Mapping and linkage disequilibrium analysis with a genome-wide collection of SNPs that detect polymorphism in cultivated tomato

Abstract

The history of tomato (Solanum lycopersicum L.) improvement includes genetic bottlenecks, wild species introgressions, and divergence into distinct market classes. This history makes tomato an excellent model to investigate the effects of selection on genome variation. A combination of linkage mapping in two F₂ populations and physical mapping with emerging genome sequence data was used to position 434 PCR-based markers including SNPs. Three-hundred-and-forty markers were used to genotype 102 tomato lines representing wild species, landraces, vintage cultivars, and contemporary (fresh market and processing) varieties. Principal component analysis confirmed genetic divergence between market classes of cultivated tomato (P <0.0001). A genome-wide survey indicated that linkage disequilibrium (LD) decays over 6–8 cM when all cultivated tomatoes, including vintage and contemporary, were considered together. Within contemporary processing varieties, LD decayed over 6–14 cM, and decay was over 3–16 cM within fresh market varieties. Significant inter-chromosomal (gametic phase) LD was detected in both fresh market and processing varieties between chromosomes 2 and 3, and 2 and 4, but in distinct chromosomal locations for each market class. Additional LD was detected between chromosomes 3 and 4, 3 and 11, and 4 and 6 in fresh market varieties and chromosomes 3 and 12 in processing varieties. These results suggest that breeding practices for market specialization in tomato have led to a genetic divergence between fresh market and processing types.

Introduction

The process of domestication and breeding has led to dramatic changes in the reproduction and morphology of crop species. The selection of individuals with favourable characteristics such as non-shattering seed pods, loss of germination inhibition in seeds, increased size of fruit, and compact plant habit has converted feral plants into forms amenable to cultivation (Tanksley and McCouch, 1997; Gepts, 2004; Doebley et al., 2006). These alterations in phenotype were the direct result of genetic changes underlying traits of interest to humans.

The effect of domestication and breeding on the genes and genomes of crop plants can be assessed using a range of approaches including linkage mapping and map-based cloning. As an alternative to analysis in controlled crosses, association mapping in unstructured and complex populations is now being applied to crops (Remington et al., 2001; Breseghello and Sorrells, 2006; Casa et al., 2008). In addition, the increased efficiency and accessibility of sequencing permits the application of advances in molecular evolution theory to detect the effects of artificial selection on genes and gene systems. Population level studies have been used to identify pathways (Whitt et al., 2002) and genes (Wang et al., 1999; Clark et al., 2004; Yamasaki et al., 2005) that were under selection during domestication and improvement. These studies are guided by observing signatures of selection in sequence data, including a reduction in diversity in cultivated germplasm relative to wild relatives, a reduction in diversity relative to control genes (neutral genes), and an excess of rare variants due to new mutations (Doebley, 2004). In addition to their value for identifying genes that were fixed during domestication, these approaches have the potential to identify the genes that explain existing phenotypic variation within breeding programmes.

Tomato (Solanum lycopersicum L.) has been a model for studying genes that distinguish domestic and wild plants. Mapping in wide crosses and the cloning of genes that affect specific traits has produced substantial insight into disease resistance, plant and fruit development, and specific biochemical pathways (Martin et al., 1993; Jones et al., 1994; Pnueli et al., 1998; Frary et al., 2000; Spassova et al., 2001). In species like tomato, fruit morphology is one of the major traits selected, and cultivated forms exhibit far greater phenotypic variation than their wild progenitors (Tanksley, 2004; Paran and van der Knaap, 2007). It is unlikely that allelic variation present in wild ancestors will explain all of the morphological changes that separate landraces, vintage cultivars or modern crop varieties from their wild relatives. For example, mutations of fruit shape genes (e.g. ovate and sun) have led to a high level of phenotypic variation (Liu et al., 2002; Xiao et al., 2008). In the case of ovate, this variation is found in wild progenitors (Tanksley, 2004) while sun originated as a gain-of-function mutation post-domestication (Xiao et al., 2008). Plant breeding balances the competing goals of introducing new variation, and selecting for specific alleles. Selection for the optimum alleles creates two problems. First, heritability declines as genetic variation declines. Thus, breeding progress will be limited as alleles are fixed throughout the genome. Second, fixation of favourable alleles at some loci may inadvertently fix undesirable genes that are linked. For example, linkage group 6 of cultivated sunflower (Helianthus annuus L.) contains several domestication-related loci, some of which provide positive effects, while others provide antagonistic effects relative to desired traits (Burke et al., 2005). Reintroduction of genetic diversity through wide crosses has been practised in cultivated tomato for nearly a century (Williams and St Clair, 1993; Sim et al., 2009). Practices that seek to introduce new variation may have negative consequences, such as the introduction of less favourable alleles and a restriction of recombination in some genomic regions. Introgression has been effective at introducing disease resistance not found in cultivated material (Francis et al., 2001; Kabelka et al., 2002), but has had mixed results with respect to fruit quality (Kabelka et al., 2004). A five-fold reduction in recombination has been documented in the region around the root-knot nematode resistance gene (Mi), which was introgressed from the wild species S. peruvianum (Messeguer et al., 1991). Thus, introgression of a trait may also lead to the inheritance of large linkage blocks associated with that trait. A major goal of marker-assisted breeding programmes is to be able to select for favourable combinations of genes, across genomes and within chromosomes (Frisch et al., 1999). Accomplishing this goal and balancing the competing demands of increasing genetic diversity while selecting desirable alleles will benefit from a description of genetic variation across the genome of breeding populations.

Several strategies have been employed to develop molecular resources for genome-wide analyses within tomato breeding germplasm. Although tomato was one of the first crops to have a saturated genetic linkage map (Tanksley et al., 1992), the nearly exclusive focus on wide crosses has left a paucity of genetic tools for investigating diversity within cultivated lineages. High-throughput markers remain a limited resource, since many markers selected based on polymorphisms in wide crosses are not polymorphic within cultivated germplasm (Jimenez-Gomez and Maloof, 2009). To overcome this limitation, several projects have identified genetic differences including simple sequence repeats (SSRs), insertion/deletion (indel), and single nucleotide polymorphisms (SNPs) among tomato varieties. Analysis of databases developed through large-scale sequencing of tomato ESTs resulted in the identification of approximately 609 potential simple sequence repeats (SSRs; Frary et al., 2005). Of these, 127 were mapped in the cultivated×wild (S. lycopersicum×S. pennellii) reference population, and 61 were polymorphic within cultivated tomato (Frary et al., 2005). Parallel strategies to develop high-throughput markers include in silico mining of SNPs from EST databases (Yang et al., 2004; Labate and Baldo, 2005), oligo-based microarray hybridization (Sim et al., 2009), and sequencing introns of conserved orthologous set (COS) genes (Van Deynze et al., 2007; Labate et al., 2009b). Since many of the SNPs from these studies have been validated in genotyping assays and show polymorphism within cultivated tomato, these marker resources provide an opportunity to assess cultivated germplasm genetically.

In order to organize these resources for the analysis of cultivated populations, a genetic map was developed based on 434 markers. Allele-specific primer extension (ASPE; Lee et al., 2004) markers were created based on previously identified SNPs and these were combined with existing framework RFLP markers, PCR-based SSR markers, and indel markers to develop an integrated linkage map based on two populations. This linkage map was combined with emerging sequence data for the tomato genome to organize markers relative to the tomato physical map. These markers have been used to genotype a collection of 93 S. lycopersicum accessions and nine wild species accessions. The resulting data were used to assess the extent of inter- and intra-chromosomal linkage disequilibrium (LD) in cultivated tomato. Given the history of tomato breeding, which includes introgression from wild species and breeding for distinct market specialization, we expected to identify differences in the pattern and distribution of genetic variation within the genomes of cultivated tomatoes representing different market classes. Specifically, the hypothesis that selection for market differentiation left a signature that could be detected through the analysis of genome-wide patterns of SNP variation was tested.

Materials and methods

Plant material

A set of 102 tomato accessions was assembled, including nine representatives of wild species, five Latin American cultivars, two unimproved breeding lines, 21 vintage cultivars, two greenhouse varieties, 24 fresh market varieties, and 39 processing varieties (Table 1; see Supplementary Table S1 at JXB online). The Latin American cultivars represent early domesticates while the vintage cultivars represent early tomato improvement. Fresh market and processing germplasm are varieties that are adapted to specific market niches and represent improvements made through contemporary plant breeding. These entries were selected from public breeding programmes that release commercially relevant parents and hybrids. Several processing lines were donated directly by seed companies. In addition, selected inbred lines were obtained through self-pollination of commercial hybrids followed by single-seed-descent selection to obtain inbred lines. These selections represent a sample of the alleles present in commercial hybrids, although they do not recreate the parents themselves. Also included were the parents of several important recombinant inbred and inbred backcross populations (Doganlar et al., 2002; Kabelka et al., 2002; Graham et al., 2004; Yang et al., 2005; Robbins et al., 2009). The collection also contained parents of populations utilized by the tomato research community such as segmental substitution lines (M82 and LA716; Eshed and Zamir, 1995) and a mutation library (Menda et al., 2004). Although a few wild tomato species were included in the collection, the focus was on cultivated materials so that the information gained may be directly applicable to tomato breeding programmes.

Table 1.

Number of markers polymorphic within different classes of tomato

	No. of entriesa	indel	SNP	SSR	Total
Processing	39	27 (22)b	104 (64)	39 (27)	170 (113)
Fresh market	24	22 (16)	101 (62)	38 (26)	161 (104)
Vintage cultivars	21	22 (16)	51 (34)	33 (22)	106 (72)
Latin American cultivars	5	18 (13)	57 (38)	42 (28)	117 (79)
All S. lycopersicumc	93	44 (34)	154 (96)	52 (37)	250 (167)
Wild species	9	63 (52)	167 (117)	65 (50)	295 (219)
All entries	102	70 (57)	205 (135)	65 (50)	340 (242)

^aThe number of entries within each class.

^bThe number in parentheses indicates the number of polymorphic markers with known genomic location either by either linkage or physical mapping.

^cAll S. lycopersicum represents cultivated tomato and includes processing, fresh market, vintage, Latin American, and greenhouse cultivars as well as unimproved breeding lines.

The germplasm collection also contained the parents of two F₂ mapping populations utilized to develop genetic linkage maps. The mapping population derived from Sun1642 (S. lycopersicum) and LA1589 (S. pimpinellifolium) consists of 100 F₂ individuals (van der Knaap and Tanksley, 2001). The second mapping population consists of 200 F₂ plants from a cross between Yellow Stuffer and LA1589 (van der Knaap and Tanksley, 2003).

Molecular marker genotyping

Markers used in this study are from various sources and are described in Supplementary Tables S2–S5 at JXB online. Framework markers (RFLP and SSR) used in map construction were from SGN (http://solgenomics.net). Additional SSRs with the prefix ‘TOM’ (Suliman-Pollatschek et al., 2002) were utilized (see Supplementary Table S5 at JXB online). Markers with the prefix ‘LEOH’ were developed based on SNPs or indels in EST sequences [Yang et al., 2004 (LEOH1-LEOH51), Francis et al., 2005 (LEOH100-LEOH365); see Supplementary Tables S2–S4 at JXB online]. Markers with the prefix ‘SL’ were developed based on SNPs and indels identified by Van Deynze et al. (2007; see Supplementary Tables S2 and S4 at JXB online). These ‘SL’ marker names correspond to the primers that amplify the locus followed by a number referring to the position of the polymorphism within the locus according to Van Deynze et al. (2007). The ‘SL’ markers spanning indels contain the suffix ‘i’ while all others are based on SNPs.

Genotyping was performed on two platforms, one for size polymorphisms (SSR, indel, and CAPS; see Supplementary Tables S3–S5 at JXB online) and a second for SNPs detected by an allele-specific primer extension (ASPE) assay (Lee et al., 2004) on the Luminex 200 system (Luminex Corporation, Austin, TX; see Supplementary Table S2 at JXB online). For markers based on indels, primers flanking the indel were designed using Primer3 (Rozen and Skaletsky, 2000). Size polymorphisms were detected using polyacrylamide gels on the Li-Cor-IR2 4200 system (Li-Cor Biosciences, Lincoln, NE) or agarose gels. To detect SNPs by ASPE (see Supplementary Table S2 at JXB online), allele specific primers were designed for each allele using Primo SNP 3.4 (Chang Bioscience; www.changbioscience.com/primo/primosnp.html) or BatchPrimer3 (You et al., 2008). SNP markers were then scored using the Luminex 200 system.

In order to determine marker genotypes, genomic DNA was isolated following the modified CTAB method described by Kabelka et al. (2002) and subjected to PCR. Conditions for PCR reactions were 10 mM TRIS-HCl (pH 8.3), 50 mM KCl, 1.5 mM MgCl₂, 50 μM of each dNTP, 0.1 μM of each forward and reverse primers, 20 ng of template DNA, and 1 unit of Taq DNA polymerase in a total volume of 10–20 μl. To visualize PCR fragments on the Li-Cor system, an additional 0.1 nM of IRD 700 or 800 dye-labelled M-13 forward primer (Li-Cor Biosciences, Lincoln, NE) was added to the PCR reaction and one of the forward or reverse primers contained the M13 sequence as a tail on the 5' end. PCR amplification was performed following Sim et al. (2009) at a suitable annealing temperature between 45 °C and 60 °C (see Supplementary Tables S2–S5 at JXB online). Markers detected as a cleaved amplified polymorphic sequence (CAPS) were digested after PCR following Yang et al. (2004). For the ASPE assay, the locus was amplified using the primers and PCR conditions developed by Van Deynze et al. (2007). The PCR products were ethanol precipitated then rehydrated in 8 μl ddH₂O. After this purification, 4 μl were used as a template in 10–15 μl ASPE reactions that included 1.25 mM MgCl₂, 5 μM each of dATP, dGTP, and dTTP, 5 μM biotin-14-dCTP (Invitrogen Corporation, Carlsbad, CA), 25 nm of each ASPE primer, and 1 U of Platinum GenoType Tsp DNA Polymerase (Invitrogen Corporation, Carlsbad, CA) in 1× supplied buffer. Cycling conditions for the ASPE reactions were 2 min at 96 °C followed by 30 cycles of 30 s at 94 °C, 1 min at 55 °C, and 2 min at 74 °C.

Mapping markers

Linkage maps were developed for the Yellow Stuffer×LA1589 and Sun1642×LA1589 populations separately, then the two maps were combined chromosome by chromosome into an integrated map (Table 2; see Supplementary Table S6 at JXB online; Fig. 1) using Joinmap 3.0 (Van Ooijen and Voorrips, 2001). For all map construction, the thresholds for parameters within JoinMap were 1.00 for LOD, 0.4 for REC, 5.0 for jump, and 1 for ripple while employing the Kosambi mapping function.

Table 2.

Number of markers and size in cM of each chromosome in two linkage maps (Sun1642×LA1589 and Yellow Stuffer×LA1589) and the integrated map

Chromosome	Framework markers	SNP and indel markers	Total markers	PCR-based markers	Markers with segregation distortion	Average cM between markers	Largest gap (cM)	Genome coverage (%)a	Total cM
Sun1642×LA1589
1	25	16	41	29	6	3.4	13.4	100.0	135.0
2	17	13	30	20	6	1.9	9.6	99.2	55.7
3	15	18	33	22	5	3.4	11.3	78.4	108.7
4	13	12	25	17	0	4.8	21.4	86.3	114.1
5	11	7	18	10	2	5.5	19.6	94.1	94.3
6	8	9	17	10	8	5.2	12.7	100.0	83.2
7	10	6	16	9	5	6.1	17.3	98.3	91.6
8	12	7	19	11	0	4.6	11.4	97.1	82.4
9	12	5	17	9	0	5.2	17.1	87.7	83.3
10	12	11	23	15	4	4.2	11.8	100.0	93.1
11	9	6	15	8	9	7.1	14.7	100.0	99.5
12	9	9	18	11	10	5.0	11.9	87.1	85.8
Total	153	119	272	171	55	4.3	21.4	93.8	1126.7
Yellow Stuffer×LA1589
1	10	6	16	6	3	7.7	15.5	95.2	115.8
2	10	7	17	7	8	5.8	10.9	99.3	93.4
3	9	10	19	10	2	5.4	12.8	79.9	97.5
4	6	4	10	4	1	12.8	26.1	74.4	114.8
5	8	7	15	7	0	6.3	30.8	86.3	88.3
6	7	3	10	3	0	9.0	21.7	61.5	80.9
7	4	3	7	3	7	8.8	23.5	41.3	52.6
8	8	5	13	5	1	5.5	19.7	51.7	66.3
9	7	4	11	4	4	9.5	19.5	96.6	95.3
10	6	4	10	4	0	9.8	33.2	47.9	88.4
11	7	6	13	6	9	7.1	17.6	71.7	84.9
12	8	1	9	1	0	11.7	20.7	29.3	93.5
Total	90	60	150	60	35	7.8	33.2	71.9	1071.6
Integrated
1	26	22	48	35	–	2.9	14.7	100.0	137.2
2	19	20	39	27	–	2.8	11.4	98.7	105.2
3	15	29	44	33	–	2.4	9.1	96.4	99.2
4	13	16	29	21	–	4.0	16.9	100.0	107.0
5	12	14	26	17	–	3.5	15.9	95.0	86.6
6	8	12	20	13	–	5.0	12.4	91.2	89.5
7	11	9	20	12	–	4.5	16.4	100.0	84.9
8	12	12	24	16	–	3.6	12.7	100.0	82.7
9	12	9	21	13	–	5.0	18.4	89.7	100.1
10	12	15	27	19	–	3.4	12.6	100.0	81.8
11	9	12	21	14	–	4.4	12.2	100.0	88.5
12	9	10	19	12	–	4.9	17.0	78.6	88.0
Total	158	180	338	232	–	3.6	18.4	96.0	1150.8

^aPercentage of the genome within 10 cM of at least one PCR-based (SSR, SNP, or indel) marker.

Fig. 1.

Integrated linkage map based on two F₂ populations. Genetically mapped markers are on the right of the linkage groups while physically mapped markers are on the left. Marker names in grey text are framework markers used to merge the two F₂ maps. Markers with the prefix ‘LEOH’ were previously developed [Yang et al., 2004 (LEOH1-LEOH51), Francis et al., 2005 (LEOH100-LEOH365)]. Markers with the prefix ‘SL’ were developed based on SNPs or indels (contains the suffix ‘i’) identified by Van Deynze et al. (2007). Markers in parentheses indicate that, although both a SNP and indel were created from the same locus, only one of the markers was mapped. Physically mapped markers are placed relative to the chromosome in their approximate position based on the framework markers. Markers on the far right of chromosome 9 were mapped using segmental substitution lines (Eshed and Zamir, 1995). Vertical lines indicate approximate boundaries of introgressions based on framework markers with the name of the segmental substitution line in vertical text to the left of the line. (Continued on page 6).

Several strategies were employed during the construction of each map to increase reliability. Segregation distortion was tested for each marker within JoinMap and the effect of skewed markers was investigated by comparing the map with and without the marker. If any marker noticeably expanded the map and had a relatively high mean χ² contribution, the marker was removed from the map. Maps were first created with no order restraints and then compared with the Tomato-EXPEN 2000 map (SGN; http://solgenomics.net) by visually inspecting the order of the framework markers on each chromosome. For chromosome 4 where notable differences were detected, mapping was repeated using a fixed order of six framework markers (TG15, TG483, CT157, CT178, CT50, and TG163) based on the Tomato-EXPEN 2000 map. The order of these framework markers in the EXPEN 2000 map represents a robust order since this order is supported by several other genetic maps: EXPEN 1992 (Tanksley et al., 1992), EXPEN 2000 (Fulton et al., 2002), EXPIMP 2001 (Grandillo and Tanksley, 1996; Tanksley et al., 1996; Doganlar et al., 2002), EXPIMP 2008 (Gonzalo and van der Knaap, 2008), and EXHIR 1997 (Bernacchi and Tanksley, 1997). In addition, the position of TG163 is well established relative to the physical map. It was therefore decided to use a fixed order of framework markers based on these multiple maps and physical information from a BAC map. This new map was accepted only if the χ² value decreased or increased reasonably. After the maps were constructed, genome coverage was calculated as the percentage of the genome that was within 10 cM of at least one PCR-based (SSR, SNP, or indel) marker.

The approximate position of markers that showed no segregation in either of the two mapping populations was identified based on the Tomato physical map (SGN; http://solgenomics.net). Tomato sequences with verified polymorphisms from ESTs (Yang et al., 2004; Francis et al., 2005) and conserved orthologous set (COS) introns (Van Deynze et al., 2007) were aligned with tomato genome sequence from the Tomato BAC sequences database (03-01-09; SGN; http://solgenomics.net) using BLASTN with the BLOSUM62 substitution matrix and a minimum expectation value (e-value) of 1e⁻¹⁰. The resulting hits were subjected to a two-step filtering process to identify highly probable marker–BAC alignments. Any BAC with >98% identity and >90% coverage of the query sequence was considered to contain the query locus. Because many BACs were in several stages of sequencing when these analyses were conducted (SGN; http://solgenomics.net), the remaining putative hits with >250 bp alignments were manually inspected to determine if the query sequence aligned to the edge of one of the unordered fragments of an unfinished BAC. In such instances, the BAC was considered to contain the query if the two sequences shared >98% identity. The BAC chromosome designation and data from the overgo analysis (bulk download SGN FTP site; http://solgenomics.net/bulk/input.pl?mode=ftp), were used to determine if each BAC containing a marker had a known chromosomal position on the tomato physical map, thereby indirectly placing the marker on the physical map.

Principal component analysis

Genotypic data from the germplasm collection was converted into allele frequencies based on their occurrence in the genome (0, 0.5, and 1) and analysed using the SAS PRINCOMP procedure (Version 9.1 for Windows, SAS Institute, Cary, NC). This approach allows for incorporation of SSR data that may be multi-allelic into the analysis. The eigenvalues of the first three principal components were extracted for each variety, and an analysis of variance (ANOVA) was performed using the General Linear Models procedure in order to test whether the market classes were significantly different.

Linkage disequilibrium analysis

Marker genotypes were used to measure the extent of LD within cultivated tomatoes (processing, fresh market, and vintage cultivars combined) as well as processing and fresh market cultivars separately. All other entries (greenhouse varieties, unimproved breeding lines, Latin American cultivars, and wild species) lacked sufficient representatives (<10 entries for each class) and were eliminated from the analysis. Only markers that were both placed on the integrated linkage map and polymorphic within cultivated tomato were used for LD analysis. Both the GGT 2.0 (van Berloo, 2008) and TASSEL (Bradbury et al., 2007) software were used to calculate pair-wise r² values between 114 markers distributed throughout the genome. P values for each r² estimate were calculated using 1000 permutations in TASSEL.

The decay of LD over genetic distance was investigated by plotting pair-wise r² values against the distance (cM) between markers on the same chromosome (Fig. 3). A smooth line was fit to the data using second-degree locally weighted scatterplot smoothing (LOESS; Breseghello and Sorrells, 2006) as implemented in SAS. To describe the relationship between LD decay and genetic distance, two methods of establishing baseline r² values were investigated. Critical values of r² were based on a fixed value of 0.1 (Nordborg et al., 2002; Palaisa et al., 2003; Remington et al., 2001) and from the parametric 95th percentile of the distribution of the unlinked markers (Breseghello and Sorrells, 2006). The relationship between these baseline r² values and genetic distance was determined using the LOESS curve and a 1 cM moving means approach. For the LOESS estimation of LD decay, genetic distance was estimated as the point where the LOESS curve first crosses the baseline r² value. For the moving means approach, the distance between linked markers was used to divide marker pairs into bins of 1 cM. Markers separated by 0–0.9 cM were placed in the first bin, marker distances from 1–1.9 were in the second bin, etc. The mean of the r² values within each bin was calculated and LD decay was estimated as the first bin where the baseline r² value was lower than the bin mean.

Fig. 3.

Plots of linkage disequilibrium (LD) values (r²) against genetic distance (cM) between pairs of markers in multiple classes of cultivated tomato. All possible pair-wise combinations of markers on the same chromosome were plotted to visualize LD decay within chromosomes over the entire genome. The r² values were calculated separately for processing and fresh market cultivars (B and C, respectively) as well as processing, fresh market, and vintage cultivar classes combined (A). Curves were fit for each plot by second-degree LOESS. The horizontal dotted lines indicate the baseline r² values based on the 95th percentile of the distribution of unlinked r² values (black) and the fixed r² value of 0.1 (grey).

To visualize LD throughout the genome, heat maps were produced based on pair-wise r² estimates and their P values for all marker pairs (Fig. 4). These heat maps were used to identify variation in disequilibrium between tomato classes at specific genomic locations. Differences were tested by comparing r² estimates of marker pairs in the region using a paired t test in SAS. Only marker pairs with r² estimates in both classes were included in the comparison.

Fig. 4.

Heat maps of linkage disequilibrium (LD) values (r²) throughout the tomato genome. Markers were ordered on the x and y axes based on genomic location so that each cell of the heat map represents a single marker pair. The r² values for each marker pair are on the bottom half of the heat map and are represented by shades of grey from 0.0 (white) increasing in darkness in equal increments of 0.1 to 1.0 (black). The P-values of each r² estimate are on the top half of the heat map and are represented by shades of grey from non-significant (P >0.05; white) to highly significant (P <0.0001; black). The combined analysis includes processing, fresh market, and vintage cultivars.

Results

A germplasm collection representing currently relevant and historical tomato germplasm was genotyped with 340 indel, SNP, and SSR markers (see Supplementary Tables S2–S5 at JXB online). Markers had been pre-selected based on their potential for polymorphism within cultivated tomato. The majority of the markers were polymorphic within our collection of cultivated tomato varieties (74%) while over 85% were polymorphic within wild species. Fifty per cent of the markers were polymorphic within processing and 47% were polymorphic within fresh market germplasm (Table 1).

Genotypic information from the germplasm collection was utilized to identify markers that could be mapped in either of two F₂ populations. For the Sun1642×LA1589 population, a total of 153 framework (SSR and RFLP) and 119 SNP and indel markers were mapped (Table 2). The order of the framework markers generally matched that of the Tomato-EXPEN 2000 map (SGN; http://solgenomics.net) without using a fixed marker order for all chromosomes except for chromosome 4. Using a fixed order of TG15, TG483, CT157, CT178, CT50, and TG163 derived from Tomato-EXPEN 2000 reduced the χ² value from 123.7 to 25.5 and increased the map length from 56.7 cM to 114.1 cM. The total length of the Sun1642×LA1589 map was 1127 cM with an average of 4.3 cM between markers and the largest gap of 21.4 cM on chromosome 4. Segregation distortion was detected on chromosomes 6, 7, 11, and 12 (Table 2), with distorted markers adjacently located and skewed in the direction of the same parental allele indicating biased transmission. Ninety-four per cent of the genome was within 10 cM of at least one SSR, SNP or indel marker.

The Yellow Stuffer×LA1589 population map contains 90 framework markers with 60 new SNP and indel markers (Table 2). As with the Sun1642×LA1589 population, chromosome 4 was the only chromosome where a fixed marker order was employed. Using a fixed order increased the χ² value from 36.7 to 124.8 and increased the map length from 97.4 cM to 114.8 cM. The Yellow Stuffer×LA1589 map had an average of 7.8 cM between markers, the largest gap of 33.2 cM on chromosome 10, and a total length of 1072 cM. Twenty-three per cent of the markers did not fit expected segregation ratios with the highest distortion on chromosomes 2, 7, 9, and 11 and distortion patterns indicating biased transmission. In this map, 72% of the genome was within 10 cM of a SSR, SNP or indel marker.

Eighty-five framework markers common to both maps allowed the creation of an integrated map with 338 markers including 180 new SNPs and indels (Table 2; Fig. 1). The average distance between markers was 3.6 cM with the largest gap on chromosome 9 of 18.4 cM. The total map length was 1151 cM with 96% of the genome within 10 cM of a PCR-based marker.

Emerging sequence data from the BAC-by-BAC international genome sequencing project were also used to identify the location of makers (see Supplementary Table S7 at JXB online). The sequence of 415 marker loci with verified polymorphisms was used as a BLAST query against the tomato genome sequence and 136 (33%) loci met the threshold for association with a BAC (see Materials and methods). The SGN data provided a chromosome assignment for 129 loci (31%), 60 (14%) of which had a precise location on the physical map. Forty-nine of the loci with a known chromosome from physical mapping were also placed on the genetic linkage map, allowing the two mapping methods to be compared. Out of these 49 loci, the chromosome designation of 48 (98%) matched. For the loci that were not placed on the linkage map, physical mapping provided the chromosome designation of 80 loci, 35 of which had a physical map position (see Supplementary Table S7 at JXB online). These loci were placed next to our integrated linkage map relative to the framework markers (Fig. 1). In addition, 18 polymorphic loci, whose physical position was previously determined (Van Deynze et al., 2007), were integrated into the map. Thus, 53 additional loci were added to the map based on physical position.

Principal Components Analysis (PCA) was used to visualize and test relationships between market classes within the collection of varieties. When processing, fresh market and vintage varieties were analysed together, the first three principal components explained 21.8% of the total variation and clear clusters emerged (Fig. 2). The hypothesis that market classes were distinct was tested by performing an analysis of variance (ANOVA) based on PCA. Both PC1 and PC2 were significantly different (P <0.0001). Mean separations demonstrated that all three classes were separated along PC1. For PC2, contemporary fresh market varieties were significantly different from contemporary processing and vintage varieties, but the latter two were not significantly different.

Fig. 2.

Graph of the first three Principal Components based on marker frequencies. Fresh market (circles), processing (squares), and vintage (triangles) cultivar groups are indicated by unique symbols.

Analysis of LD was performed for a data set consisting of contemporary and vintage varieties and separately for the two contemporary market classes. A difference was observed in both the decay of LD over genetic distance and the amount of inter-chromosomal LD between the three analyses. Based on the LOESS curves, the rate of LD decay was more pronounced for the combined entries followed by processing and then fresh market germplasm. The LOESS curves also indicate that LD decays over multiple centimorgans. The baseline r² values of 0.160 (combined), 0.248 (processing), and 0.464 (fresh market) estimated by the 95th percentile method correspond to 6.9, 6.9, and 3.0 cM on the LOESS curves, respectively (Fig. 3; Table 3). By contrast, a fixed baseline r² value of 0.1 equates to 8.0 (combined), 14.2 (processing), and 16.1 (fresh market) cM on the LOESS curve. Using a 1 cM moving means method, the 95th percentile baseline r² values correspond to the 6 (combined), 6 (processing), and 2 (fresh market) cM bins, while the fixed baseline fell in bins 6, 9, and 10, respectively. In general, using a fixed r² baseline provided larger decay estimates than the 95th percentile method. The difference in estimates between methods was especially large in fresh market varieties and probably reflects the distribution associated with unlinked loci. The baseline r² values estimated by the 95th percentile method are based on the unlinked loci, and larger baseline estimates for fresh market cultivars reflect a high level of LD between markers on different chromosomes (inter-chromosomal LD) in this group. The patterns of LD can also be visualized across the genome from the diagonal of the heat maps (Fig. 4). Processing and fresh market germplasm share a similar degree of LD on chromosomes 3, 4, and 11. Processing cultivars have greater LD on chromosomes 1, 2, and 5, while LD is higher on chromosomes 6 and 9 for fresh market cultivars.

Table 3.

Summary of genome-wide linkage disequilibrium analysis

Market class	No. marker pairsa	r2 estimatesb				Linkage disequilibrium decay (cM)e
		Median	St. Dev.	95th percentilec	P <0.01d	LOESSf		Moving meansg
						95th percentile method	Fixed r2 (0.1) method	95th percentile method	Fixed r2 (0.1) method
Combinedh	5248	0.011	0.102	0.160	8.1%	6.6	8.0	6	6
Processing	3294	0.037	0.131	0.248	5.5%	6.9	14.2	6	9
Fresh market	2622	0.031	0.187	0.464	2.0%	3.0	16.1	2	10

^aThe number of marker pairs includes only markers polymorphic within each market class.

^bLinkage disequilibrium was estimated as r² values for all possible marker pairs using TASSEL (Bradbury et al., 2007) and GGT (van Berloo, 2008) software.

^cThe 95th percentile of the distribution of r² values for the unlinked markers. This value is the baseline r² to estimate LD decay.

^dPercentage of r² estimates with P value <0.01. P values of r² estimates were calculated from 1000 permutations using TASSEL software (Bradbury et al., 2007).

^eLinkage disequilibrium decay was estimated over genetic distance by the relationship of a baseline r² estimate to linked marker pairs using two methods, LOESS and 1 cM moving means. The baseline r² value was either fixed at 0.1 or estimated using the 95th percentile of the unlinked markers. Values for r² that exceed the baseline are considered to be in linkage disequilibrium.

^fFor the LOESS estimation of LD decay, genetic distance was estimated as the point where the LOESS curve first crosses the baseline r² value.

^gFor the means estimation of LD decay, the r² values of linked markers were grouped into bins of 1 cM based on the distance between markers. LD decay was estimated as the first bin where the baseline r² value was lower than the bin mean.

^hThe combined analysis includes processing, fresh market, and vintage cultivars.

The heat maps also reveal patterns of LD between markers on different chromosomes in the combined, processing, and fresh market groups, suggesting that inter-chromosomal LD is present within cultivated germplasm. Separating the market classes removed some of the observed inter-chromosomal LD, though residual patterns remain. Values of inter-chromosomal r² tend to be higher in the fresh market germplasm, though statistically significant inter-chromosomal LD was detected for both market classes (Fig. 4). The location of the inter-chromosomal disequilibrium differs between these two classes (Fig. 4; Table 4). Pair-wise t tests of r² values indicate that processing lines have significant disequilibrium between chromosomes 2 and 3, 2 and 4, and 3 and 12 (Table 4). Fresh market varieties have significant disequilibrium between chromosomes 2 and 3, 2 and 4, 3 and 4, 3 and 11, and 4 and 6. The regions of chromosomes 2, 3, and 4 that are in disequilibrium differ for the market classes with shifts on chromosome 2 and 4 being particularly important in distinguishing patterns (Table 4).

Table 4.

Comparison of inter-chromosomal linkage disequilibrium between processing and fresh market tomato germplasm

Chromosomea	Positionb	Chromosomea	Positionb	No.c	Processing		Fresh Market		P-valuee
					Mean r2d	St. Dev.	Mean r2d	St. Dev.
2	36.3–47.3	3	71.2–87.9	33	0.0648	0.0682	0.5776	0.2813	<0.0001
2	47.3–51.6	3	71.2–76.7	10	0.2094	0.0287	0.0203	0.0167	<0.0001
2	36.3–45.2	4	100.0–105.7	10	0.2278	0.1610	0.0569	0.0525	0.0372
2	36.3–47.3	4	53.2–61.7	30	0.0294	0.0249	0.4362	0.2324	<0.0001
3	76.7–87.9	4	53.2–61.7	17	0.0506	0.0407	0.4837	0.2346	<0.0001
3	76.7–87.9	11	46.4–48.5	8	0.0581	0.0777	0.3346	0.1080	0.0009
3	52.5–94.9	12	49.7–65.8	13	0.1596	0.1496	0.0257	0.0196	0.0012
4	53.2–68.5	11	46.4–48.5	18	0.0249	0.0228	0.2358	0.1306	<0.0001

^aChromosomes being compared.

^bGenetic map position (cM) within the specified chromosomes. The position is derived from the integrated linkage map (Fig. 1).

^cNumber of marker pairs in the comparison. Only marker pairs with r² estimates in both classes were included.

^dMean r² values of all marker pairs between the two chromosomal regions.

^eP value of a paired t test of the mean r² estimates of processing versus fresh market entries.

Discussion

In order to develop resources for the evaluation of genetic variation within cultivated tomato further, 434 markers were integrated based on a combination of linkage mapping in F₂ populations and physical mapping relative to emerging sequence data. Three-hundred-and-forty markers, including 226 that were mapped based on linkage and/or physical location were used to genotype a collection of tomato lines representing wild species, landraces, vintage cultivars, and contemporary varieties. The markers differentiated the collection into market classes and >70% were polymorphic within cultivated tomatoes. These mapping and genotypic data are presented in Supplementary Tables S2–S7 at JXB online and are also available on the Tomato Mapping Resource Database under the sections Polymorphic Marker Search and Search Marker (http://www.tomatomap.net).

Our linkage map was generally consistent with the Tomato-EXPEN 2000 map. The integrated map is 21% shorter than the 1460.5 total cM of the Tomato-EXPEN 2000 map. This discrepancy may simply be due to the characteristics of the mapping populations (e.g. mapping parents of different species) or the general expansion of linkage maps with the addition of more markers. Our integrated map length is comparable with the Tomato-EXPIMP2001 (1275 total cM) and Tomato-EXPIMP2008 (1228 total cM) maps which have the same S. pimpinellifolium parent and fewer markers (145 and 181, respectively).

Segregation distortion was detected on chromosomes 6, 7, 11, and 12 for the Sun1642×LA1589 population, and chromosomes 2, 7, 9, and 11 for the Yellow Stuffer×LA1589 population. Segregation distortion is commonly observed in wide crosses of tomato and other species as the consequence of linkage between loci that operate in pre- and post-zygotic phases of reproduction (Zamir and Tadmor, 1986; Chetelat et al., 1989, 2000). The implications of distorted segregation on the map were tested by removing markers and repeating the mapping process. For the reported markers, segregation distortion does not significantly alter the map.

The use of BLAST to anchor markers to publicly available genome sequence data from the International Tomato Genome Sequencing Project (http://solgenomics.net/about/tomato_sequencing.pl) resulted in a physical association for 33% of our markers. At the time of our analysis, the sequencing effort was estimated to be 41% complete, suggesting that >80% of our markers will eventually be represented in BAC sequence. A high level of agreement (98%) was observed between markers that were mapped both physically and genetically. Thus, using the tomato genome sequence provides a robust method to identify the genomic location of unmapped loci. This approach will become the preferred method to map markers with the completion of a robust integrated tomato genome sequence in the near future.

Knowledge of the extent and structure of LD is important to assess the usefulness of association mapping strategies (Rostoks et al., 2006). The decay of LD over physical or genetic distance determines the depth of resolution as well as the density of markers needed for association analysis (Yu and Buckler, 2006). LD decay was estimated at 6–8 cM across all varieties, 6-14 cM within processing varieties, and 3–16 cM within fresh market varieties with the range dependent on the methods used to estimate threshold values and decay. The large range in fresh market estimates illustrates the difference between the methods used to establish a critical r² value. Rather than selecting an arbitrary fixed value, the 95th percentile method relies on unlinked markers. As such, the estimate is influenced by inter-chromosomal LD and takes into account properties of the entries measured that may lead to population structure (Breseghello and Sorrells, 2006). Thus, estimates based on this method are more reflective of the sample. Our LD decay estimates are consistent with previous studies. In commercial European greenhouse varieties LD decayed over 15–20 cM (van Berloo et al., 2008). Labate et al. (2009a) found that intra-locus LD was high with a plateau at r²=0.6 over 1000 bp in 31 tomato landraces. Since LD decays over centimorgans in cultivated tomato, association mapping is theoretically feasible with a small number of markers.

Although our results suggest that marker numbers may be favourable for association mapping in cultivated tomato, the extent of inter-chromosomal LD between unlinked markers is likely to confound association analyses. For example, linkage disequilibrium between two genomic locations in a tomato mapping population resulted in a significant, but spurious marker–trait association that was not confirmed in subsequent populations (Robbins et al., 2009). Significant inter-chromosomal LD was identified within cultivated tomato that differed between fresh market and processing tomatoes. In a previous study, different patterns of inter-chromosomal LD were identified between cherry and beef-round tomatoes (van Berloo et al., 2008). The majority of chromosome pairs with disequilibrium differed from those we detected, suggesting that inter-chromosomal LD is population dependent and should be determined for each population of interest. In a separate study among tomato landraces, 19% of inter-locus marker pairs showed significant LD while only 10% of these were located on the same chromosome (Labate et al., 2009a). These results suggest that inter-chromosomal LD will complicate association analyses in cultivated tomato.

Linkage disequilibrium is caused by many factors including recombination rate, drift, mating system, selection, effective population size, and population structure (reviewed by Rafalski and Morgante, 2004). It appears that, in tomato, genetic bottlenecks, introgressions from wild species, and intense selection for market specialization have established haplotype blocks with disequilibrium over long physical distances. Such haplotype blocks have been identified in the genome of humans (Patil et al., 2001), mice (Wiltshire et al., 2003), dogs (Lindblad-Toh et al., 2005), rice (Tang et al., 2006; Li et al., 2009), and maize (Gore et al., 2009). It is hypothesized that, in tomato, some of the observed inter-chromosomal disequilibrium was produced by selection for the desired combinations of characters. The differences observed in LD patterns between fresh market and processing market types suggest that plant breeders may have selected for separate combinations of genes during the development of ideotypes for specialized markets.

Tomato has gone through several genetic bottlenecks during domestication, its introduction into Europe from Latin America, and its introduction into North America from Europe and the Caribbean (Rick, 1976; Miller and Tanskley, 1990; Labate et al., 2007). Early tomato improvement depended largely on mutation, spontaneous outcrossing, and recombination of available genetic variation to provide variability for selection (Rick, 1976). It was not until the 1920s that breeding programmes were established for tomato cultivar development (Stevens and Rick, 1986). Since then, the application of genetic principles and the continued innovation of breeding practices accelerated the pace of tomato improvement (Rick, 1976). High selection pressure for desired phenotypes in a limited germplasm pool, coupled with the high degree of self-pollination and multiple bottlenecks within the cultivated species have contributed to the narrow genetic base of tomato (Rick, 1976; Miller and Tanskley, 1990; Park et al., 2004). To overcome this challenge, breeding practices dating back to the 1930s have utilized wild tomato species for the introgression of new genetic variation, especially for disease resistance. At the same time, these practices reduced recombination in linkage blocks associated with introgressed segments (MacArthur and Butler, 1938; Alexander, 1949; Miller and Tanskley, 1990; Williams and St Clair, 1993; Park et al., 2004; Sim et al., 2009). Efforts to develop tomatoes specifically for mechanical harvest began in the late 1940s and by the mid 1960s, acceptable varieties were available (Rasmussen, 1968). The emphasis in breeding processing tomatoes suitable for mechanical harvest caused a divergence between fresh market and processing types. Results from this study support the hypothesis that breeding for market specialization is a major driving force for genetic differentiation between fresh market and processing varieties.

Our mapping and genetic data will provide a resource for researchers interested in using molecular markers for tomato improvement. Different patterns of LD between fresh market and processing varieties highlight how breeding practices have altered the genomes of market classes within cultivated tomato germplasm. The extent of inter-chromosomal LD in contemporary varieties leads us to hypothesize that market specialization has preserved certain favourable combinations of alleles. Breeders may choose to preserve these combinations, while also accessing and testing the affect of variation derived from different market classes. Extensive inter-chromosomal LD also suggests that association mapping should be conducted with caution to avoid detection of spurious marker–trait linkage.

Supplementary data

Supplementary data can be found at JXB online.

Supplementary Table S1. Description of 102 tomato accessions used in this study.

Supplementary Table S2. SNP markers in this study detected by allele specific primer extension (ASPE) assay.

Supplementary Table S3. SNPs detected as CAPS markers in this study.

Supplementary Table S4. Indel markers used in this study.

Supplementary Table S5. SSR markers used in this study.

Supplementary Table S6. Marker locations on the Sun1642×LA1589, Yellow Stuffer×LA1589, and integrated maps.

Supplementary Table S7. Location of markers placed on the tomato physical map compared to the integrated linkage map.