Main

Theobroma cacao L. is a diploid tree fruit species (2n = 2x = 20 (ref. 1)) endemic to the South American rainforests. Cocoa was domesticated approximately 3,000 years ago2 in Central America3. The Criollo cocoa variety, having a nearly unique and homozygous genotype, was among the first to be cultivated4. Criollo is now one of the two cocoa varieties providing fine flavor chocolate.

However, due to its poor agronomic performance and disease susceptibility, more vigorous hybrids created with foreign (Forastero) genotypes have been introduced. These hybrids, named Trinitario, are now widely cultivated5. Here we report the sequence of a Belizean Criollo plant6.

Consumers have shown an increased interest for high-quality chocolate, and for dark chocolate, containing a higher percentage of cocoa7. Fine-cocoa production is nevertheless estimated to be less than 5% of the world cocoa production due to the low productivity and disease susceptibility of the traditional fine-flavor cocoa varieties. Therefore, breeding of improved Criollo varieties is important for sustainable production of fine-flavor cocoa.

3.7 million tons of cocoa are produced annually (see URLs). However, fungal, oomycete and viral diseases, as well as insect pests, are responsible for an estimated 30% of harvest losses (see URLs). Like many other tropical crops, knowledge of T. cacao genetics and genomics is limited. To accelerate progress in cocoa breeding and the understanding of its biochemistry, we sequenced and analyzed the genome of a Belizean Criollo genotype (B97-61/B2). This genotype is suitable for a high-quality genome sequence assembly because it is highly homozygous as a result of the many generations of self-fertilization that occurred during the domestication process.

Results

Sequencing and assembly

We used a genome-wide shotgun strategy incorporating Roche/454, Illumina and Sanger sequencing technologies. The International Cocoa Genome Sequencing consortium (ICGS) produced a total of 17.6 million 454 single reads, 8.8 million 454 paired end reads, 398.0 million Illumina paired end reads and about 88,000 Sanger BAC end reads, corresponding to 26 Gb of raw data (Supplementary Note, Supplementary Tables 1 and 2 and Supplementary Figs. 1 and 2). We used the Roche/454 and Sanger raw data to produce the assembly. This represented ×16.7 coverage of the 430-Mb genome of B97-61/B2, whose size was estimated by flow cytometry (Supplementary Note and Supplementary Table 3).

This assembly, performed with Newbler software (Roche, Inc.), consists of 25,912 contigs and 4,792 scaffolds (Table 1, Supplementary Note and Supplementary Table 4). Eighty percent of the assembly is in 542 scaffolds, and the largest scaffold measures 3.4 Mb. We determined the N50 (the scaffold size above which 50% of the total length of the sequence assembly can be found) to be 473.8 kb. The total length of the assembly was 326.9 Mb, which represents 76% of the estimated genome size of the T. cacao genotype B97-61/B2 (430 Mb). In addition, we used a high coverage of Illumina data (×44 coverage of the genome), which has a different error profile than 454 data, to improve accuracy of the T. cacao genome sequence (Online Methods).

Table 1 Global statistics of the genome assembly and annotation of Theobroma cacao

The resulting assembly appears to cover a very large proportion of the euchromatin of the T. cacao genome. We confirmed the high genome coverage of this assembly by comparing it to the unigene resource (38,737 unigenes assembled from 715,457 expressed sequence tag (EST) sequences from the B97-61/B2 genotype). We recovered 97.8% of the unigene resource in the T. cacao genome assembly.

Using a set of 1,259 molecular markers from a consensus genetic linkage map established from two mapping populations, 94% of the markers (1,192) were unambiguously located on the assembly, allowing us to anchor 67% of the assembled 326 Mb along the ten T. cacao linkage groups. The remaining 33% of the genome assembly was in 2,207 scaffolds. Fifty percent of the assembly could be anchored and oriented (Supplementary Note, Supplementary Table 5 and Supplementary Fig. 3). The average ratio of genetic-to-physical distance was 1 cM per 444 kb in centromeric regions and 1 cM per 146 kb in distal chromosomal regions.

Gene content and repeated sequences annotation

We performed the identification and annotation of transposable elements using a two step approach: the first approach was based on the de novo identification of transposable elements from the assembled scaffolds and the second one was based on the search for transposable elements from the unassembled reads (Supplementary Note).

This de novo search lead to the identification of a total of 67,575 transposable element–related sequences in the assembled cocoa sequences (Table 1). The second step led to the identification of three highly repeated transposable element families from the unassembled reads dataset. The most common transposable element was a long terminal repeat (LTR) retrotransposon that we named Gaucho. It is a Copia-like element 11,297 bp in length that is repeated approximately 1,100 times, based upon its occurrences in the 454 unassembled sequences. Fluorescence in situ hybridization (FISH) analysis revealed that Gaucho is distributed on all chromosome arms but is found mainly in their median region (as opposed to the centromeric and telomeric regions), a classical feature shared by many LTR retrotransposons in plants (Fig. 1). The other two highly repeated families were another Copia-like LTR retrotransposon and a Mu-type transposon.

Figure 1: FISH analysis of T. cacao chromosomes.
figure 1

(a) In situ hybridization of T. cacao chromosomes stained with DAPI (blue) using a ThCen repeat probe (red). (b) In situ hybridization using Gaucho LTR retrotransposon (green) and ThCen repeat (red) probes.

Class I elements were the most abundant, representing 69% of the total transposable elements in the cocoa genome, with a total of 290 Gypsy-like and 159 Copia-like families. In addition, we identified 36 transposons and 1,353 miniature inverted-repeat transposon element (MITE) families (class II) (Supplementary Table 6). The most highly repeated transposon families were Mutator and Vandal (Mu type), with copy numbers of 994 and 1,978, respectively. Transposable elements were particularly abundant in centromeric regions, as illustrated in Figure 2; this feature was already observed in other sequenced genomes8.

Figure 2: T. cacao genome heat map.
figure 2

The ten T. cacao chromosomes harboring 11 chromosome fusions (in black dotted boxes) identified in these genomes are illustrated according to their ancestral chromosomal origin (see paleo-chromosome color code in Fig. 4). Centromeres are marked 'Cent'. For the ten chromosomes, heat maps are provided for the CDS (blue <60%, yellow 60%–90% and red >90%), class I and II transposable elements (blue <80%, yellow >80% and red 100%), ThCen and Gaucho elements (blue <50% of maximum, yellow ≥50% of maximum and red = maximum) and telomeric repeats (blue = 0, yellow <40% and red >40%). Only the elements present in the assembled part of the genome are represented. Therefore, the genome distribution of the repeated sequences represented in this figure could be biased due to the major limitations of de novo sequencing of complex genomes using next-generation sequencing (NGS), which is limited in its ability to assemble highly repeated sequences.

Altogether, the transposable elements identified in both assembled (84 Mb) and un-assembled (20.3 Mb) reads represent about 24% of the T. cacao genome. This value is substantially lower than that for other sequenced genomes of similar size, for example, rice (35%, for 380 Mb)9 and grape (41.4%, for 475 Mb)10. However, sequencing and assembling of highly repeated sequences can be expected to be the major limitation of de novo sequencing of a complex genome using next-generation sequencing. This is particularly true for transposon element families that have undergone very recent amplification, like in the case of Gaucho. Therefore, we conclude that the total contribution of repetitive elements to the whole cocoa genome may be underestimated.

In addition, we identified a tandem repeat sequence (that we named ThCen) from the 454 repeat reads. This tandem repeat is 212 bp long and, when used as probe in a FISH experiment, was found to be located in the centromeres of all cocoa chromosomes (Fig. 1). Tandem repeats and retrotransposons are the major components of plant centromeres11. The copy number of Gaucho and ThCen repeat sequences varied up to 2.5-fold among T. cacao genotypes from various genetic origins. We observed a positive Pearson correlation (r = 0.56) between genome size and ThCen repeat copy number, suggesting a possible contribution of ThCen repeats in genome size variation (Supplementary Note, Supplementary Fig. 4 and Supplementary Table 7).

We annotated the genome sequence using the integrative gene prediction package EUGene12 following specific training for T. cacao (Supplementary Note). Homology searching and functional annotation (Supplementary Fig. 5) led to the identification of 28,798 T. cacao protein-coding genes (Table 1), with an average gene size of 3,346 bp and a mean of 5.03 exons per gene (Supplementary Table 8). Compared to the smaller Arabidopsis thaliana genome, the T. cacao genome has a higher gene number, a similar exon number per gene and a lower mean gene density per 100 kb (Supplementary Table 8). The genes in T. cacao were more abundant in subtelomeric regions (Fig. 2), as previously observed in other sequenced plant genomes13.

The comparison of cocoa, A. thaliana, grape, soybean and poplar proteomes revealed 6,362 clusters of genes (totaling 52,176 genes) distributed among all five eudicot genomes and 682 gene families (totaling 2,053 genes) specific to the cocoa genome (Fig. 3, Supplementary Note and Supplementary Table 9). Most of these 682 gene families encode hypothetical proteins, as supported at the transcript level. The functional analysis of these five proteomes using gene ontology terms revealed a similar pattern among them (Supplementary Note and Supplementary Fig. 6). A specific feature of the T. cacao clusters common to the other species was the relatively high level of metabolic and cellular processes (Supplementary Note and Supplementary Figs. 6 and 7). On the other hand, grape- and cocoa-specific clusters showed the highest unknown-function percentages (Supplementary Note and Supplementary Fig. 7).

Figure 3: Venn diagram showing the distribution of shared gene families among Theobroma cacao, Arabidopsis thaliana, Populus trichocarpa, Glycine max and Vitis vinifera.
figure 3

Numbers in parentheses indicate the number of genes in each cluster. The Venn diagram was created with web tools provided by the Bioinformatics and Systems Biology of Gent (see URLs).

MicroRNAs (miRNAs) are short noncoding RNAs that regulate target genes transcriptionally or post-transcriptionally. Many of them play important roles in development and stress responses14. A total of 83 T. cacao miRNAs from 25 families were computationally predicted based on sequence similarity to known plant miRNAs in miRBase 14 (ref. 15) (Supplementary Note, Supplementary Table 10 and Supplementary Figs. 8 and 9). Ninety-one T. cacao miRNA targets were predicted. Most predicted targets were homologous to known miRNA targets in other plant species, but there was a profound bias toward putative transcription factors compared to the other species (Supplementary Table 11), suggesting that miRNAs are major regulators of gene expression in T. cacao.

Disease resistance–related genes

Fungal and oomycete diseases are a major constraint to world cocoa production, and the search for natural disease resistance is one of the main objectives of all T. cacao breeding programs. Resistance genes (R genes) are divided into 2 classes: NBS-LRR, the nucleotide-binding site leucine-rich repeat class of genes, and RPK, the receptor protein kinase class of genes16.

Within the RPK class, one family of plant-specific transmembrane receptors was shown to also possess LRR-class genes in its extracellular domain and have important roles in defense responses or in plant development17. The LRR-RLK family consists of more than 200 members in A. thaliana and more than 400 members in poplar18. Here we show that the T. cacao genome contains at least 253 LRR-RLK genes orthologous to Arabidopsis LRR-RLK genes (Supplementary Note). Reports indicate that the LRR-RLK family is divided into 19 subfamilies18. In Viridiplantae, some of these subfamilies have expanded dramatically. As in Populus trichocarpa, the LRR-XII subfamily has greatly expanded in the T. cacao genome (Supplementary Note, Supplementary Table 12 and Supplementary Table 13), with approximately 36% of the LRR genes belonging to this subfamily.

The NBS genes, encoding nucleotide-binding site proteins, also play an important role in resistance to pathogens and in the cell cycle19. The NBS gene family is rather abundant in plant genomes, ranging from 0.6% to approximately 2% of the total gene number (Supplementary Note).

We identified a total of 297 non-redundant NBS-encodingorthologous genes in the T. cacao genome (Supplementary Note, Supplementary Table 14 and Supplementary Figs. 10 and 11). Among these genes, one class, characterized by the TIR (encoding the toll interleukin receptor) motif, is markedly underrepresented in the T. cacao genome compared to other eudicot plants, with only 4% of NBS orthologous genes containing TIR motifs, in contrast to grape, poplar, Medicago (20%) or Arabidopsis (65%) (Supplementary Table 14). The TIR motifs have been shown to be present in basal angiosperms and eudicots, but are nearly absent in monocots20. It has been suggested that the TIR-NBS-LRR resistance genes are more ancient than the divergence of angiosperm and gymnosperms and that they have been lost in the cereal genomes21. Their lower level in the cocoa genome compared to other eudicots, and the close relatedness of cocoa to a common eudicot ancestor as shown by paleohistory studies (see below), suggests a divergent evolution of NBS-TIR orthologous cocoa genes from an ancestral locus, leading to a lower expansion of this gene family in T. cacao.

Another gene family that plays a major role in plant defense is the NPR gene family. NPR1 is an Arabidopsis BTB/POZ domain protein that acts as a central mediator of the plant defense signal transduction pathway22. We surveyed the T. cacao genome sequence and found four related T. cacao orthologous genes corresponding to each of the NPR1 subfamilies found in Arabidopsis (Supplementary Note and Supplementary Fig. 12). Recently, we showed that one of these genes (located on chromosome 9) is a functional ortholog of Arabidopsis NPR1 by transgenic complementation23.

We mapped the NBS, LRR-LRK and NPR1 orthologous genes along the T. cacao pseudomolecules (Supplementary Note and Supplementary Fig. 13). They were distributed across the ten chromosomes, with a large number being organized in clusters, as is classically observed for these classes of genes24.

A meta-analysis of quantitative trait loci (QTL) related to disease resistance previously identified in T. cacao was recently done25. We compared the QTL genetic localizations found in this previous study with the distribution of NBS-LRR, LRR-LRK and NPR orthologous cocoa genes (Supplementary Fig. 13). Considering an average confidence interval of about 20 cM for the 76 QTLs identified25, most of the QTLs are located in genome regions containing candidate resistance genes. However, due to the fact that a large number of QTLs and candidate resistance genes are widespread across the genome, many colocalizations may have occurred at random. Therefore, the candidate genes potentially underlying QTLs need to be further studied by functional genomics approaches to confirm their potential roles in disease resistance in cocoa.

Genes potentially involved in cocoa qualities

T. cacao seeds are fermented, dried and then processed into cocoa mass (ground, dehusked seeds), cocoa butter (triacylglycerol storage lipids from cotyledonary endosperm cells) and cocoa powder (defatted mass consisting primarily of cell walls, endosperm storage proteins, starch and proanthocyanins, as well as other flavonoids, aromatic terpenes, theobromine and many other metabolites). In order to characterize the gene families involved in cocoa quality traits, we used a translational approach to survey the T. cacao genome using molecular and biochemical knowledge from Arabidopsis, poplar, grape and other model plant species.

Oils, proteins, starch and various secondary metabolites such as flavonoids, alkaloids and terpenoids comprise the principal molecular components of cocoa affecting flavor and quality. Storage lipids (triacylglycerols) provide carbon and energy reserves for germinating cocoa embryos (Supplementary Fig. 14). T. cacao seed storage lipids (cocoa butter), representing 50% of the dry seed weight, are exceptional in their very high level of stearate (30–37%), which gives cocoa butter its relatively high melting point (34–38 °C)26. The unique fatty acid profile of cocoa butter enhances the olfactory qualities of chocolate and confectionaries and makes it valuable for cosmetic and pharmaceutical products. We discovered a total of 84 orthologous T. cacao genes potentially involved in lipid biosynthesis, which is 13 more than those discovered in Arabidopsis27 (Supplementary Table 15). Consistent with the large amount of storage lipids produced in T. cacao seeds, the genome contained five additional genes encoding acyl-ACP thioesterase fat B (FATB) and three additional genes encoding ketoacyl-ACP synthase, the two key workhorse enzymes leading to the synthesis of triacylglycerols.

Flavonoids are a diverse group of plant secondary metabolites that play many important roles during plant development28. They are involved in plant defense against insects, pathogens and microbes, in absorption of free radicals and ultraviolet light, and in attraction of beneficial symbionts and pollinators. Proanthocyanidins are flavonoid polymers that are present in large amounts in T. cacao seeds (Supplementary Fig. 15). Recent evidence suggests that proanthocyanidins may be beneficial to human health by improving cardiovascular health, providing cancer chemopreventative effects and also through neuroprotective activities29,30. We identified 96 T. cacao genes orthologous to Arabidopsis genes that are involved in the flavonoid biosynthetic pathway (Supplementary Table 16), which is 60 more than are present in Arabidopsis. Of these genes, we evaluated the function of TcANS, TcANR and TcLAR in transgenic Arabidopsis and Tobacco, demonstrating that they are functional orthologs of Arabidopsis genes31. Notably, although Arabidopsis has only one gene encoding dihydroflavonol-4-reductase (DFR), the T. cacao genome contains 18 orthologous DFR genes. DFR catalyzes the reaction that produces the flavan-3,4-diols, the immediate precursors of the flavonoids catechin and epicatachin. These compounds can accumulate to as much as 8% of the dry T. cacao seed, making T. cacao one of the richest known sources of this phytonutrient32.

Terpenoids constitute a large family of natural compounds and play diverse roles in plants as hormones, pigments and in plant-environment interactions and defense. They are major components of resins, essential oils and aromas33. Among them, two subclasses of terpenoids are particularly involved in aromas: monoterpenes (C10), which represent aromatic compounds that are the basis of floral essences and essential oils, and sesquiterpenes (C15), which may also constitute a defense response of plants toward microorganisms or insect aggression. Compared to bulk cocoa, a higher level of monoterpenes (such as linalool, an acyclic monoterpene alcohol found in the floral scent of Clarkia breweri and of many other plants species) has been observed in fine-flavored cocoa varieties like Criollo and Nacional, which are characterized by fruity and floral notes34,35.

We identified 57 T. cacao genes that are orthologs of Arabidopsis genes that encode terpene synthase (TPS), which catalyze terpenoid synthesis33, and nine pseudogenes (Supplementary Table 17 and Supplementary Figs. 16 and 17). This number is higher than in Arabidopsis and poplar36, which have 30 and 40 genes, respectively, and lower than in grape10, which has 89 functional genes. The classification of TPSs in the different subclasses revealed that 34% of them correspond to monoterpenes and 31% correspond to sesquiterpenes. Two gene families are particularly expanded in T. cacao: the linalool synthase family (monoterpenes), which is represented by 7 genes clustered in a region of chromosome 6, and the cadinene synthase family (sesquiterpenes), comprising 10 members, among which 7 are localized in a same region of chromosome 7. Cadinene synthase is one of the key enzymes involved in the synthesis of gossypol, a toxic terpenoid produced in the seeds of cotton, a species belonging to Malvaceae, the same family as T. cacao. In cocoa, cadinene synthase has been found to be expressed in pod tissues37 in response to attacks by mirids (Sahlbergella singularis), a major insect pest of cocoa trees in Africa. Therefore, the cadinene synthase orthologous genes are candidates for elements of the cocoa insect resistance response.

Colocalization of quality related genes and QTL

Previous studies have reported QTLs associated with quality traits such as lipid and flavonoid content38,39 (Supplementary Note). For most of these QTLs, genes encoding key enzymes of these biosynthetic pathways were found to be colocalized with most of these QTLs (Supplementary Fig. 16). For example, a major QTL for fat content is associated on chromosome 9 with a gene orthologous to KCS, encoding beta ketoacyl-CoA synthase, and is located very close to an ortholog of one member of the FATB gene group, which is specifically expanded in T. cacao. A strong QTL for cocoa butter hardness39 was found localized in linkage group 7 near a gene orthologous to FAD4, which is involved in creating a bond between C2 and C3 of the lipid chain, resulting in lipids with a higher melting point, which, in terms of cocoa butter, represents greater hardness.

Similarly, we found each of the QTLs for astringent taste to be closely associated with genes potentially involved in the proanthocyanidins biosynthetic pathway (Supplementary Fig. 15). For example, two genes orthologous to that encoding flavonoid 3-hydroxylase (F3H) and one orthologous to that encoding dihydro-flavonol-4-reductase (DFR), which are specifically expanded in T. cacao, co-localize on linkage group 1 with a major astringency QTL.

T. cacao genome paleo-history

Angiosperms have been shown to evolve through rounds of paleo-polyploidy10,40. Two types of events have been reported in the literature for eudicots: an ancestral event (referenced as γ) and lineage-specific events (referenced as α and β). In order to investigate the paleo-history of the T. cacao genome, we characterized shared paleo-polyploidies based on the integration of orthologous relationships identified between T. cacao and five eudicot sequenced genomes (Arabidopsis41, grape10, poplar36, soybean42 and papaya43), as well as paralogous relationships identified between the ten T. cacao chromosomes.

Recently, we published a method for the identification of orthologous regions between plant genomes as well as for the detection of duplicated blocks within genomes based on integrative sequence alignment criteria combined with statistical validations44. This approach has been recently applied to available monocot and eudicot genomes and has allowed us to propose a common ancestor with five (core gene set of 9,138) and seven proto-chromosomes (core gene set of 9,731) for monocots and eudicots, respectively45. We have integrated the T. cacao genome sequence information (23,529 gene models anchored) into our previous paleo-genomics analysis in order to investigate the T. cacao evolutionary paleo-history (Online Methods).

Using the alignment parameters and statistical tests reported previously44, 7,866 orthologous relationships covering 80% of the T. cacao genome were identified between the T. cacao and the Arabidopsis, poplar, grape, soybean and papaya genomes. The chromosome-to-chromosome orthologous relationships that were established between the T. cacao and the five sequenced eudicot genomes are illustrated in Figure 4a and are available in Supplementary Table 18.

Figure 4: T. cacao genome paleohistory.
figure 4

(a) T. cacao genome synteny. A schematic representation of the orthologs identified between cacao chromosomes (c1 to c10) at the center and the grape (g1 to g19), Arabidopsis (a1 to a5), poplar (p1 to p19), soybean (s1 to s20) and papaya (p1 to p9) chromosomes. Each line represents an orthologous gene. The seven different colors used to represent the blocks reflect the origin from the seven ancestral eudicot linkage groups. (b) T. cacao genome duplication. The seven major triplicated chromosomes groups in T. cacao (c1 to c10) are illustrated (colored blocks) and related with paralogous gene pairs identified between the T. cacao chromosomes (colored lines). The seven different colors reflect the seven ancestral eudicot linkage groups. (c) T. cacao genome evolutionary model updated from Abrouk et al.46. The eudicot chromosomes are represented with a seven-color code to illustrate the evolution of segments from a common ancestor with seven protochromosomes (top). The different lineage-specific shuffling events that have shaped the structure of the six genomes during their evolution from the common paleo-hexaploid ancestor are indicated as R (for rounds of whole-genome duplication (WGD)) and F (for fusions of chromosomes). The current structure of the eudicot genomes is represented at the bottom of the figure.

Moreover, we aligned the 23,529 gene models from the T. cacao genome onto themselves. Seven blocks of duplicated genes (344 gene pairs) were identified and characterized in T. cacao, covering 64% of the genome and involving the following chromosome to chromosome (c) relationships: c2-c3-c4 (yellow), c1-c3 (blue), c1-c2-c8 (green), c6-c9 (light blue), c1-c4-c5 (purple), c5-c9-c10 (orange) and c1-c6-c7 (red) (Fig. 4b and Supplementary Table 19). We found this ancestral paleo-polyploidy event shared at orthologous positions between eudicot genomes on the following chromosome pair combinations in T. cacao compared to the seven ancestral triplicated chromosome groups reported in grape (g)10: g1-g14-g17/c2-c3-c4 (yellow), g2-g12-g15-g16/c1-c3 (blue), g3-g4-g7-g18/c1-c2-c8 (green), g4-g9-g11/c6-c9 (light blue), g5-g7-g14/c1-c4-c5 (purple), g6-g8-g13/c5-c9-c10 (orange), g10-g12-g19/c1-c6-c7 (red) (Fig. 4c). This result confirms the paleo-hexaploid origin of the eudicot species recently reported for the grape10 and soybean42 genomes. Moreover, we confirmed the known γ paleo-hexaploidization event in the T. cacao genome through classical Ks-based (synonymous substitution rate) data analysis between paralogous genes (Supplementary Fig. 18).

Based on the ancestral (γ) and lineage-specific duplications (α, β) reported for eudicots, it became possible to propose an evolutionary scenario that shaped the ten T. cacao chromosomes from the seven chromosomes of the eudicot ancestor and, more precisely, to the 21 chromosomes of the paleo-hexaploid ancestor (Fig. 4c). We suggest, from the 21 chromosomes intermediate ancestor, at least 11 major chromosome fusions (referenced as 'F' in Fig. 4c) to reach the actual ten-chromosome structure (compared to 30, 4, 71, 82 and 56 reported, respectively, for Arabidopsis, grape, poplar, soybean and papaya genomes)46.

Finally, in order to gain insight into our understanding of the molecular mechanisms driving the chromosome number reduction from the 21 chromosome ancestor intermediate to the actual 10 chromosome structure of the T. cacao genome, we produced heat maps scoring particular features such as CoDing Sequences (CDS), transposable element repeats for class I and II, tandem-Gaucho elements and telomeric repeats (Fig. 2). We observed a classical distribution pattern of CDS and transposon elements that were more abundant at the telomeric and centomeric regions of the chromosomes, respectively. Moreover, we identified a clear correlation between the position of chromosome fusion points (dotted rectangles) and the occurrence of telomeric repeats (telomeric remnants collocating with eight out of the eleven sites) consistent with a telomere-telomere recombination process leading to the chromosome fusion events reported previously in eudicots47.

Discussion

We sequenced the genome of T. cacao, resulting in the assembly of 76% of its genome and identification of 28,798 protein-coding genes, among which 23,529 (82%) were anchored onto the ten cocoa chromosomes. A large proportion of the euchromatin of the T. cacao genome is likely covered by this assembly, allowing for the recovery of 97.8% of the T. cacao unigene resource. We found that 682 gene families are specific to T. cacao, as compared to A. thaliana, grape, soybean and poplar proteomes. Only 24% of the T. cacao genome consists of transposable elements, a lower percentage than in other genomes of similar size. The analysis of specific gene families that are potentially linked to cocoa qualities and disease resistance showed that particular expansion or reduction of some gene families appears to have occurred. The mapping of these gene families along the cocoa chromosomes and comparison with the genome regions involved in trait variation (QTLs) constitutes an invaluable source of candidate genes for further functional studies that aim to discover the specific genes directly involved in trait variation. This draft genome sequence will facilitate a better understanding of trait variation and will accelerate the genetic improvement of T. cacao through efficient marker-assisted selection and exploitation of genetic resources. Using an updated version of the Newbler software (released by Roche on 8/17/2010), we performed a second-generation assembly of the cacao genome data (ICGS Assembly 1.2). The new assembly covers 84.3% of the T. cacao B97-61/B2 genome, with a N50 scaffold size of 5.624 Mb and the largest scaffold of 18.20 Mb. This enhanced assembly enabled us to improve its' anchorage onto the genetic map, which now includes approximately 87% of the assembled sequences on ten pseudochromosomes. Additional details of this and further assembly improvements will be available on the ICGS website (see URLs).

This study has highlighted the close evolutionary relationship of the T. cacao genome to the eudicot putative ancestor, showing a limited number of recombinations between ancestral chromosomes, as has also been observed in grape10. T. cacao, which has only ten pairs of chromosomes, is easily propagated by both sexual and vegetative methods and can be transformed48, and therefore, it represents a new and simple model to study the evolutionary processes, gene function, genetics and biochemistry of tree fruit crops.

URLs.

Cocoa statistics, http://www.icco.org/economics/market.aspx, http://www.dropdata.org/cocoa/cocoa_prob.htm; ICGS website, http://cocoagendb.cirad.fr/gbrowse/cgi-bin/gbrowse/theobroma/; web tools provided by the Bioinformatics and Systems Biology of Gent, http://bioinformatics.psb.ugent.be/webtools/Venn/; MUST, http://csbl1.bmb.uga.edu/ffzhou/MUST/.

Methods

High molecular weight DNA preparation.

High molecular weight DNA was prepared from nuclei of B97-61/B2 cocoa leaves according to previously described protocols49, except that the steps of filtration were replaced by five successive filtrations with nylon filters (SEFAR NITEX) having a decreasing mesh diameter: 250 μM, 100 μM, 50 μM and two times 11 μM. (Supplementary Note).

Construction of BAC libraries.

Two BAC libraries were constructed from cocoa leaves. DNA was isolated from nuclei collected in agarose plugs and DNA digestions were performed with HindIII or EcoRI, followed by ligation to the pAGIBAC1 vector (a modified pIndigoBAC536Blue with an additional SwaI site49). Ligation products were transformed into DH10B T1 phage-resistant Escherichia coli cells (Invitrogen) and plated on Lysogeny broth agar that contained chloramphenicol (12.5 μg ml−1), X-gal (20 mg ml−1) and Isopropyl β-D-1-thiogalactopyranoside (0.1 M). For characteristics, quality assessment and estimated genome coverage see the Supplementary Note and Supplementary Table 1.

Genome sequencing.

The genome was sequenced using a genome-wide shotgun strategy. All data were generated using next-generation sequencers: Roche/454 GSFLX (×16.5 coverage) and Illumina GAIIx (×44 coverage), except for data from BAC ends (×0.2 coverage), which were produced by paired-end sequencing of cloned inserts using Sanger technology on ABI3730xl sequencers (Supplementary Note and Supplementary Table 2).

Genome assembly and automatic error corrections with Solexa/Illumina reads.

Sanger and 454 reads were assembled with Newbler version 2.3. From the initial 26,519,827 reads, 80.65% (21,387,691) were assembled by Newbler. The 454 assembly was improved by automatic error corrections with Solexa/Illumina reads, which have a different bias in error type, as described previously50. Short-read sequences were aligned on the cocoa genome assembly using the SOAP software (with a seed size of 12 bp and a maximum gap size allowed on a read of 3 bp). Only uniquely mapped reads were retained (Supplementary Note).

Estimation of nuclear DNA content by flow cytometry.

The genome sizes of B97-61/B2 and a panel of diverse cocoa clones were estimated by flow cytometry51. Leaves of studied samples and internal standards were chopped with a razor blade and then stained using propidium iodide. DNA content of 5,000–10,000 stained nuclei was determined using a CyFlow SL3 flow cytometer with a 532-nm green solid state laser (100 mWatt). The monoploid C-value (1C) was calculated and expressed in Mb using the conversion factor 1 pg DNA = 978 Mb (Supplementary Note).

Anchoring of the assembly in the genetic map.

A consensus map was established from two progenies: an F1 progeny of 256 individuals (UPA 402 × UF676) and an F2 progeny of 136 individuals recently produced (Scavina 6 × ICS1). The F1 progeny was previously used to established the cocoa reference map, which includes 600 markers52,53. New SSR and SNP markers were mapped in these two progenies, and a consensus map including 1,259 markers was established54. BLAT software was used to align the markers of the genetic map with the scaffolds (Supplementary Note).

Prediction of transposable elements.

The annotation of transposable elements in the cocoa genome was achieved in two stages. First, a combination of de novo analyses (for example, LTR_finder, LTRharvest, MUST (see URLs)) and extrinsic comparisons (BLAST) was conducted. Then, a de novo approach was used to construct highly repeated elements from the 3,220,522 unassembled reads. A total of 67,575 transposable elements were annotated (Supplementary Note).

In situ hybridization of transposable element probes.

FISH was performed on mitotic metaphase spreads prepared from meristem root tip cells. The probes were labeled with Alexa-488 dUTP and Alexa-594 dUTP by random priming (Fisher Bioblock Scientific) and the in situ hybridizations were performed according to D'Hont et al.55 (Supplementary Note).

Prediction of protein-coding genes.

Gene structures were predicted using EUGene12. Translation start sites and RNA splice sites were predicted by SpliceMachine56. Available T. cacao ESTs were aligned onto the scaffolds using GenomeThreader57. Similarities to proteins from Swiss-Prot, TAIR, Malvacae GenBank extraction, Glycine max high confidence gene models58 and translated T. cacao EST contigs were searched using BLASTX. Similarities to A. thaliana, Gossypium, V. vinifera, Citrus and T. cacao ESTs were searched using TBLASTX. A total of 50,582 genes were predicted, giving a final count of 28,798 T. cacao genes after filtering (Supplementary Note).

Gene family analysis.

Protein domains were searched using InterProscan against the InterPro database. Cocoa, Arabidopsis, grape, poplar and soybean Best Blast Mutual Hit (BBMH) were computed, and protein clustering was done using OrthoMCL59.

Synteny and duplication analysis.

Arabidopsis, grape, poplar, soybean and papaya proteomes were aligned using an approach based on BLASTP in order to identify accurate paralogous and orthologous relationships44,45. KS divergence (million year ago (MYA) scale) for paralogous and orthologous gene pairs as well as speciation events dating were calculated with PAML60.

Accession codes.

The Theobroma cacao whole-genome sequences are deposited in the EMBL, GenBank and DDBJ databases under accession numbers CACC01000001–CACC01025912. A genome browser and further information on the project are available from http://cocoagendb.cirad.fr/gbrowse and http://cocoagendb.cirad.fr.