DISSECTING VARIATION IN TOMATO FRUIT COLOR QUALITY THROUGH DIGITAL PHENOTYPING AND GENETIC MAPPING DISSERTATION

Transcription

1 DISSECTING VARIATION IN TOMATO FRUIT COLOR QUALITY THROUGH DIGITAL PHENOTYPING AND GENETIC MAPPING DISSERTATION Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University By Audrey Darrigues, M.S. ***** The Ohio State University 2007 Dissertation Committee: Dr. David M. Francis, Adviser Dr. Mark A. Bennett Dr. Clay H. Sneller Dr. Steven J. Schwartz Approved by Adviser Horticulture & Crop Science Graduate Program

2 ABSTRACT Color is among the most important attributes of tomatoes for processing. Both color and color uniformity are affected by yellow shoulder disorder (YSD), a ripening disorder that results in discoloration of the proximal end tissues of the fruit. Cells from YSD tissue are smaller and more randomly organized, and the development of the chromoplast is altered. We show juice from YSD-affected tomato had 13-24% significantly less lycopene relative to juice from non-affected tomato. Beta-carotene content was reduced by 4-8% in juice from YSD-affected tomato, although this reduction was not statistically significant. Quantification of lycopene and betacarotene concentration in tomato juice samples was more precise by increasing biological replications rather than analytical replications. Variance partitioning suggests that YSD incidence and severity are affected by both genetics and environment. In order to assess genetic contributions to YSD, a color measurement module in the Tomato Analyzer software was tested to accurately quantify color and color uniformity from digital images. This approach improved the efficiency of collecting data, provided high correlations with data collected by colorimeter, and improved estimates of genetic contributions to color uniformity. We hypothesize that with increased precision and accuracy in measuring color, sampling strategies ii

3 for higher carotenoid content can be optimized and the genetic control underlying color and color uniformity in tomato can be uncovered. To elucidate the genetic basis of YSD, molecular markers were exploited for application in breeding populations. An advanced backcross population (BC2) derived from Solanum lycopersicum (OH88119) x S. pimpinellifolium (PI128216) was evaluated for color (L*, a*, b*, hue and chroma) and color uniformity (%YSD and %RED). Indices were developed to circumvent the highly correlated traits and to simplify the trait complexity based on principal component analysis. These indices capture the essential features of color intensity and color uniformity. The BC2 population was genotyped with 70 polymorphic markers for marker-trait analysis. The population was selfed through four generations to generate an inbred backcross population (BC2S4), which was evaluated for the same traits and genotyped with the same markers as the BC2 population. An F2 population and elite processing varieties were evaluated for color as confirmation of the marker-trait associations. We found QTL for color intensity on chromosomes 2, 8, and 9, and a QTL for color uniformity on chromosome 6. Positive gain under selection was realized for selection by phenotype and by marker-assisted selection (MAS). Higher gains were realized from MAS. Directional selection strategies are being used to further characterize these QTL and evaluate genetic correlations to other fruit quality traits, disease resistance, and yield. iii

4 À ma famille. À la mémoire de Papa iv

5 ACKNOWLEDGMENTS I thank my advisor, David Francis, for his guidance and intellectual diligence throughout my academic program. I thank the members of my Student Advisory Committee, Mark Bennett, Clay Sneller and Steve Schwartz, for their thoughtful comments, suggestions and revisions of my research plans, results, and manuscripts. I thank post-doctoral researchers working in David Francis lab: Wencai Yang, for his patience while providing me with training in the molecular lab and for his work on the marker development used to genotype my populations; Alba Clivati-McIntyre, for her expertise in soil science that brought understanding in soil fertility management and YSD; Matthew Robbins and Sung-Chur Sim, for their work on marker development to further genotype my populations. I thank Susana De Jesus (MS 2005) for her endless enthusiasm in the lab and beyond. Last but not least, I thank Troy Aldrich for his assistance with greenhouse and field plot establishments, seed extractions, and analyses in the quality lab. I acknowledge the Ohio Agricultural Research and Development Center (OARDC) Research Enhancement Competitive Grant for funding parts of my research. Carotenoid extractions and HPLC analyses were conducted in Steve Schwartz lab at the OSU Food Science and Technology Department. I thank v

6 Marjory Renita and Rachel Kopec for giving my initial training in extracting and quantifying carotenoids. I thank Bert Bishop from the OARDC Computing and Statistical Services for his assistance and guidance in analyzing my data. I am grateful to my family for their endless encouragement and support, especially when I needed it the most. Dad provided me with much inspiration in the realm of plant breeding, a field that inspired him for his entire career. I will stand proud in his honor, in his memory. Mom gave me endless support, strength and motivation despite the personal hardship sustained while at OSU. I will also stand proud and carry on her legacy. I am also grateful to my sisters, Carole and Sophie, and their families, and to my brother, Pierre-Emmanuel. Je vous remercie de tout mon coeur... I thank my friends. You know who you are and I wish for our prolonged friendship that we ve built over the years. Somewhere in this world, we shall meet again vi

7 VITA B.S. Biology, Iowa State University M.S. Plant Breeding, Iowa State University present.... Graduate Research Associate, The Ohio State University. PUBLICATIONS Darrigues, A., A. Clivati-McIntyre, S. Schwartz and D.M. Francis Increasing the carotenoid content of tomato by managing variety choice and soil fertility for color and color uniformity. Acta Horticulturae 744: Darrigues, A., K.R. Lamkey, M.P. Scott Breeding for grain amino acid composition in maize IN Plant breeding : the Arnel R. Hallauer International Symposium (2003: Mexico City, Mexico), edited by K. R. Lamkey and M. Lee, chapter 24. Ames, Iowa: Blackwell Publishers. Darrigues, A., C. Buffard, K.R. Lamkey, M.P. Scott Variability and genetic effects for tryptophan and methionine in commercial maize germplasm. Maydica. 50: vii

8 FIELDS OF STUDY Major Field: Horticulture and Crop Science viii

9 TABLE OF CONTENTS Page Dedication...iv Acknowledgments....v Vita. vii List of Tables.. xii List of Figures xiv CHAPTERS 1. Introduction Objectives...2 Tomato breeding & genetics Genes and alleles affecting color Population genetics and improvement Marker development Marker application Caveats of QTL analysis...10 Marker-assisted selection...12 Characteristics of tomato fruit quality...14 Yellow shoulder disorder Health-promoting carotenoids in tomato...15 Objective trait measurement...17 Measuring color and color uniformity Quantifying carotenoids Rationale and significance of research...19 ix

10 2. Optimizing sampling of tomato fruit for carotenoid content with application to assessing the impact of ripening disorders Abstract. 21 Introduction...22 Material and Methods...24 Plant material...24 Trait evaluation: carotenoid quantification..25 Digital phenotyping for YSD...26 Statistical analysis Results...28 Variance in lycopene and β-carotene Effect of YSD on lycopene and β-carotene Relationship between the extent of YSD and lycopene content..30 Discussion...31 Acknowledgments Tomato Analyzer Color Test: a new tool for efficient digital phenotyping Abstract...39 Introduction...40 Material and Methods...44 Software implementation...44 Tomato Analyzer Color Test...44 Customizing Tomato Analyzer Color Test...47 Tomato Analyzer Color Test versus colorimeter...47 Plant material...48 Phenotypic data collection...49 Statistical analysis...50 Results...51 Correlation between methods...51 Calibration...52 Variance partitioning with Tomato Analyzer Color Test...52 Application of Tomato Analyzer Color Test to other crops...54 Discussion...56 Acknowledgments...59 x

11 4. Dissecting variation in tomato fruit color quality through digital phenotyping and mapping Abstract...68 Introduction...69 Material and Methods...72 Plant material...72 Field trials...73 Digital phenotyping for color...74 DNA isolation and marker analysis...75 Statistical analysis...77 Marker-trait associations...79 Gain under selection and heritability...81 Results...83 Phenotypic data analysis...83 Genotypic analysis...88 Marker-trait associations with color in the BC2 and BC2S4 populations...89 Non-parametric analysis (NPA)...92 Marker-trait associations with color in the TC19F2 population...93 Association with color intensity in processing varieties...93 Gain under selection and heritability...94 Discussion...95 Merits of selection indices...96 QTL discovery and confirmation...97 Color and carotenoids...99 Utility of QTL in elite germplasm Acknowledgments CONCLUSIONS..116 APPENDIX A. Tomato Analyzer Color Test: user manual B. The PI IBC population C. Marker-trait associations in the BC2 and BC2S4 populations REFERENCES xi

12 LIST OF TABLES Table Page 2.1 Optimizing sampling for lycopene and β carotene content in juice samples of tomato Effect of yellow shoulder disorder (YSD) on lycopene and β-carotene content in juice samples of tomato over two years Means of lycopene and β carotene content in juice made from tomatoes affected by yellow shoulder disorder (YSD) and not affected by YSD (nonysd) Correlation values and regression characteristics of L*, a*, b* values for standard color plates obtained with Tomato Analyzer Color Test and colorimeter Proportion of variance estimates for color measurements using Tomato Analyzer Color Test with the TA-defined boundaries (minimum blue value = 30; TA_Unadj) and adjusted boundaries (TA_Adj) Average values of color parameters obtained from the output of Tomato Analyzer Color Test for a variety of fruit and vegetable. Images of these crops appear on Figure Polymorphic molecular markers tested in the populations derived from the cross Solanum lycopersicum (OH88119) x S. pimpinellifolium (PI128216) Pearson correlations by population between replications within location in Ohio (BC2, Fremont) and between locations (BC2 and BC2S4, Fremont vs. Wooster) for color and color uniformity xii

13 4.3 Pearson correlations between all color and color uniformity traits by population (BC2 and BC2S4) and by location (Fremont and Wooster, Ohio) Principal component analysis (PCA) of color and color uniformity traits in the BC2, BC2S4 and TC19F2 populations evaluated in Fremont and/or Wooster, Ohio Marker-trait associations for color and color uniformity detected in both BC2 and BC2S4 populations derived from Solanum lycopersicum (OH88119) x S. pimpinellifolium (PI128216) Non-parametric analysis (NPA) for markers associated with indices designed to capture color and color uniformity for the BC2 and BC2S4 populations evaluated in Fremont and Wooster, Ohio A.1 Munsell notations for the custom 28-patch color checker used at OSU A.2 L*, a*, b* values obtained from a colorimeter for each patch of the color checker. The illuminant is C and the observer angle 2 o B.1 Marker data of the PI IBC lines C.1 Marker-trait associations for color traits in the BC2 population C.2 Marker-trait associations for color traits in the BC2S4 population C.3 Marker-trait associations for the indices icolorint, icoloruni and icoloropt in the BC2 population C.4 Marker-trait associations for the indices icolorint, icoloruni and icoloropt in the BC2S4 population C.5 Marker-trait associations for color traits and indices detected via nonparametric analysis in the BC2 and BC2S4 populations xiii

14 LIST OF FIGURES Figure Page 2.1 Relationship between the extent of yellow shoulder disorder (YSD) in fruit and lycopene content in tomato juice Tomato Analyzer and its Color Test. The dialog box in the center of the image allows the user to customize the color parameters for analysis (top tier) and to enter the correction values for calibrating the scanner, as well as other options (bottom tier) Representation of the tomato proximal end (shoulder) analyzed for color with Tomato Analyzer Color Test. Images C and D show symptoms of yellow shoulder disorder (YSD), a ripening disorder that affects color uniformity. A, Uniform fruit analyzed with TA_Unadj method. B, Uniform fruit analyzed with TA_Adj. C, YSD-affected fruit analyzed with TA_Unadj. D, YSD-affected fruit analyzed with TA_Adj Correlation between Tomato Analyzer Color Test and colorimeter values for L*, a*, b* of the CIELab color space using 247 standard color plates Regression of different scanners for L*, a*, b* values of the CIELab color space from standard color plates spanning a range of colors observed in tomato fruits measured with Tomato Analyzer Color Test and a colorimeter. The scanners used to assess scanning quality for color were HP ScanJet 3970 (asterisk), HP ScanJet 5300C (triangle) and Microtek 6000 (square). The regression values are summarized in Table Images of various fruits and vegetables evaluated with Tomato Analyzer Color Test for objective color measurements. A, Red-skinned potato. B, Cucumber. C, Red plum. D, Cantaloupe. E, Carrot. F, Strawberry...67 xiv

15 4.1 Frequency histograms for color and color uniformity of the BC2 population. Tomato fruits were evaluated in Fremont and Wooster, Ohio, in Each bin represents one standard deviation. Dashed bars correspond to the trait distribution in Fremont (F), whereas solid bars correspond to Wooster (W). The population mean (X) is included in the legend. The asterisk corresponds to the bin position of the recurrent parent, OH88119, used in the development of the advanced backcross population Frequency histograms for color and color uniformity of the BC2S4 population. Tomato fruits were evaluated in Fremont and Wooster, Ohio, in Each bin represents one standard deviation. Dashed bars correspond to the trait distribution in Fremont (F), whereas solid bars correspond to Wooster (W). The population mean (X) is included in the legend. The asterisk corresponds to the bin position of the recurrent parent, OH88119, used in the development of the advanced backcross population Visual inspection of selection for color and color uniformity based on three indices: icoloruni for uniformity, icolorint for intensity, and icoloropt for optimal color. The range of values is based on the evaluation of the BC2S4 population in Fremont (F) and Wooster (W), Ohio, in The column on the left shows the genotype with the best index value and the right with the worst value. The sign preceding the index names indicates the direction of the index value for best or worst color Approximate map position of markers used in this study (underlined). Regions associated with color intensity and uniformity are delimited with solid black vertical line. Markers noted below a chromosome have been mapped to that chromosome but the relative position remains uncertain. The map position of genes involved in the carotenoid biosynthetic and phytochrome signaling pathways is delimited by a dashed vertical line (hp-1, high pigment 1; dg, dark green; hp-2, high pigment 2 (non-allelic to hp-1); r, yellow flesh; B, beta; og c, old gold crimson (allelic to B); t, tangerine; Del, Delta) A.1 Tomato Analyzer Color Test application A.2 Representation of the CIELab color space A.3 Representation of hue and chroma, two attributes of perceived color A.4 28-patch color checker from X-Rite xv

16 A.5 Tomato Analyzer Color Test window. Correction values used to calibrate the scanner are entered in the lower panel of the dialog box xvi

17 CHAPTER 1 INTRODUCTION There is often a conflict between the goals of genetic research and the goals of plant breeding. A geneticist will develop structured populations that seek to maximize genetic differences and trait values. These populations often involve best by worst crosses. In contrast, a plant breeder will seek to work within adapted germplasm, and will tend to develop populations based on only the best germplasm ( best by best crosses). There are three trends in plant genetic research that will help bridge the gap between the goals of genetic and crop improvement research: population structures that favor simultaneous genetic analysis and breeding progress, the development of statistical methods to map genes in unstructured populations, and molecular marker technology that dramatically alters our approach to genetic characterization. 1

18 OBJECTIVES The overall objective of this dissertation research was to apply new approaches to the genetic dissection and improvement of a major quality limitation in the production of tomatoes (Solanum lycopersicum). Yellow shoulder disorder (YSD) is a blotchy ripening disorder that is characterized by discolored regions under the epidermis of mature fruits (Francis et al, 2000). The central hypothesis was that the genetic component of YSD can be dissected through digital phenotyping and the application of molecular markers to structured populations to facilitate both genetic analyses and plant breeding goals. The specific objectives of the research were: 1. To optimize sampling of tomato fruit for carotenoid content and assess the impact of YSD on health-promoting carotenoids; 2. To implement Tomato Analyzer Color Test, a tool developed to collect objective color measurements from digital images, and to determine estimates of genotypic variances associated with color and color uniformity. 3. To elucidate the genetic component of color and color uniformity via quantitative trait loci (QTL) mapping in populations derived from Solanum lycopersicum and S. pimpinellifolium. 2

19 TOMATO BREEDING AND GENETICS Genes and alleles affecting color A number of monogenic variants have been characterized that affect color in the tomato fruit. Notable genes and alleles are beta-carotene (B) and its allele old gold crimson (og c ), high pigment -1 (hp-1), high pigment-2 (hp-2) and its allele dark green (dg), delta (Del), yellow flesh (r), and tangerine (t), among others (reviewed by Stevens and Rick, 1986). An early carotenoid profile was published demonstrating the effect of these mutants not only on the color but also on the carotenoid content of the tomato fruit (Tomes, 1963). The variants og c, hp-1, hp-2, t, and B give the most notable effects on the lycopene and β-carotene contents of the tomato fruit. Genes involved in the accumulation of carotenoids via the carotenoid biosynthetic pathway or phytochrome signaling pathway have been well characterized by the enzymes they encode and by their genetic map position. On chromosome 2, the recessive mutation of the hp-1 gene results in accumulation of both lycopene and β-carotene (Yen et al, 1997). The non-allelic hp-2 gene has a similar phenotype to hp-1 but maps to chromosome 1 (Soressi, 1975; van Tuienen et al, 1997). The dg mutant is allelic to hp-2 and results in elevated levels of carotenoids and flavonoids in mature ripe fruits (Levin et al, 2003; Wann et al, 1985). On chromosome 6, B was characterized by increased β-carotene content due to enhanced transcription of lycopene-β-cyclase, the enzyme that converts lycopene to β-carotene (Lincoln and Porter, 1950; Ronen et al, 2000). Two recessive allelic 3

20 mutations of B, old-gold (og) and og c, have been characterized as null mutants of B, which results in increased lycopene content (Thompson et al, 1965; Ronen et al, 2000). On chromosome 10, the gene that encodes the recessive mutation t, CRTISO, was mapped and, later, cloned. Fruits of t mutants accumulate tetra-cis-lycopene (i.e. pro-lycopene) instead of all-trans-lycopene and are orange in color (Isaacson et al, 2002). On chromosome 12, Del results in an increased level of δ-carotene at the expense of lycopene. The mutation Del is due to over-expression of CrtL-e, the enzyme for lycopene-ε-cyclase (Ronen et al, 1999). Population genetics and improvement The structures of population determine their effectiveness for deciphering the genetic component of a trait and their utility in crop improvement. Traditional population structures for genetic studies require parents that differ widely in the trait, with the effectiveness of a population determined by the separation of parents (Fisher, 1918). For breeding purposes, we seek to maintain combinations of favorable genes and best by best crosses are often made without regard to genetic or phenotypic differences. Population structures that are used for genetic studies very often have a standard structure. For example, populations such as F 2, backcross (BC), and recombinant inbred lines (RIL) are common. These populations have several advantages that include a well established theoretical basis for genetic inference (Fisher, 1918), readily available computational tools (e.g. MapMaker, 4

21 Lander et al, 1987; JoinMap, van Ooijen and Voorrips, 2001), balanced allele frequencies, and a structure that maximizes genetic variation between individuals. However, these populations have disadvantages relative to crop improvement goals. For example, such populations fail to maintain favorable combinations of alleles. A less than desirable parent may be chosen to emphasize variation in a specific trait, and this parent subsequently contributes too much to the population (50% in RIL and F 2 ; 25% in BC). Thus, understanding the genetic basis of a trait comes at a cost to improving germplasm. There is a trend toward developing populations that allow isolation of genetic factors and/or simultaneous breeding and discovery of genetic factors that underlie a trait. The inbred backcross (IBC) population was proposed by Wehrhahn and Allard (1965) as a way of isolating discreet genetic factors that contribute to quantitative inheritance. It consists of a donor parent crossed to a recurrent parent, followed by consecutive backcrosses to the recurrent parent. The progeny population is then selfed to homozygosity. If the trait of interest were affected by few QTL, then the lines derived from the IBC population would differ for donor alleles at a single (or a few) QTL only. IBC breeding evolved as a method for the introgression of exotic germplasm to improve quantitative traits in crop plants. This method has been utilized in bean (Bliss, 1981; Sullivan and Bliss, 1983), oilseed rape (Butruille et al, 1999), rice (Lin et al, 1998) and tomato (Hartman and St Clair, 1999; Doganlar et al, 2002; Kabelka et al, 2002; Kabelka et al, 2004; Yang et al, 2005b) for classical breeding and QTL studies. The advantage of the IBC population is that it provides a 5

22 genetic structure that allows replication and optimum allocation of sampling resources for mapping and subsequent selection of traits. The approach of simultaneous gene discovery through marker-trait analysis in breeding populations was first described for advanced backcross (AB) populations (Tanksley and Nelson, 1996). The AB population structure is identical to early stages of IBC population development. The strength of this simultaneous approach is that it combines the isolation of individual loci as described by Wehrhahn and Allard (1965) with the power of modern genetic analysis. The IBC and AB populations designed for breeding have also been exploited for QTL and gene discovery. Marker development Various marker techniques can facilitate a distinction between genotypes. The most common types of markers can be categorized as morphological or molecular. Generally, morphological markers provide a distinctive phenotype that can be scored easily. Sometimes these visual markers are associated with a trait of interest and indirect selection can be practiced (e.g. color and lycopene content; Arias, 2000). Molecular markers are the result of protein or DNA sequence variants, i.e. polymorphisms, which distinguish alleles in a population. These markers require technical manipulation of protein or DNA extracts to detect genetic differences, but offer the possibility of very large numbers. The identification of marker-trait 6

23 associations as a tool for discovery can therefore encompass a scan across the whole genome. Although selectable markers linked to a trait of interest have been used in plant and animal breeding programs for years, the increased efficiency of detection has improved the power of this approach in recent years. Biochemical and molecular markers are very diverse. Protein markers, or isozymes, were the first tools used to measure genetic diversity. DNA-based molecular markers may be classified based on detection methods: hybridization, polymerase chain reaction (PCR), sequencing, and DNA array (reviewed in Gupta and Roy, 2002). The hybridization-based markers are random fragment length polymorphism (RFLP) and oligonucleotide fingerprinting. RFLP markers allowed the creation of the first high-density genetic maps of tomato with over 1000 markers (Tanksley et al, 1992). Some of the PCR-based markers are random amplified polymorphic DNAs (RAPDs), amplified fragment length polymorphism (AFLP), simple sequence repeat (SSR), and microsatellite-primed PCR (MP-PCR). The PCR-based approaches allow for increased efficiency of technical detection. Sequence-based DNA markers include single nucleotide polymorphisms (SNPs), insertions/deletions (INDELS) and SSR. There are a number of detection methods for sequence-based markers that make them more prevalent in genetic mapping studies. The use of DNA-microarrays is emerging as a method to discover SNP markers. A microarray refers to an experimental approach in which selected molecules are affixed to a substrate (nylon membrane, glass slide, or silicone chips) 7

24 at high density in a known configuration. A DNA microarray may contain genomic DNA, cdna or oligonucleotides that are labeled with a fluorescent dye. The DNA samples are ordered in a two-dimensional matrix and detection is based on the hybridization between the labeled DNA molecules and the array probes. This array technique can be utilized to detect chromosome abnormalities, to sequence DNA, and to detect SNPs for genetic mapping. The detection of genetic differences using oligonucleotide arrays was first described for yeast (Winzeler et al, 1998). In this approach the traditional hybridization process is reversed and probes that span sequence variants are affixed to a substrate. Progeny DNA is then labeled and hybridized. More recently, the microarray-based method has been modified to meet the demands of high throughput genotyping for SNP development (Flavell et al, 2003; Jenkins and Gibson, 2002). The advantage of the microarray-based approach is that hundreds to thousands of polymorphisms can be assayed simultaneously. Another method to identify markers is to utilize expressed sequence tags (ESTs) available in public sequence databases. This in silico bioinformatics approach uses sequence redundancy among EST data available between specific genotypes. Discrepancies in the sequence alignment from the two sets of ESTs are indicative of potential polymorphisms. Further verification and confirmation of candidate SNPs is necessary for application in genetic studies. This method has been used successfully, for example in tomato (Yang et al, 2004), sunflower (Pashley et al, 2006), and wheat (Somers et al, 2003). Utilizing EST databases has also been exploited for the development of SSR markers in wheat (Yu et al, 2004a). 8

25 A third approach for marker discovery is to exploit the tendency of intron position to be conserved in plants. Conserved ortholog set are defined as genes having a single copy across most plant taxa (Fulton et al, 2002). Primers are developed based on the expected position of the intron sequence. They are then used to amplify DNA to test for polymorphism in germplasm of interest. The identification of polymorphic markers based on introns in COS (COSi) was reported for tomato (Yang et al, 2005b). This method was also used in comparative genome mapping between tomato and Arabidopsis thaliana (Fulton et al, 2002) and A. thaliana and species of the Compositeae family (Timms et al, 2006). Marker application Molecular markers can be used for a number of applications. Some of these applications are DNA fingerprinting, positional cloning, population genetics, tagging single gene traits, tagging polygene traits (QTL mapping), genetic diversity analysis, comparative genome mapping, and marker-assisted selection (MAS). These applications can lead to discovering and characterizing new genes, deciphering gene structural evolution, and discerning the genetic basis of complex traits with a quantitative nature. In tomato breeding and genetics, the use of SNPs to detect QTL associated with quality traits has been achieved successfully. Yang et al (2004) utilized the in silico approach to develop SNPs and found two markers that were useful in detecting 9

26 QTL for color in an F 2 population consisting of an IBC line (IBL2349) and an elite variety OH8245. Yu et al (2004b) utilized SSRs developed from ESTs to construct a comparative framework map between wheat and rice. New platforms, detection methods and genotyping assays make the DNA microarray coupled with multiplexing the future trend for high throughput genetic analyses (Jenkins and Gibson, 2002; Shen et al, 2006). Additionally, extensive sequence resources are available for various crops, which are archived in public databases (SGN for Solanaceae; AtGDB for Arabidopsis; CGPDB for Compositeae species; MGDB for maize; etc.). For tomato alone, a total of 356,890 ESTs (NCBI; dbest release ) can be exploited. The tomato genome sequencing project of the gene-rich euchromatin portions is 28% complete (verified , SGN). These recent advances in genome sequencing show promise in utilizing this information for applied outcomes. Caveats of QTL analysis In quantitative genetic analyses, many factors affect the detection and confirmation of QTL. The structure of the mapping population affects the ability to detect QTL. For example, F2, backcross, recombinant inbred lines (RIL), introgression lines (IL), and inbred backcross (IBC) populations are commonly used in QTL analysis. Each structure provides its strength and weaknesses (reviewed in Asins, 2002). For example, an F2 may be better suited than a backcross to detect recessive QTL. Epistatic interactions can be detected in RIL but not in an F2 10

27 population. For this reason, it is important to implement different population structures in a study for the purpose of confirming QTL. This approach helps detect recessive QTL and disclose confounding dominant and additive effects, as well as epistatis. The choice of population structure needs to allow for replicated testing. QTL are invariably affected by the environment. Testing within and across environments helps discern the extent of the genotype by environment (GxE) interaction, whereby trait values for a given genotype differ across environments. There are four patterns of GxE interaction (Ouyang et al, 1995). For the first pattern, one genotype is consistently superior to the other genotype across environments. Second, one genotype is superior to the other genotype across environments but the difference between their performance is different. Third, the performance between two genotypes is reversed across environments, but the absolute difference remains the same. Fourth, the performance between two genotypes is reversed across environments and the difference in performance between the two genotypes is different. The last two patterns are descriptive of crossover interaction, in which there is a rank shift of the genotypes across environments. Other factors that influence the accuracy of detecting and confirming QTL are the heritability of the trait, the population size, the genetic map available and the marker coverage, and the methodology for measuring the trait and for detecting QTL. Different methods for measuring a trait of common interest often lead to discrepancies among QTL studies (e.g. Fulton et al, 2000). The methodology implemented in a QTL study to measure a trait should be objective and reproducible. 11

28 A large mapping population is an important factor in estimating accurate QTL positions, whereas little gain in precision is achieved by increasing the marker density beyond cm (Darvasi et al, 1993; Kearsey, 1998; Kearsey and Farquhar, 1998). Moreover, effective population size is often affected by missing data and distorted allelic segregation (McCouch and Doerge, 1995). Despite these caveats, a well-designed experiment for QTL analysis and mapping can further enhance breeding progress. Marker-assisted selection With continuing progress in the development, reliability, and efficiency of molecular markers in crop plants, breeders and geneticists are exploring markerassisted selection (MAS). MAS provides an opportunity to increase the efficiency of selection for specific genes of interest as compared to classical plant breeding that relies on trait-based selection. For agronomic and horticultural crops, MAS can be optimized by conducting concurrent field evaluation to assess the overall performance of the crop. Generally, a MAS program is conducted by (i) identifying molecular markers linked to a gene/qtl associated with a trait of interest, (ii) screening a population, and (iii) selecting the desirable individuals in the population. There are several cases where MAS is more efficient than trait-based approaches. For example, when the heritability of the trait is low, early generation selection with markers will result in improved genetic gain (Stromberg et al, 1994; Francis et al, 2003). Also, MAS may be more efficient where there is tight linkage between a 12

29 QTL and molecular markers (Knapp, 1998). MAS can also facilitate the process of pyramiding QTL for multiple traits (e.g. Yang and Francis, 2005), or it can be used to break linkage to undesirable traits that accompany gene introgression from wild species (e.g. Frary et al, 2003). In tomato breeding, MAS has been utilized, but not extensively (Foolad and Sharma, 2005). The first application was reported for the use of an isozyme marker (APS-1, a gene encoding acid phosphatase) associated with nematode resistance (Rick and Fobes, 1974). Selection for the root knot nematode resistance gene, Mi, using the APS-1 marker remains standard practice. The first application of molecular markers in tomato for selection of a non-pathogen related trait was for soluble solids (Tanksley and Hewitt, 1988). The use of RFLP markers linked to an introgressed chromosomal fragment from Lycopersicon chmielewskii resulted in increased levels of soluble solids in different lines of tomato. Other reports have suggested the potential of selecting for a specific trait via MAS, e.g. acylsugarmediated pest resistance (Lawson et al, 1997), blackmold resistance (Robert et al, 2001). These reports have been largely theoretical in nature as sugar content, pest and disease resistance was conferred by wild species and the populations used for genetic dissection retained too much of the unadapted germplasm for commercial application. In addition, these reports pertain more to genetics than breeding, whereby the loci conferring resistance were linked to markers but no selection was practiced for crop improvement. In contrast, a successful application of MAS was conducted in tomato breeding populations (Yang and Francis, 2005). Markers were 13

30 used to select recombinants that bring together resistance to both bacterial speck (caused by Pseudomonas syringae pv. tomato) and race T1 of bacterial spot (caused by Xanthomonas campestris pv. vesicatoria). The combination of MAS and field selection yielded breeding line with resistance to both bacterial diseases. CHARACTERISTICS OF TOMATO FRUIT QUALITY Yellow shoulder disorder Color and color uniformity of tomato fruit affect grade and appearance of the end-product and are therefore important quality attributes in the tomato processing industry. A major quality constraint to producing tomatoes for whole-peel or diced products is the presence of yellow shoulder disorder (YSD). YSD is a blotchy ripening disorder that is characterized by discolored regions under the epidermis of mature fruits (Francis et al, 2000). It is an economic problem because growers receive payment premiums for color quality, and USDA processor grades are largely determined by the amount of off-color tissue in products (AMS-USDA, 2005). The causes of YSD are diverse. The incidence of YSD is influenced by soil fertility, especially phosphorous and potassium nutrition (Clivati-McIntyre et al, 2007; Hartz et al, 1999); environmental factors, including low to high temperature fluctuations (Jones and Alexander, 1962), high pericarp temperature, and high relative humidity (Picha, 1987); and genetic background (Sacks and Francis, 2001). 14

31 Some varieties appear more or less susceptible, though resistance to YSD has not been explicitly reported (Sacks and Francis, 2001; Hartz et al, 2005). Health-promoting carotenoids in tomato The color of tomato fruit is determined by carotenoid pigments. Ripe fruit contains high levels of lycopene, the pigment that gives tomato its red color. There is considerable interest in the dietary role of lycopene in reducing the risk of certain cancers. Reports show that lycopene is correlated with reduced incidence of prostate cancer (Clinton, 1998; Stacewicz-Sapuntzakis and Bowen, 2005; Wu et al, 2004) and breast cancer (Sesso et al, 2005). However, lycopene is not recognized as a nutrient despite the ongoing medical and epidemiological studies confirming health benefits. The chemistry and biology of lycopene, as well as its implications for human health and disease, has been reviewed extensively (Clinton, 1998; Giovannucci, 1999; Rao and Agarwal, 2000). Briefly, lycopene has 11 conjugated double bonds and two unconjugated double bonds. Lycopene can function as an antioxidant, which reduces oxidative damage. The property may prevent chronic diseases such as cancer and cardiovascular disease. Ripe fruits also contain β-carotene, which is synthesized from lycopene. Beta-carotene is the carotenoid recognized as a nutrient in tomato fruit due to its provitamin A activity. Each year 750 million people suffer from vitamin A deficiency, and a single serving of tomato products can supply in excess of 30% of 15

32 recommended daily allowances. Tomato varieties are available that could meet vitamin A dietary requirements with a single serving. Tomato varieties have been developed via classical breeding schemes or transgenic approaches to elevate certain carotenoids. Hybrids derived from parents carrying different ripening and carotenoid-related mutants resulted in increased lycopene (Faria et al, 2003). Introgression lines of cherry tomatoes were developed from Solanum galapagense (formally known as Lycopersicon cheesmanii f. minor) as the high beta-carotene donor parent. Their total carotenoid content consisted of 94.6% β-carotene (Stommel et al, 2005). Exploiting the transgenic approach, overexpression of phytoene synthase introduced from the bacterium Erwinia uredovora resulted in increased β-carotene and lycopene content by 1.8- and 2.2- fold, respectively (Fraser et al, 2002). Similarly, the phytoene desaturase gene from E. uredovora was overexpressed in tomato fruit and resulted in increased β-carotene content by 3-fold (Romer et al, 2000). Francis and Kabelka (2001) reported a positive and significant correlation between color and lycopene content in a breeding population consisting of adapted germplasm. Their data suggest that epistatic interactions and quantitative trait loci (QTL) may provide alternative genes for the improvement of color in tomato breeding programs. Thus variation in existing elite germplasm remains to be exploited for the improvement of the crop. 16

33 OBJECTIVE TRAIT MEASUREMENT Measuring color and color uniformity Precise and accurate measurements are needed in order to assess phenotypic variability and to dissect the genetic basis of a trait. In the case of color, there are different approaches to obtain objective measurements: spectrophotometer, colorimeter and computerized image analysis (e.g. computer vision systems), which uses digital or video camera. Colorimeters are used to collect chromaticity values of color based on a standard color space. They provide a single-point measurement, whereas digital images can provide average color values over a selected area, depending on the software available for image analysis. The use of flat-bed scanners to collect images has been reported (Shahin and Symons, 2001; Kleeberger and Moser, 2002; Kwack et al, 2005). However, the software developed to analyze the images is not fully automated and requires many manual adjustments (Shahin and Symons, 2003). There are a number of digital image acquisition and analysis techniques for color and other physical factors in foods such as apples (Leemans et al, 2002; Li et al, 2002), banana (Mendoza and Aguilera, 2004), chicory (Zhang et al, 2003), seeds (Granitto et al, 2002; Sako et al, 2001; Shahin and Symons, 2001), and meats (O Sullivan et al, 2003; Tan, 2004). Image analysis is also common in floricultural crops such as Lisianthus (Yoshioka et al, 2006) and Begonia (Lootens et al, 2007). The instruments designed to measure color have specific settings in term of their standard illumination and viewing angles. When these specifications are not 17

34 reported along with color measurements, comparative studies for color in crops are challenging, if not impossible. Quantifying carotenoids The analytical methods for identifying and quantifying carotenoids have been reviewed by Gross (1991) and provide an overview of the steps involved in the analysis of carotenoids. The analysis can be difficult due to the instability of carotenoids, the tendency to undergo stereomutation, the photo- and thermolability, and the readiness with which the carotenoids undergo oxidation. Isolation of carotenoids involves extraction, saponification, and separation. The most important technique for separating and purifying carotenoids is chromatography. Column chromatography, thin-layer chromatography (TLC), and high-pressure liquid chromatography (HPLC) are techniques that have been successfully applied in carotenoid separation, though HPLC is the most prevalent method used. HPLC provides rapid, reproducible, and highly sensitive carotenoid separation, identification, and quantification, which can be achieved simultaneously. Conventionally, a reverse-phase C 18 column is used to separate specific carotenoids. However, to obtain superior selectivity of isomers, especially the cis and trans isomers of lycopene, a reverse-phase C 30 stationary phase column is usually used (Nguyen and Schwartz, 1999). 18

35 RATIONALE AND SIGNIFICANCE OF RESEARCH Color is among the most important attributes of tomato in the processing industry. Color is greatly affected by YSD, a ripening disorder that results in discoloration of the proximal end tissues of the tomato fruit. YSD is an economic problem and may reduce the nutritional quality of tomato products. Understanding the genetic component of this disorder and developing varieties that are less susceptible to the disorder will benefit the grower, the processor, and the consumer. The goal of crop improvement is to decipher the most efficient means to solve a problem and ameliorate a trait. There are several issues pertinent to tomato breeding research: (1) MAS application is limited to crosses of Solanum lycopersicum to its wild relatives because of an insufficient number of markers available within the adapted germplasm; and (2) the conflict between breeding goals and genetic goals arises with the choice of population structures. These issues can be addressed by the development of new molecular markers and by applying a population structure that will bridge the gap between the geneticist and the breeder. In well-designed breeding populations, in which the parents are selected based on the traits of interest, the adapted S. lycopersicum germplasm was shown to have genetic variability for color. This variability can be exploited efficiently in a breeding program (Kabelka et al, 2004; Yang et al, 2004). The inbred backcross (IBC) population has the advantage of providing a genetic structure that allows replication and the optimum allocation of sampling resources for mapping and subsequent selection of traits within germplasm pools. The population structure can also be 19

36 efficiently combined with marker-trait analysis to test hypotheses and improve color and nutritional quality. 20

37 CHAPTER 2 OPTIMIZING SAMPLING OF TOMATO FRUIT FOR CAROTENOID CONTENT WITH APPLICATION TO ASSESSING THE IMPACT OF RIPENING DISORDERS ABSTRACT Color defines one aspect of quality for tomato and tomato products. Carotenoid pigments are responsible for the red and orange colors of tomato fruit, and thus color is also of dietary interest. The aims of this study were (1) to optimize sampling of lycopene and β-carotene in tomato fruit between field and analytical replications and (2) to determine the effect of yellow shoulder disorder (YSD) on the content of lycopene and β-carotene in tomato juice and tissue. Our results show that increasing biological replications while minimizing analytical replications is an efficient strategy for reducing the experimental error associated with measurements of lycopene and β-carotene. We found that YSD significantly reduces lycopene in affected tissue and in juice made from affected fruit. In contrast, β-carotene concentrations were only reduced in affected tissue, but were not significant reduced 21

38 in juice. With increasing interest in biofortified crops, modulating the carotenoid profile in tomato by minimizing YSD symptoms represents a strategy for improving tomato fruit quality that is currently supported by grower contract structure and processor grades. INTRODUCTION Color and color uniformity affect grade and appearance of the end-product and are therefore important quality attributes in the tomato processing industry. A major quality constraint to producing tomatoes is the presence of yellow shoulder disorder (YSD). YSD is a blotchy ripening disorder that is characterized by discolored regions under the epidermis of mature fruits (Francis et al, 2000). It is an economic problem because growers receive payment premiums for color quality, and USDA processor grades are largely determined by the amount of off-color tissue in products (AMS-USDA, 2005). The causes of YSD are diverse. The incidence of YSD is influenced by soil fertility, especially potassium nutrition (Clivati-McIntyre, 2007; Hartz et al, 1999); environmental factors, including low to high temperature fluctuations (Jones and Alexander, 1962), high pericarp temperature, and high relative humidity (Picha, 1987); and genetic background (Sacks and Francis, 2001). Some varieties appear more or less susceptible, though resistance to YSD has not been explicitly reported (Sacks and Francis, 2001; Hartz et al, 2005). 22

39 The color of tomato fruit is determined by carotenoid pigments. Ripe fruit contains high levels of lycopene, the pigment that gives tomato its red color. There is considerable interest in the dietary role of lycopene in reducing the risk of certain cancers, including prostate cancer (Clinton, 1998; Stacewicz-Sapuntzakis and Bowen, 2005; Wu et al, 2004) and breast cancer (Sesso et al, 2005). Ripe fruit also contains β-carotene, which is synthesized from lycopene. Beta-carotene is the carotenoid recognized as a nutrient in tomato fruit due to its pro-vitamin A activity. Each year 750 million people suffer from vitamin A deficiency, and a single serving of tomato products can supply in excess of 30% of recommended daily allowances. Tomato varieties are available that could meet vitamin A dietary requirements with a single serving. With the increasing interest in biofortified crops, modulating the carotenoid profile in tomato is becoming a major focus of germplasm improvement efforts. A limitation to biofortified tomato is the lack of economic incentives within current contract and pricing structures. Picha (1987) reports a subjective deficiency in lycopene in yellow shoulder tissue based on discoloration and a reduction in total carotenoids in tomato fruit from plants that received low K fertility. However this report does not provide evidence of the effect and variability of YSD on carotenoid content nor the potential effects on the nutritional value of affected fruit. Because YSD affects grower premiums and processor grades, it may be economically advantageous to quantify how a common ripening disorder, YSD, affects the content of carotenoids in tomato products. The aims of the present study were (1) to 23

40 optimize sampling for lycopene and β-carotene by addressing the relative importance of biological and technical replications and (2) to determine the effect of YSD on the content of lycopene and β-carotene. MATERIAL AND METHODS Plant material Five processing tomato varieties were used in the study: OH8245, PS696, FG99-36, FG00-118, and FG PS696 is a commercial hybrid, which is susceptible to YSD. OH8245 is an open pollinated variety (Berry and Gould, 1991) and is also susceptible to YSD. FG99-36, FG and FG are experimental varieties with above average agronomic performance. A complete randomized experimental design consisting of two replications each year was grown at the Ohio Agricultural Research and Development Center (OARDC) North Central Agricultural Experimental Station in Fremont, Ohio in 2003 and Each plot consisted of 20 plants per genotype spaced 30 cm apart, with plots spaced 150 cm apart. All field plots were planted and maintained following conventional practices (Precheur et al, 2004). The plots were harvested when 80% of the fruits were ripe. Fruits were collected from each plot and the proximal end of each fruit was cut. The fruits were then categorized as either not affected by YSD (nonysd) or affected by YSD (YSD). In 2003, flesh at the proximal end of nonysd and YSD fruit was dissected and saved in a tube as tissue. In both 2003 and 2006, juice was processed from each 24

41 plot and each category within a plot using a commercial blender. For each sample, 100 ml of juice (2x 50 ml) was kept for further analysis. In total, four juice samples were collected from each plot, two from each phenotype (YSD and nonysd). The samples were stored at -20 o C until carotenoid extraction. Trait evaluation: carotenoid quantification Carotenoid extraction was carried out under red light following a hexane/acetone-based protocol modified from Ferruzzi et al (1998). Juice samples were thawed to room temperature and 5.0g of sample were homogenized in 50mL of methanol with 1.0g of CaCO 3 and 4.0g of celite. The methanol extract was filtered as described by Nguyen et al (2001). The filtrate was suspended in 50mL of 1:1 acetone/hexane and allowed to stand for 1 min prior to homogenization. Repeated acetone/hexane extractions (up to three times) were required to recover the majority of carotenoids. Three milliliters of extract were stored in five 12mL-glass vials and dried under nitrogen. The vials were then wrapped in aluminum foil and at -20 o C freezer until further use. We used a Waters 2690 reverse-phase HPLC system equipped with a photodiode array detector for analysis. The carotenoid extracts were reconstituted with appropriate volumes of MTBE, depending on the volume of the dried extracts and on the desired concentration range of the analysis. An absorbance reading using a UV-spectrophotometer was recorded for each carotenoid extract at 471 nm 25

42 wavelength, a dilution of the solution followed if the absorbance reading was greater than 1.0. Separations were achieved using a C 18 column (Vydac 201TP54; 4.6mm x 250 mm). The separation of total carotenoids was carried out at a flow rate of 1.0 ml per minute using a multi-step linear gradient of 80 to 100% MTBE in 98% Methanol:2% 1 M ammonium acetate for 30 minutes. Standards of β-carotene and lycopene were purchased from Sigma Chemical Co. The peak identification and the subsequent quantification of β-carotene and lycopene were achieved using the standard curve for each compound and their molar absorptivity coefficients (reviewed in Nguyen et al, 2001). The carotenoid content was reported in units of mg/100g of sample on a fresh-weight basis. Digital phenotyping for YSD We employed the software Tomato Analyzer (TA) to collect objective measurements of color. TA is an image processing software that recognizes and collects data from JPEG files (Brewer et al, 2006). We used a flatbed scanner to scan the cut surface of the proximal end for 12 fruits within each category/plot and saved a JPEG image. The color function in TA records RGB values of each pixel of the selected object on the image and translates these values into average L*a*b* values from the CIELab color space (Commission Internationale de l Éclairage; CIE, 1978). The L*a*b* values are then used to calculate chroma as (a 2 +b 2 ). Hue is calculated as 180/pi*acos(a/ (a 2 +b 2 )) for a*>0 and as 360- ((180/pi)*acos(a/ (a 2 +b 2 ))) for a*<0. Percent YSD tissue was calculated from the 26

43 proportion of pixels that fall into a defined range of hue values. YSD symptoms are discolorations in the yellow to green region of the hue color wheel. To capture the maximum discoloration due to YSD, we tested different ranges of hue values: , , , , and Statistical analysis All statistical analyses were performed using SAS software (v. 9.1; SAS Institute Inc, Cary, NC). The following model was used to test the sampling effect on lycopene and β-carotene: Y ijklm = μ + P i + R j(i) + E k(ij) + I l(ijk) + ε ijklm where Y ijklm was the trait measured (lycopene or β-carotene), μ was the overall mean, P i was the effect due to i th plot, R j(i) was the effect due to j th sample withinplot, E k(ij) was the effect due to k th extraction, I l(ijk) was the effect due to HPLC injection and ε ijklm was the experimental error. In this nested design, all effects were random and the approximate F-tests were adjusted for the corresponding error terms. The model designed to test the effect of YSD on lycopene and β-carotene included the main effects YSD phenotype, genotype nested within YSD phenotype, year and block. The within-plot variation represents the juice sampling effect and was nested within block, genotype and YSD phenotype. Year, block and within-plot variation were considered random factors. The approximate F-tests were adjusted for their corresponding error term. The model was: Y ijklmn = μ + P i + G j(i) + Y k + B l + R m(l) + ε ijklmn 27

44 where Y ijklmn was the trait measured (lycopene or β-carotene), μ was the overall mean, P i was the effect due to YSD phenotype, G j(i) was the effect due to genotype, Y k was the year effect, B l was the block effect, R m(l) was the within-plot effect, and ε ijklmn was the experimental error. For each model, the estimates of variance were obtained using the restricted maximum likelihood (REML) method with the Mixed procedure (PROC MIXED) of SAS. The percent total variance is reported to represent the proportion of variance explained by each factor and nested factor. RESULTS Variance in lycopene and β-carotene Total phenotypic variation for lycopene and β-carotene was partitioned to ascertain the relative importance of field and analytical sampling (Table 2.1). In the field sampling, there were significant differences among and within plots for both lycopene and β-carotene. The most significant field variation was observed between plots for lycopene (49.83% of total variation) and β-carotene (52.52% of total variation). The proportion of within-plot variation was 7.3% and 3.0% for lycopene and β-carotene, respectively. In total, experimentally controlled aspects of field sampling accounted for 57% and 55% of the total variation for lycopene and β- carotene, respectively. Uncontrolled error accounted for 43% and 45% of variation for lycopene and β-carotene, respectively. In contrast, there was no significant 28

45 variance due to analytical sampling for either carotenoid. Neither replicated extraction nor replicated HPLC injection contributed to the total phenotypic variation for either lycopene or β-carotene. Effect of YSD on lycopene and β-carotene We sought to address the effect of YSD on lycopene and β-carotene content (Table 2.2). There were significant differences in the lycopene content of juice made from YSD and nonysd fruit over two years. In contrast, juice made from YSD or nonysd fruit did not differ significantly for β-carotene content. The variation due to YSD phenotype accounted for 4.15% of the total phenotypic variation for lycopene, but none for β-carotene. YSD reduced lycopene content by 13% in the first year and 24% in the second year, and these reductions were highly significant (Table 2.3). Although there was a trend toward reduced β-carotene (4-9%) in juice made from YSD affected fruit, this reduction was not significant. We also compared carotenoid concentration in YSD tissue versus nonysd tissue. There was a highly significant difference between tissue types, with a reduction in both lycopene and β-carotene in YSD tissue (P<0.0001; data not shown). Compared to nonysd tissue, YSD tissue showed a reduction of 61.5% and 71.5% for lycopene and β-carotene, respectively. Genotypes were tested within each level of the YSD phenotype. There was no significant difference among genotypes for lycopene; however, significant 29

46 differences were observed for β-carotene (Table 2.2). In 2003, PS696 had the highest β-carotene content (1.67 mg/100g f.w.) whereas FG had the lowest (1.27 mg/100g f.w.). In 2006, similar trends were observed. OH8245 had the highest β-carotene content (0.472 mg/100g f.w.), followed by PS696 (0.354 mg/100g f.w.). FG again had the lowest β-carotene content (0.137 mg/100g f.w.). The year effect was highly significant for both lycopene and β-carotene (Table 2.2). Year to year differences explained 78.4% of the total phenotypic variation for lycopene and 94% for β-carotene. In the combined analysis, we did not detect significant block effects for either carotenoid. However, sampling within a plot was significant for both carotenoids, explaining 3.99% and 2.48% of the total phenotypic variation for lycopene and β-carotene, respectively. Less than 2% of the variation was left unexplained by the model for both traits. The significant year effect was evident in lycopene content with a 1.6-fold increase in the YSD samples and a 1.9- fold increase in the nonysd samples from 2003 to 2006 evaluations. As for β- carotene, there was a 5.1-fold and 4.9-fold decrease between years in the YSD and nonysd samples, respectively (Table 2.3). Relationship between the extent of YSD and lycopene content We investigated several hue ranges to determine the relationship between tissue color and carotenoid content. As we broadened the hue range by decreasing the lower boundary values (70-120, , , , ), the correlation 30

47 decreased and its corresponding p-value increased (r 2 =0.2837, P=0.0189; r 2 =0.2817, P=0.0194; r 2 =0.2760, P=0.0209; r 2 =0.2296, P=0.0379; r 2 =0.1206, P=0.1452, respectively for each hue range). For the , , , hue windows, the linear correlation was significant, suggesting a decrease in lycopene content with increasing % YSD (Figure 2.1). Increasing the upper boundary values to hue=180 did not improve the precision of correlations. The correlation with β- carotene was not significant. DISCUSSION Measurement of carotenoids using HPLC is labor intensive and expensive. Therefore sampling strategies that optimize the balance between biological replication (either field or genotype) and technical replication (extraction and injection) maximize resources and permit more precise measurements of variation. Our results demonstrated that biological replications should be increased relative to technical replications to reduce the error associated with quantifying lycopene and β- carotene. The detection of significant plot to plot variation emphasizes the need to measure replicates within a location. Our results reveal that variation in field sampling is as high as 58% for lycopene and 55% for β-carotene. Increasing field replications may require more field space and seed for the germplasm to evaluate but is not costly to the researcher. The variation due to analytical sampling, which represents carotenoid extraction and HPLC injection, was negligible among 31

48 replicates. Minimizing replication for analytical sampling does not have a negative effect on the precision of the carotenoid estimates. Value-added whole and diced tomato products require tomatoes with high color quality. Both color and color uniformity are affected by yellow shoulder disorder (YSD), a ripening disorder that results in discoloration of the proximal end tissues of the fruit, thus reducing appearance and value. This study indicates that YSD also affects the value of tomato products from the perspective of a healthconscious consumer. Because carotenoids are the primary determinant for tomato color, we investigated the effect of YSD on lycopene and β-carotene. We demonstrated a significant negative effect of YSD on lycopene and β-carotene content in tissue. Only lycopene content was significantly reduced when affected fruit were used to make juice. In the first year of evaluation, the mean lycopene content in juice made from YSD-affected fruits was reduced by nearly 13% and by 24% in the second year. With the potential health benefits of lycopene (reviewed in Clinton, 1998), modulating the lycopene profile of tomato products by reducing the incidence of YSD is of potential interest. Currently the market for processing tomatoes does not offer incentives for increasing bioactive carotenoids. However grower contracts often reward or penalize color quality, which we have shown affects lycopene content. Processor grades of peeled products are also affected by color uniformity (Barrett et al, 2006). Fruit affected by YSD often are processed into sauces and cover juice where appearance is less of an issue. Our results suggest that this approach does not significantly affect the nutritional status of the product as 32

49 defined by β-carotene content. It is unclear what will be the consequences of lower lycopene levels in sauces and juice made from YSD-affected fruit. Studies aimed at assessing human absorption of lycopene suggest that the trans-lycopene isomer is inefficiently absorbed, and that high levels of lycopene do not result in increased absorption (Unlu et al, 2007). Therefore, despite reduced lycopene, it is unclear that juice made from YSD affected fruit is of lower value from a dietary perspective. Reduced lycopene levels and current contract and grading practices suggest that breeding varieties for resistance to YSD might be an effective strategy for increasing the value of tomatoes for growers, processors, and possibly consumers. The effect of genotype nested within each level of phenotype, Genotype(YSD) and Genotype(nonYSD), contributed significantly to the total phenotypic variation, though the contribution to total variation was low. This study was not designed to select for genotypes with resistance to YSD, but our results suggest that a sampling strategy for genetic variation should increase biological replications and reduce technical replications in order to sample a large number of genotypes (varieties). Our results also suggest that attention should be paid to specific genes that are currently in use in breeding programs. For example, the old gold crimson (og c ) is a mutant gene that results in enhanced red color with increased lycopene and reduced β- carotene (Thompson et al, 1967; Ronen et al, 2000). FG00-124, an og c hybrid, had lower β-carotene in both 2003 and This is not a surprising result given the function the og c protein as a fruit-specific β-cyclase (Ronen et al, 2000). These results are also consistent with Sacks and Francis (2001) who reported that og c 33

50 explained 18-27% of the total phenotypic variation for color. The year to year and within field environmental effects were also significant and explained a high proportion of phenotypic variation for lycopene and β-carotene. Although the reduction in lycopene due to YSD was consistent from one year to the other, there was a marked increase in the overall mean of lycopene the second year compared to the first. Environmental influences such as weather, soil fertility management practices, and/or soil properties have been reported to affect the severity of YSD (Hartz et al, 1999; Clivati-McIntyre, 2007). In Fremont, Ohio, there was twice as much precipitation in June 2006 compared to June The precipitation in July was similar both years, but August 2003 had 1.3 times more precipitation than Overall, the average maximum air temperature for the summer months in 2006 was two degrees Fahrenheit higher than Fertility management practices were very similar both years, as conventional practices were followed (Precheur et al, 2004). Thus year to year environmental fluctuations appear important in determining concentrations of carotenoids. We obtained color data from the same fruits used to make up the juice for each sample. Analysis of the digital images with the Tomato Analyzer allowed us to determine the degree of tissue affected by YSD at the proximal end of the tomato fruit. The correlation between the proportion of YSD tissue and lycopene content can be utilized as a prediction tool. Our results suggest that for every 10% increase in area affected by YSD, there is a decrease in lycopene of 1.03 mg/100g f.w. These 34

51 results can be used to draw thresholds for an acceptable amount of YSD to preserve a certain level of lycopene. In conclusion, quantification of lycopene and β-carotene concentration in tomato juice samples can be more precise by increasing biological replications while minimizing analytical replications. YSD affects the health-beneficial carotenoids predominantly by reducing lycopene content. ACKNOWLEDGMENTS I would like to thank the following people for contributing to this chapter. David Francis provided scientific guidance, and the resources to complete this study. Steven Schwartz (Department of Food Science and Technology, The Ohio State University) allowed me access to his laboratory for the analysis of carotenoids. I thank Marjory Renita and Rackel Kopec for their technical expertise and training with the HPLC analyses, and Troy Aldrich for his assistance in the field experiments and the quality lab. I also thank Bert Bishop for his assistance with the statistical analyses. Finally, I extend my appreciation to Mark Bennett, Clay Sneller, and three anonymous reviewers for their critical review of the manuscript. 35

52 TABLES Trait Lycopene β-carotene Sources of Term in Mean % Total Mean % Total DF Variation Model Squares Variance Squares Variance Plot P i * * 52.5 Rep(Plot) R j(i) ** * 2.97 Extraction E k(ij) ns ns 0.00 HPLC Injection I l(ijk) ns ns 0.00 Experimental Error *, ** Significant at α=0.10, α=0.05 respectively; ns, not significant Table 2.1: Optimizing sampling for lycopene and β-carotene content in juice samples of tomato. Trait Lycopene β-carotene Sources of Term in Mean % Total Mean % Total DF Variation Model Squares Variance Squares Variance YSD Phenotype P i **** ns 0.00 Geno. (YSD Pheno.) G j(i) ns ** 1.44 Year Y k **** **** 94.5 Block B l(k) ns ns 0.00 Rep(Plot) R m(ijkl) **** *** 2.13 Experimental Error **, ***, **** Significant at α=0.05, α=0.01, α=0.001 respectively; ns, not significant Table 2.2: Effect of yellow shoulder disorder (YSD) on lycopene and β-carotene content in juice samples of tomato over two years. 36

53 Lycopene β-carotene Lycopene β-carotene Genotype Phenotype (mg/100g f.w.) (mg/100g f.w.) (mg/100g f.w.) (mg/100g f.w.) FG nonysd FG YSD FG nonysd FG YSD FG99-36 nonysd FG99-36 YSD OH8245 nonysd OH8245 YSD PS696 nonysd PS696 YSD mean YSD mean nonysd LSD (α=0.05) % Difference 12.8% 3.63% 24.0% 8.77% P-value Table 2.3: Means of lycopene and β-carotene content in juice made from tomatoes affected by yellow shoulder disorder (YSD) and not affected by YSD (nonysd). 37

54 FIGURES Lycopene content (mg / 100g f.w.) % Yellow Shoulder Disorder (YSD) vs. lycopene content R 2 = % YSD Figure 2.1: Relationship between the extent of yellow shoulder disorder (YSD) in fruit and lycopene content in tomato juice. 38

55 CHAPTER 3 TOMATO ANALYZER COLOR TEST: A NEW TOOL FOR EFFICIENT DIGITAL PHENOTYPING ABSTRACT Characterization of some traits can be achieved via image analysis, which increases the objectivity of phenotypic evaluations. We propose a new tool, which is implemented in the Tomato Analyzer (TA) software application, called Color Test (TACT). This tool allows for accurate quantification of color as well as uniformity of color. A module is built into TACT that allows scanning devices to be calibrated using color standards. Scanned images of tomato fruits were analyzed for internal fruit color with both TACT and a colorimeter. Our objectives were to test the accuracy and precision of TACT compared to a colorimeter and to determine estimates of genotypic variances associated with different color parameters. We show high correlations (r > 0.96) and linearity of L*, a*, b* values obtained with TACT and a colorimeter. The proportion of total phenotypic variance attributed to genotype for color and color uniformity traits measured with TACT was significantly different than obtained from a colorimeter. Genotypic variance nearly doubled for 39

56 all color and color uniformity traits when collecting data with TACT. Color uniformity was also better characterized when the entire surface of the fruit was analyzed, compared to the point measurements obtained from the colorimeter. This digital phenotyping technique can also be applied to the characterization of color in other fruit and vegetable crops. INTRODUCTION Digital phenotyping refers to a description of a trait based on digital images. Computer-based analysis of objects from digital images has the potential for more efficient collection of data and reduction of subjective characterization that is typically prone to bias. There are a number of computer image acquisition and analysis techniques for color in foods such as apples (Leemans et al, 2002; Li et al, 2002), banana (Mendoza and Aguilera, 2004), chicory (Zhang et al, 2003), as well as seed analysis (Sako et al 2001; Shahin and Symons, 2001; Granitto et al, 2002), and meat (O Sullivan et al, 2003; Tan, 2004). Color image analysis is also prevalent in floricultural crops such as Lisianthus (Yoshioka et al, 2006), and Begonia (Lootens et al, 2007). Digital color analysis is also performed in plant pathology to quantify lesions on diseased leaves (Kwack et al, 2005). Coupling digital phenotyping and color analysis from images has the potential to reduce subjectivity and inaccuracies, while simultaneously increasing the efficiency of data collection. Tomato has become the prominent model horticultural crop for studies in genetics and genomic sciences. With extensive resources, including 356,890 ESTs 40

57 (expressed sequence tags; NCBI, dbest release ) and a genome sequencing project focused on euchromatin, which is 28% complete (verified , SGN; Mueller et al, 2005), there is increased promise for research that seeks to translate genome sequence into applied outcomes. Integration of emerging resources from genome studies requires the identification of nucleotide polymorphism, the development of molecular markers, and an extensive set of phenotypic data for the traits studied. Populations of recombinant inbred lines (RIL), introgression lines (IL), and inbred backcross lines (IBC) have been developed and maintained as resources for breeding and genetic mapping studies (RIL, Villalta et al, 2007; IL, Eshed and Zamir, 1995; IBC, Kabelka et al, 2002). All three population structures are considered immortal populations, meaning the populations consist of nearly homozygous lines thus preserving genetic integrity. These populations allow for replication of experiments and extensive analyses from different laboratories. RIL and IL are the most common types of immortal populations exploited in quantitative trait locus (QTL) analyses (reviewed in Keurentjes et al, 2007). Phenotypic and molecular characterization of such populations can be stored in public databases for use by other researchers. There is a need to standardize the way traits are defined (i.e. trait ontology) and measured. In recent years, trait ontologies have been developed in diverse disciplines for database retrieval and archiving (reviewed in Ilic et al, 2007; Brewer et al, 2006). One challenge to achieving a standard approach for measuring a trait objectively is to develop tools and make them available as a community resource. 41

58 To allow for direct comparisons, color analyses should be standardized, measured and interpreted appropriately. Computers use the RGB color space to display color. Each pixel on a screen can be represented in the computer or interface hardware as values of red, green and blue. The RGB color space is not standardized due to differences in hardware and software. It is also non-linear and does not mimic human color perception. In contrast, color spaces such as CIELab (Commission Internationale de l Éclairage; CIE, 1978) were designed to approximate human perception of color. CIELab color space is a reference standard and is most commonly used for measuring object color. Color data collected in the dimensions of the CIELab color space can be archived and used for quantitative analysis of color. Color holds an important economic role in horticultural crops. For fresh or processed products, color is one of the primary determinants of quality, along with texture, size and flavor (Picha, 2006). In the case of tomato, both color and color uniformity contribute to quality. The presence of yellow shoulder disorder (YSD) is a major quality constraint. YSD is a blotchy ripening disorder that is characterized by discolored regions under the epidermis of mature fruits. Cells from YSD tissue are smaller and more randomly organized, and the conversion from chloroplasts to chromoplasts is altered (Francis et al, 2000). Variation for YSD within fruit and among fruits in plots explained more than 75% of the environmental variation for color (Sacks and Francis, 2001). Color disorders are also an economic problem. USDA processor grades are largely determined by the amount of off-color tissue in 42

59 products (AMS-USDA, 2005). To improve the output of high quality product, some processors structure contracts such that growers receive premiums for fruit based on color and color uniformity. Discoloration due to YSD also reduces lycopene and b- carotene concentrations in tissue affected by YSD (Darrigues et al, 2007). A reduction in the incidence of YSD could benefit producers, processors and consumers. Reducing YSD by means of genetic manipulation will require a rigorous assessment of color variability. This requirement can be fulfilled with tools that provide precise, accurate and objective measurements. There are different approaches to obtain objective color measurements including colorimeter, spectrophotometer, or computerized image analysis. Computer vision systems may use digital cameras or scanners. Colorimeters provide a single-point measurement, whereas digital images can provide average color values over a selected area, depending on the software available for image analysis. The use of flat-bed scanners to acquire data has been reported (Shahin and Symons, 2001; Kleeberger and Moser, 2002; Kwack et al, 2005). However, the software available to analyze color images is not fully automated in these applications and requires many manual adjustments (Shahin and Symons, 2003). Our objective was to implement a new digital image analysis tool, Color Test as part of the Tomato Analyzer software application (Brewer et al, 2006). Tomato Analyzer Color Test (TACT) is capable of collecting and analyzing various color parameters in an efficient, accurate and high-throughput manner. We evaluated a 43

60 tomato population for color and color uniformity using both TACT and a colorimeter to estimate variance components associated with different parameters of color. To assess the applicability of TACT to crops other than tomato, we tested the software for its applicability to other fruits and vegetables for which color is an important trait. MATERIALS & METHODS Software implementation The software Tomato Analyzer (TA) was previously described by Brewer et al (2006). Briefly, it was implemented in C++ using Visual Studio 6.0. The image input/ouput was made possible via the image processing library Computer Vision and Image Processing (CVIP) 3.7c. TA was designed to run on the Windows operating system. It has been tested successfully with Windows Vista. The program is free and can be used for academic or private purposes. It can be downloaded from the URL: Tomato Analyzer Color Test The TA module called Color Test (TACT) is designed to collect objective color measurement from JPEG images, which are collected from scanning fruits on the flatbed surface of a scanner (Figure 3.1). A cardboard box covers the scanner to minimize the effect of shadow. Instructions for collecting, importing, and 44

61 conducting the automatic analysis of images are available in the TACT manual (Appendix A). TACT records RGB values of each pixel of the selected object on the image and translates these values into average L*, a*, b* values from the CIELab color space (Commission Internationale de l Éclairage; CIE, 1978). The algorithm implemented in TACT to convert RGB values to L*, a*, b* values can be adjusted to account for the illuminant (D65 or C) and observer angle (2 o or 10 o ). Converting RGB to L*, a*, b* was accomplished in three steps. First, RGB values were scaled to a perceptually uniform color space (equation 1): Var_R = ((((R/255)+0.055)/1.055)^2.4)*100 Var_G = ((((G/255)+0.055)/1.055)^2.4)*100 (1) Var_B = ((((B/255)+0.055)/1.055)^2.4)*100 Scaled RGB values were then converted to XYZ trismulus values using the following relationships (equation 2): X = (Var_R*0.4124)+(Var_G*0.3576)+(Var_B*0.1805) Y = (Var_R*0.2126)+(Var_G*0.7152)+(Var_B*0.0722) (2) Z = (Var_R*0.0193)+(Var_G*0.1192)+(Var_B*0.9505) The XYZ values were converted to L*, a*, b* values using the following relationships (equation 3): L* = 116 f (Y/Yn) 16 a* = 500 [ f (X/Xn) f (Y/Yn)] (3) b* = 200 [ f(y/yn) f (Z/Zn)] 45

62 where f (q) = (q)^1/3 q > f (q) = 7.787q + (16/116) q Yn, Xn and Zn are the trismulus values of the illuminant and observer angle. For illuminant C, observer angle 2 o, Xn=98.04, Yn=100.0 and Zn= For illuminant D65, observer angle 10 o, Xn=94.83, Yn=100.0 and Zn= The L*a*b* values were then used to calculate chroma as (a 2 +b 2 ). Hue was calculated as 180/pi*acos(a/ (a 2 +b 2 )) for a*>0 and as 360- ((180/pi)*acos(a/ (a 2 +b 2 ))) for a*<0. The L* coordinate indicates darkness (~0) to lightness (~100) of color. The a* and b* are the chromaticity coordinates and their axis indicates color directions: +a* is the red direction, -a* is the green direction, +b* is the yellow direction and b* is the blue direction. Hue is an angular measure from 0 to 360, which represents basic color. Chroma is the saturation or vividness of color. In addition to converting RGB values to L*, a*, b* and calculating chroma and hue from these components, an algorithm was written for TACT to compute luminosity from the following relationship (equation 4): Luminosity = (maxcol + mincol) * / (2.0 * 255.0) (4) where maxcol is the highest of the R, G, B values of an analyzed pixel, and mincol is the lowest value. Luminosity accounts for the variable sensitivity of the human eye to radiation at various wavelengths; it defines brightness. The output of the color analysis provides averaged values of R, G, B, luminosity, L*, a*, b*, hue and chroma. 46

63 Customizing Tomato Analyzer Color Test Several parameters in the Color Test can be adjusted by the user prior to analysis. TACT was designed to allow the user to adjust the minimum blue value from the RGB color space to help differentiate the cut surface of the proximal end of the tomato fruit from the peel. In generating the data set for this study, we used a blue value of 30 to define the boundaries. Another option was developed to let the user define two parameters with specific hue ranges of interest. TA returns the proportion (%) of pixels that fall into that hue range. We defined our parameters as percent YSD (%YSD), which represents yellow, green-yellow color, and percent red (%RED), which corresponds to the desired red color of tomato internal tissue. Our hue ranges were 60 to 120 for %YSD and 0 to 48 for %RED. Tomato Analyzer Color Test versus colorimeter. To determine the relationship between colorimeter and TA data, we collected color readings from 247 standard Munsell color plates (X-Rite, Grand Rapids, MI) ranging from 2.5R to 10R (Red), 5Y to 10Y (Yellow), 2.5GY to 10GY (Green- Yellow), 2.5YR to 5YR (Yellow-Red), 5Y (Yellow), 5GY (Green-Yellow), 5G (Green), 5BG (Blue-Green), 5B (Blue), 5PB (Purple-Blue), 5P (Purple) and 5RP (Red-Purple). Absolute color measurements were collected for each standard plate with a colorimeter (Minolta CR300; Ramsey, NJ). In addition, a JPEG image was collected from scanning each plate with a flatbed scanner (HP Scanjet 3970) and analyzed with the color function of TA. L*, a*, b* values were recorded from each 47

64 method. Of the 247 standard color plates, 28 were chosen to span a range of colors observed in tomatoes. These plates were custom-made into a 28-patch color checker to be used for scanner calibration. We tested the precision of our color measurements with three different scanners: HP Scanjet 3970, HP Scanjet 5300C, and Microtek ScanMaker We collected a JPEG image of the 28-patch color checker at 200 dpi with each scanner. Each of the 28 patches was considered an individual object and was analyzed for color. We collected RGB data and converted it to estimates of L*, a*, b* measurements for each patch. We also obtained colorimeter L*, a*, b* data from each patch, and applied linear regression to determine slopes, y-intercepts and regression coefficients. Plant material An inbred backcross (IBC) population derived from crosses with Solanum lycopersicon processing cultivars (OH832, OH8245, OH9241, OH9242) was evaluated for color and color uniformity. The original F1 crosses were OH9242xOH8245, OH832xOH8245, and OH9241xOH8245. Each F1 was backcrossed twice to the recurrent parent, OH8245, to obtain BC2 populations, which were then selfed to the BC2S4 generation. The IBC population was grown near Fremont, Ohio, at the Ohio Agricultural Research and Development Center (OARDC) North Central Agricultural Experimental Station in 2004 and An augmented design was implemented in both years to evaluate each IBC genotype 48

65 (n=179, r=1) and replicated checks (n=4, r 5). Each plot consisted of 20 plants per genotype spaced 30 cm apart, with plots spaced 150 cm apart. All field plots were planted and maintained following conventional practices (Precheur et al, 2004). The plots were harvested when 80% of the fruits were ripe. Phenotypic data collection. Measurements of color were collected from digital images that were analyzed by TACT and the colorimeter (Minolta CR300; Ramsey, NJ). We used a flatbed scanner (HP Scanjet 3970) to scan the cut surface of the proximal end for 12 fruits within each plot and saved a JPEG image. Images were analyzed and color data were collected for two data sets: TA_Unadj, for which no modification was made to the fruit boundaries, and TA_Adj, for which the boundaries were adjusted when needed to better represent the proximal end of the fruit to analyze. Figure 3.2 illustrates the boundaries with the two TACT methods used to generate the TA_Unadj and TA_Adj data sets. The batch feature of TACT was used to analyze both data sets, whereby images can be selected per batch and analyzed for color. Each data set generated from TACT consisted of L*, a*, b*, hue and chroma values to represent absolute color. For tomato, improved color is characterized by lower L*, b* and hue values and higher a* values for a more intense, red color. The interpretation of chroma is ambiguous as high chroma due to high b* values represents poor color, while high chroma due to high a* values represents good 49

66 color. In addition, we measured color uniformity defined by the parameters %YSD and %RED. Two color measurements using the colorimeter were also collected from the same fruits that were scanned for TACT color analysis. The two point measurements were taken on mature red tissue and any discoloration present on the fruit shoulder. The difference between the two measurements, ΔHue and ΔChroma, provided an estimate of internal fruit color uniformity that is consistent with visual symptoms of YSD. Statistical analysis All statistical analyses were performed using SAS software (v. 9.1; SAS Institute Inc, Cary, NC). Color standard data from colorimeter and TA were tested for normal distribution using the UNIVARIATE procedure. To determine the relationship between color data generated from the colorimeter and TACT, we used the regression procedure (PROC REG) to test the significance of the regressions. The estimates of variance and standard errors were obtained using the restricted maximum likelihood (REML) method with the mixed model analysis of variance procedure (PROC MIXED). The model to estimate variance components for the IBC population was: Y ijk = μ + G i + T j + GT ij + ε ijk where Y ijk was the color trait measured, μ was the overall mean, G i was the effect due to the i th genotype, T j was the effect due to the j th year, GT ij was the effect due to 50

67 the genotype by year interaction, and ε ijk was the experimental error. The percent total variance was reported to allow direct comparison between data sets. Using the estimates of variance components, broad-sense heritability (H) was determined using the following relationship: H = σ 2 G / (σ 2 error/2 + σ 2 G + σ 2 Y/2 + σ 2 GY/2) We assumed selection based on IBC line means across two years. To test whether methods of collecting data influenced the amount of phenotypic variance that we could partition into genotypic variance, the standard error associated with these estimates was used to perform a mean separation at α=0.05. RESULTS Correlation between methods To determine the precision and accuracy of the color data generated with TACT, we compared it to a colorimeter. Color data (L*, a*, b* values) were collected for each of 247 color standards using a colorimeter as well as TACT. The regressions of L*, a*, b* values from the colorimeter onto TACT values showed a significant (P<0.0001) linear relationship for all three parameters, with correlation coefficients greater than 0.96 (Figure 3.3). Despite the linear relationship, the values between the colorimeter and TACT differed because the slope was not equal to 1 and the y-intercept was not equal to 0. The strong linear relationship suggested that calibration of the scanner used to generate digital images could be accomplished with simple adjustments to the equations used to calculate L*, a*, b* values. 51

68 Calibration Three flatbed scanners were used to assess reproducibility and systematic differences in scanning devices for measuring L*, a*, b* values from JPEG images (Figure 3.4). The correlations between L*, a*, b* values measured from the colorimeter and TA were high (r>0.94) for all three scanners for each trait. Correlation values among scanners were the highest for L*, lightness, followed by a*, which measures color range from green to red. The lowest correlation values were found for b*, which measures color range from blue to yellow. However, we observed differences among the three scanners in slope and y-intercept values as with the correlation of colorimeter and TACT. For this reason, we implemented an option in the TACT dialog box (Figure 3.1) to enter correction values for slope and y-intercept as a way to calibrate the device used in collecting images. The correction values are entered as the inverse of the slope and the opposite of the absolute value of the y-intercept for each of L*, a* and b* (Table 3.1). Variance partitioning with Tomato Analyzer Color Test To test whether TACT offered advantages over the colorimeter, we evaluated a breeding population for color and color uniformity using both approaches. Variance components for genotype, year, and the interaction genotype by year were estimated to elucidate the proportion of genotypic variance associated with each color parameter (Table 3.2). Among the three methods, the total phenotypic variation partitioned into genotype and genotype-by-year interaction ranged from 52

69 12% to 30%. The variance partitioned into year ranged from 0% to 9.6%. However, significantly more variance was partitioned into genotype using TACT than the colorimeter for all traits except chroma (Table 3.2). The proportion of genotypic variance for L*, a*, b* measured with TACT was 2- to 4-fold greater than with the colorimeter. The ability to partition a greater portion of the phenotypic variation into genetic effects increases the potential for improving a trait by means of genetic manipulation. In addition to evaluating absolute color (e.g. L*, a*, b*, hue and chroma), color uniformity was also measured by adjusting two user-defined parameters in TACT. We defined those parameters as %YSD and %RED. In addition, we estimated color uniformity from colorimetric data as Δ Hue and Δ Chroma. Using TACT to measure color, we partitioned approximately 10% of the total variation for %YSD and 17% for %RED into the genotypic variance. With the colorimeter, only 0% and 4% of the total variance was partitioned into the genotypic variance for Δ Chroma and Δ Hue, respectively. In addition, the error was lower for TACT than for the colorimeter. Overall, 66-83% of the total phenotypic variance was in the error term for TACT, compared to 77-96% for the colorimeter. In our experimental design, error variance is equal to within-plot variance and complex interactions not controlled for in our sampling. The total phenotypic variance was partitioned significantly better into genotype, year and genotype-by-year interaction with TACT compared to the colorimeter. The estimates of genotypic variance correspond to estimates of heritability that ranged from 0.11 to 0.17 for chroma, a* and %YSD, 53

70 and to for L, b*, hue and %RED when TACT was used. With the colorimeter, H estimates were lower and ranged from to to L*, a* and chroma, and to for b* and hue. With greater heritability estimates, we could achieve greater gains under selection for genetic improvement by measuring color with TACT. The user has two options for analyzing digital images, one in which the images are adjusted to better approximate the boundaries prior to color analysis, the other in which the images are analyzed according to the automated boundary definition. We generated data sets to test these two approaches: TA_Adj and TA_Unadj (Table 3.2). The TA_Adj data set was compiled from images where boundaries were manually adjusted as needed. The second method, TA_Unadj, was used without making adjustments to the images prior to color analysis. TA_Unadj is less time-consuming, but is prone to include parts of the fruits, such as peel, deep cracks, or reflected light from the scanner that may bias the color values. However, between the two TA methods, there was no significant difference in the variance partitioned into genetic effects for all color parameters (Table 3.2). Application of Tomato Analyzer Color Test to other crops We demonstrated that TACT is capable of measuring color and color uniformity at the proximal end of tomato fruits with accuracy and precision. We sought to evaluate TACT with fruit and vegetable crops other than tomato. We tested a red-skinned potato, cucumber, red plum, cantaloupe, carrot and strawberry. 54

71 These crops consist of various colors and color uniformity (Figure 3.5). TACT recognized the boundaries with precision for each crop, as expected with the contrasting colors between the background and the object (Brewer et al, 2006). By converting the RGB values to L*, a*, b* values (Table 3.3), the red-skinned potato had the highest L* value, which represents the brightest, closest to white tone. TACT was able to detect a* values from -8 to 38, with the highest value given to the carrot with its orange tone. The carrot also had the highest b* value, lowest hue and highest chroma, which conveys deep orange-red tone. To test the capability of measuring color uniformity, we defined the hue range of the first parameter to for the proportion of pixels that fell into yellow-light yellow range. The red plum and cantaloupe had 72-77% of the pixels falling into that range; the strawberry had 30% (data not shown). For the second parameter, we defined the hue range to 0-48 for the proportion of pixels that fell into red-orange range. The carrot had 25% of the pixels falling into that range, whereas the strawberry had only 15% of red-orange tissue. Subjective evaluations or singlepoint measurements, as in a colorimeter, do not provide this level of precision in determining the proportion of specific hues. This shows that TACT can perform color analysis on a broad range of hue values and is not confined to the hue ranges common to the tomato fruits. 55

72 DISCUSSION Accuracy, precision and reproducibility are important criteria when measuring a trait. Color may be measured with a colorimeter, spectrophotometer, or computerized image analysis. We developed a new module in Tomato Analyzer to collect objective color measurement based on digital images. This automated tool analyzes each pixel of a selected object, then translates it from RGB to L*, a*, b* values. Our first objective was to test the accuracy of TACT against a colorimeter and to provide a calibration to account for differences in scanner. Empirical results suggest that differences exist between devices either due to hardware, software, or non-standardized RGB values. Digital images collected from different sources can vary in color depending on the resolution, light source, and light quality. Three different scanners were used to scan color standards and test the precision among scanners with TACT. Although the L*, a*, b* values computed from TACT correlated highly with those of the colorimeter for all three scanners, the slope and y- intercept values varied among scanners. Therefore, an option in TACT was developed to incorporate these values as a correction for L*, a*, and b*. Previously, an attempt to eliminate variability in brightness and color distribution due to scanner differences was reported (Shahin and Symons, 2003). However, the various calibration techniques exploited did not result in satisfying performance. For example, the color ranges varied dramatically between the corrected and reference images, which is not amenable to standardization or comparative studies. In 56

73 contrast, our technique in calibrating TACT was successful, user-friendly, reliable and malleable. Our second objective was to assess which technique can better partition observed phenotypic variation for color into genotypic variances. Breeders and geneticists seek tools that allow precise trait measurement in order to extract the most genotypic variance. Compared to colorimetric data, the proportion of total phenotypic variance attributed to genotypic variance was significantly improved as it nearly doubled for all color and color uniformity traits when collecting data with TACT. Estimates of genotypic and phenotypic variances are the basis for determining heritability, which in turn provides an insight into the expected genetic gain and genetic improvement in a breeding program. We show greater heritability estimates for all traits measured with TACT, with the highest value for %RED. Moreover, the parameters for color uniformity measured with TACT, %YSD and %RED, are more informative in terms of variance estimates than the negligible estimates for ΔHue and ΔChroma measured with the colorimeter. We hypothesize that is the case because color uniformity is better characterized when the entire surface is evaluated rather than the difference between two point measurements for hue and chroma. Also, among the methods available with TACT, correcting the boundaries of the tomato fruit cut surface prior to color analysis (TA_Adj) does not partition significantly more of the genotypic variance from the total phenotypic variance compared to the analysis with unadjusted boundaries (TA_Unadj). The 57

74 TA_Unadj is an appropriate method for high throughput phenotyping as it does not require manual adjustments before color analysis. Image analysis for color is becoming more prevalent than the use of colorimeter or spectrophotometer. It facilitates data collection and management, and it requires equipment that is relatively more affordable. TACT was designed to be user-friendly with minimum requirements for running it, yet accurate and precise for collecting objective measurements. In comparison to the various methods presented in the literature, the tools implemented for measuring color require extensive environmental control, especially for the quality and quantity of light, shadow and reflection. The flatbed scanner used to generate digital images for TACT color analysis only requires a cardboard box to cover the scanner and minimize the effect of shadow. For precision in the analysis of images generated from multiple scanners, we designed a user-friendly option for calibration in the dialog box of TACT. Other methods require a calibration of the system before each image is taken (Lootens et al, 2007) or with minimal calibration feasible (Wang-Pruski, 2006). We foresee the application of TACT for color analysis from digital images to be more accurate and precise, and less expensive. Tomato Analyzer, as the name implies, was originally designed for the analysis of the morphology of tomato fruits. We implemented a module for color measurement for objective phenotypic analyses. It was tested with other fruits and vegetables of various color and color uniformity. Overall, it was able to accurately capture and describe the characteristic color for each crop. Color uniformity was 58

75 also well characterized for fruits that tend to have non-uniform pigmentation, as in the strawberry (Figure 3.5). Its application could go beyond the color analysis of fresh crops. In food science, discoloration after processing or cooking can occur and is an important issue. Wang-Pruski (2006) reports the acquisition of potato tuber images to evaluate after-cooking darkening. Such discoloration could be measured with TACT by defining the specific range of hue values that best represent the undesired discoloration. We show TACT is a tool that is reliable, precise, amenable and affordable for digital image analysis of color. ACKNOWLEDGMENTS I would like to thank the following people for contributing to this chapter. David Francis provided scientific guidance, and the resources to complete this study. Esther van der Knaap initiated and pursued the project for the development of the Tomato Analyzer. Simon Gray, Nancy Dujmovic and David Sullivan, from the Department of Mathematics and Computer Science at the College of Wooster, Ohio, designed and optimized the software application for the Tomato Analyzer Color Test. Jack Hall collected data from color standards for calibrating scanners. I thank Bert Bishop from Statistics and Computing Services at the OARDC/OSU in Wooster, Ohio, for his suggestions on the analysis of the IBC population. I appreciate the help from Troy Aldrich for establishing field plots. The development of Tomato Analyzer and its Color Test was supported by NSF (grant no. DBI ) to Esther van der Knaap. 59

76 TABLES L* a* b* y- y- Scanner (Make, Model) r2 Slope intercept r2 Slope intercept r2 Slope y- intercept HP Scanjet HP Scanjet 5300C Microtek Scan Maker Table 3.1: Correlation values and regression characteristics of L*, a*, b* values for standard color plates obtained with Tomato Analyzer Color Test and colorimeter. 60

77 61 Proportion (%) of REML variance estimates Method Variance Component y L* a* b* Hue Chroma Δ Hue Δ Chroma % YSD % Red TA_Adj Var(Genotype) a,z a a a a a a Var(Year) Var(Genotype*Year) Var(Error) TA_Unadj Var(Genotype) a a a a a a a Var(Year) Var(Genotype*Year) Var(Error) Colorimeter Var(Genotype) b b b b a Var(Year) Var(Genotype*Year) Var(Error) y z Data were collected in 2004 and 2005 from plots evaluated in Fremont, Ohio. The letter following the estimate of Var(Genotype) represents the statistical grouping for the comparison of each method per trait. Methods for estimating Var(Genotype) in different groupings are significantly different (α=0.05). Table 3.2: Proportion of variance estimates for color measurements using Tomato Analyzer Color Test with the TA-defined boundaries (minimum blue value = 30; TA_Unadj) and adjusted boundaries (TA_Adj).

78 Crop R G B Luminosity L* a* b* Hue Chroma A, Red-skinned potato B, Cucumber C, Red plum D, Cantaloupe E, Carrot F, Strawberry Table 3.3: Average values of color parameters obtained from the output of the Tomato Analyzer Color Test for a variety of fruits and vegetables. Images of these crops appear on Figure

79 FIGURES Figure 3.1: Tomato Analyzer and its Color Test. The dialog box in the center of the image allows the user to customize the color parameters for analysis (top tier) and to enter the correction values for calibrating the scanner, as well as other options (bottom tier). 63

80 Figure 3.2: Representation of the tomato proximal end (shoulder) analyzed for color with Tomato Analyzer Color Test. Images C and D show symptoms of yellow shoulder disorder (YSD), a ripening disorder that affects color uniformity. A, Uniform fruit analyzed with TA_Unadj method. B, Uniform fruit analyzed with TA_Adj. C, YSD-affected fruit analyzed with TA_Unadj. D, YSD-affected fruit analyzed with TA_Adj. 64

81 Tomato Analyzer values L* y = x R 2 = P< Tomato Analyzer values a* y = x R 2 = P< Tomato Analyzer values b* y = 0.976x R 2 = P< Colorimeter values Figure 3.3: Correlation between Tomato Analyzer Color Test and colorimeter values for L*, a*, b* values of the CIELab color space using 247 standard color plates. 65

82 L* 100 Tomato Analyzer values a* 100 Tomato Analyzer values Tomato Analyzer values b* Colorimeter values Figure 3.4: Regression of different scanners for L*, a*, b* values of the CIELab color space from standard color plates spanning a range of colors observed in tomato fruits measured with Tomato Analyzer Color Test and a colorimeter. The scanners used to assess scanning quality for color were HP ScanJet 3970 (asterisk), HP ScanJet 5300C (triangle) and Microtek 6000 (square). The regression values are summarized in Table

83 Figure 3.5: Images of various fruits and vegetables evaluated with Tomato Analyzer Color Test for objective color measurements. A, Red-skinned potato. B, Cucumber. C, Red plum. D, Cantaloupe. E, Carrot. F, Strawberry. 67

84 CHAPTER 4 DISSECTING VARIATION IN TOMATO FRUIT COLOR QUALITY THROUGH DIGITAL PHENOTYPING AND MAPPING ABSTRACT Color is used as a criterion for quality in fruits and vegetables. In the processing tomato, color quality affects the grower through contract incentives, the processor through grade, and the consumer through concentration of health-promoting carotenoids. We investigated color variation and conducted a QTL analysis to elucidate the genetic basis of tomato fruit color and color uniformity. We evaluated backcross and inbred backcross populations developed from an interspecific cross between Solanum lycopersicum (OH88119) and S. pimpinellifolium (PI128216), as well as elite and experimental processing varieties. We used Tomato Analyzer, a software application with its newly implemented Color Test, to measure the proximal end of tomato fruits. We collected data on L*, a*, b*, hue, chroma (based on CIELab color space) and the percentage of red and percentage of discoloration due to yellow shoulder disorder (YSD). Pearson correlations for all trait combinations revealed color parameters to be highly correlated. Therefore, we conducted principal component analysis to design indices in an attempt to reduce the 68

85 number of parameters defining color intensity and uniformity. We used single factor analysis of variance corrected for uneven number of progeny and non-parametric analysis, which was based on progeny that were worse than, equal to, or better than the recurrent parent, as statistical methods to detect marker-trait associations. Our results suggest three quantitative trait loci (QTL) for color intensity on chromosomes 2, 8 and 9, and a QTL for color uniformity on chromosome 6. We also show positive gain under selection for improved color based on phenotypic selection and marker-assisted selection. Our results can be exploited in elite germplasm for crop improvement and to further the understanding of QTL affecting color. INTRODUCTION Color in horticultural and vegetable crops is a prime quality attribute (Picha, 2006). Color quality of the processing tomato is defined for producers by contracts that provide incentives for uniform, deep red fruit. For processors, color quality is defined by grades that weigh off-color tissue; whereas for consumers, it is defined by its appeal and health-promoting benefits. Disorders such as yellow shoulder disorder (YSD) affect color uniformity due to discoloration at the proximal end of the fruit (Francis et al, 2000). YSD reduces carotenoids of tomato tissue and juice, mainly lycopene (Chapter 2; Darrigues et al, 2007). Thus understanding the genetic basis of YSD through precise trait measurement and quantitative trait loci (QTL) mapping will facilitate the development of varieties with improved color quality. 69

86 Carotenoids are the primary determinants of color, with lycopene and betacarotene being the prominent carotenoids in the tomato fruit. The carotenoid biosynthesis pathway in plants is well characterized (Cunningham and Gantt, 1998), with enzymes identified and genes cloned (Isaacson et al, 2002; Ronen et al, 1999; Ray et al 1992; Ronen et al, 2000; van Tuinen et al, 1997; Yen et al, 1997). However, individual candidate genes do not explain the range of phenotypic variation, which shows continuous distribution (Liu et al, 2003). The continuous variation of color and color uniformity in tomato populations makes it amenable to quantitative studies aimed at unraveling the complexity of the trait. A number of studies have reported quantitative trait loci (QTL) associated with color using different sources of germplasm, population structures and methodologies for measuring color (Tanksley et al, 1996; Fulton et al, 1997; Bernacchi et al, 1998; Fulton et al, 2000; Saliba-Colombani, 2001; Liu et al, 2003; Kabelka et al, 2004; Yang et al, 2004). These studies report color QTL on the majority of the tomato s 12 chromosomes, with no general consensus for the location of the QTL and the trait associated with the QTL. Due to the fact that different population structures, parental lines, and evaluation strategies are used, the lack of consistency between studies is expected. Additionally, the methodology for color measurement implemented in these studies varies considerably. Data collection for fruit color is reported from visual inspection of fresh tissue using a gradient scale (Liu et al, 2003), visual inspection of puree using a gradient scale (Fulton et al, 1997), objective evaluation of puree using 70

87 a colorimeter based on CIELab values (Fulton et al, 1997) or objective evaluation of internal tissue using a colorimeter based on CIELab values (Saliba-Colombani et al, 2001; Liu et al, 2003; Kabelka et al, 2004; Yang et al, 2004). A standard approach to collecting color measurements would facilitate comparisons of QTL among studies for tomato color. In this study, our objectives were to identify genomic regions associated with color and color uniformity in populations derived from the cross Solanum lycopersicum (OH88119) x S. pimpinellifolium (PI128216). We report the application of the Tomato Analyzer Color Test (TACT) to measure color and color uniformity objectively from digital images. The Tomato Analyzer software application is capable of translating the RGB values of each pixel from a digital image to L*, a*, b* values of the CIELab color space, which is a uniform color space. Color uniformity can also be evaluated from the proportion of pixels that fall into a specific range of desired or undesired color values. TACT permits reproducible and objective phenotypic evaluations of populations (Chapter 3). We performed principal component analysis (PCA) to determine which of the color parameters contributed most to the total variation within a population. Results from the PCA were used to develop indices that accounted for correlation among traits and provided a comprehensive description of color intensity and color uniformity. We conducted QTL analysis based on a single-factor analysis of variance (ANOVA) and a non-parametric analysis of an advanced backcross population (BC2), an inbred backcross population derived from the BC2 (BC2S4), an F2 population developed 71

88 from a BC2 individual, and elite germplasm. Our results suggest three QTL for color intensity on chromosomes 2, 8 and 9, and a fourth QTL for color uniformity on chromosome 6. Our findings show the potential of exploiting these QTL in elite, adapted germplasm for the improvement of tomato color quality. MATERIAL AND METHODS Plant material. The populations were developed from a cross between Solanum lycopersicum (OH88119) and S. pimpinellifolium accession (PI128216). Seeds from the F1 cross were grown and plants were crossed to the recurrent parent, OH88119, to generate a BC1 family of 94 individuals. These individuals were grown and plants were crossed to OH88119 to generate 94 BC2 families, referred to as the BC2 population. From each BC2 family, two random siblings were harvested and the seed was saved separately to generate 188 BC2S1 individuals. A BC2S4 population was derived from four generations of selfing via single seed descent from the 188 BC2S1 individuals with no selection. The BC2S4 population consisted of 178 individuals as some were lost in the process of generating the succeeding filial generations. To develop a population for QTL confirmation, an individual from the BC2S1 population was selected based on color and color uniformity. The selection ( ) was crossed with OH8243. OH8243 produces large fruits with uniform color (Berry and Gould, 1988). Although OH8243 has excellent color, it does not carry the og c allele, which confers high lycopene through a reduced activity of the 72

89 fruit-specific lycopene-beta-cyclase (Ronen et al, 2000). The testcross was selected based on the polymorphism of specific markers of interest and was selfed to obtain an F2 population (TC19F2; n=104). Twenty three inbred parents, commercial hybrids and experimental varieties were evaluated for color. The commercial varieties included: OH7983 (Berry et al, 1992), OH8245 (Berry et al, 1991), OX38 (Berry et al, 1995), OX52 (Francis and Berry, 2000) and OX150 (Francis et al, 2002). The experimental parents and varieties included: OH8556, OH9241, OH9242, OH87160, OH88119, OX42, OX237, OX239, OX241, OX262, OX266, OX268, OX271, OX285, OX286, OX289, OX297, and OX318. These varieties consist of elite germplasm with few introgressions from wild species. Field trials The BC2 and BC2S4 populations were grown and evaluated in 2004 and 2006, respectively, at the Ohio Agricultural Research and Development Center (OARDC) North Central Agricultural Experimental Station in Fremont, Ohio, and at the Department of Horticulture and Crop Science farm in Wooster, Ohio. The BC2 and BC2S4 populations were evaluated in an augmented experimental design each year at each location. In each experiment, four to six checks were replicated (r=3). The BC2 population was evaluated in two replicates in Fremont and one replicate in Wooster in The BC2S4 population was evaluated in a single replicate in both 73

90 Fremont and Wooster in Each plot consisted of 10 plants per genotype in Wooster and 20 plants in Fremont. Plants were spaced 30 cm apart within each row with 150 cm separating each row. The TC19F2 population was grown on a single plant basis and evaluated in Fremont only in The elite varieties were evaluated in advanced trials grown in Fremont, Wooster, and Wood County, Ohio. At each location, plots consisted of 20 plants with 30 cm spacing within row, 150 cm separating each row. All field plots were planted and maintained following conventional practices (Precheur et al, 2004). The plots were harvested when 80% of the fruits were ripe. Digital phenotyping for color Populations and elite varieties were evaluated for fruit color and color uniformity using the Color Test implemented in the software application Tomato Analyzer (TA) to collect objective measurements of color (Chapter 2; Brewer et al, 2006). As described in Chapter 3, Tomato Analyzer Color Test (TACT) is an image processing software application that recognizes and collects data from JPEG files. We used a flatbed scanner (HP Scanjet 3970) to scan the cut surface of the proximal end for 12 random fruits per plot and saved a JPEG image. All images were collected from scanning the fruits at a resolution of 200 dpi. TACT records RGB values of each pixel of the selected object on the image and translates these values into average L*, a*, b* values from the CIELab color space (Commission Internationale de l Éclairage; CIE, 1978). The L*, a*, b* values are then used to 74

91 calculate chroma as (a 2 +b 2 ). Hue is calculated as 180/pi*acos(a/ (a 2 +b 2 )) for a*>0 and as 360-((180/pi)*acos(a/ (a 2 +b 2 ))) for a*<0. The L* coordinate indicates darkness (~0) to lightness (~100) of color. The a* and b* axes are the chromatic coordinates and indicate color directions: +a* is the axis in the red direction, -a* is the axis in the green direction, +b* is the axis in the yellow direction and b* is the axis in the blue direction. Hue represents basic color as an angular measure, whereas chroma is the saturation or vividness of color. The output of the color analysis provides averaged values of R, G, B, L*, a*, b*, hue and chroma. For tomato internal tissue, improved color is characterized by lower L*, b* and hue values and higher a* values. Chroma is more difficult to interpret as high values of chroma resulting from high values of a* are desirable, while high values of chroma resulting from high value of b* are undesirable. In addition, we measured color uniformity. TACT allows the user to define two parameters based on specific hue ranges. We defined our parameters in TACT to measure discoloration due to yellow shoulder disorder (YSD). Percent YSD (%YSD) and percent red (%RED) were defined based on the hue ranges 60 to 120 and 0 to 48, respectively. Improved color uniformity was characterized by lower %YSD and higher %RED. DNA isolation and marker analysis Procedures for DNA extraction, PCR and electrophoresis used in the analysis of these populations were described previously (Kabelka et al, 2002; Yang et al, 2004; Yang et al, 2005a). A total of 352 primer pairs, consisting of SSR (simple 75

92 sequence repeat), INDEL (insertion/deletion) and SNP (single nucleotide polymorphism) type of markers, were screened for polymorphism between the two parental lines. We identified 70 markers that distinguished OH88119 and PI and used them to genotype the BC2 and BC2S4 populations (Table 4.1). The markers were added to a previous map from an F2 population derived from LA1589 (Solanum pimpinellifolium) and Sun1642 (S. lycopersicum) (Yang et al, 2004). The linkage map was constructed using the Kosambi mapping function of JoinMap (version 3.0; van Ooijen and Voorrips, 2001). DNA from TC19F2 plants was isolated and genotyped with markers that were significantly associated with color traits in the BC2 population, LEOH112 and LEOH200. The complete population of 104 TC19F2 individuals was genotyped, but only 23 homozygotes were retained for color analysis. In the BC2 population, genotypic data from molecular markers were scored as homozygous for Solanum lycopersicum OH88119 (LL) and heterozygous for S. lycopersicum x S. pimpinellifolium (LP). The expected allelic segregation was 87.5% (LL) and 12.5% (LP). The BC2S4 population was scored as homozygous for OH88119 (LL) and homozygous for S. pimpinellifolium PI (PP). The expected segregation was 93.4% (LL) and 5.86% (PP). Any heterozygotes detected in the BC2S4 population (expected frequency 0.78%) were entered as missing values due to their low frequencies, which tended to cause spurious results during statistical analyses. Segregation ratios of marker genotypes were analyzed by chi-square analysis. 76

93 Commercial and experimental varieties were evaluated for color and markertrait associations. Marker data for the parental varieties had been collected and archived in the Tomato Map database ( as these parental varieties were used extensively in commercial and experimental hybrids. We used LEOH23.3, SSR5 and SSR349a, markers mapped to chromosome 2 and polymorphic among the parental lines used to develop the varieties. This data set was utilized to confirm QTL on chromosome 2 affecting color intensity. Statistical analysis All statistical analyses were performed using SAS software (v. 9.1; SAS Institute Inc, Cary, NC). Analysis of variance (ANOVA) using the general linear model procedure (PROC GLM) was performed to test the significance of the main effects genotype (G) and location (L), and their interaction (GxL) for the BC2 and BC2S4 populations separately using the following statistical model: Y ijk = μ + G i + L j + GL ij + ε ijk where Y ijk was the trait or index measured, μ was the overall mean, G i was the effect due to genotype, L j was the effect due to location (Fremont and Wooster), GL ij was the interaction effect due to genotype by location and ε ijk was the experimental error. Genotype, location and GxL were considered random effects. The data set used for the ANOVA consisted of color measurements and index values for each fruit (12) per plot. The approximate F test for genotype and locations used the interaction GxL as the error term. 77

94 Pearson and Spearman rank correlations among locations for each population were estimated between the two replicates evaluated in Fremont for the BC2 population, and between Fremont and Wooster for the BC2 and BC2S4 populations. In addition, Pearson correlations were estimated for all trait combinations by location (Wooster and Fremont) and by population (BC2 and BC2S4) using the correlation procedure (PROC CORR). To determine which color traits or combination of traits contributed most to the phenotypic variance, we conducted principal component analysis (PCA) for the BC2 and BC2S4 populations in each location using the procedure PROC PRINCOMP. Inspection of the Eigenvector coefficients (loadings) were used to develop two indices: icoloruni and icolorint. The loadings (absolute value) from principal component 1 (PC1) for %YSD, %RED and hue averaged across locations and populations were used as weights assigned to the corresponding trait value. This index, icoloruni, describes color uniformity: icoloruni = 0.438(%YSD) (Hue) (%RED). The loadings from principal component 2 (PC2) for a*, b* and chroma averaged across locations and populations were used as weights assigned to the corresponding trait value. This index, icolorint, describes color intensity: icolorint = 0.514(b*) (a*) (Chroma). Lower values of icoloruni and icolorint were interpreted as improved color. 78

95 We developed a third index based on a subjective determination of optimal color for a tomato destined for whole-peel and diced product. This index incorporated cut-off values for color and color uniformity. This index, icoloropt, is: icoloropt = %YSD/20 * (100-%RED)/55 * Hue/48 * L/40 The coefficient below each parameter indicates the optimal values observed in commercial varieties evaluated in 2004 and 2006 in Fremont. The index incorporates the most important features for color and color uniformity, including L*, hue, %YSD and %RED. icoloropt can be distinguished from the PC-based indices with the inclusion of L*, a parameter that describes brightness of color. icoloropt is normalized to a scale centered at 1. An index value of 1 represents good color and color uniformity, whereas an index value less than 1 is considered improved color and color uniformity. An index value greater than 1 indicates poor color relative to commercial tomato varieties. Marker-trait associations The statistical models for testing marker-trait associations using analysis of variance and the rationale for implementing these models were described previously (Kabelka et al, 2002). The following model was used to test the linkage relationship between the genotypic classes and the color parameters for the BC2 population evaluated in Fremont in two replicates and for the combined locations: Y ijk = μ + M i + G(M) j(i) + R k + ε ijk 79

96 where Y ijk was the trait measured (L*, Hue, Chroma, %YSD, %RED and the three indices), μ was the overall mean, M i was the effect due to marker class, G(M) j(i) was the effect due to genotype nested within marker, R k was the effect due to replication (replicate or location) and ε ijklm was the experimental error. All factors in the model were considered random. The approximate F test for marker-trait associations (M i ) used the genotype within marker term (G(M)) as the error term. For the evaluations of the BC2 population in Wooster, the BC2S4 population in Fremont and Wooster, and the TC19F2 population in Fremont, and the elite varieties, the following model was used: Y ij = μ + M i + ε ij where Y ij was the trait measured (L*, Hue, Chroma, %YSD, %RED and the three indices), μ was the overall mean, M i was the effect due to marker class, and ε ij was the experimental error. Degrees of freedom were determined using the Satterthwaite approximation in PROC GLM, which provided a correction for lack of balance between marker classes that characterize advanced backcross and inbred backcross population structure. The approximate F test for marker-trait associations (M i ) was Mi/ε ij. Significant F tests (P<0.05) provided evidence for linkage between a marker and a color trait parameter or index. We defined a robust association of a marker with a trait when it was detected in at least two locations. The total phenotypic variation explained by marker loci (Vm/Vp) was determined from variance 80

97 component estimates by restricted maximum likelihood (REML) using the mixed model analysis of variance procedure (PROC MIXED). Because of the differences in the distribution of variance for the color traits measured in Fremont and Wooster, a non-parametric analysis (NPA) for single marker analyses was also performed to confirm results from the ANOVA. For each location, mean comparisons were based on the least significance difference (LSD) at α=0.05 between individual lines in the BC2 and BC2S4 populations and the recurrent parent (OH88119). The LSD was used to categorize the genotypes as better than, worse than, or equal to OH To test whether the markers were significantly associated with the traits, we used the non-parametric, one-way procedure (PROC NPAR1WAY) and the Kruskal-Wallis test. Only associations found in a combination of two locations in the BC2 and/or BC2S4 populations are reported. Gain under selection and heritability Because we did not impose any selection during the development of the BC2 and BC2S4 populations, gain under selection was determined based on reference to the unselected BC2S4 population mean performance averaged over the two locations. Phenotypic selections based on the three indices in the BC2 population consisted of lines with values greater than or equal to one standard deviation from the BC2 population mean. The performance of the progeny from the selected BC2 81

98 lines was determined in the BC2S4 population, which represented the selected mean (U S,BC2S4 ). The mean of all BC2S4 lines constituted the unselected mean (U BC2S4 ). Estimates of gain under selection were determined based on phenotype only and on genotype plus phenotype (MAS). MAS consisted of an initial selection based on marker data, followed by a selection based on phenotype in the reference population. Because of the more extensive marker coverage on chromosome 2, we determined gain under selection for color intensity based on marker intervals along the chromosome. LEOH147 on chromosome 8 was used to test MAS for color intensity, and LEOH112 on chromosome 6 for color uniformity. We determined the mean of the BC2 lines with the favorable marker allele(s) for each QTL (interval or single marker) associated with each index. The selections were based on values greater than or equal to one standard deviation from the mean of the BC2 population. The selected and unselected means of the BC2S4 population were determined as described above. The realized genetic gain (R) based on phenotype and MAS was calculated as the difference between the selected mean and the unselected mean of the BC2S4 population (U S,BC2S4 - U BC2S4 ). The standard error of R was calculated as: [(nk)/nk] 1/2 (SEM), where n and k represent the number of individuals in the unselected and selected means, respectively, and SEM is the standard error of the means. A value of R<0 indicates positive gain under selection for all three indices. Selection differential (S) was determined as the difference between the mean of the selections in the BC2 population (U S,BC2 ) and the mean of all the BC2 lines, which constituted 82

99 the unselected mean (U BC2 ). Realized broad-sense heritability (h 2 Realized) was estimated as h 2 Realized= R/S (Falconer, 1981). When estimating and interpreting realized heritability, we assumed no epistatic effects. RESULTS Phenotypic data analysis The frequency distribution of each color and color uniformity trait in the BC2 and BC2S4 populations showed continuous quantitative variation (Figures 4.1 and 4.2). The spread of the distribution of the two populations at each location exceeded the recurrent parent, OH88119, with lines showing both improved and worse color. The ANOVA test for the main effects of genotype and location in each population showed significant differences for all traits except %RED, for which the genotype effect was not significant in the BC2 population, and chroma, for which the genotype and location effects were not significant in the BC2S4 population (data not shown). Analysis of phenotypic data revealed that the incidence of YSD was significantly lower (P<0.0001) in 2004 compared to In 2004, the average %YSD was 17.9 and 12.6 in Fremont and Wooster, respectively. In 2006, the average %YSD was 54.8 and 20.8 in Fremont and Wooster, respectively. The proportion of good red color (%RED) was higher (P<0.0001) in 2004 with an average %RED of 45.7 and 56.0 in Fremont and Wooster, respectively. In 2006 the average %RED was 13.6 and 40.1 in Fremont and Wooster, respectively. Within 83

100 each year, the Wooster location had significantly (P<0.0001) lower %YSD and higher %RED than Fremont. The Fremont location is known to have a higher incidence of YSD than Wooster, probably due to lower levels of available phosphorous and potassium in soils (Clivati-McIntyre et al, 2007). We observed different weather patterns between Wooster and Fremont in 2004 and Overall, Wooster tended to have higher precipitation and higher average air temperature in 2004 compared to Fremont. In 2006, the reverse trend was observed. The most notable differences between locations were observed in the month of September, when the bulk of the fruits were harvested. The presence of YSD and %RED are good indicators of color quality and uniformity. The analysis of phenotypic data revealed several features that needed to be taken into account prior to marker-trait analysis. The analysis of GxE by ANOVA across locations for each population showed a highly significant interaction effect for all traits and indices (P<0.0001), which implies a rank shift for lines in the population evaluated in different environments. Pearson correlations were estimated between locations for each population (Table 4.2). For the BC2 population, no significant correlations were detected for any of the traits between Fremont and Wooster. For the BC2S4 population, there were significant correlations for %YSD, %RED, L*, a* and hue but no significant correlations for b* and chroma across the two locations. Spearman correlations were also estimated to confirm a possible rank shift of the population lines. Between the two replicates evaluated in Fremont for the BC2 84

101 population, Spearman correlations were significant for a* (P=0.0078), b* (P=0.0010) and chroma (P<0.0001). Among locations in the BC2 population, all correlations were non-significant. In the BC2S4 population, Spearman correlations were significant for %YSD (P<0.0001), %RED (P=0.0005), and a* (P=0.0336). When significant effects of GxE have been reported in the literature, researchers have opted to either ignore GxE by using means of trait values across environments (e.g. Yousef and Juvik, 2001), avoid GxE by keeping environments (year, location, etc) separate (e.g. Sim et al, 2007), or exploit GxE via stability analyses (e.g. Rane et al, 2007). Because our aim was to determine the genetic component of color and color uniformity as opposed to selecting for superior lines, we chose to avoid GxE by performing a QTL analysis for each population for all color and color uniformity traits by location separately. We also estimated Pearson correlations among the color traits for each population and each location (Table 4.3). Overall, all traits were significantly correlated except for chroma in the BC2 population and between a* and b* in the BC2S4 population. When examining the high coefficients (r>0.80), hue (the lower the hue value, the better) was highly correlated with %YSD and negatively correlated with %RED in both populations across the two locations. When comparing the correlation between %RED and %YSD across the populations, the coefficients were smaller in the BC2 population and higher in the BC2S4 population. The correlation between chroma and b* was also high across populations and locations, with higher correlation values in the BC2S4 population. 85

102 PCA is useful to determine if a combination of the color and color uniformity traits could better explain the phenotypic variability for color. Thus principal components (PC) were calculated for all trait values for each population (BC2, BC2S4 and TC19F2) in each location (Table 4.4). The absolute magnitude and direction of the Eigenvector coefficients (loadings) determine the importance and relationship between the traits within a component (Stevens, 1986). The first two PCs contributed to an averaged total of 85% of the phenotypic variation in the BC2 and TC19F2 populations and 91% in the BC2S4 population, with 52-62% of the variation in PC1 alone (Table 4.4). The most important contributors to PC1 were components that account for color uniformity, %YSD, %RED, and hue. These results suggest that 52-62% of the variation for color in our populations was characterized by color uniformity (PC1). PC1 provides objective data suggesting the importance of color uniformity in contributing to variation in color. The proportion of variance for PC1 was 15% higher in the BC2S4 population than in the BC2 and TC19F2 populations. This trend was consistent with higher incidence of YSD in 2006 when the BC2S4 population was evaluated. The most important contributors to PC2 were components that account for color intensity, a*, b*, and chroma. PC2 explains 30% of the variation for color, which was fairly consistent across all three populations. In spite of the location effect, there was a marked consistency for PC1 and PC2 between locations for the BC2 and BC2S4 populations. For the TC19F2 population, the 86

103 absolute magnitude of the loadings was similar to the BC2 and BC2S4 populations, but the direction was opposite. The third PC (PC3) contributed 4% to 8% to the total variation (data not shown). The traits most affected by PC3 were L* and %RED with loadings ranging from 0.73 to 0.83 for L* and from 0.46 to 0.62 for %RED. Because of the small contribution from PC3 and the lack of consistency between loadings, we concentrated on further analysis using the first two PCs, which explained the most variation for color between lines in the populations. The results of PCA for BC2 and BC2S4 suggest that PC1 and PC2 could be exploited in the development of indices for color, with the PC1-based index representing color uniformity and the PC2-based index representing color intensity. To exemplify the merits of the PC-based indices (icoloruni and icolorint), as well as the index derived from optimal values of color parameters (icoloropt), digital images of fruits were ranked according to their index values (Figure 4.3). All three indices were able to distinguish lines with a high average of %YSD from those lines with uniform red fruits. icoloruni and icoloropt are similar indices as they both capture color uniformity and the desired red color (hue<48). Of the best and worst selections for icoloruni and icoloropt, the same lines from the BC2S4 population were selected in three out of the four selections. Selections based on icolorint had the highest values of a* (more red), lower values of b* (less yellow), and the highest values of chroma, a parameter that is calculated from a* and b* values. The icolorint index allows the intensity of color to be expressed to identify desirable 87

104 fruit. Chroma alone cannot be used for this purpose because high values can result from high values of b* and low values of a*, which is not desired due to its poor color. Overall, we were able to combine traits that describe color intensity and color uniformity, the latter accounting for most of the variation when measuring color, and develop efficient indices with potential for phenotypic selection. Genotypic analysis The marker data was obtained from genotyping the populations with 70 markers. Most of these markers were mapped with an average of 5 markers per chromosome (Figure 4.4). The genetic map spanned 1114 cm, resulting in an average marker distance of 16 cm. Our goal was to identify two markers per chromosome arm; however the marker coverage was uneven with only one or two markers on chromosomes 5 and 7, and 12 markers on chromosome 3. Four of the 70 markers were polymorphic among the parents but were fixed in the population, suggesting that PI was a heterogeneous accession or that the genome of PI was not captured in its entirety. Of the four monomorphic markers, one mapped to chromosome 1 and one to the top of chromosome 8; the map position of the other two markers is currently unknown. In the BC2 population, chi-square analysis showed significant deviation from the expected segregation based on the genotype of the BC2 individuals (75% LL and 25% LP) for 93% of the markers scored, with only 7% of the markers segregating as expected. Segregation distortion did not favor the LL genotype over the LP 88

105 genotype. In the BC2S4 population, deviation from the expected segregation (86% LL and 11% PP using the BC2S3 genotypic segregation frequencies expected in the evaluation of the BC2S4 population) was significant for 8.3% of the markers scored, with 92% of the markers segregating as expected. Segregation distortion in the BC2S4 population was in favor of the LL genotype for each marker with distorted segregation. Marker-trait associations with color in the BC2 and BC2S4 populations Analysis of marker-trait associations with multiple related traits can lead to difficulties in interpreting the results. Our original color data set consisted of five color parameters (L*, a*, b*, hue and chroma) and two color uniformity parameters (%YSD and %RED). We found these traits to be highly correlated and influenced by the environment. The distribution of mean values and variance for each trait differed considerably between the Fremont and Wooster locations and between populations (Figures 4.1 and 4.2). The distribution of trait values and variance in the data led to a situation where a marker might be associated with some parameters of color in one environment (i.e. any combination of location and population) and other parameters in another environment. For example, the association of SSR192 with L* was detected in Fremont and with a* and chroma in Wooster in the BC2S4 population. Despite these associations, it was not considered as evidence of a robust association for color intensity despite the fact that a locus for color was implicated in both environments. 89

106 To facilitate the analysis of marker-trait associations, we proposed two objective indices that captured the essential features of color, namely color uniformity (icoloruni) and color intensity (icolorint). Optimal color (icoloropt) was a third index developed based on our subjective determination of an ideal whole-peel tomato. Using these indices, we detected four genomic regions on chromosomes 2, 6, 8 and 9 that were significantly associated with one or more of the indices (Table 4.5). Associations were detected on chromosome 2 for color intensity with icolorint. The LL genotype was associated with improved color with lower values of icolorint. In both populations, the associations with marker fw2.2 were detected in Fremont only. Associations with marker SSR349a were specific to the BC2 population. The region covered by markers fw2.2 and SSR349a explained 16% to 27% of the phenotypic variation for icolorint. Examination of associations between markers on chromosome 2 and other color variables supported the involvement of a locus or loci on chromosome 2 with color intensity. For example, we detected associations in one environment only for LEOH348 with a* (P=0.0026), chroma (P=0.0079), and icolorint (P=0.0040); for SSR5 with a* (P=0.0273), and icolorint (P=0.0338); for SSR26 with a* (P=0.0126), b* (P=0.0143), chroma (P=0.0061), and icolorint (P=0.0156); and for SSR66 with a* (P=0.0493) (Appendix C). These results provided further evidence of a region on chromosome 2 controlling color intensity that is localized to the interval spanned by SSR66 to fw2.2 (Figure 4.4). 90

107 A region controlling color uniformity was found on chromosome 6 based on associations detected for icoloruni and icoloropt. The LP genotype in the BC2 population and the PP genotype in the BC2S4 population were consistently associated with improved color defined by lower values of icoloruni and icoloropt. In the BC2 population, the associations were detected in Wooster and/or Fremont. In the BC2S4 population, the associations were detected in Wooster only. The region covered by markers LEOH112, LEOH200 and LEOH17 explained 3% to 12% of the phenotypic variation for color uniformity. We found further evidence in one environment only for a region on chromosome 6 associated with color variables %RED (0.0164<P<0.0390), %YSD (0.0061<P<0.0475), Hue (0.0149<P<0.0236), L* (0.0018<P<0.0433), and chroma (0.0007<P<0.0100) that spans markers CT10242i to LEOH112 (Appendix C; Figure 4.4). A region on chromosome 8 was also shown to contribute to color intensity. The LL genotype in both populations contributed to improved color. The associations for icolorint were Fremont-specific in both populations. Marker LEOH147 explained 6% to 9% of the phenotypic variation. Another region on chromosome 9 controlled color intensity, for which the LL genotype contributed to more intense color in both populations. The associations were detected at both locations for the BC2 population. In the BC2S4 population, they were detected in Fremont only. The association with marker SSR70 explained 2.3% to 12% of the phenotypic variation. 91

108 Non-parametric analysis (NPA) The NPA provided a second approach to detect marker-trait associations in the BC2 and BC2S4 populations. The logic of this approach is that marker genotypes associated with improved or decreased color should be over-represented in extreme phenotypic classes relative to the recurrent parent, OH88119, for each color index. Results of the NPA revealed five chromosomal regions controlling color and color uniformity (Table 4.6). Chromosomes 2, 8, and 11 showed significant associations for color intensity with icolorint. Regions on chromosomes 4 and 7 were associated with color uniformity with icoloruni and icoloropt. Another region was associated with SSR71, an unmapped marker. Overall, using the two statistical methods to detect QTL (ANOVA and NPA), regions on chromosomes 2 and 8 were detected consistently in the BC2 and BC2S4 populations for improved color intensity and uniformity. The QTL on chromosome 6 detected by ANOVA was also found in the BC2 and BC2S4 populations using NPA (Appendix C). These associations were detected at a single location only across the populations. Associations detected using NPA with other color variables %RED (0.0146<P<0.0201), %YSD (P=0.0498), L* (0.0041<P<0.0147), and chroma (0.0001<P<0.0171) provided further evidence for a color QTL on chromosome 6 that spans markers CT10242i to LEOH112 (Appendix C). 92

109 Marker-trait associations with color in the TC19F2 population The TC19F2 population was developed to confirm the region on chromosome 6 controlling color uniformity. The associations with the color indices detected in the BC2 and BC2S4 populations were also found in the TC19F2 population (data not shown). The significance of the associations with LEOH112 and LEOH200 ranged from P= to P= for color uniformity, and from P= to P= for color intensity. In this population, the LL genotype contributed consistently to improved color and color uniformity. This result is inconsistent with the allelic effect observed in the BC2 and BC2S4 populations, where the LP and PP genotypes had improved color, respectively. The source of the L allele in the TC19F2 LL homozygote is OH8243, whereas OH88119 was the source of the L allele in the BC2 and BC2S4 populations. The region covered by markers LEOH112 and LEOH200 controlled 33% to 71% of the total phenotypic variation for both color uniformity and color intensity in the TC19F2 population, which is considerably higher than that shown in the BC2 and BC2S4 populations. Also, the association contributing to improved color intensity with icolorint had not been previously detected in the BC2 and BC2S4 populations. Association with color intensity in processing varieties A query in the Tomato Map database allowed us to find multiple markers on chromosome 2 that were polymorphic among parental lines of commercial and experimental varieties. In this collection of varieties, we found markers SSR5 and 93

110 SSR349a co-segregated, which is evidence for linkage disequilibrium. However, we found recombinants between these two markers and LEOH23. We detected significant associations with L* (P= and P=0.0077) and chroma (P=0.0016) with markers that span a chromosomal region from LEOH23 (which maps in proximity of CT10682i) to SSR5. SSR349a could not be mapped in our population because it was not polymorphic. However, in a genetic map available on SGN (Tomato EXPEN 2000 v50), SSR349a was found to map in close proximity to SSR5. The association with L* and chroma in that region is indicative of a QTL for color intensity on chromosome 2. Gain under selection and heritability We compared the gain under selection based on phenotype and MAS for the three indices. We observed a gain under phenotypic selection using both icoloruni and icolorint. There was no gain under selection when using icoloropt. The greatest gain was realized with icoloruni (R=-2.57, SE=0.879) compared to icolorint (R=-0.95, SE=0.169). For MAS, we used marker intervals on chromosome 2 and single markers for chromosomes 6 (LEOH112) and 8 (LEOH147). Overall, we observed a similar trend for MAS as for phenotypic selection with positive gain realized with icolorint and icoloruni and no gain with icoloropt. Scanning chromosome 2 with marker intervals showed that the greatest gains were realized in the regions spanned by markers SSR5 to CT10279i and CT10279i to SSR32 (R=- 2.82, SE=1.30). Positive gain was also observed by selecting the favorable allele of 94

111 LEOH112 (LP) for color uniformity (R=-4.85, SE=1.38), as well as the favorable allele of LEOH147 (LL) for color intensity (R=-1.62, SE=0.309). An explanation for the lack of gain under selection using icoloropt may be that the correlation between Fremont and Wooster within each population was not significant. In contrast, the correlations were significant for icoloruni between locations for the BC2 and BC2S4 populations (P= and P=0.0002, respectively). For icolorint, a significant correlation was found for the BC2S4 population only (P<0.0001). In general, making selections by combining phenotypic and genotypic data resulted in a greater gain than making selections based on phenotype. For icoloruni and icolorint, we determined estimates of realized broadsense heritability, which are indicative of selection effects that are transmitted to the next generation. Heritability estimates determined from selection by phenotype only were 0.30 and 0.52 for icoloruni and icolorint, respectively. Heritability estimate determined from MAS using the marker LEOH112 was 0.73 for icoloruni. Using markers fw2.2 and LEOH147, heritability estimates for icolorint were 0.68 and 0.81, respectively. These estimates can be exploited for further selection for color uniformity and intensity. DISCUSSION The aim of this study was to elucidate the genetic basis of color and color uniformity in populations derived from Solanum lycopersicum (OH88119) and S. 95

112 pimpinellifolium (PI128216) via QTL mapping. The populations were evaluated in Fremont and Wooster, Ohio. These two environments were quite distinct, probably due to differences in soil properties that affect the incidence of yellow shoulder disorder (Clivati-McIntyre et al, 2007). The distance between the two locations is 150 km, yet Fremont and Wooster received different patterns of precipitation and average air temperature each year ( Merits of selection indices We designed three indices to combine parameters describing color intensity (icolorint) and uniformity (icoloruni and icoloropt). Two of the indices were based on objective analysis and captured the essential features of color that maximize variation between lines in the populations tested. icoloruni is based on the first principal component and confirms the importance of YSD in contributing to variation between lines in our populations. PCA showed that 52-62% of the variation for color was explained by uniformity. The indices were able to discriminate color quality effectively. A notable asset of icolorint is that it allowed us to interpret chroma for improved color, which was not possible using values of chroma alone. Moreover, positive gains under selection were realized for both icolorint and icoloruni. Estimates of realized heritability were high when exploiting MAS and the indices for improved color uniformity and intensity. Developing indices based on principal components rather than optimal and visual inspection resulted in greater gains and can be exploited for further selection. 96

113 QTL discovery and confirmation In total, regions in seven of the 12 chromosomes were associated with color and color uniformity detected with either the ANOVA or the non-parametric analyses in at least two environments. QTL on chromosomes 2 and 8 for color intensity were detected with both statistical methods and found in the BC2 and BC2S4 populations. In addition, an analysis with elite varieties confirmed the QTL for color intensity on chromosome 2. The QTL on chromosome 6 controlling color uniformity remains problematic due to differences in the favorable allelic effect found across the BC2, BC2S4 and TC19F2 populations. The LP genotype in the BC2 population and the PP genotype in the BC2S4 population were favorable for improved color uniformity. However, the LL genotype was favorable in the TC19F2 population. One explanation would be to postulate an allelic series may be the cause. The LL genotype from OH8243, the parent used to develop the F2 population, may have had a stronger effect on color uniformity than the P allele from PI128216, which in turn could have a stronger effect than the LL genotype from OH This supposition can be tested by crossing the BC2 individual to a different tester to assess the effect of genetic background or by evaluating the F3 population derived from the TC19F2 population. The TC19F2 population was evaluated on a singleplant basis; the F3 population would allow for a better sampling and statistical power for detecting allelic effects. In addition to QTL on chromosomes 2, 6 and 8, we detected other QTL for color intensity on chromosomes 9 and 11 and for color uniformity on chromosomes 97

114 4 and 7. Although these QTL conformed to our definition of a robust association for the statistical method used to detect them, they were not consistently detected across populations or detection methods. The lack of confirmation between the BC2 and BC2S4 populations may be due to the effect of either recombination, distribution of the trait variance and its associated error variance, or false positives in statistical analysis. Inconsistency in detecting QTL across populations generated with the same parents but with different structures was reported for popcorn (Li et al, 2007). In this study, inconsistencies in QTL detection between BC2F2 and F2:3 populations were attributed to selection, population structure, and the mode of gene action. In our study, the effect of the environment was also an influencing factor. Conservation of QTL across environments was also an issue presented in a study for traits of agronomic importance in tomato (Bernacchi et al, 1998). They report a low correlation (r=0.17; P=0.05) between the number of QTL detected in multiple locations and the proportion of phenotypic variance explained by the QTL. Thus there is a weak tendency for QTL with major effect to be detected more consistently across locations. In the BC2 and BC2S4 populations, the QTL effect ranged from 5% to 27%. QTL on chromosome 2 had the highest effect (17-27%), whereas QTL on chromosome 8 had a lower effect (6-9%). Nonetheless, the two statistical methods were able to detect both QTL across the two locations. The position and effect of the QTL on chromosome 2 was found to be more consistent than the other color QTL detected. 98

115 Color and carotenoids The QTL on chromosome 2 affecting color intensity resides on the long arm, with data suggesting associations over the majority of the region. Chromosome 2 contains candidate genes such as high pigment 1 (hp-1), and sequencing of the genome (28% complete as of ; SGN) is likely to reveal others in the near future. Our results are in agreement with a color QTL reported in the chromosomal region encompassing LEOH23 (close to TG165 and SSR66; Figure 4.4), which controlled 14.6% of the phenotypic variation for L* (Yang et al, 2004). The QTL contributed to improved color within elite germplasm. The QTL on chromosome 6 controlling color uniformity is in close proximity to the Beta (B) gene and its allele old gold crimson (og c ). The B gene results in orange fruits due to the accumulation of β-carotene at the expense of lycopene. It is caused by enhanced transcription of lycopene-β-cyclase, the enzyme that converts lycopene to β-carotene. The recessive og c lacks β-carotene due to a loss of function mutation in the fruit-specific lycopene-β-cyclase (Ronen et al, 2000). Thus it is possible that we detected allelic variation. However our marker-trait association was predominantly for color uniformity, thus any link remains speculative. Genes associated with the synthesis of carotenoids are not the sole source of genetic variation for color (Liu et al, 2003). We detected another QTL on chromosome 8. Of the known structural and regulatory genes involved in the carotenoid pathway that have been characterized and mapped, none are presently mapped to chromosome 8. However, a tomato light signal transduction gene, 99

116 LeHY5, was mapped to chromosome 8. Repression of LeHY5 in tomato fruit resulted in a 12-32% decrease in total carotenoids compared to normal controls (Liu et al, 2004). Two other loci involved in light signal transduction mapped to chromosome 8 in tomato (Jim Giovannoni, personal communication). Such findings prompt the need to investigate pathways other than the carotenoid biosynthesis pathway for possible candidate genes. Also, there is a need for confirming the QTL in various genetic backgrounds and population structures, which will lend itself to further characterization of the QTL via fine mapping. Utility of QTL in elite breeding germplasm A number of studies have reported QTL for improved color traits for which the effect was due to an introgression from wild germplasm (Bernacchi et al, 1998; Fulton et al, 1997; Fulton et al, 2000; Liu et al, 2003; Tanksley et al, 1996). Although these introgressions hold promise (Tanksley and McCouch, 1997), they can also be detrimental for utility in crop improvement where linkage drag and pleiotropic effects are present. Other QTL associated with fruit color were detected in elite germplasm (Yang et al, 2004) and introgressed populations in which the beneficial allele was from the adapted parent line (Kabelka et al, 2004). In our study, we found that the positive effect of the QTL detected on chromosomes 2, 8, and 9 was due to the adapted OH88119 allele. This result led us to query the Tomato Map database to test the utility of the markers mapped to chromosome 2 that are polymorphic within the core collection of adapted varieties. We found multiple 100

117 polymorphic markers on chromosome 2 that can be exploited within elite germplasm. These color QTL with positive effect from adapted germplasm show potential for crop improvement using elite by elite crosses. In conclusion, we report the development of objective indices that capture the most important features of color. Our selection indices used in the QTL analysis allowed us to select genotypes with improved color and color uniformity. We show positive gain under selection based on phenotype and on the combination of phenotype and genotype by implementing the index for color uniformity, icoloruni, and for color intensity, icolorint. Developing indices based on principal components (PC) captured the essential features of color better than the index based on optimal values, with which no gain under selection was realized. The new software application Tomato Analyzer and its Color Test (TACT) allowed us to quantify color uniformity in a way that can be standardized and archived in database as a community resource. We report QTL for color intensity on chromosomes 2, 8, and 9 and a QTL for color uniformity on chromosome 6. Our findings suggest the feasibility of crop improvement for color within the adapted germplasm and the continuous exploitation of genome sequencing resources for applied outcomes in breeding and genetics. ACKNOWLEDGMENTS I would like to thank the following people for contributing to this chapter. David Francis provided scientific guidance, and the resources to complete this study. 101

118 Matthew Robbins and Wencai Yang contributed to the development of markers used in this study. I thank Bert Bishop from Statistics and Computing Services at the OARDC/OSU in Wooster, Ohio. I appreciate the help from Troy Aldrich for establishing field plots. 102

119 Locus Type Chr. Forward primer Reverse primer Tm y RE z CT10030I.2 InDel 1 CAAGTCTATGGGGATTGTAGGG GCAAATTAGGGACCAAAAAGG 52.. CT10126I InDel 1 CATGACTGAGCATCTGCGTTC GCCGCCACTTATTGTAGGAT 52.. CT10975I InDel 1 CGTGAACCCGGAACTCTGAAC TCATTGCCACACAGAAGCAG 52.. LEOH106 SNP 1 AGGGAGAAATTTGACATACGG GGACCAACAGCAAATACAAAA 52 Alu I SSR117 SSR 1 AATTCACCTTTCTTCCGTCG GCCCTCGAATCTGGTAGCTT 45.. SSR134 SSR 1 CCCTCTTGCCTAAACATCCA CGTTGCGAATTCAGATTAGTTG 45.. SSR192 SSR 1 ACAACATGGGAAGCACTTGA ATTAAATTGGGCCATGGTGA 45.. CosOH44 SSR 2 TGCTTCTTGCACCACAAACT TGTTGTCATGGTCCCTTTGA 45.. CT10279I InDel 2 CTGATGCCGCAATTACTTTTAG GAAAGCCAAACCAGGATTTTC 52.. CT10682I InDel 2 CGCCGCTCGTACAAGGTTATTC TCGATTTCCCAAATTGAAGC 52.. fw2.2 InDel 2 CACATCTTACGATTATTGGGGTAA GTGCACACATCTTAACAAATCA 52.. LEOH348 SNP 2 TGTTTCCCTTCATTCATGCT CCAATTGGATAAATTGGTGGT 52 Tai I SSR26 SSR 2 CGCCTATCGATACCACCACT ATTGATCCGTTTGGTTCTGC 45.. SSR32 SSR 2 TGGAAAGAAGCAGTAGCATTG CAACGAACATCCTCCGTTCT 45.. SSR349A SSR 2 GAGTGATCATCCATCCTCTCA GGAAGAGACTTTGGACTAAGGGA 45.. SSR5 SSR 2 TGGCCGGCTTCTAGAAATAA TGAAATCACCCGTGACCTTT 45.. SSR66 SSR 2 TGCAACAACTGGATAGGTCG TGGATGAAACGGATGTTGAA 45.. CT10151I InDel 3 CGCTACAGGCAGACAGCATAGAC CTGCATTCCCTTGTGTTTTC 52.. CT10402I InDel 3 CGGCCACATCTGGATATTTTTC ATGGCAACTTCCTTCAATGC 52.. CT10690I InDel 3 CCAAGCAGGTAGCAAAAGACTG AAGAAAAGGGAATGTCAGTTTG 52.. CT10772I InDel 3 CTGCAAAAACATATCGCAATTC GAGGGTCATTTCAGTAGACAGC 52.. LEOH124 InDel 3 CCCGTCTCCTTCTCCCTCTTT CTGGCTGGTGTCTTCTCCAT 52.. LEOH127 SNP 3 CAAGGCATCAACCTAATTGGA TGTAGGCTTGAAAAATAAGAGGAGA 52 Hinc II LEOH185 InDel 3 CGTCACAGTCGCGTAAATGA CCTTCTTCCCCAATTTCCTC 52.. LEOH223 SNP 3 ACAAGAGTCGGGTGATGGAC GCGATGGAAATAGCATCACA 52 Tru1 I SSR22 SSR 3 GATCGGCAGTAGGTGCTCTC CAAGAAACACCCATATCCGC 45.. SSR320 SSR 3 ATGAGGCAATCTTCACCTGG TTCAGCTGATAGTTCCTGCG 45.. SSR601 SSR 3 TCTGCATCTGGTGAAGCAAG CTGGATTGCCTGGTTGATTT 45.. CT10888 InDel 4 AAGCCTCCTTTACCAATCAGC GCCATGAAAGTGAACTGAAGC 52.. LEOH101 SNP 4 AAGAGGCTGCCTTTCTGAAT GCCATCACAAAAGCATCAAA 52 Tru1 I LEOH361 SNP 4 TCAATTTCGTCCAATGCAAA CTTGGGCCTATCTCAAAACG 52 Rsa I SSR43 SSR 4 CTCCAAATTGGGCAATAACA TTAGGAAGTTGCATTAGGCCA 45.. SSR450 SSR 4 AATGAAGAACCATTCCGCAC ACATGAGCCCAATGAACCTC 45.. y Tm, melting temperature; z RE, Restriction enzyme. Continued Table 4.1: Polymorphic molecular markers tested in the populations derived from the cross Solanum lycopersicum (OH88119) x S. pimpinellifolium (PI128216). 103

120 Table 4.1 continued Locus Type Chr. Forward primer Reverse primer Tm y RE z TOM184 SSR 4 CAACCCCTCTCCTATTCT CTGCTTTGTCGAGTTTGAA 45.. TOM194 SSR 4 ACGAAGTAATACAGCCAATG AGCCATCCAACACAAAACAC 45.. CosOH73 SNP 5 CTTCCCGACAAGCACAAAAA CGAATGCTCTGTACCATTTCC 52 Alu I LEOH63 SNP 5 CCCCTCGTGTAGGTGTCACT ATCCTCCGATCATCAGCAGT 52 Taq I CT10242I InDel 6 CATGCGTTATCAGTTTGAGACG CAGATCTGCAGCCTGAAACC 52.. CT10328I InDel 6 ACCGTGAATCTGAGGTTGCT CCGTGCCAATGTCCAACTAAG 52.. LEOH112 SNP 6 GCCAATTGAACTGACCATCTG CCCATGTATTTGGCTGTAGAA 52 Tai I LEOH17 SNP 6 CAGACGAGAAGCAAGTTGAGG CTACCACTGCGTGCTTTGAC 52 BseN I LEOH200 SNP 6 GGGTTTATGTTGGTGATATGGTG TCAGCAGCTAAAAGTCGAACC 52 EcoR V SSR45 SSR 7 TGTATCCTGGTGGACCAATG TCCAAGTATCAGGCACACCA 45.. CosOH42 SNP 8 GGAATTCCACATGAAGTAATGGA TTGATCAAATCGGGCTTAGG 52 Tsp45 I CT10152I InDel 8 CTGAGAAATGGAAGGTAAAAACTGTC TGATTTAAGAGATAAGGCATCAAGC 52.. CT10367I InDel 8 CCTCAAGCTGGTGCTTCTGTC CCCTGGGTCTTCTCCTCTTC 52.. LEOH147 SNP 8 AGTTCCCGTTGGTGTTCAAG CCCTTGCCAGTGGATGTTAG 52 Tsp45 I LEOH70 SNP 8 CCTCTCCCAATCCCAATTCT CAGAAGCAGCAATTCGACAA 52 Tru1 I Cosi52 InDel 9 GCCTTTCTTCCAGGATGCTA CCCATTTTCCTTCTTCCTAGA 52 - LEOH144 SNP 9 ATGGCCTAGGATTGCATCTG TTGCATACACTTGGATAAAAGCA 52 Fok I SSR70 SSR 9 TTTAGGGTGTCTGTGGGTCC GGAGTGCGCAGAGGATAGAG 45.. CT10105I InDel 10 CCCAAGCCCTTCTGATTTAGTG CTTTACATAATTGGCCGACAAAC 52.. CT10386I InDel 10 CTGGAGTTCTGGGTCACTTTG ATAGCCATCCCAAACCACAC 52.. SSR223 SSR 10 TGGCTGCCTCTTCTCTGTTT TTTCTTGAAGGGTCTTTCCC 45.. SSR34 SSR 10 TTCGGATAAAGCAATCCACC TCGATTGTGTACCAACGTCC 45.. CT10615I InDel 11 CTTTCCACAGGTCATTCTTCC TGGGGGATGAGAGTTGTAATG 52.. CT10683I InDel 11 CTGGATATTCGTATATTCGAGACAGG ATTCCGATCCATCCAATCTG 52.. CT10737I InDel 11 CCCACTCCTGGGACTCAAATC TGGACCCACAGGTAATGAGG 52.. TOM196 SSR 11 CCTCCAAATCCCAAAACTCT TGTTTCATCCACTATCACGA 45.. CT10796I InDel 12 CAAATCCTCACCGACAAGGTG GAAACGAAAATCGCTCAAAAG 52.. CT10953I InDel 12 CCTGTCTCTCGCTTTTCTCCTG ACGGAACACACCCTAAGTGC 52.. LEOH197 InDel 12 CGTCACAGTCGCGTAAATGA CCTTCTTCCCCAATTTCCTC 52.. LEOH275 SNP 12 TCCTCTGAAAACAACTTCACGA AGTGTGAGCCTCAAATTCCA 52 Tru1 I SSR20 SSR 12 GAGGACGACAACAACAACGA GACATGCCACTTAGATCCACAA 45.. CT10844I InDel ND GCTGAAGAAAGGGATACTCCAA CCATCCACTGAGAGCCCAAAG 52.. LEOH26 SNP ND GAAGATTCGGAGGTCAAACG AAAGGCTTCACATCTTCACCA 52 Fok I SSR71 SSR ND CAAATGGCATGGAGAATGGAA CAT CCA CTG AGA GCC CAA AG 45.. CT10925I InDel ND CTGGATCTCAATACAACAATCAAGG TCAGTTGAACCGACAGAAGC 52.. y Tm, melting temperature; z RE, Restriction enzyme. 104

121 %YSD %RED L* a* b* Hue Chroma BC2 Fremont: Rep1 vs. Rep (0.2678) (0.0200) (<0.0001) (0.0018) (<0.0001) (0.1428) (<0.0001) BC2: Fremont vs. Wooster (0.3960) (0.4862) (0.1373) (0.9531) (0.8725) (0.4061) (0.9878) BC2S4: Fremont vs. Wooster (<0.0001) (<0.0001) (0.0002) (0.0006) (0.8001) (<0.0001) (0.6412) z The significance of the Pearson correlation estimates (P-value) is indicated in parentheses below the coefficient. Table 4.2: Pearson correlations by population between replications within location in Ohio (BC2, Fremont) and between locations (BC2 and BC2S4, Fremont vs. Wooster) for color and color uniformity. 105

122 Population FRE \ WOO z %YSD %RED L* a* b* Hue Chroma BC2 %YSD ns %RED ns L* ns a* ns b* Hue ns Chroma ns ns ns - BC2S4 %YSD %RED L* ns a* b* ns Hue Chroma z Pearson coefficients noted above and below the diagonal are for Wooster and Fremont, respectively. Only highly significant (P<0.0001) correlations are shown, or otherwise italicized when significant at P<0.05 or noted non-significant (ns). 106 Table 4.3: Pearson correlations between all color and color uniformity traits by population (BC2 and BC2S4) and by location (Fremont and Wooster, Ohio).

123 Principal Component 1 y Principal Component 2 BC2 BC2S4 TC19F2 BC2 BC2S4 TC19F2 Fremont Wooster Fremont Wooster Fremont Fremont Wooster Fremont Wooster Fremont %YSD %RED L* a* b* Hue Chroma Proportion z Cumulative y Eigenvector coefficients are reported for each trait. z The contribution of each PC is represented by the proportional and cumulative Eigenvalues. 107 Table 4.4: Principal component analysis (PCA) of color and color uniformity traits in the BC2, BC2S4 and TC19F2 populations evaluated in Fremont and/or Wooster, Ohio.

124 108 Fremont Wooster Combined Pop. Chr. Marker Index P-value x LL y LP/PP y Vm/Vp z P-value LL LP/PP Vm/Vp P-value LL LP/PP Vm/Vp BC2 2 fw2.2 icolorint < SSR349A icolorint LEOH112 icoloropt LEOH200 icoloropt LEOH200 icoloruni LEOH147 icolorint SSR70 icolorint BC2S4 2 fw2.2 icolorint SSR349A icolorint LEOH112 icoloropt LEOH200 icoloropt LEOH200 icoloruni LEOH147 icolorint SSR70 icolorint x Only marker-trait associations detected in at least two locations are reported. y Actual index values for the corresponding genotypic classes: LL=homozygous Solanum lycopersicum; LP (for the BC2 population only)=heterozygous S. lycopersium/s. pimpinellifolium; PP (for the BC2S4 population only)=homozygous S. lycopersicum. In bold are the beneficial alleles (significant at P<0.05). z Proportion of total phenotypic variation explained by marker locus. Table 4.5: Marker-index associations for color and color uniformity detected in both BC2 and BC2S4 populations derived from Solanum lycopersicum (OH88119) x S. pimpinellifolium (PI128216).

125 BC2 population BC2S4 population Fremont Wooster Fremont Wooster Chr. Marker Index/Trait P-value y LL z LP P-value LL LP P-value LL PP P-value LL PP 2 fw2.2 icolorint SSR450 icoloruni SSR45 icoloropt LEOH147 icolorint CT10737I icolorint ND SSR71 icoloropt y P-value derived from the Kruskal-Wallis test obtained from mean comparisons between the population lines and the recurrent parent, OH Only marker-trait associations detected in at least two locations are reported. z Actual index values for the corresponding genotypic classes: LL=homozygous Solanum lycopersicum; LP=heterozygous S. lycopersium/s. pimpinellifolium; PP=homozygous S. lycopersicum. In bold are the beneficial alleles (significant at P<0.05). 109 Table 4.6: Non-parametric analysis (NPA) for markers associated with indices designed to capture color and color uniformity for the BC2 and BC2S4 populations evaluated in Fremont and Wooster, Ohio.

126 a* * * F (X=25.0) W (X=26.8) b* * F (X=32.2) W (X=32.4) * L* * * F (X=42.9) W (X=41.2) Chroma F (X=41.8) * W (X=42.8) * Hue * F (X=52.5) W (X=50.6) * %RED F (X=44.2) * W (X=48.9) * %YSD * F (X=15.8) * W (X=10.5) icoloruni * * F (X=12.8) W (X=7.48) icoloropt icolorint * * F (X=1.07) W (X=0.796) F (X=18.8) W (X=20.1) * * Figure 4.1: Frequency histograms for color and color uniformity of the BC2 population. Tomato fruits were evaluated in Fremont and Wooster, Ohio, in Each bin represents one standard deviation. Dashed bars correspond to the trait distribution in Fremont (F), whereas solid black bars correspond to Wooster (W). The population mean (X) is included in the legend. The asterisk corresponds to the bin position of the recurrent parent, OH88119, used in the development of the advanced backcross population. 110

127 a* * * F (X=17.4) W (X=24.7) * * b* F (X=37.9) W (X=33.9) L* F (X=49.0) W (X=44.0) * * Chroma * * F (X=43.6) W (X=43.2) Hue F (X=64.9) * W (X=53.6) * * %RED F (X=13.6) W (X=40.1) * %YSD F (X=54.8) * W (X=20.8) * icoloruni * * F (X=48.9) W (X=17.5) icoloropt * * F (X=2.97) W (X=1.33) icolorint * * F (X=13.6) W (X=17.5) Figure 4.2: Frequency histograms for color and color uniformity of the BC2S4 population. Tomato fruits were evaluated in Fremont and Wooster, Ohio, in Each bin represents one standard deviation. Dashed bars correspond to the trait distribution in Fremont (F), whereas solid black bars correspond to Wooster (W). The population mean (X) is included in the legend. The star corresponds to the bin position of the recurrent parent, OH88119, used in the development of the inbred backcross population. 111

128 112

129 Figure 4.3: Visual inspection of selection for color and color uniformity based on three indices: icoloruni for uniformity, icolorint for intensity, and icoloropt for optimal color. The range of values is based on the evaluation of the BC2S4 population in Fremont (F) and Wooster (W), Ohio, in The column on the left shows the genotype with the best index value and on the right with the worst value. The sign preceding the index names indicates the direction of the index values for best or worst color. 113

130 114

131 Figure 4.4: Approximate map position of markers used in this study (underlined). Regions associated with color intensity and uniformity are delimited with solid black vertical lines. Markers noted below a chromosome have been mapped to that chromosome but the relative position remains uncertain. The map position of genes involved in the carotenoid biosynthetic and phytochrome signaling pathways is delimited by a dashed vertical line (hp-1, high pigment 1; dg, dark green; hp-2, high pigment 2 (non-allelic to hp-1); r, yellow flesh; B, beta; og c, old gold crimson (allelic to B); t, tangerine; Del, delta). 115

132 CONCLUSIONS This dissertation research was initiated to gain a better understanding of tomato fruit color as a quality trait for breeding and genetic studies. Multiple perspectives define the importance of color. The processing tomato grower wishes for varieties that will produce intense and uniform red tomatoes in order to meet the demands of contract incentives. The processor grades tomato color based on uniformity, with the most uniform fruits used for whole-peel or diced products, and the non-uniform fruits for tomato juice. The consumer benefits from the healthpromoting carotenoids found in tomato products. These criteria represent the essence of fruit color quality for the processing tomato. A limitation to quality is the presence of yellow shoulder disorder (YSD). YSD is a blotchy ripening disorder that is characterized by discoloration under the epidermis at the proximal end of mature fruits (Francis et al, 2000). Although the potential of a genetic constituent has been reported (Hartz et al, 2005; Sacks and Francis, 2001), the main causes known to increase the incidence of YSD are soil fertility, especially phosphorous and potassium nutrition (Clivati-McIntyre et al, 2007; Hartz et al, 1999) and environmental factors such as fluctuations in 116

133 temperature, high pericarp temperature and high relative humidity (Jones and Alexander, 1962; Picha, 1987). The first objective of this study was to optimize sampling of tomato fruit carotenoid content and to assess the impact of YSD on lycopene and beta-carotene. Lycopene and beta-carotene content was evaluated in juice obtained from field and analytical replications. The greatest source of variation for lycopene and betacarotene was found among plots, which accounted for approximately 50% of the total phenotypic variation. In contrast, variation among analytical replications, evaluated between replicates of carotenoid extractions and HPLC quantifications, was negligible. We thus suggested increasing the biological replications, while minimizing the analytical replications, which tend to be more costly and technically demanding. In addition, lycopene and beta-carotene content was determined from juice made with fruits affected by YSD and with fruits free of YSD symptoms. The former had 13-24% significantly less lycopene than juice made from YSD-free fruits. Beta-carotene content was also reduced (4-9%), although the reduction was not significant. We show that for every 10% increase in area affected by YSD, there is a decrease in lycopene of 1.03 mg/100g fresh weight. These results can be used to draw thresholds for an acceptable amount of YSD that will not impede the potential health benefits of lycopene in tomato products. The second objective aimed to develop a tool allowing us to measure the extent of YSD efficiently and objectively. The software application Tomato 117

134 Analyzer was initially designed to measure fruit morphological attributes. A color module was developed and implemented to measure color and color uniformity (referred to as TACT). TACT collects RGB values for each pixel on a digital image and converts them to L*, a*, b* values of the CIELab color space, as well as hue, chroma and two user-defined ranges. The values of L*, a*, b* collected with TACT were highly correlated with the colorimeter values. Moreover, TACT provided more precision to color measurement as it collects averaged values of the entire fruit surface analyzed. Additionally, TACT was designed to offer a measure of color uniformity. A range of hue values was defined to represent YSD, and another to represent the optimal red color. An inbred backcross population derived from elite processing varieties was evaluated for color and color uniformity using both TACT and a colorimeter. The analysis of variance components showed that TACT partitioned a greater proportion of the total phenotypic variation into genotype. On the basis that estimates of genotypic variance are the principle for genetic gain and genetic improvement in a breeding program, TACT offers breeders and geneticists a tool to improve the acquisition of phenotypic data. Lastly, the third objective was to evaluate color and color uniformity using TACT in breeding populations and to elucidate the genetic basis of color via quantitative trait loci (QTL) mapping. The breeding populations consisted of an advanced backcross population (BC2) derived from Solanum lycopersicum (OH88119) and S. pimpinellifolium (PI128216), an inbred backcross population (i.e. PI IBC population), and an F2 population derived from a selected BC2 118

135 individual and a tester OH8243. Elite processing varieties were also evaluated. Thorough analysis of phenotypic data revealed significant GxE interaction and highly correlated traits. To simplify the complexity of the correlated color traits, a principal component analysis was conducted for each population evaluated at each location. The first principal component (PC1) attributed up to 62% of the total phenotypic variation to color. Inspection of the Eigenvector loadings revealed that PC1 was dominated by hue, %YSD and %RED, and that PC1 measured color uniformity. The fact that PC1 was heavily weighted to color uniformity emphasizes the importance of YSD in the variation observed within populations. The second principal component (PC2) explained up to 33% of the total phenotypic variation. The main contributors to PC2 were chroma, a* and b*. Two indices were designed based on these principal components and exploited them for QTL analysis. A third index incorporated optimal color uniformity and intensity based on visual inspection and TACT data of commercial varieties. QTL analysis was based on two statistical methods to detect significant marker-trait associations: single factor analysis of variance and non-parametric analysis. The findings suggest three QTL for color intensity on chromosomes 2, 8, and 9, and a QTL for color uniformity on chromosome 6. Additionally, positive gains under selection were realized by phenotypic selection as well as marker-assisted selection (MAS). Compared to phenotypic selection, we observed greater gains by assisting our selection scheme utilizing each of the three QTL identified for improved color uniformity and color intensity. Also, scanning chromosome 2 with marker intervals as opposed to single 119

136 markers allowed us to determine the region on the chromosome that resulted in the greatest gain under selection. The marker fw2.2 was consistently associated with color intensity across populations. However, the chromosome scan showed that the region encompassing CT10279i and either SSR5 or SSR32 resulted in higher gains under selection. The estimates of realized heritability were also highest when determined from MAS. The analysis of elite varieties showed that markers used to detect QTL for color intensity on chromosome 2 can be exploited in elite germplasm for crop improvement. With the completion of the tomato genome sequencing project, continuous progress in marker development, and application of tools such as TACT for standardized phenotypic analysis, applied outcomes are imminent that will benefit the grower, the processor, and the consumer with efficient development of tomato varieties with improved color and color uniformity. 120

137 APPENDIX A TOMATO ANALYZER COLOR TEST: USER MANUAL Part 1: Overview of color and Tomato Analyzer Color Test (TACT) Digital color and the RGB color space TACT and the CIELab color space Standard illuminant and observer angle TACT application Part 2: Basic features Collecting and formatting images for TACT Generating automated boundaries Adjusting boundaries Part 3: Calibration Obtaining color standards Scanning color checker and collecting TACT L*, a*, b* values Collecting colorimeter L*, a*, b* values Determining correction values for calibration Part 4: Color Test Defining parameters Analyzing a single image Using the batch analysis Part 5: Color standard specifications Part 6: Mathematical formulas 121

138 PART 1: OVERVIEW OF COLOR AND TOMATO ANALYZER COLOR TEST The Tomato Analyzer module called Color Test (TACT) is designed to collect objective color measurement from JPEG images (Figure A.1), obtained from scanning fruits on the flatbed surface of a scanner. Figure A.1: Tomato Analyzer Color Test application. Digital color and the RGB color space. Computer color measurements are based on the RGB color space. This system is additive, as it measures the strength of each R (red), G (green), B (blue) color in each pixel to reproduce other colors. The additive RGB color space is a cube with each axis representing variance in one of the primary colors and a white reference point. This color space is nonlinear and does not mimic the nature of color perception. It is not generally standardized. While there is a standardized version (srgb) for which conversion formulas exist, measurements may differ among hardware and software. These differences can be corrected by calibrating the devices involved in the process of collecting and analyzing color images. 122

139 TACT and the CIELab color space. TACT takes the average RGB values for each pixel and translates the color measurements to L*, a*, b* values of the CIELab color space. Unlike the RGB color space, the CIELab color space is able to approximate human visual perception. It is a spherical color space with the vertical axis representing lightness (+L*) to darkness (-L*). The chromaticity coordinates are a* and b* and their axis indicates color directions: +a* is the red direction, -a* is the green direction, +b* is the yellow direction and b* is the blue direction (Figure A.2). Figure A.2: Representation of the CIELab color space (original photo reproduced with permission from Konica Minolta). Hue and chroma are descriptors of color based on a* and b* values (Figure A.3). Hue represents the basic color. It is an angular measurement in the quadrant between the a* and b* axes. Chroma is the saturation or vividness of color. It is measured radially from the center of each quadrant with the a* and b* axes. 123

140 Figure A.3: Representation of hue and chroma, two attributes of perceived color (original photo reproduced with permission from Konica Minolta). Standard illuminant and observer angle. Different light sources will make colors appear different. A standard illuminant has a specific spectral distribution. Standard illuminant D65 represents natural daylight. It should be used for specimens that will be illuminated by daylight, including ultraviolet radiation. Illuminant C was also constructed to represent natural daylight, but its spectral distribution excludes ultraviolet radiation. In addition to the light source, the angle of view will also affect color sensitivity of the eye. Colors are perceived most precisely if they strike the area of the fovea in the eye, which is most sensitive to color. The 2 o Standard Observer angle is used for viewing angles between 1 o and 4 o, whereas the 10 o Standard Observer is used for angles larger than 4 o. Tomato Analyzer Color Test application TACT collects RGB values for each pixel of an object. It then translates them to L*, a*, b* values of the CIELab color space, as well as luminosity. 124

141 Algorithms were written for TACT to compute hue and chroma based on L*, a*, b* values. The output of each image analyzed with the TACT consists of the averaged values of RGB, luminosity, L*, a*, b*. In addition, two parameters based on specific ranges of hue values can be defined by the user (see Part 4: Defining parameters). PART 2: BASIC FEATURES The basic features of the Tomato Analyzer (TA) were presented in the original user manual (Version ). It can be downloaded from the following URL: The features will be briefly described, unless they have to be adapted specifically for TACT. Collecting and formatting images for TACT Set the resolution of the scanner at 200 dpi and the output image size at a million colors. Scan the surface of the fruit to be analyzed with a black background. A label with descriptive information for the fruits (year, plot number, etc) can be included on the scanner. A ruler and a color standard can also be included. TACT will not consider the objects as fruits to analyze. Save the image as JPEG. Using any image editing software, crop the image to remove all objects (label, ruler, color standard, etc) and save it as the image for color analysis with TACT. 125

142 Generating automated fruit boundaries The automated analysis will use the default settings for shape attributes. The settings should be changed to meet the user s demand and interest. For the color test and prior to analysis, open the TACT dialog box under the Settings menu. Set the minimum blue value to 30 and save the setting. Open the image to analyze. It will be displayed in the left panel. Click on the Analyze icon. The TACT-defined boundaries of the fruits will appear in yellow. Lower the minimum blue value if the fruits are large (in which case TA may crash) or if the boundaries are much erroneous. Adjusting boundaries If the boundaries need to be adjusted, select the fruit by left-clicking on it. The selected fruit will be displayed in the upper-right panel. Click on the Revise icon, selecting Boundary (shift+b). Click on the boundary at the start of the area to adjust and at the end. The delimited boundary will disappear. The distance between the start and end points should be left than half the entire boundary, as only the shortest boundary distance will disappear. Left-click at multiple points on the boundary to delimit the desired contour. Any action can be undone by right-clicking. Press the Enter key to accept the changes. Click the Save Fruit to save any changes on the image. A TMT file containing all information and adjustments will be saved along with the image. Both files should be kept in the same folder to avoid reanalyzing the image. 126

143 PART 3: CALIBRATION Obtaining color standards Color standards should be chosen based on the broad range of colors observed in the crop of interest. For the color analysis of tomato, we chose 28 color standards (Figure A.4). They were custom-made into a 28-patch color checker (X- Rite, Grand Rapids, MI). The Munsell notations for each patch are provided in Table A.1. Figure A.4: 28-patch color checker from X-Rite Scanning color checker and collecting TACT L*, a*, b* values Scan the color checker and analyze the image as described previously. Make sure TACT recognizes each patch as an object to analyze (yellow boundary). Under the Settings menu, select Color Test. The dialog box will appear. Select the illuminant and observer angle specific to the scanner used and to be used in the future for color analyses. Make sure the correction values are set to 1 for the slope (left boxes) and 0 for the y-intercept (right boxes). Click on the Analyze button. A new window will appear to save the output in a comma-delimited (CSV) Excel 127

144 document. Specify the name and the directory for the data file. The output data file will contain L*, a*, b* values, as well as values for the other parameters, to be used for calibration. Collecting colorimeter L*, a*, b* values Verify the settings of the colorimeter for its source of illuminant and observer angle. They should be consistent with the scanner. Calibrate a colorimeter with a white tile, following the manufacturer s protocol. Collect L*, a*, b* values for each patch of the color checker. Make sure to report the tile number with its corresponding color values. If using the 28-patch color checker developed at OSU (Figure A.4; Table A.1), the L*, a*, b* values are available in Table A.2 (Part 5). Determining correction values for calibration Plot the TACT against colorimeter values for L*, a*, and b*. Determine the regression equation and record the slope and y-intercept for each parameter. Enter the inverse of the slope and the reverse the sign of the y-intercept values; they are used as correction values in the dialog window of the TACT (Figure A.5). 128

145 Figure A.5: Tomato Analyzer Color Test window. Correction values used to calibrate the scanner are entered in the lower panel of the dialog box. PART 4: COLOR TEST Defining parameters. TACT allows the user to define two parameters based on specific ranges of hue values. Enter the lower and upper bounds of the selected ranges. Save settings for these values to become default settings until the program is closed. Analyzing a single image For a single image to analyze for color, open the image and make the necessary adjustment. Open TACT and click on the Analyze button. A new window will appear to save the output in a comma-delimited (CSV) Excel document. Specify the name and the directory for the data file. 129