infrequently in breeding programs, as the participant training and evaluation cycle is time intensive, costly to maintain, and often facilitates phenotyping of few lines. Deployment of descriptive analysis in breeding programs is also hindered by available resources and access to sensory expertise. Alternative rapid profiling methods seek to alleviate researchers of traditional panelist constraints [i.e. experts vs. consumers (Moskowitz et

11

al., 2004b)], lengthy panelist training periods, and restrictive testing environments

(Dehlholm et al., 2012, Delarue, 2015). Rapid sensory methods, including Free Choice

Profiling, Projective Mapping, and Free Multiple Sorting, have opened additional avenues to perform integrated sensory analysis with diverse stakeholders on timelines amenable with breeding cycles. Few programs have conducted rapid sensory methods to date, but it is a promising area of research when concerning plant breeders (Dawson and Healy, 2018). Deployment of these strategies is especially useful in new and emerging crop market classes.

Sensory studies commonly use late-stage breeding materials, recently released, or established commercial and heirloom cultivars in their evaluations. Simon and

Peterson (1981) noted that sensory evaluation with a trained panel was impractical during early-stage evaluations of carrot breeding populations. The same sentiment may be extended to large-scale consumer acceptance studies. Reasons may include the logistical issues regarding availability of product during the early stages of development, when seed supply and plant material available for sensory testing may be limited. Early generations of selection and evaluation are prone to single-plant selections, leaving very few seeds or clones for replication (Fehr, 1987). Sensory evaluation may also be impractical in early generations due to underlying genetic factors and lack of sufficient vigor in the field. Genes implicated in flavor or quality attributes may not be homogenous within breeding population, and the genetic instability of genotypes during early generations may introduce variability and further complexity in the analysis of sensory data. Therefore, it is assumed that sensory analysis performed with late-stage materials provides a proxy for the selection of

12

parents to be used in future population development.

As plant scientists and food chemists begin to connect the biochemistry and genetics associated with some quality traits (Tieman et al., 2017), sensory analysis will remain a key component in connecting intrinsic quality parameters to consumer perception. The integrated, consumer-oriented approaches to plant breeding, such as

“consumer-assisted selection” proposed by Folta and Klee (2016), have shifted existing models of plant breeding to use consumer perception of quality to guide the breeding and development of materials. These approaches emphasize the role of consumers, the ultimate end-user, in defining supply chain attributes (i.e. plant traits) through their purchasing and consumption decisions. Identification of plant traits integral to the consumer perception of quality provides the foundation for greater utilization of consumer-oriented strategies. Plant breeding programs have an opportunity to partner with sensory analysis experts to not only bridge fundamental research with consumers and markets, but to also provide an opportunity to creatively redesign selection procedures for quality within breeding pipelines.

Botany and breeding of Brassica oleracea L. vegetables

Domestication and taxonomy

Members of the Brassicaceae plant family, formerly known as Cruciferae, include many crops of agronomic and horticultural importance, including radish

(Raphanus spp.), canola and rutabaga (Brassica napus), Ethiopian mustard (Brassica carinata), broccoli and cabbage (Brassica oleracea), turnip and oilseed rape (Brassica rapa), arugula (Eruca sativa), black mustard (Brassica nigra), and the model species

Arabidopsis thaliana. Species within the Brassicaceae family were domesticated over

13

a large region spanning from current day western Europe to northern Africa and eastern Asia (Dixon, 2007). The Brassicaceae plant family today exhibits a rich diversity of plant morphotypes resulting from direct selection for seed and vegetative tissue over thousands of years. As of 2006, the plant family consisted of nearly 3,700 plant species classified within 338 genera (Al-Shehbaz et al., 2006).

The Brassica genus, in particular, contains numerous species of economic importance in current cropping systems. Interspecific hybridization

(allopolyploidization) events among domesticated diploid (2x) progenitors (B. oleracea, B. campestris, B. nigra) produced allopolyploids (4x) species (B. carinata,

B. juncea, and B. napus). Genetic relationships among relevant species in the Brassica genus have been well described (Kim et al., 2018; U, 1935). The inherent diversity within the genus lends itself to continued diversification of modern cultivated types and improvement for yield, quality, and seed production through introgression of disease resistance genes and cytoplasmic male sterility (Dickson and Wallace, 1986).

The diversity among Brassica market classes includes quality traits that are desirable in downstream consumer markets.

Brassica oleracea (2n = 2x = 18) remains one of the most diverse members of the Brassica genus and contains a significant number of horticultural crop market classes. Cultivated forms of B. oleracea include morphotypes distinguished by their inflorescences (broccoli and cauliflower), terminal or axillary buds (cabbage and

Brussel’s sprouts), enlarged stems (kohlrabi), and non-heading leaf traits (kale and collard). Wild progenitors of these morphotypes grow today along cliff-sides of the

Mediterranean basin and European West Atlantic Coasts. Genetic and literary

14

evidence suggests domestication of the species occurred in the cool, moist climates of the Mediterranean basin (Gómez-Campo and Prakash, 1999; Maggioni et al., 2010;

Snogerup, 1980). Progenitors of the modern B. oleracea varieties most closely resemble a modern leafy kale type with very thick leaves, small leaf areas, and robust xylem structures.

Plant morphotypes within the B. oleracea species are also known as ‘cole crops’, a term originating from the Greek word καυλός (kaulos) and later derived

Latin word caulis (Maggioni et al., 2018). The term highlights the prominent primary stem characteristic of the species. The earliest written recorsd of cole crops date to the sixth century B.C.E. Greek literature, and anthropological studies cite descriptions of cole crops dating back to the Roman Empire and Celtic tribes (Dixon, 2007; Sauer,

1993). By the Middle Ages, texts and paintings depicted distinguishable B. oleracea morphotypes. Most of the modern morphotypes were heavily cultivated in European gardens as storage and winter crops and brought to North America as early as the

1500s. Cole crops (Brassica oleracea) are commonly touted for their health- promoting nutritional and phytochemical profiles; these vegetables have been associated with low-caloric diets, chemo-protective strategies, and enhanced antimicrobial activity (Björkman et al., 2011; Walley and Buchanan-Wollaston, 2011).

Today, nearly all plant parts of B. oleracea, including the stem, axillary buds, leaves, and floral buds are harvested for human consumption on every arable continent.

Diverse trait-delineated market classes exist within each B. oleracea morphotype. Cabbage, for example, is a widely consumed crop with diverse geographical taste preferences. Market classes in cabbage are determined largely by

15

simple trait differences: pointed or sweetheart cabbage (shape), red cabbage (color), savoy cabbage (leaf texture), storage cabbage (size), processing cabbage (soluble solids), flat Dutch cabbage (tenderness), etc. These are all domestication traits that can be transferred into other B. oleracea morphotypes increasing in popularity, including the leafy Brassica vegetables at the focus of this research. Directed breeding to introgress these domestication traits can also improve our understanding of the relative importance of each trait in different plant morphotype backgrounds.

Botany and mating systems

Nineteenth century public and private breeding efforts in Europe and the U.S. produced productive, commercially available cultivars to support growing industries.

B. oleracea is characterized by both annual (broccoli and cauliflower) and biennial

(cabbage, kale, Brussels sprouts) reproductive systems. The biennial life cycle is generally considered the ancestral trait and necessitates a winter vernalization period, or a minimum chilling phase, to trigger reproductive development. There can be significant variability in the required vernalization temperature and time requirement among biennial types, though generally a temperature of 8-9°C (48°F) for 8-10 weeks is sufficient to induce flowering in fully grown plants. Due to strict vernalization requirements, the seed-to-seed life cycle of biennial plants such as cabbage and borecole kales ranges between 12-24 months but can be reduced to 9-12 months when using artificial cooling and greenhouse production. Vernalization requirements of other kale types, such as Tuscan kale, have minimal or no chilling requirements and appear as an annual type in some environments. The chilling conditions for flowering of B. oleracea market classes vary widely from almost no chill requirement in crops

16

like kailaan and tropical cauliflower, to cooler climate growth conditions for crops like broccoli and some collards, to cold seasonal winter vernalization for Brussels sprouts and cabbage.

During reproductive stages, perfect flowers are borne along racemes emerging from the central stem ad axillary branches. The flowers consist of four symmetrical petals, typically yellow or white in color with six stamens. Sporophytic self- incompatibility (SI) restricts the capacity of the outcrossing market classes to self- pollinate. The SI system, however, is often imperfect in unfavorable environmental conditions, less common among annual types, and can vary in degree depending on plant age, generation of inbreeding, and vigor (Dickson and Wallace, 1986). The degree of SI is typically higher among genotypes with high vernalization requirements. Self-incompatibility necessitates insect-mediated cross pollination for adequate seed set in most outcrossing Brassicas, though self-compatibility can be frequently observed in annual types such as broccoli and kailaan. Hand pollination of immature flowers prior to onset of SI activation or a CO2 treatment (Nakanishi and

Hinata, 1973) enables directed self-pollination and cross pollination in breeding materials regardless of SI in the diploid plant.

Commercially available cultivars today are developed as open-pollinated (true breeding or mixed lines) or hybrid genotypes, the latter increasing in commercial prevalence. Hybrid seed production remains complicated in that it requires the timing of flowering between two inbred parents (‘nicking’), the identification of female inbred lines with high quality seed set and quality. Parental combinations with good combing ability or trait complementation are also important in developing a successful

17

cultivar. Hybrid genotypes are advantageous in that they are often highly uniform, exhibit heterosis (hybrid vigor) leading to higher yield, provide increased intellectual property protection, and show complementation of horticultural traits.

The seed industry has increasingly shifted toward hybrid seed production using cytoplasmic male sterility (CMS) due to the breakdown of a SI systems and resulting off-types in hybrid populations (Myers, 2014). The identification and introgression of CMS, namely the ‘Ogura’ source derived from protoplast fusion with radish (Raphanus sativus), created a stable means of developing hybrid cultivars for commercial sale when integrated into a self-compatible, high yielding female seed parent (Yamagishi and Bhat, 2014). Maintenance of CMS inbred lines (A-lines) and their fertile maintainers (B-lines) was also greatly improved through selection of self- compatible female lines enabling more efficient maintenance of parental seed stock.

Adoption of CMS systems is more common in self-compatible annual Brassica parental lines (e.g. broccoli, cauliflower) where the SI system is less reliable for hybrid seed production, though hybrid development using SI is still widely used in biennial Brassica species where SI is often more reliable (e.g. cabbage, Brussel’s sprouts, etc.).

Breeding for quality traits in cole crops

Plant breeders are tasked with balancing the requirements of horticultural production traits, supply chain attributes, and end-user quality expectations while maintaining high yields of shelf-stable products. There is often a tension in breeding for quality traits, particularly consumer-focused quality traits such as flavor, color, and nutrition. Selection for consumer-defined quality traits must account for correlated,

18

sometimes detrimental changes in plant physiology and performance including those that alter crop productivity, disease resistance, and shelf-stability (Shewfelt, 2000;

Ulrich and Olbricht, 2011). Nutritional quality and phytochemical profiles of cole crops have been extensively studied and dully compiled. Leafy Brassicas are diverse in color and flavor imparting phytochemicals, particularly carotenoids, anthocyanins, and glucosinolates. These compounds are diverse among morphotypes and their relative concentrations are subject to changes in genetics, environment, plant age, and post-harvest handling. Modification of phytochemicals in food crops can have downstream effects on horticultural productivity and consumer perception. Without understanding how modifications to the underlying phytochemical profiles can affect intrinsic quality perceptions, plant breeders may face a lack of market acceptance and/or adoption (Ceccarelli and Grando, 2007; Persley and Anthony, 2017).

Pigment profiles

As one of the first visible intrinsic cues encountered in a purchasing situation, color plays an important role in the evaluation of food quality. Color or color combinations affect perception of flavor intensity, mediate color-taste associations in humans, and have the potential to generate novelty in the marketplace (Higgins and

Hayes, 2019; Wadhera and Capaldi-Phillips, 2014; Zampini et al., 2007). Four primary classes of secondary metabolites govern most pigment in plants: chlorophylls, carotenoids, flavonoids, and species-specific betalains. The complex biosynthetic pathways and genetic regulation of these metabolites are well described in plant systems (Allan and Espley, 2018; Sun and Li, 2020; Tanaka et al., 2008). Human perception and hedonic responses are additional criterion to consider when breeding

19

for pigment profiles and color-related attributes in plants. Pigmentation patterns that are perceivably different to the human eye may be the result of genetic background, environmental conditions, modifications in waxy cuticle layers of fruits and leaves, co-pigmentation effects outside specific absorption regions, or perceptual differences caused by an interaction among other plant pigments in various cell layers (Gould,

2004). Modification of the underlying phytochemicals responsible for color can influence consumer perception of freshness and quality of vegetable crops.

Yellow, orange, and red plant pigments are produced by photoprotective carotenoid compounds. Carotenoids appear in nearly all plant tissues, but pigments in photosynthetic tissues are often masked by chlorophyll or are colorless (i.e. lutein).

These lipid-soluble phytochemicals are precursors for Vitamin A, associated with retinal health (Johnson et al., 2000), and have been connected to preventative therapies for a diverse array of human diseases (Krinsky and Johnson, 2005). Carotenoids in green leaf tissue are produced primarily in the chloroplast where their metabolism is tightly linked to chlorophyll production (Sun and Li, 2020). Regulatory control of the isoprenoid pathway via the PSY enzyme produces two primary carotenoid classes: carotenes (α- and β-carotene) and oxygenated xanthophylls (i.e. lutein and zeaxanthin)

(Yuan et al., 2015).

Relative concentrations of carotenoids in cole crops are highly dependent on genetic background, often tissue-specific, and vary with environmental conditions

(Guzman et al., 2012; Kopsell et al., 2004; Lefsrud et al., 2007). In comparison to other leafy greens, such as spinach and lettuce, leafy Brassicas contain high levels of lutein, zeaxanthin, and β-carotene (Khachik et al., 1986; Kurilich et al., 1999).

20

Yellowing leafy greens have generally been identified by consumers as a negative quality trait; lighter green and yellow leaves are anecdotally associated with aging, nutritionally deficient sensory profiles. Some consumers, however, might view genetically stable yellow types as a bright and welcoming aesthetic addition to a dish.

Heirloom ‘cabbage collards’, for example, have been maintained by seed savers in

North Carolina for their yellowish hue and sweet flavor profile in comparison to other collard cultivars. It is unclear whether the characteristic yellowish hue of cabbage collards is due to carotenoid, chlorophyll, flavonoids, or other polyphenolic content, but the quality trait remains of interest to seed savers and home gardeners (Davis and

Morgan, 2015).

Carotenoid expression in tissues other than the leaves has also diversified the appearance and nutrition of cole crops. The incompletely dominant Or (Orange) gene affects the accumulation of β-carotene in curd tissues (Li et al., 2003). The novel orange curded cauliflower genotype was first described in 1975 and has since been introgressed into commercial material to diversify cauliflower markets (Crisp et al.,

1975; Dickson et al., 1988). Popularity of niche curd color market types have led to the development of additional phenotypic diversity in Western cauliflower markets, including purple-curded and green-curded (broccoflower or Romanesco) phenotypes.

Additional pigmentation in cole crops is controlled by the expression of anthocyanins, a class of water-soluble flavonoids present in nearly all plants. Pigments resulting from anthocyanin expression are wide ranging (e.g. red, purples, blues), depend on the polyphenol grouping, and are highly influenced by the environment.

Anthocyanin compounds are secondary metabolites which can serve as

21

photoprotectors in the leaves, mitigate metal toxicity, and may play a defensive role during herbivory (Hatier and Gould, 2009; Landi et al., 2015; Lev-Yadun and Gould,

2009). Health-associated properties and antioxidant capacity of anthocyanins have been described in therapeutic and preventative capacities; anthocyanins have been associated with the prevention of certain cancers, diabetes, and neurodegenerative disease (Lila et al., 2016; Pojer et al., 2013). These polyphenolic compounds are present in nearly all higher-order plants but vary in regulation of their synthetic pathways, functional structure, and stability under different environmental conditions.

The capacity for anthocyanin production is universal among B. oleracea morphotypes. Anthocyanin profiles in the species are primarily attributed to acylation and glycosylation of the cyanidin base molecule (Wu et al., 2006). The genetic control of anthocyanin synthesis in red cabbage was described by Yuan et al. (2009) as an interplay between expression of structural genes and three families of transcriptional factors: MYB domains, basic helix-loop-helix (bHLH) domains, and WD40 proteins

(Koes et al., 2005). Similar to the Or carotenoid gene, an incompletely dominant Pr

(Purple) gene that encodes a R2R3 MYB transcription factor responsible for the purple and violet hues in purple-curded cauliflowers (Chiu et al., 2010). Molecular studies using visual phenotyping in ornamental kale have mapped the pink inner-leaf phenotype to a single incompletely dominant gene (Pi) (Zhu et al., 2016), while single dominant genes were reported as responsible for red (Re) and purple (BoPr) phenotypes in ornamental kale (Liu et al., 2017; Wang et al., 2013). These studies in ornamental kale, however, represent genetic mapping performed with narrow germplasm, limited environmental replication/stress, and little discussion of

22

anthocyanin pathways. Guo et al. (2019) conducted a more exhaustive expression analysis in ornamental kale and revealed numerous biosynthetic and regulatory genes involved in the expression of ornamental kale anthocyanins. Pigmentation in B. oleracea is subject to dominant gene inheritance but regulation is more likely the result of incompletely dominant gene action and modifiers (Baggett, 1978; Sampson,

1967). Population development has also indicated that pigment genes can act in a complementary manner (P.D. Griffiths, personal communication).

Anthocyanin synthesis in B. oleracea, via upregulation of the entire biosynthetic pathway, is often stimulated in stress environments (Chalker‐Scott, 1999;

Yuan et al., 2009). Anthocyanin concentrations are sensitive to changes in pH and increase as a response to cold during development and post-harvest handling (Mazza and Brouillard, 1987; Socquet-Juglard et al., 2016; Zhang et al., 2012). Regulating concentrations of anthocyanins and standardizing their production outside of a controlled environment setting is a challenge to food supply chains; their production typically incurs a metabolic and production cost (Chalker‐Scott, 1999; Gould, 2004) and is subject to modifications during storage and distribution. Anthocyanin production in B. oleracea has also been linked to certain self-incompatibility alleles

(Ockendon, 1977; Thompson and Taylor, 1965), which can complicate inbreeding and selection processing in a breeding program. Continued improvement of anthocyanin germplasm, however, can directly support the development of natural food coloring and add market diversity available to health-conscious consumers (Bridle and

Timberlake, 1997; Charron et al., 2007).

23

Flavor constituents

The perception of flavor is traditionally understood as the synthesis of taste and aroma (i.e. smell) stimuli, though it has also been argued that flavor perception is a multimodal sensory experience drawing on trigeminal, tactile, visual, and auditory cues (Auvray and Spence, 2008; Small and Prescott, 2005). Basic tastes (sweet, sour, bitter, salty, umami) are detected by taste receptor cells across the fungiform papillae on the human tongue (Chandrashekar et al., 2006). Taste stimuli are integrated with olfactory and retronasal cues, often elicited by volatile compounds in foods, to generate perceptible flavors (Pierce and Halpern, 1996). Flavor perception ultimately affects the preference and acceptance of food products, and it can alter the perceived nutritional and health value of foods.

Flavor compounds and their biosynthesis in vegetables are well detailed, but their interactions and relationship to human perception remain complex. Secondary metabolites commonly associated with vegetable flavor include classes of terpenoids, glucosinolates, phenolics, and alk(en)yl cystine sulphoxides (Jones, 2008). Some terpenoids and phenols may play a role in Brassica aroma, and lipid oxygenation or tissue rupture can lead to formation of volatile alcohols or aldehydes producing

“green” or “grassy” flavor notes (Baik et al., 2003). The majority of the flavor profile in cole crops, however, can be dominated by the enzymatic breakdown of glucosinolates (GLS) or methyl cysteine sulphoxide (MSCO).

Glucosinolates are a diverse group of nearly 130 organosulfur compounds identified by various amino side groups bonded to a core thioglucoside structure

(Agerbirk and Olsen, 2012; Hopkins et al., 2009; Sønderby et al., 2010). Relative

24

concentrations of GLS and their downstream products vary by genotype, environment, and plant tissue sampled (Björkman et al., 2011; Farnham et al., 2004; Rosa et al.,

1994; Tian et al., 2016; Velasco et al., 2007). Modification and regulation of glucosinolates is of particular interest due to their health and nutritional benefits.

Glucosinolates have been associated with reducing the risk of certain cancers and preventing degenerative diseases (Björkman et al., 2011; Cartea and Velasco, 2008;

Traka and Mithen, 2009). They also mediate complex insect-plant defense interactions

(Hopkins et al., 2009; Rask et al., 2000). Glucosinolates can stimulate feeding and oviposition in specialist Brassicaceae insects (e.g. diamondback moth (Plutella xylostella), cabbage root fly (Delia radicum), white butterfly (Pieris brassicae) and, in contrast, play a defensive role in mediating predation from generalist insects

(Björkman et al., 2011; Jeschke et al., 2017). Alteration or modified regulation of GLS production can lead to dramatic changes in the nutritional quality, productivity, and sensory profile of Brassica vegetables.

Bitter taste receptivity in humans evolved as a response to detect toxic or harmful food products. The ability to sense bitter taste is a highly conserved trait, but variations in sensitivity are genotype specific and associated with vegetable consumption frequency (Chandrashekar et al., 2006; Dinehart et al., 2006). Damage or maceration of fresh tissue releases the myrosinase enzyme and induces hydrolysis of

GLS to form glucose, sulphate, and byproducts such as isothiocyanates, nitriles, and thiocyanates (Wieczorek et al., 2018). These compounds have shown great diversity in their capacity to produce the characteristic bitter, pungent flavors of cole crops.

Breakdown products of MSCO, namely volatile sulfides, also contribute to the

25

sulfurous and “off-flavors” of cole crops. The C-S lysase and myrosinase enzymes responsible for the breakdown of MSCO and GLS, respectively, are denatured via heat and the characteristic pungent flavors of Brassica crops can be reduced greatly as a result of cooking or processing (Barba et al., 2016). Sweet stimuli have also been demonstrated to suppress the perception of bitter taste (Green et al., 2010); bitter and sweet tastes are detected by G protein coupled receptors (GPCR) in the same family and elicit similar signaling pathways (Scott, 2004; Zhang et al., 2003). Sucrose has been demonstrated to mask the bitter taste of sinigrin and goitrin in cabbage (Beck et al., 2014), but modification of free sugar levels in the plants may alter hormone signaling, plant defense, storage/post-harvest capacity, end markets (Voorrips et al.,

2008).

Brassica oleracea morphotypes contain a subset of specific GLS compounds, though the composition and concentration of individual GLS vary. Predominant GLS in the edible portions of B. oleracea include sinigrin, glucobrassicin, gluconapin, and progoitrin (Kushad et al., 1999). Significant work remains to distinguish the effect of individual GLS compounds on the perception bitterness in Brassica vegetables.

Recognition of specific GLS compounds and their thresholds for detection has allowed for the directed breeding and development of GLS profiles. For example, the relative ratios/concentration of sinigrin and progroitrin were highlighted as major constituents of bitter taste in Brussels sprouts and led to the development of less bitter cultivars in the late 1990’s (Doorn et al., 1998; Fenwick et al., 1983). In contrast, glucoraphanin was determined to have little-to-no effect on the perception of bitterness or consumer acceptance (Bell et al., 2018) and two high-glucoraphanin

26

broccoli cultivars, marketed under the name Beneforté ®, were developed in alignment with nutritional and health-relating breeding goals (Traka et al., 2013).

These examples represent a continued need to understand expectations for nutritional quality, plant productivity requirements, and complex interactions with human taste perception during breeding and development of new cultivars.

Leaf morphology

Leaves serve as the photosynthetic organs of plants. Alteration of leaf shape and size affects many underlying physiological processes critical to the growth and development of a plant, including light absorption and gas exchange (Tsukaya, 2006).

Leaf traits must be evaluated over a temporal scale, as they are subject to hormonal and environmental constraints (temperature, precipitation, nutrient availability etc.) during ontogeny. These developmental constraints influence not only leaf morphology, anatomy, and physiology, but also the productivity and quality of the leaves. Changes in leaf development can affect the texture and phytochemical compounds relevant to post-harvest management and quality.

The foundation of plant leaf shape is generally understood to be multigenic and heritable between parents and progeny in a breeding program (Chitwood and Sinha,

2016). Numerous quantitative trait loci (QTL) have been detected for laminar, petiole, and node-formation traits in B. oleracea (Lan and Paterson, 2001; Sebastian et al.,

2002; Stansell et al., 2019). Researchers commonly use automated software and imaging platforms for the analysis of leaf shape [LeafAnalyser (Weight et al., 2008) or

LAMINA (Bylesjö et al., 2008) or PlantSize (Faragó et al., 2018) and many others], and plant classification on the basis of leaf shape and texture has also developed as a

27

discipline in the computer sciences. These technologies are promising in the screening of new morphologies and textures in plant breeding programs, but continued efforts are needed to adapt these to complex leaf morphologies with extreme leaf curvature or leaf margins, such as borecole (curly) kale. Further, correlating these technologies with consumer liking could serve to better inform the plant breeding process.

Modification of leaf structure, including shape, size, epicuticular wax, and trichomes alters insect predation dynamics and disease progression in Brassica crops.

Highly curled, folded leaf margins of six-week old kale plants are postulated to interfere with flea beetles (Phyllotreta cruciferae) walking between feeding sites. In a study conducted by Vaughn and Hoy (1993), flea beetles tended to spend longer periods of time on flat, collard leaves than curly kale leaves, though feeding on the cotyledons of each morphotype remain the same. Across morphotypes of B. oleracea, larvae of diamond back moth (Plutella xylostella) are recognized to prefer younger, less fibrous leaves on the upper portion of the plants, despite higher levels of glucosinolates (Moreira et al., 2016). The glossy trait, generally considered a recessive trait conferring a non-glaucous, bright green lamina surface, has been associated with reduced infestation from some lepidopteran pests, whiteflies, and cabbage root fly eggs (Eigenbrode et al., 1991; Stoner, 1990). Glossiness undoubtedly changes the product appearance of Brassica leaves and is associated with vegetable freshness

(Barrett et al., 2010), but various sources suggest this trait reduces yield and may not be well liked by consumers (Dixon, 2007).

Modification of leaf morphology and its relation to consumer perception of texture has received minimal attention. Texture is a broad term used to describe the

28

mechanical, geometrical, and surface-related characteristics of a food product. The textural sensation results from a combination of tactile perception, visual assessment, and, occasionally, auditory cues (Lawless and Heymann, 2010e). Texture-related traits are often judged by consumers as the presence or absence of defects (Shewfelt, 2000), but humans have an immense capacity to detect a wide range of texture and are very sensitive to changes (Wilkinson et al., 2000). Perceptible textural attributes are modified by changes in anatomical characteristics in plant tissues, including cell size, water concentration, and dry matter. Texture analysis and sensory profiling is commonplace in processed foods and some high-value raw agricultural products but understanding of textural sensory properties in raw vegetable products, particularly leafy greens, remains understudied. Near-infrared (NIR) spectroscopy has been deployed to evaluated texture, dry matter, and color in spinach (Sánchez et al., 2018), but little instrumental work has been conducted to evaluate the texture of leafy

Brassicas. Instrumental evaluation of texture serves as the foundation for screening textural traits, but these measurements must be aligned and related to human perception (Szczesniak, 2002).

Assessment of human perception of leaf morphology is critical to gauge the acceptance of new products; changes in leaf morphology alter the appearance and may drastically change how consumers perceive a product. Alteration of leaf morphology maybe one of the easiest means by which plant breeders can introduce new diversity to markets and reframe existing expectations for certain plant ideotypes. For example,

Tozer Seeds (UK) recently released a suit of Kalettes®, a unique leafy morphotype that blurs the boundaries between kale and Brussels sprouts. Continued diversification

29

of leaf morphology, however, must integrate holistic evaluations of plant productivity and human perception before market introduction.

Translating sensory science to a leafy Brassica breeding program

The ability to hybridize with both wild species and other cultivated members of the Brassica genus has encouraged continued diversification of B. oleracea.

Continued production and selection for diverse leaf morphologies has resulted in numerous modern varieties of leafy cole crops within the ‘Acephala’ (“non-heading”) botanical group: acephala (borecole or Galega kales), viridis (collard), costata

(Portuguese cabbage/kale), medullosa (marrow stem kale), ramosa (thousand headed kale), palmifolia (Tuscan kale) (Diederichsen, 2001; Hammer et al., 2013). Nearly all of these leafy types are biennial but vary in the degree to which they require vernalization. Most are well suited to spring and fall plantings in the Northern

Hemisphere and can withstand heavy frosts. Certain types, such as collard, are better adapted to warmer conditions, but extremely hot environments result in leaf injury, suppression of plant growth, and may trigger a bolting response in some leafy Brassica varieties. These leafy types are grown globally for human consumption and animal fodder in cool, maritime climates. They are characterized by their long shelf-life, nutritional profiles, and durability when cooking, but they vary dramatically in their form. The Acephala group is diverse in leaf shape, size, and pigmentation due to variation in underlying phytochemicial profiles, but directed intraspecific crossing among morphotypes in the species has the potential to generate additional diversity and plant materials in plant breeding programs.

30

Kale (B. oleracea var. acephala L.) and collards (B. oleracea var. viridis) are common leafy green vegetables whose recent rise in popularity has reinvigorated interest from both fresh and processed markets across the U.S. Both kale and collards are low calorie produce items high in fiber, Vitamins A, C, and K, and contain a diverse phytochemical profile that appeal to nutrition-conscious consumers (Walley and Buchanan-Wollaston, 2011). Data from the 2017 Census of Agriculture estimates that approximately 11,318 acres of collard and 15,325 acres of kale were harvested for

U.S. fresh and processed markets (USDA-NASS, 2019); the majority of production in

2017 was intended for fresh produce markets. Collard is commonly grown and consumed in the southern U.S., though acreage in northern regions is beginning to increase. It retains a strong cultural identity in Georgia, South Carolina, and North

Carolina among seed savers and gardeners (Davis and Morgan, 2015; Diederichsen,

2001). Kale is a more recently popularized vegetable, generating significant horticultural production in Northeastern direct-to-consumer and organic markets. In

December of 2015, a stakeholder listening session focused on breeding needs for organic production identified kale as an important crop to organic growers in the northeast (Hultengren et al., 2016). Acreage dedicated to kale production increased five-fold between 2012 (104 ac.) and 2017 (573 ac.) within the state of New York alone (USDA-NASS, 2019).

Very few public plant breeding programs in the U.S. focus on biennial

Brassica crops – currently three public plant breeders developing new B. oleracea cultivars operate in the public sector. The Cornell AgriTech vegetable breeding program is housed in Geneva, New York at the New York State Agricultural

31

Experiment Station. Today the breeding program focuses on developing improved varieties of kale, cabbage, and broccoli. New York is historically an important place for Brassica crop production. It rivals California’s acreage dedicated to cabbage production, and the primary B. oleracea germplasm collection is housed at the United

States Department of Agriculture’s National Plant Germplasm (NPGS) NE-9 repository in Geneva, NY. Close relationships with industry and access to diverse germplasm have aided the program’s success since it began focusing on Brassica crops in the 1970’s. The program today consists of advanced leafy Brassica materials developed from various research projects, including breeding for heat-tolerant broccoli for the East Coast industry, improving collard/sukuma wiki for eastern Africa, development of high-anthocyanin lines for natural color use, and improvement of cabbage materials for resistance to black rot (Xanthomonas campestris). Advanced materials are diverse in their leaf shapes and colors, and their sensory profiles may play into new and emerging markets for leafy Brassica types.

The research that follows was designed to explore different facets of sensory evaluation and consumer research that may be applicable breeding programs interested in engaging consumers and/or integrating consumer feedback into larger plant breeding pipelines. This research has been conducted within the context of an applied plant breeding program focused on the breeding and improvement of leafy Brassicas, namely kale and collard genotypes.

This research was driven by expanding consumer markets for Brassica leafy greens, for which very little consumer and/or market information was available to guide new selections. Recognizing there was little consumer insight into the

32

perception of these materials, this research sought to obtain baseline sensory and quality information to inform continued development of novel market types. Each of the following research projects is not only analytic, but also a creative effort to think about how an applied plant breeding program can use sensory analysis tools to build a new, integrative means of connecting consumers and plant breeders in fresh market crops. Presented herein are multiple studies within three primary project areas: an exploration into the quantitative and qualitative sensory landscape of leafy Brassicas

(Chapter 2), a cross-cultural examination of regional preferences and sensory lexicons

(Chapter 3), and development of a diallel mating design to assess the validity of consumer liking as a phenotype for selection (Chapter 4). These projects established a foundation for quality-driven breeding in the leafy Brassica breeding program and highlight the need for continued multidisciplinary research in the development of novel vegetable cultivars.

33

References

Acquaah, G., 2009. Horticulture: Principles and Practices. Pearson Prentice Hall. Agerbirk, N., Olsen, C.E., 2012. Glucosinolate structures in evolution. Phytochemistry 77, 16–45. https://doi.org/10.1016/j.phytochem.2012.02.005 Allan, A.C., Espley, R.V., 2018. MYBs Drive Novel Consumer Traits in Fruits and Vegetables. Trends in Plant Science 23, 693–705. https://doi.org/10.1016/j.tplants.2018.06.001 Al-Shehbaz, I.A., Beilstein, M.A., Kellogg, E.A., 2006. Systematics and phylogeny of the Brassicaceae (Cruciferae): an overview. Plant Syst. Evol. 259, 89–120. https://doi.org/10.1007/s00606-006-0415-z Amyotte, B., Bowen, A.J., Banks, T., Rajcan, I., Somers, D.J., 2017. Mapping the sensory perception of apple using descriptive sensory evaluation in a genome wide association study. PLoS One; San Francisco 12. http://dx.doi.org.proxy.library.cornell.edu/10.1371/journal.pone.0171710 Auvray, M., Spence, C., 2008. The multisensory perception of flavor. Consciousness and Cognition 17, 1016–1031. https://doi.org/10.1016/j.concog.2007.06.005 Azanza, F., Klein, B.P., Juvik, J.A., 1996. Sensory Characterization of Sweet Corn Lines Differing in Physical and Chemical Composition. Journal of Food Science 61, 253–257. https://doi.org/10.1111/j.1365-2621.1996.tb14772.x Baggett, J.R., 1978. Inheritance of internal pigmentation in green cabbage (Brassica oleracea var. capitata L.). Euphytica 27, 593–599. Baik, H.-Y., Juvik, J.A., Jeffery, E.H., Wallig, M.A., Kushad, M., Klein, B.P., 2003. Relating Glucosinolate Content and Flavor of Broccoli Cultivars. Journal of Food Science 68, 1043–1050. https://doi.org/10.1111/j.1365- 2621.2003.tb08285.x Barba, F.J., Nikmaram, N., Roohinejad, S., Khelfa, A., Zhu, Z., Koubaa, M., 2016. Bioavailability of Glucosinolates and Their Breakdown Products: Impact of Processing. Front Nutr 3. https://doi.org/10.3389/fnut.2016.00024 Barrett, D.M., Beaulieu, J.C., Shewfelt, R., 2010. Color, Flavor, Texture, and Nutritional Quality of Fresh-Cut Fruits and Vegetables: Desirable Levels, Instrumental and Sensory Measurement, and the Effects of Processing. Critical Reviews in Food Science and Nutrition 50, 369–389. https://doi.org/10.1080/10408391003626322 Bayarri, S., Costell, E., 2010. Sensory Evaluation of Fruit and Vegetable Flavors, in: Hui, Y.H. (Ed.), Handbook of Fruit and Vegetable Flavors. John Wiley & Sons, Inc., pp. 45–57. https://doi.org/10.1002/9780470622834.ch3 Beck, T.K., Jensen, S., Bjoern, G.K., Kidmose, U., 2014. The Masking Effect of Sucrose on Perception of Bitter Compounds in Brassica Vegetables. Journal of Sensory Studies 29, 190–200. https://doi.org/10.1111/joss.12094 Bell, L., Oloyede, O.O., Lignou, S., Wagstaff, C., Methven, L., 2018. Taste and Flavor Perceptions of Glucosinolates, Isothiocyanates, and Related Compounds. Molecular Nutrition & Food Research 62, 1700990. https://doi.org/10.1002/mnfr.201700990

34

Bernardo, R.N., 2010. Breeding for quantitative traits in plants, 2nd ed. ed. Stemma Press, Woodbury, Minn. Björkman, M., Klingen, I., Birch, A.N.E., Bones, A.M., Bruce, T.J.A., Johansen, T.J., Meadow, R., Mølmann, J., Seljåsen, R., Smart, L.E., Stewart, D., 2011. Phytochemicals of Brassicaceae in plant protection and human health – Influences of climate, environment and agronomic practice. Phytochemistry 72, 538–556. https://doi.org/10.1016/j.phytochem.2011.01.014 Brandt, M.A., Skinner, E.Z., Coleman, J.A., 1963. Texture Profile Method. Journal of Food Science 28, 404–409. https://doi.org/10.1111/j.1365- 2621.1963.tb00218.x Bridle, P., Timberlake, C.F., 1997. Anthocyanins as natural food colours—selected aspects. Food Chemistry 58, 103–109. https://doi.org/10.1016/S0308- 8146(96)00222-1 Bylesjö, M., Segura, V., Soolanayakanahally, R.Y., Rae, A.M., Trygg, J., Gustafsson, P., Jansson, S., Street, N.R., 2008. LAMINA: a tool for rapid quantification of leaf size and shape parameters. BMC Plant Biol 8, 82. https://doi.org/10.1186/1471-2229-8-82 Cartea, M.E., Velasco, P., 2008. Glucosinolates in Brassica foods: bioavailability in food and significance for human health. Phytochem Rev 7, 213–229. https://doi.org/10.1007/s11101-007-9072-2 Causse, M., Buret, M., Robini, K., Verschave, P., 2003. Inheritance of Nutritional and Sensory Quality Traits in Fresh Market Tomato and Relation to Consumer Preferences. Journal of Food Science 68, 2342–2350. https://doi.org/10.1111/j.1365-2621.2003.tb05770.x Ceccarelli, S., Grando, S., 2007. Decentralized-participatory plant breeding: an example of demand driven research. Euphytica 155, 349–360. https://doi.org/10.1007/s10681-006-9336-8 Chadwick, M., Gawthrop, F., Michelmore, R.W., Wagstaff, C., Methven, L., 2016. Perception of bitterness, sweetness and liking of different genotypes of lettuce. Food Chemistry 197, 66–74. https://doi.org/10.1016/j.foodchem.2015.10.105 Chalker‐Scott, L., 1999. Environmental Significance of Anthocyanins in Plant Stress Responses. Photochemistry and Photobiology 70, 1–9. https://doi.org/10.1111/j.1751-1097.1999.tb01944.x Chandrashekar, J., Hoon, M.A., Ryba, N.J.P., Zuker, C.S., 2006. The receptors and cells for mammalian taste. Nature 444, 288–294. https://doi.org/10.1038/nature05401 Charron, C.S., Clevidence, B.A., Britz, S.J., Novotny, J.A., 2007. Effect of Dose Size on Bioavailability of Acylated and Nonacylated Anthocyanins from Red Cabbage (Brassica oleracea L. Var. capitata). J. Agric. Food Chem. 55, 5354– 5362. https://doi.org/10.1021/jf0710736 Chitwood, D.H., Sinha, N.R., 2016. Evolutionary and Environmental Forces Sculpting Leaf Development. Current Biology 26, R297–R306. https://doi.org/10.1016/j.cub.2016.02.033

35

Chiu, L.-W., Zhou, X., Burke, S., Wu, X., Prior, R.L., Li, L., 2010. The Purple Cauliflower Arises from Activation of a MYB Transcription Factor. Plant Physiology 154, 1470–1480. https://doi.org/10.1104/pp.110.164160 Colquhoun, T.A., Levin, L.A., Moskowitz, H.R., Whitaker, V.M., Clark, D.G., Folta, K.M., 2012. Framing the perfect strawberry: An exercise in consumer-assisted selection of fruit crops. Journal of Berry Research 2, 45–61. https://doi.org/10.3233/JBR-2011-027 Crisp, P., Walkey, D.G.A., Bellman, E., Roberts, E., 1975. A mutation affecting curd colour in cauliflower (Brassica oleracea L. var. Botrytis DC). Euphytica 24, 173–176. https://doi.org/10.1007/BF00147182 Davis, E.H., Morgan, J.T., 2015. Collards: A Southern Tradition from Seed to Table. University of Alabama Press. Dawson, J.C., Healy, G.K., 2018. Flavour Evaluation for Plant Breeders, in: Plant Breeding Reviews. John Wiley & Sons, Ltd, pp. 215–261. https://doi.org/10.1002/9781119414735.ch5 Delarue, J., 2015. 1 - The use of rapid sensory methods in R&D and research: an introduction, in: Rapid Sensory Profiling Techniques, Woodhead Publishing Series in Food Science, Technology and Nutrition. Woodhead Publishing, pp. 3–25. https://doi.org/10.1533/9781782422587.1.3 Despommier, D., 2013. Farming up the city: The rise of urban vertical farms. Trends in biotechnology 31, 388–9. https://doi.org/10.1016/j.tibtech.2013.03.008 Dickson, M.H., Lee, C.Y., Blamble, A.E., 1988. Orange-curd high carotene cauliflower inbreds, NY 156, NY 163, and NY 165. HortScience 23, 778–779. Dickson, M.H., Wallace, D.H., 1986. Cabbage breeding, in: Bassett, M.J. (Ed.), Breeding Vegetable Crops. Avi Publishing Company, Inc., Westport, Conneticut, pp. 395–422. Diederichsen, A., 2001. Cruciferae: Brassica, in: Hanelt, P., Büttner, R., Mansfeld, R. (Eds.), Mansfeld’s Encyclopedia of Agricultural and Horticultural Crops (except Ornamentals). Springer, Berlin, pp. 1435–1446. Dinehart, M.E., Hayes, J.E., Bartoshuk, L.M., Lanier, S.L., Duffy, V.B., 2006. Bitter taste markers explain variability in vegetable sweetness, bitterness, and intake. Physiology & Behavior 87, 304–313. https://doi.org/10.1016/j.physbeh.2005.10.018 Dixon, G.R., 2007. Vegetable Brassicas and Related Crucifers. CABI. Doorn, H.E. van, Kruk, G.C. van der, Holst, G.-J. van, Raaijmakers‐Ruijs, N.C.M.E., Postma, E., Groeneweg, B., Jongen, W.H.F., 1998. The glucosinolates sinigrin and progoitrin are important determinants for taste preference and bitterness of Brussels sprouts. Journal of the Science of Food and Agriculture 78, 30–38. https://doi.org/10.1002/(SICI)1097-0010(199809)78:1<30::AID- JSFA79>3.0.CO;2-N Doorn, J.E. van, 1999. Development of vegetables with improved consumer quality : a case study in Brussels sprouts (phd). S.n., S.l. Earle, M.D., 2001. Food product development. Woodhead Publishing, Cambridge, England.

36

Eggink, P.M., Maliepaard, C., Tikunov, Y., Haanstra, J.P.W., Bovy, A.G., Visser, R.G.F., 2012. A taste of sweet pepper: Volatile and non-volatile chemical composition of fresh sweet pepper (Capsicum annuum) in relation to sensory evaluation of taste. Food Chemistry 132, 301–310. https://doi.org/10.1016/j.foodchem.2011.10.081 Eigenbrod, C., Gruda, N., 2015. Urban vegetable for food security in cities. A review. Agron. Sustain. Dev. 35, 483–498. https://doi.org/10.1007/s13593-014-0273-y Eigenbrode, S.D., Stoner, K.A., Shelton, A.M., Kain, W.C., 1991. Characteristics of Glossy Leaf Waxes Associated with Resistance to Diamondback Moth (Lepidoptera: Plutellidae) in Brassica oleracea. J Econ Entomol 84, 1609– 1618. https://doi.org/10.1093/jee/84.5.1609 Faragó, D., Sass, L., Valkai, I., Andrási, N., Szabados, L., 2018. PlantSize Offers an Affordable, Non-destructive Method to Measure Plant Size and Color in Vitro. Front. Plant Sci. 9. https://doi.org/10.3389/fpls.2018.00219 Farnham, M.W., Wilson, P.E., Stephenson, K.K., Fahey, J.W., 2004. Genetic and environmental effects on glucosinolate content and chemoprotective potency of broccoli. Plant Breeding 123, 60–65. https://doi.org/10.1046/j.0179- 9541.2003.00912.x Fehr, W.R., 1987. Principles of cultivar development. Volume 1. Theory and technique. Macmillan, New York. Fenwick, G.R., Griffiths, N.M., Heaney, R.K., 1983. Bitterness in brussels sprouts (Brassica oleracea L. var. gemmifera): The role of glucosinolates and their breakdown products. Journal of the Science of Food and Agriculture 34, 73– 80. https://doi.org/10.1002/jsfa.2740340111 Folta, K.M., Klee, H.J., 2016. Sensory sacrifices when we mass-produce mass produce. Hortic Res 3, 16032. https://doi.org/10.1038/hortres.2016.32 Fuller, G.W., 2011. New food product development: from concept to marketplace, 3rd ed. ed. CRC Press, Boca Raton, FL. Gepts, P., 2002. A Comparison between Crop Domestication, Classical Plant Breeding, and Genetic Engineering. Crop Science 42, 1780–1790. https://doi.org/10.2135/cropsci2002.1780 Gilbert, J.L., Guthart, M.J., Gezan, S.A., Carvalho, M.P. de, Schwieterman, M.L., Colquhoun, T.A., Bartoshuk, L.M., Sims, C.A., Clark, D.G., Olmstead, J.W., 2015. Identifying Breeding Priorities for Blueberry Flavor Using Biochemical, Sensory, and Genotype by Environment Analyses. PLOS ONE 10, e0138494. https://doi.org/10.1371/journal.pone.0138494 Gómez-Campo, C., Prakash, S., 1999. 2 Origin and domestication, in: Gómez-Campo, C. (Ed.), Developments in Plant Genetics and Breeding, Biology of Brassica Coenospecies. Elsevier, pp. 33–58. https://doi.org/10.1016/S0168- 7972(99)80003-6 Gould, K.S., 2004. Nature’s Swiss Army Knife: The Diverse Protective Roles of Anthocyanins in Leaves. J Biomed Biotechnol 2004, 314–320. https://doi.org/10.1155/S1110724304406147 Green, B.G., Lim, J., Osterhoff, F., Blacher, K., Nachtigal, D., 2010. Taste mixture interactions: Suppression, additivity, and the predominance of sweetness.

37

Physiology & Behavior 101, 731–737. https://doi.org/10.1016/j.physbeh.2010.08.013 Guo, N., Han, S., Zong, M., Wang, G., Zheng, S., Liu, F., 2019. Identification and differential expression analysis of anthocyanin biosynthetic genes in leaf color variants of ornamental kale. BMC Genomics 20, 1–17. https://doi.org/10.1186/s12864-019-5910-z Guzman, I., Yousef, G.G., Brown, A.F., 2012. Simultaneous Extraction and Quantitation of Carotenoids, Chlorophylls, and Tocopherols in Brassica Vegetables. J. Agric. Food Chem. 60, 7238–7244. https://doi.org/10.1021/jf302475d Hammer, K., Gladis, T., Laghetti, G., Pignone, D., 2013. The wild and the grown - Remarks about the botatical classification of Brassica. Acta Hortic. 49–59. https://doi.org/10.17660/ActaHortic.2013.1005.2 Hampson, C.R., Quamme, H.A., Hall, J.W., MacDonald, R.A., King, M.C., Cliff, M.A., 2000. Sensory evaluation as a selection tool in apple breeding. Euphytica 111, 79–90. https://doi.org/10.1023/A:1003769304778 Hatier, J.-H.B., Gould, K.S., 2009. Anthocyanin Function in Vegetative Organs, in: Winefield, C., Davies, K., Gould, K. (Eds.), Anthocyanins: Biosynthesis, Functions, and Applications. Springer, New York, NY, pp. 1–19. https://doi.org/10.1007/978-0-387-77335-3_1 HHS & USDA, 2015. 2015-2020 Dietary Guidelines for Americans, 8th Edition [WWW Document]. URL https://health.gov/our-work/food-nutrition/2015- 2020-dietary-guidelines/guidelines/ (accessed 4.2.20). Higgins, M.J., Hayes, J.E., 2019. Learned color taste associations in a repeated brief exposure paradigm. Food Quality and Preference 71, 354–365. https://doi.org/10.1016/j.foodqual.2018.08.013 Hopkins, R.J., van Dam, N.M., van Loon, J.J.A., 2009. Role of glucosinolates in insect-plant relationships and multitrophic interactions. Annu. Rev. Entomol. 54, 57–83. https://doi.org/10.1146/annurev.ento.54.110807.090623 Hultengren, R., Glos, M., Mazourek, M., 2016. Breeding Research and Education Needs Assessment for Organic Vegetable Growers in the Northeast. Janke, R.R., Altamimi, M.E., Khan, M., 2017. The Use of High Tunnels to Produce Fruit and Vegetable Crops in North America. Agricultural Sciences 08, 692. https://doi.org/10.4236/as.2017.87052 Jeschke, V., Kearney, E.E., Schramm, K., Kunert, G., Shekhov, A., Gershenzon, J., Vassão, D.G., 2017. How Glucosinolates Affect Generalist Lepidopteran Larvae: Growth, Development and Glucosinolate Metabolism. Front. Plant Sci. 8. https://doi.org/10.3389/fpls.2017.01995 Johnson, E.J., Hammond, B.R., Yeum, K.-J., Qin, J., Wang, X.D., Castaneda, C., Snodderly, D.M., Russell, R.M., 2000. Relation among serum and tissue concentrations of lutein and zeaxanthin and macular pigment density. Am J Clin Nutr 71, 1555–1562. https://doi.org/10.1093/ajcn/71.6.1555 Jones, M.G., 2008. Chapter 5. Formation of Vegetable Flavour, in: Brückner, B., Wyllie, S.G. (Eds.), Fruit and Vegetable Flavour - Recent Advances and Future Prospects. Woodhead Publishing, Cambridge, England.

38

Jongen, W.M.F., 2000. Food Supply Chains: From Productivity Toward Quality, in: Shewfelt, R.L., Brückner, B. (Eds.), Fruit and Vegetable Quality: An Integrated View. Technomic Pub. Co., Lancaster, PA, pp. 3–20. https://doi.org/10.1201/9781482293937-8 Kader, A.A., 2002. Postharvest Technology of Horticultural Crops. University of California Agriculture and Natural Resources. Khachik, F., Beecher, G.R., Whittaker, N.F., 1986. Separation, identification, and quantification of the major carotenoid and chlorophyll constituents in extracts of several green vegetables by liquid chromatography. J. Agric. Food Chem. 34, 603–616. https://doi.org/10.1021/jf00070a006 Kim, C.-K., Seol, Y.-J., Perumal, S., Lee, J., Waminal, N.E., Jayakodi, M., Lee, S.-C., Jin, S., Choi, B.-S., Yu, Y., Ko, H.-C., Choi, J.-W., Ryu, K.-Y., Sohn, S.-H., Parkin, I., Yang, T.-J., 2018. Re-exploration of U’s Triangle Brassica Species Based on Chloroplast Genomes and 45S nrDNA Sequences. Scientific Reports 8, 1–11. https://doi.org/10.1038/s41598-018-25585-4 Klee, H.J., Tieman, D.M., 2013. Genetic challenges of flavor improvement in tomato. Trends in Genetics 29, 257–262. https://doi.org/10.1016/j.tig.2012.12.003 Koes, R., Verweij, W., Quattrocchio, F., 2005. Flavonoids: a colorful model for the regulation and evolution of biochemical pathways. Trends in Plant Science, Special Issue: Plant model systems 10, 236–242. https://doi.org/10.1016/j.tplants.2005.03.002 Kopsell, D.A., Kopsell, D.E., Lefsrud, M.G., Curran-Celentano, J., Dukach, L.E., 2004. Variation in Lutein, β-carotene, and Chlorophyll Concentrations among Brassica oleracea Cultigens and Seasons. HortScience 39, 361–364. https://doi.org/10.21273/HORTSCI.39.2.361 Krinsky, N.I., Johnson, E.J., 2005. Carotenoid actions and their relation to health and disease. Molecular Aspects of Medicine 26, 459–516. https://doi.org/10.1016/j.mam.2005.10.001 Kurilich, A.C., Tsau, G.J., Brown, A., Howard, L., Klein, B.P., Jeffery, E.H., Kushad, M., Wallig, M.A., Juvik, J.A., 1999. Carotene, Tocopherol, and Ascorbate Contents in Subspecies of Brassica oleracea. J. Agric. Food Chem. 47, 1576– 1581. https://doi.org/10.1021/jf9810158 Kushad, M.M., Brown, A.F., Kurilich, A.C., Juvik, J.A., Klein, B.P., Wallig, M.A., Jeffery, E.H., 1999. Variation of Glucosinolates in Vegetable Crops of Brassica oleracea. J. Agric. Food Chem. 47, 1541–1548. https://doi.org/10.1021/jf980985s Lado, J., Vicente, E., Manzzioni, A., Ares, G., 2010. Application of a check-all-that- apply question for the evaluation of strawberry cultivars from a breeding program. J. Sci. Food Agric. 90, 2268–2275. https://doi.org/10.1002/jsfa.4081 Lan, T.-H., Paterson, A.H., 2001. Comparative mapping of QTLs determining the plant size of Brassica oleracea. Theor Appl Genet 103, 383–397. https://doi.org/10.1007/s001220100615 Landi, M., Tattini, M., Gould, K.S., 2015. Multiple functional roles of anthocyanins in plant-environment interactions. Environmental and Experimental Botany,

39

Functional roles of secondary metabolites in plant-environment interactions 119, 4–17. https://doi.org/10.1016/j.envexpbot.2015.05.012 Lawless, H.T., Heymann, H., 2010a. Physiological and Psychological Foundations of Sensory Function, in: Sensory Evaluation of Food, Food Science Text Series. Springer, New York, NY, pp. 19–56. https://doi.org/10.1007/978-1-4419- 6488-5_2 Lawless, H.T., Heymann, H., 2010b. Descriptive Analysis, in: Sensory Evaluation of Food, Food Science Text Series. Springer, New York, NY, pp. 227–257. https://doi.org/10.1007/978-1-4419-6488-5_10 Lawless, H.T., Heymann, H., 2010c. Acceptance Testing, in: Sensory Evaluation of Food, Food Science Text Series. Springer, New York, NY, pp. 325–347. https://doi.org/10.1007/978-1-4419-6488-5_14 Lawless, H.T., Heymann, H., 2010d. Preference Testing, in: Sensory Evaluation of Food, Food Science Text Series. Springer, New York, NY, pp. 303–324. https://doi.org/10.1007/978-1-4419-6488-5_13 Lawless, H.T., Heymann, H., 2010e. Texture Evaluation, in: Sensory Evaluation of Food, Food Science Text Series. Springer, New York, NY, pp. 259–281. https://doi.org/10.1007/978-1-4419-6488-5_11 Lefsrud, M., Kopsell, D., Wenzel, A., Sheehan, J., 2007. Changes in kale (Brassica oleracea L. var. acephala) carotenoid and chlorophyll pigment concentrations during leaf ontogeny. Scientia Horticulturae 112, 136–141. https://doi.org/10.1016/j.scienta.2006.12.026 Lev-Yadun, S., Gould, K.S., 2009. Role of Anthocyanins in Plant Defence, in: Winefield, C., Davies, K., Gould, K. (Eds.), Anthocyanins: Biosynthesis, Functions, and Applications. Springer, New York, NY, pp. 22–28. https://doi.org/10.1007/978-0-387-77335-3_2 Li, L., Lu, S., O’Halloran, D.M., Garvin, D.F., Vrebalov, J., 2003. High-resolution genetic and physical mapping of the cauliflower high-β-carotene gene Or (Orange). Mol Genet Genomics 270, 132–138. https://doi.org/10.1007/s00438- 003-0904-5 Lila, M.A., Burton-Freeman, B., Grace, M., Kalt, W., 2016. Unraveling Anthocyanin Bioavailability for Human Health. Annual Review of Food Science and Technology 7, 375–393. https://doi.org/10.1146/annurev-food-041715-033346 Liu, X., Gao, B., Han, F., Fang, Z., Yang, L., Zhuang, M., Lv, H., Liu, Y., Li, Zhan- sheng, Cai, C., Yu, H., Li, Zhi-yuan, Zhang, Y., 2017. Genetics and fine mapping of a purple leaf gene, BoPr, in ornamental kale (Brassica oleracea L. var. acephala). BMC Genomics 18, 230. https://doi.org/10.1186/s12864-017- 3613-x Low, S., Adalja, A., Beaulieu, E., Key, N., Martinez, S., Melton, A., Perez, A., Ralston, K., Stewart, H., Suttles, S., Jablonski, B., 2015. Trends in U.S. Local and Regional Food Systems: A Report to Congress. Articles and Chapters. Lukyanenko, A.N., 1991. Breeding Tomato for Mechanized Harvesting, in: Genetic Improvement of Tomato, Monographs on Theoretical and Applied Genetics. Springer, Berlin, Heidelberg, pp. 213–230. https://doi.org/10.1007/978-3-642- 84275-7_17

40

Maggioni, L., von Bothmer, R., Poulsen, G., Branca, F., 2010. Origin and Domestication of Cole Crops (Brassica oleracea L.): Linguistic and Literary Considerations1. Econ Bot 64, 109–123. https://doi.org/10.1007/s12231-010- 9115-2 Maggioni, L., von Bothmer, R., Poulsen, G., Lipman, E., 2018. Domestication, diversity and use of Brassica oleracea L., based on ancient Greek and Latin texts. Genet Resour Crop Evol 65, 137–159. https://doi.org/10.1007/s10722- 017-0516-2 Mazza, G., Brouillard, R., 1987. Recent developments in the stabilization of anthocyanins in food products. Food Chemistry 25, 207–225. https://doi.org/10.1016/0308-8146(87)90147-6 Moreira, L.F., Teixeira, N.C., Santos, N.A., Valim, J.O.S., Maurício, R.M., Guedes, R.N.C., Oliveira, M.G.A., Campos, W.G., 2016. Diamondback moth performance and preference for leaves of Brassica oleracea of different ages and strata. Journal of Applied Entomology 140, 627–635. https://doi.org/10.1111/jen.12294 Moskowitz, H.R., Moskowitz, H.R., Muñoz, A.M., Gacula, M.C., 2004a. Contrasting R&D, Sensory Science, and Marketing Research Approaches, in: Viewpoints and Controversies in Sensory Science and Consumer Product Testing. Food & Nutrition Press, Inc., pp. 77–95. https://doi.org/10.1002/9780470385128.ch4 Moskowitz, H.R., Moskowitz, H.R., Muñoz, A.M., Gacula, M.C., 2004b. Descriptive Panels/Experts versus Consumers, in: Viewpoints and Controversies in Sensory Science and Consumer Product Testing. Food & Nutrition Press, Inc., pp. 109–123. https://doi.org/10.1002/9780470385128.ch7 Muñoz, A., Civille, G., 1992. The Spectrum Descriptive Analysis Method. Manual on Descriptive Analysis Testing for Sensory Evaluation 22. https://doi.org/10.1520/MNL10524M Muñoz, A., Szczesniak, A., Einstein, M., Schwartz, N., 1992. The Texture Profile, Manual on Descriptive Analysis Testing for Sensory Evaluation. ASTM International, West Conshohocken, PA. https://doi.org/10.1520/MNL10525M Myers, J.R., 2014. Status of Cytoplasmic Male Sterility (CMS) in Vegetable Brassicas. Presented at the 7th Organic Seed Growers Conference, Corvallis, OR, pp. 134–143. Nagaharu, U., 1935. Genome analysis in Brassica with special reference to the experimental formation of B. napus and peculiar mode of fertilization. Japanese Journal of Botany 7, 389–452. Nakanishi, T., Hinata, K., 1973. An effective time for CO2 gas treatment in overcoming self-incompatibility in Brassica. Plant Cell Physiol 14, 873–879. https://doi.org/10.1093/oxfordjournals.pcp.a074929 Ockendon, D.J., 1977. Rare self-incompatibility alleles in a purple cultivar of Brussels sprouts. Heredity 39, 149–152. https://doi.org/10.1038/hdy.1977.51 Olmstead, M.A., Gilbert, J.L., Colquhoun, T.A., Clark, D.G., Kluson, R., Moskowitz, H.R., 2015. In Pursuit of the Perfect Peach: Consumer-assisted Selection of Peach Fruit Traits. HortScience 50, 1202–1212.

41

Ong, P.K.C., Liu, S.Q., 2010. Flavor and Sensory Characteristics of Vegetables, in: Ph.D, N.K.S. (Ed.), Handbook of Vegetables and Vegetable Processing. Wiley-Blackwell, pp. 59–82. https://doi.org/10.1002/9780470958346.ch3 Park, S., Elless, M.P., Park, J., Jenkins, A., Lim, W., Chambers IV, E., Hirschi, K.D., 2009. Sensory analysis of calcium-biofortified lettuce. Plant Biotechnology Journal 7, 106–117. https://doi.org/10.1111/j.1467-7652.2008.00379.x Pecher, S., Oppen, M., 2000. Consumer Preferences and Breeding Goals, in: Shewfelt, R.L., Brückner, B. (Eds.), Fruit and Vegetable Quality: An Integrated View. Technomic Pub. Co., Lancaster, PA, pp. 43–62. https://doi.org/10.1201/9781482293937-10 Persley, G.J., Anthony, V.M. (Eds.), 2017. The business of plant breeding: market-led approaches to new variety design in Africa. CABI, Wallingford, Oxfordshire. Pierce, J., Halpern, B.P., 1996. Orthonasal and Retronasal Odorant Identification Based upon Vapor Phase Input from Common Substances. Chem Senses 21, 529–543. https://doi.org/10.1093/chemse/21.5.529 Pojer, E., Mattivi, F., Johnson, D., Stockley, C.S., 2013. The Case for Anthocyanin Consumption to Promote Human Health: A Review. Comprehensive Reviews in Food Science and Food Safety 12, 483–508. https://doi.org/10.1111/1541- 4337.12024 Rask, L., Andréasson, E., Ekbom, B., Eriksson, S., Pontoppidan, B., Meijer, J., 2000. Myrosinase: gene family evolution and herbivore defense in Brassicaceae. Plant Mol Biol 42, 93–114. https://doi.org/10.1023/A:1006380021658 Roberts, W., 2019. Vegetables - US - August 2019. Mintel Academic. Rocha, M. de C., Deliza, R., Corrêa, F.M., Carmo, M.G.F. do, Abboud, A.C.S., 2013. A study to guide breeding of new cultivars of organic cherry tomato following a consumer-driven approach. Food Research International 51, 265–273. https://doi.org/10.1016/j.foodres.2012.12.019 Rodríguez-Burruezo, S., Prohens, J., RosellÓ, J., Nuez, F., 2005. “Heirloom” varieties as sources of variation for the improvement of fruit quality in greenhouse- grown tomatoes. The Journal of Horticultural Science and Biotechnology 80, 453–460. https://doi.org/10.1080/14620316.2005.11511959 Rosa, E.A.S., Heaney, R.K., Rego, F.C., Fenwick, G.R., 1994. The variation of glucosinolate concentration during a single day in young plants of Brassica oleracea var Acephala and Capitata. Journal of the Science of Food and Agriculture 66, 457–463. https://doi.org/10.1002/jsfa.2740660406 Sampson, D.R., 1967. New light on the complexities of anthocyanin inheritance in Brassica oleracea. Canadian J. Genetics Cytology 9, 352–358. Sánchez, M.-T., Entrenas, J.-A., Torres, I., Vega, M., Pérez-Marín, D., 2018. Monitoring texture and other quality parameters in spinach plants using NIR spectroscopy. Computers and Electronics in Agriculture 155, 446–452. https://doi.org/10.1016/j.compag.2018.11.004 Sauer, J.D., 1993. Historical geography of crop plants: a select roster. CRC Press, Boca Raton, FL. Scott, K., 2004. The sweet and the bitter of mammalian taste. Current Opinion in Neurobiology 14, 423–427. https://doi.org/10.1016/j.conb.2004.06.003

42

Sebastian, R.L., Kearsey, M.J., King, G.J., 2002. Identification of quantitative trait loci controlling developmental characteristics of Brassica oleracea L. Theor Appl Genet 104, 601–609. https://doi.org/10.1007/s001220100743 Shewfelt, R.L., 2000. Fruit and Vegetable Quality, in: Shewfelt, R.L., Brückner, B. (Eds.), Fruit and Vegetable Quality: An Integrated View. Technomic Pub. Co., Lancaster, PA, pp. 144–157. https://doi.org/10.1201/9781482293937-17 Simon, P.W., Peterson, C.E., Lindsay, R.C., 1981. The Improvement of Flavor in a Program of Carrot Genetics and Breeding, in: Quality of Selected Fruits and Vegetables of North America, ACS Symposium Series. American Chemical Society, pp. 109–118. https://doi.org/10.1021/bk-1981-0170.ch009 Small, D.M., Prescott, J., 2005. Odor/taste integration and the perception of flavor. Exp Brain Res 166, 345–357. https://doi.org/10.1007/s00221-005-2376-9 Snogerup, S., 1980. The wild forms of the Brassica oleracea group (2n = 18) and their possible relations to the cultivated ones. Brassica crops and wild allies. [I]. 121–132. Socquet-Juglard, D., Bennett, A.A., Manns, D.C., Mansfield, A.K., Robbins, R.J., Collins, T.M., Griffiths, P.D., 2016. Effects of Growth Temperature and Postharvest Cooling on Anthocyanin Profiles in Juvenile and Mature Brassica oleracea. J. Agric. Food Chem. 64, 1484–1493. https://doi.org/10.1021/acs.jafc.5b05309 Sønderby, I.E., Geu-Flores, F., Halkier, B.A., 2010. Biosynthesis of glucosinolates – gene discovery and beyond. Trends in Plant Science 15, 283–290. https://doi.org/10.1016/j.tplants.2010.02.005 Stansell, Z., Farnham, M., Björkman, T., 2019. Complex Horticultural Quality Traits in Broccoli Are Illuminated by Evaluation of the Immortal BolTBDH Mapping Population. Front. Plant Sci. 10. https://doi.org/10.3389/fpls.2019.01104 Steenkamp, J.-B.E.M., 1990. Conceptual model of the quality perception process. Journal of Business Research 21, 309–333. https://doi.org/10.1016/0148- 2963(90)90019-A Stone, H., 1992. Quantitative Descriptive Analysis (QDA), in: Hootman, R. (Ed.), Manual on Descriptive Analysis Testing for Sensory Evaluation. ASTM International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken, PA 19428-2959, pp. 15–7. https://doi.org/10.1520/MNL10523M Stone, H., Bleibaum, R.N., Thomas, H.A., 2012. Chapter 1 - Introduction to Sensory Evaluation, in: Sensory Evaluation Practices (Fourth Edition), Food Science and Technology. Academic Press, San Diego, pp. 1–21. https://doi.org/10.1016/B978-0-12-382086-0.00001-7 Stone, H., Sidel, J., Oliver, S., Woolsey, A., Singleton, R.C., 1974. Sensory evaluation by quantitative descriptive analysis. Food technology. Stoner, K.A., 1990. Glossy Leaf Wax and Plant Resistance to Insects in Brassica oleracea Under Natural Infestation. Environ Entomol 19, 730–739. https://doi.org/10.1093/ee/19.3.730 Sun, T., Li, L., 2020. Toward the ‘golden’ era: The status in uncovering the regulatory control of carotenoid accumulation in plants. Plant Science 290, 110331. https://doi.org/10.1016/j.plantsci.2019.110331

43

Szczesniak, A.S., 2002. Texture is a sensory property. Food Quality and Preference 13, 215–225. https://doi.org/10.1016/S0950-3293(01)00039-8 Tanaka, Y., Sasaki, N., Ohmiya, A., 2008. Biosynthesis of plant pigments: anthocyanins, betalains and carotenoids. The Plant Journal 54, 733–749. https://doi.org/10.1111/j.1365-313X.2008.03447.x The Packer, 2013. Better flavor’s the gold ring for melon breeders [WWW Document]. The Packer. URL https://www.thepacker.com/article/better- flavors-gold-ring-melon-breeders (accessed 4.12.20). Thompson, K.F., Taylor, J.P., 1965. Linkage between S alleles and pigmentation genes in marrow-stem kale, brassica oleracea. Euphytica 14, 285–292. https://doi.org/10.1007/BF00149513 Tian, M., Xu, X., Liu, Y., Xie, L., Pan, S., 2016. Effect of Se treatment on glucosinolate metabolism and health-promoting compounds in the broccoli sprouts of three cultivars. Food Chem 190, 374–380. https://doi.org/10.1016/j.foodchem.2015.05.098 Tieman, D., Zhu, G., Resende, M.F.R., Lin, T., Nguyen, C., Bies, D., Rambla, J.L., Beltran, K.S.O., Taylor, M., Zhang, B., Ikeda, H., Liu, Z., Fisher, J., Zemach, I., Monforte, A., Zamir, D., Granell, A., Kirst, M., Huang, S., Klee, H., 2017. A chemical genetic roadmap to improved tomato flavor. Science 355, 391– 394. https://doi.org/10.1126/science.aal1556 Traka, M., Mithen, R., 2009. Glucosinolates, isothiocyanates and human health. Phytochem Rev 8, 269–282. https://doi.org/10.1007/s11101-008-9103-7 Traka, M.H., Saha, S., Huseby, S., Kopriva, S., Walley, P.G., Barker, G.C., Moore, J., Mero, G., Bosch, F. van den, Constant, H., Kelly, L., Schepers, H., Boddupalli, S., Mithen, R.F., 2013. Genetic regulation of glucoraphanin accumulation in Beneforté® broccoli. New Phytologist 198, 1085–1095. https://doi.org/10.1111/nph.12232 Tsukaya, H., 2006. Mechanism of leaf-shape determination. Annual review of plant biology. Ulrich, D., Olbricht, K., 2011. Fruit Organoleptic Properties and Potential for Their Genetic Improvement, in: Jenks, tthew A., Bebeli, P.J. (Eds.), Breeding for Fruit Quality. John Wiley & Sons, Inc., pp. 39–59. https://doi.org/10.1002/9780470959350.ch2 USDA-NASS, 2019. 2017 Census Volume 1, Chapter 2: State Level Data [WWW Document]. Census of Agriculture. URL www.nass.usda.gov/AgCensus (accessed 3.24.20). USDA-NASS, 2015. 2015 Local Food Marketing Practices Survey [WWW Document]. USDA - NASS, Census of Agriculture - Local Food Marketing Practices. URL https://www.nass.usda.gov/Publications/AgCensus/2012/Online_Resources/Lo cal_Food/index.php (accessed 4.13.20). Vaughn, T.T., Hoy, C.W., 1993. Effects of Leaf Age, Injury, Morphology, and Cultivars on Feeding Behavior of Phylloteta cruciferae (Coleoptera: Chrysomelidae). Environ Entomol 22, 418–424. https://doi.org/10.1093/ee/22.2.418

44

Velasco, P., Cartea, M.E., González, C., Vilar, M., Ordás, A., 2007. Factors Affecting the Glucosinolate Content of Kale (Brassica oleracea acephala Group). J. Agric. Food Chem. 55, 955–962. https://doi.org/10.1021/jf0624897 Voorrips, R.E., Steenhuis-Broers, G., Tiemens-Hulscher, M., van Bueren, E.T.L., 2008. Plant traits associated with resistance to Thrips tabaci in cabbage (Brassica oleracea var capitata). Euphytica 163, 409. https://doi.org/10.1007/s10681-008-9704-7 Wadhera, D., Capaldi-Phillips, E.D., 2014. A review of visual cues associated with food on food acceptance and consumption. Eating Behaviors 15, 132–143. https://doi.org/10.1016/j.eatbeh.2013.11.003 Walley, P.G., Buchanan-Wollaston, V., 2011. Brassicas, in: Terry, L.A. (Ed.), Health- Promoting Properties of Fruits and Vegetables. CABI, Wallingford, Oxfordshire, pp. 74–90. Wang, J., Yue, C., Gallardo, K., McCracken, V., Luby, J., McFerson, J., 2017. What Consumers Are Looking for in Strawberries: Implications from Market Segmentation Analysis. Agribusiness 33, 56–69. https://doi.org/10.1002/agr.21473 Wang, Y.S., Liu, Z.Y., Li, Y.F., Zhang, Y., Yang, X.F., Feng, H., 2013. Identification of sequence-related amplified polymorphism markers linked to the red leaf trait in ornamental kale (Brassica oleracea L. var. acephala). Genet. Mol. Res. 12, 870–877. Weight, C., Parnham, D., Waites, R., 2008. TECHNICAL ADVANCE: LeafAnalyser: a computational method for rapid and large-scale analyses of leaf shape variation. The Plant Journal 53, 578–586. https://doi.org/10.1111/j.1365- 313X.2007.03330.x Wieczorek, M.N., Walczak, M., Skrzypczak-Zielińska, M., Jeleń, H.H., 2018. Bitter taste of Brassica vegetables: The role of genetic factors, receptors, isothiocyanates, glucosinolates, and flavor context. Critical Reviews in Food Science and Nutrition 58, 3130–3140. https://doi.org/10.1080/10408398.2017.1353478 Wilkinson, C., Dijksterhuis, G.B., Minekus, M., 2000. From food structure to texture. Trends in Food Science & Technology 11, 442–450. https://doi.org/10.1016/S0924-2244(01)00033-4 Wu, X., Beecher, G.R., Holden, J.M., Haytowitz, D.B., Gebhardt, S.E., Prior, R.L., 2006. Concentrations of Anthocyanins in Common Foods in the United States and Estimation of Normal Consumption. J. Agric. Food Chem. 54, 4069–4075. https://doi.org/10.1021/jf060300l Yamagishi, H., Bhat, S.R., 2014. Cytoplasmic male sterility in Brassicaceae crops. Breed Sci 64, 38–47. https://doi.org/10.1270/jsbbs.64.38 Yuan, H., Zhang, J., Nageswaran, D., Li, L., 2015. Carotenoid metabolism and regulation in horticultural crops. Horticulture Research 2, 1–11. https://doi.org/10.1038/hortres.2015.36 Yuan, Y., Chiu, L.-W., Li, L., 2009. Transcriptional regulation of anthocyanin biosynthesis in red cabbage. Planta 230, 1141. https://doi.org/10.1007/s00425- 009-1013-4

45

Zampini, M., Sanabria, D., Phillips, N., Spence, C., 2007. The multisensory perception of flavor: Assessing the influence of color cues on flavor discrimination responses. Food Quality and Preference 18, 975–984. https://doi.org/10.1016/j.foodqual.2007.04.001 Zhang, B., Hu, Z., Zhang, Y., Li, Y., Zhou, S., Chen, G., 2012. A putative functional MYB transcription factor induced by low temperature regulates anthocyanin biosynthesis in purple kale (Brassica Oleracea var. acephala f. tricolor). Plant Cell Rep 31, 281–289. https://doi.org/10.1007/s00299-011-1162-3 Zhang, Y., Hoon, M.A., Chandrashekar, J., Mueller, K.L., Cook, B., Wu, D., Zuker, C.S., Ryba, N.J.P., 2003. Coding of Sweet, Bitter, and Umami Tastes: Different Receptor Cells Sharing Similar Signaling Pathways. Cell 112, 293– 301. https://doi.org/10.1016/S0092-8674(03)00071-0 Zhu, P., Cheng, M., Feng, X., Xiong, Y., Liu, C., Kang, Y., 2016. Mapping of Pi, a gene conferring pink leaf in ornamental kale (Brassica oleracea L. var. acephala DC). Euphytica 207, 377–385. https://doi.org/10.1007/s10681-015- 1555-4

46

Chapter 2

BRIDGING SENSORY EVALUATION AND CONSUMER RESEARCH FOR

STRATEGIC LEAFY BRASSICA (BRASSICA OLERACEA) IMPROVEMENT

(Formatted for publication in the Journal of Food Science)

Abstract

Plant breeders working with new or underrepresented horticultural crops often have minimal sensory resources available to aid in the breeding and selection of new varieties. Kale (Brassica oleracea var. acephala) is a recently popularized horticultural crop in Western markets, however, plant breeding programs have little knowledge regarding the underlying sensory characteristics motivating this trend. We employed a multilayered, sensory‐driven approach to understand the inherent consumer values, sensory attributes, and consumer preferences for kale types currently available on the market and novel genotypes from the Cornell AgriTech vegetable breeding program. Underlying consumer values related to storability, health and wellbeing, and sensory characteristics were identified through Qualitative Multivariate

Analysis (QMA). A trained descriptive panel developed a lexicon of 44 sensory attributes common within kale germplasm, 21 of which exhibited significant differences among the 15 tested kale genotypes. Following a consumer test, four clusters of kale consumers were identified with agglomerative hierarchical clustering

(AHC) and external preference mapping was used to connect consumer hedonic scores with descriptive data. Consumers demonstrated a preference for familiar kale types

47

(that is, curly types), while new test hybrids scored favorably within flavor and appearance modalities. Preference mapping highlighted the utility of plant breeding in developing products to expand the existing sensory space. This work provides important resources for horticultural crop selection efforts, and it serves as a strategic model for breeding programs working with new or unfamiliar traits.

Introduction

The health and nutritional benefits of Brassicaceae (Cruciferous) vegetables have propelled their popularity in Western markets and generated an interest in the development of novel cultivars to further expand these markets (Higdon, Delage,

Williams, & Dashwood, 2007; Li, Hullar, Schwarz, & Lampe, 2009; Zhang et al.,

2011). The Brassicaceae plant family exhibits a rich diversity of plant morphotypes resulting from direct selection for seed and vegetative tissue over thousands of years.

Kale (Brassica oleracea var. acephala L.) in particular is a common leafy Brassica vegetable whose recent increase in consumption has generated interest from both fresh and processed markets across the United States (Eigenbrod and Gruda, 2015). As a low calorie food rich in vitamins, minerals, and phytochemicals (Björkman et al.,

2011; Jahangir, Kim, Choi, & Verpoorte, 2009; Kopsell et al., 2003; Thavarajah et al.,

2016), kale has found niche market value in dried chip, powdered/fresh smoothie mixes, and as a primary ingredient in fresh salads (Mintel, 2018). Approximately 19% of consumers purchased fresh kale within the last 12 months, and it has exhibited strong sales in organic grocery outlets (Produce Market Guide, 2018).

Kale represents broad-leaved, non-heading vegetable crops of the Brassica oleracea species, a diverse plant species containing horticulturally relevant vegetable

48

crops such as cabbage (var. capitata), broccoli (var. italica), cauliflower (var. botrytis), and Brussels sprouts (var. gemmifera) (Dixon, 2006). The common term

“kale” is used to describe nearly all non-heading leafy Brassicas; it encompasses not only market classes within Brassica oleracea var. acephala (i.e. green curly, red curly,

Tuscan/Lacinato/dinosaur), but it is also describes B. oleracea var. alboglabra (Gai lan or Chinese kale) some B. oleracea var. viridis (collard) varieties and related species, Brassica napus var. pabularis (Russo-Siberian or rape kale) (Diederichsen,

2001; Hammer, Gladis, Laghetti, & Pignone, 2013). The two most predominant leafy market classes within the U.S. are B. oleracea var. acephala (kale) and B. oleracea var. viridis (collard) (USDA-NASS, 2019). Both market classes exhibit robust nutritional and phytochemical profiles, though strong cultural and market divisions appear to distinguish these two botanical varieties. From a plant breeding perspective, however, these two related varieties have the same mating system and are readily hybridized (Gómez-Campo, 1999).

Intraspecific hybridization, coupled with the ability to hybridize with wild species and other cultivated members of the Brassica genus, has encouraged continued diversification of the B. oleracea species in modern plant breeding programs. Today, nearly all plant parts of B. oleracea, including the stem, axillary buds, leaves, and floral buds are harvested and marketed as separate market classes for human consumption. The intraspecific diversity within B. oleracea provides an opportunity for plant breeders to capitalize on divergent plant morphologies and secondary metabolite profiles responsible for distinct flavors and pigmentation profiles in plants

(Jones, 2008; Tanaka, Sasaki, & Ohmiya, 2008) to develop unique genotypes with no

49

prior market exposure. In practice, plant breeding pipelines closely mimic traditional food product development pipelines in that they rely on an iterative and integrated process to identify valuable plant material (Earle, Earle, & Anderson, 2001; Gepts,

2002). While sensory methodologies have been used for screening existing varieties and breeding material in many horticultural crops (e.g. apples (Amyotte, Bowen,

Banks, Rajcan, & Somers, 2017), blueberries (Gilbert et al., 2015, 2014), carrot

(Simon, Peterson, & Lindsay, 1981), lettuce, (Chadwick, Gawthrop, Michelmore,

Wagstaff, & Methven, 2016), strawberries (Civille & Oftedal, 2012; Colquhoun et al.,

2012), tomato (Causse, Buret, Robini, & Verschave, 2003; Tieman et al., 2017), sweetcorn (Azanza, Tadmor, Klein, Rocheford, & Juvik, 1996), etc.), the development of a comprehensive sensory and consumer research program including direct consumer feedback has not been well established in Brassica breeding programs.

There is a high degree of complexity when moving a product from developer to consumer, and an understanding of how new market genotypes affect consumer perception can aid plant breeders during the selection and development of new market classes.

Significant efforts have been dedicated to understanding the flavor-active chemical constituents in broccoli, cabbage, and Brussels sprouts, with a particular emphasis on the volatile breakdown products of glucosinolates (GS) and S- methylcysteine sulfoxide (MCSO) (Jones, 2008; Radovich, 2010). Previous studies have recognized sensory analysis as a direct outlet to measure consumer perception of quality traits within B. oleracea, namely with regard to the effect of secondary metabolites on consumer preference (Baik et al., 2003; Hansen, Laustsen, Olsen, Poll,

50

& Sørensen, 1997; Martens, Martens, & Wold, 1983; Schonhof, Krumbein, &

Brückner, 2004; van Doorn et al., 1999). Several studies have also assessed quality in kale and kale products directly (Albornoz, 2014; Kobori, Huber, Sarantópoulos, &

Rodriguez-Amaya, 2011; Kopsell et al., 2003; Sistrunk, 1980), but few studies to date have integrated sensory analysis in the interpretation of their results. Studies evaluating the effect of cooking methods (Armesto, Gómez-Limia, Carballo, &

Martínez, 2016) and production practices (i.e. nitrogen, sulfur, and frost treatments)

(Groenbaek et al., 2016) are some of the few that have included trained sensory panels, though lexicons were not standardized and only a small number of cultivars were included in each sample set.

The sensory profile of leafy Brassicas has not been well evaluated, and a standardized sensory lexicon for members of the Acephala group does not exist.

Aforementioned studies within B. oleracea used study-specific lexicons or did not provide a detailed attribute list. A standardized lexicon for leafy Brassica vegetables would improve reproducibility and aid in the recognition of true differences between samples (Lawless & Civille, 2013). Talavera-Bianchi et al. (2010) developed a product lexicon for use in the evaluation of fresh leafy vegetables, including four cabbage, one collard, and three unnamed kale varieties, though these samples were part of a product set comprising nearly 30 varieties of leafy vegetables from 12 plant species. Observations from Hongsoongnern and Chambers’ (2008) comprehensive effort to dissect and describe the “green” character may be useful in the description of leafy Brassica genotypes.

Significant phenotypic diversity exists among leafy Brassica breeding

51

materials in the Cornell AgriTech vegetable breeding program, however, it is unknown whether this is enough to affect consumer sensory perception/preference and necessitate the continued breeding and selection for new market classes or genotypes.

In this study, we describe a multi-faceted sensory-driven approach that included

Qualitative Multivariate Analysis (QMA), descriptive analysis, and consumer acceptance testing to evaluate the diversity among current market classes of kale and new testcross hybrids. Home-use tests (HUTs) and a facilitated focus group performed during QMA were designed to understand underlying values and motivators associated with kale consumption, and this project sought to develop a standardized kale lexicon and understand distinguishing characteristics among kale products in a standardized manner that could subsequently be employed by other groups. When coupled with marketing insights and consumer sensory analysis, descriptive analysis affords plant breeders and product developers the ability to understand consumer clustering and opportunities for product development that can fill gaps in current product offerings (Bowen, Blake, Tureček, & Amyotte, 2019; Drake, Lopetcharat, &

Drake, 2009; Jaeger, Rossiter, Wismer, & Harker, 2003; van Kleef, van Trijp, &

Luning, 2006). We hypothesized existing market divisions between kale and collard types would be confirmed by consumer clustering. Further, we anticipated flavor and texture would be important discriminating factors among kale types. The project presented herein was a concerted effort between plant breeders and sensory scientists to understand the sensory landscape of leafy Brassica genotypes and develop a reference for future studies that seek to understand the role of novel horticultural products in the current marketplace from a sensory standpoint.

52

Materials and Methods

Human subject assurance. All procedures involving human subjects were reviewed and approved by the Cornell University Institutional Review Board. In each phase of the study, participants were provided with written consent forms and the opportunity to ask researchers any questions prior to their participation. All recruitment materials, consent forms, product codes, journal entries, researcher notes, and data are available in the Supplementary Materials or Mendeley Data.

Germplasm. Plant materials evaluated throughout this study included publicly available kale cultivars, inbred lines developed by the Cornell AgriTech breeding program (Geneva, NY), and testcross hybrids currently under development within the program (Table 2.1), herein referred to as “genotypes.” With the exception of one

Brassica napus cultivar (‘White Russian’), all plant materials were classified as

Brassica oleracea var. acephala (kale) or Brassica oleracea var. viridis (collard).

Commercially available cultivars were sourced from seed companies in the winter of

2016. Testcross hybrids were developed during the winter of 2016 in Geneva, NY from field-grown, inbred parental lines (selfed to the S4 generation or beyond). Field materials were vernalized for approximately ten weeks in a cool nursery cellar (8˚C) prior to their placement in the Geneva Greenhouse complex (42°52’31.7”N,

77°00’28.0”W) for crossing and seed production in the spring of 2017.

53

Table 2.1. Kale germplasm included in Qualitative Multivariate Analysis (QMA), descriptive analysis (DA), and consumer central location test (CLT) studies. All genotypes represent breeding lines or cultivars within Brassica oleracea, with the exception of 'White Russian' (Brassica napus). Seed Market Study Variety Breeder Description* Photo Source Class Usage Hybrid, Vates-type kale Bejo Zaden, B.V. with slightly greener leaves DA ‘Darkibor F ' Seedway (Warmenhuizen, Green Curly 1 and less curl than other CLT Netherlands) green curly types

Centre for Genetic Germplasm Accession: Germplasm Resources (CGN) CGN11130 (Passport Info: ‘Lage Krul’ Green Curly DA Repository (Wageningen, NLD-Holsel); slightly Netherlands) darker Vates-type leaves

Open-pollinated, German High heirloom with long, narrow High Mowing Seeds ‘Meadowlark’ Mowing Green Curly leaves. Long stems and DA (Wolcott, VT, US) Seeds leaves with smooth, curled margins Centre for Genetic Germplasm Accession: Germplasm Resources (CGN) CGN11152 (Passport Info: ‘Laerkentunge' Green Curly DA Repository (Wageningen, DNK-DAEHNF); long

Netherlands) stems and pointed leaves Selfed seed (S4) from Bejo Zaden, B.V. 'Reflex F ' bred by Bejo Bejo Zaden, 1 ‘Reflex S ' (Warmenhuizen, Green Curly Vegetable Seeds, Vates- DA 4 B.V. Netherlands) type kale highly curled leaf margin Hybrid, Vates-type kale, Bejo Zaden, B.V. Bejo Zaden, significant, fine curl to dark ‘Starbor F ' (Warmenhuizen, Green Curly DA 1 B.V. blue-green leaves, relatively Netherlands) compact plants

Hybrid, Vates-type kale, Bejo Zaden, B.V. Bejo Zaden, slightly taller plant type ‘Winterbor F ' (Warmenhuizen, Green Curly QMA 1 B.V. than ‘Darkibor’ with blue- Netherlands)

green leaves Jagged, flat leaves that are High QMA Wild Garden Seeds thin and representative of ‘White Russian' Mowing Siberian DA (Philomath, OR, US) Siberian types, extra leaflets Seeds CLT are common on leaves

Bejo Zaden, B.V. Hybrid red curly kale, long DA ‘Redbor F ' Seedway (Warmenhuizen, Red Curly internodes and stems, color 1 CLT Netherlands) intensifies with cold

Open-pollinated Lacinato type, blistered and bumpy Tozer Seeds DA ‘Black Magic' Seedway Tuscan leaf surface, but slightly (Pysports, Surrey, UK) CLT more erect plant type with a darker leaf color Classic open-pollinated ‘Nero di Territorial Lacinato type, blistered and Italian Heirloom Tuscan QMA Toscano' Seeds bumpy leaf surface, open palm-like plant structure

Georgia-type hybrid collard Sakata Seed America with slightly shorter stem DA ‘Tiger F ' Seedway Collard 1 (Morgan Hill, CA, US) length and rounder leaves CLT than ‘Top Bunch F ’ 1 Georgia-type hybrid Sakata Seed America ‘Top Bunch F ' Seedway Collard collard; oblong leaves with QMA 1 (Morgan Hill, CA, US) slight savoy along edges

Couve Troncuda kale with Johnny’s Takii & Co., Ltd. Portuguese very large leaves and ‘Beira F ' Selected QMA 1 (Kyoto, Japan) Tronchuda prominent white midribs; Seeds

short, loosely-headed plant

54

Seed Market Study Variety Breeder Description* Photo Source Class Usage Golden-green, flat-leafed Breeding Cornell Univ. Program collard (open-pollinated) ‘Cornell P13-2' Inbred QMA Material (Geneva, NY, US) with slight leaf wrinkling and short plant stature

Golden-green, flat-leafed collard (open-pollinated) Cornell Breeding Cornell Univ. Program DA Inbred with slight wrinkling in ‘Yellow_IL’ Material (Geneva, NY, US) CLT leaves; slightly larger than P13-2 leaves Golden kale with pink vein; Cornell Breeding Cornell Univ. Program slight wave along leaf edges Test Hybrid DA ‘Hyb E' (F1) Material (Geneva, NY, US) Testcross: KP1702xKP1704

Savoy Tuscan-collard leaf Cornell Breeding Cornell Univ. Program with more oval shape than DA Test Hybrid ‘Hyb I’ (F1) Material (Geneva, NY, US) traditional Tuscan CLT Testcross: KP1703xKP4

Jagged-leaf broccoli x DA Cornell Breeding Cornell Univ. Program Test Hybrid Tuscan kale ‘Hyb K1’ (F ) Material (Geneva, NY, US) 1 Testcross: KP1703xP1A CLT

Soft red curly kale; Cornell Breeding Cornell Univ. Program grey/lavender color DA Test Hybrid ‘Hyb L’ (F1) Material (Geneva, NY, US) Testcross: KP1704 x CLT

KP1706 *Descriptions do not represent information from seed source catalogs or online stores.

Plant production and management. During each year of the study (2016-2018), mature plant materials were produced on certified organic land at the Homer C.

Thompson Vegetable Research Farm in Freeville, NY (42°31’24.2”N,

76°19’40.2”W). Untreated and/or organic seed of each cultivar/breeding line were planted in 72-cell plastic trays in McEnroe OMRI-approved potting mix (Premium

Organic Potting Soil, McEnroe Organic Farm, Millerton, NY). Transplant materials were grown in a certified organic greenhouse at the Guterman Bioclimatic Laboratory

Complex (42°26’55.8”N, 76°27’39.9”W). All plant materials were evaluated for uniformity and sufficient vigor prior to field transplanting. Four-week old seedlings were transplanted into raised, black plastic beds spaced at 50cm with 2m row centers and outfitted with drip-fed irrigation. Immediately after transplanting, rows were covered with a floating row cover (Agribond®+ AG-19, Berry Plastics, San Luis

55

Potosí, México) to avoid herbivory damage prior to sensory evaluation.

Approximately eight-to-ten weeks after transplanting, the third and fourth most fully expanded leaves were harvested from each genotype, bunched, and assigned a random three-digit code before being placed in a 4°C refrigerator for further processing. All leaves used in sensory testing were free of decay and herbivory damage.

Qualitative Multivariate Analysis (QMA)

Participant recruitment and screening. General subscribers to the Cornell

Sensory Evaluation Center’s listserv were solicited for participation in an abbreviated

QMA (Lopetcharat & Beckley, 2012) during August of 2016 (Fig 2.1). Candidate participants were required to be 18 years of age or older, like and consume kale at least once per month out of season, eat kale at least once per week during the peak season (fall/winter), prepare kale themselves, and feel comfortable expressing their opinions in a group setting. Fourteen (n = 9F) self-identified kale enthusiasts, twelve of which were between 18 and 34 years of age, were selected from the Ithaca, NY community to participate in the study. The study consisted of a 14-day home use portion and a three-hour on-site focus group. Prior to home-use testing, participants were asked to complete a “warm-up” activity to engage their critical thinking skills and provide introductory material for the focus group discussion. Participants were responsible for the pick-up, preparation, and documentation of in-home kale product usage in journals. Attendance at a three-hour focus group discussion upon the completion of home testing was also mandatory. All participants were compensated

$75.00 upon completion of the study.

56

Figure 2.1. Qualitative Multivariate Analysis (QMA) method process, as adapted from Lopetcharat and Beckley (2012). Six leafy Brassica samples representing the kale product space were subject to home-use tests (HUT) by n = 14 consumers. In-home journals, photos, and observations provided a foundation for discussion during a facilitated three hour focus group, which included lexicon development and perceptual mapping exercises.

Home-Use Test (HUT). Over the course of two weeks in September 2016, selected participants were asked to document their in-home experience with six kale genotypes (Table 2.1) via paper journal or an online platform (Google Drive, Google LLC,

Mountain View, CA, USA). Genotypes used in the HUT were selected to be sufficiently distinct from one another and span the kale product space. Respondents were provided with enough whole, fresh kale (~400-500g) for two sizeable culinary preparations of each genotype; each participant evaluated three genotypes during each week of the study. Participants were asked to use each

57

genotype in two different preparations (“raw” and “cooked”) and prompted to describe their experiences with seven journal questions after each culinary preparation.

Questions were designed to understand the type of culinary preparation, sensory details, and the participant’s relative satisfaction with each kale genotype.

Facilitated focus group testing. At the conclusion of HUT, all participants were asked to join a focus group discussion hosted on the Cornell University Campus

(Stocking Hall, Ithaca, NY) on 24 September 2016. The three-hour focus group discussion was video recorded for future reference. A general discussion outline adhered to the QMA recommendations by Lopetrcharat and Beckley (2012), whereby the focus group started with consumer lexicon development, transitioned to building the value diagram, and concluded with perceptual mapping exercises. The majority of the focus group outline emphasized sensory modalities (i.e. appearance, flavor/taste, texture, and scent/aroma) and their relative importance when purchasing or consuming kale. Participants were given their original HUT journals to use as reference during lexicon development and value diagramming. Key values identified during the Love

It! or Hate It! exercise were used to inform panelist construction of perceptual maps.

Perceptual mapping was first performed on an individual basis, after which participants were split into two groups to discuss, identify valued attributes, and organize samples on a large piece of paper containing blank axes. The resulting two map projections were later compiled to create a singular perceptual map according to common trends and product arrangements identified during the focus group.

Descriptive Analysis (DA)

Panel recruitment and screening. In July 2017, thirty respondents were invited

58

through Cornell Sensory Evaluation Center’s listserv to participate in a brief sensory acuity test, including common red/green color blindness assessment using Ishihara color plates (Ishihara, 1936), identification of basic taste solutions (i.e. sweet, sour, salty, bitter, and umami), ranking of sour and bitter solutions of various intensity, and a series of triangle tests. Eligible panelists were 18 years of age or older and were required to attend two training sessions per week over the course of nine weeks to qualify for the study. In addition, panelists were required to attend three duplicate product evaluation sessions. Panelists were compensated for both training and evaluation sessions. Nine panelists (n = 7F) were able to complete all required training necessary to participate in product evaluations.

Panelist training and lexicon development. Selected panelists began descriptive sensory training in late July 2017. Two training sessions were hosted each week by the Cornell Sensory Center. Training sessions were designed in accordance with quantitative descriptive analysis methodology (QDA®) (ISO, 2009, 2014; Stone,

Sidel, Oliver, Woolsey, & Singleton, 2004). The panel initially focused on developing an attribute lexicon and identifying appropriate physical reference samples for each attribute. Panelists were exposed to numerous leafy greens, including those outside of the kale market product category, in order to formulate an adequate product concept.

Individual and group consensus lexicon development occurred during the first eight training sessions, whereby attributes were first generated individually and then proposed to the larger group of panelists for discussion (Fig 2.2, Appendix 2.1).

Submitted attributes were grouped, screened for redundancies, verbally defined, and associated with a physical reference to be used during later training phases. The final

59

attribute list consisted of 44 attributes (and associated physical references) spanning visual (5), aroma (5), flavor/taste (13), aftertaste (5), and texture (16) modalities

(Table 2.2). Elements of the Texture Profile® were used to help define textural attributes and aid panelists in understanding attribute definitions, but texture scales were not standardized and panelists did not receive as rigorous textural training as the

Texture Profile® method necessitates (Civille & Szczesniak, 1973; ISO, 1994).

Trained panel product evaluations. The remaining nine training sessions were used to calibrate panelists and refine attribute definitions/references. Panelists were instructed to evaluate each attribute on an unstructured numerical line scale marked with labeled endpoints indicating attribute intensity (0 = “none” to 100 = “extremely”) using

RedJade® Sensory Software Suite (RedJade®, Redwood Shores, CA, USA).

Conventional verbal and visual feedback from practice evaluations was provided to the panel at the following training session.

Consumer Central Location Test (CLT)

Panel recruitment and screening. A total of 90 kale consumers (n = 63F) were recruited through the Cornell Sensory Evaluation Center’s listserv to participate in a consumer CLT of organic kale in late September 2018. Eligible panelists were required to be 18 years of age or older, like and consume kale at least once per month, and were able to attend both evaluation sessions over the course of two days. A total of n = 88 participants completed both days of the study (Appendix 2.2). Unbalanced panelist data (n = 2) were also included in the final models; one participant was later removed from the study after reporting they “did not consume kale.” A $10 incentive was offered to panelists completing both days of the study.

60

Figure 2.2. Kale sensory wheel depicting lexicon developed through Qualitative Multivariate Analysis (QMA) and descriptive analysis. Descriptors (n=77) in the outermost ring of the wheel may exist within the kale product space, although not all were vetted through descriptive panel evaluations. Those descriptors protruding (solid color) from the outermost ring were observed to be significantly (P _ 0.05) different between kale samples subject to descriptive evaluation and may serve as foundational descriptors in future kale sensory studies.

61

Table 2.2. Attribute lexicon developed by trained kale sensory panel (n=9) over 17, one-hour training sessions in accordance with descriptive analysis. Reference products and attribute definitions were adjusted after panel discussion and agreement. ATTRIBUTE DEFINITION REFERENCE PRODUCT

VISUAL All the visible attributes of a substance or object (ISO

ATTRIBUTES* 5492:2008, 3.1) Young baby lettuce (thin); cabbage Leaf Thickness Degree to which leaf appears thick (thick) Waxy Wax-like, opaque bloom or coating over leaf Baby spinach (low); waxed apple (high)

Baby spinach (low); savoy cabbage Vein Presence Degree to which veins are present/prominent on leaves (high) General appearance of smooth surface; antonyms: Smooth Baby spinach (low); Tuscan kale (high) bumpy, prickly, bubbly Visual proxy for water content; degree to which leaf Old lettuce (wilted); fresh lettuce Turgidity appears turgid; antonyms: wilt, limp (turgid) AROMA Sensory attribute perceptible by the olfactory organ via

ATRIBUTES* the back of the nose when tasting (ISO 5492:2008, 3.25) Overall Intensity Overall strength of a sample's aroma Various leafy Brassica samples

Aroma associated with fresh cut grass, wheat grass Green Lawn clippings, green pepper shots, etc. Aroma associated with boiled cabbage, sautéed onions, Sulfurous Boiled cabbage hard-boiled eggs, etc. Aroma associated with damp/flooded basements, Musty Image of flooded basement; mold image storage sheds, and warehouses Aroma evoking stinging or biting sensations; synonym: Pungent Mustard; horseradish; garlic; chili spicy, sharp, biting, stingy Optional Attributes

Aroma associated with raw nuts (e.g. walnuts or Nutty Walnuts; peanuts; almond peanuts) Floral/Sweet Aroma often associated with sweet smell of flowers Rose water; peony; grape flowers

Aroma of fresh dairy products (fresh milk, fresh sweet Sweet Buttery/Milk Unsalted butter; whole milk butter) Green Apple Aroma associated with green apples Granny Smith apple slices

Earthy Aroma associated with dirt, mushrooms, etc. Beets; garden soil

Moldy Aroma associated with mold

Complex combination of the olfactory, gustatory, and FLAVOR trigeminal sensations perceived during tasting (ISO ATTRIBUTES 5492:2008, 3.20) Taste (non-retronasal) associated with citric acid Sour 0.1g/L (L), 0.3g/L (M), 0.5g/L (H) dilutions in water; synonym: acidic, tart Taste (non-retronasal) associated caffeine dilutions in Bitter 0.1g/L (L), 0.3g/L (M), and 0.5g/L (H) water Taste (non-retronasal) associated with monosodium Umami 0.3g/L (L), 0.6g/L(M), and 0.9g/L (H) glutamate (MSG) dilutions in water Taste (non-retronasal) associated with sucrose dilutions Sweet 5g/L (L), 10g/L (M), and 15g/L (H) in water Flavor associated with fresh cut grass, wheat grass, or Green Lawn clippings; green pepper green pepper, etc. Green Apple Flavor associated with green apples Granny Smith apple slices

Buttery Flavor associated with fresh milk, fresh butter Unsalted butter; whole milk

Flavor associated with dirt, mushrooms, etc.; synonym: Earthy Beets; mushrooms forest, moss, dirt Flavor associated with fermented foods like sauerkraut, Fermented Sauerkraut kimchi, etc. Flavor associated with boiled cabbage, sautéed onions, Sulfurous Boiled cabbage hard-boiled eggs, etc. Musty Flavor associated with damp basements Image of flooded basement; mold image

Moldy Flavor associated with mold e.g. moldy cheese Moldy cheddar cheese

Flavor evoking stinging or biting sensations; synonym: Pungent Mustard; horseradish; garlic; chili spicy, sharp, biting, stingy

62

ATTRIBUTE DEFINITION REFERENCE PRODUCT

Olfactory and/or gustatory sensation occurring after elimination of the product; differs from the sensations AFTERTASTE perceived while the product was in the mouth (ISO 5492:2008, 3.65) Lingering taste produced by dilute aqueous solutions of Bitter 0.1g/L (L), 0.3g/L (M), and 0.5g/L (H) quinine or caffeine Dilute aqueous solutions of various acid (citric, tartaric, Sour 0.1g/L (L), 0.3g/L (M), 0.5g/L (H) etc.) Dilute aqueous solutions of monosodium glutamate Umami 0.3g/L (L), 0.6g/L(M), and 0.9g/L (H) (MSG) Shrinking, drawing or puckering of the skin or mucosal Astringent Cranberry; pomegranate surface in the mouth 1.0g FeSO /L H O; penny in water; cut- Metallic A lingering taste of metal in the mouth 4 2 lip Mechanical, geometrical, and surface attributes of a TEXTURE* product perceptible by means of mechanical, tactile, visual, and auditory receptors (ISO 5492:2008, 3.41) Force required to achieve a given deformation or Hard Processed cheese (soft); carrot (hard) penetration of a product Young baby lettuce (thin); cabbage Leaf Thickness Perception of leaf thickness when in mouth (thick) Dense Higher mass or compactness of the leaf Baby spinach (low); cabbage (high)

Saltine (low); raw, peeled cucumber Moisture Water absorbed by or released from the product (high) Degree to which a substance can be deformed before it Cohesive Processed cheese (low); caramel (high) breaks Product breaking between teeth accompanied by an Crisp Crunch Wilted lettuce (low); carrot (high) acute breaking sound Hardness and the force necessary to break a product into Brittle Caramel (low); carrot (high) crumbs or pieces Effort required to disintegrate the product to the state Ice cream (low); gummy fruit candy Chewy ready for swallowing (high) Number of chews required to reduce product to Chew Count N/A consistency appropriate for swallowing Force required to remove material that sticks to the Adhesive Raisins; peanut butter mouth or teeth Perception of particles in a product; a continuous even Smooth Avocado peel (low); baby spinach (high) surface Waxy Sleek, smooth, and plastic-like product surface Cucumber and zucchini peels

Smooth, flat particles that can feel powdery and not Chalky Image of chalk dissolve in mouth Fibrous Long particles or strands oriented in the same direction Raw celery sticks; unripe mango

Quantity of fat in a product, as perceived by soaking and Oily Coating Water (low); butter (high) runny fat in a product Slimy Coating Slippery mucous-like viscous feel in mouth Water (low); okra (high)

*Assess the condition of the food prior to swallowing

Consumer panel product evaluations. The consumer panel evaluated ten of the genotypes tested in the descriptive study (Table 2.1); six genotypes from the “Green

Curly” market class were not evaluated by the consumer panel. Similar to sample preparation during descriptive analysis, plant material was harvested, bunched, coded, and refrigerated (4°C) from Cornell’s organic research farm so that sufficient sample

63

was available for both days of the study. Samples were prepared and portioned as previously mentioned; prepared materials were returned to the 4°C refrigerator and relocated to room temperature one hour before evaluation.

Five raw kale samples, identified with randomly assigned three-digit codes, were tested during each evaluation session. Samples were monadically served to panelists in individual computer-equipped testing booths under red lights with adequate ventilation, temperature regulation, and protection from sound. Sample order was randomized for each panelist. Filtered water and unsalted crackers were available to panelists during mandatory breaks between samples. Panelists were instructed to rate (9-point hedonic scale) their liking (overall, appearance, color, aroma, flavor, and texture) of each sample, and they were also instructed to report the intensity of a small number of flavor and texture attributes on unstructured numerical line scales marked with labeled endpoints indicating attribute intensity (0 = “none” to 100 = “extremely).

All data were collected using RedJade® Sensory Software Suite.

Statistical procedures and multivariate analysis

Intensity ratings and consumer hedonic scores were treated as continuous variables and subject to multivariate statistical analysis. Quantitative data were analyzed using nested mixed models composed in the ‘lme4’ package (Bates, Mächler,

Bolker, & Walker, 2015) of R software (R version 3.5.0) (R Core Team, 2018), whereby the effect of genotype (‘Sample’) was fixed and the effect of evaluation session (‘Session’) was modeled as a random effect nested within panelist (‘Panelist’).

Random effect variance components were extracted from each model, and a two-way analysis of variance (ANOVA) was performed on product sample intensity ratings

64

(Appendix 2.3) and consumer hedonic scores (Appendix 2.4) to test differences among sample means; sample effects were declared significant below ⍺ = 0.05. Where appropriate, the ‘emmeans’ package (Lenth, 2018) was used to perform post hoc

Tukey’s honestly significant difference (HSD) tests on estimated means and detect significant differences (P ≤ 0.05) among product samples for the descriptive

(Appendix 2.5) and consumer CLT (Appendices 2.6 and 2.7) studies.

To understand segmentation among consumer preferences and broader trends of variation among product samples, agglomerative hierarchical clustering (AHC), principal components analysis (PCA), and external preference mapping were subsequently performed using XLSTAT-Sensory (XLSTAT Version 2018.5.52460,

Addinsoft, Paris, France). Cluster analysis was performed on hedonic scores for overall liking, appearance, flavor, and texture using a Euclidean dissimilarity matrix,

Ward’s agglomeration method, and the automatic truncation (i.e. entropy) settings in

XLSTAT. Chi-square tests of independence were performed in R software to examine relationships between consumer demographics and AHC consumer clusters (Appendix

2.8). A correlation PCA (Pearson’s r) was performed on the estimated mean values of the 15 attributes judged by the trained panel. Appearance attributes were not included in the PCA, as their interpretation would be confounded by differences in the sampling and testing procedures.

Resulting PCA factor scores were used in conjunction with aforementioned

AHC results to produce an external preference map, whereby consumer hedonic scores from the CLT were related to QDA attribute intensity ratings through multiple regression. AHC class centroids (i.e. mean hedonic ratings for each sample) were

65

regressed onto PCA factor scores using the best-fit polynomial model to produce preference maps (Lawless & Heymann, 2010; van Kleef et al., 2006). External preference maps for each sensory modality were superimposed onto a contour plot, displaying clusters with above-average preference throughout regions of the map.

Graphics were developed using a combination of XLSTAT-Sensory and the ‘ggplot2’ package (Wickham, 2016) of R software.

Results and Discussion

HUT journal feedback and value diagramming

QMA was first performed to understand values and motivators associated with kale consumption, link them to specific sensory attributes, and build hypotheses for future research. Ten of the fourteen QMA participants completed HUT journals through the online outlet, and all participants successfully uploaded culinary preparation photos through Google Drive. Several participants were not exposed to kale until recently and expressed hesitancy in exploring various culinary preparations during HUT. Few participants used similar preparations among samples; the majority of participants felt as though the different genotypes afforded them the opportunity to explore other preparations.

Raw kale salad was the most repeated preparation method across samples.

Generally speaking, the most common preparation methods included sautéed in a stew, used as a soup ingredient, chipped/roasted, and used as a base for a raw salad.

Larger, mature leaves lent themselves to heated preparations, and participants readily suggested pre-packaged or baby kale would be more appropriate for raw preparations or salads. Participants noted a distinction between product samples and other plant

66

relatives (e.g. cabbage and broccoli), but often mentioned similarities to leafy greens such as chard and spinach. In many culinary preparations, consumers substituted kale for chard or spinach. Kale’s robust texture and ability to maintain its shape and integrity after cooking was recorded as a primary differentiating feature between kale and other leafy greens. Kale often served as the foundation for many of the participant’s dishes, adding bulk and color impact to preparations. Participant’s culinary preparations reflected the versatility of kale as an ingredient in world cuisines.

Participants strongly adhered to an “iconic” kale identity through the focus group, often describing ‘Winterbor’ as the iconic genotype. ‘Winterbor’ was seen as the familiar or traditional kale most commonly present in grocery stores or displayed in the media. Key features of this sample signaling its kale identity included its voluminous appearance, heartier leaf texture, good storability, and ability to retain its shape/form after cooking. These descriptors were also identified as the “most important” features of this product class to participants; aroma, leaf size, and range of color were of lesser importance. ‘Winterbor’ was also described as voluminous, satisfying, bouncy, curly, leathery and sturdy. This variety was the most preferred in appearance (raw or cooked), texture (raw or cooked) and flavor (cooked only) sensory modalities (Appendix 2.9).

While some samples in the product set did not fit within the iconic kale identity, none were completely rejected by participants in this study. Commonly available genotypes ‘Nero di Toscano’ and ‘White Russian’ were more accepted as

“kale” than the other three genotypes in this study (‘Top Bunch’, ‘Beira’, ‘P13-2’).

67

The ‘Nero di Toscano’ Tuscan kale genotype was associated with rich, lush, and ancient descriptors. For some participants, this kale was more flavorful, and, for others, it was among the most bitter of the product set. Many sensory characteristics of the Siberian ‘White Russian’ genotype aligned with traditional kale expectations, but, according to some participants, the appearance and durability did not qualify it as a

“kale.” This sample was rich in color but wilted faster due to its thinner texture.

Delicate, dramatic, and tender were all common descriptors used to describe ‘White

Russian’.

The majority of participants strongly agreed the ‘Top Bunch’ collard genotype was more closely related to collard, cabbage, or even swiss chard than it was to kale.

Similarly, the Portuguese tronchuda genotype (‘Beira’) was considered by all participants to poorly fit their expectation of kale. Its mild flavor, extensive midribs, and large leaves were daunting to participants and left many unsure of how to use this sample in a culinary preparation. It was recorded as one of the most sturdy and turgid samples in the product set. Breeding line ‘P13-2’ was noted to be too light in color and texture to be kale. Participants found this genotype similar to a dense lettuce with slight bitterness. In some preparations, this cultivar lost its shape/integrity after cooking, but retained a floral or botanical flavor. The yellow color was noted by some participants as off-putting, unhealthy, or cheap. Others found it to have a brighter, more vibrant color than other cultivars in the product set. These observations highlight the importance of color in defining a consumer’s sensory experience (Sugrue &

Dando, 2018). Additional leaf types with diverse colors not included in the HUT were presented to participants during the focus group and positively received. Participants

68

expected kale to maintain a dark green, dark, or seafoam green color from purchasing to consumption. Deep, rich colors (including purple) were also inviting, as these colors indicated higher perceived nutritional value, but they were also considered to be fancy or decorative.

While the facilitated focus group centered on identifying and describing sensory attributes associated with each sample in the product set, discussion during the focus group consistently returned to topics within three primary categories: purchasing motivators, “iconic” kale loyalty, and health/nutrition (Table 2.3).

Table 2.3. Values associated with the purchasing and consumption of kale. Values were described by a focus group of consumers (n=14) after home-use testing (HUT) of six kale genotypes. Values are separated according to higher-level and sensory descriptors and listed in order of relative importance. Value Descriptors Lifestyle/ gardening, farmers’ market, environmental concerns, organic, local, Social Interactions family, self-care, cool/on-trend, convenient nutritious, gardening, organic, diet, lifestyle supplement, virtuousness, Health and Well-Being convenience, exercise, superfood, “good for me”, energizing, fresh durability, hardy, keeps in refrigerator for days, distinct from other Shelf-Life leafy greens, bagged salads vs. whole leaves preconceived kale identity, versatility in uses, size of midrib/stem, Appearance/Recognitio muscular vs. graceful, leaf margin (flat vs. curly), undamaged leaf n surface, kale vs. not kale, retains identity when prepared Sensory Attributes

robust/pungent/strong, spectrum of flavor intensity, sulfurous, bitter, Flavor grassy, natural tender, tough, fibrous, rubbery, mucilaginous, rigid/hard vs. soft, Texture crunchy, flaccid, stringy vibrant or dark green, red/purple pigments, retention after cooking, Color aversion to yellow type, pigmented types considered “decorative” Scent/Aroma fresh, absent, mild, hint of iconic “sulfur” smell when cooked

These categories existed within a larger framework of primary core values associated with kale: lifestyle/social interactions, health and well-being, shelf-life, and product appearance/recognition. Product sensory attributes aggregated to form consumer core value concepts. For example, thicker leaves were associated with tougher texture

69

(Value: Sensory Attributes) and, in turn, were perceived to store and retain those qualities when cooked (Value: Storability/Freshness).

Trained panel assessments

Descriptive analysis was performed to create a standard lexicon for kale and to describe sensory differences between kale samples. Nine trained panelists identified

21 sensory attributes with significantly (P ≤ 0.05) different means among kale genotypes (Table 2.4; Appendices 2.3 and 2.10). All visual attributes (turgidity, smooth, vein presence, waxy, and leaf thickness) were significantly different between samples, highlighting the diversity of morphologies and leaf types in this product set.

Overall intensity of aroma, bitterness (flavor), bitterness (aftertaste), and sweetness were the only significantly different attributes from the aroma, flavor, and aftertaste modalities. Eleven of the 21 differing attributes were texture-related attributes, the majority of which were mechanical and/or related to features of the sample's surface.

Pearson correlation (𝜌) analyses indicated a high degree of correlation (𝜌 ≥ 0.75) among hardness-density, hardness-waxy, leaf thickness-waxy, and smooth-slimy attributes (Appendix 2.11). Ratings for leaf thickness, hardness, density, waxy, and slimy were different among curly kale samples.

70

Table 2.4. Estimated means and groupings of 44 sensory attributes in kale genotypes (n=15) evaluated by a trained descriptive panel (n=9). Attribute intensity was assessed on an unstructured numeric line scale (0 = “none” to 100 = “extremely”). ‘Consumer Liking’ was evaluated using a 9-pt. hedonic scale by n=90 consumers in a central location test (CLT). Tukey's HSD was performed on nested mixed model estimate means; means with the same grouping letter are not significantly different (P ≤ 0.05).

ATTRIBUTES BREEDING LINES TUSCAN RED COLLARD SIBERIAN GREEN CURLY

Yellow Black Magic White Russian Laerketunge Meadowlark Hybrid E Hybrid I Hybrid K1 Hybrid L Inbred Redbor Tiger Darkibor Krul Reflex Starbor Smooth*** 41.3 ab 36.8 abc 42.0 ab 22.3 cd 52.3 a 18.8 d 26.4 bcd 51.4 a 28.7 bcd 23.4 cd 20.1 cd 28.8 bcd 26.5 bcd 24.3 bcd 23.2 bcd

Thick*** 25.7 c 32.4 abc 44.4 ab 24.7 c 36.5 abc 44.5 a 27.5 c 38.3 abc 33.3 abc 33.6 bc 25.3 c 29.8 bc 32.4 abc 31.1 bc 34.1 abc Turgid*** 35.6 b 36.7 b 52.0 ab 44.4 ab 54.5 ab 50.1 ab 50.8 ab 50.4 ab 39.5 ab 53.1 a 51.8 ab 42.5 ab 41.1 ab 54.8 ab 48.3 ab 52.4 ab 46.7 ab 43.6 ab 54.4 a 43.0 ab 37.2 b 50.9 ab 48.3 ab 37.7 ab 42.3 ab 48.1 ab 37.1 ab 46.3 ab 46.8 ab 39.1 ab VISUAL Vein Presence** Waxy*** 23.6 ef 40.1 bcde 65.8 a 30.4 cdef 29.2 def 63.7 a 16.2 f 56.6 ab 48.4 abc 41.3 bcd 25.8 def 27.1 def 27.9 def 35.9 cdef 34.2 cdef Yellow Black Magic White Russian Laerketunge Meadowlark Hybrid E Hybrid I Hybrid K1 Hybrid L Inbred Redbor Tiger Darkibor Krul Reflex Starbor Green ns 29.5 27.6 27.8 27.5 23.2 19.3 19.8 24.1 20.8 20.0 15.2 29.3 23.3 22.1 17.6

Musty ns 15.7 17.6 13.2 19.3 15.9 17.0 16.1 18.8 20.5 19.5 20.7 18.4 17.4 20.9 20.9 Overall*** 35.3 ab 32.7 ab 31.0 ab 33.3 ab 41.1 a 32.0 ab 42.9 a 33.4 ab 41.5 a 27.3 b 32.2 ab 35.8 ab 42.3 a 28.5 ab 26.5 ab 3.2 4.9 4.7 7.6 8.5 6.1 6.6 9.4 5.2 3.7 7.4 7.2 3.8 4.1 5.8

AROMA Pungent Sulfurous 8.2 5.3 9.2 6.9 6.3 10.3 7.1 10.1 15.0 9.4 14.8 9.6 11.6 13.2 8.1 Yellow Black Magic White Russian Laerketunge Meadowlark Hybrid E Hybrid I Hybrid K1 Hybrid L Inbred Redbor Tiger Darkibor Krul Reflex Starbor Bitter*** 20.9 ab 25.6 ab 17.2 ab 20.8 ab 24.7 ab 26.5 a 29.2 ab 15.7 ab 20.0 ab 15.2 b 26.7 ab 26.0 ab 33.6 a 17.3 ab 25.6 ab Buttery ns 4.5 7.5 9.7 5.5 2.5 4.8 2.9 2.8 4.4 5.7 5.8 4.0 3.3 7.0 5.5 Earthy ns 25.7 25.7 25.3 22.8 21.0 25.1 25.1 22.2 25.6 22.1 23.5 27.8 26.3 23.0 27.9

Fermented ns 8.7 11.1 6.6 11.1 8.6 11.3 7.6 8.9 11.5 11.1 11.2 10.4 6.6 11.9 9.8 Green ns 37.9 32.8 34.9 38.3 34.7 34.2 35.8 37.5 37.7 32.9 35.9 36.9 38.2 25.7 35.9 Green Apple ns 15.8 9.6 11.7 9.4 14.4 10.8 19.1 9.9 17.1 10.3 7.3 13.8 12.0 12.5 12.9 Moldy ns 5.9 9.6 7.3 4.6 5.6 7.9 7.6 5.8 6.4 8.6 5.3 2.9 5.9 5.9 6.7 Musty ns 16.5 17.2 10.2 13.8 18.1 15.5 16.1 17.0 17.6 17.7 18.4 16.7 16.0 17.9 18.1 Pungent ns 25.0 15.6 23.2 17.6 34.4 29.6 23.2 24.8 22.7 19.8 25.1 29.1 27.8 28.0 28.7

TASTE AND FLAVOR Sour* 17.8 ab 19.8 ab 21.0 ab 18.7 ab 26.6 ab 22.7 ab 31.9 a 16.7 b 25.2 ab 16.6 b 22.4 ab 21.1 ab 25.0 ab 16.3 ab 24.1 ab Sulfurous ns 19.9 21.1 15.6 14.6 21.9 20.4 13.2 18.9 15.6 17.8 19.0 14.9 19.3 19.6 18.9 Sweet* (+) 23.7 a 14.3 a 13.0 a 13.0 a 12.6 a 11.7 a 8.0 a 17.8 a 17.5 a 15.7 a 17.4 a 16.6 a 8.0 a 18.4 a 11.9 a Umami ns 10.9 15.7 14.0 15.3 10.4 9.6 10.7 11.4 8.1 12.3 8.8 6.9 6.3 7.8 9.1

71

Table 2.4 cont. Estimated means and groupings of 44 sensory attributes in kale genotypes (n=15) evaluated by a trained descriptive panel (n=9). Attribute intensity was assessed on an unstructured numeric line scale (0 = “none” to 100 = “extremely”). ‘Consumer Liking’ was evaluated using a 9-pt. hedonic scale by n=90 consumers in a central location test (CLT). Tukey's HSD was performed on nested mixed model estimate means; means with the same grouping letter are not significantly different (P ≤ 0.05). ATTRIBUTES BREEDING LINES TUSCAN RED COLLARD SIBERIAN GREEN CURLY Yellow White Black Magic Laerketunge Meadowlark Hybrid E Hybrid I Hybrid K1 Hybrid L Inbred Redbor Tiger Russian Darkibor Krul Reflex Starbor Astringent ns 16.8 21.8 20.2 22.4 17.1 23.3 24.2 15.7 19.2 15.8 21.3 20.3 24.8 19.3 17.2 Bitter** 22.4 ab 16.6 ab 17.6 ab 16.2 ab 23.0 ab 27.4 a 22.0 ab 17.7 ab 23.5 ab 14.7 b 22.3 ab 18.8 ab 28.4 a 18.8 ab 23.1 ab Metallic ns 3.4 3.7 5.7 3.4 6.6 6.5 5.1 5.2 5.3 3.7 5.6 6.2 4.7 4.2 5.0

Sour ns 16.9 11.9 11.5 14.9 14.5 16.7 18.8 12.0 16.4 12.6 17.5 16.5 15.5 14.6 14.0 Umami ns 7.0 6.4 9.4 8.7 7.3 7.3 10.6 8.2 8.5 11.4 9.0 8.0 6.1 8.4 10.7

AFTERTASTE Yellow White Hybrid E Hybrid I Hybrid K1 Hybrid L Black Magic Redbor Tiger Darkibor Krul Laerketunge Meadowlark Reflex Starbor Inbred Russian Adhesive* 43.6 ab 47.4 ab 46.7 ab 49.6 ab 38.5 ab 38.8 b 36.1 ab 42.1 ab 35.8 ab 46.3 ab 49.9 ab 47.4 ab 45.9 ab 50.9 ab 53.9 a Brittle*** 25.3 abc 24.6 abc 25.2 abc 18.4 abc 34.7 a 28.5 ab 16.2 bc 28.8 abc 24.7 abc 15.5 c 17.8 abc 26.1 abc 28.0 abc 15.4 bc 21.9 abc Chalky ns 9.4 12.0 9.9 11.1 6.5 7.3 10.2 4.2 8.4 10.6 9.3 11.1 9.1 7.1 8.5 Chewy*** 32.2 abc 33.5 abc 42.5 a 36.0 abc 21.4 c 31.7 abc 39.9 ab 27.6 abc 34.2 abc 38.0 ab 34.6 abc 34.9 abc 25.4 bc 32.9 abc 35.2 abc Cohesive ns 30.9 27.4 35.2 32.5 22.7 28.2 34.0 28.2 27.9 34.0 36.7 38.0 32.7 40.7 30.7 Crisp Crunch*** 42.1 a 32.0 ab 28.1 ab 27.9 ab 35.4 ab 33.6 a 40.3 a 35.6 ab 33.6 ab 25.0 b 32.4 ab 31.4 ab 37.0 ab 28.3 ab 32.3 ab Dense*** 38.0 a 40.8 a 43.2 a 34.2 ab 18.1 b 37.8 a 35.3 a 35.6 a 36.0 a 35.0 a 30.2 ab 31.1 ab 26.7 ab 34.0 ab 37.5 a Fibrous ns 30.1 36.5 36.5 31.7 23.0 26.7 33.4 31.1 33.7 30.5 36.8 30.7 28.6 31.3 34.3 Hard*** 34.6 abcd 44.5 a 41.7 ab 26.6 cde 18.1 e 32.8 bc 29.6 abcde 33.6 abcd 27.8 bcde 26.5 cde 23.1 cde 26.7 bcde 20.8 de 25.1 cde 30.5 abcde Leaf Thickness*** 28.1 bcd 37.2 abc 44.9 a 19.8 d 31.6 abcd 34.9 ab 26.3 bcd 33.5 abcd 29.7 abcd 25.5 cd 22.5 cd 23.8 bcd 20.6 d 26.9 bcd 23.7 bcd

TEXTURE Moisture ns 23.1 22.3 21.1 25.9 27.1 22.4 23.1 20.1 16.9 21.3 22.8 20.4 21.4 20.5 18.7 Number Chews** 14.5 ab 15.3 ab 18.2 a 17.0 a 12.0 b 15.4 ab 17.9 a 14.2 ab 16.5 a 16.3 a 15.9 ab 14.9 ab 14.0 ab 16.0 ab 15.5 ab Oily Coat ns 13.9 7.4 7.9 5.2 7.1 6.0 2.4 14.8 9.7 6.0 3.2 4.3 4.7 5.5 4.6 Slimy** 12.5 ab 4.6 ab 7.9 ab 2.4 b 11.6 ab 3.9 b 3.9 b 17.0 a 8.8 ab 3.2 b 3.8 ab 3.1 ab 5.6 ab 2.9 ab 2.5 b Smooth*** 42.1 a 30.4 abc 30.1 abc 19.0 c 42.5 a 24.6 c 22.0 c 38.2 ab 35.8 abc 22.9 c 23.2 bc 25.9 abc 23.2 bc 25.7 abc 20.9 bc Waxy*** 37.6 ab 49.1 a 47.9 a 26.5 b 34.1 ab 47.4 a 23.2 b 37.7 ab 34.9 ab 28.2 b 28.5 b 26.4 b 21.0 b 27.5 b 38.7 ab

CONSUMER LIKING 6.7 ab 6.8 ab 6.4 abc 7.1 a 6.0 c 6.9 ab 6.5 abc 6.6 abc 6.3 bc 6.9 ab ------Significance codes: 0.001 = '***', 0.01 = '**', 0.05 = '*', not significant = ‘ns’. (+) - adjustment for multiple comparisons resulted in a single Sample grouping

72

Most notably, the curly kale samples were readily distinct from ‘Black Magic’,

‘Hybrid I’, and ‘Hybrid K1’ genotypes, all of which had Tuscan kale in their genetic background. Curly market class types also exhibited distinct waxy (texture) grouping patterns and greater adhesiveness than the rest of the product set.

Little variation within the green curly market class existed beyond the overall intensity of aroma, bitterness (flavor), bitterness (aftertaste) attributes, with differences found only between ‘Meadowlark’ and ‘Darkibor’. This observation may be the byproduct of a general lack of diversity among germplasm within already existing breeding programs or, alternatively, a lack of sufficient breeding programs contributing to the diversification of these curly kale types. Targeted breeding for new curly kale genotypes in the U.S. has only been underway within the past few years, but there are a significant number of green curly kale varieties available through small- scale and international markets. As such, it may be valuable to validate this observation in future studies that include replication across location-environments, with additional culinary preparations and a larger product set.

PCA was conducted using estimated means of the 15 statistically significant product attributes within aroma, flavor, texture (excluding chew count), and aftertaste modalities (Fig 2.3A, Appendix 2.11). Nine factor variables were identified using a

Pearson correlation (𝜌) matrix, four of which had an Eigenvalue >1 and explained

88.1% of the variation. PCA illustrated a separation of kale samples along two primary dimensions (axes) that accounted for 61.2% of the variance. Attribute contributions to the first two PCA dimensions illustrated a slight separation of texture

(mechanical)/flavor and textural (surface attributes between Dimensions

73

Figure 2.3. Graphical representation of product separation in both qualitative and quantitative studies. Products closers together in the visual space provided are understood to be more similar to one another than those spaced further apart. (A) First two PCA dimensions (axes) explaining 61.24% of the variation among 10 kale samples (blue) subject to descriptive analysis. Estimated means of 15 sensory attributes (orange) within aroma, flavor, aftertaste, and texture modalities, all of which were significantly different among kale samples, were used to compile a correlation matrix. Attribute contributions to the first two PCA dimensions illustrate a slight separation of texture (mechanical)/flavor and textural (surface) attributes between Dimensions 1 and 2, respectively. (B) Qualitative compilation of group consensus perceptual mapping exercise from the QMA facilitated focus group. Separation of samples in the QMA product set is represented along axes of “Texture” and “Flavor Intensity,” both of which were identified as primary sensory distinguishers among samples. Orange ellipses depict range of sample sensory space identified between two consumer groups performing mapping exercises. (C) External consumer preference mapping obtained by PLS analysis of descriptive and AHC consumer hedonic data. Axes represent the first (F1) and second (F2) dimensions of PCA performed on descriptive data. Four consumer clusters (orange) were identified through AHC of consumer hedonic scores for “Overall Liking” of 10 kale samples (white). A contour plot (olive) has been overlaid to illustrate the percentage of cluster exhibiting an above-average preference for samples in a particular region of the sensory map.

74

One and Two, respectively. Dimension One correlated with overall intensity of aroma, bitter (aftertaste), adhesive, and crisp crunch. Samples ‘Darkibor’ and ‘Yellow IL’ contributed most significantly to Dimension One, though many genotypes within the curly kale market class separated along this axis. Chew count was also included as a supplementary quantitative variable in the PCA and was highly correlated (𝜌 = 0.93) with the chewy attribute along the first dimension. Attributes including sweet (flavor), waxy (visual), smooth, slimy, and brittle differentiated product samples along

Dimension Two. ‘Redbor’ and ‘Tiger’ marked the extremes of genotypes correlated with the second dimension. Bitterness (flavor), leaf thickness, hardness, density, and waxy (texture) appeared correlated with a third dimension (not shown). Testcross hybrid breeding lines, notably 'Hybrid I' and 'Hybrid K1', occupied space on the variables-observations biplot that was not otherwise well represented by existing genotypes (Fig 2.3A).

Hedonic CLT and consumer clustering

A consumer test was conducted to determine consumer acceptance of kale genotypes and validate earlier findings from the QMA and descriptive analysis studies. Approximately 65% of participating consumer panelists reported consuming kale at least once per week, regularly consuming kale in raw (69%) or sautéed (72%) form (Appendix 2.2). Liking scores from the panel (n = 90) were significantly different (P ≤ 0.05) between raw kale product genotypes for overall liking, appearance, color, flavor, and texture (Appendix 2.4). No significant difference in liking of aroma was observed between genotypes, and Tukey’s HSD mean separation analysis for liking of texture revealed no differences between individual samples.

75

Mean hedonic scores for overall liking ranged from 6.0 - “like slightly”

(‘YellowIL’) to 7.1 - “like moderately” (‘Hybrid L’) (Table 2.4). Across modalities,

‘Hybrid L’, ‘Darkibor’, and ‘Black Magic’ were consistently rated in the top half of all samples, though ‘Hybrid I’ was given the highest hedonic score for liking of color

(Appendix 2.6). Acceptability of aftertaste did not differ (P ≥ 0.05) between genotypes, but one-third of panelists declared the aftertaste of ‘White Russian’ unacceptable. Concept fit and purchase intent of each genotype reflected trends in overall liking; in descending order, ‘Darkibor’, ‘Hybrid L’, ‘Black Magic’, and

‘Hybrid I’ were selected as those which best fit panelist concept of kale and indicated greatest purchase intent. ‘YellowIL’ was consistently rated among the least-liked samples across all hedonic categories. In addition, ‘YellowIL’ was not consistent with panelist concept of kale exhibited the lowest purchase intent among all genotypes.

All attribute intensities were rated significantly different (P ≤ 0.05) between genotypes by the consumer panel. Consumers used a greater range of the intensity scale than trained descriptive analysis panelists. Panelists were asked to rate their ideal bitterness [xbar = 36.0 (22.6)] and sweetness [xbar= 47.9 (24.67)] intensities, both of which highlighted variable liking patterns among consumers. With the exception of

‘White Russian’, all product sample means were within one standard deviation of the mean ideal bitterness intensity. Similarly, all genotypes were within one standard deviation of the mean ideal sweetness, though all samples fell below the ideal mean.

AHC grouped consumers into four clusters based on scores of overall liking.

Within class variance represented 71.37% of the total variance, and within class variance of each cluster ranged from 14.4% (Cluster 3) to 31.6% (Cluster 4). Cluster 4

76

(n = 17) was most dissimilar to the other three clusters: Cluster 1 (n = 29), Cluster 2

(n = 11), and Cluster 3 (n = 31). Chi‐square tests of independence revealed a significant (alpha = 0.1) relationship between consumer clusters and three demographic categories: age (X2 (df = 3, N = 88) =8.0121, p = 0.04576), ethnicity

(X2 (df = 6, N = 88) = 12.526, p = 0.05121), and consumption rate (X2 (df = 6, N =

88) =12.396, p = 0.05369). While a statistically significant relationship was established between these demographics and clusters, it is difficult to discern noticeable demographic trends due to low frequency counts in some clusters. Gender and variety seeking behavior were not significantly different among consumer clusters. Clustering procedures were also performed with sample appearance, flavor, and texture hedonic scores (Appendix 2.12).

Perceptual mapping and external preference maps

Perceptual mapping exercises were performed during the QMA and descriptive portions of this study to illustrate relationships among kale products. Projective mapping performed during the QMA facilitated focus group resulted in a two- dimensional perceptual map characterized by ‘Texture’ and ‘Flavor Intensity’ (Fig

2.3B). Appearance (color and visual texture of the leaf) was also a key sensory modality differentiating among test samples, though it did not appear as frequently as flavor and texture. Similar to the quantitative PCA developed during the descriptive portion of this study, green curly kale (‘Winterbor’) and the yellow breeding line

(‘P13-2’) were differentiated along the ‘Flavor Intensity’ axis. ‘White Russian’ and

‘Nero di Toscano’ separated along the ‘Texture’ axis, whereby ‘White Russian’ was soft/tender and ‘Nero di Toscano’ was considered hard/rough. This trend was a

77

departure from the PCA results, where ‘White Russian’ and ‘Black Magic’ (a Tuscan kale type similar to ‘Nero di Toscano’) were closely grouped near the map’s origin, due to differences in the context of each study.

PCA factor scores were integrated with results from AHC to perform external preference mapping with consumer hedonic scores. Hedonic scores were modeled for each consumer cluster, using PCA factor scores as the explanatory variables. Clusters

1 and 2 exhibited similar trends in sample preference order. Curly genotypes

(‘Darkibor’ and ‘Hybrid L’) were most preferred in both clusters, and hybrid breeding lines (‘Hybrid K1’ and ‘Hybrid I’) ranked in the top four most preferred samples. Both clusters ranked ‘Yellow IL’ as their least preferred kale type, and small rank changes among ‘Hybrid E,’ ‘White Russian,’ and ‘Tiger’ were the only differentiating features between Cluster 1 and Cluster 2. The iconic green curly kale and flat‐leaved collard types, however, created a dichotomy in consumer clustering and preferences.

A contour plot was produced to demonstrate the percentage of clusters with an above‐average preference for samples on the sensory map (Fig 2.3C). Collard‐like genotypes, including ‘YellowIL’, ‘HybE’, and ‘Tiger’ exhibited little‐to‐no (that is

<40%) above average preference for any consumer cluster. Within the context of this study, which was advertised as a “kale” consumer study, it is reasonable to assume that these collard types do not fit the consumer kale concept and were, therefore, less preferred among this product set. Interestingly, Cluster 3 rated ‘Redbor’ and ‘HybL’ as the most preferred types. ‘Redbor’ was not highly liked within other clusters and represents the most pigmented (red/purple) sample in the product set. The separation between Clusters 3 and 4 was notable, and a narrow band along the “Hybrid K1” and

78

“Hybrid I” portion of the sensory map indicated between 60–80% of consumer clusters had an above‐average preference for kale types in that region. In general, samples ‘Darkibor,’ ‘Hybrid K1,’ ‘Hybrid I,’ and ‘Hybrid L’ satisfied at least two‐thirds of all consumer panelists.

The performance of traditional Tuscan kale type (‘Black Magic’) was interesting, as it was a moderate performer in all three consumer CLT clusters. Despite having similar descriptive profiles to ‘Black Magic,’ test hybrids ‘Hybrid I’ and

‘Hybrid K1’ performed moderately‐to‐well among all four consumer clusters and were located in the area of the preference map satisfying the greatest percentage of consumer clusters. These samples appeared to have recovered much of the Tuscan sensory profile after crossing with collard and broccoli, respectively. During the QMA

“Warm‐Up” exercises, participants were asked to describe their ideal kale. One participant commented, “My ideal kale would look something like dinosaur [Tuscan] kale, with a more flat leaf blade rather than the convex leaves you usually see in dinosaur kale. The leaves would be large and have a small stem/mid

‐vein.” This description accurately describes the ‘Hybrid I’ leaf type. Three test hybrids (‘Hybrid I,’ ‘Hybrid K1,’ and ‘Hybrid L’), in fact, performed well in the consumer CLT. Each of these test hybrids retained some resemblance of available

(and popularized) green curly and Tuscan kale market classes, and these products are in essence subcategories of these existing market classes.

Higher liking scores among these test hybrids may be affected by a prior favorable experience or exposure to these familiar types (Garber et al., 2003).

Regardless, these samples occupy new space on the sensory map and demonstrate the

79

capacity of consumer clustering in understanding varietal distinction, market class plasticity, and consumer-defined quality within leafy Brassicas. Clustering of liking patterns is reasonable in a vegetable such as kale, as bitter compounds capable of activating receptors with known polymorphisms in human populations can drive disliking in foods (Franks, Lawrence, Abbaspourrad, & Dando, 2019; Wieczorek,

Walczak, Skrzypczak-Zielińska, & Jeleń, 2017). Regular kale consumers appreciate the curly, bitter identity of traditional kale genotypes (e.g. ‘Darkibor’), but also prefer thin leaves that leave little bitter aftertaste. The sensory profile of ‘Hybrid L’, which did not have a noticeably different profile from other genotypes, was most preferred by most consumers. With its thin leaf and acute bitter aftertaste, ‘Hybrid L’ essentially represents an extension of the curly kale market class.

Integrative observations

The addition of QMA to traditional quantitative sensory studies provides external validity and aids in the contextualization of the study’s quantitative components (Garber, Hyatt, & Starr, 2003). For example, the overall intensity of aroma was significantly different among samples in the descriptive analysis, but these differences did not influence overall consumer liking of aroma during CLT. A general theme noted in the QMA corroborates these observations, whereby aroma was consistently recorded as the least important sensory trait when consuming or purchasing kale. Provided breeding materials do not deviate significantly from current aroma profiles, this trait does not need to be a primary emphasis in future breeding programs. Continued application of QMA methodology within the leafy brassica breeding program may be beneficial on a regular basis (e.g. every five years), as

80

markets, trends, and consumer preferences can shift dramatically.

Though sensory diversity within the curly kale market class was minimal, kale and collard market classes were clearly distinct in their sensory characteristics.

Without understanding the underlying genetic or botanical differences, participants across studies identified sensory differences among kale, collard, and a closely related species (B. napus). This was most evident within the flavor and texture modalities of the descriptive analysis. Collard-like genotypes, including ‘Tiger’ and ‘Yellow IL’ were smooth, crisp, brittle, and retained a greater leaf thickness than traditional curly kale types. Genetic studies indicate rich diversity among landrace varieties

(Christensen et al., 2011; Dias & Monteiro, 1994; Pelc, Couillard, Stansell, &

Farnham, 2015) and a distinction among non-heading B. oleracea genotypes as a result of phytochemical and phenotypic differences (Hahn, Müller, Kuhnert, &

Albach, 2016). Additional molecular and phylogenetic studies are necessary to elucidate underlying taxonomic relationships among leafy Brassicas, but the sensory- oriented work presented herein reinforces many of the standard market and genetic divisions between kale and collard.

The current sensory landscape of leafy Brassicas is not inflexible and bound to the established kale-collard split. Products included in this study were intentionally selected to provide assessors with the range of currently available market classes and test hybrids with novel morphologies. Product sample ‘Yellow IL’, for example, was an inbred selection from Southern “cabbage collards.” It had a much brighter, yellow leaf-type than anything available in the commercial market. ‘P13-2’, a product similar to ‘Yellow IL’, was particularly divisive during QMA discussions, and was either the

81

top or bottom-rated sample among consumer clusters during preference mapping. This sample was associated with brittle, crunchy, smooth, and sour attributes during the descriptive study, and was compared to lettuce on multiple occasions during the CLT.

‘Yellow IL’ failed to fit the kale product concept, but it was highly liked by over one- third of participants. A comment by one consumer panelist summarized this trend:

“Seems like it's for people who don’t like kale but feel like they should eat it.” The targeted use of this inbred line in future breeding efforts could result in materials that expand the kale market class and cater to consumers who currently prefer softer leafy greens, such as lettuce or spinach.

Application to future breeding efforts

The aforementioned activities highlight both the benefits and limitations of the sensory analysis in providing succinct, directed plant traits to the plant breeder. Even with two distinct clusters driving the market, specific “drivers of liking” exist within leafy Brassicas. Sensory attributes such as thin leaves, little-to-no bitter aftertaste, little sour flavor, and darker colors were preferred across clusters. Test hybrids

(‘Hybrid I’, ‘Hybrid K1’, and ‘Hybrid L’) retained these attributes in ratios acceptable to most consumers in this study, and, encouragingly, represented new space within the market for new leafy Brassica products. It is now imperative that we begin identifying tangible and replicable means of quantifying these attributes within the leafy Brassica breeding program. The instrumental measurement of texture in kale is very laborious and requires costly texture analyzers, microscopy, and anatomical work to correlate these analyses with human perception. Other traits may be easier to screen using analytical chemistry, such as the flavor contributing to “greenness” or bitterness-

82

imparting compounds (Hongsoongnern & Chambers, 2008; Wieczorek et al., 2017).

Simon et al. (1981) describe the establishment of a program dedicated to improving flavor in carrots, and, in doing so, outlined a balanced approach between human sensory analysis and correlated laboratory analyses. Such an approach to breed for quality traits may be feasible within a leafy Brassica breeding program in the future.

The logistics and feasibility of formal sensory studies executed directly within a leafy Brassica breeding program must also be evaluated. Evaluating the stage at which sensory analysis is most pertinent to the breeding program and objectives is imperative. The breeding timeline varies dramatically depending on crop type, limited seed/plant material may hinder early generation sensory testing, and the number of traits under selection is often limited within a generation. Selection for consumer- defined quality traits must also account for correlated, sometimes detrimental, changes in plant physiology and performance, including those that alter crop productivity and disease resistance (Shewfelt & Brückner, 2000; Ulrich & Olbricht, 2011). Further, screening selective traits must be simple, rapid, and reproducible. Costs associated with formal sensory analysis may be unsustainable for publicly funded breeding programs, and sensory methodology amenable to the non-traditional settings in which plant breeders operate (e.g. open-field tastings, grower field days, conferences/expos, etc.) need to be explored. Rapid sensory profiling, a term encompassing alternative consumer sensory and marketing techniques, could contribute to the search for such amenable methods (Dawson & Healy, 2018; Frøst, Giacalone, & Rasmussen, 2015).

The understood definition of quality (Barrett, Beaulieu, & Shewfelt, 2010), which recognizes quality as the collective excellence of sensory characteristics, must

83

also be contextualized within the larger social and cultural framework of food systems

(Steenkamp, 1990). This includes consumer expectations, values, and consumer- defined quality. Sensory characteristics were significantly different among samples within the descriptive analysis and CLT portions of this study, but additional drivers of consumer choice, including health and nutrition credence attributes, may play a larger role in the purchasing and consumption of kale (Grunert, 2002). Plant breeders looking to engage and connect with consumers may find QMA to be beneficial as a hypothesis-generating activity, and facets of the QMA methodology, such as the focus group and value-building activity, may also be useful to plant breeders looking to probe items other than important sensory traits (e.g. new breeding methods and associated technologies). The supply chain from production-to-consumer is admittedly complex, but plant breeders, have a unique opportunity to address quality at the onset of plant development and set a precedent for quality products in downstream supply channels.

Conclusion

The multi-faceted structure of this research provided a comprehensive modelling of the leafy Brassica sensory space to be strategically deployed in the development of new leafy Brassica cultivars in the future. Consumer expectations for kale and related leafy Brassica genotypes appear to be well-defined in the US market.

Kale is envisioned as a shelf-stable, durable, bitter, and slightly sweet leafy green.

This report identified values and sensory traits associated with kale consumption through qualitative assessments, developed a standard lexicon for sensory testing of leafy Brassicas, revealed sensory differences between commercial and test hybrid kale

84

samples evaluated by a trained descriptive panel, and mapped consumer preference to identify clustering patterns in kale consumers. The sensory studies presented herein provide insight into the relative importance of sensory traits to kale consumers, which can inform future breeding and selection efforts within leafy Brassica breeding programs.

85

References

Albornoz, K. (2014). Postharvest characterization of kale (Brassica oleracea L. var. acephala DC): A study of the impact of leaf maturity, fresh-cut preparation and storage conditions. Master’s thesis. Available from ProQuest Dissertations and Theses database. (UMI No. 1560060). Amyotte, B., Bowen, A. J., Banks, T., Rajcan, I., & Somers, D. J. (2017). Mapping the sensory perception of apple using descriptive sensory evaluation in a genome wide association study. PLoS One; San Francisco, 12(2). https://doi.org/10.1371/journal.pone.0171710 Armesto, J., Gómez-Limia, L., Carballo, J., & Martínez, S. (2016). Effects of different cooking methods on some chemical and sensory properties of Galega kale. International Journal of Food Science & Technology, 51(9), 2071–2080. Azanza, F., Tadmor, Y., Klein, B. P., Rocheford, T. R., & Juvik, J. A. (1996). Quantitative trait loci influencing chemical and sensory characteristics of eating quality in sweet corn. Genome / National Research Council Canada = Genome / Conseil National de Recherches Canada, 39(1), 40–50. Baik, H.-Y., Juvik, J. A., Jeffery, E. H., Wallig, M. A., Kushad, M., & Klein, B. P. (2003). Relating Glucosinolate Content and Flavor of Broccoli Cultivars. Journal of Food Science, 68(3), 1043–1050. Barrett, D. M., Beaulieu, J. C., & Shewfelt, R. (2010). Color, Flavor, Texture, and Nutritional Quality of Fresh-Cut Fruits and Vegetables: Desirable Levels, Instrumental and Sensory Measurement, and the Effects of Processing. Critical Reviews in Food Science and Nutrition, 50(5), 369–389. Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software, Vol. 67, pp. 1–48. https://doi.org/10.18637/jss.v067.i01 Björkman, M., Klingen, I., Birch, A. N. E., Bones, A. M., Bruce, T. J. A., Johansen, T. J., … Stewart, D. (2011). Phytochemicals of Brassicaceae in plant protection and human health--influences of climate, environment and agronomic practice. Phytochemistry, 72(7), 538–556. Bowen, A. J., Blake, A., Tureček, J., & Amyotte, B. (2019). External preference mapping: A guide for a consumer-driven approach to apple breeding. Journal of Sensory Studies, 34(1), e12472. Causse, M., Buret, M., Robini, K., & Verschave, P. (2003). Inheritance of Nutritional and Sensory Quality Traits in Fresh Market Tomato and Relation to Consumer Preferences. Journal of Food Science, 68(7), 2342–2350. Chadwick, M., Gawthrop, F., Michelmore, R. W., Wagstaff, C., & Methven, L. (2016). Perception of bitterness, sweetness and liking of different genotypes of lettuce. Food Chemistry, 197, 66–74. Christensen, S., von Bothmer, R., Poulsen, G., Maggioni, L., Phillip, M., Andersen, B. A., & Jørgensen, R. B. (2011). AFLP analysis of genetic diversity in leafy kale (Brassica oleracea L. convar. acephala (DC.) Alef.) landraces, cultivars and wild populations in Europe. Genetic Resources and Crop Evolution, 58(5), 657–666.

86

Civille, G. V., & Oftedal, K. N. (2012). Sensory evaluation techniques - make “good for you” taste “good.” Physiology & Behavior, 107(4), 598–605. Civille, G. V., & Szczesniak, A. S. (1973). Guidelines to training a texture profile panel. Journal of Texture Studies, 4(2), 204–223. Colquhoun, T. A., Levin, L. A., Moskowitz, H. R., Whitaker, V. M., Clark, D. G., & Folta, K. M. (2012). Framing the perfect strawberry: An exercise in consumer- assisted selection of fruit crops. Journal of Berry Research, 2(1), 45–61. Dawson, J. C., & Healy, G. K. (2018). Flavour Evaluation for Plant Breeders. In I. Goldman (Ed.), Plant Breeding Reviews (Vol. 22, pp. 215–261). Hoboken, NJ, USA: John Wiley & Sons, Inc. Dias, J. S., & Monteiro, A. A. (1994). Taxonomy of Portuguese Tronchuda cabbage and Galega kale landraces using morphological characters, nuclear RFLPs, and isozyme analysis: A review. Euphytica/ Netherlands Journal of Plant Breeding, 79(1), 115–126. Diederichsen, A. (2001). Mansfeld’s Encyclopedia of Agricultural and Horticultural Crops: (Except Ornamentals) (pp. 1435–1446; R. Mansfeld & R. Büttner, Eds.). Springer Science & Business Media. Dixon, G. R. (Ed.). (2006). Vegetable brassicas and related crucifers. Wallingford: CABI. Drake, S. L., Lopetcharat, K., & Drake, M. A. (2009). Comparison of two methods to explore consumer preferences for cottage cheese. Journal of Dairy Science, 92(12), 5883–5897. Earle, M. D., Earle, R. L., & Anderson, A. M. (2001). Food Product Development (pp. 380). Cambridge, England: Woodhead Publishing. Franks, M., Lawrence, P., Abbaspourrad, A., & Dando, R. (2019). The Influence of Water Composition on Flavor and Nutrient Extraction in Green and Black Tea. Nutrients, 11(1). https://doi.org/10.3390/nu11010080 Frøst, M. B., Giacalone, D., & Rasmussen, K. K. (2015). 17 - Alternative methods of sensory testing: Working with chefs, culinary professionals and brew masters. In J. Delarue, B. Lawlor, & M. Rogeaux (Eds.), Rapid sensory profiling techniques (pp. 363–382). doi: 10.1533/9781782422587.3.363 Garber, L. L., Hyatt, E. M., & Starr, R. G. (2003). Measuring consumer response to food products. Food Quality and Preference, 14(1), 3–15. Gepts, P. (2002). A Comparison between Crop Domestication, Classical Plant Breeding, and Genetic Engineering. Crop Science, 42(6), 1780–1790. Gilbert, J. L., Guthart, M. J., Gezan, S. A., Carvalho, M. P. de, Schwieterman, M. L., Colquhoun, T. A., … Olmstead, J. W. (2015). Identifying Breeding Priorities for Blueberry Flavor Using Biochemical, Sensory, and Genotype by Environment Analyses. PloS One, 10(9), e0138494. Gilbert, J. L., Olmstead, J. W., Colquhoun, T. A., Levin, L. A., Clark, D. G., & Moskowitz, H. R. (2014). Consumer-assisted Selection of Blueberry Fruit Quality Traits. HortScience: A Publication of the American Society for Horticultural Science, 49(7), 864–873. Gómez-Campo, C. (Ed.). (1999). Biology of Brassica coenospecies. Amsterdam: Elsevier.

87

Groenbaek, M., Jensen, S., Neugart, S., Schreiner, M., Kidmose, U., & Kristensen, H. L. (2016). Nitrogen split dose fertilization, plant age and frost effects on phytochemical content and sensory properties of curly kale (Brassica oleracea L. var. sabellica). Food Chemistry, 197(Pt A), 530–538. Grunert, K. G. (2002). Current issues in the understanding of consumer food choice. Trends in Food Science & Technology, 13(8), 275–285. Hahn, C., Müller, A., Kuhnert, N., & Albach, D. (2016). Diversity of Kale (Brassica oleracea var. sabellica): Glucosinolate Content and Phylogenetic Relationships. Journal of Agricultural and Food Chemistry, 64(16), 3215– 3225. Hammer, K., Gladis, T., Laghetti, G., & Pignone, D. (2013). The wild and the grown – remarks on Brassica. International Journal of AgriScience, 3(6), 453–480. Hansen, M., Laustsen, A. M., Olsen, C. E., Poll, L., & Sørensen, H. (1997). Chemical and sensory quality of broccoli (Brassica oleracea L. var Italica). Journal of Food Quality, 20(5), 441–459. Higdon, J. V., Delage, B., Williams, D. E., & Dashwood, R. H. (2007). Cruciferous vegetables and human cancer risk: epidemiologic evidence and mechanistic basis. Pharmacological Research: The Official Journal of the Italian Pharmacological Society, 55(3), 224–236. Hongsoongnern, P., & Chambers, E., Iv. (2008). A lexicon for green odor or flavor and characteristics of chemicals associated with green. Journal of Sensory Studies, 23(2), 205–221. Ishihara, S. (1936). Series of plates designed as tests for colour-blindness. Tokyo, JP: Kanehara Trading Inc. ISO. (1994). Sensory analysis - Methodology - Texture profile (No. 11036). Retrieved from International Organization for Standardization website: https://www.iso.org/standard/19016.html ISO. (2009). Milk and milk products - Sensory analysis. Part 1: General guidance for the recruitment, selection, training, and monitoring of assessors (No. 22935- 1). https://www.iso.org/standard/41216.html: International Organization for Standardization. ISO. (2014). Sensory analysis - General guidelines for the selection, training, and monitoring of selected assessors and expert sensory assessors (No. 8586). https://www.iso.org/standard/45352.html: International Organization for Standardization. Jaeger, S. R., Rossiter, K. L., Wismer, W. V., & Harker, F. R. (2003). Consumer driven product development in the kiwifruit industry. Food Quality and Preference, 14(3), 187–198. Jahangir, M., Kim, H. K., Choi, Y. H., & Verpoorte, R. (2009). Health-Affecting Compounds in Brassicaceae. Comprehensive Reviews in Food Science and Food Safety, 8(2), 31–43. Jones, M. G. (2008). Formation of vegetable flavour. In: B. Brückner & S. G. Wyllie (Eds.), Fruit and vegetable flavor: Recent advances and future prospects (pp. 85-88). doi: 10.1533/9781845694296.2.71 Kobori, C. N., Huber, L. S., Sarantópoulos, C. I. G. L., & Rodriguez-Amaya, D. B.

88

(2011). Behavior of flavonols and carotenoids of minimally processed kale leaves during storage in passive modified atmosphere packaging. Journal of Food Science, 76(2), H31–H37. Kopsell, D. E., Kopsell, D. A., Randle, W. M., Coolong, T. W., Sams, C. E., & Curran Celentano, J. (2003). Kale carotenoids remain stable while flavor compounds respond to changes in sulfur fertility. Journal of Agricultural and Food Chemistry, 51(18), 5319–5325. Lawless, H. T., & Heymann, H. (2010). Data Relationships and Multivariate Applications. In Food Science Text Series. Sensory Evaluation of Food (pp. 433–449). Springer, New York, NY. Lawless, L. J. R., & Civille, G. V. (2013). Developing Lexicons: A Review. Journal of Sensory Studies, 28(4), 270–281. Lenth, R. (2018). emmeans: Estimated Marginal Means, aka Least-Squares Means. Retrieved from https://CRAN.R-project.org/package=emmeans Li, F., Hullar, M. A. J., Schwarz, Y., & Lampe, J. W. (2009). Human gut bacterial communities are altered by addition of cruciferous vegetables to a controlled fruit- and vegetable-free diet. The Journal of Nutrition, 139(9), 1685–1691. Lopetcharat, K., & Beckley, J. (2012). Qualitative Multivariate Analysis. In Product Innovation Toolbox (pp. 100–121). Oxford, UK: Wiley-Blackwell. Martens, M., Martens, H., & Wold, S. (1983). Preference of cauliflower related to sensory descriptive variables by partial least squares (PLS) regression. Journal of the Science of Food and Agriculture, 34(7), 715–724. Mintel. (2018). Executive Summary: Healthy Dining Trends, US, March 2018. Retrieved from Mintel website: http://academic.mintel.com/display/860323/ Pelc, S. E., Couillard, D. M., Stansell, Z. J., & Farnham, M. W. (2015). Genetic diversity and population structure of collard landraces and their relationship to other Brassica oleracea crops. The Plant Genome, 8(3), 1-11. doi:10.3835/plantgenome2015.04.0023 Produce Market Guide. (2018). Commodity: Kale. Retrieved from The Packer website: https://www.producemarketguide.com/produce/kale Radovich, T. J. K. (2010). Cabbage Flavor. In Handbook of Fruit and Vegetable Flavors (pp. 741–750). Hoboken, NJ, USA: John Wiley & Sons, Inc. R Core Team. (2018). R: A Language and Environment for Statistical Computing. Retrieved from https://www.R-project.org/ Schonhof, I., Krumbein, A., & Brückner, B. (2004). Genotypic effects on glucosinolates and sensory properties of broccoli and cauliflower. Die Nahrung, 48(1), 25–33. Shewfelt, R. L., & Brückner, B. (Eds.). (2000). Fruit & vegetable quality: An integrated view. Lancaster, PA: Technomic Pub. Co. Retrieved from http://newcatalog.library.cornell.edu/catalog/3808444 Simon, P.W., Peterson, C. E., & Lindsay, R. C. (1981). The improvement of flavor in a program of carrot genetics and breeding. In R. Teranishi and H. Barrera- Benitez (Eds), ACS Symposium Series: Quality of selected fruits and vegetables of North America (pp. 109–118). American Chemical Society, Washington, DC. doi: 10.1021/bk-1981-0170.ch009

89

Sistrunk, W. A. (1980). Kale greens quality, vitamin retention and nitrate content as affected by preparation, processing, and storage. Journal of Food Science, 45(3), 679–681. Steenkamp, J.-B. E. M. (1990). Conceptual model of the quality perception process. Journal of Business Research, 21(4), 309–333. Stone, H., Sidel, J., Oliver, S., Woolsey, A., & Singleton, R. C. (2004). Sensory Evaluation by Quantitative Descriptive Analysis. In M. C. Gacula Jr. (Ed.), Descriptive Sensory Analysis in Practice (pp. 23–34). Trumbull, Connecticut, USA: Food & Nutrition Press, Inc. Sugrue, M., & Dando, R. (2018). Cross-modal influence of colour from product and packaging alters perceived flavour of cider. Journal of the Institute of Brewing. Institute of Brewing, 124(3), 254–260. Talavera-Bianchi, M., Chambers, E., IV, & Chambers, D. H. (2010). Lexicon to describe flavor of fresh leafy vegetables. Journal of Sensory Studies, 25(2), 163–183. Tanaka, Y., Sasaki, N., & Ohmiya, A. (2008). Biosynthesis of plant pigments: anthocyanins, betalains and carotenoids. The Plant Journal: For Cell and Molecular Biology, 54(4), 733–749. Thavarajah, D., Thavarajah, P., Abare, A., Basnagala, S., Lacher, C., Smith, P., & Combs, G. F. (2016). Mineral micronutrient and prebiotic carbohydrate profiles of USA-grown kale (Brassica oleracea L. var. acephala). Journal of Food Composition and Analysis: An Official Publication of the United Nations University, International Network of Food Data Systems, 52, 9–15. Tieman, D., Zhu, G., Resende, M. F. R., Lin, T., Nguyen, C., Bies, D., … Klee, H. (2017). A chemical genetic roadmap to improved tomato flavor. Science, 355(6323), 391–394. Ulrich, D., & Olbricht, K. (2011). Fruit organoleptic properties and potential for their genetic improvement. In T. A. Jenks & P. J. Bebeli (Eds.), Breeding for Fruit Quality (pp. 39–59). John Wiley & Sons, Inc. doi: 10.1002/9780470959350.ch2 USDA-NASS. (2019). Table 36. Vegetables, Potatoes, and Melons Harvested for Sale: 2017 and 2012 [Data set]. 2017 Census of Agriculture. Retrieved from https://www.nass.usda.gov/Publications/AgCensus/2017/Full_Report/Volume_ 1,_Chapter_1_US/st99_1_0036_0036.pdf van Doorn, H. E., van der Kruk, G. C., van Holst, G. J., Raaijmakers‐Ruijs, N., Postma, E., Groeneweg, B., & Jongen, W. (1999). The glucosinolates sinigrin and progoitrin are important determinants for taste preference and bitterness of Brussels sprouts. Journal of the Science of Food and Agriculture, 78(1), 30– 38. van Kleef, E., van Trijp, H. C. M., & Luning, P. (2006). Internal versus external preference analysis: An exploratory study on end-user evaluation. Food Quality and Preference, 17(5), 387–399. Wickham, H. (2016). ggplot2: Elegant Graphics for Data Analysis. Retrieved from http://ggplot2.org Wieczorek, M. N., Walczak, M., Skrzypczak-Zielińska, M., & Jeleń, H. H. (2017).

90

Bitter taste of Brassica vegetables: The role of genetic factors, receptors, isothiocyanates, glucosinolates, and flavor context. Critical Reviews in Food Science and Nutrition, 58(18), 3130-3140. Yan, K. S., & Dando, R. (2015). A crossmodal role for audition in taste perception. Journal of Experimental Psychology. Human Perception and Performance, 41(3), 590–596. Zhang, X., Shu, X.-O., Xiang, Y.-B., Yang, G., Li, H., Gao, J., … Zheng, W. (2011). Cruciferous vegetable consumption is associated with a reduced risk of total and cardiovascular disease mortality. The American Journal of Clinical Nutrition, 94(1), 240–2.

91

Chapter 3

FLASH-PROFILING COLLARD (BRASSICA OLERACEA VAR. VIRIDIS)

MARKETS TO IDENTIFY REGION-SPECIFIC PREFERENCES

Abstract

Collards have formed regional identities and food cultures in both the Southern

U.S. and East Africa. Through in-situ consumer testing of six collard genotypes, we sought to identify consumer lexicons, underlying drivers of liking, and regional preferences specific to Upstate New York and Western Kenyan consumers. Flash

Profiling (FP) with untrained consumer panels was deployed to obtain region-specific descriptive attributes of raw and cooked collard product. Consensus configurations accounted for 29-40% of the original variance in FP attribute datasets. Raw collard products were well differentiated in both regions, and descriptive attributes in the sensory modalities of color (green, light-green) and shape (curly, broad, long/tall) were common to both regions. New York panelists generated approximately twice as many descriptive attributes as Kenyan panelists who, despite much greater product familiarity, frequently generated hedonic terms. Cooked products generated far fewer descriptors and were not as readily discriminated in consensus plots. Consumer acceptance tests identified a significant (p < 0.05) cultivar-by-country interaction.

Participants in Western Kenya readily identified their local variety, ‘Mfalme’, and intraspecific collard-broccoli hybrids scored well in both regions. This project highlights the importance of recognizing familiarity, cultural relevance, and linguistic

92

consideration in cross-cultural sensory studies. In addition, it provides descriptive and hedonic product information to maintain horticultural quality in each region.

Introduction

The Brassicaceae plant family consists of morphologically diverse crop species with global relevance. Cole crops (Brassica oleracea L.), such as broccoli, cabbage, cauliflower, and kale, are among the most widely consumed vegetable species in the

Brassicaceae family and have significant economic impact across food systems.

Botanical varieties of leafy vegetable Brassicas, including cabbage (B. oleracea var. capitata), kale (B. oleracea var. acephala), collard (B. oleracea var. viridis), and kailaan (B. oleracea var. alboglabra) are consumed worldwide and well known for their nutritive and phytochemical properties. Non-heading varieties of leafy cole crops are uniquely distinct from other widely consumed leafy Brassica vegetables such as

Ethiopian kale (B. carinata), rape kale (B. napus), mustards (B. juncea), or B. rapa vegetables. These loose-leafed, non-heading botanical varieties are differentiated by their prolonged shelf-life, waxy leaf surface, tolerance to heat, and flexible production/harvest windows (Dixon, 2007). References to cultivation of leafy

Brassicas extend to the Greek and Roman Empires (Maggioni et al., 2018; Snogerup,

1980), though by late 1500s these vegetable crops were cultivated in gardens of western Europe and the British Isles (Mitchell, 1976). Regional distribution of leafy varieties exists today across Europe, Africa, and North America.

Morphologically similar leaf types, often referred to as collard or “leafy cabbage” types, are widely grown in both the southeast U.S. and easern Africa. These types are characterized by flat, oblong leaves with entire margins or a slight curl to the

93

leaf edges. Open-pollinated collard genotypes grown in both the Southern U.S. and

Eastern Africa are likely admixtures of viridis (collard), costata (Portuguese cabbage/kale), and medullosa (marrow stem kale) botanical varieties. Collards show genetic resemblance to cabbage and Brussel’s sprouts (Farnham, 1996), and U.S. landraces contain considerable genetic diversity not represented in commercial cultivars (Pelc et al., 2015). Market classes in the Southeastern U.S. are recognized under numerous common names (e.g. Georgia Southern, Thousand Headed Kale,

Vates, Morris Heading, etc.). In the East African region, these leafy greens are commonly referred to as “sukuma wiki”, “African kale”, or “African collard.”

Many conjectures as to how collard seed of these types arrived in the Southern

U.S. and Eastern Africa exist, but the dispersion of non-heading collard varieties from

British and European cultivation regions, namely Portuguese and Spanish, is not well understood. Most dispersal theories emphasize the role of African slaves, European explorers, and colonial settlers in introducing collards to North American gardens between 1500 and 1820 (Davis and Morgan, 2005; Sauer, 1993). Numerous countries in Eastern and Southern Africa commercially produce non-heading varieties today, but

Davis and Morgan (2015) reported a palpable lack of records pertaining to Brassica vegetable production in Sub Saharan Africa prior to 1830. Leafy cabbages may have been introduced historically (i.e. pre-European colonialism) to East African highlands, where seed production is possible, but it is more likely that these types were introduced to the region by European settlers in the nineteenth century (Davis and

Morgan, 2015). Since their introduction to these regions, collards have generated unique food cultures in Southeastern U.S. and East African foodways.

94

Collards remain an important crop to southern U.S. food systems, backyard gardeners, and seed savers (Davis and Morgan, 2015; Farnham et al., 2008).

Approximately 58% of U.S. collard acreage (11,318 acres) resides in Georgia, South

Carolina, and North Carolina (USDA-NASS, 2019). Braised collard greens, often slow cooked with a ham hock or pork rind, are a common side dish during mealtimes and a staple during family and community gatherings. Production in the northern U.S. is much less common and consumer expectations in the northern U.S. are poorly defined. African sukuma wiki is a main food crop in the East African region and especially important among subsistence farmers in rural Kenya (One Acre Fund,

2014). Sukuma wiki’s namesake translates from Swahili as “to push the week,” which aptly describes its ability to withstand continuous harvests over many months. In many Kenyan households, this leafy green is commonly sautéed and served with ground corn (“ugali”) as the main meal and serves as a side dish when more expensive meat, eggs, or beans are available. Despite dramatically different production practices, market outlets, and end-users, collards retain a strong identity as a traditional food product in the Southeast U.S. and Eastern and Southern Africa.

Existing expectations for quality traits and embedded food culture have constrained plant ideotypes in both regions. Farmer-directed seed saving has historically occurred in both regions and occasional vegetative propagation of sukuma wiki for personal consumption can be found among Kenyan households (Mvere and van der Werff, 2004). It is difficult to assess the extent to which seed saving practices, genotypic admixtures, and selection criteria have driven regional ideotypes, but shifts in genetic diversity may have occurred in response to regional preferences and end-use

95

of the product. Both regions have seen a decline in seed saving as commercially available seed, including new hybrid varieties, have come to market. The U.S. collard market is dominated by two or three hybrid varieties (e.g. ‘Top Bunch’ (Sakata Seed,

Japan), ‘Flash’ (Sakata Seed, Japan), and ‘Hi Crop’ (Takii Seed, Japan)] (Farnham et al., 2008). Diverse open pollinated varieties are sold in Eastern Africa as sukuma wiki, and a hybrid variety of sukuma wiki [‘Mfalme’ (East African Seed Co., Kenya)] recently entered Kenyan markets (Farmers Trend, 2016). The success of ‘Mfalme’ and other hybrid cultivars speaks broadly to increased acceptance and use of hybrid seed in vegetable production systems, but it also raises opportunities to understand regional preferences and market flexibility in current collard market classes. Intraspecific diversity within the B. oleracea species and, more specifically, diversity within leafy

B. oleracea germplasm, could be utilized in directed breeding efforts to fundamentally shift regionally adapted ideotypes.

Understanding and managing consumer quality expectations, is critical during the development and introduction of horticultural crops for new markets. The alteration of a traditional food product in such a way that it detracts from the habitual and cultural relevance could be damaging to the underlying food culture (Guerrero et al., 2009). The introduction of novel collard/sukuma wiki cultivars must recognize existing intrinsic product expectations (e.g. taste, aroma, nutrition) and contextualize these expectations within the larger cultural identity of the food product. Materials in the Southeastern U.S. and East Africa are marketed and prepared in dramatically different manners, complicating the direct comparison of quality perceptions across collard production regions. It is presumptuous to assume materials from external

96

breeding programs are productive and applicable in markets beyond their locality.

This is particularly relevant in the case of collards, a vegetable crop known for high glucosinolate concentration (Carlson et al., 1987; Stansell et al., 2015) and the identifiable bitter/pungent flavors which result from their breakdown products (Cartea and Velasco, 2008; Wieczorek et al., 2018). Haplotypes of the bitter taste receptor gene (TAS2R38) have been identified in human populations around the globe (Risso et al., 2016). Regional differences in flavor sensitivity and detection may have altered current plant morphotypes and could potentially affect the cultural acceptance of new cultivars in these markets. The establishment of region-specific breeding priorities, from both agronomic and end-user viewpoints, are critical to the long-term success and adoption of new cultivars (Brouwer et al., 2016; Ceccarelli and Grando, 2007).

Region-specific breeding priorities may also help develop new and emerging markets, such as the northeastern U.S. collard market.

Sensory analysis is a tool that can lead to strategic decision-making, helping to generate new products accepted by consumers while ensuring quality standards. It is also a tool that can inform the plant breeding process and promote directed cultivar adoption within new markets (Bechoff et al., 2016; Dawson and Healy, 2018).

Descriptive analysis is a key area of sensory analysis that seeks to obtain objective, replicable sensory descriptors of products. Techniques such as Quantitative

Descriptive Analysis (QDA®; Stone, 1992) and Spectrum Descriptive Analysis

Method (Muñoz and Civille, 1992) are well established means of conducting descriptive analysis and profiling products. Formal descriptive analysis and sensory profiling, however, is still a relatively under-utilized means of evaluating new

97

vegetable breeding lines or varieties. Descriptive studies are often conducted in highly-controlled settings, involve training a panel of experienced raters, require experienced sensory coordinators, and can be resource-intensive (Lawless and

Heymann, 2010a; Varela and Ares, 2012). The end products of a descriptive study, however, are beneficial to researchers and product developers in that replicable, quantitative means of profiling underlying product sensory attributes.

Rapid sensory profiling employs alternative consumer sensory and marketing techniques to acquire descriptive product information (Delarue et al., 2014). Flash

Profiling (FP) is a rapid sensory technique that has recently gained popularity as a viable alternative to conventional descriptive analysis. The FP method is a variation on Free-Choice Profiling, whereby sensory panelists develop their own list of terms to describe a product set and subsequently use these terms to rank the relative intensity of each product (Dairou and Sieffermann, 2002; Delarue, 2015). Though certainly not used as extensively as other descriptive methods, FP has the potential to provide quality descriptive information about multiple products (i.e. collard genotypes) in a faster, more flexible manner than traditional descriptive analysis (Blancher et al.,

2007; Bredie et al., 2018; Dehlholm et al., 2012). Further, FP may potentially reduce confounding factors of language and panelist expertise. Several studies have used FP to address issues in cross-cultural sensory profiling (Jamir et al., 2020; Park et al.,

2017) and profiling methods conducted with untrained consumer panelists (Ballay et al., 2015; Blancher et al., 2007; Ramírez‐Rivera et al., 2017).

The following study addresses key stages in understanding quality traits pertinent to the dissemination of new collard cultivars and, notably, seeks to engage

98

local end-users in the advancement of breeding lines to cultivars. In an effort to tailor

Cornell AgriTech’s leafy Brassica breeding program to the diverse environments in which collards are grown, this study sought to elucidate consumer lexicons and preferences from two distinct regions: Western Kenya and New York, USA.

Evaluations conducted in these two regions enabled the comparison of panelists with varying degrees of product familiarity and the ease with which rapid profiling techniques may be implemented when language and cultural distinctions exist. Using

FP and in-situ acceptance testing, we hypothesized consumers would develop similar lexicons in both Western Kenya and New York, though overall cultivar preferences would differ between the two regions. Through understanding preferences and market requirements in each region, we hope to explore opportunities to improve the production and consumption of leafy greens in both regions prior to significant investment in the development, promotion and commercialization of new genotypes.

Materials and Methods

Plant materials and production

Plant materials evaluated throughout this study included commercially available collard cultivars and testcross hybrids currently under development in the

Cornell AgriTech breeding program (Geneva, NY) (Table 3.1), herein referred to as

“genotypes.” Seed of testcross hybrids were generated during the winter of 2018 in

Geneva, NY from field-grown, inbred parental lines (selfed to the S4 generation or beyond) based on promising hybrid combinations from previous trialing. Field materials were vernalized for approximately ten weeks in a temperature controlled nursery cellar (8˚C) prior to their placement in the Geneva Greenhouse complex

99

(42°52’31.7”N, 77°00’28.0”W) to initiate flowering for crossing and seed production in the spring of 2018. Commercially available cultivars were sourced from seed companies in Kenya and the U.S. in the spring of 2018. All plant materials were evaluated for uniformity and seedling vigor in the greenhouse prior to establishment of field trials. A seed import permit (Permit No. KEPHIS/SEED/0007664) was generated and a phytosanitary export certificate (PPQ Form 577) was approved in the winter of

2019 to ship materials to collaborators at Advantage Crops Limited (ACL) in Rodi

Kopany, Homa Bay County, Kenya for field testing.

Field trials of mature plant materials were established at three sites in Western

Kenya and two sites in Upstate New York (Appendix 3.1). Kenyan field sites were established from transplanted seedlings grown at ACL certified production site.

During the second week of April 2019, five-week old seedlings were planted in a randomized complete block design (r = 3) at each location. Plots consisted of six plants spaced 60cm within and between mounded rows. Farmer managers at each site were given discretion to maintain plots using standard production practices and encouraged to minimize weed and pest pressure.

The two field sites in Upstate New York consisted of one certified organic site

(Freeville) and one conventional field site (Geneva). Seedlings were produced in either the certified organic greenhouse at the Guterman Bioclimatic Laboratory

Complex or greenhouses at Cornell AgriTech. During the first week of July, four- week old seedlings were planted in randomized complete block design (r = 3).

Materials were transplanted into raised, black plastic beds spaced at 50cm with 2m row centers with drip-fed irrigation. Immediately after transplanting, organic materials

100

Table 3.1. Hybrid collard germplasm used in consumer acceptance and Flash Profiling evaluations in Western Kenya (May ‘19) and Upstate New York (Sept. ‘19). Flash Seed Field Consumer Genotype Breeder Profile Description* Photo Source Code Code Code Circular leaf shape from a cross between a ‘Champion S2 self-pollinated Breeding Cornell AgriTech X V03 709 B OP Georgia- Material (Geneva, NY, US) KP1702’ type collard and a golden cabbage-collard

inbred line Oblong leaf shape with soft- curled edge ‘KP1706 from a cross Breeding Cornell AgriTech X V04 591 D between Material (Geneva, NY, US) KP15’ borecole curly kale and an inbred collard

line Oblong leaf shape with undulate edges ‘P8CMS from a cross Breeding Cornell AgriTech X V06 141 E between male- Material (Geneva, NY, US) KP1’ sterile broccoli and an inbred Georgia-type

collard Long leaf with undulate edges from a cross ‘P8CMS Breeding Cornell AgriTech between male- X V07 982 A Material (Geneva, NY, US) sterile broccoli KP1702’ and a golden cabbage-collard

inbred line Crinkled Tuscan-collard leaf with oval East African Seed shape and plant ‘Mfalme F1’ Syova Co. V09 267 C habit (Nairobi, Kenya) resembling a cabbage in

cooler climates Portuguese Troncuda kale with very large Bejo Zaden, B.V. leaves and ‘Beira F1’ Seedway (Warmenhuizen, V11 356 F prominent Netherlands) white midribs;

short, loosely headed plant Georgia-type Sakata Seed hybrid collard; Stokes America oblong leaves ‘Top Bunch F1’ V12 NA C or A Seeds (Yokohama, with slight Japan) savoy along edges

101

in Freeville were covered with a floating row cover (Agribond®+ AG-19, Berry

Plastics, San Luis Potosí, México) to avoid herbivory damage prior to sensory evaluation. The conventional field site was maintained using standard practices at the

Cornell AgriTech research farm.

Approximately eight-to-ten weeks after transplanting, the third and fourth most fully expanded leaves were harvested from each genotype, bunched, and assigned a random three-digit code for sensory testing. In all locations, field materials were harvested early in the morning and within four hours of conducting sensory testing.

All leaves used in sensory testing were free of decay and herbivory damage.

Human subject assurance

All procedures involving human subjects were reviewed and approved by the

Cornell University Institutional Review Board (Protocol ID#: 1903008671).

Participants were provided with written consent forms and the opportunity to ask researchers any questions prior to their participation. All recruitment materials, consent forms, product codes, and data are available on Mendely Data.

Consumer acceptance testing

On-site acceptance testing of six collard/sukuma wiki genotypes was performed in May/June 2019 (Kenya, n = 3 testing sites) and August/September 2019

(New York, n = 3 testing sites) (Table 3.2). Evaluations capitalized on field days, existing collaborations, and public gatherings to conduct sensory testing outside of a controlled sensory center. Participants were required to be 18 years of age or older and consume collard/sukuma wiki at least a few times per year. Approximately n = 30 participants, the majority of whom were food and agricultural professionals, engaged

102

in hedonic evaluations of collard materials at each testing site.

Table 3.2. Site information for acceptance evaluations of hybrid collard germplasm in Western Kenya (May ‘19) and Upstate New York (Sept. ‘19). Date Code Location Event Description Participants (mm.dd.yyyy) Aram Aram, Kenya Field Site Gathering 5.31.2019 n = 28 (7F) Samanga Nyangweso, Kenya Field Site Gathering 6.3.2019 n = 25 (7F) ACL Rodi Kopany, Kenya Field Site Gathering 6.5.2019 n = 31 (20F) Freeville Freeville, NY Organic Field Day 7.31.2019 n = 29 (15F) Ithaca Ithaca, NY Outdoor Testing at The Westy Bar 9.11.2019 n = 44 (14F) Geneva Geneva, NY Cornell AgriTech Coffee Break 9.20.2019 n = 29 (14F)

All genotypes were sampled in cooked product form, whereby ~0.45kg of raw, whole leaves were thinly sliced (<1cm wide), quick sautéed (6 minutes) in 4T vegetable oil and 1tsp table salt. Materials in Kenya were prepared over charcoal and immediately placed in hotpots to keep warm. Instant Pots® (Duo™SV 6-Quart,

Instant Brands Inc, Ontario, Canada) were used to sauté and keep materials warm during New York testing. Approximately 4-5g of each cooked sample was placed in a plastic portion cup (S&B Computer and Office Product, Inc, NY, US) labeled with a single alphabet letter (A-F) before serving to each panelist. Participants were instructed to rate (9-point hedonic scale) their overall liking of each sample on a paper ballot. Due to the in-situ nature of these sensory tests, concessions were made with regard to sampling order and monadic serving. All participants were encouraged to sample in any order and rate each sample independently.

Rapid sensory analysis: Flash Profiling (FP)

A focus group consisting of untrained panelists (n = 12-15) who like and consume collards/sukuma wiki was assembled in each testing region (Table 3.3).

Through a facilitated Flash Profiling (FP) session, panelists were asked to provide descriptive attributes and rankings of six collard/sukuma wiki genotypes in both raw

103

and cooked preparations. Panelists were asked to assess raw products visually and tactilely, and cooked product was prepared in accordance with aforementioned methods used in consumer acceptance testing. Flash Profiling was conducted in a single session divided into two steps: ballot creation and product ranking. Panelists were given an overview of the study and verbal directions prior to receiving written directions in English; participants in Kenya were encouraged to complete both the ballot development and ranking in either English, Swahili, or their local Dholuo language (Luo). Ballots and rankings were translated as needed by in-house by employees of Advantage Crops Limited (ACL).

Table 3.3. Summary of Flash Profiling descriptive analysis with untrained panelists in Rodi Kopany, Kenya (Homa Bay County) and Geneva, NY (New York). Panelist descriptors were subject to GPA transformations and a PCA was produced to highlight product separation Location Method Configurations Sensory Attributes (Frequencies) Rc1 F1 & F2 (Participants) Unfiltered Filtered % Hedonic Value Variance2 Homa Bay Raw n = 12 109 85 17% 0.294 67.43% New York Raw n = 15 176 163 1% 0.398 63.20% Homa Bay Cooked n = 12 70 63 43% 0.263 88.45% New York Cooked n = 14 108 98 1% 0.292 58.33% 1 Rc = proportion of original variance explained by GPA consensus configuration 2 Total variance explained by the first two (F1 & F2) PCA axes after GPA procedure

In our modified procedure, a pooled list of generated attributes was not developed among panelists (Bredie et al., 2018; Dairou and Sieffermann, 2002). For each preparation method, panelists were simultaneously presented with the entire product set (n = 6 genotypes, each assigned a random three-digit code) and asked to create a list of non-hedonic attributes to describe the genotypes. These attributes were then categorized into various sensory modalities, including flavor, aroma, texture, and

104

others. In the second step, panelists ranked the intensity of each genotype from low-to- high for each attribute created during ballot development. Panelists were allowed to include new attributes defined during the ranking step. This method produced an individual matrix for each panelist, whereby each genotype was ranked according to panelist-defined attributes. Panelists were provided with water, unsalted crackers as palette cleansers, snacks, and were compensated for their participation.

Statistical procedures and multivariate analysis

Consumer hedonic scores were treated as continuous variables and subject to multivariate statistical analysis. Quantitative data were analyzed using mixed models composed in the lme4 package (Bates et al., 2015) of R Software (v3.5.0) (R Core

Team, 2018). A complete model including the fixed interaction between genotype and country, fixed effects of age and gender, and random effects of testing site and participant was generated. Additional mixed models without the interaction effect were set up to evaluate the effect of variety within each testing region. Random effect variance components were extracted from each model, and a two-way analysis of variance (ANOVA) was performed on consumer hedonic scores (Appendix 3.2) to test differences among sample means; sample effects were declared significant below ⍺ =

0.05. Where appropriate, the emmeans package (Lenth et al., 2020) was used to generate a compact letter display with a Tukey’s adjusted p-value for comparing estimates (Appendix 3.3).

Flash Profile ranking data were subject to Generalized Procrustes Analysis

(GPA; (Gower, 1975) in XLSTAT-Sensory software (XLSTAT Version

2018.5.52460, Addinsoft, Paris, France), whereby a consensus configuration of

105

panelist ratings was obtained through a series of transformations intended to reduce scaling effects among panelists. Panelist attributes without complete rankings or ties were removed from the dataset. Obvious hedonic attributes were removed from downstream analysis. Following GPA transformations using the Commandeur method

(Commandeur, 1991), a principal component analysis (PCA) was produced from the consensus configuration.

To understand the relationship between consumer hedonic scores and FP ratings, agglomerative hierarchical clustering (AHC) and external preference mapping were performed on cooked product evaluations in Kenya. There was insufficient representation of each cultivar in New York evaluations to perform hierarchical clustering and preference mapping. Cluster analysis was performed on hedonic scores for overall liking using a Euclidean dissimilarity matrix, Ward’s agglomeration method (Ward, 1963), and the automatic truncation (i.e. entropy) settings in XLSTAT-

Sensory. Resulting AHC class centroids (i.e. mean hedonic ratings for each sample) were regressed onto the PCA object coordinates from the GPA consensus configuration using PREFMAP function in XLSTAT-Sensory to produce external preference maps (Lawless and Heymann, 2010b; van Kleef et al., 2006). External preference maps for each testing region were superimposed onto a contour plot, displaying clusters with above-average preference throughout regions of the map.

Graphics were developed using a combination of XLSTAT-Sensory and the ggplot2 package (Wickham, 2016) of R software.

106

Results

Regional consumer acceptance

Consumer acceptance evaluations of six collard hybrids engaged 200 participants in Kenya [n = 98 (36F)] and New York [n = 102 (43F)]. After filtering for missing data, a total of 186 participants were used in final modeling and data analysis

(Table 3.3). Kenyan participants [n = 84 (34F)] consisted primarily of farmers, market vendors, and gardeners associated with the Development in Gardening (DIG)

501(c)(3) organization. Approximately 80% of these participants reported growing sukuma wiki for personal consumption and nearly all participants consumed sukuma wiki at least one time per week (Fig 3.1). Kenyan participants used a greater range of the 9-pt. hedonic scale to evaluate collard genotypes; ratings were highly variable across participants. New York participants [n = 102 (43F)] were principally associated with Cornell University’s New York State Agricultural Experiment Station, though participants also included organic farmers and community-supported agriculture

(CSA) members. Less than 20% of New York participants reported growing collards for personal consumption or consuming collards at least once per week (Fig 3.1). New

York participants rated nearly all genotypes favorably and generally gave higher liking scores for all genotypes than Kenyan participants.

107

Figure 3.1. Demographic information for farmers, educators, and consumers who participated in consumer acceptance testing of n=6 collard genotypes in Western Kenya (May ‘19) and Upstate New York (Sept. ‘19). Demographics highlight major discrepancies in product familiarity between the two regions.

Mixed model analysis of liking scores from both regions indicated a significant effect of cultivar-by-country (p < 0.001)

(Appendix 3.2). Local cultivar ‘Mfalme’ was more liked in Kenya than in New York. Kenyan consumer participants and farmers

readily identified and preferred their regional hybrid variety (‘Mfalme’) during acceptance testing. Two hybrid breeding lines, both

of which were generated with a broccoli parent (P8CMS), were also highly liked among Kenyan participants (Fig 3.2). The two

most highly liked cultivars in New York were developed using an inbred parent from a cabbage-collard accession (KP1702). ‘Top

Bunch’, the most common hybrid cultivar in U.S. markets, was also highly liked among New York participants; this genotype was

only sampled in New York evaluations.

108

Figure 3.2. Estimated mean liking (9pt hedonic scale) of each collard genotype subject to consumer acceptance testing in New York and Western Kenya (Homa Bay). ‘Top Bunch’ was only sampled in New York evaluations. Bars = SE

Underlying trends in raw product pigmentation translated to distinct preferences between New York and Kenyan consumers during cooked product evaluations (Appendix 3.4). New York consumers liked collards with lighter, yellower colors in their raw form (i.e. hybrids containing ‘KP1702’ in their pedigree) and

Kenyan consumers enjoyed dark green or blue-green types such as cultivars 267

(‘Mfalme’) or 141 (‘P8CMS x KP1’). Two hybrids included in the study, genotype

591 (‘KP1706 x KP15’) and cultivar 356 (‘Beira’), were equally disliked among consumers in Kenya and New York. Genotype 591 (‘KP1706 x KP15’) was developed as an intraspecific cross between borecole (curly) kale and collard inbred lines. During one tasting in Aram, Kenya, a consumer participant noted that cultivar 591 (‘KP1706 x KP15’) “tasted like a foreign vegetable.” The Portuguese tronchuda cultivar 356

(‘Beira’) was also disliked due to its thick midrib and generally poor flavor.

Consumer acceptance evaluations of collard genotypes occurred in non- traditional settings, adding variability and noise to the sensory testing environments.

109

Minimal variance was captured by the random effects of participant and location in

Kenya (Table 3.3), suggesting there was additional variability in the model due to external factors. Despite variability among testing locations, the location effect accounted for less than 5% of the random effect variance in both Kenya and New

York. Random error variance was reduced ~30% in New York evaluations, and participant variance accounted for 26.6% of the total random variance in the New

York model.

Lexicon development and variation by region

Flash Profiling (FP) of six collard genotypes in both raw and cooked preparations were completed in Rodi Kompany, Kenya and Geneva, NY. Panelists in

Kenya included n=12 participants completing their surveys in English (n = 5) or

Dholuo (n = 7); New York panelists (n = 15) completed surveys in English. Ballot development and ranking of both product preparations took approximately four hours with the Kenyan panelists and two hours with the New York panelists. In both regions, panelists generated additional descriptive attributes during the ballot development stage which were not included during the ranking session. Incomplete rankings were common among panelists in both regions, though it disproportionately reduced the number of attributes used in data analysis for the Kenyan profiles. After filtering, New

York panelists developed more than twice the number of descriptive attributes than

Kenyan panelists (Appendices 3.5 and 3.6).

Raw product descriptive attributes were categorized within color, durability, shape, visual, and textural sensory modalities. New York panelists provided between

2-18 attributes, averaging eleven attributes per panelist. Singular attributes (i.e.

110

descriptive attributes generated only once) comprised 70% of the total attribute list (n

= 105) in New York. Raw product attributes were dominated by shape and color descriptors, and many more textural attributes were defined among New York panelists than Kenyan panelists. Descriptive attributes developed by Kenyan panelists were also present in the New York lexicon, especially within the sensory modalities of color (green, light-green) and shape (curly, broad, long/tall). Kenyan panelists generated between 2-13 descriptive attributes, with an average of seven per panelist.

Approximately 49% of attributes (n = 41) were singular among Kenyan panelists.

Market production attributes (easy to care for, harvest for long, high production, etc.) were common attributes supplied by Kenyan panelists; these attributes could be considered hedonic in nature.

Cooked products were described with far fewer descriptive attributes than raw products. Descriptive attributes among cooked products were categorized within flavor, aroma, aftertaste, texture, and visual sensory modalities. New York panelists provided between 3-13 attributes, averaging seven attributes per panelist. Singular attributes comprised 80% of the total attribute list (n = 66) in New York. Kenyan panelists each generated between 1-13 descriptive attributes, averaging five per panelist. The attribute list (n=33) was comprised of 58% singular descriptors. Very few textural attributes were defined by Kenyan panelists. This is in contrast to the

New York panelists that generated approximately 15 texture-related attributes, namely crunchy and chewy. Flavor attributes among the Kenyan panelists were hedonic (bad, nice, tasty) or difficult to interpret (soft, poor) after translation. Sweet flavor was the most commonly reported among Kenyan panelists, but the translation of Dholou’s mit

111

can also imply delicious or favorable taste. Eleven of the thirty-three unique attributes among Kenyan participants were categorized as hedonic terms, including sweet.

Collard product differentiation

Generalized Procrustes Analysis (GPA) was performed with descriptive attribute ranking data to obtain a consensus configuration across FP panelists. Across regions and product preparations, the GPA procedure explained 29-40% of the original variance in the FP datasets (Table 3.3). Principal Components Analysis (PCA) was subsequently performed on the optimal GPA consensus configuration to identify genotype (i.e. object) coordinates and correlated attributes (Fig 3.3). In both product preparations, attributes generated by New York panelists better differentiated among genotypes.

Raw products were reasonably differentiated in both regions (Fig 3.3A and B).

Approximately the same amount of variance among raw products was explained by the first two principal components (PC) after GPA analysis in both Kenya (68%) and

New York (63%). Both regions exhibited overlap in genotypes 982 (‘P8CMS x

KP1702’) and 141 (‘P8CMS x KP1’), which contain a male sterile broccoli parent in their pedigree. This was a common feature in both raw and cooked preparations.

Principal component analysis of raw product in New York was driven by color and texture modalities (PC1) and shape or durability modalities (PC2); products were well separated by underlying product attributes. Clear resolution or trends among raw sensory modalities were not as obvious among Kenyan participants.

112

Figure 3.3. Cultivar coordinates for each Flash Profiling participant configuration (n = 12-15) in Rodi Kopany, Kenya (Homa Bay) and Geneva, NY (New York) (A, C) and New York, USA (B, D) after Generalized Procrustes Analysis (GPA) transformations and principal components analysis (PCA). Cultivar coordinates are summarized with 90% CI ellipses to highlight product separation. Participant- provided sensory attributes correlated (r > 0.6) with each dimension are noted along their respective axis for raw (A, B) and cooked (C, D) product preparations. External preference mapping was performed for cooked product preparation in Homa Bay (C) using PREFMAP in XLSTAT. A contour plot (gray) has been overlaid to illustrate the percentage of consumer clusters exhibiting an above- average preference for samples in a particular region of the sensory map. Dark gray 113 indicates 80-100% of consumer clusters exhibiting above average preference in that region; additional preferences are shades in increments of 20%.

Variance explained by the first two PCs in the cooked product preparation were significantly more variable between regions. In Kenya, approximately 88% percent of the variance among cooked collard genotypes were explained by the first two PCs. In comparison, the first two PCs explained 58% of the variance among collard genotypes evaluated in New York. Genotypes 709 (‘Champion S2 x KP1702’) and 267 (‘Mfalme’) were distinct in cooked preparations in Kenya and New York, respectively. Kenyan color and flavor descriptive attributes were strongly correlated with PC1 and PC2 axes, respectively.

Using data from the consumer acceptance evaluations, three consumer clusters were identified by agglomerative hierarchical clustering (AHC). External preference mapping models used to characterize clusters on the preference map were not significant (data not shown), and all three consumer clusters grouped similarly on the sensory map. A contour plot was overlaid on the PCA sensory map to illustrate the percentage of Kenyan consumer clusters with an above-average liking for samples in particular map regions (Fig 3.3C). The map illustrated a strong consumer preference for genotype 267 (‘Mfalme’) and, to a lesser extent genotypes 591 (‘KP1706 x KP15’) and 141 (‘P8CMS x KP1’). This highly liked region of the map was associated with hedonic terms, such as sweet and tasty. Genotypes 356 (‘Beira’) and 709 (‘Champion

S2 x KP1702’) were least liked among all three consumer clusters in Kenya.

Preference mapping was not performed with cooked attribute data from New York due to insufficient sample sizes for consumer clustering. Estimates of liking from acceptance testing, however, indicated that New York consumers preferred genotypes

982 (‘P8CMS x KP1702’) and 709 (‘Champion S2 x KP1702’). These genotypes were

114

distinguished from the product set along the first PC associated with earthy, dense, and bitter/sharp descriptive attributes.

Discussion

Understanding which quality attributes are of importance to consumers of different cultures and familiarity is admittedly complex, but it is also critical in understanding the changes and opportunities occurring across global food systems. To inform breeding programs and product development, this research sought to explore the existence of region-specific lexicons and preferences through the comparison of consumer sensory studies conducted in two markedly different environments: Western

Kenya and Upstate New York. Studies presented herein provided a preliminary look into the underlying sensory attributes consumers use to describe collard/sukuma wiki products and consumer acceptance of market types. They represent the evaluation of a raw agricultural product with contrasting consumer participants in dramatically different cultural contexts. To respect cultural nuances in perception and retain adequate capacity to discriminate among products (Jamir et al., 2020), descriptive data were not combined or analyzed across regions and acceptance data were interpreted primarily within each region.

To discriminate among products and generate a specific lexicon suitable for future product profiling, we employed rapid Flash Profiling (FP) of raw and cooked collard products with untrained consumer panelists. New York panelists generated more than twice the number of descriptive attributes from Kenyan panelists, despite much greater product familiarity among Kenyan panelists. However,the quality and utility of each region’s lexicon is not measured by the sheer number of attributes used

115

in product discrimination. Attribute lists from New York panelists contained a significant number of singular attributes with embedded sensory complexity (e.g. pliable, ruffled, vegetal, citrus, opalescent, mineral, etc.), but these attributes may not be representative across product samples and they may not be perceptible by all consumers. Evaluation of singular attributes should be explored by a trained descriptive panel and subject to additional validation. Textural attributes, which were common among New York panelists, appeared infrequently in the Kenyan description of collard products. These sensory elements were instead projected onto perception of durability and/or market production characteristics. Kenyan panelists also generated hedonic attributes at a much higher rate than New York panelists, rendering many attributes generated by Kenyan panelists unacceptable for use in future descriptive analysis. Further, these hedonic attributes confounded acceptability and descriptive traits during product separation.

Previous efforts to develop a sensory lexicon for fresh leafy vegetables produced a validated list of 32 flavor-related attributes (Talavera‐Bianchi et al., 2010).

Descriptive analysis in leafy Brassicas identified 21 sensory attributes to be significantly different among diverse leaf types, the majority of which were texture related (Swegarden et al., 2019). While each region produced 30+ attributes for raw and cooked collard/sukuma wiki products, far fewer descriptive attributes may in fact be needed by consumers to differentiate among collard products. Relevant descriptors generated by more than one panelist within each region were common in the sensory modalities of color (green, light-green, dark green, yellow, purple), shape (thick vs. thin midrib, broad vs. tall, curly vs. flat margins), flavor (degree of bitter, sweet, and

116

sour), and texture (tough vs. delicate, chewy, waxiness, and crunchy). Future consumer testing may wish to focus their lexicon to these attributes but should be wary in the deployment of textural attributes in Kenyan markets. These textural attributes may not be well defined among consumers and could present a source of variability in future consumer testing.

New York panelists’ descriptive attributes and rankings provided greater product separation, but all six genotypes evaluated in consumer acceptance testing were well liked. No cultivar was overtly rejected during consumer testing in New

York. Conversely, Kenyan participants effortlessly recognized and favored the local sukuma wiki variety (‘Mfalme’). Inherent variability in familiarity between participants may have influenced the discriminatory power and, therefore, liking scores in each region (Prescott and Bell, 1995). Product familiarity may also be intricately tied to food neophobia or variety seeking behaviors among consumer participants. Inclusion of food neophobia (FNS; Pliner and Hobden, 1992) or variety- seeking (VARSEEK; Van Trijp and Steenkamp, 1992) scales would have significantly accented our ability to determine whether changes in product liking were due to an aversion to new products or association with product novelty. According to

Januszewska et al. (2012) consumers characterized as “variety seeking” tend to provide higher overall liking scores compared to food “neophobic” consumers when sampling traditional food products. Translation and interpretation of these scales in future studies will need to be carefully implemented, but they would help highlight underlying decision rationale among consumer participants.

This preliminary research highlights major limitations to the implementation

117

and interpretation of descriptive profiling and consumer acceptance testing performed in dramatically different cultural contexts. Similar to understanding food aversion characteristics of consumers, this study would have benefitted from directed cultural characterization or value mapping to obtain definitive similarities and differences between Kenyan and New York cultures (Hofstede, 2011; Schwartz, 2006). This characterization would have perhaps aided in the development different ballots that catered to each culture. Further, it would have informed the interpretation of FP panelist and consumer liking data. Such analysis was beyond the scope of this study.

Feasibility assessments were made prior to establishing field trials and conducting consumer studies in each region, and great care was taken to develop materials that could be applied in both cultures. This research provides valuable lessons in how to mitigate constraints during the implementation of consumer studies, manage existing cultural expectations, and address linguistic interpretations in future cross-cultural studies.

The implementation of consumer studies in each region required markedly different approaches. Consumer acceptance testing and FP were performed outside traditional testing environments and in the context of a plant breeding program. Many tests were conducted adjacent to field sites or during field days with agricultural experts to simultaneously obtain feedback on agronomic quality. Obtaining sufficient numbers of participants was difficult at individual locations, particularly in Western

Kenya where populations are significantly less dense, and transportation to-and-from testing sites is much more difficult. Further, Kenyan participants may have been intimidated by the formality of the FP and acceptance evaluation methods. Assurance

118

of participant compensation coupled with an empathetic and detailed introduction to the study helped to ease apprehensions among FP panelists and consumer participants.

Panelists in both regions still struggled to complete rankings for some descriptive attributes. This led to dropped attributes during final data analysis and fewer reported cooked product attributes. As previously mentioned, this disproportionately affected

Kenyan panelists. These observations may be an artifact of either panelist fatigue or reduced variation among products after cooking. Future studies may be advised to evaluate only one preparation method and consider exploring the use of alternative scaling methods, such as a line scale, that may be later reverted to ranking data.

Cultural contexts affect participant response styles, and flexible scaling methods may facilitate easier comparisons of products for consumers during the FP ranking portion.

Assessment of quality in each region would also benefit from longer consumer-product interactions or qualitative product testing (e.g. home-use-tests

(HUTs), moderated focus group, collaborative meal preparation, etc.). At numerous points during the FP session, New York panelists eagerly shared familial collard recipes and personal stories relating to collards. Several commented that they would be interested in preparing particular genotypes using their family recipe(s) or according to their personal preferences. Kenyan participants were more holistically familiar with sukuma wiki, as many participants produced, prepared, and consumed sukuma wiki in their own households. These participants were keen to obtain seed of each evaluated variety and trial them in their own gardens/farms/kitchens. Field trial sites were maintained by local growers for many months after the sensory studies were conducted. Unexpectedly, collaborators at the Advantage Crops Limited seed

119

company received positive feedback from numerous growers and households regarding the quality genotype 709 (‘Champion x KP1702’), a genotype which was not among the top-three most liked genotypes during the acceptance test. Qualitative methods, while logistically challenging across regions, may elucidate nuanced differences among produced across a temporal scale and help to highlight regional preferences for quality traits.

Consumer acceptance evaluations used only one form of product preparation: the quick sauté method common in Western Kenya. As mentioned previously, collards in the Southern U.S. are commonly braised and cooked for long periods of time. A portion of New York panelists expressed alternative expectations for product preparation but were also familiar with the quick sauté method, as it is a common preparation method for other greens (e.g. kale, mustard, etc.). The decision to only use one preparation method was rationalized by the fact that a strong collard culture is not endemic to the northeastern U.S. and direct comparison of FP and consumer acceptance data would only be valid if the same preparation method was used in both regions. Cooking with a braised method would have presented sampling concerns regarding cooking time and temperature, and it would have made it impractical to conduct acceptance testing in-situ. Access to ovens and consistent cooking facilities are limited in Western Kenya. Inclusion of this preparation method, however, may have elucidated different descriptive attributes and resulted in different cultivar preference trends. This research warrants cross-cultural studies performed across geographic sub-cultures in the U.S. (Sobal, 1998). Differences in liking and lexicon may not exist in geographically similar regions or with different preparation methods.

120

Evaluation and comparisons among collard consumers in the northeastern U.S. and southeastern U.S. would explore varying levels of familiarity and expectations but eliminate language interpretation barriers.

Linguistics remains one of the most difficult components of cross-cultural sensory analysis. When translated between languages, descriptive attributes may hold many different meanings (i.e. verbal richness), may not have a direct translation (e.g. umami), and may undergo translation bias depending on the translator and their bilingual capacities (Ares, 2018). English and Swahili are the national languages of

Kenya, but outside of business and politics, local languages dominate daily conversation. Panelists were encouraged to complete FP in whatever language they felt most comfortable; seven of the twelve Kenyan panelists completed their surveys in the local Dholuo language (Luo). Taste descriptors exist in Luo, but they appeared infrequently on Kenyan panelists ballots. The terms sweet and tasty are used often by

Luo speakers to describe acceptance, positive perception, or intensity of a food product’s sensory characteristics. Panelists in both regions were directed to avoid hedonic or preference terms during ballot development, but due to the nature of the study. It was difficult to determine whether the term sweet (Luo: mit) was hedonic in nature or in reference to the basic sweet taste. Similarly, the Luo word both was also translated to English as not tasty. This word may be more appropriately translated to bland (Owuor and Olaimer-Anyara, 2007). These are two examples of how verbal richness and translation bias may have influenced the interpretation of Kenyan panelist ballots. Ultimately, this research would have benefitted from conducting studies in the local language and partnering with experienced food scientists or culinary experts in

121

Western Kenya to improve the interpretation of some descriptive attributes. Defining sensory modalities (e.g. flavor, aroma, texture, etc.) prior to ballot development may have also helped panelists better categorize their attribute list and avoid hedonic terms.

It is important to contextualize these consumer studies within the cultural relevance of collards and sukuma wiki. Collards are a culturally significant crop in the southeastern U.S., and seed savers have cultivated a rich amount of morphological and genetic diversity in the region (Farnham et al., 2008; Pelc et al., 2015). Little work has been performed to evaluate this diversity in the context of consumer preference or market potential. The collard genotypes evaluated herein represent a very small portion of potential collard diversity. Further, the materials were developed in New

York and evaluated primarily by participants with little intricate knowledge of southern collard culture. Preferred genotypes and descriptive attributes collected in the

New York region may not be representative of perceptions in the Southeastern U.S.

Trends in liking of lighter green/yellow types by New York participants may be driven by underlying phytochemical profiles associated with pigment formation, but further acceptance testing with cultivars distinct in their color profiles is warranted. Inclusion of different preparation methods is also necessary to holistically understand regional preferences across regions in the U.S.

Sukuma wiki is of one several horticultural crops transitioning from subsistence enterprises to sustainable cash crops throughout regions of East Africa

(One Acre Fund, 2014; Tschirley et al., 2004). Increasing production, access to quality seed, and sustainable practices is critical to increasing market participation and economic sustainability for these growers. Equally important, however, is recognition

122

that this food crop is a recent introduction to the region and is in its infancy of establishing itself within indigenous folklore (Owuor and Olaimer-Anyara, 2007).

Breeding and development efforts should recognize this as an opportunity for sukuma wiki consumers to define (or redefine) their crop preferences. For example, this research indicated that the introgression of curly, borecole kale types into sukuma wiki backgrounds is ill-advised in Kenyan markets. The development and improvement of cultivars with broccoli backgrounds, however, may be a promising outlet for diversification of the flavor profiles in sukuma wiki.

The introduction of new collard varieties in the U.S. and East Africa could present valuable market opportunities for leafy vegetable producers and help promote a more diversified, nutritious diet. Product expectations for intrinsic and extrinsic quality traits affect consumer acceptance, preparation and end-use, and likelihood to purchase (Barrett et al., 2010). Consumers maintain recognizable expectations for freshness, taste/flavor, nutrition, and convenience, but these expectations can vary dramatically according to region and cultural context. The establishment of descriptive attributes and a vetted lexicon to describe the sensory characteristics of collard/sukuma wiki are the first steps in evaluating and maintaining relevant quality traits. Descriptive product information coupled with acceptance data provides valuable information to plant breeding programs which seek to develop regionally appropriate cultivars.

Conclusion

The research presented herein profiled the sensory characteristics of collards in two production regions: Western Kenya and Upstate New York. Within each region,

123

consumer acceptance testing and Flash Profiling (FP) were performed. Significant differences in liking of collard cultivars were noted across regions and the capacity to provide descriptive product attributes varied immensely between panelists in Kenya and New York. Flash Profiling is a relatively underutilized descriptive, sensory profiling method, and to our knowledge, this research represents the first application of this method with an unprocessed vegetable product. Further, this research contributes to the limited literature exploring rapid profiling methods in cross-cultural settings.

124

References

Ares, G., 2018. Methodological issues in cross-cultural sensory and consumer research. Food Quality and Preference 64, 253–263. https://doi.org/10.1016/j.foodqual.2016.10.007 Ballay, S., Loescher, E., Gazano, G., 2015. 19 - Projective Flash Profile from experts to consumers: a way to reveal fragrance language, in: Delarue, J., Lawlor, J.B., Rogeaux, M. (Eds.), Rapid Sensory Profiling Techniques, Woodhead Publishing Series in Food Science, Technology and Nutrition. Woodhead Publishing, pp. 401–425. https://doi.org/10.1533/9781782422587.3.401 Barrett, D.M., Beaulieu, J.C., Shewfelt, R., 2010. Color, Flavor, Texture, and Nutritional Quality of Fresh-Cut Fruits and Vegetables: Desirable Levels, Instrumental and Sensory Measurement, and the Effects of Processing. Critical Reviews in Food Science and Nutrition 50, 369–389. https://doi.org/10.1080/10408391003626322 Bates, D., Mächler, M., Bolker, B., Walker, S., 2015. Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software 67, 1–48. https://doi.org/10.18637/jss.v067.i01 Bechoff, A., Tomlins, K., Fliedel, G., Lopez-lavalle, L.A.B., Westby, A., Hershey, C., Dufour, D., 2016. Cassava traits and end-user preference: Relating traits to consumer liking, sensory perception, and genetics. Critical Reviews in Food Science and Nutrition 0, 1–21. https://doi.org/10.1080/10408398.2016.1202888 Blancher, G., Chollet, S., Kesteloot, R., Hoang, D.N., Cuvelier, G., Sieffermann, J.- M., 2007. French and Vietnamese: How do they describe texture characteristics of the same food? A case study with jellies. Food Quality and Preference 18, 560–575. https://doi.org/10.1016/j.foodqual.2006.07.006 Bredie, W.L.P., Liu, J., Dehlholm, C., Heymann, H., 2018. Flash Profile Method, in: Descriptive Analysis in Sensory Evaluation. John Wiley & Sons, Ltd, pp. 513– 533. https://doi.org/10.1002/9781118991657.ch14 Brouwer, B.O., Murphy, K.M., Jones, S.S., 2016. Plant breeding for local food systems: A contextual review of end-use selection for small grains and dry beans in Western Washington. Renewable Agriculture and Food Systems 31, 172–184. https://doi.org/10.1017/S1742170515000198 Carlson, D.G., Daxenbichler, M.E., VanEtten, C.H., Kwolek, W.F., Williams, P.H., 1987. Glucosinolates in crucifer vegetables: broccoli, brussels sprouts, cauliflower, collards, kale, mustard greens, and kohlrabi. Cartea, M.E., Velasco, P., 2008. Glucosinolates in Brassica foods: bioavailability in food and significance for human health. Phytochem Rev 7, 213–229. https://doi.org/10.1007/s11101-007-9072-2 Ceccarelli, S., Grando, S., 2007. Decentralized-participatory plant breeding: an example of demand driven research. Euphytica 155, 349–360. https://doi.org/10.1007/s10681-006-9336-8 Commandeur, J.J.F., 1991. Chapter 2: Generalized Procrustes Analysis, in: Matching Configurations. DSWO Press, Leiden, The Netherlands.

125

Dairou, V., Sieffermann, J.-M., 2002. A Comparison of 14 Jams Characterized by Conventional Profile and a Quick Original Method, the Flash Profile. Journal of Food Science 67, 826–834. https://doi.org/10.1111/j.1365- 2621.2002.tb10685.x Davis, E.H., Morgan, J.T., 2015. Collards: A Southern Tradition from Seed to Table. University of Alabama Press. Davis, E.H., Morgan, J.T., 2005. Collards in North Carolina. Southeastern Geographer 45, 67–82. https://doi.org/10.1353/sgo.2005.0004 Dawson, J.C., Healy, G.K., 2018. Flavour Evaluation for Plant Breeders, in: Plant Breeding Reviews. John Wiley & Sons, Ltd, pp. 215–261. https://doi.org/10.1002/9781119414735.ch5 Dehlholm, C., Brockhoff, 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 Profiling and conventional profiling. Food Quality and Preference 26, 267–277. https://doi.org/10.1016/j.foodqual.2012.02.012 Delarue, J., 2015. 6 - Flash Profile, its evolution and uses in sensory and consumer science, in: Delarue, Julien, Lawlor, J.B., Rogeaux, M. (Eds.), Rapid Sensory Profiling Techniques, Woodhead Publishing Series in Food Science, Technology and Nutrition. Woodhead Publishing, pp. 121–151. https://doi.org/10.1533/9781782422587.2.121 Delarue, J., Lawlor, B., Rogeaux, M., 2014. Rapid Sensory Profiling Techniques: Applications in New Product Development and Consumer Research. Elsevier. Dixon, G.R., 2007. Vegetable Brassicas and Related Crucifers. CABI. Farmers Trend, 2016. Complete guide on how to plant the best sukuma wiki (kales) in Kenya. [WWW Document]. Farmers Trend. URL https://farmerstrend.co.ke/complete-guide-plant-best-sukuma-wiki-kales- kenya/ (accessed 11.14.17). Farnham, M.W., 1996. Genetic Variation among and within United States Collard Cultivars and Landraces as Determined by Randomly Amplified Polymorphic DNA Markers. Journal of the American Society for Horticultural Science 121, 374–379. https://doi.org/10.21273/JASHS.121.3.374 Farnham, M.W., Davis, E.H., Morgan, J.T., Smith, J.P., 2008. Neglected landraces of collard (Brassica oleracea L. var. viridis) from the Carolinas (USA). Genet Resour Crop Evol 55, 797–801. https://doi.org/10.1007/s10722-007-9284-8 Gower, J.C., 1975. Generalized procrustes analysis. Psychometrika 40, 33–51. https://doi.org/10.1007/BF02291478 Guerrero, L., Guàrdia, M.D., Xicola, J., Verbeke, W., Vanhonacker, F., Zakowska- Biemans, S., Sajdakowska, M., Sulmont-Rossé, C., Issanchou, S., Contel, M., Scalvedi, M.L., Granli, B.S., Hersleth, M., 2009. Consumer-driven definition of traditional food products and innovation in traditional foods. A qualitative cross-cultural study. Appetite 52, 345–354. https://doi.org/10.1016/j.appet.2008.11.008

126

Hofstede, G., 2011. Dimensionalizing Cultures: The Hofstede Model in Context. Online Readings in Psychology and Culture 2. https://doi.org/10.9707/2307- 0919.1014 Jamir, S.M.R., Stelick, A., Dando, R., 2020. Cross-cultural examination of a product of differing familiarity (Hard Cider) by American and Chinese panelists using rapid profiling techniques. Food Quality and Preference 79, 103783. https://doi.org/10.1016/j.foodqual.2019.103783 Januszewska, R., Viaene, J., 2012. Sensory Evaluation of Traditional Products by Variety-Seekers and Food Neophobics. Journal of Culinary Science & Technology 10, 192–210. https://doi.org/10.1080/15428052.2012.706117 Lawless, H.T., Heymann, H., 2010a. Descriptive Analysis, in: Sensory Evaluation of Food, Food Science Text Series. Springer, New York, NY, pp. 227–257. https://doi.org/10.1007/978-1-4419-6488-5_10 Lawless, H.T., Heymann, H., 2010b. Physiological and Psychological Foundations of Sensory Function, in: Sensory Evaluation of Food, Food Science Text Series. Springer, New York, NY, pp. 19–56. https://doi.org/10.1007/978-1-4419- 6488-5_2 Lenth, R., Singmann, H., Love, J., Buerkner, P., Herve, M., 2020. emmeans: Estimated Marginal Means, aka Least-Squares Means. Maggioni, L., von Bothmer, R., Poulsen, G., Lipman, E., 2018. Domestication, diversity and use of Brassica oleracea L., based on ancient Greek and Latin texts. Genet Resour Crop Evol 65, 137–159. https://doi.org/10.1007/s10722- 017-0516-2 Mitchell, N.D., 1976. The status of Brassica oleracea L. subsp. oleracea (Wild Cabbage) in the British Isles. Watsonia 11, 97–103. Muñoz, A., Civille, G., 1992. The Spectrum Descriptive Analysis Method. Manual on Descriptive Analysis Testing for Sensory Evaluation 22. https://doi.org/10.1520/MNL10524M Mvere, B., van der Werff, M., 2004. Brassica oleracea - leaf cabbage [WWW Document]. Brassica oleracea L. (leaf cabbage). URL https://uses.plantnet- project.org/en/Brassica_oleracea_-_leaf_cabbage_(PROTA) (accessed 3.25.20). One Acre Fund, 2014. Collard Greens Trial – Long Rainy Season, Kenya (2013). One Acre Seed Fund. One Acre Seed Fund. Owuor, B.O., Olaimer-Anyara, E., 2007. The value of leafy vegetables: an exploration of African folklore. African Journal of Food, Agriculture, Nutrition and Development 7, 33–38. Park, H.-J., Ko, J.-M., Jang, S.-H., Hong, J.-H., 2017. Comparison of consumer perception and liking of bulgogi marinade sauces between Korea and Japan using flash profiling. Food Sci Biotechnol 26, 427–434. https://doi.org/10.1007/s10068-017-0058-6 Pelc, S.E., Couillard, D.M., Stansell, Z.J., Farnham, M.W., 2015. Genetic Diversity and Population Structure of Collard Landraces and their Relationship to Other Brassica oleracea Crops. The Plant Genome 8. https://doi.org/10.3835/plantgenome2015.04.0023

127

Pliner, P., Hobden, K., 1992. Development of a scale to measure the trait of food neophobia in humans. Appetite 19, 105–120. https://doi.org/10.1016/0195- 6663(92)90014-W Prescott, J., Bell, G., 1995. Cross-cultural determinants of food acceptability: Recent research on sensory perceptions and preferences. Trends in Food Science & Technology 6, 201–205. https://doi.org/10.1016/S0924-2244(00)89055-X R Core Team, 2018. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. Ramírez‐Rivera, E. de J., Ramón‐Canul, L.G., Díaz‐Rivera, P., Juárez‐Barrientos, J.M., Herman‐Lara, E., Prinyawiwatkul, W., Herrera‐Corredor, J.A., 2017. Sensory profiles of artisan goat cheeses as influenced by the cultural context and the type of panel. International Journal of Food Science & Technology 52, 1789–1800. https://doi.org/10.1111/ijfs.13452 Risso, D.S., Mezzavilla, M., Pagani, L., Robino, A., Morini, G., Tofanelli, S., Carrai, M., Campa, D., Barale, R., Caradonna, F., Gasparini, P., Luiselli, D., Wooding, S., Drayna, D., 2016. Global diversity in the TAS2R38 bitter taste receptor: revisiting a classic evolutionary PROPosal. Scientific Reports 6, 1–9. https://doi.org/10.1038/srep25506 Sauer, J.D., 1993. Historical geography of crop plants: a select roster. CRC Press, Boca Raton, FL. Schwartz, S., 2006. A Theory of Cultural Value Orientations: Explication and Applications. Comparative Sociology 5, 137–182. https://doi.org/10.1163/156913306778667357 Snogerup, S., 1980. The wild forms of the Brassica oleracea group (2n = 18) and their possible relations to the cultivated ones. Brassica crops and wild allies. [I]. 121–132. Sobal, J., 1998. Cultural Comparison Research Designs in Food, Eating, and Nutrition. Food Quality and Preference 9, 385–392. https://doi.org/10.1016/S0950- 3293(98)00029-9 Stansell, Z., Cory, W., Couillard, D., Farnham, M., 2015. Collard landraces are novel sources of glucoraphanin and other aliphatic glucosinolates. Plant Breeding 134, 350–355. https://doi.org/10.1111/pbr.12263 Stone, H., 1992. Quantitative Descriptive Analysis (QDA), in: Hootman, R. (Ed.), Manual on Descriptive Analysis Testing for Sensory Evaluation. ASTM International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken, PA 19428-2959, pp. 15-15–7. https://doi.org/10.1520/MNL10523M Swegarden, H., Stelick, A., Dando, R., Griffiths, P.D., 2019. Bridging Sensory Evaluation and Consumer Research for Strategic Leafy Brassica (Brassica oleracea) Improvement. Journal of Food Science 84, 3746–3762. https://doi.org/10.1111/1750-3841.14831 Talavera‐Bianchi, M., Iv, E.C., Chambers, D.H., 2010. Lexicon to Describe Flavor of Fresh Leafy Vegetables. Journal of Sensory Studies 25, 163–183. https://doi.org/10.1111/j.1745-459X.2009.00249.x Tschirley, D.L., Muendo, K.M., Weber, M.T., 2004. Improving Kenya’s Domestic Horticultural Production and Marketing System: Current Competitiveness,

128

Forces of Change, and Challenges for the Future (Volume II: Horticultural Marketing) (No. 55156), Food Security Collaborative Working Papers, Food Security Collaborative Working Papers. MSU - Dept. Agr. Food Res. Econ. USDA-NASS, 2019. 2017 Census Volume 1, Chapter 2: State Level Data [WWW Document]. Census of Agriculture. URL www.nass.usda.gov/AgCensus (accessed 3.24.20). van Kleef, E., van Trijp, H.C.M., Luning, P., 2006. Internal versus external preference analysis: An exploratory study on end-user evaluation. Food Quality and Preference 17, 387–399. https://doi.org/10.1016/j.foodqual.2005.05.001 Van Trijp, H.C.M., Steenkamp, J.-B.E.M., 1992. Consumers’ variety seeking tendency with respect to foods: Measurement and managerial implications. Eur Rev Agric Econ 19, 181–195. https://doi.org/10.1093/erae/19.2.181 Varela, P., Ares, G., 2012. Sensory profiling, the blurred line between sensory and consumer science. A review of novel methods for product characterization. Food Research International 48, 893–908. https://doi.org/10.1016/j.foodres.2012.06.037 Ward, J.H., 1963. Hierarchical Grouping to Optimize an Objective Function. Journal of the American Statistical Association 58, 236–244. https://doi.org/10.1080/01621459.1963.10500845 Wickham, H., 2016. ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag New York. Wieczorek, M.N., Walczak, M., Skrzypczak-Zielińska, M., Jeleń, H.H., 2018. Bitter taste of Brassica vegetables: The role of genetic factors, receptors, isothiocyanates, glucosinolates, and flavor context. Critical Reviews in Food Science and Nutrition 58, 3130–3140. https://doi.org/10.1080/10408398.2017.1353478

129

Chapter 4

DEVELOPMENT OF A LEAFY BRASSICA DIALLEL TO ASSESS GENETIC

ARCHITECTURE OF CONSUMER LIKING

Abstract

The integration of consumer acceptance data into horticultural crop breeding and parental selection has historically relied on anecdotal or informal methods.

Establishing a heritable nature of consumer liking and understanding traits underlying consumer liking are key to building a leafy Brassica (Brassica oleracea L.) breeding program directed by consumer acceptance and preferences. Eight morphologically distinct B. oleracea inbred lines were used as parents to develop a half diallel mating design to assess the heritability of consumer liking. Leaf images of parents and F1 progeny were subject to an online consumer acceptance survey, whereby n = 564 consumers rated their overall visual liking and familiarity with each genotype. In addition, diallel materials were characterized with genotyping-by-sequencing (GBS) and preliminary morphological and nutritional (glucosinolate and carotenoid) data were collected. Survey results revealed a strong correlation between overall liking and familiarity, and, even after accounting for familiarity in a mixed model analysis,

“familiar” kale and collard market class genotypes were highly liked. We identified a small (h2 < 0.1) but heritable genetic component to consumer liking. Obvious trends in parental selection underscore the success of hybrid progeny; novel leaf types that were well received by consumer participants typically incorporated some “familiar” aspect

130

from kale or collard breeding lines. Sequencing studies indicated that leafy Brassica breeding materials were genetically distinct from currently available commercial cultivars and their robust nutritional profiles exhibited moderate genetic variance due to parental type. This work highlights the importance of parental selection in developing new crop types to suit current consumer preferences and the use of genetic studies to better understand genotypic diversity as it relates to consumer liking. This work demonstrates the capacity of existing breeding materials to introduce novel cultivars with market potential.

Introduction

Rapidly changing food systems and consumer preferences can benefit from inherent diversity in horticultural crop species. Brassica oleracea L. is characterized by a rich diversity of plant morphologies. Domestication of leafy, cliff-dwelling progenitors in the Mediterranean region, followed by generations of targeted selection for floral characteristics (broccoli and cauliflower), heading/bud sprout traits (cabbage and Brussels sprouts), and leaf shapes (kale and collard), has resulted in diverse cultivated forms of modern varieties. These morphotypes are consumed around the globe and have cultural relevance on every arable continent.

Western markets have seen a marked rise in popularity and production of new leafy Brassica types, primarily in kale (B. oleracea var. acephala L.) and collard (B. oleracea var. viridis) market classes. Though functionally biennial crops requiring a cold period of vernalization for flowering, singly borne leaves of these varieties are harvested in the first year of annual production for human consumption. Data from the

2017 USDA Census of Agriculture estimates approximately 11,318 acres of collard

131

and 15,325 acres of kale were harvested for U.S. fresh and processed markets (USDA-

NASS, 2019). The majority of acreage in 2017 was intended for fresh produce markets

(baby and mature leaves) and minimally processed products, including powdered smoothie mixes and bagged salads. Markets, sensory characteristics, and end-use products of these two leafy greens are well established and distinct in the U.S.

(Swegarden et al., 2019), but little information exists to inform what potential new quality traits or combination of quality traits that currently exist within B. oleracea biodiversity of breeding pipelines may resonate with consumers.

Market separation and utility of kale and collards are primarily predicated on their characteristic leaf morphologies. The “iconic” curly (borecole) kale retains a strong identity among health-conscious consumers and is touted for its durability, long shelf-life, and capacity to withstand heavy cooking. Though there is market recognition of other kale morphotypes such as Tuscan/Lacinato kale (var. palmifolia) and Russian types (B. napus), mature curly kale types are the most widely grown and consumed. Flat, broad collard leaves with smooth margins are more commonly associated with cuisines from the southern U.S. such as braised collard greens.

Collards have significant cultural relevance among gardeners and seed savers in the

Southeastern U.S. and could be considered a traditional food product in the region

(Davis and Morgan, 2015). Private sector breeding and development of plant materials for these two markets adheres to visual expectations for each morphotype and has traditionally focused on breeding for horticultural traits, such as higher yields and disease resistance. Given the inherent intraspecific diversity of B. oleracea, the potential to generate and introduce new morphological and color diversity to leafy

132

greens markets is substantial.

Intrinsic (i.e. flavor, texture, appearance, etc.) and extrinsic (i.e. label, name, packaging, etc.) product cues shape consumer acceptance and purchasing decisions

(Jongen, 2000; Steenkamp, 1990). Visual product cues encompass appearance-related traits such as color, shape, surface texture, and optical properties (Barrett et al., 2010;

Hutchings, 1999). Appearance can provide the first indication of a product’s intrinsic quality and, in some cases, influence the perception of qualities such as flavor and texture (Wadhera and Capaldi-Phillips, 2014). The importance of visual cues is well documented in food products and vegetables across the supply chain, and it is particularly relevant when discussing fresh market produce. Search attributes of minimally processed vegetables, including appearance and packaging, are important during the purchasing stage (Ragaert et al., 2004). Fresh market vegetables comprise

78% of the U.S. sales of vegetables, and the majority of vegetables are still consumed as loose, whole vegetables (Roberts, 2019). Identification of desirable consumer traits based on appearance and visual sensory testing of leafy Brassicas may provide direction and inspiration to develop new market classes that could diversify consumer options. In turn, these may provide value-added products for supply chain stakeholders and promote regional production.

Leaf phenotype diversity in B. oleracea includes variation in leaf shape, texture, and phytochemical profiles imparting organoleptic and nutritional characteristics. Modification in these traits has biological significance; leaves are the photosynthetic organs of a plant and play an important role in plant growth and development (Tsukaya, 2006). Leaf morphogenesis and biosynthetic expression of

133

phytochemicals are subject to hormonal fluctuations, environmental conditions, and complex genetic regulation. Leaf shape, for example, is generally understood to be multigenic and heritable, but it is also modified by hormonal responses to precipitation, salinity, and light availability (Bar and Ori, 2014; Chitwood and Sinha,

2016). Numerous quantitative trait loci (QTL) have been detected for laminar, petiole, and node-formation traits in B. oleracea (Lan and Paterson, 2001; Sebastian et al.,

2002; Stansell et al., 2019). Modification of leaf characteristics, including shape/size, epicuticular wax, and trichomes, may alter insect predation and disease progression dynamics in leafy Brassica crops (Eigenbrode et al., 1991; Vaughn and Hoy, 1993).

They may also influence yield, shelf-life, and consumer acceptance.

The biosynthesis of color-imparting compounds in leaves, such as carotenoids

(orange/yellow), anthocyanins (red/purple), and chlorophylls (green), have been well described (Allan and Espley, 2018; Sun and Li, 2020; Tanaka et al., 2008). These secondary metabolites play key roles in leaf photosynthesis and photoprotection but are highly influenced by the environment and plant age (Chalker‐Scott, 1999; Esteban et al., 2015; Hatier and Gould, 2009; Socquet-Juglard et al., 2016). They are also well- known for their antioxidant capacity and relevance to human health and nutrition

(Khoo et al., 2017; Krinsky and Johnson, 2005). Phytochemicals can also impact the visual phenotype of a plant and shape the perception/acceptance of products.

Diversity for leaf shape and color in materials from the Cornell AgriTech leafy

Brassica breeding program could feed directly into current consumer trends (e.g. fresh, colorful, nutritional, seasonal produce) and novel applications in the vegetable sector (Roberts, 2019). The large number of breeding lines available for further

134

development and established kale and collard markets makes it difficult to assess which lines have potential in the marketplace. Evaluating market potential is further complicated by the variation in reproductive life cycles, plant habits, and mating systems in Brassica vegetables. Intraspecific hybridization within B. oleracea can be performed with relative ease, but the development and testing of a biennial variety can take anywhere between seven and ten years (P.D. Griffiths, personal communication).

Understanding the potential success of plant products, particularly those without prior market exposure, is invaluable to plant breeders working with new crop types with a long product development timeframe. This can allow a focus on valuable end-market traits early in the breeding process.

Consumer research tools, including those used in marketing, research, sales, and sensory science are recognized for their capacity to obtain consumer insights and develop strategic plans to bring products to marketplace (Lopetcharat et al., 2012).

Consumer surveys are product testing tools that assess how consumers interact with and perceive products. When coupled with quantitative affective testing, surveys allow for direct product comparisons and understanding of consumer acceptance. Affective testing seeks to measure the magnitude with which consumers like or dislike a product using a scaled method, commonly a 9-point hedonic scale or a labeled magnitude scale

(Lawless and Heymann, 2010). The degree of acceptance inferred from these scaling techniques represents a highly complex measure of human perception. Consumer liking is fluid and can shift in accordance with changing demographics, socio-political movements, food and marketing trends, and product context (Kim et al., 2015).

Consumer acceptance is also driven by prior product experience and exposure, or

135

“familiarity.” Product familiarity biases liking and acceptance upward (Rindfleisch and Inman, 1998) and repeated exposure to vegetable products have been demonstrated to increase overall liking (Bingham et al., 2005; Wardle et al., 2003) .

Familiarity and product usage may also lead to greater discriminatory power among products (Prescott and Bell, 1995). As a quantitative, highly subjective trait, it is of interest to plant breeders to evaluate the utility of consumer liking in making breeding selections and decisions.

The use of mating or cross-classified designs is well established within both horticultural and agronomic crop breeding programs. These designs are used to understand the inheritance of quantitative traits and determine the combining ability

(i.e. potential) of parental types in a population (Bernardo, 2010). They are also valuable tools to estimate genetic variance of a phenotype, determine whether parental selection plays a role in the expression of a phenotype, and enable the identification of parents that provide promising genetic complementation for traits of interest.

The diallel experiment is one such mating design, whereby a set of inbred parental genotypes (n) are crossed in all possible cross combinations to produce a set of F1 hybrid progeny [n*(n-1)] (Gardner and Eberhart, 1966; Hayman, 1954; Sprague and Tatum, 1942). Parental types and their progeny are subsequently phenotyped for traits of interest and subject to statistical analysis to estimate variance in the population and optimal parental combinations. Diallel experiments are commonly conducted to evaluate genetic variance of agronomic traits and have recently been executed in vegetable crops such as eggplant (Kaushik et al., 2018) and carrots

(Turner et al., 2018). Within Brassica oleracea, diallel designs have been developed to

136

understand inheritance of the self-incompatibility allele in cabbage (Hodgkin, 1978), numerous diseases resistance traits (Chiang and Crête, 1976; Grandclement et al.,

1996), and leaf/stem traits in an intermorphotypic diallel (Johnston, 1968).

Statistical analysis of diallel designs can take numerous forms depending on whether reciprocal and/or parental inbreds are included in the analysis and whether parental lines were selected from fixed or random populations (Christie and Shattuck,

1992; Griffing, 1956). Estimates of variance from a diallel design are used to calculate the general combining ability (GCA) of parental genotypes and the specific combining ability (SCA) of individual hybrid combinations (Sprague and Tatum, 1942). Parental

GCA can be understood as a genotype’s performance across all hybrid combinations and is largely the result of additive genetic effects. In contrast, the SCA compares the performance of specific hybrid combinations to the overall mean of all hybrid combinations and estimates the genetic effect of dominance in these combinations.

These two combining ability estimates inform the selection of parental types and breeding of hybrid populations.

To the best of our knowledge, a mating design phenotyped for consumer acceptance has not yet been reported. The necessarily large sample size, potential for consumer fatigue, or missing/incomplete data may have previously discouraged this usage. The current capacity of multivariate mixed-modelling software, however, overcomes issues of missing data and complex experimental designs (Möhring et al.,

2011). Recent advances also allow for the inclusion of variance-covariance structure

(e.g. genetic relationships among parents) in the random effects used to estimate variance (Covarrubias-Pazaran, 2016). Therefore, modern diallel designs allow the

137

integration of large-scale consumer data and next generation genotyping to estimate the genetic component of consumer liking.

This research sought to answer two key questions. First, how relevant is the intermorphotypic diversity in B. oleracea to the development of new genotypes and how do we adequately survey the potential of this diversity? Second, is the phenotype of consumer liking useful to the leafy Brassica breeding program and can it be used guide breeding decisions? To address these questions, this study developed a half- diallel mating design with eight inbred B. oleracea parental lines capturing the current market class diversity of leafy Brassica vegetables in the Cornell AgriTech vegetable breeding program. These diallel materials, in addition to several commercially available B. oleracea cultivars and wild relatives, were subject to genotyping-by- sequencing (GBS) to characterize diversity in the breeding program and assess population structure within the mating design. Diallel materials were additionally screened for leaf morphological characteristics and nutritional (carotenoid and glucosinolate) content. Finally, parental lines and F1 hybrid progeny were imaged and subject to an online consumer survey whereby consumer acceptance and the relative importance of product familiarity was assessed. It was our expectation that established market classes, namely curly kale and collard parents, exhibit high general combining ability and a positive correlation with familiarity. This research represents an accent to previous work focused on the intrinsic sensory qualities (flavor, texture, aroma) in leafy Brassicas (Swegarden et al., 2019), and products developed as part of this research may directly shape new market types. This work represents a novel methodology connecting consumer evaluation with traditional and next-generation

138

breeding strategies that will inform the development of leafy Brassica genotypes.

Materials and Methods

Plant materials and field phenotyping

Eight morphologically distinct inbred (self-pollinated to the S4 generation or beyond) parental lines were selected from Cornell AgriTech’s vegetable breeding program (Geneva, NY) to represent the current and potential market class diversity of leafy Brassicas (Table 4.1). Diallel F1 hybrids were developed during the winter/spring of 2018 in Geneva, NY from field-grown parental lines. Field materials were vernalized for approximately 70 days in a cool nursery cellar (8˚C) prior to their placement in the Geneva Greenhouse complex (42°52’31.7”N, 77°00’28.0”W) for crossing and seed production in the spring of 2018. Eight parental lines were crossed in all pairwise cross combinations, resulting in 56 F1 hybrid progeny. All plant materials (F1 hybrid progeny and parental inbred lines) were evaluated for sufficient vigor in the greenhouse prior to establishment of a single field replicate in the summer of 2018 to verify uniformity and absence of visual maternal effects among progeny.

Untreated seed of each diallel genotype were planted in 72-cell styrofoam trays with Cornell Mix potting media (Boodley and Sheldrake, 1972). Transplant materials were grown at the Guterman Bioclimatic Laboratory Complex (42°26’55.8”N,

76°27’39.9”W). In June 2018, four-week old seedlings were transplanted into raised, black plastic beds spaced at 50cm with 2m row centers with drip irrigation.

Approximately nine weeks after transplanting, the third and fourth most fully expanded leaves were harvested from each diallel line, bunched, and photographed for further documentation and study. Leaves were free of defects, including herbivory

139

Table 4.1. Inbred parental genotypes (Brassica oleracea) used in the development of a diallel mating design in early 2018. Each genotype was selfed to at least the S4 generation and evaluated for uniformity/stability. Inbreeding coefficients of each parent were calculated from SNP marker data of three biological replicates. Female Inbred Pedigree Inbreeding Leaf Image Designator Descriptor Information Coefficient

Yellow-leaf collard (var. viridis). Yellow 1 Inbred line KP1702-A derived from F = 0.56 Collard cabbage-collard line GK32558

Tuscan/Lacinato kale (var. Tuscan palmifolia). Inbred line KP1703 2 F = 0.62 Kale derived from Bavicchi’s OP cultivar ‘Cavolo di Nero’

Red borecole (curly) kale (var. Curly Red acephala). Inbred line KP1704 3 F = 0.53 Kale derived from OP kale cultivar ‘Scarlet’

Green borecole (curly) kale (var. Curly acephala). Dwarf inbred line 4 Green F = 0.54 KP1706 derived from cross between Kale ‘Redbor’ x ‘Ripbor’ kale cultivars

Jagged leaf annual broccoli (var. italica). Inbred line KP1707 Jagged developed from unknown broccoli 5 F = 0.69 Broccoli outcross between ‘P1’ broccoli breeding line and HRI kale accession Stabilized deep purple kale leaf Deep (var. acephala). Inbred line KP1713 6 Purple (17HP41) derived from multiple F = 0.54 Kale crosses among pigmented kales, cabbages, and Brussels sprouts Elongated/pointy light green collard Elongated (var. viridis). Inbred line KP1715 7 F = 0.55 Collard (17G32769-S4-P2) derived from breeding line GK32769

Georgia Southern type collard (var. Georgia viridis). Inbred line 16KP15 derived 8 F = 0.53 Collard from collard breeding line GK32762

140

damage and disease. Three leaves of each genotype were photographed with a

Canon DSLR T5i camera (f/5.6, ISO 200, EXP 1/10). Leaves were photographed under two daylight (5500 K) lights placed 0.75m above leaf samples at a 45° angle.

The background was removed from each image and the percentage of each image occupied by the leaf sample was calculated. This was termed the “visual volume” of the image and incorporated into downstream analysis of morphological traits

(Appendix 4.1). Predominant colors (HEX) from each leaf image were summarized using k-means clustering (K = 5 clusters) through the Image Color Summarizer

(v0.76) developed by Krzywinski (2017).

Full diallel materials were reduced to a half-diallel design due to field and resource constraints in the 2019 field season. Parental lines (n = 8) and 28 F1 hybrid progeny representing each parental combination were field-tested in a randomized complete block design (RCBD) at the Homer C. Thompson Vegetable Research Farm

(42°31'21.4"N, 76°19'44.2"W). Two replicated plantings were transplanted in July and

August 2019 for evaluation in September and October 2019. Transplant materials were again grown at the Guterman Bioclimatic Laboratory Complex (42°26’55.8”N,

76°27’39.9”W) and four-week old seedlings (12 per plot) were transplanted into double rows (0.45m within and between plants) on a raised black plastic bed outfitted with drip irrigation.

Leaf morphology and nutritional quality evaluations (Appendix 4.1) were performed approximately nine weeks after planting. Morphological traits were evaluated in both environments on either the plot-level [% dry matter, lamina curl (1-

5), and lamina blister (1-5)] or subsampled using the third most fully expanded leaves

141

from three plants per plot [lamina length (cm), lamina width (cm), petiole length (cm), and petiole width (mm)]. Additional subsampling was performed only during the

September 2019 environment to collect sufficient tissue for nutritional characterization of leaf carotenoid and glucosinolate content. Fresh leaf tissue from each plot was collected, weighed, frozen at -80°C, and shipped on dry ice to plant tissue analysis laboratory of Dr. Carl Sams (University of Tennessee, Knoxville, TN) for nutritional analysis. Protocols followed established extraction and high-performance liquid chromatography (HPLC) techniques for eight carotenoid and nine glucosinolate compounds (Kopsell et al., 2004; Sams et al., 2011). Summary variables of Total

Carotenoids, Total Cholorophylls, and Total Glucosinolates were calculated from raw data, and final weights were presented in mg/g of fresh tissue.

Genotyping-by-Sequencing (GBS)

A diverse set of 164 Brassica oleracea cultivars, including diallel plant materials and commercial cultivars, and five related wild species were genotyped using genotyping-by-sequencing (GBS) through the University of Wisconsin-Madison

Biotechnology Center (Madison, WI). Lyophilized tissue samples of three-week old, greenhouse-grown seedlings were shipped to UW-Madison’s Biotechnology Center for DNA extraction and sequencing in December 2018. Isolated DNA was digested with ApeKI restriction enzyme to construct GBS libraries (Elshire et al., 2011).

Libraries were sequenced using 100bp reads in a single lane of an Illumina HiSeq2500

(Illumina, San Diego, CA).

Reads were aligned to the Belser et al. (2018) Brassica oleracea reference genome. Single nucleotide polymorphisms (SNPs) were called using the TASSEL

142

v5.0 GBS Discovery Pipeline (Glaubitz et al., 2014). In total, 238,616 unfiltered SNPs were stored in Variant Calling Format (VCF) files and later filtered using VCFtools

(Danecek et al., 2011); SNPs with less than 4x read depth, over 75% missing data, minorallel frequencies less than 0.05, insertions, deletions, or multi-allelic SNPs were removed. The SNP density, inbreeding coefficients (F), individual depth, and missingness were calculated using VCFtools (Danecek et al., 2011).

Downstream data analysis was performed with a subset of 129 genotypes relevant the study (Appendix 4.2). Markers were converted to PLINK format (Purcell et al., 2007) and a principal component analysis (PCA) was performed using SNPRelate (Zheng et al., 2012) in R software (v3.5.2) (R Core Team, 2018) . Linkage disequilibrium (LD) decay analysis was performed with PopLDdecay (Zhang et al., 2019) with maximum distance between two SNP markers set to 500kb (Fig. 4.1).

To validate parent-hybrid relationships, structure analysis of half-diallel materials was performed using fastStructure (Raj et al., 2014). Exploratory analysis of k-values approximately equal to the number of parental inbred (n = 8) was performed; convergence occurred at K=7 populations. To construct additive and dominance relationship matrices, imputation was performed on half-diallel (n = 36) materials using Beagle 4.1 software with default parameters (Browning and Browning, 2016).

Realized relationship matrices were calculated using functions in the rrBLUP package

(Endelman, 2011).

Online consumer acceptance survey

Subscribers to the Cornell Sensory Evaluation Center’s listserv were solicited for participation in an online acceptance survey of diallel plant materials in March of

143

2019. Candidate participants were required to be 18 years of age or older and incentivized with a prize drawing. All procedures involving human subjects conducted by the Cornell Sensory Center have been deemed exempt from oversight by Cornell’s

Human Research Protection Program (HRPP) per Protocol ID# 1510005908. All participants were provided a written consent form prior to participation in the online survey. Consent forms, product codes, and coded data are available upon request.

Photographs of the half-diallel materials collected in 2018 were imported into a Qualtrics (Qualtrics, Provo, UT) survey. The survey was designed to adhere to a balanced incomplete block design (BIBD), though concessions were made to the design given the unknown number of participants that might partake in the online survey. The survey consisted of 35 treatments (t); diallel materials used in this survey omitted one F1 progeny which did not produce seed in 2018. Each participant monadically viewed a randomized subset of seven treatments (k). With each image, participants were asked to a) rate their overall liking of the leaf image using a 9pt hedonic scale (1 = dislike extremely; 9 = like extremely) and b) rate their familiarity with this leaf image using a 5pt scale (1 = very unfamiliar; 5 = very familiar).

In total, 591 individuals participated in the online survey; after filtering for missing data or incomplete surveys, 564 participants (b) were used for downstream analysis. Each genotype was rated by participants between 106-117 times (r).

Additional demographics, including consumption patterns, age, gender, and ethnicity were collected from each participant at the conclusion of the survey (Appendix 4.3).

The survey took most participants between 5-7 minutes to complete.

144

Figure 4.1. Genomic descriptions of Brassica oleracea cultivars, wild types, and diallel breeding materials. A and B) Principal component analysis (PCA) of the SNP marker matrix (32,366 SNPs) were developed from all sequenced materials (A) and full diallel mating design (B). B) Parental inbred (colored) are illustrated in triplicate alongside reciprocal hybrids (gray). (C) Linkage disequilibrium (LD) decay plot across the genome. Marker correlations were placed in bins of 1000bp for enhanced resolution. D) Inbreeding coefficients of all sequenced materials, as calculated using the observed number of heterozygous markers in each genotype. 145

Statistical procedures and mixed model analysis

All statistical analyses were performed using R-software (R Core Team, 2018), and all graphics were developed using the ggplot2 package (Wickham, 2016). A

Spearman Rank correlation test was performed to estimate the relationship between mean liking and familiarity. Mean phenotype data [morphological traits (Appendix

4.4), glucosinolate content (Appendix 4.5), and carotenoid content (Appendix 4.6)] were subject to PCA using the ‘stats::prcomp’ function and visualized with the

‘ggfortify::autoplot’ function (Horikoshi et al., 2016).

Phenotypes of the half-diallel parents and hybrid progeny were fit using a linear mixed effects model and variances were estimated with restricted maximum likelihood (REML). Mixed model analysis of product familiarity and consumer demographic data were performed using lme4 (Bates et al., 2015) with the following model structure:

푦 ~ X1훽퐺푒푛표푡푦푝푒 + X2훽퐶표푛푠푢푚푝푡𝑖표푛푅푎푡푒 + X3훽퐴𝑔푒 + (Eq. 4.1)

X4훽퐺푒푛푑푒푟 + X5훽퐸푡ℎ푛𝑖푐𝑖푡푦 + X6훽퐹푎푣푇푦푝푒 + 푍1휇푝푎푟푡𝑖푐𝑖푝푎푛푡 + 휀 where y was the known vector of familiarity observations, 훽 werethe fixed effects vectors for genotype, consumption rate, age, gender, ethnicity, and favorite type of leafy Brassica. The random effect of participant is represented by , and 휀 is the vector of random errors. Incidence matrices for fixed and random effects are described by X and Z, respectively.

Additional models were derived from Möhring et al. (2011), whereby parents and F1 progeny were evaluated in a random effects model and reciprocals were excluded. Mixed models were executed and evaluated using the sommer package with

146

the default algorithm (Covarrubias-Pazaran, 2016). For the case of consumer liking phenotype, the mixed model was written as follows:

푦 ~ X훽퐹푎푚𝑖푙𝑖푎푟𝑖푡푦 + 푍1휇퐺퐶퐴 + 푍2휇푆퐶퐴 + 푍3휇푝푎푟푡𝑖푐𝑖푝푎푛푡 + 휀 (Eq. 4.2) where y was the known vector of liking observations, 훽 was the fixed effects vector for product familiarity,  was the vectors of random effects for GCA, SCA, and participant, and 휀 was the vector of random errors. Incidence matrices for fixed and random effects are described by X and Z, respectively. Models were deployed without covariance structures (data not shown) and with additive and dominance covariance structures were calculated from SNP marker information. The ‘overlay’ function allowed for specification of male-female parental lines as a single random effect. The

‘Gu’ argument within the ‘vs’ function was used to define realized additive and dominance covariance structures among parental and F1 hybrid progeny, respectively.

2 Random effects for GCA were assumed to follow ~ N(0, Kaσ퐺퐶퐴), where Ka = additive relationship matrix, and random effects for SCA were assumed to follow ~ N(0,

2 Kdσ푆퐶퐴), where Kd = dominance relationship matrix (Table 4.2).

Table 4.2. Random effect variance components from participant liking models. Models were performed with (Liking ~ Familiarity) and without (Liking ~ 1) the fixed effect of familiarity. Random effect variances due to general combining ability 2 2 2 2 (𝜎GCA ), specific combining ability (𝜎SCA ), participant (𝜎p ), and random error (𝜎 ) 2 are reported. Percentage of total random variance (% 𝜎total ) provides an indication of the effect’s contribution to the overall random effect variance. Liking ~ 1 Liking ~ Familiarity

σ2 se %𝜎 2 σ2 se %𝜎 2 total total 2 𝜎GCA 0.038 0.023 1.2% 0.013 0.009 0.6% 2 𝜎SCA 0.067 0.025 2.2% 0.027 0.013 1.1% 2 𝜎p 1.173 0.085 37.9% 0.877 0.065 36.6% 2 𝜎 1.814 0.044 58.7% 1.478 0.036 61.7%

147

Finally, Eq. 4.2 was also run as a completely random effects model without the fixed effect of familiarity. Best linear unbiased predictors (BLUPs) for each genotype

(Table 4.3) were calculated such that the estimated breeding value equaled the sum of the random effects of both parents (휇퐺퐶퐴(푀) + 휇퐺퐶퐴(퐹)) (Appendix 4.7) and the random effect of their specific combination 휇푆퐶퐴 (Appendix 4.8):

퐵퐿푈푃 ~ 휇퐺퐶퐴(푀) + 휇퐺퐶퐴(퐹) + 휇푆퐶퐴 (Eq. 4.3)

Completely random genotypic models were used to obtain random variance components of carotenoid (Appendix 4.9), and glucosinolate (Appendix 4.10) phenotypes. Equation 4.1 was modified such that the fixed effect of familiarity

훽퐹푎푚𝑖푙𝑖푎푟𝑖푡푦 and random effect of participant (휇푝푎푟푡𝑖푐𝑖푝푎푛푡 ) were replaced with the random effect of replicate (휇푏푙표푐푘). Square root transformations were imposed on heteroscedastic carotenoid and glucosinolate data prior to model deployment.

Random effect variance estimates from the mixed models were used to calculate broad-sense heritability (H2) (Eq. 4.4) and narrow-sense heritability (h2) (Eq.

4.5) of consumer liking on an experimental (nratings = 113.2) and individual participant

(nratings = 1) basis according to the Fehr (1987):

ퟒ 2 ퟒ 2 ( 휎푔푐푎 + 휎푠푐푎) ퟐ (ퟏ+푭) (ퟏ+푭)ퟐ 푯 = (Eq. 4.4) ퟒ ퟒ 휎2 [ 휎2 + 휎2 + 푒⁄ ] (ퟏ+푭) 푔푐푎 (ퟏ+푭)ퟐ 푠푐푎 풏풓풂풕풊풏품풔

ퟒ 2 휎푔푐푎 풉ퟐ = (ퟏ+푭) (Eq. 4.5) ퟒ ퟒ 휎2 [ 휎2 + 휎2 + 푒⁄ ] (ퟏ+푭) 푔푐푎 (ퟏ+푭)ퟐ 푠푐푎 풏풓풂풕풊풏품풔

2 2 where 𝜎𝑔푐푎 is the random effect variance associated with parental lines, 𝜎푠푐푎 is the variance associated with the specific male x female interaction (i.e. hybrid progeny),

2 and 𝜎푒 is the variance among samples ratings. Variance components for GCA and

148

Table 4.3. Mean participant liking (1-9) and familiarity (1-5) of each hybrid genotype and parental inbred in the half-diallel design. Best unbiased linear predictors (BLUPs) for participant liking were calculated from liking models performed with (Liking ~ Familiarity) and without (Liking ~ 1) the fixed effect of familiarity. Mean Liking Mean Familiarity MIXED MODELS

(1-9) (1-5) Liking ~ 1 Liking ~ Familiarity Genotyp Familiarit Genotyp Liking Genotype BLUP Genotype BLUP e y e 77 7.08 44 4.38 77 0.51 11 0.39

38 7.07 42 4.32 38 0.45 13 0.34

87 6.98 47 4.19 11 0.42 88 0.29

11 6.96 41 4.19 87 0.39 77 0.29

47 6.93 87 4.07 84 0.37 38 0.27

84 6.92 38 4.06 88 0.35 81 0.22

13 6.84 77 4.05 13 0.35 73 0.22

42 6.84 84 4.03 73 0.32 87 0.21

73 6.82 54 3.98 41 0.32 17 0.21

88 6.81 33 3.94 47 0.29 16 0.19

41 6.80 72 3.91 44 0.27 84 0.17

44 6.72 53 3.89 42 0.25 86 0.17

81 6.70 34 3.89 81 0.22 41 0.09

76 6.68 73 3.89 17 0.16 66 0.08

46 6.67 81 3.88 46 0.07 76 0.05

23 6.63 82 3.88 54 0.02 26 0.03

12 6.55 76 3.85 76 0.01 46 0.02

53 6.51 46 3.83 23 0.00 12 0.01

85 6.51 11 3.82 53 0.00 23 0.00

26 6.50 88 3.78 82 -0.01 47 -0.01

82 6.48 13 3.70 33 -0.01 85 -0.02

72 6.44 23 3.68 86 -0.04 51 -0.04

33 6.42 56 3.63 16 -0.05 82 -0.07

17 6.40 12 3.62 72 -0.07 33 -0.08

54 6.39 25 3.59 12 -0.09 42 -0.10

86 6.35 85 3.55 85 -0.10 53 -0.12

16 6.27 17 3.54 34 -0.15 72 -0.14

51 6.25 57 3.42 26 -0.18 44 -0.15

34 6.21 26 3.41 51 -0.24 54 -0.19

66 6.12 51 3.40 66 -0.41 57 -0.24

57 6.01 16 3.38 57 -0.45 34 -0.24

25 5.97 86 3.36 56 -0.49 56 -0.34

56 5.96 55 3.35 25 -0.62 55 -0.42

55 5.92 66 2.97 55 -0.71 25 -0.47

22 5.25 22 2.84 22 -1.08 22 -0.53

Range 1.83 Range 1.54 Range 1.59 Range 0.92

SCA were adjusted according to Bernardo (2010) to estimate additive and dominance variance. The inbreeding coefficient (Favg = 0.568) among parental types was calculated from SNP marker data.

149

Results

Genetic characterization of diallel materials

Aligned reads produced a total of 34,983 filtered SNP markers for downstream analysis of commercial hybrids (n = 20), commercial open-pollinated cultivars (n =

25), inbred diallel parents (n = 8, in triplicate), diallel hybrid lines (n = 56) materials, and wild species (n = 5) (Appendix 4.2). Markers developed were well distributed across each chromosome (Appendix 4.11) with an average individual depth of 30.4x and approximately 7.5% frequency of missingness per individual.

Principal component (PC) analysis identified distinct, centralized clustering of wild types and separated inbred diallel parents along each axis (Fig. 4.1A). The first five PCs captured 35.1% of the variance among genotypes. Approximately 17.37% of genotype variance among the commercial, breeding, and wild genotypes was captured by the first two PCs. Diallel inbreds and hybrids (Fig. 4.1A; purple and gray) exhibited greater distribution along both PC1 and PC2 in comparison to commercial hybrids, open-pollinated cultivars, and wild types. The jagged leaf broccoli and

Tuscan inbred parental lines were individually resolved along PC3.

An additional PC analysis was performed with only diallel inbred parental lines and hybrid progeny (Fig. 4.1B). The first two PCs explained 25.18% of the variation among these breeding materials. Parental inbred replicates (Fig 4.1B; color points) were similarly clustered, confirming their relatedness and stability, and reciprocal hybrid progeny were distributed evenly between both inbred parents, confirming hybrid progeny.

Genome-wide LD decay was evaluated independently for diallel materials and

150

all genotyped cultivars (n = 129). On average, LD in both groups decayed rapidly within the first 25 kb and reached r2 ~ 0.1 by 100kb. As expected, decay among diallel materials (reciprocal hybrids and inbred parents) was slower than the full dataset; LD among diallel materials plateaued more rapidly (r2 ~ 0.1) compared to the full dataset

(r2 ~ 0.06). Recombination among hybrid progeny was likely constrained by the number of parents in the diallel mating design. This rapid decay is characteristic across out-crossing species, including B. oleracea (Cheng et al., 2016; Pelc et al.,

2015; Stansell et al., 2018).

Inbreeding coefficients among commercial hybrids (F range: -0.08 and +0.71) and open-pollinated cultivars (F range: -0.53 and +0.69) varied considerably (Fig

4.1D). In both commercial breeding classes, broccoli and cauliflower lines were among the most inbred; open-pollinated collards from Kenya demonstrated the highest heterozygosity (low inbreeding coefficient) among all genotypes and breeding classes.

As expected, diallel hybrid progeny had a significantly lower inbreeding coefficient (F

= -0.144) than the inbred parental types (F = 0.568).

FastStructure analysis and K-means exploration with parental inbreds and hybrid progeny consistently retrieved seven (K=7) clusters. These clusters aligned closely with parental inbreds. Two collard inbred parents, elongated and green genotypes, were grouped into a single cluster, thus confirming their genetic similarity

(Fig. 4.1B).

Parental effects on overall liking of leafy genotypes

Consumer liking and familiarity of each genotype in the half-diallel was quantitatively evaluated through an online consumer survey. Each genotype was

151

viewed and assessed for liking and familiarity and average of 113 participants. Survey participants (n = 564) were primarily female (n = 417), white/Caucasian (n = 347), and reported eating leafy Brassicas between a few times per month to 2-3x per week

(Appendix 4.3). The survey received representation from participants in 30 U.S. states, the overwhelming majority of which were from New York. Very few participants

(3%; n = 17) reported they “do not consume” leafy Brassicas.

Figure 4.2. Diallel mating design images used in online consumer survey with n = 8 inbred leafy B. oleracea parents pictured along the axes. Background of hybrid progeny are shaded in orange from least liked (5.25; light orange) to most liked (7.08; dark orange) on a 9pt. hedonic scale by participants in the online consumer survey. Gray background was not included in the survey due to low seed supply.

152

Mean participant liking across genotypes ranged from 5.25 (“neither like-nor dislike”) to 7.08 (“like moderately”) on a 9pt. hedonic scale (Fig. 4.2). Among the most liked genotypes were crosses with collard parents and curly kale parents.

Between 39.3-60.1% of each parental inbred was recovered in the hybrid progeny

(Fig. 4.3), and a majority of the well-liked genotypes retained some collard (yellow, elongated, or green) backgrounds. Surprisingly, parental inbreds comprised some of the most liked (yellow and elongated collards) and least liked (jagged leaf broccoli and

Tuscan kale) genotypes. Relationships between leaf pigmentation and liking were subtle, but generally, blue-green and purple leaf colors were less preferred to lighter, yellow green leaf pigments (Fig. 4.2). In addition, a strong linear relationship between liking and visual volume was observed.

Liking data were subject to mixed model statistical analysis to estimate variance in participant liking due to underlying genetic profiles. Namely, random

2 2 2 effect variance components due to GCA (𝜎GCA ), SCA (𝜎SCA ), participant (𝜎p ), and

2 random error (𝜎 ) were reported. Random effect variance components from participant liking with and without covariance structure (i.e. additive and dominance relationship matrices) were minimally different. Inclusion of SNP marker relationship matrices reduced random variance due to GCA and SCA, but the reduction was negligible. A very small, non-zero percentage (3.3%) of random effect variance was due to the genetic effects of parental selection (GCA) and hybrid specific hybrid cross combinations (SCA) (Table 4.2). Broad sense heritability was very high (H2 = 0.93) and narrow sense heritability was moderately high (h2 = 0.43) when calculated on a plot-level basis (Fig. 4.4). These values were significantly reduced when individual

153

Figure 4.3. Visual assessment of parental effects on overall consumer liking of genotypes (n = 35) in the online consumer survey. Genotypes are vertically ordered from least (bottom) to most (top) liked (raw mean data). A) Genotypes colored by the four most predominant colors (HEX) identified using k-means clustering (n = 5 clusters; background color removed). percent color composition of four k-means clustering (K = 5; removed black background cluster). B) Genotypes colored according to their parental cluster (K = 7) representation identified using FastStructure analysis. Elongated and green collard parental types grouped to a single cluster.

154

participant heritability calculations were performed, but still highlight the non-zero signal in consumer liking due to genetic components.

Figure 4.4. Broad (H2) and narrow (h2) sense heritability of consumer liking calculated using variance components from genotype-informed mixed models with and without the fixed effect of familiarity. In each model, heritability was calculated on an experimental (n = 113.2; green) and individual participant (n = 1; yellow) basis to illustrate the effect of sample size in heritability equations.

Role of familiarity in assessment of liking

A strong correlation (rho = 0.7) between mean genotype liking and familiarity was established (Fig. 4.5A). This trend was consistently observable across individual genotype data. Mean participant familiarity with each genotype ranged from 2.8 – 4.4

(5pt scale). Eight out of the top ten most familiar genotypes consisted of a cross with a curly kale (either green or red curly kale parents). The three least familiar genotypes were Tuscan kale, jagged leaf broccoli, and deep purple kale parents.

Familiarity was significantly affected by consumption rate (p < 0.001) and age

(p < 0.05). As expected, participants who consumed leafy Brassicas at a higher

155

frequency were more likely to rate genotypes with higher familiarity. Similarly, as the average age of participants increased, the likelihood of rating genotypes “more familiar” also increased. Gender, ethnicity, and favorite type of leafy Brassica did not significantly impact familiarity ratings. Inclusion of the fixed effect of familiarity in consumer liking models significantly reduced breeding values (BLUPs) (Table 4.3) and significantly impacted the relative ranking of genotypes (Fig. 4.5B). Green borecole kale types already familiar to consumers were penalized upon addition of familiarity to the model, while genotypes with pigment/red coloration generally shifted upward in their relative breeding values. Addition of familiarity to mixed

2 2 models nearly halved the random variance due to genetic effects (𝜎GCA and 𝜎SCA ) in

Figure 4.5. Relationships between liking and familiarity of B. oleracea genotypes (n = 35; gray) included in the online consumer survey. A) Correlation between mean liking and familiarity among genotypes, as evaluated by a Spearman Rank correlation test. B) Predicted mean change in genotype BLUPs between models with and without the fixed effect of familiarity. Green line represents an x = y line with a slope of one and an intercept of zero; ranks changed are noted by genotypes above or below this line.

156

comparison to models without familiarity (Table 4.2). Heritability estimates were also reduced, though estimates calculated on an experimental basis were not as significantly impacted (Fig. 4.4).

Associated morphological and nutritional traits

A preliminary evaluation of associated leaf morphological (Appendix 4.4) and nutritional quality traits (Appendices 4.5 and 4.6) was conducted to characterize the diallel plant materials beyond consumer liking. Independent principal component analyses (PCA) were performed on mean leaf morphological trait data, glucosinolate content, and carotenoid content measured during Fall 2019 field trials (Fig. 4.6).

Leaf morphological relationships highlighted the diversity of leaf shape among diallel plant materials. The first two PCs explained 72.6% of the morphological variation among diallel parents and hybrid progeny (Fig. 4.6A). The first PC (49.35%) primarily separated borecole (curly kale) genotypes associated with high dry matter

(%) and lamina curl from collard genotypes associated with large lamina width.

Lamina width was strongly correlated with leaf visual volume (%) calculated from survey images. Collard parental types and progeny were the widest genotypes among all materials. The second PC (26.26%) separated Tuscan kale genotypes on the basis of lamina blister and lamina length. Leaf blistering was only present among Tuscan kale genotypes and blistering phenotypes were equally expressed in their hybrid progeny.

157

Figure 4.6. Principal components analysis (PCA) of leaf morphological traits (A), glucosinolates (B), and carotenoids (C) among n = 36 inbred and hybrid lines developed from a half diallel B. oleracea mating design. Genotypes (green) are labeled with a two-digit identifier denoting the female (#) designators from each parental inbred it was derived. Factor loadings (yellow) indicate the influence of each variable (red) on the component.

158

Collard genotypes were clearly resolved from other diallel materials in nutritional content. The three collard parental inbreds (yellow, elongated, and green) were correlated with higher glucosinolate content and separated from other genotypes along PC1 (46.3%) (Fig. 4.6B). Neoglucobrassicin and glucoiberin exhibited a positive association with PC2 (13.55%) and correlated with jagged leaf broccoli and borecole (curly) kale genotypes. Over 75% of carotenoid content was explained by

PC1 (Fig. 4.6C). Carotenoid content and total chlorophyll among Tuscan kale genotypes was higher than other genotypes. Similarly, the deep purple kale genotypes exhibited robust carotenoid profiles due to alpha and beta-carotene content.

Genotypic models were used to obtain random variance components of transformed carotenoid (Appendix 4.9) and glucosinolate (Appendix 4.10)

2 phenotypes. Variance estimates due to the effect of parental selection (𝜎GCA ) and

2 specific hybrid cross combinations (𝜎SCA ) were calculated for each nutritional compound. Parental selection explained more than 15% of total random model variance for glucoprogoitrin, sinigrin, and gluconasturtiin. Phenotypic variance in

2 glucobarbarin and glucoraphanin, in contrast, was predominately explained by 𝜎SCA .

For carotenoid compounds, phenotypic variances were nearly equally explained by variance due to parental selection and specific hybrid cross combinations. Over 20% of the random variance in violaxanthin, neoxanthin, lutein, alpha carotene, and total chlorophyll phenotypes was explained by genetic components. Random error variance remained relatively high across nutritional compounds, although morphological and nutritional trait data are preliminary and should be verified with further replication.

These results, however, establish the framework for future research into the genetic

159

control of complex horticultural quality traits in leafy Brassicas.

Discussion

Consumer liking is inherently a quantitative trait. Product acceptance is not only a reflection of the intrinsic and extrinsic product qualities that define it, but also a reflection of the complexities that underlie human perception. This perception is partly shaped by previous product exposure and familiarity. Overall acceptance must therefore be contextualized within these underlying motivators derived from past experiences. This study sought to explore the implications of familiarity on product acceptance and investigated the heritable genetic components in determining the overall liking of fresh market leafy Brassicas.

A diallel mating design with materials in our breeding program enabled systematic evaluation of genetic diversity and leaf morphological traits and connected these results to a larger body of consumer acceptance methodologies. This study represents the first effort to genetically characterize Brassica breeding materials from the Cornell AgriTech program selected on the basis of phenotypic diversity, and it aided in the development of valuable strategies to identify parental sets based on a combination of genotypic and phenotypic information. The high-quality SNP markers developed through GBS were useful to estimate inbreeding coefficients and genetic relationships among sampled materials, and these results will inform future breeding efforts.

We used advanced, inbred germplasm to generate the diallel experiment.

Several parental genotypes, including the deep red/purple kale and Tuscan kale, exhibited moderate-to-severe phenotypic inbreeding depression and also produced

160

very little seed under standard pollination techniques, potentially constraining the capacity of inbred biennial leafy Brassicas to maintain high levels of homozygosity.

Additional research with diverse materials and alternative methods to calculate inbreeding coefficients is required to draw a clear connection between inbreeding generation and heterozygosity (Ackerman et al., 2017). Generally speaking, diallel hybrid progeny were more heterozygous than commercial hybrid cultivars, which could reflect a lack of complementary markers (i.e. reduced genetic diversity) between commercial parents used in commercial breeding development, differences in seed regeneration protocols, or bias in breeding selection approaches. Four open-pollinated commercial cultivars sourced from sukuma wiki (collard) materials in Kenya had very low inbreeding coefficients. This was to be expected, as they are often sold as open- pollinated line mixes to overcome self-incompatibility and generate higher seed quantities. These materials may represent germplasm that is unavailable or underutilized in European or North American breeding programs but could contribute to diversification of these germplasm pools.

Principal component (PC) analysis demonstrated the program’s inherent diversity and capacity to generate additional diversity beyond the sampled commercial cultivars (Fig. 4.1A). Principal component projections were sufficient to resolve broccoli/cauliflower, cabbage/collard, Tuscan kale, and borecole kale genotypic clusters. Existing genetic diversity literature in B. oleracea focuses largely on annual morphotypes, including broccoli, Chinese kale, and cauliflower, and cabbage. Biennial

Tuscan kale types, borecole kale types, and collards, are typically underrepresented or represented by only a few cultivars in most B. oleracea diversity studies (for

161

exceptions see: Christensen et al., 2011; Okumus and Balkaya, 2007; Pelc et al.,

2015). These types appear to be readily differentiated from other B. oleracea morphotypes and continued genotyping of breeding materials may provide a valuable resource to diversification efforts.

The signal in consumer liking due to genetic components in the diallel was

2 2 very low. Random variance in liking due to additive (𝜎GCA ) and dominance (𝜎SCA ) genetic effects was less than 3.5% of total phenotypic variance, and this percentage decreased to 1.7% after accounting for familiarity as a fixed effect (Table 4.2). The addition of additive/dominance relationship matrices to mixed model analysis produced only minor changes in genetic variance components, likely because the analysis was performed with a pre-existing structured design. Inclusion of genotypic relationship matrices may allow for more accurate estimation of genetic variance/breeding values and informed heritability calculations (Lee et al., 2010;

Oakey et al., 2007; Zapata-Valenzuela et al., 2013). Proper estimation of genetic variance components was significant within the context of this study, given the overarching goal of determining whether consumer liking was a heritable plant trait.

Heritability calculations sought to describe the percentage of variance in consumer liking due to genetic factors in the plant population. This quantitative descriptor can be interpreted as the percentage of phenotypic variance due to genetic variance components [i.e. broad sense (H2) or narrow sense (h2)]. Unsurprisingly, the broad sense heritability in our study was relatively high (H2 > 0.85) and, given the large number of genotypes present in the online survey, this study retained a strong capacity to identify genotypes participants tended to like more than others.

162

Interestingly, the values of narrow sense heritability (h2 < 0.45; Fig. 4.4) highlight the ability to select and combine parental inbreds which produce progeny that tend to be more preferred (Fig. 4.3B). Heritability estimates decreased dramatically when the effective sample size was reduced, and we therefore recommend consumer sample sizes greater than 100 participants in future studies.

This study presents a novel exploration between genetic components and consumer liking. In contrast to previous work, we apply heritability estimates of consumer preference to the product, rather than the individual (Simonson and Sela,

2011). Direct comparisons to other estimates of heritability remain limited and the interpretation/extension of heritability estimates must be performed with caution.

Drennan et al. (1983) performed several consumer preference studies in Gerbera clones using full-sib families and parent-offspring regression calculations, but their experimental design complicated interpretation of their generally high (h2 > 0.6) heritability estimates. The non-zero estimates of genetic variance, albeit small, in this study may warrant the incorporation of consumer liking into breeder’s selection indices, whether formal or informal in nature.

Establishing a heritable nature of consumer liking is the first step in understanding how to select and breed for consumer liking. Similar to yield components, underlying qualitative and quantitative drivers of liking must be identified (van Trijp et al., 2007). The diallel design is in many ways akin to the

IdeaMap® proposed by Moskowitz and Martin (1993), whereby a multitude of

“concept elements” can be developed and presented in various combinations to participants. This survey design relied on photographic images to represent concept

163

combinations informed by pedigree and genetic relationships. Plant traits such as color, shape, and size were inherently diverse in the design, but were not explicitly defined in relative combinations. Future studies may include additional questions to profile these plant traits or conduct an independent conjoint analysis to elucidate drivers of liking and market segments (Behe, 2006; Endrizzi et al., 2015; Oltman et al., 2014).

While evaluation of leaf morphological traits (e.g. dry matter, leaf shape, lamina curl, etc.) and nutritional quality (e.g. carotenoid and glucosinolate content) was informative to characterize the diallel design, a direct connection between consumer liking and perception was not possible within this experimental design.

Phenotypes such as dry matter (%) and glucosinolate content may be relevant to flavor, texture, and nutritional quality (Barba et al., 2016; Bell et al., 2018; Wieczorek et al., 2018), but they may not have a direct impact on visual liking scores. Although glucosinolate content was strongly correlated with highly liked collard genotypes and dry matter was highest among borecole (curly) kale types, it is unlikely that study participants were aware of this. Carotenoid content correlated with deeply pigmented and dark green/blue-green genotypes, and were among the least-liked genotypes.

Color preference has been extensively explored in food and personal care products, and a significant body of literature suggests cross-modal effects of color on the perception of flavor and texture (Chylinski et al., 2015; Koch and Koch, 2003;

Spence et al., 2010). Color in vegetables may also be used as a proxy for perceived quality and nutritional content (Drewnowski, 1996). Population specific color preferences have been described, but they may not be consistent across products or

164

universally accepted (Madden et al., 2000; Taylor et al., 2013). Color quantification from survey images produced mixed or stratified trends in this study, and further studies are warranted with materials of different ages (i.e. baby greens) grown in multiple environments.

This study demonstrates the capacity of online consumer surveys to deliver relevant consumer data to breeding programs interested in probing acceptance of appearance-related traits. Large plant population sizes, heterogeneity of early- generation materials, lack of sufficient seed supply, and seasonal variation are commonly cited obstacles to performing sensory analysis before late-stage testing or cultivar release (Klee and Tieman, 2013; Simon et al., 1981). Evaluating morphological and appearance-related characteristics, however, is possible using static images and can be performed at almost any time in the breeding process. Static images have been used extensively in horticultural crops to evaluate yield and quality traits

(Gehan et al., 2017; Rodríguez et al., 2010), and current research and development of improved imaging and image analysis platforms could dramatically improve the reproducibility of online surveys.

While this online study reached over four hundred participants in less than one-week, some issues must be addressed regarding image quality and survey interface in future studies. As expected from traditional acceptance testing, we observed a high degree of participant variance. Including a question regarding the type of device used to take this survey may have helped parse apart effects due to participant and testing “environment.” Standard lighting procedures are considered essential for in-house consumer testing, as lighting conditions can modulate the

165

perception of flavor and appearance (Hasenbeck et al., 2014; Oberfeld et al., 2009), and luminance can greatly affect the perception of freshness (Arce-Lopera et al.,

2013). Great care was taken in this survey to produce consistent, high quality images, but there is still room to improve accuracy and presentation of leaf material digitally.

Variation in leaf pigmentation and degree of epicuticular wax, for example, complicated photographic settings. In carrots, Schifferstein et al. (2019) observed reduced perception of freshness in low-saturation images and an association with artificiality in highly saturated images of carrots. Participant variance and other sources of error may be reduced with embedded image replicates. For example, images from multiple perspectives, occupying different levels of image space, under various lighting conditions may provide for a more holistic phenotype and better characterization of true genetic effects.

Due to consumer fatigue considerations, participants rated a subset of seven genotypes and were asked only about familiarity and liking during this survey. While samples were independently presented and randomized across consumers, pairing liking and familiarity questions may have conflated results by creating a false relationship between these two quantitative variables, and these two variables could be validated in isolation. Ultimately, in-person acceptance testing with a defined subset of plant materials, would validate online surveys to gauge consumer liking of plant materials. Additional questions relating to product utility and functionality, willingness to pay, and perceived nutritional content would aid in understanding the potential of these genotypes in the marketplace. Further, recruiting more geographically diverse participants would aid in consumer segmentation and

166

identification of regional preferences.

Well liked products largely grouped into three market categories: products liked independently of familiarity, products liked when paired with “familiar” traits, and products that performed well only after accounting for familiarity. Flat, broad leafed collards and genotypes with lamina curl were well liked independently of familiarity. These genotypes could be considered market standards, as their phenotypes are already commercially accepted and exhibited high GCA values. In contrast, Tuscan kale crosses with a leaf blistering phenotype generally performed poorly unless combined with “familiar” traits such as a broader leaf or slight leaf curl.

These blistering genotypes exhibited high SCA and may represent incremental product innovations that improve upon existing standards (Costa and Jongen, 2006;

Lopetcharat et al., 2012). Finally, the last category of well-liked genotypes was comprised of pigmented (i.e. red or purple) genotypes whose relative breeding values increased only after accounting for product familiarity. These pigmented genotypes were generally not well liked unless they were paired with a familiar leaf type or a contrasting leaf color such as bright green or green-yellow. This phenomenon may be interesting to explore within the context of color contrast and modulation of background color (Schifferstein et al., 2017). These products were relatively unfamiliar to participating consumers and represent novel, niche, or disruptive additions to the product space. They tended to generate polarizing responses. They may require targeted marketing or innovative techniques to shift initial consumer receptivity.

Variety in choice of fresh fruits and vegetables can increase consumption in these

167

product categories (Just et al., 2012; Meengs et al., 2012), but little information exists regarding purchasing and consumption when novel, market-class specific cultivars are added to the product space. Further, few studies have sought to understand market flexibility and capacity to absorb new fresh market vegetable cultivars before market saturation occurs. The genetic component of consumer acceptance is only one variable in defining successful products in the market. Information gleaned from this mating design and accompanying genetic analysis is directive but not necessarily predictive of products that will fit in rapidly changing food systems. Leveraging the disciplines of sociology, marketing, and economics in a systems approach is required to fully anticipate which genotypes have market potential.

Conclusion

This study highlights the inherent diversity and potential of materials from our program to diversify the leafy Brassica product space. We identified a small but heritable genetic component in leafy Brassicas influencing consumer liking of appearance attributes and characterized the inherent diversity of our breeding program.

The genotypes and markers developed herein serve as valuable research materials and a resource for the development of new leafy Brassica cultivars.

168

References

Ackerman, M.S., Johri, P., Spitze, K., Xu, S., Doak, T.G., Young, K., Lynch, M., 2017. Estimating Seven Coefficients of Pairwise Relatedness Using Population-Genomic Data. Genetics 206, 105–118. https://doi.org/10.1534/genetics.116.190660 Allan, A.C., Espley, R.V., 2018. MYBs Drive Novel Consumer Traits in Fruits and Vegetables. Trends in Plant Science 23, 693–705. https://doi.org/10.1016/j.tplants.2018.06.001 Arce-Lopera, C., Masuda, T., Kimura, A., Wada, Y., Okajima, K., 2013. Luminance distribution as a determinant for visual freshness perception: Evidence from image analysis of a cabbage leaf. Food Quality and Preference, Ninth Pangborn Sensory Science Symposium 27, 202–207. https://doi.org/10.1016/j.foodqual.2012.03.005 Bar, M., Ori, N., 2014. Leaf development and morphogenesis. Development 141, 4219–4230. https://doi.org/10.1242/dev.106195 Barba, F.J., Nikmaram, N., Roohinejad, S., Khelfa, A., Zhu, Z., Koubaa, M., 2016. Bioavailability of Glucosinolates and Their Breakdown Products: Impact of Processing. Front Nutr 3. https://doi.org/10.3389/fnut.2016.00024 Barrett, D.M., Beaulieu, J.C., Shewfelt, R., 2010. Color, Flavor, Texture, and Nutritional Quality of Fresh-Cut Fruits and Vegetables: Desirable Levels, Instrumental and Sensory Measurement, and the Effects of Processing. Critical Reviews in Food Science and Nutrition 50, 369–389. https://doi.org/10.1080/10408391003626322 Bates, D., Mächler, M., Bolker, B., Walker, S., 2015. Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software 67, 1–48. https://doi.org/10.18637/jss.v067.i01 Behe, B.K., 2006. Conjoint Analysis Reveals Consumers Prefer Long, Thin Asparagus Spears. HortScience 41, 1259–1262. https://doi.org/10.21273/HORTSCI.41.5.1259 Bell, L., Oloyede, O.O., Lignou, S., Wagstaff, C., Methven, L., 2018. Taste and Flavor Perceptions of Glucosinolates, Isothiocyanates, and Related Compounds. Molecular Nutrition & Food Research 62, 1700990. https://doi.org/10.1002/mnfr.201700990 Belser, C., Istace, B., Denis, E., Dubarry, M., Baurens, F.-C., Falentin, C., Genete, M., Berrabah, W., Chèvre, A.-M., Delourme, R., Deniot, G., Denoeud, F., Duffé, P., Engelen, S., Lemainque, A., Manzanares-Dauleux, M., Martin, G., Morice, J., Noel, B., Vekemans, X., D’Hont, A., Rousseau-Gueutin, M., Barbe, V., Cruaud, C., Wincker, P., Aury, J.-M., 2018. Chromosome-scale assemblies of plant genomes using nanopore long reads and optical maps. Nature Plants 4, 879–887. https://doi.org/10.1038/s41477-018-0289-4 Bernardo, R.N., 2010. Estimating Genetic Variances, in: Bernardo, R.N. (Ed.), Breeding for Quantitative Traits in Plants. Stemma Press, Woodbury, Minn.

169

Bingham, A., Hurling, R., Stocks, J., 2005. Acquisition of liking for spinach products. Food Quality and Preference 16, 461–469. https://doi.org/10.1016/j.foodqual.2004.09.006 Boodley, J.W., Sheldrake, R., 1972. Cornell Peat-Lite Mixes for Commercial Growing. Information Bulletin 43. Browning, B.L., Browning, S.R., 2016. Genotype Imputation with Millions of Reference Samples. Am J Hum Genet 98, 116–126. https://doi.org/10.1016/j.ajhg.2015.11.020 Chalker‐Scott, L., 1999. Environmental Significance of Anthocyanins in Plant Stress Responses. Photochemistry and Photobiology 70, 1–9. https://doi.org/10.1111/j.1751-1097.1999.tb01944.x Cheng, F., Sun, R., Hou, X., Zheng, H., Zhang, F., Zhang, Y., Liu, B., Liang, J., Zhuang, M., Liu, Yunxia, Liu, D., Wang, Xiaobo, Li, P., Liu, Yumei, Lin, K., Bucher, J., Zhang, N., Wang, Y., Wang, H., Deng, J., Liao, Y., Wei, K., Zhang, X., Fu, L., Hu, Y., Liu, Jisheng, Cai, C., Zhang, Shujiang, Zhang, Shifan, Li, F., Zhang, Hui, Zhang, J., Guo, N., Liu, Z., Liu, Jin, Sun, C., Ma, Y., Zhang, Haijiao, Cui, Y., Freeling, M.R., Borm, T., Bonnema, G., Wu, J., Wang, Xiaowu, 2016. Subgenome parallel selection is associated with morphotype diversification and convergent crop domestication in Brassica rapa and Brassica oleracea. Nature Genetics 48, 1218–1224. https://doi.org/10.1038/ng.3634 Chiang, M.S., Crête, R., 1976. Diallel Analysis of the Inheritance of Resistance to Race 6 of Plasmodiophora Brassicae in Cabbage. Can. J. Plant Sci. 56, 865– 868. https://doi.org/10.4141/cjps76-140 Chitwood, D.H., Sinha, N.R., 2016. Evolutionary and Environmental Forces Sculpting Leaf Development. Current Biology 26, R297–R306. https://doi.org/10.1016/j.cub.2016.02.033 Christensen, S., von Bothmer, R., Poulsen, G., Maggioni, L., Phillip, M., Andersen, B.A., Jørgensen, R.B., 2011. AFLP analysis of genetic diversity in leafy kale (Brassica oleracea L. convar. acephala (DC.) Alef.) landraces, cultivars and wild populations in Europe. Genet Resour Crop Evol 58, 657–666. https://doi.org/10.1007/s10722-010-9607-z Christie, B.R., Shattuck, V.I., 1992. The Diallel Cross: Design, Analysis, and Use for Plant Breeders, in: Janick, J. (Ed.), Plant Breeding Reviews. John Wiley & Sons, pp. 9–32. Chylinski, M., Northey, G., Ngo, L.V., 2015. Cross-modal Interactions between Color and Texture of Food. Psychology & Marketing 32, 950–966. https://doi.org/10.1002/mar.20829 Costa, A.I.A., Jongen, W.M.F., 2006. New insights into consumer-led food product development. Trends in Food Science & Technology 17, 457–465. https://doi.org/10.1016/j.tifs.2006.02.003 Covarrubias-Pazaran, G., 2016. Genome-Assisted Prediction of Quantitative Traits Using the R Package sommer. PLoS One 11. https://doi.org/10.1371/journal.pone.0156744

170

Danecek, P., Auton, A., Abecasis, G., Albers, C.A., Banks, E., DePristo, M.A., Handsaker, R.E., Lunter, G., Marth, G.T., Sherry, S.T., McVean, G., Durbin, R., 2011. The variant call format and VCFtools. Bioinformatics 27, 2156– 2158. https://doi.org/10.1093/bioinformatics/btr330 Davis, E.H., Morgan, J.T., 2015. Collards: A Southern Tradition from Seed to Table. University of Alabama Press. Drennan, D., Harding, J., Byrne, T., 1983. Heritability of consumer preference in Gerbera. Euphytica 32, 465–474. https://doi.org/10.1007/BF00021456 Drewnowski, A., 1996. From asparagus to zucchini: mapping cognitive space for vegetable names. Journal of the American College of Nutrition 15, 147–153. https://doi.org/10.1080/07315724.1996.10718580 Eigenbrode, S.D., Stoner, K.A., Shelton, A.M., Kain, W.C., 1991. Characteristics of Glossy Leaf Waxes Associated with Resistance to Diamondback Moth (Lepidoptera: Plutellidae) in Brassica oleracea. J Econ Entomol 84, 1609– 1618. https://doi.org/10.1093/jee/84.5.1609 Elshire, R.J., Glaubitz, J.C., Sun, Q., Poland, J.A., Kawamoto, K., Buckler, E.S., Mitchell, S.E., 2011. A Robust, Simple Genotyping-by-Sequencing (GBS) Approach for High Diversity Species. PLOS ONE 6, e19379. https://doi.org/10.1371/journal.pone.0019379 Endelman, J.B., 2011. Ridge Regression and Other Kernels for Genomic Selection with R Package rrBLUP. The Plant Genome 4, 250–255. https://doi.org/10.3835/plantgenome2011.08.0024 Endrizzi, I., Torri, L., Corollaro, M.L., Demattè, M.L., Aprea, E., Charles, M., Biasioli, F., Gasperi, F., 2015. A conjoint study on apple acceptability: Sensory characteristics and nutritional information. Food Quality and Preference 40, 39–48. https://doi.org/10.1016/j.foodqual.2014.08.007 Esteban, R., Moran, J.F., Becerril, J.M., García-Plazaola, J.I., 2015. Versatility of carotenoids: An integrated view on diversity, evolution, functional roles and environmental interactions. Environmental and Experimental Botany, Functional roles of secondary metabolites in plant-environment interactions 119, 63–75. https://doi.org/10.1016/j.envexpbot.2015.04.009 Fehr, W.R., 1987. Principles of cultivar development. Volume 1. Theory and technique. Macmillan, New York. Gardner, C.O., Eberhart, S.A., 1966. Analysis and Interpretation of the Variety Cross Diallel and Related Populations. Biometrics 22, 439–452. https://doi.org/10.2307/2528181 Gehan, M.A., Fahlgren, N., Abbasi, A., Berry, J.C., Callen, S.T., Chavez, L., Doust, A.N., Feldman, M.J., Gilbert, K.B., Hodge, J.G., Hoyer, J.S., Lin, A., Liu, S., Lizárraga, C., Lorence, A., Miller, M., Platon, E., Tessman, M., Sax, T., 2017. PlantCV v2: Image analysis software for high-throughput plant phenotyping. PeerJ 5. https://doi.org/10.7717/peerj.4088 Glaubitz, J.C., Casstevens, T.M., Lu, F., Harriman, J., Elshire, R.J., Sun, Q., Buckler, E.S., 2014. TASSEL-GBS: A High Capacity Genotyping by Sequencing Analysis Pipeline. PLOS ONE 9, e90346. https://doi.org/10.1371/journal.pone.0090346

171

Grandclement, C., Laurens, F., Thomas, G., 1996. Genetic analysis of resistance to clubroot (Plasmodiophora brassicae Woron) in two Brassica oleracea groups (ssp. acephala and ssp. botrytis) through diallel analysis. Plant Breeding 115, 152–156. https://doi.org/10.1111/j.1439-0523.1996.tb00892.x Griffing, B., 1956. A generalised treatment of the use of diallel crosses in quantitative inheritance. Heredity 10, 31–50. https://doi.org/10.1038/hdy.1956.2 Hasenbeck, A., Cho, S., Meullenet, J.-F., Tokar, T., Yang, F., Huddleston, E.A., Seo, H.-S., 2014. Color and illuminance level of lighting can modulate willingness to eat bell peppers. Journal of the Science of Food and Agriculture 94, 2049– 2056. https://doi.org/10.1002/jsfa.6523 Hatier, J.-H.B., Gould, K.S., 2009. Anthocyanin Function in Vegetative Organs, in: Winefield, C., Davies, K., Gould, K. (Eds.), Anthocyanins: Biosynthesis, Functions, and Applications. Springer, New York, NY, pp. 1–19. https://doi.org/10.1007/978-0-387-77335-3_1 Hayman, B.I., 1954. The Theory and Analysis of Diallel Crosses. Genetics 39, 789– 809. Hodgkin, T., 1978. The inheritance of partial self-compatibility inBrassica oleracea L.: results from a half diallel homozygous for a highly recessive S-allele. Theoret. Appl. Genetics 53, 81–87. https://doi.org/10.1007/BF00817836 Horikoshi, M., Tang, Y., Dickey, A., Grenié, M., Thompson, R., Selzer, L., Strbenac, D., Voronin, K., 2016. ggfortify: Data Visualization Tools for Statistical Analysis Results. Hutchings, J.B., 1999. The Philosophy of Total Appearance, In: Hutchings, J.B. (Ed.), Food Colour and Appearance. Springer US, Boston, MA, pp. 30–60. https://doi.org/10.1007/978-1-4615-2373-4_2 Johnston, T.D., 1968. Studies on a diallel cross and double-cross hybrids among various Brassica oleracea types. Euphytica 17, 63–73. https://doi.org/10.1007/BF00038967 Jongen, W.M.F., 2000. Food Supply Chains: From Productivity Toward Quality, in: Shewfelt, R.L., Brückner, B. (Eds.), Fruit and Vegetable Quality: An Integrated View. Technomic Pub. Co., Lancaster, PA, pp. 3–20. https://doi.org/10.1201/9781482293937-8 Just, D.R., Lund, J., Price, J., 2012. The Role of Variety in Increasing the Consumption of Fruits and Vegetables Among Children. Agricultural and Resource Economics Review 41, 72–81. https://doi.org/10.1017/S1068280500004196 Kaushik, P., Plazas, M., Prohens, J., Vilanova, S., Gramazio, P., 2018. Diallel genetic analysis for multiple traits in eggplant and assessment of genetic distances for predicting hybrids performance. PLOS ONE 13, e0199943. https://doi.org/10.1371/journal.pone.0199943 Khoo, H.E., Azlan, A., Tang, S.T., Lim, S.M., 2017. Anthocyanidins and anthocyanins: colored pigments as food, pharmaceutical ingredients, and the potential health benefits. Food Nutr Res 61. https://doi.org/10.1080/16546628.2017.1361779

172

Kim, M.-A., Dessirier, J.-M., van Hout, D., Lee, H.-S., 2015. Consumer context- specific sensory acceptance tests: Effects of a cognitive warm-up on affective product discrimination. Food Quality and Preference 41, 163–171. https://doi.org/10.1016/j.foodqual.2014.11.019 Klee, H.J., Tieman, D.M., 2013. Genetic challenges of flavor improvement in tomato. Trends in Genetics 29, 257–262. https://doi.org/10.1016/j.tig.2012.12.003 Koch, C., Koch, E.C., 2003. Preconceptions of Taste Based on Color. The Journal of Psychology 137, 233–242. https://doi.org/10.1080/00223980309600611 Kopsell, D.A., Kopsell, D.E., Lefsrud, M.G., Curran-Celentano, J., Dukach, L.E., 2004. Variation in Lutein, β-carotene, and Chlorophyll Concentrations among Brassica oleracea Cultigens and Seasons. HortScience 39, 361–364. https://doi.org/10.21273/HORTSCI.39.2.361 Krinsky, N.I., Johnson, E.J., 2005. Carotenoid actions and their relation to health and disease. Molecular Aspects of Medicine 26, 459–516. https://doi.org/10.1016/j.mam.2005.10.001 Krzywinski, M., 2017. Image Color Summarizer. Lan, T.-H., Paterson, A.H., 2001. Comparative mapping of QTLs determining the plant size of Brassica oleracea. Theor Appl Genet 103, 383–397. https://doi.org/10.1007/s001220100615 Lawless, H.T., Heymann, H., 2010. Acceptance Testing, in: Sensory Evaluation of Food, Food Science Text Series. Springer, New York, NY, pp. 325–347. https://doi.org/10.1007/978-1-4419-6488-5_14 Lee, S.H., Goddard, M.E., Visscher, P.M., Werf, J.H. van der, 2010. Using the realized relationship matrix to disentangle confounding factors for the estimation of genetic variance components of complex traits. Genet Sel Evol 42, 1–14. https://doi.org/10.1186/1297-9686-42-22 Lopetcharat, K., Paredes, D., Beckley, J.H., Paredes, M.D., Paredes, D., 2012. Product Innovation Toolbox : A Field Guide to Consumer Understanding and Research. John Wiley & Sons, Incorporated, Hoboken, UNITED STATES. Madden, T.J., Hewett, K., Roth, M.S., 2000. Managing Images in Different Cultures: A Cross-National Study of Color Meanings and Preferences. Journal of International Marketing 8, 90–107. https://doi.org/10.1509/jimk.8.4.90.19795 Meengs, J.S., Roe, L.S., Rolls, B.J., 2012. Vegetable Variety: An Effective Strategy to Increase Vegetable Intake in Adults. Journal of the Academy of Nutrition and Dietetics 112, 1211–1215. https://doi.org/10.1016/j.jand.2012.05.013 Möhring, J., Melchinger, A.E., Piepho, H.P., 2011. REML-Based Diallel Analysis. Crop Science 51, 470–478. https://doi.org/10.2135/cropsci2010.05.0272 Moskowitz, H.R., Martin, D.C., 1993. How computer aided design and presentation of concepts speeds up the product development process, in: 46th ESOMAR Conference. Presented at the “The many faces of quality now and in future”, Copenhagen, Denmark, pp. 405–419. Oakey, H., Verbyla, A.P., Cullis, B.R., Wei, X., Pitchford, W.S., 2007. Joint modeling of additive and non-additive (genetic line) effects in multi-environment trials. Theor Appl Genet 114, 1319–1332. https://doi.org/10.1007/s00122-007-0515- 3

173

Oberfeld, D., Hecht, H., Allendorf, U., Wickelmaier, F., 2009. Ambient Lighting Modifies the Flavor of Wine. Journal of Sensory Studies 24, 797–832. https://doi.org/10.1111/j.1745-459X.2009.00239.x Okumus, A., Balkaya, A., 2007. Estimation of genetic diversity among Turkish kale populations (Brassica oleracea var. acephala L.) using RAPD markers. Russ J Genet 43, 411–415. https://doi.org/10.1134/S1022795407040096 Oltman, A.E., Jervis, S.M., Drake, M.A., 2014. Consumer Attitudes and Preferences for Fresh Market Tomatoes. Journal of Food Science 79, S2091–S2097. https://doi.org/10.1111/1750-3841.12638 Pelc, S.E., Couillard, D.M., Stansell, Z.J., Farnham, M.W., 2015. Genetic Diversity and Population Structure of Collard Landraces and their Relationship to Other Brassica oleracea Crops. The Plant Genome 8. https://doi.org/10.3835/plantgenome2015.04.0023 Prescott, J., Bell, G., 1995. Cross-cultural determinants of food acceptability: Recent research on sensory perceptions and preferences. Trends in Food Science & Technology 6, 201–205. https://doi.org/10.1016/S0924-2244(00)89055-X Purcell, S., Neale, B., Todd-Brown, K., Thomas, L., Ferreira, M.A.R., Bender, D., Maller, J., Sklar, P., de Bakker, P.I.W., Daly, M.J., Sham, P.C., 2007. PLINK: A Tool Set for Whole-Genome Association and Population-Based Linkage Analyses. Am J Hum Genet 81, 559–575. R Core Team, 2018. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. Ragaert, P., Verbeke, W., Devlieghere, F., Debevere, J., 2004. Consumer perception and choice of minimally processed vegetables and packaged fruits. Food Quality and Preference 15, 259–270. https://doi.org/10.1016/S0950- 3293(03)00066-1 Raj, A., Stephens, M., Pritchard, J.K., 2014. fastSTRUCTURE: Variational Inference of Population Structure in Large SNP Data Sets. Genetics 197, 573–589. https://doi.org/10.1534/genetics.114.164350 Rindfleisch, A., Inman, J., 1998. Explaining the Familiarity-Liking Relationship: Mere Exposure, Information Availability, or Social Desirability? Marketing Letters 9, 5–19. https://doi.org/10.1023/A:1007958302123 Roberts, W., 2019. Vegetables - US - August 2019. Mintel Academic. Rodríguez, G.R., Moyseenko, J.B., Robbins, M.D., Huarachi Morejón, N., Francis, D.M., van der Knaap, E., 2010. Tomato Analyzer: A Useful Software Application to Collect Accurate and Detailed Morphological and Colorimetric Data from Two-dimensional Objects. J Vis Exp. https://doi.org/10.3791/1856 Sams, C.E., Panthee, D.R., Charron, C.S., Kopsell, D.A., Yuan, J.S., 2011. Selenium Regulates Gene Expression for Glucosinolate and Carotenoid Biosynthesis in Arabidopsis. Journal of the American Society for Horticultural Science 136, 23–34. https://doi.org/10.21273/JASHS.136.1.23 Schifferstein, H.N.J., Howell, B.F., Pont, S.C., 2017. Colored backgrounds affect the attractiveness of fresh produce, but not it’s perceived color. Food Quality and Preference 56, 173–180. https://doi.org/10.1016/j.foodqual.2016.10.011

174

Schifferstein, H.N.J., Wehrle, T., Carbon, C.-C., 2019. Consumer expectations for vegetables with typical and atypical colors: The case of carrots. Food Quality and Preference 72, 98–108. https://doi.org/10.1016/j.foodqual.2018.10.002 Sebastian, R.L., Kearsey, M.J., King, G.J., 2002. Identification of quantitative trait loci controlling developmental characteristics of Brassica oleracea L. Theor Appl Genet 104, 601–609. https://doi.org/10.1007/s001220100743 Simon, P.W., Peterson, C.E., Lindsay, R.C., 1981. The Improvement of Flavor in a Program of Carrot Genetics and Breeding, in: Quality of Selected Fruits and Vegetables of North America, ACS Symposium Series. American Chemical Society, pp. 109–118. https://doi.org/10.1021/bk-1981-0170.ch009 Simonson, I., Sela, A., 2011. On the Heritability of Consumer Decision Making: An Exploratory Approach for Studying Genetic Effects on Judgment and Choice. J Consum Res 37, 951–966. https://doi.org/10.1086/657022 Socquet-Juglard, D., Bennett, A.A., Manns, D.C., Mansfield, A.K., Robbins, R.J., Collins, T.M., Griffiths, P.D., 2016. Effects of Growth Temperature and Postharvest Cooling on Anthocyanin Profiles in Juvenile and Mature Brassica oleracea. J. Agric. Food Chem. 64, 1484–1493. https://doi.org/10.1021/acs.jafc.5b05309 Spence, C., Levitan, C.A., Shankar, M.U., Zampini, M., 2010. Does Food Color Influence Taste and Flavor Perception in Humans? Chem. Percept. 3, 68–84. https://doi.org/10.1007/s12078-010-9067-z Sprague, G.F., Tatum, L.A., 1942. General vs. Specific Combining Ability in Single Crosses of Corn1. Agronomy Journal 34, 923–932. https://doi.org/10.2134/agronj1942.00021962003400100008x Stansell, Z., Farnham, M., Björkman, T., 2019. Complex Horticultural Quality Traits in Broccoli Are Illuminated by Evaluation of the Immortal BolTBDH Mapping Population. Front. Plant Sci. 10. https://doi.org/10.3389/fpls.2019.01104 Stansell, Z., Hyma, K., Fresnedo-Ramírez, J., Sun, Q., Mitchell, S., Björkman, T., Hua, J., 2018. Genotyping-by-sequencing of Brassica oleracea vegetables reveals unique phylogenetic patterns, population structure and domestication footprints. Horticulture Research 5, 1–10. https://doi.org/10.1038/s41438-018- 0040-3 Steenkamp, J.-B.E.M., 1990. Conceptual model of the quality perception process. Journal of Business Research 21, 309–333. https://doi.org/10.1016/0148- 2963(90)90019-A Sun, T., Li, L., 2020. Toward the ‘golden’ era: The status in uncovering the regulatory control of carotenoid accumulation in plants. Plant Science 290, 110331. https://doi.org/10.1016/j.plantsci.2019.110331 Swegarden, H., Stelick, A., Dando, R., Griffiths, P.D., 2019. Bridging Sensory Evaluation and Consumer Research for Strategic Leafy Brassica (Brassica oleracea) Improvement. Journal of Food Science 84, 3746–3762. https://doi.org/10.1111/1750-3841.14831 Tanaka, Y., Sasaki, N., Ohmiya, A., 2008. Biosynthesis of plant pigments: anthocyanins, betalains and carotenoids. The Plant Journal 54, 733–749. https://doi.org/10.1111/j.1365-313X.2008.03447.x

175

Taylor, C., Clifford, A., Franklin, A., 2013. Color preferences are not universal. Journal of Experimental Psychology: General 142, 1015–1027. https://doi.org/10.1037/a0030273 Tsukaya, H., 2006. Mechanism of leaf-shape determination. Annual review of plant biology. Turner, S.D., Maurizio, P.L., Valdar, W., Yandell, B.S., Simon, P.W., 2018. Dissecting the Genetic Architecture of Shoot Growth in Carrot (Daucus carota L.) Using a Diallel Mating Design. G3: Genes, Genomes, Genetics 8, 411–426. https://doi.org/10.1534/g3.117.300235 USDA-NASS, 2019. 2017 Census Volume 1, Chapter 2: State Level Data [WWW Document]. Census of Agriculture. URL www.nass.usda.gov/AgCensus (accessed 3.24.20). van Trijp, H.C.M., Punter, P.H., Mickartz, F., Kruithof, L., 2007. The quest for the ideal product: Comparing different methods and approaches. Food Quality and Preference 18, 729–740. https://doi.org/10.1016/j.foodqual.2007.01.005 Vaughn, T.T., Hoy, C.W., 1993. Effects of Leaf Age, Injury, Morphology, and Cultivars on Feeding Behavior of Phylloteta cruciferae (Coleoptera: Chrysomelidae). Environ Entomol 22, 418–424. https://doi.org/10.1093/ee/22.2.418 Wadhera, D., Capaldi-Phillips, E.D., 2014. A review of visual cues associated with food on food acceptance and consumption. Eating Behaviors 15, 132–143. https://doi.org/10.1016/j.eatbeh.2013.11.003 Wardle, J., Herrera, M.-L., Cooke, L., Gibson, E.L., 2003. Modifying children’s food preferences: the effects of exposure and reward on acceptance of an unfamiliar vegetable. European Journal of Clinical Nutrition 57, 341–348. https://doi.org/10.1038/sj.ejcn.1601541 Wickham, H., 2016. ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag New York. Wieczorek, M.N., Walczak, M., Skrzypczak-Zielińska, M., Jeleń, H.H., 2018. Bitter taste of Brassica vegetables: The role of genetic factors, receptors, isothiocyanates, glucosinolates, and flavor context. Critical Reviews in Food Science and Nutrition 58, 3130–3140. https://doi.org/10.1080/10408398.2017.1353478 Zapata-Valenzuela, J., Whetten, R.W., Neale, D., McKeand, S., Isik, F., 2013. Genomic Estimated Breeding Values Using Genomic Relationship Matrices in a Cloned Population of Loblolly Pine. G3: Genes, Genomes, Genetics 3, 909– 916. https://doi.org/10.1534/g3.113.005975 Zhang, C., Dong, S.-S., Xu, J.-Y., He, W.-M., Yang, T.-L., 2019. PopLDdecay: a fast and effective tool for linkage disequilibrium decay analysis based on variant call format files. Bioinformatics 35, 1786–1788. https://doi.org/10.1093/bioinformatics/bty875 Zheng, X., Levine, D., Shen, J., Gogarten, S.M., Laurie, C., Weir, B.S., 2012. A high performance computing toolset for relatedness and principal component analysis of SNP data. Bioinformatics 28, 3326–3328. https://doi.org/10.1093/bioinformatics/bts606

176

Chapter 5

FINAL CONCLUSIONS AND FUTURE DIRECTIONS

This dissertation research was performed within the context of a Brassica breeding program targeting new phenotypic traits with little accessible consumer and market information. Breeding of cole crops requires balancing reproductive life cycles, diverse plant morphologies, and varying levels of self-incompatibility among parental lines of vegetative crops. As a result, the breeding and development of new cultivars is a time-intensive, long-term endeavor that could benefit from early population information to select and target genotypes with consumer acceptance.

Renewed relevance of several Brassica crops, driven by increased awareness of health and nutrition, has occurred in recent years. Changing food systems have increased the demand for leafy Brassica vegetables in salad markets and beyond. It was of interest to the program’s future direction to incorporate consumer information by identifying existing market standards, consumer preferences, and the acceptance and potential of new traits. Ultimately, this research was an effort to think beyond traditional or existing market classes and develop novel strategies to redefine market expectations.

Key findings from sensory and consumer research in leafy Brassicas

Sensory and quality traits in leafy Brassicas vary among genotypes and germplasm.

Our research began by exploring established sensory and consumer research methods, including home-use tests, a focus group, a trained descriptive panel, and a

177

central location test. Through these outlets, we identified variation in 21 texture, flavor, and appearance traits within the leafy Brassica germplasm studied. Inherent genetic diversity of these attributes demonstrates the potential to breed and select for consumer quality traits, and these attributes provide a baseline for future descriptive or affective sensory work.

Texture has been popularly described as the “final frontier in food science”

(Pierre-Louis, 2017), but the quantification and identification of textural attributes in leafy vegetables has received minimal attention. Textural traits are primary motivators in the acceptance of vegetable crops (Duffy et al., 2017; Kilcast and Fillion, 2001).

Eleven of the 21 significant attributes in descriptive analysis related to leaf texture

(Table 2.4). Durability and shelf life are components of kale’s core identity. Alteration of textural traits must be approached diligently to avoid changing its consumer and market identities. Similarly, bitter, sulfur, and pungent flavors are core attributes expected by kale and collard consumers. Elimination of these flavors or textures could potentially cater to certain consumer clusters but would require redefining market class perception.

Integration of advanced metabolomic profiling and flavor/texture phenotyping could aid in understanding complex relationships between texture, flavor, and consumer liking in leafy Brassicas (Klee and Tieman, 2018). A next logical step in developing capacity to select and breed for textural traits would be to identify underlying plant traits conferring each textural attribute and develop replicable, inexpensive quantification techniques that correlate with human perception. These measurements must be conducted across environments, harvest maturities, and

178

preparation methods to fully understand the sensory space. Correlated instrumental measurements for each attribute would ultimately encourage their inclusion in future breeding efforts.

Leaf color “palettes” enable deeper observation into preferences and consumer perception of nutrition.

Appearance and leaf morphological traits were highly variable across study materials and intricately connected to consumer acceptance. Further, leaf morphological traits and color may be associated with underlying phytochemical profiles. Flat collard genotypes, for example, were higher in glucosinolates while Tuscan and curly kale salad types were higher in carotenoids (Fig. 4.6). These salad types were generally higher in dry matter and leaf curl/blister than flat collard types. Additional replication and characterization of these phenotypes is required, but these preliminary data may lead to new research endeavors. The dry matter components that form the basis of texture, for example, may also be responsible for forming the leaf curl and blistering traits that define kale characteristics popular among consumers.

The development of a platform in which perceived nutritional content could be correlated with quantitative color assessment and phytochemical profiles would make for a unique study into whether consumer preferences for certain leaf colors are at the expense of nutritional quality. Light green or yellow genotypes were among the most liked materials in the online consumer survey. While it is unclear whether the lighter lamina color is conferred by an alteration of carotenoid, chlorophyll, or other phytochemical content, the yellow parental inbred and its progeny were generally lower in carotenoid content than other materials. Additional genetic studies and

179

nutritional analyses could shed light on the underlying biochemical regulation responsible for these traits.

Advanced materials in the breeding program enabled the development of near isogenic lines that differ in a singular trait (i.e. color or shape), and these materials could provide the base for experimental material to survey perceived nutritional content in leafy greens. Literature exploring carrot color demonstrated clear connections between underlying phytochemical profiles and sensory profiles (Surles et al., 2004). Additional studies demonstrated modulating effects of product color and image background color on the freshness and perceived nutritional content of carrots

(Schifferstein et al., 2019, 2017). Color and nutritional evaluations must be designed and implemented with care, especially in regard to the growing environment.

Understanding consumer perception of leaf shape, color, phytochemical profiles, and their interaction(s) would direct phenotypic selection for visual quality in new cultivars that are preferred by consumers.

Advanced modelling and integration of “omic” data could inform the interdisciplinary connections from breeder-to-consumer.

A structured mating design presented in Chapter Four enabled the elucidation of a small, but heritable genetic component of consumer liking among leafy Brassicas.

Genotyping-by-sequencing (GBS) and subsequent population structure analyses identified genetic diversity among breeding lines not present in commercially available materials. In addition, genetic relationship matrices were used to inform mixed model analysis of diallel plant materials. Though the addition of genetic information may have resulted in slightly better estimates of genetic effects, the

180

structured mating design limited the utility of genomic relationship data in mixed modelling procedures.

Future research should focus on generating additional means of integrating next-generation genomic technology into the assessment of quality traits and consumer liking (Tieman et al., 2017). Genetic information as it pertains to consumer quality may perhaps be more relevant when used in an association panel or among a large number of diverse genotypes. Resulting data may be used in a predictive capacity to select and combine superior parental types. Additionally, genomic data may serve to inform the improvement of horticultural traits after consumer ideotypes have been established.

In what is perhaps the most ambitious approach to connecting consumers and plant scientists, researchers from Wageningen University aimed to model consumer flavor perception, citing that “reverse breeding, starting from the consumers’ perspective, can help breeders deliver novel cultivars” (Tesfaye et al., 2013). Their study, conducted within the context of a tomato breeding program in the Netherlands, sought to reduce the “information asymmetry” between breeding technology (i.e. genomics, metabolomics, etc.) and consumer information when limited number of cultivars are evaluated. Its application has not yet been fully realized, but it provides an interesting theoretical framework for advanced modelling procedures to “reverse engineer” quality cultivars.

Culture and context matter.

Chapter Three introduced a rapid sensory method known as Flash Profiling

(FP) to provide descriptive information of collard genotypes. The study was conducted

181

alongside consumer acceptance testing at six heterogeneous sites throughout Western

Kenya and Upstate New York. Regional preferences were clearly drawn from consumer acceptance testing, but descriptive work to generate a region-specific lexicon did not result in clearly defined lexicons between the two regions. Issues due to language and product familiarity led to an increased number of hedonic attributes among Kenyan panelists, making it difficult to differentiate genotypes on the basis of underlying sensory traits. Key comparisons between consumers in the Northeastern

U.S. and Southeastern U.S. were missing from this study and would have helped elucidate the impact of language on the results from this study.

These experiments were experiential and informative in the development of a well-balanced and consumer-informed breeding program. Directed selection of materials suited to the production, distribution, and consumption patterns of their intended food system should continue to be emphasized within cultivar development programs. However, transparent communication that frames breeding products within the regional and cultural contexts of their development is essential (Brouwer et al.,

2016).

As a relevant and recent example, some cassava cultivars released by research focused on improving cassava for African markets failed to meet market standards and expectations set by the end-users, the result of which was a lack of adoption and utilization (Bechoff et al., 2016). One response to this dilemma was to develop a gender-responsive breeding program focused on integrating feedback relevant to roles across the value chain (Tufan et al., 2018). The FP descriptive work performed in

Western Kenya and Upstate New York could have similarly benefitted from translated

182

surveys, partnerships with local academic institutions/organizations, and a deeper understanding of collard’s role and familiarity in regional food systems. Investment and forethought into the constraints of cross-cultural sensory work would have enhanced the reproducibility and conclusions drawn from our study.

Pragmatism and creativity in designing sensory and consumer research methods for vegetable breeding programs

Many of the methods presented throughout this dissertation, including in-situ acceptance testing, descriptive FP, and an online consumer survey were purposefully conducted to evaluate their feasibility and utility in a breeding program. These methods required adapting traditional sensory and consumer methods to fit within the context and diverse environments in which a plant breeding program operates. They were an effort to maintain the scientific rigor of traditional sensory analysis in participatory contexts that engage a wider audience.

Integration of sensory methods can be time consuming and financially straining to breeding programs. Implementation of these methods required additional thought into confounding factors (e.g. participant recruitment, sample preparation, heterogenous environments etc.) and necessitated a flexible means of collecting consumer data. They also required significant planning and travel to implement.

Continued research is needed to validate and refine rapid sensory profiling methods within the context of a plant breeding program, but the initial groundwork has been established to connect plant breeders with these potentially powerful techniques. The information provided from these studies was pertinent and useful in making breeding

183

decisions, and in each study a sufficient number of participants were recruited to power statistical analyses. These methods are feasible within the context of a vegetable breeding program, but their utility depends on the breeder’s motivation to engage and integrate non-traditional stakeholders in breeding quality traits.

Though it seems readily apparent, researchers should carefully consider the underlying theory and design of a sensory experiment in non-traditional settings.

Sensory methodology outside of traditional facilities, particularly when evaluating inherently variable products such as fruit and vegetables, should consider additional replicates, the appropriateness of employed survey language, and panelist variability

(Stone, 2015). Sensory analysis conducted outside of controlled environments should ensure direct and purposeful methodology, versus ad-hoc, non-replicable interactions with end-users, but these constraints should not hinder the creativity with which researchers attempt to connect with consumers and markets.

Several models have emerged in recent years championing diverse means of connecting plant breeding programs with end-users. The Seed-to-Kitchen

Collaborative (SKC), for example, works in conjunction with several local chefs, now considered a trained expert panel, to obtain culinary and quality feedback on research and breeding materials (Healy and Dawson, 2019). Events such as SKC’s Farm-to

Flavor dinner, the Variety Showcase organized by the Culinary Breeding Network

(CBN) at Oregon State University, or the Grain Gathering hosted by Washington

State’s Bread Lab are also opportunities for plant breeders to potentially collect sensory data. These models should be better integrated and used as inspiration to continue exploring alternative means of connecting consumers and plant breeders.

184

There will be continued need to understand the economic strain and long-term impacts of decentralized, interdisciplinary programs. Researchers at the University of

Wisconsin – Madison have extended the participatory plant breeding model in a study designed to compare “single-farm” and “broad outreach” decentralized plant breeding models (Hanson and Goldman, 2017). One of the study’s main objectives is to assess the feasibility and economic efficiency of each model within the context of an applied plant breeding program. While the study has not yet concluded, initial results demonstrate the economic viability of the “outreach” model and its suitability in providing plant breeders with consumer preference feedback to guide the quality trait selection within a population. Plant breeding programs are highly heterogenous in their operations and objectives, so no single model will suffice to integrate sensory or consumer research across programs. However, embedded within decentralized models is the capacity to educate consumers and potentially re-invigorate public plant breeding programs dedicated to cultivar development (Gepts and Hancock, 2006).

Balancing stakeholder expectations in breeding for quality

Plant breeding does not exist in isolation; the utility and potential of plant products relies on feedback from diverse stakeholders and disciplines. Catering solely to consumer demand and preferences may not only be disadvantageous to food systems and plant productivity, but it may also limit the capacity of a breeding program to identify novel and potentially disruptive innovations. Several studies have alluded to an “ideal” or “perfect” fruit type that might emerge from sensory analysis or consumer research data (Colquhoun et al., 2012; Kappel et al., 1996; Olmstead et al.,

185

2015). This trend highlights the need for greater training of plant breeders to recognize both the possibilities and limitations of sensory and consumer research methods. Often the benefit of these methods lies in understanding not just want consumers expect and prefer in the current market, but in the spaces between consumer clusters and existing market standards. Continued use of sensory methods in plant breeding necessitates partnerships and collaboration with disciples of marketing, agricultural economics, human ecology, and others to holistically understand these data.

Beyond the sensory and consumer space, partnerships across the supply chain must be established and fostered. Advancements in supply chain management and digital logistics have enabled management of higher quality, more perishable products in supply chains while promoting the expansion of regional production of high value products. Stakeholders across the supply chain have unique product expectations and plant breeders have an inherent responsibility to stakeholders across food systems, not just consumers. Seed companies, processing facilities, growers, retailers, extension agents, restaurants, and other vertically integrated players can inform developing and new markets for plant breeders. Many plant breeders could better recognize these players as resources to identify opportunities in breeding work.

It remains to be seen whether breeding for consumer quality traits in leafy

Brassicas, such as texture or leaf morphology, negatively affects other aspects of supply chain quality. Our breeding program, for example, has been approached by fast-fresh restaurants to breed for softer, thinner stems to ease in-house processing. In doing so, however, the plant habit, shelf-life, and desirable eating qualities may be negatively impacted. The program has also been asked to breed less bitter leaf types

186

for fresh-pressed juice outlets, but the impacts on insect or herbivory interactions could significantly affect grower’s capacity to produce a viable crop. Shifting market expectations and improving quality traits in leafy Brassicas is possible but will take time and conscious effort to minimize unintended consequences for other stakeholders.

The research presented herein was developed as a means of highlighting the interdisciplinary nature of plant breeding and encourage continuous process of utilizing downstream feedback within the field of plant breeding. The intention was to provide examples of integrated plant breeding endeavors that better connect plant breeders with consumers and, ultimately, enhance the process of cultivar development.

These are just some of the first steps in understanding the capacity of our fresh market vegetable breeding program to connect with diverse stakeholders and breed for quality traits.

187

References

Bechoff, A., Tomlins, K., Fliedel, G., Lopez-lavalle, L.A.B., Westby, A., Hershey, C., Dufour, D., 2016. Cassava traits and end-user preference: Relating traits to consumer liking, sensory perception, and genetics. Critical Reviews in Food Science and Nutrition, 1–21. https://doi.org/10.1080/10408398.2016.1202888 Brouwer, B.O., Murphy, K.M., Jones, S.S., 2016. Plant breeding for local food systems: A contextual review of end-use selection for small grains and dry beans in Western Washington. Renewable Agriculture and Food Systems 31, 172–184. https://doi.org/10.1017/S1742170515000198 Colquhoun, T.A., Levin, L.A., Moskowitz, H.R., Whitaker, V.M., Clark, D.G., Folta, K.M., 2012. Framing the perfect strawberry: An exercise in consumer-assisted selection of fruit crops. Journal of Berry Research 2, 45–61. https://doi.org/10.3233/JBR-2011-027 Duffy, E.M., Hayes, J.E., Feeney, E.L., 2017. Understanding taste and texture perception to enhance vegetable acceptance. Proceedings of the Nutrition Society 76. https://doi.org/10.1017/S0029665117001409 Gepts, P., Hancock, J., 2006. The Future of Plant Breeding. Crop Science 46, 1630– 1634. https://doi.org/10.2135/cropsci2005-12-0497op Hanson, S., Goldman, I., 2017. Consumer-Engaged Participatory Plant Breeding Model Comparison and Beet Flavor Breeding · Ceres Trust. Healy, G.K., Dawson, J.C., 2019. Participatory plant breeding and social change in the Midwestern United States: perspectives from the Seed to Kitchen Collaborative. Agric Hum Values 36, 879–889. https://doi.org/10.1007/s10460-019-09973-8 Kappel, F., Fisher-Fleming, B., Hogue, E., 1996. Fruit Characteristics and Sensory Attributes of an Ideal Sweet Cherry. HortScience 31, 443–446. Kilcast, D., Fillion, L., 2001. Understanding consumer requirements for fruit and vegetable texture. Nutrition & Food Science 31, 221–225. https://doi.org/10.1108/00346650110396574 Klee, H.J., Tieman, D.M., 2018. The genetics of fruit flavour preferences. Nature Reviews Genetics 19, 347–356. https://doi.org/10.1038/s41576-018-0002-5 Olmstead, M.A., Gilbert, J.L., Colquhoun, T.A., Clark, D.G., Kluson, R., Moskowitz, H.R., 2015. In Pursuit of the Perfect Peach: Consumer-assisted Selection of Peach Fruit Traits. HortScience 50, 1202–1212. Pierre-Louis, K., 2017. Texture is the final frontier of food science. Popular Science. https://www.popsci.com/texture-food- science/#:~:text=Compared%20to%20our%20rich%20vocabulary,for%20food %20tactility%20is%20thin. Schifferstein, H.N.J., Howell, B.F., Pont, S.C., 2017. Colored backgrounds affect the attractiveness of fresh produce, but not it’s perceived color. Food Quality and Preference 56, 173–180. https://doi.org/10.1016/j.foodqual.2016.10.011 Schifferstein, H.N.J., Wehrle, T., Carbon, C.-C., 2019. Consumer expectations for vegetables with typical and atypical colors: The case of carrots. Food Quality and Preference 72, 98–108. https://doi.org/10.1016/j.foodqual.2018.10.002

188

Stone, H., 2015. 2 - Alternative methods of sensory testing: advantages and disadvantages, in: Rapid Sensory Profiling Techniques, Woodhead Publishing Series in Food Science, Technology and Nutrition. Woodhead Publishing, pp. 27–51. https://doi.org/10.1533/9781782422587.1.27 Surles, R.L., Weng, N., Simon, P.W., Tanumihardjo, S.A., 2004. Carotenoid Profiles and Consumer Sensory Evaluation of Specialty Carrots (Daucus carota, L.) of Various Colors. J. Agric. Food Chem. 52, 3417–3421. https://doi.org/10.1021/jf035472m Tesfaye, L.M., Bink, M.C.A.M., Lans, I.A. van der, Gremmen, B., Trijp, H.C.M. van, 2013. Bringing the voice of consumers into plant breeding with Bayesian modelling. Euphytica 189, 365–378. https://doi.org/10.1007/s10681-012-0765- 2 Tieman, D., Zhu, G., Resende, M.F.R., Lin, T., Nguyen, C., Bies, D., Rambla, J.L., Beltran, K.S.O., Taylor, M., Zhang, B., Ikeda, H., Liu, Z., Fisher, J., Zemach, I., Monforte, A., Zamir, D., Granell, A., Kirst, M., Huang, S., Klee, H., 2017. A chemical genetic roadmap to improved tomato flavor. Science 355, 391– 394. https://doi.org/10.1126/science.aal1556 Tufan, H.A., Grando, S., Meola, C., 2018. State of the Knowledge for Gender in Breeding: Case Studies for Practitioners (Working Paper). International Potato Center (CIP), Lima, Peru.

189

APPENDIX: Chapter Two

Appendix 2.1. Timeline and evaluation agenda of Qualitative Multivariate Analysis (QMA), descriptive analysis, and consumer hedonic central-location test (CLT) sensory studies. Dates are presented in mm/dd/yy format. Qualitative Multivariate Analysis (QMA) - 2016 Date Topic Preparation 8.26.16 Send out recruitment email Recruitment poster 9.2.16 Select and notify participants for sensory acuity test Recruit 30 participants, draft e-mail 9.2.16 Warm-up activities Send to participants, collect by 9.8.17 9.8.16 Sample pick-up (Day 1) harvest samples, wash/bag samples 9.15.16 Sample pick-up (Day 2) harvest samples, wash/bag samples 9.24.16 Moderated focus group prepare discussion outline, set-up notes Quantitative Descriptive Analysis (QDA) - 2017 Date Topic Preparation 7.5.17 Send out recruitment email Recruitment poster 7.14.17 Select and notify participants for sensory acuity test Recruit 30 participants, draft email 7.17.17 Sensory acuity test Multiple reference solutions, triangle test 7.20.17 Select and notify participants for training sessions Recruit 15 participants, draft email 7.24.17 Training 1: Project introduction; generate broad leafy green terminology Paper pad, variety of leafy greens 7.26.17 Training 2: Refine terminology specific to leafy Brassicas; discuss aroma- Paper pad, raw and cooked leafy Brassicas related attributes 7.31.17 Training 3: Review leafy Brassicas; introduce basic taste solutions; generate Paper pad, raw and cooked kale, references potential list of attribute reference standards 8.2.17 Training 4: Discuss flavor-related attributes; provide potential reference Paper pad, raw and cooked kale, references standards; compile deck for future reference 8.7.17 Training 5: Discuss texture-related attributes; integrate external lexicons Printed lexicon, raw and cooked kale, from literature; re-visit basic tastes references 8.9.17 Training 6: Review flavor and texture attributes with samples and potential Paper pad, raw and cooked kale, references reference standards 8.14.17 Training 7: Discuss aftertaste attributes; finalize reference standards & Scale sheet, lexicon list, reference standards attribute definitions/anchors 8.16.17 Training 8: Paper test with finalized lexicons, discuss scale & sequence Multiple scale sheets, lexicon list 8.21.17 Training 9: Revisit aroma and nuanced flavor attributes; raw kale practice Sensory booth, raw kale samples session; self-reported evaluations 8.23.17 Training 10: Revisit texture and aftertaste attributes; raw kale practice Sensory booth, raw kale samples session; self-reported evaluations 8.28.17 Training 11: Discuss appearance-related attributes; raw kale practice Full-leaf samples, sensory booth, raw kale session; self-reported evaluations samples 8.30.17 Training 12: Practice appearance-related attributes SD of panelists/panel, develop feedback, calibrate 9.6.17 Training 13: Review results; improve lexicon agreement Sensory booth, appearance testing, raw kale 9.11.17 Training 14: Revisit texture-related attributes; cooked kale practice session; Sensory booth, cooked (blanched) kale self-reported evaluations 9.13.17 Training 15: Review results; improve lexicon agreement SD of panelists/panel, develop feedback, calibrate 9.15.17 Training 16: Triangle test of basic solutions, raw kale practice session Sensory booth, kale samples, show panelists results 9.18.17 Training 17: Review results, discuss upcoming evaluations 9.20.17 Evaluation Session 1a: Hybrid breeding line evaluations (6 samples) harvest plant material, wash, cut, and pack 9.22.17 Evaluation Session 1b: Hybrid breeding line evaluations (6 samples) harvest plant material, wash, cut, and pack s 9.29.17 Evaluation Session 2a: Commercial variety evaluations (6 samples) harvest plant material, wash, cut, and pack 10.2.17 Evaluation Session 2b: Commercial variety evaluations (6 samples) harvest plant material, wash, cut, and pack 10.4.17 Evaluation Session 3a: Green curly market class evaluations (7 samples) harvest plant material, wash, cut, and pack 10.6.17 Evaluation Session 3b: Green curly market class evaluations (7 samples) harvest plant material, wash, cut, and pack s Consumer Hedonic Evaluations (CLT) - 2018 Date Topic Preparation 9.21.18 Send out recruitment email Recruitment poster 9.24.18 Notify participants of scheduled date/time for evaluation Draft email 9.27.18 Consumer evaluations (Day 1) harvest plant material, wash, cut, and pack 9.28.18 Consumer evaluations (Day 2) harvest plant material, wash, cut, and pack

I

Appendix 2.2. Consumer demographic information for n = 90 panelists included in the September 2018 consumer hedonic central-location test (CLT) study. Participants were recruited through Cornell Sensory Evaluation Center’s list-serv. Reported participant ethnicities, according to gender. Black or White and White Hispanic or Gender White Asian African Hispanic/La and N/A TOTAL Latino American tino Asian Male 14 7 1 1 1 0 0 24 (27%) Female 32 19 3 3 5 1 0 63 (70%) Other 1 0 0 0 0 0 2 3 (3%) TOTAL 47 (52%) 26 (29%) 4 (4%) 4 (4%) 6 (7%) 1 (1%) 2 (2%) 90

Participant age, according to gender. Age Male Female Other TOTAL 18-24 years 8 36 0 44 25-34 years 12 14 0 26 35-44 years 1 5 1 7 45-54 years 0 1 0 1 55-64 years 3 5 0 8 65-74 years 0 2 0 2 75+ years 0 0 0 0 N/A 0 0 2 2 TOTAL 24 63 3 90

Average consumption of kale products among participants, according to gender. Consumption Rate Male Female Other TOTAL At least once per day 0 1 0 1 4-6 times per week 1 9 1 10 2-3 times per week 7 15 1 22 Once per week 9 16 1 25 Few times per month 7 22 0 29 I do not eat kale 0 1 0 1 TOTAL 24 64 3 88

Self-described participant behavior regarding willingness to try new leafy greens and varieties Variety Seeking Behavior Male Female Other TOTAL Very rarely do I try new leafy greens 0 9 0 9 I sometimes like to try new leafy greens if someone recommends it or it’s in a recipe 5 21 0 26 I sometimes like to try new leafy greens without prompting 16 26 1 43 I seize every chance I get to try a new leafy green 2 7 0 9 Other 1 0 0 1

Preferred preparation methods among participants (check-all-that apply) Gender Raw Sautee Juiced Boiled Dried Fried Other* Female 43 48 16 17 15 5 1 Male 18 14 6 3 9 2 1 Other 1 3 0 1 1 0 1 TOTAL 62 65 22 21 25 7 3 *Other preparations included: blended in pesto (2) and slow cooked in "beans and greens" (1)

Sourcing and purchasing locations of kale among participants (check-all-that apply) Gender Supermarket Green Grocer CSA Farmers' Mkt Home Grown Other* Female 59 9 6 21 7 6 Male 21 7 6 14 2 0 Other 3 1 1 1 0 0 TOTAL 83 17 13 36 9 6 *Other sources included: dining hall (3), restaurants (1), HelloFresh® (1), and donation (1)

II

Appendix 2.3. Nested mixed model output for 44 sensory attributes evaluated by n=9 trained panelists during Quantitative Descriptive Analysis (QDA®) of kale samples. Output represents fixed-factor ANOVA (Satterthwaite's Method) and random effect 2 2 2 2 variance components attributed to nested rep [𝜎r(s|푝) ], nested session [𝜎(s|푝) ], panelist variance [𝜎푝 ], and residual variance [𝜎 ]. 2 Percentage of total random variance attributed to panelist (% 𝜎푝 ) provides an indication of total variation due to panelist effect. AROMA ATTRIBUTES ANOVA (Satterthwaite's) Variance Components 2 2 2 2 2 SS MS NumDF DenDF F-value Pr(>F) 𝜎r s 푝 𝜎 s 푝 𝜎푝  % 𝜎푝 ( | ) ( | ) 𝜎 Intensity 7303 522 14 248.4 2.91 4.20x10-4 *** 90.5 0.0 108.2 179.3 28.63%

Green 4241 303 14 226.9 1.12 0.340 ns 58.3 56.0 46.5 270.8 10.77%

Sulfurous 1679 120 14 253.7 1.27 0.227 ns 71.0 0.0 112.0 94.5 40.34%

Musty 1085 77 14 241.8 0.35 0.986 ns 54.6 84.2 142.4 222.2 28.29%

Pungent 858 61 14 248.0 1.03 0.427 ns 27.4 0.0 20.8 59.6 19.26%

TASTE/FLAVOR ATTRIBTUES ANOVA (Satterthwaite's) Variance Components 2 2 2 2 2 SS MS NumDF DenDF F-value Pr(>F) 𝜎r s 푝 𝜎 s 푝 𝜎푝  % 𝜎푝 ( | ) ( | ) 𝜎 Sour 4804 343 14 206.8 2.08 1.42x10-2 * 33.5 11.5 129.8 165.1 38.18%

Bitter 8509 608 14 204.7 2.79 7.92 x10-4 *** 57.7 14.1 166.6 217.7 36.53%

Umami 1194 85 14 243.2 1.02 0.430 ns 46.1 21.8 78.6 83.3 34.21%

Sweet 4290 306 14 232.4 1.85 3.22x10-2 * 34.2 0.0 39.9 165.2 16.67%

Green 2460 176 14 224.7 0.88 0.584 ns 41.7 35.7 300.6 200.1 52.00%

Green Apple 2381 170 14 235.9 1.07 0.386 ns 41.7 35.7 300.6 200.1 52.00%

Buttery 667 48 14 242.9 1.19 0.286 ns 7.9 17.6 30.7 40.1 31.85%

Earthy 1130 81 14 235.8 0.56 0.894 ns 40.2 0.0 460.9 144.0 71.45%

Fermented 866 62 14 254.4 0.76 0.714 ns 19.5 85.9 83.5 81.6 30.89%

Sulfurous 1778 127 14 215.4 0.97 0.487 ns 33.5 11.9 450.0 131.1 71.83%

Musty 911 65 14 245.4 0.88 0.577 ns 60.0 11.7 251.4 73.6 63.37%

Moldy 766 55 14 198.8 0.69 0.781 ns 13.6 5.0 76.7 79.1 43.97%

Pungent 6209 444 14 214.7 1.73 5.22x10-2 ns 79.0 15.1 373.5 257.0 51.55%

AFTERTASTE ATTRIBUTES ANOVA (Satterthwaite's) Variance Components 2 2 2 2 2 SS MS NumDF DenDF F-value Pr(>F) 𝜎r s 푝 𝜎 s 푝 𝜎푝  % 𝜎푝 ( | ) ( | ) 𝜎 Bitter 6151 439 14 232.4 2.63 1.47x10-3 ** 82.6 19.8 186.2 167.3 40.85% Sour 1372 98 14 212.8 1.05 0.410 ns 33.1 2.9 100.9 93.7 43.76% Umami 886 63 14 243.5 1.32 0.195 ns 40.8 5.9 52.4 47.9 35.62% Astringent 2985 213 14 250.9 1.50 0.113 ns 93.1 0.0 258.2 142.6 52.27% Metallic 354 25 14 257.3 1.01 0.442 ns 29.5 0.0 36.8 25.0 40.32%

III

TEXTURE ATTRIBUTES ANOVA (Satterthwaite's) Variance Components 2 2 2 2 2 SS MS NumDF DenDF F-value Pr(>F) 𝜎r s 푝 𝜎 s 푝 𝜎푝  % 𝜎푝 ( | ) ( | ) 𝜎 Hardness 9687 692 14 237.9 5.49 4.02x10-9*** 31.6 41.1 192.5 126.0 49.21%

Thickness 10762 769 14 219.3 5.87 9.88x10-10 *** 13.6 27.1 250.8 131.0 59.35%

Density 7761 554 14 204.5 2.93 4.46x10-4*** 25.2 21.7 221.4 189.3 48.38%

Moisture 1442 103 14 212.3 0.90 0.564 ns 4.9 25.2 30.5 114.9 17.38%

Cohesive 4927 352 14 226.0 1.56 9.12x10-2 ns 35.7 0.0 223.5 225.2 46.14%

Crunch 6774 484 14 223.3 3.12 1.82x10-4 *** 27.6 29.1 163.6 155.1 43.58%

Brittle 10223 730 14 234.2 4.20 1.40x10-6 *** 26.1 51.3 390.7 173.8 60.87%

Chewy 7146 510 14 208.8 3.46 4.40x10-5 *** 31.8 12.3 293.3 147.5 60.49%

Adhesive 5370 384 14 233.3 2.04 1.57x10-2 * 46.2 58.4 332.3 187.6 53.21%

Smooth 13671 976 14 205.7 5.47 7.02x10-9 *** 0.0 25.5 155.0 178.7 43.16%

Waxy 23829 1702 14 234.4 7.90 1.13x10-13 *** 53.2 0.0 305.8 215.6 53.22%

Chalky 998 71 14 252.5 1.30 0.205 ns 40.4 0.0 128.4 54.7 57.46%

Fibrous 3898 278 14 219.8 1.61 7.79x10-2 ns 11.8 38.3 244.5 173.0 52.29%

Oily Coating 2571 184 14 234.5 1.63 7.26x10-2 ns 18.0 34.8 103.8 112.8 38.53%

Slimy 4114 294 14 228.4 2.71 1.04x10-3 ** 15.4 27.2 59.5 108.5 28.26%

Chew Count 4114 294 14 228.4 2.71 1.04x10-3 ** 2.8 5.2 19.3 11.7 49.45%

VISUAL ATTRIBUTES ANOVA (Satterthwaite's) Variance Components 2 2 2 2 2 S.S. M.S. NumDF DenDF F-value Pr(>F) 𝜎r s 푝 𝜎 s 푝 𝜎푝  % 𝜎푝 ( | ) ( | ) 𝜎 Thickness 12500 893 14 240.4 6.18 1.75x10-10 *** 49.86 46.09 109.2 144.41 31.24% Waxy 70264 5019 14 174.2 20.10 2.20x10-16 *** 0.00 15.99 164.50 249.70 38.24% Vein Presence 8491 607 14 213.3 2.67 1.26 x10-3 ** 26.96 35.49 125.18 226.8 30.21% Smoothness 30460 2176 14 197.8 10.53 2.20x10-16 *** 6.338 30.099 84.307 206.688 25.75% Turgidity 10155 725 14 257.4 3.03 2.45x10-4 *** 330.80 0.00 12.49 15.48 3.48%

Significance codes: 0.001 = '***', 0.01 = '**', 0.05 = '*', not significant = ‘ns’

IV

Appendix 2.4. Nested mixed model output for hedonic scores (1-9 scale), attribute intensity (0-100 scale) ratings, and product scores (1-5 scale) as evaluated by a consumer panel (n=90) in a central location test (CLT). Output represents fixed-factor 2 ANOVA (Satterthwaite's Method) and random effect variance components attributed to nested session [𝜎(s|푝) ], panelist variance 2 2 2 [𝜎푝 ], and residual variance [𝜎 ]. Percentage of total random variance attributed to panelist (% 𝜎푝 ) provides an indication of total variation due to panelist effect.

HEDONIC SCORES (1-9 scale) ANOVA (Satterthwaite's) Variance Components 2 2 2 2 SS MS NumDF DenDF F-value Pr(>F) 𝜎(s|푝) 𝜎푝 𝜎 % 𝜎푝 Overall Liking 80.6 9.0 9 791.99 4.1602 2.92x10-5*** 0 0.626 2.153 22.53% Appearance 281.7 31.3 9 786.37 17.273 2.20x10-16*** 0.0599 0.44178 1.81186 19.10% Color 245.3 27.3 9 791.74 13.923 2.20x10-16*** 0 0.496 1.958 20.21% Aroma 17.8 2.0 9 779.56 1.571 0.1197ns 0.1071 0.5643 1.2605 29.21% Flavor 127.3 14.1 9 792.47 5.1411 8.60x10-7*** 0 0.5792 2.7516 17.39% Texture 39.6 4.4 9 787.2 2.083 0.02873* 0.0532 0.59421 2.1118 21.54%

ATTRIBUTE INTENSITY (0-100 scale) ANOVA (Satterthwaite's) Variance Components 2 2 2 2 SS MS NumDF DenDF F-value Pr(>F) 𝜎(s|푝) 𝜎푝 𝜎 % 𝜎푝

Overall Intensity (Aroma) 11915 1324 9 791.85 3.6892 1.53x10-4*** 0 190.7 358.9 34.70% Kale Intensity (Flavor) 15739 1749 9 774.2 5.1492 8.43x10-7*** 25.63 82.05 339.63 18.34%

Sweet (Flavor) 10020 1113 9 772.73 3.3505 4.91x10-4*** 28.95 186.9 332.29 34.10%

Bitter (FL) 37948 4216 9 773.85 9.0081 5.19x10-13*** 35.59 145.65 468.07 22.43%

Sour (FL) 5828 648 9 774.09 2.7017 4.23x10-3** 16.73 250.92 239.68 49.46%

Leaf Thickness (TX) 72387 8043 9 779.39 31.918 2.20x10-16*** 8.535 85.011 251.985 24.60%

Crisp Crunch (TX) 26350 2928 9 772.72 7.9966 2.26x10-11*** 32.56 78.08 366.13 16.38%

Chewy (TX) 8273 919 9 768.22 3.5797 2.24x10-4*** 36.8 163.8 256.8 35.81%

Slimy (TX) 2833 315 9 758.13 2.1445 2.40x10-2* 39.14 238.49 146.77 56.19%

Bitter (AT) 61052 6784 9 774.96 11.015 2.22x10-16*** 41.32 139.46 615.86 17.51%

PRODUCT SCORES (1-5 scale) ANOVA (Satterthwaite's) Variance Components 2 2 2 2 SS MS NumDF DenDF F-value Pr(>F) 𝜎(s|푝) 𝜎푝 𝜎 % 𝜎푝

Concept Fit 119.4 13.3 9 788.51 14.904 2.20E-16*** 0.0043 0.173072 0.889768 16.22%

Purchase Intent 53.4 5.9 9 792.32 5.1857 7.31E-07*** 0.00 0.18 1.14 13.89% ns Significance codes: 0.001 = '***', 0.01 = '**', 0.05 = '*', not significant = ‘ ’

V

Appendix 2.5. Ranked mean separation of 21 attributes that differed significantly (P ≤ 0.05) among kale genotypes in descriptive analysis. Tukey's HSD was performed on nested mixed model estimate means; means with the same grouping letter are not significantly different (P ≤ 0.05).

Intensity (Aroma) Sour (Flavor) Sweet (Flavor) Sample Mean LowerCL UpperCL Group Sample Mean LowerCL UpperCL Group Sample Mean LowerCL UpperCL Group

Redbor 42.9 32.4 53.5 a Redbor 31.9 21.2 42.7 a Hybrid E 23.7 15.7 31.7 a Meadowlark 42.3 31.6 53.0 a Yellow 26.6 15.8 37.3 ab Reflex 18.4 10.2 26.5 a White Russian 41.5 30.9 52.0 a White Russian 25.2 14.5 36.0 ab Tiger 17.8 9.8 25.8 a Yellow 41.1 30.5 51.7 a Meadowlark 25.0 14.1 35.8 ab White Russian 17.5 9.5 25.5 a Laerketunge 35.8 25.1 46.5 ab Starbor 24.1 13.2 34.9 ab Krul 17.4 9.2 25.5 a Hybrid E 35.3 24.7 45.9 ab Black Magic 22.7 13.1 32.2 ab Laerketunge 16.6 8.4 24.8 a Tiger 33.4 22.8 44.0 ab Krul 22.4 11.5 33.2 ab Darkibor 15.7 9.6 21.8 a Hybrid L 33.3 22.7 43.8 ab Laerketunge 21.1 10.3 32.0 ab Hybrid I 14.3 6.3 22.3 a Hybrid I 32.7 22.2 43.3 ab Hybrid K1 21.0 10.3 31.8 ab Hybrid K1 13.0 5.0 21.0 a Krul 32.2 21.5 42.9 ab Hybrid I 19.8 9.1 30.6 ab Hybrid L 13.0 5.0 21.0 a Black Magic 32.0 22.8 41.1 ab Hybrid L 18.7 7.9 29.4 ab Yellow 12.6 4.6 20.6 a Hybrid K1 31.0 20.4 41.5 ab Hybrid E 17.8 7.0 28.5 ab Starbor 11.9 3.7 20.1 a Reflex 28.5 17.8 39.2 ab Tiger 16.7 5.9 27.4 b Black Magic 11.7 5.6 17.8 a Darkibor 27.3 18.2 36.5 b Darkibor 16.6 7.1 26.2 b Meadowlark 8.0 -0.2 16.2 a Starbor 26.5 15.8 37.2 ab Reflex 16.3 5.5 27.2 ab Redbor 8.0 -0.1 16.0 a

Bitter (Flavor) Bitter (Aftertaste) Leaf Thickness (Texture) Sample Mean LowerCL UpperCL Group Sample Mean LowerCL UpperCL Group Sample Mean LowerCL UpperCL Group

Meadowlark 33.6 21.2 46.0 a Meadowlark 28.4 15.9 41.0 a Hybrid K1 44.9 31.5 58.2 a Redbor 29.2 16.9 41.5 ab Black Magic 27.4 15.9 38.9 a Hybrid I 37.2 23.8 50.5 abc Krul 26.7 14.3 39.2 ab White Russian 23.5 11.1 36.0 ab Black Magic 34.9 22.2 47.6 ab Black Magic 26.5 15.6 37.5 a Starbor 23.1 10.5 35.7 ab Tiger 33.5 20.1 46.8 abcd Laerketunge 26.0 13.6 38.4 ab Yellow 23.0 10.5 35.5 ab Yellow 31.6 18.3 45.0 abcd Hybrid I 25.6 13.3 37.9 ab Hybrid E 22.4 10.0 34.9 ab White Russian 29.7 16.3 43.0 abcd Starbor 25.6 13.2 38.0 ab Krul 22.3 9.7 34.8 ab Hybrid E 28.1 14.7 41.4 bcd Yellow 24.7 12.4 37.0 ab Redbor 22.0 9.5 34.4 ab Reflex 26.9 13.5 40.3 bcd Hybrid E 20.9 8.6 33.2 ab Reflex 18.8 6.2 31.4 ab Redbor 26.3 13.0 39.7 bcd Hybrid L 20.8 8.5 33.1 ab Laerketunge 18.8 6.2 31.4 ab Darkibor 25.5 12.7 38.2 cd White Russian 20.0 7.7 32.3 ab Tiger 17.7 5.2 30.1 ab Laerketunge 23.8 10.4 37.2 bcd Reflex 17.3 4.9 29.7 ab Hybrid K1 17.6 5.1 30.0 ab Starbor 23.7 10.3 37.1 bcd Hybrid K1 17.2 4.9 29.5 ab Hybrid I 16.6 4.1 29.1 ab Krul 22.5 9.1 35.9 cd Tiger 15.7 3.4 28.0 ab Hybrid L 16.2 3.7 28.7 ab Meadowlark 20.6 7.2 34.0 d Darkibor 15.2 4.3 26.1 b Darkibor 14.7 3.2 26.1 b Hybrid L 19.8 6.5 33.2 d

Hard (Texture) Dense (Texture) Crisp Crunch (Texture) Sample Mean LowerCL UpperCL Group Sample Mean LowerCL UpperCL Group Sample Mean LowerCL UpperCL Group

Hybrid I 44.5 32.3 56.7 a Hybrid K1 43.2 30.0 56.3 a Hybrid E 42.1 30.4 53.7 a Hybrid K1 41.7 29.5 53.9 ab Hybrid I 40.8 27.6 53.9 a Redbor 40.3 28.7 52.0 a Hybrid E 34.6 22.3 46.8 abcd Hybrid E 38.0 24.8 51.2 a Meadowlark 37.0 25.3 48.8 ab

VI

Tiger 33.6 21.4 45.8 abcd Black Magic 37.8 25.7 50.0 a Tiger 35.6 23.9 47.2 ab Black Magic 32.8 21.4 44.3 bc Starbor 37.5 24.2 50.7 a Yellow 35.4 23.8 47.1 ab Starbor 30.5 18.3 42.8 abcde White Russian 36.0 22.8 49.2 a White Russian 33.6 21.9 45.2 ab Redbor 29.6 17.3 41.8 abcde Tiger 35.6 22.5 48.8 a Black Magic 33.6 22.9 44.2 a White Russian 27.8 15.6 40.0 bcde Redbor 35.3 22.2 48.5 a Krul 32.4 20.7 44.2 ab Laerketunge 26.7 14.4 39.0 bcde Darkibor 35.0 22.9 47.2 a Starbor 32.3 20.6 44.0 ab Hybrid L 26.6 14.3 38.8 cde Hybrid L 34.2 21.1 47.4 ab Hybrid I 32.0 20.4 43.7 ab Darkibor 26.5 15.0 38.0 cde Reflex 34.0 20.8 47.3 ab Laerketunge 31.4 19.7 43.2 ab Reflex 25.1 12.8 37.4 cde Laerketunge 31.1 17.8 44.3 ab Reflex 28.3 16.6 40.0 ab Krul 23.1 10.8 35.4 cde Krul 30.2 17.0 43.5 ab Hybrid K1 28.1 16.4 39.7 ab Meadowlark 20.8 8.6 33.1 de Meadowlark 26.7 13.4 39.9 ab Hybrid L 27.9 16.2 39.5 ab Yellow 18.1 5.9 30.3 e Yellow 18.1 4.9 31.2 b Darkibor 25.0 14.3 35.7 b

Brittle (Texture) Waxy (Texture) Smooth (Texture) Sample Mean LowerCL UpperCL Group Sample Mean LowerCL UpperCL Group Sample Mean LowerCL UpperCL Group

Yellow 34.7 18.2 51.3 a Hybrid I 49.1 34.1 64.1 a Yellow 42.5 31.1 54.0 a Tiger 28.8 12.2 45.4 abc Hybrid K1 47.9 32.9 62.9 a Hybrid E 42.1 30.7 53.6 a Black Magic 28.5 12.6 44.4 ab Black Magic 47.4 33.3 61.4 a Tiger 38.2 26.8 49.6 ab Meadowlark 28.0 11.4 44.6 abc Starbor 38.7 23.6 53.7 ab White Russian 35.8 24.3 47.2 abc Laerketunge 26.1 9.4 42.7 abc Tiger 37.7 22.7 52.6 ab Hybrid I 30.4 18.9 41.8 abc Hybrid E 25.3 8.7 41.9 abc Hybrid E 37.6 22.6 52.6 ab Hybrid K1 30.1 18.7 41.6 abc Hybrid K1 25.2 8.7 41.8 abc White Russian 34.9 20.0 49.9 ab Laerketunge 25.9 14.4 37.5 abc White Russian 24.7 8.1 41.2 abc Yellow 34.1 19.1 49.0 ab Reflex 25.7 14.2 37.3 abc Hybrid I 24.6 8.0 41.2 abc Krul 28.5 13.5 43.6 b Black Magic 24.6 14.3 34.9 c Starbor 21.9 5.3 38.6 abc Darkibor 28.2 14.1 42.2 b Meadowlark 23.2 11.7 34.8 bc Hybrid L 18.4 1.8 34.9 abc Reflex 27.5 12.5 42.6 b Krul 23.2 11.7 34.8 bc Krul 17.8 1.2 34.4 abc Hybrid L 26.5 11.5 41.4 b Darkibor 22.9 12.6 33.2 c Redbor 16.2 -0.4 32.8 bc Laerketunge 26.4 11.3 41.4 b Redbor 22.0 10.5 33.4 c Darkibor 15.5 -0.4 31.4 c Redbor 23.2 8.3 38.2 b Starbor 20.9 9.3 32.4 bc Reflex 15.4 -1.3 32.0 bc Meadowlark 21.0 5.9 36.1 b Hybrid L 19.0 7.6 30.5 c

Adhesive (Texture) Slimy (Texture) Chewy (Texture) Sample Mean LowerCL UpperCL Group Sample Mean LowerCL UpperCL Group Sample Mean LowerCL UpperCL Group

Starbor 53.9 38.1 69.7 a Tiger 17.0 8.9 25.1 a Hybrid K1 42.5 28.2 56.8 a Reflex 50.9 35.1 66.7 ab Hybrid E 12.5 4.4 20.6 ab Redbor 39.9 25.6 54.2 ab Krul 49.9 34.1 65.7 ab Yellow 11.6 3.5 19.7 ab Darkibor 38.0 24.3 51.6 ab Hybrid L 49.6 33.9 65.4 ab White Russian 8.8 0.6 16.9 ab Hybrid L 36.0 21.7 50.3 abc Laerketunge 47.4 31.6 63.2 ab Hybrid K1 7.9 -0.2 16.0 ab Starbor 35.2 20.9 49.6 abc Hybrid I 47.4 31.7 63.1 ab Meadowlark 5.6 -2.6 13.8 ab Laerketunge 34.9 20.5 49.2 abc Hybrid K1 46.7 30.9 62.4 ab Hybrid I 4.6 -3.5 12.7 ab Krul 34.6 20.2 49.0 abc Darkibor 46.3 31.4 61.2 ab Black Magic 3.9 -3.0 10.9 b White Russian 34.2 19.9 48.5 abc Meadowlark 45.9 30.1 61.7 ab Redbor 3.9 -4.2 12.0 b Hybrid I 33.5 19.2 47.8 abc Hybrid E 43.6 27.9 59.4 ab Krul 3.8 -4.4 12.0 ab Reflex 32.9 18.5 47.2 abc Tiger 42.1 26.3 57.8 ab Darkibor 3.2 -3.8 10.1 b Hybrid E 32.2 17.9 46.5 abc Black Magic 38.8 23.9 53.7 b Laerketunge 3.1 -5.1 11.3 ab Black Magic 31.7 18.0 45.4 abc Yellow 38.5 22.7 54.2 ab Reflex 2.9 -5.3 11.1 ab Tiger 27.6 13.3 41.9 abc

VII

Redbor 36.1 20.3 51.8 ab Starbor 2.5 -5.7 10.7 b Meadowlark 25.4 11.0 39.7 bc White Russian 35.8 20.0 51.5 ab Hybrid L 2.4 -5.7 10.5 b Yellow 21.4 7.1 35.7 c

Chew Count (Texture) Leaf Thickness (Visual) Vein Presence (Visual) Sample Mean LowerCL UpperCL Group Sample Mean LowerCL UpperCL Group Sample Mean LowerCL UpperCL Group

Hybrid K1 18.2 14.4 22.1 a Black Magic 44.5 35.1 53.8 a Hybrid L 54.4 42.9 65.9 a Redbor 17.9 14.0 21.7 a Hybrid K1 44.4 34.0 54.9 ab Hybrid E 52.4 40.9 63.9 ab Hybrid L 17.0 13.1 20.9 a Tiger 38.3 27.8 48.8 abc Redbor 50.9 39.4 62.4 ab White Russian 16.5 12.7 20.4 a Yellow 36.5 26.0 46.9 abc Tiger 48.3 36.9 59.8 ab Darkibor 16.3 12.7 20.0 a Starbor 34.1 23.6 44.7 abc Krul 48.1 36.4 59.7 ab Reflex 16.0 12.1 19.9 ab Darkibor 33.6 24.3 42.9 bc Reflex 46.8 35.2 58.5 ab Krul 15.9 12.0 19.7 ab White Russian 33.3 22.9 43.8 abc Hybrid I 46.7 35.2 58.2 ab Starbor 15.5 11.6 19.4 ab Hybrid I 32.4 22.0 42.9 abc Meadowlark 46.3 34.7 58.0 ab Black Magic 15.4 11.7 19.0 ab Meadowlark 32.4 21.8 43.0 abc Hybrid K1 43.6 32.1 55.1 ab Hybrid I 15.3 11.5 19.2 ab Reflex 31.1 20.5 41.7 bc Yellow 43.0 31.5 54.5 ab Laerketunge 14.9 11.0 18.8 ab Laerketunge 29.8 19.2 40.3 bc Darkibor 42.3 32.5 52.1 ab Hybrid E 14.5 10.6 18.3 ab Redbor 27.5 17.0 37.9 c Starbor 39.1 27.4 50.7 ab Tiger 14.2 10.3 18.0 ab Hybrid E 25.7 15.2 36.2 c White Russian 37.7 26.2 49.2 ab Meadowlark 14.0 10.1 17.9 ab Krul 25.3 14.7 35.8 c Black Magic 37.2 27.4 47.0 b Yellow 12.0 8.1 15.9 b Hybrid L 24.7 14.3 35.2 c Laerketunge 37.1 25.4 48.7 ab

Turgid (Visual) Smooth (Visual) Waxy (Visual) Sample Mean LowerCL UpperCL Group Sample Mean LowerCL UpperCL Group Sample Mean LowerCL UpperCL Group

Reflex 54.8 44.4 65.2 ab Yellow 52.3 42.3 62.4 a Hybrid K1 65.8 53.6 77.9 a Yellow 54.5 44.3 64.7 ab Tiger 51.4 41.3 61.5 a Black Magic 63.7 53.0 74.4 a Darkibor 53.1 45.5 60.7 a Hybrid K1 42.0 32.0 52.1 ab Tiger 56.6 44.4 68.7 ab Hybrid K1 52.0 41.8 62.2 ab Hybrid E 41.3 31.3 51.4 ab White Russian 48.4 36.2 60.6 abc Krul 51.8 41.4 62.2 ab Hybrid I 36.8 26.7 46.9 abc Darkibor 41.3 30.7 52.0 bcd Redbor 50.8 40.6 61.0 ab Laerketunge 28.8 18.6 39.0 bcd Hybrid I 40.1 27.9 52.3 bcde Tiger 50.4 40.2 60.6 ab White Russian 28.7 18.6 38.7 bcd Reflex 35.9 23.6 48.2 cdef Black Magic 50.1 42.5 57.7 ab Meadowlark 26.5 16.3 36.7 bcd Starbor 34.2 21.9 46.5 cdef Starbor 48.3 37.9 58.7 ab Redbor 26.4 16.3 36.5 bcd Hybrid L 30.4 18.2 42.6 cdef Hybrid L 44.4 34.2 54.5 ab Reflex 24.3 14.1 34.5 bcd Yellow 29.2 17.1 41.4 def Laerketunge 42.5 32.1 52.9 ab Darkibor 23.4 15.1 31.6 cd Meadowlark 27.9 15.6 40.2 def Meadowlark 41.1 30.7 51.5 ab Starbor 23.2 13.0 33.4 bcd Laerketunge 27.1 14.8 39.4 def White Russian 39.5 29.3 49.7 ab Hybrid L 22.3 12.2 32.3 cd Krul 25.8 13.5 38.1 def Hybrid I 36.7 26.5 46.9 b Krul 20.1 9.9 30.3 cd Hybrid E 23.6 11.5 35.8 ef Hybrid E 35.6 25.4 45.8 b Black Magic 18.8 10.5 27.0 d Redbor 16.2 4.0 28.3 f

VIII

Appendix 2.6. Mean separation of hedonic scores (1-9 scale), attribute intensity (0-100 scale) ratings, and produce scores (1-5 scale) evaluated by a consumer panel (n=90) in a central location test (CLT). Tukey's HSD was performed on nested mixed model estimate means; means with the same grouping letter are not significantly different (P ≤ 0.05).

Black Magic Darkibor Hyb. E Hyb. I Hyb. K1 Hyb. L Redbor Tiger White Russian YellowIL

CONSUMER LIKING (APPEARANCE)*** 6.9 ab 7.6 a 6.3 c 7.0 ab 6.7 bc 7.2 ab 6.8 bc 6.6 bc 6.9 bc 5.4 d

CONSUMER LIKING (COLOR)*** 7.3 a 7.4 a 6.4 b 7.4 a 7.0 ab 7.2 a 6.9 ab 6.9 ab 7.1 a 5.6 c

CONSUMER LIKING (AROMA)ns 5.9 5.5 5.6 6.0 5.6 5.5 5.6 5.7 5.6 5.6

Overall Intensity*** 28.2 a 27.3 ab 20.1 abc 25.8 abc 26.6 ab 22.5 abc 18.3 bc 17.4 c 24.5 abc 21.3 abc

CONSUMER LIKING (FLAVOR)*** 6.6 ab 6.9 a 6.7 a 6.4 abc 6.2 abc 6.9 a 6.5 ab 6.5 ab 5.7 c 5.9 bc

Sour** 25.2 ab 19.0 b 23.8 ab 25.0 ab 25.0 ab 26.0 ab 23.8 ab 23.2 ab 28.7 a 28.3 a

Bitter Flavor*** 47.7 b 34.5 c 42.8 bc 52.0 ab 42.6 bc 45.3 b 43.3 bc 43.6 bc 60.9 a 48.4 b

Sweet*** 30.0 abc 35.9 a 34.3 ab 26.5 bc 31.1 abc 31.8 abc 30.0 abc 35.4 a 25.2 c 33.3 abc

Bitter Aftertaste*** 39.0 bc 24.6 d 35.7 cd 49.2 ab 36.0 cd 38.3 bc 32.7 cd 40.2 bc 57.0 a 39.9 bc

Intensity of Kale Flavor*** 66.3 a 57.3 bcd 53.9 cd 64.8 ab 57.7 abcd 63.0 ab 57.5 abcd 57.3 bcd 62.6 abc 53.5 d

CONSUMER LIKING (TEXTURE)* (+) 7.0 a 6.7 a 6.4 a 6.9 a 6.4 a 6.9 a 6.5 a 6.4 a 6.6 a 6.7 a

Leaf Thickness*** 68.4 a 42.8 cd 38.9 d 57.6 b 61.8 ab 44.6 cd 49.8 c 61.0 ab 48.7 c 47.4 c

Crisp Crunch*** 62.0 a 44.6 d 45.9 cd 53.7 abcd 47.4 cd 46.5 cd 50.1 bcd 54.4 abc 52.8 abcd 59.3 ab

Chewy*** 54.8 a 52.9 a 50.4 ab 52.9 a 55.2 a 49.0 ab 51.5 ab 52.4 ab 47.4 ab 44.7 b

Slimy* (+) 13.8 a 15.1 a 14.9 a 14.6 a 18.0 a 14.9 a 19.5 a 14.8 a 18.4 a 16.2 a

OVERALL CONSUMER LIKING 6.9 ab 6.9 ab 6.7 abc 6.8 ab 6.4 abc 7.1 a 6.5 abc 6.6 abc 6.3 bc 6.0 c Significance codes: 0.001 = '***', 0.01 = '**', 0.05 = '*', not significant = ‘ns’; (+) - adjustment for multiple comparisons resulted in a single sample grouping

IX

Appendix 2.7. Ranked mean separation of hedonic scores (1-9 scale), attribute intensity (0-100 scale) ratings, and produce scores (1-5 scale) evaluated by a consumer panel (n=90) in a central location test (CLT). Tukey's HSD was performed on nested mixed model estimate means; means with the same grouping letter are not significantly different (P ≤ 0.05).

Overall Liking Liking (Appearance) Liking (Color) Liking (Flavor) Lower Upper Grou Mea Lower Upper Lower Upper Lower Upper Sample Mean Sample Group Sample Mean Group Sample Mean Group CL CL p n CL CL CL CL CL CL Hybrid L 7.1 6.7 7.4 a Darkibor 7.6 7.3 7.9 a Hybrid I 7.4 7.1 7.8 a Darkibor 6.9 6.5 7.3 a Darkibor 6.9 6.5 7.2 ab Hybrid L 7.2 6.9 7.5 ab Darkibor 7.4 7.1 7.7 a Hybrid L 6.9 6.5 7.2 a Black Magic 6.9 6.5 7.2 ab Hybrid I 7.0 6.7 7.4 ab Black Magic 7.3 7.0 7.6 a Hybrid E 6.7 6.3 7.1 a Hybrid I 6.8 6.4 7.1 ab Black Magic 6.9 6.6 7.3 ab Hybrid L 7.2 6.9 7.5 a Black Magic 6.6 6.2 6.9 ab Hybrid E 6.7 6.3 7.0 abc White Russian 6.9 6.5 7.2 bc White Russian 7.1 6.8 7.4 a Tiger 6.5 6.1 6.9 ab Tiger 6.6 6.3 7.0 abc Redbor 6.8 6.5 7.2 bc Hybrid K1 7.0 6.6 7.3 ab Redbor 6.5 6.1 6.9 ab Redbor 6.5 6.1 6.8 abc Hybrid K1 6.7 6.4 7.0 bc Redbor 6.9 6.6 7.2 ab Hybrid I 6.4 6.0 6.8 abc Hybrid K1 6.4 6.0 6.7 abc Tiger 6.6 6.3 6.9 bc Tiger 6.9 6.6 7.2 ab Hybrid K1 6.2 5.9 6.6 abc White Russian 6.3 5.9 6.6 bc Hybrid E 6.3 6.0 6.6 c Hybrid E 6.4 6.0 6.7 b YellowIL 5.9 5.5 6.2 bc YellowIL 6.0 5.7 6.4 c YellowIL 5.4 5.0 5.7 d YellowIL 5.6 5.3 5.9 c White Russian 5.7 5.3 6.1 c

Kale Intensity (Flavor) Overall Intensity (Aroma) Bitter (Flavor) Sweet (Flavor)

Lower Upper Grou Mea Lower Upper Lower Upper Lower Upper Sample Mean Sample Group Sample Mean Group Sample Mean Group CL CL p n CL CL CL CL CL CL Black Magic 66.3 61.9 70.7 a Black Magic 28.2 23.4 33.1 a White Russian 60.9 55.6 66.2 a Darkibor 35.9 31.1 40.8 a Hybrid I 64.8 60.5 69.2 ab Darkibor 27.3 22.5 32.2 ab Hybrid I 52.0 46.7 57.3 ab Tiger 35.4 30.6 40.2 a Hybrid L 63.0 58.6 67.4 ab Hybrid K1 26.6 21.7 31.5 ab YellowIL 48.4 43.1 53.7 b Hybrid E 34.3 29.4 39.2 ab White Russian 62.6 58.3 67.0 abc Hybrid I 25.8 20.9 30.7 abc Black Magic 47.7 42.4 53.0 b YellowIL 33.3 28.4 38.1 abc Hybrid K1 57.7 53.3 62.1 abcd White Russian 24.5 19.6 29.3 abc Hybrid L 45.3 40.0 50.6 b Hybrid L 31.8 26.9 36.6 abc Redbor 57.5 53.1 61.8 abcd Hybrid L 22.5 17.6 27.3 abc Tiger 43.6 38.4 48.9 bc Hybrid K1 31.1 26.2 35.9 abc Tiger 57.3 53.0 61.7 bcd YellowIL 21.3 16.4 26.2 abc Redbor 43.3 38.0 48.6 bc Black Magic 30.0 25.2 34.9 abc Darkibor 57.3 52.9 61.7 bcd Hybrid E 20.1 15.2 25.0 abc Hybrid E 42.8 37.5 48.1 bc Redbor 30.0 25.1 34.8 abc Hybrid E 53.9 49.5 58.3 cd Redbor 18.3 13.5 23.2 bc Hybrid K1 42.6 37.3 47.9 bc Hybrid I 26.5 21.6 31.4 bc YellowIL 53.5 49.1 57.9 d Tiger 17.4 12.5 22.3 c Darkibor 34.5 29.2 39.8 c White Russian 25.2 20.4 30.0 c

Sour (Flavor) Bitter (Aftertaste) Leaf Thickness (Texture) Chewy (Texture)

Lower Upper Grou Mea Lower Upper Lower Upper Lower UpperC Sample Mean Sample Group Sample Mean Group Sample Mean Group CL CL p n CL CL CL CL CL L White Russian 28.7 24.0 33.4 a White Russian 57.0 51.2 62.9 a Black Magic 68.4 64.6 72.3 a Hybrid K1 55.2 50.7 59.6 a YellowIL 28.3 23.6 33.0 a Hybrid I 49.2 43.3 55.0 ab Hybrid K1 61.8 57.9 65.7 ab Black Magic 54.8 50.4 59.2 a Hybrid L 26.0 21.3 30.6 ab Tiger 40.2 34.4 46.0 bc Tiger 61.0 57.1 64.9 ab Hybrid I 52.9 48.5 57.3 a Black Magic 25.2 20.5 29.9 ab YellowIL 39.9 34.0 45.8 bc Hybrid I 57.6 53.7 61.5 b Darkibor 52.9 48.4 57.3 a Hybrid K1 25.0 20.3 29.7 ab Black Magic 39.0 33.2 44.9 bc Redbor 49.8 46.0 53.7 c Tiger 52.4 47.9 56.8 ab Hybrid I 25.0 20.3 29.7 ab Hybrid L 38.3 32.4 44.1 bc White Russian 48.7 44.9 52.6 c Redbor 51.5 47.0 55.9 ab Redbor 23.8 19.2 28.5 ab Hybrid K1 36.0 30.1 41.9 cd YellowIL 47.4 43.5 51.2 c Hybrid E 50.4 46.0 54.8 ab Hybrid E 23.8 19.1 28.5 ab Hybrid E 35.7 29.8 41.6 cd Hybrid L 44.6 40.7 48.5 cd Hybrid L 49.0 44.6 53.4 ab Tiger 23.2 18.5 27.9 ab Redbor 32.7 26.8 38.6 cd Darkibor 42.8 38.9 46.6 cd White Russian 47.4 43.0 51.8 ab Darkibor 19.0 14.3 23.7 b Darkibor 24.6 18.8 30.5 d Hybrid E 38.9 35.0 42.8 d YellowIL 44.7 40.3 49.1 b

Crisp Crunch (Texture) Slimy (Texture) Concept Fit Purchase Intent

Lower Upper Grou Mea Lower Upper Lower Upper Lower UpperC Sample Mean Sample Group Sample Mean Group Sample Mean Group CL CL p n CL CL CL CL CL L Black Magic 62.0 57.5 66.5 a Redbor 19.5 15.2 23.7 a Darkibor 4.28 4.06 4.49 a Darkibor 3.93 3.69 4.17 a YellowIL 59.3 54.8 63.9 ab White Russian 18.4 14.2 22.6 a HybL 4.26 4.05 4.48 a HybL 3.81 3.57 4.05 a Tiger 54.4 49.9 58.9 abc Hybrid K1 18.0 13.8 22.3 a Black Magic 4.09 3.88 4.30 ab Black Magic 3.68 3.44 3.92 ab Hybrid I 53.7 49.2 58.2 abcd YellowIL 16.2 11.9 20.4 a HybI 4.05 3.83 4.26 abc HybI 3.53 3.29 3.77 abc White Russian 52.8 48.3 57.3 abcd Darkibor 15.1 10.9 19.4 a HybK1 3.95 3.73 4.16 abcd Tiger 3.51 3.27 3.75 abc Redbor 50.1 45.5 54.6 bcd Hybrid L 14.9 10.7 19.1 a Redbor 3.78 3.57 4.00 bcde Redbor 3.49 3.25 3.73 abc Hybrid K1 47.4 42.8 51.9 cd Hybrid E 14.9 10.6 19.1 a Tiger 3.63 3.42 3.85 cde HybE 3.45 3.21 3.69 abc Hybrid L 46.5 42.0 51.1 cd Tiger 14.8 10.6 19.0 a White Russian 3.56 3.34 3.77 de HybK1 3.44 3.20 3.68 abc Hybrid E 45.9 41.3 50.4 cd Hybrid I 14.6 10.4 18.9 a HybE 3.43 3.21 3.65 ef White Russian 3.21 2.97 3.45 bc Darkibor 44.6 40.1 49.1 d Black Magic 13.8 9.6 18.0 a YellowIL 3.08 2.86 3.30 f YellowIL 3.05 2.81 3.29 c

X

Appendix 2.8. Chi-square tests of independence for filtered CLT consumer (n=88, four clusters) demographic data. Observed (Revised) Frequency Expected Original Demographic Data Table Frequency CLUSTER CLUSTER CLUSTER

X2 (df = 3, N = TOT TOT Age 1 2 3 4 Age 1 2 3 4 1 2 3 4 88) =8.0121, AL AL p = 0.04576 1 1 2 1 1 9. 18-24 years 8 12 44 18-44 years 30 77 25.4 27.1 14.9 0 4 2 1 4 6 1 45 - 74 1. 25-34 years 8 3 2 26 7 0 1 3 11 3.6 3.9 2.1 3 years 4 GRAND 2 1 1 35-44 years 4 0 3 0 7 31 88 TOTAL 9 1 7

45-54 years 1 0 0 0 1

55-64 years 4 0 1 3 8

65-74 years 2 0 0 0 2

GRAND 2 1 3 17 88 TOTAL 9 1 1

CLUSTER CLUSTER CLUSTER

X2 (df = 3, N = TOT TOT Gender 1 2 3 4 Gender 1 2 3 4 1 2 3 4 87) = 1.6974, AL AL p = 0.6375 2 2 2 1 8. Female 7 14 63 Female 7 21 63 20.3 22.4 12.3 1 1 1 4 0 1 3. Male 7 4 3 24 Male 7 4 10 3 24 7.7 8.6 4.7 0 0 GRAND 2 1 1 Other (FTM) 1 0 0 0 1 31 87 TOTAL 8 1 7 GRAND 2 1 3 17 88 TOTAL 9 1 1

CLUSTER CLUSTER CLUSTER

X2 (df = 6, N = TOT TOT Ethnicity 1 2 3 4 Ethnicity 1 2 3 4 1 2 3 4 88) = 12.526, AL AL p = 0.05121 1 3. Asian 4 5 7 26 Asian 4 5 10 7 26 8.6 9.2 5.0 0 3 2 5. Black 1 0 2 1 4 White 6 12 8 47 15.5 16.6 9.1 1 9 1. Hispanic 1 0 2 1 4 Other 4 0 9 2 15 4.9 5.3 2.9 9 2 1 GRAND 2 1 1 White 6 8 47 31 88 1 2 TOTAL 9 1 7

White/Asian 0 0 1 0 1

White/Hispa 2 0 4 0 6 nic GRAND 2 1 3 17 88 TOTAL 9 1 1

CLUSTER CLUSTER CLUSTER

X2 (df = 6, N = Consumptio TOT Consumpti TOT 1 2 3 4 1 2 3 4 1 2 3 4 88) =12.396, n Rate AL on Rate AL p = 0.05369 At least once >2x per 4. 0 1 0 0 1 8 7 14 5 34 11.2 12.0 6.6 per day week 3 4-6 times Once per 1 3. 3 2 3 2 10 3 8 2 25 8.2 8.8 4.8 per week week 2 1 2-3 times 1 Few times 1 3. 5 4 3 23 9 1 9 29 9.6 10.2 5.6 per week 1 per month 0 6 Once per 1 GRAND 2 1 1 3 8 2 25 31 88 week 2 TOTAL 9 1 7 Few times 9 1 9 10 29 per month GRAND 2 1 3 17 88 TOTAL 9 1 1

CLUSTER CLUSTER CLUSTER

X2 (df = 3, N = Variety TOT Variety TOT 1 2 3 4 1 2 3 4 1 2 3 4 88) = 3.988, Seeking AL Seeking AL p = 0.2628 Unpropmpt 1 4. Very often 5 1 3 0 9 5 11 4 35 11.5 12.3 6.8 ed/Often 5 4 Sometimes With 1 1 1 6. (Unprompte 4 8 4 26 Prompt/Rar 6 20 53 17.5 18.7 10.2 0 4 3 6 d) ely Sometimes 1 1 GRAND 2 1 1 (Reccomend 4 10 43 31 88 0 9 TOTAL 9 1 7 ) Very rarely 3 2 1 3 9

Other 1 0 0 0 1 GRAND 1 3 29 17 88 TOTAL 1 1

XI

Appendix 2.9. Favorite kale types by preparation method and sensory attribute according to QMA consumer (n=14) participants. “Most Preferred” and “Least Preferred” kale types were identified during the Love It! or Hate It! portion of the facilitated focus group. The most important trait distinguishing “Most Preferred” and “Least Preferred” kale types are listed under each image.

XII

Appendix 2.10. Radar (spider) plots depicting estimated means of 15 sensory attributes significantly different (P ≤ 0.05) between kale samples (n=15) in descriptive analysis. Sensory attributes were evaluated on an unstructured numerical line scales marked with labeled endpoints indicating attribute intensity (0 = “none” to 100 = “extremely). Chew count (‘NoChews’) was included as an unbound supplementary quantitative variable; values ranged from 12.0-18.2 chews and were also significantly different between kale samples.

XIII

Appendix 2.11. A) Pearson (𝜌) correlation among estimated means of sensory attributes (n=15) significantly different (P ≤ 0.05) between kale samples (n=15) in descriptive analysis. B) Variable projection map of sensory attributes (n=15) depicting correlation among variables in the factors space. C) Product factor map depicting trends among kale samples (n=10) in the factors space.

XIV

Appendix 2.12. A) Mean hedonic scores (1-9) of sensory modalities (appearance, flavor, overall liking, and texture) for kale samples (n=10) evaluated by n=90 consumers during a central location test (CLT). Four consumer clusters within each sensory modality were identified through AHC of consumer hedonic scores; clusters are independent among sensory modalities. B-D) External consumer preference mapping of appearance (B), flavor (C), and texture (D) sensory modalities. Preference maps were obtained by PLS analysis of descriptive data and AHC consumer hedonic data. Axes represent the first (F1-26.22%) and second (F2 35.02%) dimensions of PCA performed on descriptive data. Four consumer clusters (orange) were identified through AHC of consumer hedonic scores of kale samples (white). A contour plot (olive) has been overlaid to illustrate the percentage of clusters with above-average preference in a given region of the preference map. XV

APPENDIX: Chapter Three

Appendix 3.1. Site information for replicated field trials of hybrid collard germplasm in Western Kenya (May ‘19) and Upstate New York (Sept. ‘19). Code Name Location Coordinates

ACL Advantage Crops Limited Research Site Rodi Kopany, Kenya 0°38'16.8"S, 34°30'54.0"E SAMANGA Samanga (Oluch) Farmer Site Nyangweso, Kenya 0°27'36.0"S, 34°33'03.6"E ARAM Aram Market/Farmer Site Akom, Kenya 0°12'28.8"S, 34°20'20.4"E FREVILLE Homer C. Thompson Vegetable Res Farm Freeville, NY 42°31'21.4"N, 76°19'44.2"W GENEVA Cornell AgriTech – McCarthy Res Farm Geneva, NY 42°53'43.4"N, 77°00’36.2"W

XVI

Appendix 3.2. Nested mixed model output for liking scores (9pt hedonic scale) comparing collard genotypes in Western Kenya (May ‘19) and Upstate New York (Sept. ‘19). A cross-country interaction model was performed in addition to within-country models, as ‘Top Bunch' cultivar was only evaluated in New York. Output represents fixed-factor ANOVA (Satterthwaite's Method) 2 2 2 and random effect variance components attributed to panelist variance [𝜎푝 ], location [𝜎l ], and residual variance [𝜎 ]. Percentage 2 of total random variance (% 𝜎total ) provides an indication of the effect’s contribution to the overall random effect variance.

CROSS COUNTRY CROSS COUNTRY

Analysis of Variance (TYPE III TEST) Random Variance Components 2 Effect SS MS NumDF DenDF F-value Pr(>F) Effect Variance SD % 𝜎total -15 2 Genotype 329.3 65.86 5 802.17 16.2609 2.682x10 *** 𝜎푝 0.2863 0.5351 6.3 ns 2 Country 19.7 19.703 1 4.1 4.8647 0.09034 𝜎l 0.176 0.4196 3.9 ns 2 Age 20.47 4.095 5 170.07 1.011 0.41284 𝜎 4.0502 2.0125 89.8 Gender 0.06 0.029 2 171.88 0.0071 0.99295ns Cultivar*Country 218.58 43.717 5 802.08 10.7938 4.621x10-10 ***

NEW YORK, USA NEW YORK, USA

Analysis of Variance (TYPE III TEST) Random Variance Components 2 Effect SS MS NumDF DenDF F-value Pr(>F) Effect Variance SD % 𝜎total -10 2 Genotype 97.665 16.2775 6 498.94 9.0623 2.03x10 *** 𝜎푝 0.6913 0.8315 26.6 ns 2 Age 4.585 0.9169 5 93.73 0.5105 0.7677 𝜎l 0.1145 0.3384 4.4 ns 2 Gender 3.317 1.6583 2 94.23 0.9233 0.4008 𝜎 1.7962 1.3402 69.0

WESTERN KENYA WESTERN KENYA

Analysis of Variance (TYPE III TEST) Random Variance Components 2 Effect SS MS NumDF DenDF F-value Pr(>F) Effect Variance SD % 𝜎total -13 2 Genotype 451.97 90.394 5 407.52 14.4887 4.477x10 *** 𝜎푝 0.01712 0.1308 0.3 ns 2 Age 39.08 7.816 5 74.6 1.2528 0.2934 𝜎l 0.06017 0.2453 1.0 ns 2 Gender 1.75 0.877 2 72.9 0.1406 0.8691 𝜎 6.23898 2.4978 98.8 Significance codes: 0.001 = '***', 0.01 = '**', 0.05 = '*', not significant = ‘ns’

XVII

Appendix 3.3. Estimated mean liking (hedonic scale, 1-9) and Tukey's HSD groupings of cultivars in Western Kenya (May ‘19) and Upstate New York (Sept. ‘19) consumer acceptance tastings. Results were averaged over age and gender effects and degrees- of-freedom calculated using Kenward-Roger method. NEW YORK, USA Cultivar Mean Group SE DF LowerCL UpperCL 982 (‘P8CMS x KP1702’) 7.57 ab 0.362 13.07 6.79 8.35 709 ('Champion x KP1702') 7.35 a 0.28 4.79 6.62 8.08 ('Top Bunch') 7.07 abc 0.281 4.87 6.35 7.8 141 ('P8CMS x KP1') 6.79 bcd 0.281 4.84 6.06 7.52 267 ('Mfalme') 6.59 cd 0.296 5.84 5.86 7.32 591 ('KP1706 x KP15') 6.44 d 0.28 4.81 5.71 7.17 356 ('Biera') 6.26 d 0.28 4.81 5.53 6.99

WESTERN KENYA Cultivar Mean Group SE DF LowerCL UpperCL 267 ('Mfalme') 7.47 a 0.355 23.9 6.74 8.21 141 ('P8CMS x KP1') 6.65 ab 0.358 24.5 5.92 7.39 982 (‘P8CMS x KP1702’) 6.36 ab 0.355 24.1 5.63 7.1 709 ('Champion x KP1702') 5.57 bc 0.356 24.1 4.84 6.31 356 ('Biera') 4.9 c 0.353 23.5 4.17 5.63 591 ('KP1706 x KP15') 4.83 c 0.358 24.4 4.1 5.57

XVIII

Appendix 3.4. Green-red (a*) by blue-yellow (b*) components of CIELAB measurements on lamina surface of collard genotypes subject to consumer testing. Each genotype is represented by n = 6 color measurements., colored according to corresponding HEX values.

XIX

Appendix 3.5. Flash Profile panelist descriptive attributes, including respective panelist-defined sensory modalities and frequency among panelists, for raw product preparation of n = 6 hybrid collard cultivars evaluated in Rodi Kopany, Kenya (Homa Bay County) and Geneva, NY (New York). Hedonic attributes are italicized. Panelists did not consume raw products. HOMA BAY - Raw Preparation (n = 41) NEW YORK - Raw Preparation (n = 105) Modality Attribute FREQ Modality Attribute FREQ Modality Attribute FREQ Modality Attribute FREQ

Color Green 10 Color Blue-Green 5 Color Medium-Green 1 Shape Pointy Leaf 1 Durability Has Weight 5 Durability Thick 5 Color Petiole Color 1 Shape Prominent Laterals 1 Color Light Green 4 Shape Curly 5 Color Pretty 1 Shape Ragged 1 Shape Big 4 Shape Long/Tall 5 Color Shiny 1 Shape Rigid 1 Shape Broad 4 Shape Round 5 Color Translucent 1 Shape Rounded Fan 1 Shape Curly 4 Shape Broad 4 Color Underside White 1 Shape Size Area 1 Visual Attractive 4 Texture Veiny 4 Color Variegated 1 Shape Skinny 1 Shape Long/Tall 3 Color Light-Green 3 Color Yellow 1 Shape Smooth Curly 1 Texture Soft 3 Durability Tough 3 Color Yellow-Blue 1 Shape Stringy Midrib 1 Durability Big Stem 2 Shape Flat 3 Durability Dainty 1 Shape Thick Midrib 1 Durability Hard 2 Shape Veiny 3 Durability Durable 1 Shape Thin Midrib 1 Durability Lasts Longer 2 Shape Wavy 3 Durability Hardy 1 Shape Tree-Like 1 Durability Light Weight 2 Texture Smooth 3 Durability Limp 1 Texture Bumpy Leaf 1 Durability Long 2 Texture Thick 3 Durability Stocky 1 Texture Coarse 1 Durability Soft 2 Color Blue 2 Durability Structure 1 Texture Crisp 1 Durability Strong 2 Color Bright 2 Durability Veiny 1 Texture Delicate 1 Durability Strong Midrib 2 Color Dark 2 Durability Waxy 1 Texture Dense Veins 1 Shape Narrow 2 Color Dark-Green 2 Durability Weighty 1 Texture Dry 1 Shape Oval 2 Color Deep-Green 2 Shape Auricle 1 Texture Dull 1 Visual No Pest/Disease 2 Color Green 2 Shape Bare Stalk 1 Texture Easily Rolled 1 Visual Veins 2 Color Purple 2 Shape Blade Margin 1 Texture Fibrous 1 Color Beautiful 1 Color Purple Midrib 2 Shape Bottom Heavy 1 Texture Fibrous 1 Color Brown 1 Color Waxy 2 Shape Branched Veins 1 Texture Fine Veins 1 Color Dark Green 1 Durability Stiff 2 Shape Bulbous 1 Texture Fleshy 1 Color Evergreen 1 Durability Sturdy 2 Shape Cupped 1 Texture Kale-Like 1 Color No Much Color 1 Shape Large 2 Shape Deep Lobe 1 Texture Lush 1 Color Purple 1 Shape Oblong 2 Shape Elongate 1 Texture Matted 1 Color White Vein 1 Shape Wide 2 Shape Fringed 1 Texture Pliable 1 Color Yellow 1 Texture Curly 2 Shape Heavy Midrib 1 Texture Porous 1 Durability Big 1 Texture Prominent Midrib 2 Shape Indented Midrib 1 Texture Silky 1 Durability Easy to Care For 1 Texture Ruffled 2 Shape Leaf Angle 1 Texture Soft 1 Durability Fetch More Money 1 Texture Wrinkly 2 Shape Lobed Base 1 Texture Spiky 1 Durability Harvest For Long 1 Color Forest Green 1 Shape Long Stem 1 Texture Tough 1 Durability High Production 1 Color Glazy 1 Shape Oval 1 Visual Branched Veins 1 Durability Wilts Easily 1 Color Kali-ish 1 Shape Petiole Size 1 Visual Cosmetics 1 Shape ZigZag 1 TOTAL 163 Visual Colorful 1 % Hedonic 1.20% Visual Long Petiole 1 Visual Move Money 1 Shine By Jar 1 Sweet 1 TOTAL 85 % Hedonic 8.20%

XX

Appendix 3.6. Flash Profile panelist descriptive attributes, including respective sensory modalities and frequency among panelists, for cooked product preparation of n = 6 hybrid collard cultivars evaluated in Rodi Kopany, Kenya (Homa Bay) and Geneva, NY (New York). Hedonic attributes are italicized. HOMA BAY NEW YORK Cooked Preparation (n = 33) Cooked Preparation (n = 66)

Modality Attribute FREQ Modality Attribute FREQ Modality Attribute FREQ Aroma Sweet 7 Flavor Bitter 7 Flavor Pungent 1 Visual Green 7 Flavor Sweet 5 Flavor Salty 1 Flavor Sweet 5 Texture Crunchy 5 Flavor Savory Umami 1 Aftertaste Salty 3 Visual Dark Green 5 Texture Crisp 1 Aftertaste Sweet 3 Aftertaste Bitter 4 Texture Dense 1 Visual Mixes w/ Oil 3 Texture Chewy 4 Texture Fibrous 1 Aroma Bad 2 Aroma Earthy 3 Texture Gamey 1 Flavor Bad 2 Visual Dark 3 Texture Moist 1 Flavor Nice 2 Aroma Citrus 2 Texture Slippery 1 Flavor Salty 2 Aroma Sulfur 2 Texture Soft Mushy 1 Flavor Tasty 2 Flavor Nutty 2 Texture Stringy 1 Visual Purple 2 Visual Green 2 Texture Sturdy 1 Visual Watery 2 Aftertaste Long Aftertaste 1 Texture Tender 1 Visual Yellow 2 Aftertaste Present 1 Texture Thick 1 Aftertaste Remain w/ Appetite 1 Aftertaste Sharp 1 Texture Tough 1 Aftertaste Sour 1 Aroma Bright 1 Texture Waxy 1 Aftertaste Tasteless 1 Aroma Coffee 1 Visual Appealing 1 Aftertaste Tasty 1 Aroma Grassy 1 Visual Blue-Green 1 Aroma Beet 1 Aroma Green 1 Visual Bold-Green 1 Aroma Poor 1 Aroma Intense 1 Visual Bright 1 Aroma Salty 1 Aroma Nutty 1 Visual Brown 1 Flavor Not Tasty 1 Aroma Sharp 1 Visual Chopped 1 Flavor Poor 1 Aroma Uncooked 1 Visual Green Olive 1 Flavor Soft 1 Aroma Vegetal 1 Visual Large Stem 1 Flavor Sour 1 Aroma Warm Umami 1 Visual Moist 1 Flavor Strong Appetite 1 Flavor Bland 1 Visual Opalescent 1 Flavor Tasteless 1 Flavor Bright 1 Visual Slimy 1 Texture Hard Chew 1 Flavor Buttery 1 Visual Stemy 1 Texture Soft Chew 1 Flavor Citrus 2 Visual Thick Midrib 1 Visual Color Fades 1 Flavor Fresh 1 Visual Variegated 1 Visual Dark Green 1 Flavor Juicy 1 Visual Vein Contrast 1 Visual Light Green 1 Flavor Mellow 1 Visual Yellow 1 Visual Retains Color 1 Flavor Mineral 1 Visual Yellow-Green 1 TOTAL 63 TOTAL 97 % Hedonic 19.0% % Hedonic 1.0%

XXI

APPENDIX: Chapter Four

Appendix 4.1. Phenotype descriptions of morphological traits evaluated in replicate field environments during the 2019 field season.

Scale Phenotype Description

Image Visual Percentage of each genotype image in the online consumer (%) Volume survey occupied by the leaf sample.

Plot Lamina Degree of lamina blistering on a scale of 1 (flat) to 5 (extreme (0-5) Blistering blistering)

Plot Lamina Degree of lamina curling along margins measured at harvest (0-5) Curling on a scale from 1 (none/flat) to 5 (extremely curly).

Measured on three (3) plants per plot. Use a flexible ruler to Subsample Lamina measure the distance from the base of the leaf blade (at petiole (cm) Length joint) to the leaf blade apex (tip). The measurement should run parallel to the length of the leaf mid-rib. Measured on three (3) plants per plot. Use a flexible ruler to Subsample Lamina measure the cross-sectional distance (cm) of the largest portion (cm) Width of the leaf blade. The measurement should run perpendicular to the leaf mid-rib and approximately halfway the leaf length. Measured on three (3) plants per plot. Use a large caliper to Subsample Petiole measure the distance from the base of the leaf blade (at the (cm) Length petiole joint) to the to the stipule joint (no longer connected to the main stem).

Measured on three (3) plants per plot. Use a large caliper to Subsample Petiole measure the cross-sectional distance (cm) of the petiole at the (cm) Width base of the leaf blade (where petiole joins with lamina).

Three (3) marketable leaves from each plot were weighed fresh Plot and dried in paper bags at 60°C until weights stabilized. Dry Dry Matter (%) leaf tissue was re-weighed, and the percentage of dry matter was calculated by dividing the fresh weight.

XXII

Appendix 4.2. Genotypes (n = 129) subject to genotyping by sequencing (GBS) and population structure analysis. Genotype numbers for diallel materials consist of a female (F) and male (M) two-digit designator. Summary genotype descriptors were calculated after SNP filtering (i.e. minor allele frequency < 0.05; minimum depth = 4; maximum missing: 0.75). Geno Breeder/ Market Indv. Indiv # Indv. Freq. Inbreeding # Name F M Subspecies Breeding Class # Collection Class Depth Missing Missing Coeff (F) Sites 11 KP1702ALT_P1 1 1 P. Griffiths Collard_Yellow acephala Inbred 16.6 7586 0.22 0.57 27397 11 KP1702ALT_P3 1 1 P. Griffiths Collard_Yellow acephala Inbred 30.6 3456 0.10 0.59 31527 11 KP1702ALT_P2 1 1 P. Griffiths Collard_Yellow acephala Inbred 10.0 13685 0.39 0.59 21298 12 KP1702AxKP1703 1 2 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 36.9 1196 0.03 -0.16 33787 13 KP1702AxKP1704 1 3 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 38.5 1209 0.03 -0.14 33774 14 KP1702AxKP1706 1 4 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 26.3 2560 0.07 -0.09 32423 15 KP1702AxKP1707 1 5 P. Griffiths Hybrid italica x viridis Hybrid_Diallel 41.5 997 0.03 -0.11 33986 16 KP1702AxKP1713 1 6 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 71.6 380 0.01 -0.15 34603 17 KP1702AxKP1715 1 7 P. Griffiths Hybrid viridis x viridis Hybrid_Diallel 48.9 682 0.02 -0.09 34301 18 KP1702Ax16KP15 1 8 P. Griffiths Hybrid viridis x viridis Hybrid_Diallel 37.9 1545 0.04 -0.02 33438 21 KP1703xKP1702A 2 1 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 59.4 267 0.01 -0.21 34716 22 KP1703_P3 2 2 P. Griffiths Kale_Tuscan acephala Inbred 42.2 2151 0.06 0.64 32832 22 KP1703_P2 2 2 P. Griffiths Kale_Tuscan acephala Inbred 24.7 5486 0.16 0.64 29497 22 KP1703_P1 2 2 P. Griffiths Kale_Tuscan acephala Inbred 33.2 2725 0.08 0.64 32258 23 KP1703xKP1704 2 3 P. Griffiths Hybrid acephala x acephala Hybrid_Diallel 71.2 257 0.01 -0.17 34726 24 KP1703xKP1706 2 4 P. Griffiths Hybrid acephala x acephala Hybrid_Diallel 63.8 340 0.01 -0.18 34643 25 KP1703xKP1707 2 5 P. Griffiths Hybrid italica x acephala Hybrid_Diallel 30.3 2587 0.07 -0.05 32396 26 KP1703xKP1713 2 6 P. Griffiths Hybrid acephala x acephala Hybrid_Diallel 39.5 872 0.02 -0.17 34111 27 KP1703xKP1715 2 7 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 44.2 993 0.03 -0.13 33990 28 KP1703x16KP15 2 8 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 51.4 592 0.02 -0.14 34391 31 KP1704xKP1702A 3 1 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 34.2 409 0.01 -0.22 34574 32 KP1704xKP1703 3 2 P. Griffiths Hybrid acephala x acephala Hybrid_Diallel 26.1 833 0.02 -0.23 34150 33 KP1704-P1_P2 3 3 P. Griffiths Kale_RedCurly acephala Inbred 40.3 1999 0.06 0.53 32984 33 KP1704-P1_P1 3 3 P. Griffiths Kale_RedCurly acephala Inbred 18.2 7448 0.21 0.56 27535 33 KP1704-P1_P3 3 3 P. Griffiths Kale_RedCurly acephala Inbred 21.4 5391 0.15 0.56 29592 34 KP1704xKP1706 3 4 P. Griffiths Hybrid acephala x acephala Hybrid_Diallel 25.6 1023 0.03 -0.15 33960 35 KP1704xKP1707 3 5 P. Griffiths Hybrid italica x acephala Hybrid_Diallel 37.3 378 0.01 -0.22 34605 36 KP1704xKP1713 3 6 P. Griffiths Hybrid acephala x acephala Hybrid_Diallel 25.2 1127 0.03 -0.11 33856 37 KP1704xKP1715 3 7 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 19.2 2226 0.06 -0.21 32757 38 KP1704x16KP15 3 8 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 26.9 689 0.02 -0.25 34294 41 KP1706xKP1702A 4 1 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 61.9 297 0.01 -0.17 34686 42 KP1706xKP1703 4 2 P. Griffiths Hybrid acephala x acephala Hybrid_Diallel 44.6 836 0.02 -0.16 34147 43 KP1706xKP1704 4 3 P. Griffiths Hybrid acephala x acephala Hybrid_Diallel 43.1 1131 0.03 -0.09 33852 44 KP1706_P1 4 4 P. Griffiths Kale_GreenCurly acephala Inbred 60.5 1540 0.04 0.55 33443 44 KP1706_P2 4 4 P. Griffiths Kale_GreenCurly acephala Inbred 25.7 4975 0.14 0.56 30008 44 KP1706_P3 4 4 P. Griffiths Kale_GreenCurly acephala Inbred 44.2 1775 0.05 0.56 33208 45 KP1706xKP1707 4 5 P. Griffiths Hybrid italica x acephala Hybrid_Diallel 45.4 777 0.02 -0.12 34206 46 KP1706xKP1713 4 6 P. Griffiths Hybrid acephala x acephala Hybrid_Diallel 44.9 720 0.02 -0.09 34263

XXIII

Appendix 4.2 cont. Genotypes (n = 129) subject to genotyping by sequencing (GBS) and population structure analysis. Genotype numbers for diallel materials consist of a female (F) and male (M) two-digit designator. Summary genotype descriptors were calculated after SNP filtering (i.e. minor allele frequency < 0.05; minimum depth = 4; maximum missing: 0.75). Geno Breeder/ Market Indv. Indiv # Indv. Freq. Inbreeding # Name F M Subspecies Breeding Class # Collection Class Depth Missing Missing Coeff (F) Sites 47 KP1706xKP1715 4 7 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 57.3 298 0.01 -0.19 34685 48 KP1706x16KP15 4 8 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 41.2 931 0.03 -0.14 34052 51 KP1707xKP1702A 5 1 P. Griffiths Hybrid italica x viridis Hybrid_Diallel 45.9 487 0.01 -0.15 34496 52 KP1707xKP1703 5 2 P. Griffiths Hybrid italica x acephala Hybrid_Diallel 31.0 2189 0.06 -0.06 32794 53 KP1707xKP1704 5 3 P. Griffiths Hybrid italica x acephala Hybrid_Diallel 60.6 523 0.01 -0.15 34460 54 KP1707xKP1706 5 4 P. Griffiths Hybrid italica x acephala Hybrid_Diallel 43.3 722 0.02 -0.14 34261 55 KP1707_P3 5 5 P. Griffiths Broccoli_Jagged italica Inbred 32.0 3000 0.09 0.70 31983 55 KP1707_P2 5 5 P. Griffiths Broccoli_Jagged italica Inbred 39.1 2605 0.07 0.70 32378 55 KP1707_P1 5 5 P. Griffiths Broccoli_Jagged italica Inbred 12.4 10802 0.31 0.71 24181 56 KP1707xKP1713 5 6 P. Griffiths Hybrid italica x acephala Hybrid_Diallel 46.5 846 0.02 -0.13 34137 57 KP1707xKP1715 5 7 P. Griffiths Hybrid italica x viridis Hybrid_Diallel 32.8 1829 0.05 -0.06 33154 58 KP1707x16KP15 5 8 P. Griffiths Hybrid italica x acephala Hybrid_Diallel 16.5 7019 0.20 -0.01 27964 61 KP1713xKP1702A 6 1 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 31.5 467 0.01 -0.18 34516 62 KP1713xKP1703 6 2 P. Griffiths Hybrid acephala x acephala Hybrid_Diallel 20.4 1816 0.05 -0.19 33167 63 KP1713xKP1704 6 3 P. Griffiths Hybrid acephala x acephala Hybrid_Diallel 31.0 574 0.02 -0.12 34409 64 KP1713xKP1706 6 4 P. Griffiths Hybrid acephala x acephala Hybrid_Diallel 23.0 1483 0.04 -0.12 33500 65 KP1713xKP1707 6 5 P. Griffiths Hybrid italica x acephala Hybrid_Diallel 25.1 944 0.03 -0.15 34039 66 KP1713_P2 6 6 P. Griffiths Kale_Red acephala Inbred 42.1 2126 0.06 0.55 32857 66 KP1713_P1 6 6 P. Griffiths Kale_Red acephala Inbred 39.2 2363 0.07 0.57 32620 67 KP1713xKP1715 6 7 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 24.9 791 0.02 -0.17 34192 68 KP1713x16KP15 6 8 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 26.2 696 0.02 -0.18 34287 71 KP1715xKP1702A 7 1 P. Griffiths Hybrid viridis x viridis Hybrid_Diallel 48.8 940 0.03 -0.07 34043 72 KP1715xKP1703 7 2 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 47.9 534 0.02 -0.15 34449 73 KP1715xKP1704 7 3 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 16.5 7469 0.21 -0.08 27514 74 KP1715xKP1706 7 4 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 49.6 639 0.02 -0.17 34344 75 KP1715xKP1707 7 5 P. Griffiths Hybrid italica x viridis Hybrid_Diallel 49.0 558 0.02 -0.11 34425 76 KP1715xKP1713 7 6 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 37.3 1147 0.03 -0.10 33836 77 KP1715_P2 7 7 P. Griffiths Collard_Elongate viridis Inbred 21.9 5527 0.16 0.53 29456 77 KP1715_P3 7 7 P. Griffiths Collard_Elongate viridis Inbred 21.4 5615 0.16 0.58 29368 77 KP1715_P1 7 7 P. Griffiths Collard_Elongate viridis Inbred 30.4 4862 0.14 0.59 30121 78 KP1715x16KP15 7 8 P. Griffiths Hybrid viridis x viridis Hybrid_Diallel 27.3 2852 0.08 0.07 32131 81 16KP15xKP1702A 8 1 P. Griffiths Hybrid viridis x viridis Hybrid_Diallel 17.0 6665 0.19 0.02 28318 82 16KP15xKP1703 8 2 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 51.1 730 0.02 -0.15 34253 83 16KP15xKP1704 8 3 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 27.8 2287 0.07 -0.12 32696 84 16KP15xKP1706 8 4 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 36.8 1280 0.04 -0.16 33703 85 16KP15xKP1707 8 5 P. Griffiths Hybrid italica x acephala Hybrid_Diallel 32.9 2328 0.07 -0.07 32655 86 16KP15xKP1713 8 6 P. Griffiths Hybrid viridis x acephala Hybrid_Diallel 45.3 799 0.02 -0.09 34184 87 16KP15xKP1715 8 7 P. Griffiths Hybrid viridis x viridis Hybrid_Diallel 10.0 12708 0.36 0.18 22275

XXIV

Appendix 4.2 cont. Genotypes (n = 129) subject to genotyping by sequencing (GBS) and population structure analysis. Genotype numbers for diallel materials consist of a female (F) and male (M) two-digit designator. Summary genotype descriptors were calculated after SNP filtering (i.e. minor allele frequency < 0.05; minimum depth = 4; maximum missing: 0.75). Indv. Geno Breeder/ Market Indv. Indiv # Inbreeding # Name F M Subspecies Breeding Class Freq. # Collection Class Depth Missing Coeff (F) Sites Missing

88 16KP15_P2 8 8 P. Griffiths Collard_Green viridis Inbred 29.6 3129 0.09 0.51 31854 88 16KP15_P1 8 8 P. Griffiths Collard_Green viridis Inbred 26.6 3264 0.09 0.55 31719 88 16KP15_P3 8 8 P. Griffiths Collard_Green viridis Inbred 26.0 3250 0.09 0.57 31733 101 Green_Harmony_F1 Broccoflower botrytis x italica Hybrid_Commercial 27.1 1141 0.03 0.03 33842 102 Millenium_F1 Sakata Broccoli italica Hybrid_Commercial 33.7 2548 0.07 0.71 32435 103 Gypsy_F1 Sakata Broccoli italica Hybrid_Commercial 30.2 1136 0.03 0.24 33847 104 Waltham_29 UNK Broccoli italica OP 21.6 2171 0.06 0.17 32812 105 Bay_Meadows_F1 Syngenta Broccoli italica Hybrid_Commercial 20.5 2447 0.07 0.21 32536 106 Packman_F1 Peto Seed Broccoli italica Hybrid_Commercial 21.7 2748 0.08 0.48 32235 107 De_Cicco Peto Seed Broccoli italica OP 17.0 3653 0.10 0.33 31330 108 Jade_Cross _F1 P. Griffiths Brussels_Green gemmifera Hybrid_Commercial 19.2 2376 0.07 0.04 32607 109 Bedfordshire_Group UNK Brussels_Green gemmifera OP 28.3 966 0.03 -0.14 34017 110 Falstaf Brussels_Red gemmifera OP 15.6 2713 0.08 0.01 32270 111 Redbull Brussels_Red gemmifera OP 16.4 2598 0.07 -0.04 32385 112 Rubine Brussels_Red gemmifera OP 20.3 1738 0.05 0.05 33245 113 Golden_Acre P. Griffiths Cabbage_Green capitata OP 30.1 1019 0.03 -0.02 33964 115 Futurima_F1 Cabbage_Red capitata Hybrid_Commercial 14.2 4475 0.13 0.23 30508 117 RedExpress Cabbage_Red capitata OP 16.2 2941 0.08 0.25 32042 118 Early_Green_Glazed South Africa Cauliflower botrytis OP 19.7 3498 0.10 0.65 31485 119 Fremont_F1 Seminis Cauliflower botrytis Hybrid_Commercial 19.2 3080 0.09 0.44 31903 120 White_Amazing Cauliflower botrytis OP 24.3 2815 0.08 0.69 32168 121 Snow_Crown_F1 Takii Cauliflower botrytis Hybrid_Commercial 23.8 2321 0.07 0.44 32662 122 White_Coral_F1 Evergrow Cauliflower botrytis Hybrid_Commercial 14.5 5271 0.15 0.56 29712 123 Atlantis_F1 Johnnys ChineseBroc alboglabra x italica Hybrid_Commercial 16.2 3602 0.10 0.07 31381 124 16KG088 Phillip Griffiths ChineseBroc alboglabra OP 15.3 4311 0.12 0.16 30672 126 A12DHd D. Keith, JIC ChineseBroc alboglabra OP 16.5 5459 0.16 0.69 29524 127 TopBunch_F1 Collard_Green viridis Hybrid_Commercial 19.5 1598 0.05 -0.03 33385 129 EASeed_Collard EastAfricaSeed Collard_Kenyan viridis OP 29.1 486 0.01 -0.49 34497 130 Gromost_GACollards Gromost Collard_Kenyan viridis OP 25.8 1085 0.03 -0.06 33898 131 SafariSeeds_1000Head SafariSeeds Collard_Kenyan viridis OP 29.3 656 0.02 -0.37 34327 132 SeedCo_SoGeorgia SeedCo Collard_Kenyan viridis OP 36.3 261 0.01 -0.53 34722 133 Simlaw_Collard SimlawSeeds Collard_Kenyan viridis OP 23.4 1063 0.03 -0.48 33920 134 Syova_Mfalme_F1 Syova Collard_Kenyan viridis Hybrid_Commercial 19.6 3978 0.11 -0.03 31005 135 Beira_Johnnys Johnnys Collard_PortTronc viridis Hybrid_Commercial 22.6 1814 0.05 0.11 33169 136 Darkibor_F1 Kale_GreenCurly acephala Hybrid_Commercial 18.5 1589 0.05 -0.07 33394 138 Dwarf_Curled P. Griffiths Kale_GreenCurly acephala OP 23.5 1404 0.04 -0.16 33579 139 BalticRed Adaptive Seeds Kale_RedCurly acephala OP 14.1 3861 0.11 -0.02 31122

XXV

Appendix 4.2 cont. Genotypes (n = 129) subject to genotyping by sequencing (GBS) and population structure analysis. Genotype numbers for diallel materials consist of a female (F) and male (M) two-digit designator. Summary genotype descriptors were calculated after SNP filtering (i.e. minor allele frequency < 0.05; minimum depth = 4; maximum missing: 0.75). Indv. Geno Breeder/ Market Indv. Indiv # Inbreeding # Name F M Subspecies Breeding Class Freq. # Collection Class Depth Missing Coeff (F) Sites Missing

140 CurlyRoja High Mowing Kale_RedCurly acephala OP 13.9 3715 0.11 -0.01 31268 141 Redbor_F1 Seedway Kale_RedCurly acephala Hybrid_Commercial 18.0 2683 0.08 0.52 32300 142 Roja_Curly Kale_RedCurly acephala OP 37.0 495 0.01 -0.09 34488 143 Redbor_F1 Kale_RedCurly acephala Hybrid_Commercial 40.6 661 0.02 0.11 34322 144 Redbor_F1 Kale_RedCurly acephala Hybrid_Commercial 33.4 522 0.01 -0.04 34461 147 BlackMagic Kale_Tuscan acephala OP 16.8 2766 0.08 -0.05 32217 148 NerodiToscano Kale_Tuscan acephala OP 11.2 5953 0.17 -0.05 29030 151 White_Quickstar Johnny's Kohlrabi_Green gongylodes Hybrid_Commercial 26.0 1281 0.04 0.12 33702 152 Kossak_F1 Kohlrabi_Green gongylodes Hybrid_Commercial 20.4 1879 0.05 0.11 33104 153 AzureStar Kohlrabi_Red gongylodes OP 16.0 3092 0.09 0.14 31891 154 Kolibri Kohlrabi_Red gongylodes OP 16.1 2814 0.08 0.07 32169 155 W001_WT Wild Wild_incana incana Wild 19.8 4506 0.13 0.44 30477 156 W002_WT Wild Wild_macrocarpa macrocarpa Wild 15.0 12672 0.36 0.76 22311 157 W003_WT Wild Wild_insularis insularis Wild 21.8 8653 0.25 0.56 26330 158 W004_WT Wild Wild_cretica cretica Wild 21.0 3767 0.11 0.25 31216 159 W005_WT Wild Wild_incana incana Wild 15.7 5758 0.16 0.19 29225

XXVI

Appendix 4.3. Demographics of participating consumers (n = 564) from an online consumer survey of Brassica oleracea half diallel genotypes in the spring of 2019.

XXVII

Appendix 4.4. Mean morphological leaf trait data of n = 36 hybrid and inbred genotypes from the Brassica oleracea half diallel materials evaluated in replicate field environments during the 2019 field season. Genotypes with the top-three highest measurements are bolded within each trait column Visual Lamina Lamina Petiole Lamina Lamina Dry Genotype Volume1 Length Width Width Blister2 Curl3 Matter (%) (cm) (cm) (mm) (1-5) (1-5) (%) 11 60.9 26.2 21.9 10.4 1.0 1.0 10.2 12 58.1 27.7 17.1 8.8 3.8 1.5 11.7 13 53.9 29.6 23.1 10.0 1.2 3.0 12.0 16 53.9 28.8 25.0 11.4 1.3 1.8 11.2 17 66.5 29.2 24.6 11.4 2.0 1.2 10.9 22 41.1 40.0 9.3 10.2 5.0 1.8 11.8 23 47.5 27.5 16.6 8.4 4.8 2.8 13.3 25 49.1 28.6 17.2 9.3 4.3 1.8 12.8 26 45.1 28.1 17.2 9.2 4.0 3.0 12.8 33 48.9 22.7 17.2 8.3 1.0 4.3 13.9 34 49.8 26.2 17.6 8.6 1.0 5.0 13.5 36 40.8 26.1 19.3 8.7 1.2 4.0 12.9 38 59.1 27.8 23.7 10.8 1.5 2.7 12.6 41 54.1 28.4 21.1 10.0 1.0 3.5 12.0 42 50.6 26.1 14.4 8.4 3.8 3.7 12.8 44 55.6 21.8 14.4 7.5 1.2 4.8 13.3 46 53.3 25.7 17.5 8.9 1.0 4.5 13.6 47 53.2 30.3 19.9 9.6 1.8 3.3 12.8 51 54.6 31.2 23.7 11.5 1.2 1.2 10.9 53 53.1 25.3 21.9 9.8 3.3 3.0 12.5 54 54.5 26.6 21.1 10.5 2.3 3.2 12.7 55 46.2 27.4 19.2 9.5 1.2 1.2 11.5 56 45.9 27.0 23.3 10.5 1.8 2.0 12.2 57 50.5 28.5 20.6 10.7 1.2 1.0 11.2 66 52.0 21.7 15.9 6.5 1.0 2.3 12.9 72 51.6 35.4 16.1 10.3 4.2 1.5 12.3 73 58.3 31.3 21.6 9.7 1.7 2.7 12.8 76 59.5 32.4 22.1 10.3 1.5 2.0 12.0 77 51.3 33.3 18.7 11.0 1.3 1.0 11.5 81 61.9 28.1 24.3 11.5 1.2 1.0 10.9 82 58.6 29.7 17.5 9.7 3.5 1.0 12.7 84 55.1 24.9 19.1 10.0 1.2 3.3 12.3 85 55.8 27.2 24.1 11.8 1.0 1.0 11.3 86 53.8 28.5 22.7 10.7 1.2 2.2 12.4 87 60.4 31.7 24.2 11.7 1.2 1.2 11.3 88 60.9 24.8 22.4 10.8 1.2 1.0 11.6 1Visual Volume = percentage of each image occupied by the leaf sample 2Lamina Blister scale: 1 = none/flat lamina surface; 5 = extremely blistered or bullate lamina surface 3Lamina Curl scale: 1 = none/entire leaf margin; 5 = extremely curled leaf margin

XXVIII

Appendix 4.5. Mean glucosinolate concentration (mg/g FW) in leaves of n = 36 hybrid and inbred genotypes from the Brassica oleracea half-diallel. Leaves were harvested from replicated plots (n = 3) during the September 2019 field trials in Freeville, NY and subject to HPLC quantification. Genotypes with the top-three highest measurements are bolded within each trait column. Neo- Total Genotype Glucoiberin Glucoprogoitrin Glucoepiprogoitrin Sinigrin Glucoraphanin Glucobarbarin Glucobrassicin Gluconasturtiin glucobrassicin GS 11 0.254 17.441 1.072 10.781 0.366 0.526 34.968 2.039 1.082 68.530 12 0.341 5.509 0.738 15.643 1.650 0.741 32.366 1.296 0.939 59.221 13 0.366 10.795 0.831 5.562 0.310 0.213 12.353 1.773 0.799 33.003 16 0.243 10.163 0.986 3.231 0.287 0.537 24.766 1.418 0.615 42.246 17 0.279 22.091 0.993 11.167 0.839 0.758 43.003 5.014 1.815 85.959 22 0.376 0.760 0.096 2.509 0.235 0.431 11.035 0.443 0.833 16.718 23 0.256 2.975 0.636 5.279 0.391 0.306 13.297 0.911 0.860 24.912 25 0.217 0.621 0.242 4.242 0.000 0.055 10.869 1.238 1.460 18.945 26 0.160 1.348 0.447 2.372 0.106 0.307 7.208 0.308 0.439 12.696 33 0.300 4.247 0.770 1.494 0.112 0.305 6.555 1.885 1.324 16.992 34 0.346 4.308 0.808 1.631 0.088 0.315 5.487 1.104 0.902 14.989 36 0.291 6.705 0.809 3.458 0.187 0.259 11.153 2.050 0.872 25.785 38 0.332 10.231 0.781 5.057 0.309 0.342 13.987 1.998 0.908 33.945 41 0.316 12.285 1.114 5.590 0.232 0.340 12.767 1.242 1.676 35.561 42 0.334 1.655 0.654 3.130 0.103 0.522 7.488 0.539 0.498 14.924 44 0.465 3.276 1.238 3.481 0.155 0.369 15.892 1.659 5.614 32.148 46 0.342 5.638 0.946 1.931 0.209 0.401 21.025 1.138 1.353 32.983 47 0.219 9.228 0.912 5.242 0.309 0.335 18.555 1.967 1.613 38.381 51 0.296 2.451 1.230 9.887 0.680 0.471 9.726 1.969 1.257 27.967 53 0.292 1.894 1.232 7.787 0.437 0.319 13.220 2.454 1.253 28.888 54 0.259 2.155 1.134 3.268 0.651 0.061 9.551 1.183 1.013 19.274 55 0.273 1.154 0.337 2.211 0.701 0.389 21.411 4.494 0.844 31.814 56 0.269 1.188 0.457 2.702 0.214 0.533 17.241 1.753 0.851 25.208 57 0.217 2.568 0.811 6.869 0.984 0.298 20.598 4.572 3.166 40.083 66 0.331 1.185 0.844 0.305 0.000 0.478 24.044 0.778 0.391 28.357 72 0.281 5.946 0.868 8.593 2.479 0.985 65.265 2.822 0.991 88.231 73 0.246 7.007 4.008 7.075 0.521 0.306 19.659 3.073 0.781 42.677 76 0.253 9.031 1.201 3.709 0.447 0.478 33.800 2.991 0.480 52.389 77 0.302 14.030 1.033 10.796 0.752 0.779 43.356 7.256 1.379 79.683 81 0.333 27.671 1.098 7.935 0.405 0.513 40.746 2.809 0.866 82.377 82 0.227 4.303 0.779 8.290 0.167 0.388 20.647 0.988 0.492 36.281 84 0.274 10.151 0.896 5.079 0.271 0.431 18.022 1.845 0.993 37.962 85 0.297 2.804 0.994 7.819 0.174 0.409 17.999 1.955 1.633 34.085 86 0.204 6.444 0.862 1.913 0.275 0.436 26.909 1.539 0.591 39.172 87 0.256 22.394 0.987 6.565 0.689 0.501 30.232 3.798 0.925 66.347 88 0.367 9.945 1.091 3.615 0.249 0.569 28.338 2.476 0.602 47.253

XXIX

Appendix 4.6. Mean carotenoid concentration (mg/g FW) in leaves of n = 36 hybrid and inbred genotypes from the Brassica oleracea half-diallel. Leaves were harvested from replicated plots (n = 3) during the September 2019 field trials in Freeville, NY and subject to HPLC quantification. Genotypes with the top-three highest measurements are bolded within each trait column. Alpha Beta Total Total Genotype Neoxanthin Violaxanthin Antheraxanthin Lutein Zeaxanthin Phytoene Carotene Carotene Carotenoids Chlorophyll1 11 0.024 0.037 0.005 0.049 0.002 0.001 0.033 0.024 0.175 0.716 12 0.040 0.046 0.011 0.121 0.004 0.002 0.071 0.097 0.392 1.149 13 0.028 0.017 0.007 0.055 0.003 0.002 0.032 0.047 0.190 0.509 16 0.021 0.018 0.007 0.058 0.002 0.000 0.034 0.034 0.174 0.808 17 0.027 0.031 0.022 0.063 0.002 0.002 0.044 0.035 0.225 0.691 22 0.116 0.154 0.030 0.405 0.011 0.003 0.196 0.137 1.053 4.464 23 0.040 0.076 0.014 0.184 0.004 0.014 0.105 0.095 0.531 1.666 25 0.086 0.103 0.022 0.232 0.007 0.003 0.127 0.066 0.646 2.963 26 0.073 0.151 0.025 0.282 0.011 0.003 0.173 0.100 0.818 3.287 33 0.029 0.031 0.008 0.101 0.002 0.002 0.051 0.058 0.281 0.977 34 0.037 0.055 0.010 0.127 0.003 0.003 0.078 0.045 0.357 1.523 36 0.028 0.038 0.008 0.100 0.003 0.002 0.054 0.058 0.292 1.048 38 0.027 0.038 0.010 0.103 0.004 0.002 0.060 0.057 0.303 1.053 41 0.015 0.022 0.006 0.059 0.003 0.002 0.042 0.061 0.209 0.504 42 0.049 0.090 0.011 0.164 0.003 0.003 0.100 0.069 0.490 1.966 44 0.031 0.045 0.008 0.106 0.003 0.002 0.074 0.060 0.330 1.273 46 0.044 0.061 0.015 0.127 0.005 0.008 0.059 0.049 0.367 1.607 47 0.026 0.034 0.008 0.077 0.007 0.002 0.052 0.063 0.269 0.788 51 0.035 0.032 0.014 0.118 0.005 0.003 0.086 0.072 0.365 1.377 53 0.058 0.066 0.016 0.171 0.003 0.003 0.105 0.061 0.482 2.057 54 0.058 0.070 0.013 0.152 0.006 0.002 0.103 0.088 0.494 1.772 55 0.039 0.039 0.010 0.149 0.004 0.003 0.091 0.068 0.402 1.926 56 0.071 0.123 0.015 0.203 0.007 0.002 0.126 0.037 0.586 2.760 57 0.056 0.070 0.012 0.159 0.004 0.003 0.104 0.053 0.460 2.301 66 0.038 0.068 0.018 0.158 0.006 0.002 0.088 0.124 0.503 1.370 72 0.045 0.080 0.016 0.130 0.003 0.002 0.066 0.059 0.400 1.511 73 0.024 0.026 0.008 0.068 0.002 0.004 0.037 0.029 0.199 0.844 76 0.097 0.098 0.020 0.269 0.008 0.026 0.145 0.095 0.758 3.210 77 0.017 0.034 0.009 0.070 0.003 0.002 0.044 0.043 0.221 0.697 81 0.011 0.026 0.008 0.068 0.003 0.002 0.043 0.042 0.203 0.669 82 0.030 0.041 0.009 0.175 0.007 0.009 0.092 0.098 0.461 1.688 84 0.048 0.071 0.022 0.148 0.002 0.002 0.082 0.062 0.437 1.851 85 0.029 0.017 0.007 0.065 0.002 0.002 0.038 0.047 0.207 0.628 86 0.043 0.052 0.010 0.120 0.003 0.002 0.074 0.049 0.354 1.464 87 0.024 0.038 0.006 0.085 0.002 0.001 0.047 0.050 0.253 1.089 88 0.014 0.020 0.007 0.046 0.001 0.001 0.025 0.029 0.144 0.560 1Total Chlorophyll = sum of independently determined Chlorophyll a and Chlorophyll b concentrations

XXX

Appendix 4.7. General combining ability (GCA) random effect predictors of eight B. oleracea inbred parental genotypes used in the development of a diallel mating design. Liking ~ Parental Inbred Liking ~ 1 Parental Inbred Familiarity #7 - Elongated Collard 0.250 #1 - Yellow Collard 0.171 #8 - Georgia Collard 0.205 #8 - Georgia Collard 0.140 #1 - Yellow Collard 0.202 #7 - Elongated Collard 0.126 #4 - Curly Green Kale 0.153 #6 - Deep Purple Kale 0.045 #3 - Curly Red Kale 0.039 #3 - Curly Red Kale -0.011 #6 - Deep Purple Kale -0.140 #4 - Curly Green Kale -0.060 #5 - Jagged Broccoli -0.285 #5 - Jagged Broccoli -0.185 #2 – Tuscan Kale -0.424 #2 – Tuscan Kale -0.225

XXXI

Appendix 4.8. Specific combining ability (SCA) random effect predictors for n = 35 genotypes from the Brassica oleracea half diallel. Liking ~ 1 Liking ~ Familiarity Genotype SCA Genotype SCA 42 0.525 23 0.237 23 0.386 26 0.205 26 0.384 42 0.186 53 0.246 13 0.176 38 0.208 38 0.138 82 0.204 73 0.103 54 0.153 84 0.090 12 0.132 53 0.078 13 0.107 12 0.068 72 0.100 54 0.053 25 0.089 11 0.043 46 0.057 77 0.040 73 0.035 46 0.031 84 0.018 85 0.021 77 0.015 82 0.012 11 0.014 88 0.011 85 -0.022 66 -0.006 41 -0.032 86 -0.018 44 -0.038 41 -0.020 88 -0.055 16 -0.023 56 -0.064 51 -0.030 87 -0.064 44 -0.031 33 -0.094 72 -0.042 76 -0.097 55 -0.048 86 -0.102 87 -0.052 16 -0.108 33 -0.055 47 -0.109 25 -0.062 66 -0.133 47 -0.073 55 -0.142 22 -0.083 51 -0.155 17 -0.084 81 -0.191 81 -0.090 22 -0.230 76 -0.116 17 -0.288 34 -0.169 34 -0.345 57 -0.177 57 -0.416 56 -0.200

XXXII

Appendix 4.9. Random effect variance components from mixed model analysis of square root transformed carotenoid concentrations. Random effect variances due to 2 2 2 general combining ability (𝜎GCA ), specific combining ability (𝜎SCA ), block (𝜎block ), 2 2 and random error (𝜎 ) are reported. Percentage of total random variance (% 𝜎total ) provides an indication of the effect’s contribution to the overall random effect variance. Violaxanthin Neoxanthin 2 2 2 2 σ se % 𝜎total σ se % 𝜎total 2 -04 -04 -04 -04 𝜎GCA 5.77x10 3.8x10 11.3% 3.99x10 2.70x10 10.5% 2 -04 -04 -04 -04 𝜎SCA 8.06x10 6.9x10 15.7% 5.32x10 5.29x10 14.0% 2 -04 -04 -06 -04 𝜎block 2.45x10 3.5x10 4.8% 2.59x10 8.60x10 0.1% 2 -03 -04 -03 -04 𝜎 3.49x10 5.9x10 68.2% 2.87x10 4.86x10 75.5%

Lutein Antheraxanthin 2 2 2 2 σ se % 𝜎total σ se % 𝜎total 2 -03 -03 -05 -05 𝜎GCA 1.76x10 1.1x10 21.2% 5.95x10 4.39x10 5.3% 2 -03 -03 -05 -04 𝜎SCA 1.53x10 1.1x10 18.4% 4.22x10 1.16x10 3.7% 2 -04 -04 -04 -04 𝜎block 1.64x10 3.0x10 2.0% 1.78x10 2.02x10 15.8% 2 -03 -04 -04 -04 𝜎 4.85x10 8.3x10 58.4% 8.48x10 1.40x10 75.2%

Phytoene Zeaxanthin 2 2 2 2 σ se % 𝜎total σ se % 𝜎total 2 -05 -05 -05 𝜎GCA 0.00 1.5x10 0.0% 3.82x10 2.47x10 8.9% 2 -05 -04 -05 𝜎SCA 3.20x10 1.2x10 2.6% 0.00 4.39x10 0.0% 2 -04 -04 -05 -05 𝜎block 1.76x10 2.1x10 14.2% 1.06x10 2.16x10 2.5% 2 -03 -04 -04 -05 𝜎 1.03x10 1.7x10 83.2% 3.80x10 6.20x10 88.6%

Alpha Carotene Beta Carotene 2 2 2 2 σ se % 𝜎total σ se % 𝜎total 2 -04 -04 -04 -04 𝜎GCA 8.10x10 5.1x10 15.7% 2.67x10 1.90x10 5.7% 2 -03 -04 -04 𝜎SCA 1.09x10 7.2x10 21.0% 0.00 4.62x10 0.0% 2 -04 -04 -04 -04 𝜎block 2.10x10 3.0x10 4.1% 3.31x10 4.48x10 7.1% 2 -03 -04 -03 -04 𝜎 3.06x10 5.x 10 59.2% 4.05x10 6.59x10 87.1%

Total Chlorophyll (A + B) Total Carotenoids 2 2 2 2 σ se % 𝜎total σ se % 𝜎total 2 -02 -02 -03 -03 𝜎GCA 1.87 x10 1.2x10 14.0% 4.03x10 2.43x10 21.0% 2 -02 -02 -03 -03 𝜎SCA 2.07x10 1.8x10 15.5% 3.27x10 2.53x10 17.0% 2 -03 -04 𝜎block 0.00 2.7x10 0.0% 0.00 3.48x10 0.0% 2 -02 -02 -02 -03 𝜎 9.42x10 1.6x10 70.5% 1.19x10 2.04x10 62.1%

XXXIII

Appendix 4.10. Random effect variance components from mixed model analysis of square root transformed glucosinolate concentrations. Random effect variances due to 2 2 2 general combining ability (𝜎GCA ), specific combining ability (𝜎SCA ), block (𝜎block ), 2 2 and random error (𝜎 ) are reported. Percentage of total random variance (% 𝜎total ) provides an indication of the effect’s contribution to the overall random effect variance. Glucoiberin Sinigrin 2 2 2 2 σ se % 𝜎total σ se % 𝜎total 2 -04 -02 -02 𝜎GCA 0.00 1.59x10 0.0% 9.06x10 5.71x10 15.7% 2 -04 -03 -02 -02 𝜎SCA 3.17x10 1.22x10 2.6% 9.95x10 8.04x10 17.3% 2 -03 -03 -03 -02 𝜎block 1.10x10 1.41x10 9.0% 4.03x10 1.54x10 0.7% 2 -02 -03 -02 𝜎 1.08x10 1.76x10 88.4% 0.382 6.59x10 66.3%

Glucoprogoitrin Glucoepiprogoitrin 2 2 2 2 σ se % 𝜎total σ se % 𝜎total 2 -03 -03 𝜎GCA 0.264 0.149 28.1% 8.12x10 5.85x10 7.0% 2 -02 -02 -03 -02 𝜎SCA 2.23x10 8.45x10 2.4% 5.99x10 1.45x10 5.2% 2 -02 -02 -03 𝜎block 4.44x10 6.23x10 4.7% 0.00 3.02x10 0.0% 2 -02 𝜎 0.611 0.103 64.9% 0.102 1.71x10 87.8%

Glucoraphanin Glucobrassicin 2 2 2 2 σ se % 𝜎total σ se % 𝜎total 2 -03 -03 𝜎GCA 5.83x10 6.13x10 4.0% 0.199 0.133 10.3% 2 -02 -02 𝜎SCA 4.36x10 2.49x10 30.2% 0.251 0.255 13.0% 2 -03 𝜎block 0.00 2.83x10 0.0% 0.126 0.166 6.5% 2 -02 -02 𝜎 9.49x10 1.64x10 65.7% 1.36 0.234 70.3%

NeoGlucobrassicin Glucobarbarin 2 2 2 2 σ se % 𝜎total σ se % 𝜎total 2 -03 -03 -04 -03 𝜎GCA 5.51x10 5.86x10 2.1% 1.76x10 1.38x10 0.3% 2 -02 -02 -03 𝜎SCA 0.00 2.61x10 0.0% 1.63x10 9.93x10 27.1% 2 -02 -02 -03 -03 𝜎block 1.68x10 2.37x10 6.6% 1.71x10 2.95x10 2.8% 2 -02 -02 -03 𝜎 0.235 3.83x10 91.3% 4.21x10 7.21x10 69.8%

Gluconasturtiin Total Glucosinolates 2 2 2 2 σ se % 𝜎total σ se % 𝜎total 2 -02 -02 𝜎GCA 4.47x10 2.52x10 33.7% 0.418 0.243 17.3% 2 -03 -02 𝜎SCA 8.17x10 1.29x10 6.2% 0.00 0.218 0.0% 2 -03 𝜎block 0.00 2.37x10 0.0% 0.161 .0215 6.7% 2 -02 -02 𝜎 7.97x10 1.36x10 60.1% 1.83 .303 76.0%

XXXIV

Appendix 4.11. Distribution of SNP markers across each chromosome (n = 9) in diverse Brassica oleracea cultivars (n = 124) and related wild species (n = 5). Marker numbers are depicted in bins of 2Mb.

XXXV