A chromosome-level genome of a Kordofan melon illuminates the origin of domesticated watermelons

For phylogenetic analyses of Citrullus, we assembled entire plastid genomes (121 genes and 33 spacers) and 6,183 single-copy nuclear genes for nine accessions from a taxonomic collection (Dataset S1 and Data Availability). Maximum likelihood (ML) analysis of the plastid alignment as well as coalescence-based analysis of the nuclear data yielded the same topology (with a minor difference in the concatenated ML analysis of the nuclear alignment, consistent with gene flow; Fig. 1A and SI Appendix, Fig. S1) and a backbone similar to earlier phylogenies that, however, lacked multiple Northeast African samples (18, 21). It now appears that Sudanese watermelons (collected in 1958, 1982, and 2017 in Darfur; SI Appendix, Table S1 and Dataset S1) are closest to the domesticated watermelon (Fig. 1A), with West African C. mucosospermus the next-closest relative. Similar to the West African egusi melon (C. mucosospermus) (23), which is cultivated in its local range, the Kordofan melon is grown by local farmers in Darfur where the very short rainy season requires locally adapted plants.

To complement our collection-based approach, we took advantage of publicly available genome resequencing data of various wild and cultivated watermelon accessions (8). We aligned the cordophanus genomic reads to the genome of C. lanatus cultivar “97103,” a typical East Asian cultivar with sweet, red, and crispy flesh, to identify single nucleotide polymorphisms (SNPs), which were combined with SNPs identified in 415 accessions reported in Guo et al. (8) The phylogenetic relationship between cordophanus and other Citrullus accessions were inferred using 114,193 SNPs at fourfold degenerate sites. Our C. lanatus subsp. cordophanus accession was located in the deepest branches of the C. lanatus clade and most closely related to PI 254622 and PI 481871 from Sudan (Fig. 2 A and B). Principal component analysis (PCA) showed that cordophanus located between C. mucosospermus and C. lanatus (Fig. 2C). Consistently, cordophanus was inferred to share ancestry with both C. mucosospermus and C. lanatus based on STRUCTURE (24) analyses (Fig. 2A). To test for possible gene flow between the Kordofan melon and other accessions, we performed ABBA-BABA tests (25). The results suggest significant gene flows between C. mucosospermus and C. lanatus subsp. cordophanus, C. lanatus landraces and C. lanatus subsp. cordophanus, and C. mucosospermus and C. lanatus landraces (SI Appendix, Fig. S2). These results are consistent with Guo et al. (8) and indicate admixture that could have occurred because of larger population sizes during the African Humid period (14,800 to 5,500 y ago) (26) or cultivation close to wild populations. Altogether, these results support that the Sudanese Kordofan melon could be a direct progenitor of the domesticated watermelon.

An alternative to the hypothesis that Kordofan melons are the watermelon’s progenitor is that they are instead a feral form (implying that the watermelon would have been domesticated in West Africa and subsequently brought to Sudan). If this were the case, the nucleotide diversity of Kordofan melons should be similar to that of watermelon landraces. We therefore calculated the nucleotide diversity (π) using the three cordophanus accessions and obtained a value of 0.68 × 10⁻³. The nucleotide diversity of our 87 accessions of landraces (Dataset S1) is 0.56 × 10⁻³. We then randomly selected three accessions from the 87 landraces, calculated their nucleotide diversity, and repeated this 10 times. This yielded a mean nucleotide diversity of 0.44 × 10⁻³ and an SE of 0.02 × 10⁻³. The nucleotide diversity in cordophanus is thus higher than that in the 87 sampled landraces. This finding is consistent with cordophanus being a progenitor rather than a feral form.

Prior to this study, two Ancient Egyptian illustrations of plausible watermelons were known. One comes from the tomb of Chnumhotep near Saqqara, dated to 4360 to 4350 BP (27, Fig. 1B). It shows an oblong fruit with dark green stripes on a flat surface. To the left of this fruit in this illustration are seven lotus flowers, and to the left of these are two snake melons (Cucumis melo var. flexuosus) on a tray. One can also see grapes, suggesting that this was a table laden with sweet foods. The other illustration comes from a tomb at Meir, northwest of Asyut, and shows a large oblong watermelon with dark longitudinal stripes served on a tray (28, 29) (Fig. 1C). Its precise age has not been resolved, but it probably dates to 4350 to 4200 BP (R. Schiestl, Alte Geschichte und Altertumskunde, University of Munich, February 2018). A third relevant illustration shows a perfectly globose fruit with longitudinal stripes on a stalk with two leaves slightly longer than the fruit (Fig. 1D). It comes from a papyrus of the 21st dynasty, 1069 to 945 BC (30), reproduced by Keimer (31), who thought it showed a wild form of C. lanatus. In its small shape (relative to the leaves) and stripes on the fruit, it matches Kordofan melons in photos of Ter-Avanesyan, a Russian breeder who in the early 1960s obtained seeds from the Kordofan region, which were propagated at a research station of the Vavilov Center in Tashkent (22). These watermelons have fruits of 23.5 × 21 cm in size with white, nonbitter pulp that has an aromatic taste. Both Ter-Avanesyan (32) and Fursa (33, 34) considered these watermelons to represent the progenitor of cultivated watermelon, describing them formally as subspecies or varietas cordophanus (C. lanatus subsp. cordophanus Ter-Avan., C. lanatus subsp. vulgaris var. cordophanus [Ter-Avan.] Fursa). Because they imply early consumption of raw (hence nonbitter) watermelon in the Nile Valley and because one depiction morphologically matches the Kordofan melon (Fig. 1D), these archaeological records are consistent with the Kordofan melon being a direct progenitor of the cultivated watermelon.

Our identification of Kordofan melons from collection-based phylogenomics and large-scale resequencing as potential progenitors of domesticated watermelon prompted us to generate a high-quality genome assembly to compare its genome structure and content to the domesticated form, assessing structural variants (SVs) in genes linked to the domestication. We generated ∼165 Gb Pacific Biosciences (PacBio) sequences for C. lanatus subsp. cordophanus, covering ∼388.8× of the genome, which were de novo assembled into contigs, followed by polishing with both PacBio and Illumina reads. The resulting assembly contained 86 contigs with a total size of 367.9 Mb and an N50 length of 9.34 Mb, which had higher contiguity than the watermelon “97103” PacBio genome assembly (8) (SI Appendix, Table S2). Combining the Hi-C contact maps and collinearity with the “97103” genome, 98.94% of the cordophanus contigs were clustered into 11 pseudomolecules (Fig. 3 and SI Appendix, Table S3 and Figs. S3 and S4). The completeness of the cordophanus genome assembly was assessed with BUSCO (35). About 97.2% of the core conserved plant genes were found complete in the assembly (SI Appendix, Table S4). We then aligned the RNA sequencing (RNA-seq) data generated from various tissues of cordophanus to the assembly, which showed mapping rates of up to 97.9% (SI Appendix, Table S5). These results indicate that the cordophanus genome assembly is of high quality.

About 57.7% of the cordophanus assembly were repetitive (SI Appendix, Table S6). A total of 23,043 protein-coding genes were predicted in the cordophanus genome, of which 20,554 (89.2%) were assigned with a putative function.

Bitterness in cucurbits depends on terpene compounds called cucurbitacins. The Bi gene, which encodes an oxidosqualene cyclase (OSC) that catalyzes the first committed step in cucurbitacin C biosynthesis, is critical for determining bitterness (36–38). Two bHLH transcription factors (Bl and Bt) regulate cucurbitacin C biosynthesis by up-regulating Bi expression in the leaves (Bl) and fruits (Bt) directly via binding to the E-box elements of the Bi promoter (37, 38). In addition to up-regulating Bi, the watermelon ClBl and ClBt transcription factors up-regulate most other cucurbitacin metabolism genes, including those encoding the two cytochrome P450 enzymes that convert cucurbidienol in cucurbitacin precursor (Cl890A and Cl890B) and the acyltransferase that converts the cucurbitacin precursor in cucurbitacin E (38). In cucumber (Cucumis sativus), honey melon (Cucumis melo), and watermelon, the examination of different lines with varying bitterness has revealed that the domestication of nonbitter fruits occurred via a convergent nucleotide substitution, leading to a premature stop codon in the Bt gene and resulting in a truncated, nonfunctional protein (36–38). Building on this knowledge, we compared genes in the cucurbitacin biosynthetic pathway and its regulation across Citrullus species. All cucurbitacin metabolic genes were conserved across species.

Analysis of the Bt gene across all Citrullus species revealed that the Kordofan melon Bt gene shares the substitution leading to a premature stop codon with the modern watermelon as well as C. mucosospermus (SI Appendix, Fig. S5A and https://figshare.com/s/7dc6e938e304d7854920), implying nonbitter fruits. The Kordofan melon has the homozygous nonbitter genotype at Chr01:3216322 (same as “97103”) (SI Appendix, Table S7), while C. mucosospermus is variable for bitterness, with about 20% of individuals having acid, plain, or sweet pulp, while the rest are bitter (table 5 in ref. 23). These results suggest a scenario wherein loss of pulp bitterness occurred in the most recent common ancestor of C. lanatus subsp. cordophanus and C. lanatus subsp. vulgaris following a transition phase in C. mucosospermus in which the trait was variable. Loss of pulp bitterness therefore appears to be a preadaptation for the domestication of watermelon, implying that early farmers probably took into cultivation nonbitter plants from the wild. However, our data on Kordofan bitterness come from few accessions, and we therefore cannot ascertain if the absence of fruit bitterness is fixed in the population. Nevertheless, the presence of nonbitter forms suggests that those were the ones brought into cultivation.

The red pulp color in modern watermelon is due to lycopene accumulation, likely by blocking the conversion of lycopene into β-carotene, a step mediated by the enzyme LCYB (lycopene β-cyclase) (39–41). LCYB is encoded by the lcyb gene, and watermelon accessions with red flesh are characterized by having a polymorphism in the lcyb gene on chromosome 4, resulting in a valine instead of a phenylalanine (V226F) (39, 41). These distinct LYCB alleles affect flesh color by modulating the abundance of the LYCB protein (41). Comparison of lycopene metabolic genes across all Citrullus species revealed a high conservation of these metabolic genes, with the same copy number for all genes in the lycopene pathway across all species, all of which were functional (lacked a premature stop codon; https://figshare.com/s/7dc6e938e304d7854920). With the exception of the domesticated, red-fleshed sweet watermelon, all Citrullus species including the Kordofan melons lack the V226F mutation of LYCB and have white to greenish pulp (SI Appendix, Fig. S5B). Indeed, the Kordofan melon has a homozygous non-red genotype at Chr04:15442987 (SI Appendix, Table S7). These results suggest that the red-fleshed color of watermelon could have arisen during the early stage of the domestication process.

To characterize genome-wide differences and look for differences in key domestication-related genes between the Kordofan melon and modern domesticated watermelon, we investigated SVs. The cordophanus and “97103” genome assemblies were compared with each other to identify SVs, including medium-sized [10 to 49 base pairs (bp)] and large SVs (≥50 bp). In addition, PacBio genomic reads of cordophanus and “97103” were mapped to the opposite genomes for SV detection. We found 13,884 medium-sized and 1,940 large SVs, hence a total of 15,824 SVs (Fig. 3 D–F and SI Appendix, Fig. S6). Based on the gene annotations of cordophanus and “97103,” 242 SVs led to changes in the coding sequences. Another 3,110 SVs in gene bodies were found only in introns or untranslated regions (UTRs), and an additional 3,363 SVs were located within 3 kb upstream of the translation start site (Dataset S2). Approximately 47.6% of the SV sequences were gypsy-like retrotransposons (SI Appendix, Fig. S7), higher than in the entire genome (37.8%), while the contents of other types of transposable elements were similar between the indel regions and the whole genome (SI Appendix, Fig. S7), suggesting that SVs are overrepresented in genome regions occupied by gypsy-like retrotransposons, similar to the pattern found in tomato (42).

The 15,824 identified SVs were then genotyped in 408 watermelon accessions using the Citrullus resequencing data reported in Guo et al. (8) (Dataset S1) After removing accessions with insufficient genotyping data, 393 accessions were used to investigate the SV dynamics in different populations, including C. amarus, C. colocynthis, C. mucosospermus, C. lanatus landraces, and C. lanatus cultivars. To evaluate the SV genotyping accuracy using genome resequencing data, we performed SV genotyping in the two reference accessions (“97103” and cordophanus) using Illumina short reads, which showed an accuracy rate of 97.8% for “97103” and 90.3% for cordophanus. The lower SV genotyping accuracy in cordophanus than in “97103” was mainly due to the higher heterozygosity of the cordophanus genome, which sometimes caused one allele to be assembled in the genome and another to be captured in the short reads.

The cordophanus alleles of a total of 1,414 (8.9%) SVs were not found in cultivated watermelons, of which 1,026 were also absent in watermelon landraces, indicating that the cordophanus genome provides breeders with new genetic elements that have been lost during watermelon domestication and improvement. In C. mucosospermus, SVs with homozygous cordophanus alleles comprised 54% of the SVs in each accession. In C. lanatus landraces, the cordophanus alleles become significantly (ANOVA; P < 0.0001) less prevalent (40%) and in C. lanatus cultivars, even less (31%) (Fig. 4A). Pairwise comparison of the cordophanus allele frequencies with all other species and populations identified 7,302 SVs with significantly changed frequencies (P < 0.01) in at least one comparison (Fig. 4B, SI Appendix, Fig. S8A, and Dataset S2). At the loci of these differential SVs, the frequency of cordophanus alleles was lower in C. lanatus (cultivars and landraces) than in C. mucosospermus and also in cultivars compared to landraces (SI Appendix, Fig. S8A). Out of the 3,075 SVs with significantly changed frequencies during domestication (C. mucosospermus to C. lanatus landraces), 312 (10.15%; compared to 4.97% of the total SVs) were found to be located in the previously identified domestication sweeps (8), and 319 (7.18%; compared to 3.70%) out of the 4,440 SVs with significantly changed frequencies from C. lanatus landrace to C. lanatus cultivar were found in the improvement sweeps. Our data showed that SVs with changed frequencies were significantly (Fisher’s exact tests; P < 0.0001) enriched within the regions under selection during domestication and improvement.

These changes of cordophanus allele frequencies are the result of evolution, domestication, and modern breeding. Fruit flesh sweetness was selected during domestication and continues to be an important breeding target. We identified a 12-bp indel (sv20309) in the promoter (∼1.7 kb upstream of the translation start site) of a sucrose synthase gene, Clc10G10370/Cla97C10G194010, located in a previously identified genomic region that is significantly associated with flesh sugar content (8) (Dataset S2). The “97103” allele (12-bp deletion) at sv20309 is the predominant genotype in C. lanatus landraces and cultivars (“97103” allele frequency > 0.97 in both), and the cordophanus allele is not found in the East Asian cultivars but is present in some American cultivars (Fig. 4C). We also identified five SVs whose cordophanus allele frequencies were significantly reduced during domestication (C. mucosospermus to C. lanatus landraces) and/or improvement (C. lanatus landraces to C. lanatus cultivars) (Fig. 4C and Dataset S2) in the promoter or intron regions of sugar metabolism/transport genes, including one sucrose phosphate synthase gene (Clc06G08580/Cla97C06G117750; sv11422), one arabinose 5-phosphate isomerase gene (Clc07G01070/Cla97C07G129400; sv13442), two nucleotide sugar transporter genes (Clc10G03570/Cla97C10G187870; sv19452 and Clc02G22490/Cla97C02G047270; sv04863), and one sugar transporter gene (Clc06G19460/Cla97C06G127910; sv13366). These results point to a role of these variants in determining pulp sugar content during watermelon domestication and improvement.

In addition to the SVs in genes potentially contributing to pulp sweetness, a 12-bp indel (sv18841) was detected in the 5′ UTR of an expansin gene, Clc09G19820/Cla97C09G179520, that may play a role in fruit development through modifying cell walls; its allele frequency increased from 0.70 in C. lanatus landraces to 0.99 in C. lanatus cultivars (SI Appendix, Fig. S8B and Dataset S2).

We also detected three SVs in the promoters or introns of disease resistance genes (CC-NBS-LRR and TIR-NBS-LRR) that had significantly reduced cordophanus allele frequency after domestication and/or improvement (SI Appendix, Fig. S8C). Functional genomic work, aided by CRISPR-cas9 genome editing—which is now established in watermelon (43)—will enable researchers to test whether the successive changes in the frequency of SVs in disease resistance genes during watermelon domestication mirrors a genetic erosion of the plants’ defense genetic toolkit as has been postulated (9). This in turn could prove useful to engineer disease-resistant watermelons.

This study identifies Kordofan melons from Sudan as closest relatives, and perhaps progenitors, of modern watermelons and implies that progenitor populations still exist in the eastern Sahel savannas in the Darfur region, possibly as relicts of a formerly more widespread range into parts of the Sahara during the Holocene African humid period. The sweet, red watermelon may have been domesticated there by Nubian or other Nilo-Saharan ethnicities (44), and the crop could then have spread northward, where it appears to have been consumed as a dessert by 4360 BP (Fig. 1B). Alternatively, but less likely, watermelon could have been domesticated in West Africa where the soft-seeded C. mucosospermus is endemic (Fig. 2), the natural range of which might have extended north into Libya, where ancient seeds of Citrullus have been collected at Uan Muhugiagg (15) close to the Takarkori rock shelter, which covers four millennia of human occupation in the central Sahara (45). Human population levels reached a peak around 7500 BP, with an expansion of populations into new Saharan territories (46), and early farmers could have brought seeds with them as well as encountering new local variants to select on. While it is still unsure which people domesticated the watermelon, our high-resolution, chromosome-level genome for the Kordofan melon provides a substantial resource for watermelon breeding. For example, Kordofan melons possess significantly different disease resistance alleles compared to domesticated watermelons.

This work was funded by DFG 603/27-1, the US Department of Agriculture National Institute of Food and Agriculture Specialty Crop Research Initiative (Grant 2020-51181-32139), and generous support from the Elfriede and Franz Jakob Foundation. G.C. is funded by a Natural Environment Research Council Independent Research Fellowship (Grant NE/S014470/1). We thank two anonymous reviewers for their excellent suggestions and Richard Parkinson, Egyptologist at the University of Oxford, for discussion.