The tomato genome sequence provides insights into fleshy fruit evolution.

Tomato (Solanum lycopersicum) is a major crop plant and a model system for fruit development. Solanum is one of the largest angiosperm genera and includes annual and perennial plants from diverse habitats. Here we present a high-quality genome sequence of domesticated tomato, a draft sequence of its closest wild relative, Solanum pimpinellifolium, and compare them to each other and to the potato genome (Solanum tuberosum). The two tomato genomes show only 0.6% nucleotide divergence and signs of recent admixture, but show more than 8% divergence from potato, with nine large and several smaller inversions. In contrast to Arabidopsis, but similar to soybean, tomato and potato small RNAs map predominantly to gene-rich chromosomal regions, including gene promoters. The Solanum lineage has experienced two consecutive genome triplications: one that is ancient and shared with rosids, and a more recent one. These triplications set the stage for the neofunctionalization of genes controlling fruit characteristics, such as colour and fleshiness.

Introductory Paragraph

Tomato (Solanum lycopersicum) is a major crop plant and a model system for fruit development. Solanum is one of the largest angiosperm genera[1] and includes annual and perennial plants from diverse habitats. We present a high quality genome sequence of domesticated tomato, a draft sequence of its closest wild relative, S. pimpinellifolium[2], and compare them to each other and to potato (S. tuberosum). The two tomato genomes show only 0.6% nucleotide divergence and signs of recent admixture, but show >8% divergence from potato, with nine large and several smaller inversions. In contrast to Arabidopsis, but similar to soybean, tomato and potato, small RNAs map predominantly to gene-rich chromosomal regions, including gene promoters. The Solanum lineage has experienced two consecutive genome triplications: one that is ancient and shared with rosids, and a more recent one. These triplications set the stage for the neofunctionalization of genes controlling fruit characteristics, such as colour and fleshiness.

Main Text

The genome of the inbred tomato cultivar ‘Heinz 1706’ was sequenced and assembled using a combination of Sanger and “next generation” technologies (Supplementary Section 1). The predicted genome size is ~900 Mb, consistent with prior estimates[3], of which 760 Mb were assembled in 91 scaffolds aligned to the 12 tomato chromosomes, with most gaps restricted to pericentromeric regions (Fig. 1A; Supplementary Fig. 1). Base accuracy is approximately one substitution error per 29.4 kb and one indel error per 6.4 kb. The scaffolds were linked with two BAC-based physical maps and anchored/oriented using a high-density genetic map, introgression line mapping and BAC fluorescence in situ hybridisation (FISH).

Figure 1

A. Multi-dimensional topography of tomato chromosome .

(a) Left: contrast-reversed, DAPI-stained pachytene chromosome; centre and right: FISH signals for repeat sequences on diagrammatic pachytene chromosomes: TGR1 purple, TGR4 blue, telomere repeat red, Cot 100 DNA (including most repeats) green. (b) Frequency distribution of recombination nodules representing crossovers on 249 chromosomes. Red stars mark 5 cM intervals starting from the end of the short arm (top). Scale is in micrometers. (c) FISH-based locations of selected BACs (horizontal blue lines on left). (d) Kazusa F2-2000 linkage map. Blue lines to the left connect linkage map markers on the (c) BAC-FISH map, (e) heat maps and (f) DNA pseudomolecule. (e) From left to right: linkage map distance (cM/Mb, turquoise); repeated sequences (% nucleotides/500 kb, purple); genes (% nucleotides/500 kb, blue); chloroplast insertions; RNA-Seq reads from leaves and breaker fruits of S. lycopersicum and S. pimpinellifolium (number of reads/500 kb, green and red, respectively); microRNA genes (transcripts per million/500 kb, black); small RNAs (thin horizontal black lines, sum of hits-normalized abundances). Horizontal grey lines represent gaps in the pseudomolecule (f). (f) DNA pseudomolecule consisting of nine scaffolds. Unsequenced gaps (approximately 9.8 Mb, Supplementary Table 13) are indicated by white horizontal lines. Tomato genes identified by map-based cloning (Supplementary Table 14) are indicated on the right. For more details, see legend to Supplementary Figure 1.

B. Syntenic relationships in the .

COSII-based comparative maps of potato, eggplant, pepper and Nicotiana with respect to the tomato genome (Supplementary section 4.5, Supplementary Fig. 14). Each tomato chromosome is assigned a different colour and orthologous chromosome segment(s) in other species are shown in the same colour. White dots indicate approximate centromere locations. Each black arrow indicates an inversion relative to tomato and “+1”indicates a minimum of one inversion. Each black bar beside a chromosome indicates translocation breakpoints relative to tomato. Chromosome lengths are not to scale, but segments within chromosomes are.

C. Tomato-potato syntenic relationships.

Dot plot of tomato and potato genomic sequences based on collinear blocks Supplementary Section 4.1). Red and blue dots represent gene pairs with statistically significant high and low ω (Ka/Ks) in collinear blocks, which average Ks≤0.5, respectively. Green and magenta dots represent genes in collinear blocks which average 0.51.5, respectively. Yellow dots represent all other gene pairs. Blocks circled in red are examples of pan-eudicot triplication. Inserts represent schematic drawings of BAC-FISH patterns of cytologically demonstrated chromosome inversions (also in Supplementary Fig. 15).

The genome of S. pimpinellifolium (accession LA1589) was sequenced and assembled de novo using Illumina short reads, yielding a 739 Mb draft genome (Supplementary Section 3). Estimated divergence between the wild and domesticated genomes is 0.6% (5.4M SNPs distributed along the chromosomes (Fig. 1A, Supplementary Fig. 1)). Tomato chromosomes consist of pericentric heterochromatin and distal euchromatin, with repeats concentrated within and around centromeres, in chromomeres and at telomeres (Fig. 1A, Supplementary Fig. 1). Substantially higher densities of recombination, genes and transcripts are observed in euchromatin, while chloroplast insertions (Supplementary Sections 1.22-1.23) and conserved miRNA genes (Supplementary Section 2.9) are more evenly distributed throughout the genome. The genome is highly syntenic with those of other economically important Solanaceae (Fig. 1B). Compared to the genomes of Arabidopsis[4] and sorghum[5], tomato has fewer high-copy, full-length LTR retrotransposons with older average insertion ages (2.8 versus 0.8 mya) and fewer high-frequency k-mers (Supplementary Section 2.10). This supports previous findings that the tomato genome is unusual among angiosperms by being largely comprised of low-copy DNA[6,7].

The pipeline used to annotate the tomato and potato[8] genomes is described in Supplementary Section 2. It predicted 34,727 and 35,004 protein-coding genes, respectively. Of these, 30,855 and 32,988, respectively, are supported by RNA-Seq data, and 31,741 and 32,056, respectively, show high similarity to Arabidopsis genes (Supplementary section 2.1). Chromosomal organisation of genes, transcripts, repeats and sRNAs is very similar in the two species (Supplementary Figures 2-4). The protein coding genes of tomato, potato, Arabidopsis, rice and grape were clustered into 23,208 gene groups (≥2 members), of which 8,615 are common to all five genomes, 1,727 are confined to eudicots (tomato, potato, grape and Arabidopsis), and 727 are confined to plants with fleshy fruits (tomato, potato and grape) (Supplementary Section 5.1, Supplementary Fig. 5). Relative expression of all tomato genes was determined by replicated strand-specific Illumina RNA-Seq of root, leaf, flower (2 stages) and fruit (6 stages) in addition to leaf and fruit (3 stages) of S. pimpinellifolium (Supplementary Table 1).

sRNA sequencing data supported the prediction of 96 conserved miRNA genes in tomato and 120 in potato, a number consistent with other plant species (Fig. 1A, Supplementary Figures 1 and 3, Supplementary Section 2.9). Among the 34 miRNA families identified, 10 are highly conserved in plants and similarly represented in the two species, whereas other, less conserved families are more abundant in potato. Several miRNAs, predicted to target TIR-NBS-LRR genes, appeared to be preferentially or exclusively expressed in potato (Supplementary Section 2.9).

Supplementary section 4 deals with comparative genomic studies. Sequence alignment of 71 Mb of euchromatic tomato genomic DNA to their potato[8] counterparts revealed 8.7% nucleotide divergence (Supplementary Section 4.1). Intergenic and repeat-rich heterochromatic sequences showed more than 30% nucleotide divergence, consistent with the high sequence diversity in these regions among potato genotypes[8]. Alignment of tomato-potato orthologous regions confirmed 9 large inversions known from cytological or genetic studies and several smaller ones (Fig. 1C). The exact number of small inversions is difficult to determine due to the lack of orientation of most potato scaffolds. 18,320 clearly orthologous tomato-potato gene pairs were identified. Of these, 138 (0.75%) had significantly higher than average non-synonymous (Ka) versus synonymous (Ks) nucleotide substitution rate ratios (ω), suggesting diversifying selection, whereas 147 (0.80%) had significantly lower than average ω, suggesting purifying selection (Supplementary Table 2). The proportions of high and low ω between sorghum and maize (Zea mays) are 0.70% and 1.19%, respectively, after 11.9 Myr of divergence[9], suggesting that diversifying selection may have been stronger in tomato-potato. The highest densities of low-ω genes are found in collinear blocks with average Ks >1.5, tracing to a genome triplication shared with grape (see below) (Fig. 1C, Supplementary Fig. 6, Supplementary Table 3). These genes, which have been preserved in paleo-duplicated locations for more than 100 Myr[10,11] are more constrained than ‘average’ genes and are enriched for transcription factors and genes otherwise related to gene regulation (Supplementary Tables 3-4).

Sequence comparison of 32,955 annotated genes in tomato and S. pimpinellifolium revealed 6,659 identical genes and 3,730 with only synonymous changes. A total of 22,888 genes had non-synonymous changes, including gains and losses of stop codons with potential consequences for gene function (Supplementary Tables 5-7). Several pericentric regions, predicted to contain genes, are absent or polymorphic in the broader S. pimpinellifolium germplasm (Supplementary Table 8, Supplementary Fig. 7). Within cultivated germplasm, particularly among the small-fruited cherry tomatoes, several chromosomal segments are more closely related to S. pimpinellifolium than to ‘Heinz 1706’ (Supplementary Figures 8-9), supporting previous observations on recent admixture of these gene pools due to breeding[12]. ‘Heinz 1706’ itself has been reported to carry introgressions from S. pimpinellifolium[13], traces of which are detectable on chromosomes 4, 9, 11 and 12 (Supplementary Table 9).

Comparison of the tomato and grape genomes supports the hypothesis that a whole-genome triplication affecting the rosid lineage occurred in a common eudicot ancestor[11] (Fig. 2B). The distribution of Ks between corresponding gene pairs in duplicated blocks suggests that one polyploidisation in the solanaceous lineage preceded the rosid-asterid (tomato-grape) divergence (Supplementary Fig. 10).

Figure 2

The Solanum whole genome triplication

A. Based on alignments of multiple tomato genome segments to single grape genome segments, the tomato genome is partitioned into three non-overlapping ‘subgenomes’ (T1, T2, T3), each represented by one axis in the 3D plot. The ancestral gene order of each subgenome is inferred according to orthologous grape regions, with tomato chromosomal affinities shown by red-shaded (inner) bars. Segments tracing to pan-eudicot triplication (γ) are shown by green-shaded (outer) bars with colours representing the seven putative pre-γ eudicot ancestral chromosomes[10], also coded a-g.

B. Speciation and polyploidisation in eudicot lineages. Confirmed whole-genome duplications and triplications are shown with annotated circles, including “T” (this paper) and previously discovered events α, β, γ[10,11,14]. Dashed circles represent one or more suspected polyploidies reported in previous publications that need further support from genome assemblies[27,28]. Grey branches indicate unpublished genomes. Black and red error bars bracket, respectively, the likely timings of divergence of major asterid lineages and of “T”. The post-“T” subgenomes, designated T1, T2, and T3, are further detailed in Supplementary Fig. 10.

Comparison to the grape genome also reveals a more recent triplication in tomato and potato. While few individual tomato/potato genes remain triplicated (Supplementary Tables 10-11), 73% of tomato gene models are in blocks that are orthologous to one grape region, collectively covering 84% of the grape gene space. Among these grape genomic regions, 22.5% have one orthologous region in tomato, 39.9% have two, and 21.6% have three, indicating that a whole genome triplication occurred in the Solanum lineage, followed by widespread gene loss. This triplication, also evident in potato (Supplementary Fig. 11) is estimated at 71 (+/-19.4) mya based on Ks of paralogous genes (Supplementary Fig. 10), and therefore predates the ~7.3 mya tomato-potato divergence. Based on alignments to single grape genome segments, the tomato genome can be partitioned into three non-overlapping ‘subgenomes’ (Fig. 2A). The number of euasterid lineages that have experienced the recent triplication remains unclear and awaits complete euasterid I and II genome sequences. Ks distributions show that euasterids I and II, and indeed the rosid-asterid lineages, all diverged from common ancestry at or near the pan-eudicot triplication (Fig. 2B), suggesting that this event may have contributed to formation of major eudicot lineages in a short period of several million years[14], partially explaining the explosive radiation of angiosperm plants on earth[15].

Supplementary section 5 reports on the analysis of specific gene families. Fleshy fruits (Supplementary Fig. 12) are an important means of attracting vertebrate frugivores for seed dispersal[16]. Combined orthology and synteny analyses suggest that both genome triplications added new gene family members that mediate important fruit-specific functions (Fig. 3). These include transcription factors and enzymes necessary for ethylene biosynthesis (RIN, CNR, ACS) and perception (LeETR3/NR, LeETR4)[17], red light photoreceptors influencing fruit quality (PHYB1/PHYB2) and ethylene- and light-regulated genes mediating lycopene biosynthesis (PSY1/PSY2). Several cytochrome P450 subfamilies associated with toxic alkaloid biosynthesis show contraction or complete loss in tomato and the extant genes show negligible expression in ripe fruits (Supplementary Section 5.4).

Figure 3

Whole genome triplications set the stage for fruit-specific gene neofunctionalisation

The genes shown represent a fruit ripening control network regulated by transcription factors (MADS-RIN, CNR) necessary for production of the ripening hormone ethylene, the production of which is regulated by ACC synthase (ACS). Ethylene interacts with ethylene receptors (ETRs) to drive expression changes in output genes, including phytoene synthase (PSY), the rate-limiting step in carotenoid biosynthesis. Light, acting through phytochromes, controls fruit pigmentation through an ethylene-independent pathway. Paralogous gene pairs with different physiological roles (MADS1/RIN, PHYB1/PHYB2, ACS2/ACS6, ETR3/ETR4, PSY1/PSY2), were generated during the eudicot (γ, black circle) or the more recent, Solanum (T, red circle) triplications. Complete dendrograms of the respective protein families are shown in Supplementary Figures 16 and 17.

Fruit texture has profound agronomic and sensory importance and is controlled in part by cell wall structure and composition[18]. More than 50 genes showing differential expression during fruit development and ripening encode proteins involved in modification of wall architecture (Fig. 4A and Supplementary Section 5.7). For example, a family of xyloglucan endotransglucosylase-/hydrolases (XTHs) has expanded both in the recent whole genome triplication and through tandem duplication. One of the triplicated members, SlXTH10, shows differential loss between tomato and potato (Fig. 4A, Supplementary Table 12), suggesting genetically driven specialisation in the remodelling of fruit cell walls.

Figure 4

The tomato genome allows systems approaches to fruit biology

A. Xyloglucan transglucosylase-hydrolases (XTHs) differentially expressed between mature green and ripe fruits (Supplementary Section 5.7). These XTH genes and many others are expressed in ripening fruits and are linked with the Solanum triplication, marked with a red circle on the phylogenetic tree. Red lines on the tree denote paralogs derived from the Solanum triplication, and blue lines are tandem duplications.

B. Developmentally regulated accumulation of sRNAs mapping to the promoter region of a fruit-regulated cell wall gene (Pectin acetylesterase, Solyc08g005800). Variation of abundance of sRNAs (left) and mRNA expression levels from the corresponding gene (right) over a tomato fruit developmental series (T1 – bud, T2 – flower, T3 – fruit 1- 3mm, T4 – fruit 5-7mm, T5 – fruit 11-13mm, T6 – fruit mature green, T7 – breaker, T8 – breaker+3days, T9 – breaker+7days). The promoter regions are grouped in 100nt windows. For each window the size class distribution of sRNAs is shown (21 – red, 22 – green, 23 – orange, 24 – blue). The height of the box corresponding to the first time point shows the cumulative sRNA abundance in log scale. The height of the following boxes is proportional to the log offset fold change (offset = 20) relative to the first time point. The expression profile of the mRNA is shown in log2 scale.

Similar to soybean and potato and in contrast to Arabidopsis, tomato sRNAs map preferentially to euchromatin (Supplementary Fig. 2). sRNAs from tomato flowers and fruits[19] map to 8,416 gene promoters. Differential expression of sRNAs during fruit development is apparent for 2,687 promoters, including those of cell wall-related genes (Fig. 4B) and occurs preferentially at key developmental transitions (e.g. flower to fruit, fruit growth to fruit ripening, Supplementary Section 2.8).

The genome sequences of tomato, S. pimpinellifolium and potato provide a starting point for comparing gene family evolution and sub-functionalization in the Solanaceae. A striking example is the SELF PRUNING (SP) gene family, which includes the homolog of Arabidopsis FT, encoding the mobile flowering hormone florigen[20] and its antagonist SP, encoding the ortholog of TFL1. Nearly a century ago, a spontaneous mutation in SP spawned the “determinate” varieties that now dominate the tomato mechanical harvesting industry[21]. The genome sequence has revealed that the SP family has expanded in the Solanum lineage compared to Arabidopsis, driven by the Solanum triplication and tandem duplication (Supplementary Fig. 13). In potato, SP3D and SP6A control flowering and tuberisation, respectively[22], whereas SP3D in tomato, known as SINGLE FLOWER TRUSS, similarly controls flowering, but also drives heterosis for fruit yield in an epistatic relationship with SP[23,24,25]. Interestingly, SP6A in S. lycopersicum is inactivated by a premature stop codon, but remains functionally intact in S. pimpinellifolium. Thus, allelic variation in a subset of SP family genes has played a major role in the generation of both shared and species-specific variation in Solanaceous agricultural traits.

The genome sequences of tomato and S. pimpinellifolium also provide a basis for understanding the bottlenecks that have narrowed tomato genetic diversity: the domestication of S. pimpinellifolium in the Americas, the export of a small number of accessions to Europe in the 16th Century, and the intensive breeding that followed. Charles Rick pioneered the use of trait introgression from wild tomato relatives to increase genetic diversity of cultivated tomatoes[26]. Introgression lines exist for seven wild tomato species, including S. pimpinellifolium, in the background of cultivated tomato. The genome sequences presented here and the availability of millions of SNPs will allow breeders to revisit this rich trait reservoir and identify domestication genes, providing biological knowledge and empowering biodiversity-based breeding.

Methods Summary

A total of 21 Gb of Roche/454 Titanium shotgun and matepair reads and 3.3 Gb of Sanger paired-end reads, including ~200,000 BAC and fosmid end sequence pairs, were generated from the ‘Heinz 1706’ inbred line (Supplementary Sections 1.1-1.7), assembled using both Newbler and CABOG and integrated into a single assembly (Supplementary Sections 1.17-1.18). The scaffolds were anchored using two BAC-based physical maps, one high density genetic map, overgo hybridization and genome-wide BAC FISH (Supplementary Sections 1.8-1.16 and 1.19). Over 99.9% of BAC/fosmid end pairs mapped consistently on the assembly and over 98% of EST sequences could be aligned to the assembly (Supplementary Section 1.20). Chloroplast genome insertions in the nuclear genome were validated using a matepair method and the flanking regions were identified (Supplementary Sections 1.22-1.24). Annotation was carried out using a pipeline based on EuGene that integrates de novo gene prediction, RNA-Seq alignment and rich function annotation (Supplementary Section 2). To facilitate interspecies comparison, the potato genome was re-annotated using the same pipeline. LTR retrotransposons were detected de novo with the LTR-STRUC program and dated by the sequence divergence between left and right solo LTR (Supplementary Section 2.10). The genome of S. pimpinellifolium was sequenced to 40x depth using Illumina paired end reads and assembled using ABySS (Supplementary Section 3). The tomato and potato genomes were aligned using LASTZ (Supplementary Section 4.1). Identification of triplicated regions was done using BLASTP, in-house generated scripts and three way comparisons between tomato, potato and S. pimpinellifolium using MCscan (Supplementary Sections 4.2-4.4). Specific gene families/groups (genes for ascorbate, carotenoid and jasmonate biosynthesis, cytochrome P450s, genes controlling cell wall architecture, hormonal and transcriptional regulators, resistance genes) were subjected to expert curation/analysis, (Supplementary Section 5). PHYML and MEGA were used to reconstruct phylogenetic trees and MCSCAN was used to infer gene collinearity (Supplementary Section 5.2).