Chromosomal-scale and haplotype-resolved genome assembly of the first Trinitario hybrid cacao ICS 1

Abstract

Cocoa is a vital agricultural commodity that yields cocoa butter and powder, both essential to the confectionery industry. The Trinitario cacao varieties are key sources of fine flavor beans, among which ICS 1 stands out due to its high productivity and superior bean quality. ICS 1 is also recognized as a valuable parent in breeding programs for desirable agronomic traits such as fine flavor bean quality. Using sequencing reads from PacBio HiFi and Hi-C technologies, we generated two haploid genome assemblies (374.4 Mb and 410.8 Mb), characterized by low scaffold numbers, high N50 values (39.5 Mb and 39.7 Mb), and 10 pseudo-chromosomes. Genome annotation identified 22,477 and 22,263 protein-coding genes, with repeat content of 62.19% and 65.14%, respectively. BUSCO completeness exceeded 98%, confirming high assembly quality. Cacao breeders will benefit from these haploid genomes to develop high-yielding, climate-resilient, fine flavor cacao varieties, addressing challenges such as declining soil fertility, rising disease pressures, and accelerating climate change.

Background & Summary

Cacao (Theobroma cacao L.) is an economically significant tree native to the lowland tropical regions of the Americas, particularly within the Amazon Basin. This diploid (2n = 2x = 20) evergreen species belongs to the mallow family (Malvaceae)¹. Traditionally, cacao germplasm has been classified into three primary groups based on geographical origin and morphological characteristics: Criollo, Forastero, and Trinitario². The Criollo landraces, cultivated for millennia, are highly valued for their fine flavor, while Forastero cacao—likely introduced from Brazil—is prized for its robustness and higher yield^2,3. The natural hybrid, Trinitario, originated in Trinidad through natural cross-pollination, combines Criollo’s desirable flavor and Forastero’s adaptability. Modern genetic studies, however, have expanded this classification, identifying more than ten distinct genetic groups, further underscoring the extensive genetic diversity within the species^4–7.

Trinitario cacao is the key source of raw material for the high-end chocolate industry. The global cocoa market distinguishes between two broad categories of cocoa beans: “fine-flavor” cocoa and “bulk” or “ordinary” cocoa⁸. Fine-flavor cocoa exhibits a complex sensory profile characterized by well-balanced attributes with aromatic and nuanced flavor notes. It is primarily produced from the Criollo, Trinitario and Nacional genetic groups⁹, representing approximately 12% of global cacao production⁸. In contrast, bulk cocoa—known for its basic chocolate flavor—originates mainly from the Forastero group⁸. However, the classification between fine flavor and bulk cocoa is not solely determined by their genetic or germplasm groups. Several cacao varieties from the Upper Amazon, such as Chuncho, Piura Porcenala, and Nacional Boliviano are fine flavor cocoa and sold with higher prices in the international market¹⁰. This underscores the importance of studying the diverse sensory attributes of fine flavor cocoa to understand the wide range of unique flavors in fine flavor cocoa beans¹¹.

Global demand for fine-flavor cacao beans is increasing, driven by consumer preference for high-quality chocolate with distinctive flavor profiles. In the international cocoa market, fine-flavor cocoa usually commands a moderate price premium over bulk cocoa. Prices for flavor cocoa generally begin at a 20% premium, rising to up to treble the price per metric ton¹². However, this price differential remains relatively small, providing insufficient incentive for cacao farmers to cultivate traditional fine-flavor varieties, such as Criollo and Trinitario. These varieties tend to be less vigorous, less productive, and more susceptible to biotic and abiotic stresses compared to modern hybrid varieties^13,14. As a result, traditional varieties are being replaced by more productive modern varieties, which offer higher yields but usually produce beans classified as bulk cocoa¹⁰.

To address these challenges, cacao breeders are working to develop new generation of varieties that combine self-compatibility, high productivity, disease resistance, and superior flavor quality¹⁵. This goal can be achieved through the strategic crossbreeding of traditional fine-flavor varieties with resilient germplasm that exhibits enhanced yield and stress resistance¹⁶. Notably, the fine-flavor variety ICS 1 has played a crucial role in introgressing bean quality attributes and self-compatibility into modern hybrid populations. According to the International Cocoa Germplasm Database (ICGD), ICS 1 has been used as a parent clone in over 70 hybrid combinations in Asia, Africa and the Americas (https://research.reading.ac.uk/cocoa/icgd/).

Since 2011, various cacao reference genomes have been published, representing Criollo, Amelonado, and different wild populations in Upper Amazon^7,17–20. The advancement of cacao reference genomes has provided valuable tools for understanding cacao genetic diversity and improving breeding strategies aimed at enhancing yield, stress tolerance, and quality traits^21–25. However, a reference genome specifically for Trinitario cacao—the largest hybrid cacao group known for its fine-flavor beans—has not yet been developed. High-quality Trinitario cacao genomes will significantly accelerate breeding programs, offering precise insights into its genetic architecture, and enabling more efficient genomic selection, prediction, and identification of genes responsible for desirable traits, particularly those related to cacao bean quality.

In addition, cacao also represents a powerful model organism for investigating plant evolutionary processes²⁶. New cacao genomes will help elucidate the evolutionary dynamics shaped by long-term climatic oscillations and domestication bottlenecks⁴. For example, high-quality nuclear genome resources have revealed deep population stratification among major cacao genetic groups, reflecting complex histories of divergence, introgression, and hybridization¹⁹. Cacao genomic resources have also enabled the study of how selective pressures imposed by pathogens and ecological changes have driven the diversification of gene families involved in defense, metabolism, and reproductive biology^7,21. Thus, the Trinitario genomes presented in this study will contribute to a better understanding of genome evolution, speciation, and adaptation of the long-lived cacao plants.

In this study, we combined PacBio HiFi sequencing data (24 Gb) and Hi-C data (18.6 Gb) to generate high-quality, chromosome-level, haplotype-resolved genome assemblies of ICS 1, one of the most significant Trinitario cacao varieties (Fig. 1). The two haplotype-resolved assemblies are 374.4 Mb and 410.8 Mb in size, with high assembly quality, as indicated by low scaffold counts (312 and 302) and large scaffold N50 values (39.5–39.7 Mb). The lengths of the 10 pseudo-chromosomes of ICS 1 closely match estimated consensus diploid genome size (351.8 Mb) based on a k-mer frequency analysis. However, publicly available estimates of genome size of different cacao cultivars, based on flow cytometry analysis, range from 415 Mb to 445 Mb^18,27,28, suggesting that different cacao varieties may exhibit a wider range of genome sizes.

Fig. 1.

The pods and seeds of the ICS 1 cacao cultivar.

The image of pods and seeds of ICS 1 were sourced from USDA-ARS Tropical Agriculture Research Station, Mayaguez, Puerto Rico, 2010. The image was obtained from International Cocoa Germplasm Database (ICGD) https://www.icgd.reading.ac.uk/photo.php?nacode=5186&&id=1295 with permission.

Genome annotation reveals that the two haplotype assemblies share similar characteristics, including the number of protein-coding genes (22,477 and 22,263), average gene length (3,310 bp and 3,485 bp), average exon length (250 bp and 254 bp), and average intron length (407 bp and 431 bp). Additionally, repetitive element content is similar in both haplotype assemblies, at 62.19% and 65.14%, respectively. BUSCO v5.3.2²⁹ analysis further confirms the high completeness of these assemblies, with scores exceeding 98%, using the Viridiplantae, Eukaryota and Embryophyte databases.

In summary, we successfully developed chromosome-level, haplotype-resolved assemblies for the first Trinitario cacao genomes, providing a valuable resource for cacao researchers and breeders. This work is poised to facilitate the development of new cacao varieties that are both high-yielding and of fine flavor, while simultaneously addressing sustainability challenges such as biotic and abiotic stresses and the broader impacts of climate change in cacao-producing regions.

Materials and Methods

Cacao materials

The T. cacao clone ICS 1 is a hybrid of Criollo and Forastero variety and the clone was obtained from USDA-ARS Tropical Agriculture Research Station, Mayaguez, Puerto Rico. This clone was originally introduced by USDA as budwood from the Imperial College of Tropical Agriculture (ICTA) in Trinidad in 1956. ICS 1 is maintained in the greenhouse of USDA-ARS Sustainable Perennial Crops Laboratory (SPCL), Beltsville, Maryland (Fig. 1).

DNA extraction and sequencing

Cacao leaves were collected for genomic DNA extraction. DNA fragments averaging 20–30 kb was selected and sequenced using PacBio Sequel II 8 M SMRT cells. A total of 24 Gb HiFi reads were generated (Table 1).

Table 1.

Summary of the sequencing data of the ICS 1 cultivar.

Cacao cultivar	ICS 1
PacBio HiFi (Gb)	24
HiFi average length (Kb)	12
Hi-C (Gb)	18.6
k-mer estimated genome size (Mb)	351.8
Haploid type	Haploid 1	Haploid 2
Assembled genome size (Mb)	374.4	410.8
HiFi reads mapping rate	99.98	99.98
Sequencing coverage (X)	64.1	58.4

Additionally, genomic DNA was sequenced using the Omni-C technique from Dovetail Genomics to capture chromatin conformation. The protocol followed the Omni-C kit guidelines for plant samples. Briefly, approximately 300 mg of frozen cacao leaves were ground into a fine powder in liquid nitrogen. The powdered sample was then fixed with 37% formaldehyde for 10 minutes at room temperature. After centrifugation, the supernatant was discarded, and the resulting pellet was washed twice with 1X Wash Buffer. The washed pellet was then filtered with a syringe and digested using DNase I (Nuclease Enzyme Mix). Following DNA quality control (QC), the proximity ligation step was performed through the following steps: binding chromatin to Chromatin Capture Beads, end polishing with biotinylated adapters, bridge ligation with bridge ligase, and crosslink reversal. After proximity ligation, the DNA was purified using SPRI select beads. The purified DNA was used for library preparation without fragmentation. It underwent end repair and adapter ligation (for Illumina), followed by further DNA purification and ligation capture with Streptavidin beads. The samples then underwent Index PCR and DNA size selection, targeting a range of 350 bp to 1,000 bp. The final DNA library was sequenced on an Illumina HiSeqX system, and only reads with a mapping quality (MQ) score greater than 50 were used for scaffolding. The Hi-C sequencing produced 18.6 Gb raw reads (Table 1).

Genome size estimation

To estimate the ICS 1 genome size, a k-mer frequency analysis was performed. The PacBio HiFi reads were first mapped using minimap2³⁰ to the reference plastome of ICS 1 (JQ228381.1)³¹ and the reference mitogenome of Nanay POUND 7 (NC_066894.1)³². The unmapped reads were collected as nuclear reads and were used as input to KmerGenie³³ in diploid mode. The histogram of the best k-mer size was subsequently visualized by using GenomeScope2 in diploid mode³⁴. The k-mer frequency analysis estimates the consensus diploid genome size for ICS 1 is ~351.8 Mb (Table 1, Fig. 2).

Fig. 2.

ICS 1 consensus diploid genome size estimation using k-mer frequency distribution. The x-axis shows k-mer coverage depth, and y-axis represents frequency at each coverage level. ICS 1 exhibits a higher first peak than the second peak, corresponding to a heterozygosity (0.485%) of the diploid genome.

Genome assembly and scaffolding

The PacBio HiFi reads were assembled using the hifiasm software v0.20.0-r639³⁵, which produced a pair of haplotype-resolved assemblies. Possible contamination was assessed using BLASTn searching against the NCBI nt database that is implemented in BlobTools2 v1.1.134³⁶. The cleaned two haploid genome assemblies were scaffolded with Dovetail Omni-C reads using HiRise v2.0.5³⁷. Briefly, Omni-C reads were aligned to the draft assemblies using BWA v0.7.11³⁸, and HiRise analyzed the mapped reads to generate a likelihood model for genomic distance, enabling the identification of mis-joins and the scaffolding of genomic contigs.

Pseudo-chromosome assignment and assessment

The scaffolded ICS 1 genome assemblies were compared to the reference plastome (JQ228381.1) and the reference mitogenome (NC_066894.1) using BLASTn to identify chloroplast and mitochondrial scaffolds. The remaining ICS 1 nuclear genome assemblies were compared to the reference nuclear genome of Cacao Matina 1–6 with MUMmer v4.0.0rc1³⁹. This allowed us to re-orient and assign a number for each ICS 1 pseudo-chromosomes according to the Matina 1–6 reference pseudo-chromosomes. The final ICS 1 haplotype-resolved genome assemblies show a size of 374.4 Mb and 410.8 Mb, respectively (Table 1). The sizes of the largest 10 pseudo-chromosomes (Fig. 3), 343.8 Mb and 358.7 Mb, for the two haploids respectively, closely matched the consensus diploid genome size estimated from k-mer frequency analysis, 351.8 Mb (Fig. 2). The two ICS 1 haploid genome assemblies have lower scaffold counts (312 and 302) and higher N50 values (39.5 Mb and 39.7 Mb) than the two cacao reference genomes (Criollo and Matina1-6), and comparable qualities to the three recently published wild cacao genomes (Contamana, Iquitos, Nanay)^17–20 (Table 2).

Fig. 3.

The linkage density heatmaps of Omni-C reads against the 10 pseudo-chromosomes of the two ICS 1 haploid genomes. (A) Omni-C reads mapped to the hap1 genome using Juicer v1.6 with default parameters. (B) Omni-C reads mapped to the hap2 genome using Juicer v1.6 with default parameters. (C) Omni-C reads mapped to the combined genomes of hap1 and hap2 using Juicer v1.6 with mapping quality ≥ 0 and including duplicated reads. (D) Omni-C reads mapped to the combined genomes of hap1 and hap2 using Juicer v1.6 with mapping quality ≥ 1 and excluding duplicated reads.

Table 2.

Comparison of the quality of the two ICS 1 haploid genomes assemblies with five published cacao genomes.

Genomes *	ICS 1 hap1	ICS 1 hap2	Criollo v2.0	Matina1-6	Contamana SCA6	Iquitos IMC67	Nanay Pound7
HiFi (Gb)	24		—	—	21.4	19.1	20.4
Length of Genome (bp)	374,435,859	410,791,423	324,879,930	345,993,675	383,477,156	381,649,652	419,275,778
Number of scaffolds	312	302	431	711	175	170	1,895
Length of 10 Chromosomes (bp)	343,785,016	358,679,425	314,189,522	330,456,197	352,058,514	351,354,296	338,407,810
GC (%)	34.18	33.74	34.09	34.13	33.8	33.82	34.9
N50 (bp)	39,518,752	39,748,545	36,364,294	35,454,299	39,464,999	39,489,205	34,430,447
L50	5	5	5	5	5	5	6
N90 (bp)	22,419,816	2,705,499	21,614,486	25,578,749	24,455,724	23,705,257	41,995
L90	10	13	9	9	10	10	580
k-mercompleteness (%)	83.7	84.3	92.0	74.5	86.4	82.9	87.6
k-merconsensus QV	64	63.4	38.9	32.4	61	61.6	53
BUSCO genome **	98.6	98.5	98.5	98.4	98.5	98.6	98
BUSCO proteins **	98.7	97.6	99.7	98.4	91.7	96.1	95.4

^* Genome assembly accession numbers: ICS 1 hap1 (GCA_050656495.1)⁶¹, ICS 1 hap2 (GCA_050656505.1)⁶², Criollo v2.0 (GCA_000208745.1)²⁷, Amelonado Matina1-6 (GCF_000403535.1)¹⁸, Contamana SCA6 (GCA_958329045.1)²⁰, Iquitos IMC67 (GCA_958328385.1)²⁰, Nanay Pound7 (GCA_958329735.1)²⁰. ** BUSCO completeness assessment was conducted on the genome and the protein sequences, separately, using the embryophyta_odb10 database.

To assess the scaffolding quality of the 10 pseudo-chromosomes in the two ICS 1 haploid genome assemblies, Omni-C reads were first mapped to hap1 and hap2 individually using Juicer v1.6⁴⁰ with default parameters, and the resulting contact density heatmaps were visualized using Juicebox⁴¹. From the linkage heatmaps, strong diagonal contact signals were observed within each chromosome (Fig. 3AB). Then, Omni-C reads were mapped to the combined genomes of hap1 and hap2 using the Juicer pre command with two mapping quality thresholds⁴²: (1) mapping quality ≥ 0 and including duplicated reads (pre -n -q 0), and (2) mapping quality ≥ 1 and excluding duplicated reads (pre -n -q 1). When allowing duplicated reads with lower mapping quality threshold, the diagonal contact signals were still the strongest, while strong contact signals were also observed between homologous chromosomes (Fig. 3C). This is expected as highly similar reads were mapped to both homologous chromosomes with mapping quality ≥ 0. After excluding the duplicated reads using mapping quality ≥ 1, only strong diagonal contact signals remained (Fig. 3D). This suggests the two haploid genomes have a high scaffolding quality for their 10 pseudo-chromosomes with no switch errors between homologous chromosomes.

Scaffolds unanchored in the pseudo-chromosomes

In addition to the 10 pseudo-chromosomes, about 30 Mb and 52 Mb scaffolds (Table 2) were not placed in the 10 pseudo-chromosomes of the two haploid genomes, respectively. Aligning these unanchored scaffolds against the 10 pseudo-chromosomes found that most unanchored scaffolds (hap1: 62%, hap2: 58%) aligned to chromosome-anchored regions with >90% sequence identity and >30% sequence coverage. This suggests that many of the unanchored scaffolds represent assembly redundancy due to highly similar sequence repeats.

We have further mapped the HiFi reads to all scaffolds of the two haploid genomes using minimap2 using default parameters. The overall read mapping rate was 99.98% for both genomes (Table 1). When only considering HiFi reads mapped to the 10 pseudo-chromosomes, the mapping rates were 98.99% for hap1 and 99.11% for hap2. This suggests that only a small fraction of HiFi reads contributed to the scaffolds unanchored in the 10 pseudo-chromosomes.

We plotted the HiFi read depth/coverage distribution across the 10 pseudo-chromosomes and the unanchored scaffolds (Fig. 4). The average read depths were 64.6 × and 62.6 × for the 10 pseudo-chromosomes of hap1 and hap2, respectively. The average read depths were 55.1 × and 28.7 × for the unanchored scaffolds of hap1 and hap2, respectively. The read depth distribution of the 10 pseudo-chromosomes is more uniform than that of the unanchored scaffolds. As expected, the centromere regions of the 10 pseudo-chromosomes have less uniform read depths due to a higher content of sequence repeats.

Fig. 4.

Read depth distributions across the 10 pseudo-chromosomes and the unanchored scaffolds of the hap1 and hap2 genomes. A 50-kb window size and a 25-kb step size were used in the plots.

Genome annotation

Protein coding gene models in the two ICS 1 haploid genomes were predicted by the MAKER-P pipeline⁴³. It integrates multiple gene prediction approaches, combining ab initio approach, homology-based and transcripts-based evidence. ICS 1 RNA-seq data was obtained from BioProject PRJNA785999 with the Sequence Read Archive (SRA) accession SRR17587553⁴⁴. RNA-seq reads were trimmed using Trim_galore v0.6.844⁴⁵. These trimmed RNA-seq reads were mapped to the genomes using STAR v2.7.2a⁴⁶ with parameter–alignIntronMax 100000 and assembled into transcripts by Trinity v2.8.5⁴⁷ with parameters–genome_guided_max_intron 5000. The trimmed RNA-seq reads were also de novo assembled by Trinity v2.8.5⁴⁷ for the transcripts-based evidence. The PASA v2.5.3 pipeline with–ALIGNERS gmap,blat,minimap2 (https://github.com/PASApipeline/PASApipeline)⁴⁸ was run to take the RNA-seq reads and transcripts for predicting gene structures. In addition, proteins from well-annotated Criollo v2²⁷, Matina 1-6 v2¹⁸, Arabidopsis thaliana (TAIR10)⁴⁹, Oryza sativa v7⁵⁰, Glycine max v7⁵¹, Amborella trichopoda v6.1 (CoGe genome ID 50948)⁵², and Ginkgo biloba (Ver. 2021)⁵³ were used for the homology-based evidence. To enhance the confidence of protein coding gene model predictions, MAKER-P was run for three iterations with the transcript evidence and homology evidence. Finally, all predicted gene models were integrated to run EVM (EvidenceModeler, v2.1.0)⁵⁴ with–segmentSize 100000–overlapSize 10000 to obtain a final gene model prediction.

Before gene prediction, repetitive elements were identified using RepeatMasker v.4.0.9⁵⁵ with a custom repeat library generated by RepeatModeller⁵⁶. The masked genome was used for the following genome annotation. Noncoding RNA and transfer RNA (tRNA) were identified using Rfam⁵⁷ with infernal and tRNA-Scan2⁵⁸.

As the result, the two ICS 1 haploid genomes have very similar protein coding-gene counts, 22,477 and 22,263, respectively (Table 3). They also have similar average gene lengths, average CDS lengths, average exon lengths, average intron lengths, average exons per gene (Table 3).

Table 3.

Genome annotation of the two ICS 1 haploid genomes.

Haploid type	Haploid 1	Haploid 2
Protein-coding gene count	22,477	22,263
Average gene length (bp)	3,310	3,485
Average CDS length (bp)	1,126	1,170
Average exon length (bp)	250	254
Average intron length (bp)	407	431
Average exon per gene	5.7	5.7
Average intron per gene	4.7	4.7
Number of overlapping genes	525	610
Number of genes without intron	2,315	1,842

Genome synteny between the two haploid genomes and against representative cacao genomes

To compare the genome synteny between the two ICS 1 haploid genomes, WGDI v0.75 (wgdi -icl with BLASTP E-value < 1e-5 and score > 100)⁵⁹ was employed to identify syntenic blocks between the two genomes. SOI v1.2.3 was then used to filter the syntenic blocks with the Orthology Index (OI) > 0.6 for orthologous syntenic blocks, which were then plotted with the soi dotplot command. The dotplot revealed a high degree of orthologous gene synteny between the two homologous chromosomes (Fig. 5A). No inter-chromosomal syntenic blocks were found with the OI > 0.6 threshold, suggesting no inter-chromosomal rearrangements.

Fig. 5.

Genome synteny plots between the two ICS 1 haploid genomes. (A) Orthologous syntenic blocks identified by SOI and presented as the dotplot. (B) Structural rearrangements identified by SyRI.

To identity the structural rearrangements between homologous chromosomes of the two haploid genomes, SyRI v1.7.0 (with the options -k -F S)⁶⁰ was used to detect inversions, translocations, and duplications with minimap2³⁰ for the haploid genome alignment (Fig. 5B). Between the two haploid genomes, homologous chromosomes are generally very conserved in chromosomes 1, 2, 5, and 10. Notably, chromosome 6 has a ~2.1 Mb region being inverted between the two haploid genomes. Large unaligned regions exist in chromosomes 3, 7, 8, and 9. These unaligned regions contain gaps inserted between scaffolds and are enriched with repeat sequences.

In addition, we have further aligned the two ICS 1 haploid genomes using MUMmer v4.0.0rc1³⁹ nucmer command at the nucleotide level, with a minimum sequence identity threshold of 90% (-i 90) and a minimum alignment length of 20 kb (-l 20000). We also aligned the hap1 genome against three previously published cacao genomes (Criollo, Matina1-6, Contamana SCA6). The dot plots (Fig. 6) revealed strong genome-wide syntenies between the hap1 genome and these compared genomes, supporting the high quality of the two ICS 1 haploid genomes.

Fig. 6.

Dot plots of genomic synteny between the ICS 1 hap1 genome and other representative cacao genomes. Synteny plots were generated using the MUMmer nucmer command, with a minimum sequence identity of 90% (-i 90) and a minimum alignment length of 20 kb (-l 20000).

Repeat annotation

Repetitive elements in the genomes were first identified using RepeatModeler v2.0.1⁵⁶, which integrates RepeatScout, RECON, LTRharvest and LTR_retriever tools. The resulting repeat libraries were further processed by RepeatMasker v4.0.9⁵⁵ for repeat annotation. The two haploid genomes exhibited similar repeat content, 62.19% and 65.14%, respectively. No SINEs (Short Interspersed Nuclear Elements) were found in the two genomes. Most classified repeat elements were annotated as LTRs (Long Terminal Repeats), while nearly half of the repeats were unclassified/unannotated (Table 4).

Table 4.

Repeat elements in the two ICS 1 haploid genomes.

Haploid type	Haploid 1	Haploid 2
SINEs (%)	0	0
LINEs (%)	1.05	1.33
LTR elements (%)	24.16	21.91
DNA elements (%)	3.69	1.97
Small RNA (%)	1.56	1.26
Satellites (%)	0	0.06
Simple repeats (%)	0.76	0.69
Low complexity (%)	0.21	0.19
Unclassified (%)	30.75	37.33
Total repeat (%)	62.19	65.14

Data Records

The ICS 1 cacao assembled genomes, along with the raw Pacbio HiFi reads and Dovetail Hi-C reads are available in NCBI through the Umbrella BioProject PRJNA1136499 with BioSample SAMN48090412. The genome assemblies are available in GenBank with haplotype 1 assembly (PRJNA1254434; JBNHNC000000000)⁶¹ and haplotype 2 assembly (PRJNA1254437; JBNHND000000000)⁶². The Pacbio HiFi sequence reads are available through SRA SRR33364810⁶³. The Dovetail Hi-C short reads are available through SRA SRR33323860⁶⁴. The gene annotations (protein sequence and GFF files) are available via Figshare⁶⁵.

Technical Validation

To validate the quality of the ICS 1 genome assemblies, the raw PacBio HiFi reads were mapped to the genomes, revealing a high mapping rate over 99% (Table 1). Additionally, the sizes of largest 10 pseudo-chromosomes closely matched the consensus haploid genome size estimates obtained from k-mer frequency analysis (Fig. 2). Moreover, the Omni-C read linkage density histogram of the ICS 1 genomes confirmed the presence of 10 pseudo-chromosomes (Fig. 3), consistent with the chromosome numbers of other published cacao cultivars²⁰. The quality of the ICS 1 genome assemblies was further accessed using Quast analysis⁶⁶, which showed that the two haploid genome assemblies exhibited a higher N50 and N90, lower L50 and L90 at both contig and scaffold levels (Table 2).

To further assess the quality and completeness of the ICS 1 assemblies, a k-mer–based evaluation was performed using KAT v2.4.2⁶⁷ and Merqury v1.3⁶⁸. K-mers shared between the PacBio HiFi reads and the genome assemblies indicated a high k-mer completeness for hap1 and hap2, at 83.7% and 84.3%, respectively (Table 2). Assembly consensus accuracy was further assessed by estimating the consensus quality value (QV) from unique k-mers, yielding QV scores of 64.0 and 63.7 for hap1 and hap2, respectively (Table 2). The genome completeness was evaluated using BUSCO (Benchmarking Universal Single-Copy Ortholog) v5.3.2²⁹ with the embryophyta_odb10 database. The hap1 and hap2 genome assemblies recovered 98.6% and 98.5% of the 1,614 conserved single-copy gene orthologs, respectively. Similarly, the annotated protein sets showed high BUSCO completeness at 98.7% and 97.6% for hap1 and hap2 genomes, respectively. These metrics (Table 2) indicates a high-quality assembly of the two ICS 1 haploid genomes.

Author contributions

Y.Y., D.Z., L.W.M. conceived and designed the project. D.Z., B.B., S.P., S.P.C., L.W.M. collected the Cacao materials and generated the sequencing data. X.F., Y.Yan, R.S.K.R.P performed genome assembly evaluation, annotation, and all data analysis, V.A. contributed to the data analysis and management. X.F., D.Z., Y.Y. draft the manuscript. All authors contributed and approved the final manuscript.

Data availability

All genome sequencing data are available in NCBI through the Umbrella BioProject PRJNA1136499. The gene annotations (protein sequence and GFF files) are available via Figshare⁶⁵.

Code availability

No original code was developed; all computational analyses were performed with published computer tools.

Competing interests

The authors declare no competing interests.