{"paper_id":"0cfd93e6-394e-464a-83c5-9ed2a43ca72a","body_text":"1 \n \nPopulation-level super-pangenome reveals  genome evolution and empowers precision \nbreeding in watermelon \n \nHonghe Sun1,2,3,#, Jie Zhang2,#, Shengjin Liao2,#, Shaogui Guo2,#, Zhe Zhou1,4,#, Xuebo Zhao1, Shan \nWu1, Jiantao Zhao1, Guoyi Gong2, Jinfang Wang2, Maoying Li2, Yongtao Yu2, Yi Ren2, Shouwei \nTian2, Shaofang Li2, Haiying Zhang2, Sue A. Hammar5, Cecilia McGregor6, Robert Jarret7, Patrick \nWechter8, Sandra E. Branham8, Chandrasekar Kousik9, Amnon Levi9, Rebecca Grumet5, Zhangjun \nFei1,*, Yong Xu2,* \n \n1Boyce Thompson Institute, Cornell University, Ithaca, NY , USA. \n2State Key Laboratory of Vegetable Biobreeding, National Engineering Research Center for \nVegetables, Beijing Key Laboratory of Vegetable Germplasms Improvement, Beijing Vegetable \nResearch Center, Beijing Academy of Agriculture and Forestry Science, Beijing, China. \n3Plant Biology Section, School of Integrative Plant Science, Cornell University, Ithaca, NY , USA. \n4Zhengzhou Fruit Research Institute, Chinese Academy of Agricultural Sciences, Zhengzhou, \nHenan, China. \n5Department of Horticulture, Graduate Program in Plant Breeding, Genetics and Biotechnology, \nMichigan State University, East Lansing, MI, USA. \n6Department of Horticulture, University of Georgia, Athens, GA, USA. \n7U.S. Department of Agriculture -Agricultural Research Service, Plant Genetic Resources \nConservation Unit, Griffin, GA, USA. \n8Coastal Research and Education Center, Clemson University, Charleston, SC, 29414, USA. \n9U.S. Department of Agriculture-Agricultural Research Service, U.S. Vegetable Lab, Charleston, \nSC, USA. \n#These authors contributed equally. \n*Email: xuyong@nercv.org; zf25@cornell.edu \n \n \nAbstract \nPangenomes are increasingly critical for harnessing crop genetic diversity, yet their resolution and \nutility are often limited by insufficient sampling of high -quality genome assemblies.  Here, we \nreport a population -level watermelon super -pangenome constructed from 138 reference -grade \nassemblies, including 135 newly generated  near-gapless genomes representing all seven \nwatermelon species. The super-pangenome captures approximately one million structural variants \n(SVs), enabling accurate variant genotyping across ~900 watermelon accessions and substantially \nexpanding variant discovery both across and within species. Broader sampling within the \npangenome provides insights into genome evolution among watermelon species and sheds light \non the origin of cultivated watermelon . SV-inclusive genome-wide association studies enhance \ntrait mapping resolution and identif y a copy number variation upstream of ClFCI1 that regulates \nflesh color intensity in a dosage-dependent manner. Leveraging this comprehensive variation map, \nwe developed high -accuracy genomic prediction models for 18 agronomic traits. Together, our \nfindings and genomic resources establish a foundational framework for dissecting complex traits \nand accelerating precision breeding in  watermelon, while offering a valuable model for SV -\nresolved pangenomics in crop species. \n \n \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n2 \n \nIntroduction \nWatermelon (Citrullus lanatus subsp. vulgaris) is one of the most commercially important fruit \ncrops worldwide, with a global production of approximately 105 million tonnes in 2023, ranking \nthird among all fruit crops (FAOSTAT, 2023). Its sweet, juicy, and brightly colored flesh has driven \nstrong consumer demand. Watermelon belongs to the genus Citrullus, which originates in Africa \nand comprises six additional extant species: C. naudinianus , C. colocynthis , C. rehmii , C. \necirrhosus, C. amarus, and C. mucosospermus. Archaeobotanical and genomic evidence indicates \nthat sweet watermelon was domesticated in the northeastern African region from the wild form C. \nlanatus subsp. cordophanus1–3, while a recent comparative genomic analysis suggests that C. \nmucosospermus could be an additional ancestor of sweet watermelon4. \nModern watermelon cultivars exhibit substantial diversity in traits such as flesh color, sugar \ncontent, fruit size, shape, and rind pattern. However, intensive selection for these fruit quality \nattributes has led to a marked reduction in genetic diversity, accompanied by the loss of numerous \ndisease-resistance and abiotic stress -tolerance traits, raising concerns about the long -term \nsustainability of watermelon production5. Wild Citrullus species possess valuable adaptive traits, \nincluding resistance to various pathogens, tolerance to environmental stresses, as well as enhanced \nlevels of health -promoting compounds such as citrulline, which are critical resources for the \ngenetic improvement of cultivated watermelon. Therefore, comprehensive characterization of \ngenetic variation, both within cultivated germplasm and between cultivated and wild relatives, is \nurgently needed to enable effective genomics-assisted breeding and genetic engineering aimed at \nimproving fruit quality and resilience in watermelon. \nHigh-quality reference genomes and SNP -based variation maps have substantially \nadvanced our understanding of watermelon domestication and key agronomic traits 6–8. However, \ngenomes from one or a few accessions do not offer sufficient coverage to efficiently assess the \ngenetic diversity within a species or genus.  Genetic bottlenecks in domestication and breeding \nhave further constrained genetic diversity within cultivated watermelon germplasms. These \nlimitations significantly restrict the genetic information available for watermelon breeding and \nimpede the detection of causative variants/genes underlying important agronomic traits. To \novercome these challenges, pangenomic approaches are essential for expanding the repertoire of \ngenetic diversity accessible for watermelon improvement. Additionally, graph-based pangenomes \ngreatly enhance the detection of structural variants (SVs) —including large insertions, deletions, \ninversions, and translocations—which comprise a substantial portion of genomic diversity and are \nincreasingly recognized as major contributors to phenotypic variation9–11. \nIn this study, we assembled 135 near -gapless reference -quality genomes encompassing \ncultivated watermelon and its wild relatives. Leveraging these assemblies, we constructed a \npopulation-level graph-based super-pangenome of watermelon that captures nearly one million \nlarge SVs, which were confidently genotyped across more than 900 watermelon accessions. This \nresource enabled high -resolution analyses of genomic variation, providing new insights into the \norigin of cultivated watermelon. By integrating SVs into genome-wide association studies (GWAS) \nof 18 agronomic traits, we identified a causal copy number variation (CNV) underlying flesh color \nintensity. Furthermore, leveraging the extensive variation map within our graph -based super -\npangenome, we demonstrated the potential of genomic selection for enhancing fruit quali ty and \ndisease resistance in watermelon breeding. This study establishes a foundational genomic resource \nto accelerate future genomics-assisted breeding strategies aimed at optimizing watermelon quality \nand productivity. \n \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n3 \n \nResults \nGenome assembly and annotation of 135 watermelon accessions \nTo capture the genetic diversity within the Citrullus genus, we selected a total of 135 representative \naccessions spanning all seven extant species for genome assembly. These included 1 C. \nnaudinianus, 1 C. rehmii, 2 C. ecirrhosus, 5 C. colocynthis, 16 C. amarus, 9 C. mucosospermus, 6 \nC. lanatus subsp. cordophanus, and 95 C. lanatus subsp. vulgaris (comprising 7 landraces and 88 \ncultivars) (Supplementary Table 1). The panel was designed to integrate core germplasm, founder \ninbred lines and accessions with v aluable traits such as disease resistance 12,13. HiFi reads were \ngenerated for all 135 accessions at an average depth of ~30.3×. Additionally, Oxford Nanopore \nTechnologies (ONT) ultra -long reads (~54.6× on average) and high -throughput chromosome \nconformation capture (Hi-C) reads (~153.8×) were generated for ten of these accessions, spanning \ncultivar, landrace, C. lanatus subsp. cordophanus, and three wild relatives widely used in disease-\nresistance breeding, C. mucosospermus, C. amarus, and C. colocynthis (Supplementary Table 1). \nThe assembled genomes had an average size of 374 Mb, with an average contig N50 size \nof 31.2 Mb (Supplementary Table 1). Approximately 99.2% of the assembled sequences were \nanchored and ordered onto the 11 watermelon chromosomes. Of the assembled chromosomes, 78.2% \n(1162 out of 1485) contained telomeres at both ends and 52.6% (781) were completely gapless. \nAmong the ten genomes assembled from HiFi, ONT, and Hi-C reads, seven were gapless telomere-\nto-telomere (T2T) genomes. The remaining three contained gaps –two with a single gap and one \nwith three–likely due to unresolved centromeric and ribosomal DNA (rDNA) repeats 14. Genome \nassemblies derived from the accession ‘97103’ have served as the primary references for \nwatermelon genomic studies . Compared to the previous version 6, the gapless T2T genome of \n‘97103’ assembled in this study (version 3) increased the size from 360 Mb to 370 Mb with \nmarkedly improved base-level accuracy (consensus quality value [QV] of 65 vs. 45.3). \nBUSCO15 evaluation revealed high completeness across these 135 genome assemblies, \nwith an average completeness rate of 99.1%. Assessment using a k-mer-based approach16 indicated \nan average QV of 64.9 ( Supplementary Table 1). These metrics were comparable to or higher \nthan those reported for the recently published watermelon genome assemblies 4, underscoring the \nrobustness and high accuracy of our assemblies . The transposable element (TE) content ranged \nfrom 59.0% to 64.7%, and between 20,834 and 23,330 protein -coding genes were predicted in \nthese 135 Citrullus genomes (Supplementary Table 2). \n \nChromosomal evolution of the Citrullus genus \nLarge chromosomal rearrangements, including translocations and inversions, play crucial roles in \nplant evolution and domestication 17,18. Understanding these structural variations in Citrullus \naccessions can enhance their effective utilization in breeding programs. Comparative analysis of \nthe 135 genome assemblies generated in this study, along with three previously published \ngenomes2,19,20, identified a total of 11 translocations and 101 large inversions (>100 kb) ( Fig. 1a \nand Supplementary Tables 3 and 4). Five of  the translocations were specific to C. lanatus . \nNotably, one event between chromosomes 2 and 3 has been reported to disrupt the structure of the \ngynoecious gene ClWIP1, thereby leading to the gynoecious phenotype 21. Another C. lanatus-\nspecific translocation, between chromosomes 6 and 10, has been found to cause chromosomal \nsynapsis abnormalities during meiotic diakinesis in hybrids, resulting in fruits with reduced seed \nnumbers22. Of the 101 large inversions, 52 were specific to wild relatives ( C. colocynthis , C. \namarus, and C. mucosospermus ), with 22 overlapping with known disease -resistance QTLs \n(Supplementary Table 4). \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n4 \n \n \nFigure 1 Genome evolution in the Citrullus genus. (a) Whole -genome alignments showing inter -\nchromosomal translocations and large inversions. One representative genome from each wild \nspecies/subspecies (abbreviations defined in b), along with genomes from one landrace and five cultivars, \nare displayed. (b) Time-calibrated species tree. ( c) Estimated divergence time between C. lanatus subsp. \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n5 \n \ncordophanus and cultivated watermelon using SMC++. Top: Demographic history of the effective \npopulation size in C. lanatus subsp. cordophanus and cultivated watermelon. The blue vertical line marks \nthe estimated split time between the two. Bottom: Histogram of split time estimates based on random \naccession sampling, with a mean divergence time of ~4.9 thousand years ago (Kya). ( d) Distribution of \nlarge inversions across genome assemblies (53 cultivars without large inversions are not shown). \n \nWild relatives have long been recognized as invaluable source for disease resistance in \nwatermelon. However, linkage drag often complicates the breeding process by inadvertently \nintroducing undesirable alleles, leading to trade -offs in key traits such as yield and fruit quality. \nFor instance, QTLs associated with Fusarium wilt resistance (qFon2-6; 9.77-25.00 Mb), sweetness \n(QBrix6; 10.21 -11.24 Mb), and flesh firmness (13.00 -20.54 Mb) are closely located on \nchromosome 6 (refs. 23–25). Introgression analysis using RFMix26 revealed that the genomic region \nencompassing these three QTLs originated from C. amarus (Supplementary Fig. 1). Three large \ninversions within this region (10.49 -13.15 Mb, 17.43 -19.64 Mb, and 21.96 -23.29 Mb) likely \nsuppress recombination27, resulting in the acquisition of desirable traits like disease resistance and \nfirm flesh but at the cost of decreased sweetness. The cultivar ‘SugarleeXZ’ inherited all these \nthree inversions. In contrast, C. amarus accessions ‘PI 296341-FR’ and ‘USVL252’ lacked the first \nand the first two inversions, respectively ( Supplementary Fig. 1). Therefore, using ‘PI 296341 -\nFR’ and ‘USVL2 52’ in backcross breeding could facilitate the introgression of resistance and \nfirmness traits without compromising sweetness. \nThe evolutionary timeline and genomic relationships among Citrullus species \nreconstructed in this study ( Fig. 1b ) broadly aligned with previous reports 4,28. Notably, we \nestimated that C. mucosospermus diverged from C. lanatus approximately 120,000 years ago, well \nbefore the domestication of any crops (<12,000 years ago) 29. In contrast, the divergence between \ndomesticated watermelon and C. lanatus subsp. cordophanus was estimated at ~4,900 years ago \n(Fig. 1c), closely aligning with archaeological evidence of  watermelon domestication (~4,000 –\n6,000 years ago) in northeastern Africa 20,30. These evolutionary timelines further support C. \nlanatus subsp. cordophanus as the likely direct wild progenitor of cultivated watermelon, while C. \nmucosospermus is unlikely to be a direct ancestor.  \nA recent study proposed C. mucosospermus as an additional progenitor based on seven \ndiagnostic variants shared with cultivated watermelon but absent from C. lanatus  subsp. \ncordophanus4. However, this conclusion relied on a single C. lanatus subsp. cordophanus genome, \noverlooking potential intraspecific variation within this subspecies. In this study, we analyzed \nseven C. lanatus subsp. cordophanus genome assemblies, revealing that four of the seven variants \nwere, in fact, segregating within the population ( Supplementary Table 5). ABBA-BABA tests31 \nrevealed that the remaining three variants did not fall within genomic regions introgressed from C. \nmucosospermus, suggesting that they are unlikely derived from that species ( Supplementary \nTable 6). Thus, all seven variants can be plausibly explained by variation within C. lanatus subsp. \ncordophanus, and the absence of three from current assemblies likely reflects incomplete sampling. \nThese findings highlight the importance of comprehensive population sampling when inferring \ncrop ancestry, as reliance on limited genomes can obscure intraspecific diversity and lead to \nmisleading conclusions. \n \nComprehensive gene-based super-pangenome \nThrough gene clustering across the 138 Citrullus genomes, we constructed a comprehensive gene-\nbased super -pangenome for wild and cultivated watermelons, comprising 35,919 pangenes –\napproximately 1.6 times the number of genes in the ‘97103’ reference genome (v3). The number \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n6 \n \nof pangenes increased with the inclusion of additional genomes , plateauing around 80 (Fig. 2a). \nPangenes were classified into four categories: core (39.3%), softcore (5.4%; presented in 137 \naccessions), shell (54.4%; present in 2 -136 accessions), and private (0.9%) ( Fig. 2b). Within \ndomesticated watermelon, the inclusion of 96 genomes identified 7,279 additional pangenes \nbeyond those found in the ‘97103’ genome ( Supplementary Fig. 2 ). As the wild progenitor of \ncultivated watermelon, C. lanatus  subsp. cordophanus contributed 1,537 (4.3%) additional \npangenes absent in cultivated accessions ( Fig. 2c  and Supplementary Fig. 2a ). Three wild \nrelatives widely used in watermelon breeding programs, C. mucosospermus, C. amarus, and C. \ncolocynthis, collectively contributed 4,732 pangenes (13.2%) not found in C. lanatus. In contrast, \nC. naudinianus, C. rehmii, and C. ecirrhosus contributed only 666 additional pangenes, likely due \nto limited sampling. \nNucleotide-binding site leucine-rich repeat (NLR) genes play a pivotal role in plant disease \nresistance32. In addition to the 46 NLR pangenes present in the ‘97103’ reference genome, we \nidentified 41 novel NLR pangenes from the pangenome: 3 from non -reference cultivated \nwatermelons, 4 from C. lanatus subsp. cordophanus, and 34 from six wild species ( Fig. 2d and \nSupplementary Table 7 ), highlighting wild species as a rich reservoir of disease resistance \ndiversity. \n \n \nFigure 2 Gene pool dynamics in the Citrullus genus. (a) Modeling of gene-based pan- and core-genome \nsizes as additional genomes are incorporated. ( b) Composition of the gene -based Citrullus super-\n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n7 \n \npangenome. (c) Presence-absence variation (PA V) of pangenes across cultivated watermelon and its wild \nrelatives. Top 30 intersected groups are plotted. ( d) Presence-absence variation of N LR pangenes across \nCitrullus species. ( e) Flesh sugar content (°Brix) in accessions carrying one or two copies of the \nhexosyltransferase gene. Data are from two field trials, conducted in Hainan and Yanqing. ‘*’ and ‘**’ \nindicate P < 0.05 and P < 0.01, respectively (Student’s t test). (f) Schematic diagram of the cultivar-specific \n65-kb insertion in different Citrullus groups. Each rectangle represents a gene, and purple rectangles \nindicate the hexosyltransferase gene. Cultivar A: ‘97103’; Cultivar B: ‘TWFYingRou’. \n \nTandem gene duplication can increase gene dosage and enhance the expression of \nbeneficial traits. Using the gene-based super-pangenome, we identified 31 pangenes that exhibited \nhigher frequencies of tandem duplication in cultivated watermelons compared to wild species and \nwere expressed at higher levels in the flesh of ‘97103’ relative to the wild accession ‘PI 296341 -\nFR’ (Supplementary Table 8 ). These included the previously reported tandem  duplication of  \nClTST2, which is associated with increased flesh sweetness and was strongly favored during \ndomestication, becoming nearly fixed in cultivars (allele frequency of 97.1%) 2, thereby limiting \nits potential in future improvement of fruit sweetness. Among the newly identified tandem \nduplicates, we discovered a hexosyltransferase gene duplication ( XG0025C01G000760 and \nXG0025C01G000860) specific to cultivars, with an allele frequency of 28%. Hexosyltransferases \ncatalyze the transfer of hexose sugars to various acceptor molecules and are involved in sucrose \nmetabolism33. Notably, we found that this duplication was significantly associated with increased \nflesh sweetness (Fig. 2e). These findings suggest that the hexosyltransferase duplication represents \na promising target for enhancing flesh sweetness in future watermelon breeding. \n \nGraph-based pangenome facilitates trait-variation association \nTo capture the full spectrum of genetic diversity within the Citrullus genus, we performed pairwise \ngenome alignments using the ‘97103’ genome as the reference, leading to the identification of \n37,699,340 SNPs, 8,294,544 small indels (<20 bp), and 910,844 large SVs (≥20 bp; including \n502,800 insertions and 408,044 deletions). The number of SVs per accession was positively \ncorrelated with their genetic distance from ‘97103’ ( Fig. 3a and Supplementary Table 9). The \ncumulative SV count within each group significantly surpassed the average observed in individual \naccessions (Fig. 3b). Notably, the inclusion of 89 cultivars captured 60,297 SVs –over ten times \nthe average (5,933)–highlighting a substantially enhanced representation of genetic diversity. \nWe then constructed a graph -based pangenome by integrating SNPs, indels, and SVs, enabling \naccurate SV genotyping in 7 76 re-sequenced accessions, including 313 newly sequenced in this \nstudy (Supplementary Table 10). Combined with variants from the 138 genome assemblies, this \nyielded a comprehensive variation map encompassing 91 4 wild and cultivated watermelon  \naccessions. Using this SV-inclusive variation map, we uncovered a previously unreported 216-bp \ninsertion in the first exon of ClBt (the bitterness gen e), which introduces three premature stop \ncodons. This insertion was in complete linkage with the previously reported nonsense SNP in the \nsecond exon6, together forming a haplotype that underlies the loss of bitterness . This haplotype \nwas completely fixed in cultivated watermelons and C. lanatus subsp. cordophanus, but was rare \nin wild relatives, particularly those other than C. mucosospermus (Fig. 3c). The discovery of this \ninsertion provides new insight into the genetic basis of bitterness loss and reveals a key haplotype \nat the ClBt locus. \n \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n8 \n \n \nFigure 3 Landscape of structural variation across Citrullus species. (a) Number of large insertions and \ndeletions identified in each of the 138 Citrullus accessions. (b) Average (black line) and cumulative (red \ndots) number of distinct structural variants (SVs) across different  Citrullus groups. Detailed numbers are \nprovided in Supplementary Table 9. (c) Haplotypes of the ClBt gene and their distribution among different \nwatermelon groups. (d) Copy number variation of the 65-kb insertion across cultivars, landraces, and wild \nCitrullus accessions. (e,f) Local Manhattan plots (left) and corresponding box plots showing the distribution \nof accessions carrying distinct alleles (right) for flesh sugar content (e) and FON 2 (Fusarium oxysporum f. \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n9 \n \nsp. niveum race 2) resistance (f). Horizontal solid and dashed lines represent Bonferroni-corrected genome-\nwide significance thresholds at α = 0.05 and α = 0.10, respectively.  \n \nWe further identified 622 and 983 SVs under selection during watermelon domestication \n(C. lanatus subsp. cordophanus vs. landrace) and improvement (landrace vs. cultivar), respectively, \nencompassing 184 and 395 genes. Among SVs under selection during improvement, we identified \na 65-kb insertion that led to the tandem duplication of the aforementioned hexosyltransferase gene \n(Fig. 2f). Allele frequency analysis indicated that this insertion originated in landrace and was \nsubsequently favored during watermelon improvement (Fig. 3d). Additionally, this variation map \nprovided insights into the genetic basis of fruit shape: both the nonsynonymous SNP and 159 -bp \ndeletion in ClFS1 (a key gene for fruit shape) reported previously 34,35 were captured. The \nnonsynonymous SNP was present in both C. amarus (allele frequency of 42%) and cultivated \nwatermelons (11%), while the 159 -bp deletion was found exclusively in cultivated watermelons \n(9%), indicating that this variation likely originated during domestication ( Supplementary Fig. \n3a,b). Furthermore, the phenotypic effect of the deletion appeared to be stronger than that of the \nSNP ( Supplementary Fig. 3c ). Together, these findings highlight how the graph -based \npangenome elucidates key domestication and improvement alleles underlying important traits such \nas fruit bitterness, sweetness, and morphology in watermelon. \nEmpowered by the comprehensive variant set –particularly SVs –captured in the graph -\nbased pangenome, we conducted genome -wide association studies (GWAS) for 12 fruit -quality \nand 6 disease -resistance traits ( Supplementary Table 11 and Supplementary Fig s. 4 –5). We \nidentified a total of 93 loci significantly associated with at least one trait, 14 (14.9%) of which had \nlead signals marked by SVs ( Supplementary Table 1 2). Notably, a GWAS signal for flesh \nsweetness on chromosome 10 was detected exclusively in the SV -based GWAS. This signal \nincluded a 29 -bp deletion located 187 bp downstream of the transcription factor ClNOR \n(XG0025C10G005980) ( Fig. 3e ), a gene recently shown to regulate fruit ripening and sugar \naccumulation in watermelon36. For resistance to Fusarium oxysporum f. sp. niveum (FON) race 2, \na 135-bp insertion at 10.8 Mb on chromosome 10 showed the strongest association, and accessions \ncarrying this insertion exhibited significantly enhanced resistance ( Fig. 3f). This SV is located \n4,224 bp upstream of an AAA -type ATPase gene (XG0025C10G006750), whose homologs have \nbeen implicated in broad -spectrum disease resistance in rice and Arabidopsis37,38. For nematode \nresistance, an association signal was detected on chromosome 3, overlapping with the previously \nmapped QTL 3.1 for nematode resistance39. The lead variant was an SV located 7,014 bp upstream \nof XG0025C03G016370, which encodes a calmodulin-binding protein 60 (CBP60), a member of \na tandem gene cluster in this region. Additional associated SVs were identified within the promoter, \nintronic, and coding regions of other CBP60 members in the cluster ( Supplementary Fig. 6). \nGiven that CBP60 genes are known regulators of plant immune responses40, this locus likely plays \na pivotal role in nematode defense. Together, these results underscore the enhanced resolution and \ndiscovery power of SV-inclusive GWAS in uncovering trait-associated variants in watermelon. \n \nA copy number variant regulates flesh color intensity \nFlesh color is a key fruit -quality trait in watermelon, with b righter flesh colors enhancing visual \nappeal and reflecting higher levels of nutritional compounds such as carotenoids. Using chroma \nvalue as a quantitative metric (Fig. 4a), our GWAS identified two significant signals for flesh color \nintensity: a major locus on chromosome 6 at 24.5 Mb ( FCI1) and a minor one on chromosome 8 \nat 24.9 Mb ( FCI2) (Fig. 4b). The major peak at FCI1 corresponded to a 2, 516-bp insertion that \nexhibited low linkage disequilibrium with nearby SNPs, likely explaining why it was not detected  \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n10 \n \n \nFigure 4 Copy number variation in the promoter region of ClFCI1 controls watermelon flesh color \nintensity. (a) Photos of representative watermelon accessions showing the gradient of flesh color intensity. \n(b) Manhattan plot of GWAS for flesh color intensity. Red and green horizontal lines indicate genome-wide \nsignificance thresholds at α = 0.05 and α = 0.10, respectively. (c) Zoomed-in Manhattan plot (top) and LD \nheatmap (bottom) at the ClFCI1 locus (24.5-24.6 Mb on chromosome 6). Black triangles in LD heatmap \nindicate LD blocks. The 2,516-bp insertion was the only variant in this region significantly associated with \nflesh color intensity. (d) Structural diagram showing copy-number variation of the 1,258-bp segment (blue \nboxes) in the ClFCI1 promoter across representative watermelon accessions. ( e) Expression levels of \nClFCI1 in fruit flesh of parents and F 1 lines from the cro sses ‘Ming 58’ × ‘JX2’ (left) and ‘JLM’ × ‘CS’ \n(‘Cream of Saskatchewan’; right). Error bars indicate the standard deviation of three biological replicates. \n(f) Violin plots of chroma values in accessions carrying one to four copies of the 1,258-bp segment. Black \ndots indicate chroma values in individual accessions, and horizontal red bars indicate mean chroma values. \n(g) Dual-luciferase (LUC) reporter activity driven by ClFCI1 promoters carrying one to three copies of the \n1,258-bp segment. Error bars indicate standard deviation of ten biological replicates. ** denotes P < 0.01 \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n11 \n \n(Student’s t-test). (h) Representative fruits and chroma values of ClFCI1-knockdown (ClFCI1-KD), wild \ntype (WT, ‘ZZJM’), and ClFCI1-overexpression (ClFCI1-OE) lines. (i) Carotenoid contents (mg/kg flesh \nweight) and relative ClFCI1 expression in wild-type and transgenic lines. Fruits were sampled at 34 days \nafter pollination. Error bars indicate standard deviation of three biological replicates. Different lowercase \nletters indicate significant differences according to Turkey’s multiple range test (P < 0.05). \n \nin SNP-based GWAS (Fig. 4c). Bulked-segregant analysis (BSA) using an F2 population derived \nfrom a cross between ‘Ming 58’ (scarlet red flesh) and ‘JX2’ (pink flesh) independently mapped \nthe FCI1 locus to a 2.8-Mb interval (23.8–26.6 Mb) (Supplementary Fig. 7a). Fine mapping using \nthis F2 population (n = 141) and an additional F 2 population derived from ‘JLM’ (yellow flesh) ´ \n‘Cream of Saskatchewan’ (pale yellow flesh) (n = 636) narrowed the locus to a 146 -kb region \n(24.45-24.60 Mb) containing 13 genes (Supplementary Fig. 7b,c). Of these, five were expressed \nin fruit flesh and only XG0025C06G012030 (hereafter ClFCI1) showed an expression pattern \nconsistent with contrasting dark - and light-flesh phenotypes (Supplementary Fig. 7d). ClFCI1, \nwhich encodes a tetratricopeptide repeat (TPR) protein, is homologous to the Arabidopsis \nREDUCED CHLOROPLAST COVERAGE (REC) genes known to regulate chloroplast \ncompartment size and chlorophyll content 41. The 2, 516-bp insertion corresponding to the FCI1 \npeak resulted in a tandem triplication of a 1,258 -bp promoter segment located ~1.8 kb upstream \nof ClFCI1 (Fig. 4d). It is worth noting that no other sequence polymorphisms were found within \nor near the ClFCI1 coding region between the mapping parents.  \nDifferent copy numbers of the 1,258-bp promoter segment, ranging from one to four, were \nobserved among watermelon accessions ( Fig. 4d). The copy number of this segment showed a \npositive correlation with both flesh color intensity and ClFCI1 transcript abundance in the mapping \nparents and their F1 hybrids (Fig. 4e). Similarly, across the panel of natural watermelon accessions, \nincreased copy number was positively associated with greater flesh color intensity ( Fig. 4f ). \nTransient dual-luciferase assays further validated a dose-dependent increase in promoter activity, \nwith multi-copy alleles driving significantly higher reporter expression compared to the single -\ncopy allele (Fig. 4g). \nTo elucidate the functional role of ClFCI1 in regulating flesh color intensity, we generated \nboth antisense knockdown and overexpression lines. Knockdown of ClFCI1 in the red-fleshed line \n‘ZZJM’ reduced pigmentation and carotenoid content in the fruit flesh. In contrast, overexpression \nof ClFCI1 enhanced flesh pigmentation and increased carotenoid accumulation ( Fig. 4h,i ). \nTranscriptome analysis of fruits at 34 days after pollination (DAP) identified 1,131 and 832 \ndifferentially expressed genes (DEGs) in th e ClFCI1 knockdown and overexpression lines, \nrespectively, compared to the wild type  (Supplementary Tables 13 and 14). These DEGs were \nsignificantly enriched for genes involved in photosynthesis and plastid development \n(Supplementary Table 15), suggesting that ClFCI1 regulates watermelon flesh color intensity \nprimarily by modulating these processes.  \nAcross the 914-accession panel, multi -copy alleles of the 1,258 -bp ClFCI1-promoter \nsegment were predominantly found within C. lanatus, while in wild Citrullus species, only single \ninstances of tandem duplication and triplication were observed in C. mucosospermus  and C. \namarus, respectively (Supplementary Table 16). Within C. lanatus, the frequency of multi-copy \nalleles increased significantly from 12.5% in the wild progenitor cordophanus to 50.25% in \ndomesticated accessions (Fisher’s exact test, P = 0.0037), and showed a moderate increase during \nimprovement, from 40.38% in landraces to 51.74% in modern cultivars ( P = 0.14). Among \ncultivars, tandem triplication was more prevalent than tandem duplication, consistent with \nselection favoring intensified flesh coloration in modern watermelon breeding programs. These \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n12 \n \nfindings suggest that copy number variation in the ClFCI1 promoter represents a promising target \nfor marker -assisted breeding aimed at enhancing flesh color intensity and nutritional value in \nwatermelon. \n \nGenomic selection empowered by the graph pangenome \nGenomic selection has emerged as a transformative strategy for accelerating crop improvement42. \nHere, leveraging the comprehensive variation map captured in the graph pangenome, we \nestablished a robust genomic selection framework for watermelon by training prediction models \nfor 18 traits related to fruit quality and disease resistance. For each trait, we employed CropGBM43 \nfor both marker selection and genomic prediction, identifying informative genome -wide variants \nbased on feature importance. We built genomic selection models using either an SNP -only panel \nor a combined SNP+SV panel. Most traits required 476–756 markers, with two traits needing fewer \nthan 100 ( Supplementary Table 17). Five -fold cross -validation with five repeats yielded \ngenerally high prediction accuracies, ranging from 0.56 to 0.97 (Fig. 5a). Inclusion of SVs did not \nimprove the prediction accuracies for 17 of the 18 traits but did slightly enhance performance for \nthe flesh color category trait, indicating that large-effect structural polymorphisms not sufficiently \ncaptured by SNPs alone may underlie this trait. \n \n \nFigure 5 Graph-based pangenome empowers genomic selection in watermelon. (a) Genomic prediction \naccuracies for 12 fruit -quality traits and 6 disease -resistance traits using models built with high -effect \nvariants selected from two different sets: SNP-only and SNP+SV . Trait abbreviations: WMV II, resistance \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n13 \n \nto watermelon mosaic virus II. PRSV , resistance to papaya ringspot  virus-watermelon strain. BFB, \nresistance to bacterial fruit blotch. PM2W, resistance to powdery mildew race 2W. FON 2, resistance to \nFusarium oxysporum f. sp. niveum race 2. For each boxplot, the lower and upper bounds indicate the first \nand third quartiles, respectively, the center line indicates the median, and the whiskers extend to 1.5× the \ninterquartile range. ( b) Genomic prediction accuracies for flesh sugar content (°Brix) in cultivat ed \nwatermelon using the model built with selected SNP+SV markers. Flesh sugar contents were measured in \nthree independent experiments: GRIN, data from the USDA GRIN database; Yanqing, field trial in 2019 in \nYanqing, China; Hainan; field trial in 2022 in Ha inan, China. (c) Linear regression analysis of predicted \nand observed flesh sugar contents in a RIL population derived from a cross between cultivar ‘97103’ and \nC. amarus ‘PI 296341-FR’. \n \nTo further validate model robustness, we focused on the flesh sugar content trait and tested \nprediction accuracy using datasets of non -overlapping accessions from three phenotyping \nexperiments independent from training: historical records from the USDA GRIN database, a 2019 \nfield trial in Yanqing, China 6, and a 2022 field trial in Hainan, China conducted in this study. \nWithout retraining, the SNP+SV -based genomic selection model achieved consistently high \naccuracies across these datasets, ranging from 0.89 to 0.93 ( Fig. 5b ). To simulate real -world \nbreeding applications, we further evaluated the model in a recombinant inbred line (RIL) \npopulation derived from a cross between the sweet cultivar ‘97103’ and the non-sweet C. amarus \naccession ‘PI 296341 -FR’. The F 1 hybrids, same as ‘PI 296341 -FR’, exhibited the non-sweet \nphenotype, indicating epistatic suppression of sweetness —a challenge for genomic prediction. \nDespite this, the SNP+SV model maintained a moderate prediction accuracy of 0.53 (Fig. 5c). \nCollectively, these results demonstrate that our compact marker panels (~500 –700 high-\neffect variants) enable stable genomic predictions in watermelon. The inclusion of SVs provides \nlimited additional predictive power, likely because most SVs are effectively linked to nearby SNPs. \n \nDiscussion \nIn this study, we assembled 135 high -quality, near -gapless genomes representing all extant \nCitrullus species and constructed a population-scale graph-based super-pangenome. This enabled \naccurate detection and genotyping of nearly one million SVs across 914 accessions , substantially \nexpanding upon previous genomic and pangenomic resources of watermelon2,4,6. The breadth and \nquality of these genomic resources allowed us to re-examine the origin of cultivated watermelon, \nproviding strong evidence that C. lanatus subsp. cordophanus is the direct wild progenitor. In \ncontrast to the findings of the previous pangenomics study4, our broader sampling suggests that C. \nmucosospermus is unlikely to be an additional direct ancestor of cultivated watermelon. Our \nfindings highlight the importance of extensive intraspecific sampling for accurately reconstructing \ndomestication history and capturing genetic diversity , providing valuable insights for leveraging \nwild Citrullus species in watermelon resistance breeding. \nFlesh color is a prominent quality trait in watermelon. While previous studies have \nprimarily focused on identifying loci that control discrete color categories44–50, the genetic basis of \nflesh color intensity —a quantitative trait relevant to both breeding objectives and consumer \npreferences—has remained poorly understood. Here, leveraging accurate SV genotyping enabled \nby the graph-based super-pangenome, we conducted GWAS and identified a copy number variant \n(CNV) involving a 1 ,258-bp sequence in the promo ter of ClFCI1 that modulates flesh color \nintensity. This CNV , present in one to four tandem copies, was strongly associated with ClFCI1 \nexpression and, consequently, with flesh color intensity and carotenoid accumulation. Notably, the \nfrequency of multi -copy alleles increased during watermelon domestication and improvement, \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n14 \n \nconsistent with breeding preferences for brighter flesh colors. Transcriptomic analyses of ClFCI1 \nknockdown and overexpression lines revealed that ClFCI1 regulates flesh color intensity primarily \nby modulating genes involved in photosynthesis and plastid development. These findings \nidentified a novel regulatory variant of both functional and breeding relevance and demonstrate \nthe power of SV-informed pangenomics in uncovering the genetic basis of complex trait variation. \nThe comprehensive variation map also enabled genomic prediction across a broad range of \nfruit-quality and disease -resistance traits. For most traits, compact marker panels composed of \nseveral hundred high -effect variants achieved high predictive accuracy. Notably, the model for \nflesh sugar content performed consistently well across independent datasets derived from both \nnatural accessions and breeding populations. Although the inclusion of SVs yielded only modest \noverall gains in predictive power due to man y causative SVs being in strong linkage with nearby \nSNPs, it improved the capture of variation poorly tagged by SNPs alone, as demonstrated by \nenhanced prediction accuracy for traits such as flesh color category.  \nTogether, our results demonstrate how population -scale, SV-resolved pangenomics can \nprovide both evolutionary insights and practical tools for trait -variant association and crop \nimprovement. As watermelon breeding advances toward greater precision and effi ciency, \nintegrating pangenomic resources with genomics -assisted selection and genome editing will be \nessential for harnessing wild genetic diversity and accelerating the development of cultivars with \nimproved fruit quality, disease resistance, and environmental adaptability. \n \nMethods \nPlant materials and phenotyping \nCultivated and wild watermelon accessions were obtained from Beijing Vegetable Research Center \n(BVRC) and the U.S. National Plant Germplasm System  (NPGS). For phenotyping, accessions \nwere planted in triplicate at the Hainan Experiment Station of BVRC (18° 27′ N, 108° 57′E) in \n2022. One fruit per plant was harvested 34 days after pollination, with three biological replicates \nper accession. Each fruit was cu t longitudinally, photographed, and sampled to determine flesh \nsugar content, flesh color intensity, rind firmness, fruit length, and fruit width. Flesh sugar content \nwas measured at the center of the flesh in °Brix using a hand -held digital PAL-1 refractometer \n(Atago, Bellevue, WA, USA ). Flesh color intensity  was assessed from fruit  images with a \ncolorimeter (Minolta CR -400, Tokyo, Japan) to measure CIE L*, a*, b*, C* (chroma) and h* \nvalues. Rind firmness was measured at the equatorial region of each fruit using a hand -held fruit \nsclerometer with a 3.0 mm diameter tip (FR-5120, Lutron Electronic Enterprise Co., Ltd., Taiwan), \nand flesh firmness was measured at the center flesh. The fruit shape index was calculated as the \nratio of fruit length to fruit width. \n \nGenome sequencing and assembly \nHigh-molecular-weight genomic DNA was extracted from young fresh leaves using the \ncetyltrimethylammonium bromide (CTAB) method51. SMRTbell libraries were prepared using the \nSMRTbell Express Template Prep Kit 2.0 (PacBio) and sequenced on the PacBio Sequel II \nplatform in circular consensus sequencing (ccs) mode to generate high -fidelity (HiFi) reads. For \nthe selected ten accessions, ONT and Hi-C sequencing libraries were constructed according to the \nmanufacturers’ instructions and sequenced on the Oxford Nan opore PromethION and Illumina \nNovaSeq platforms, respectively. HiFi reads for each accession were processed using \nHiFiAdapterFilt52 to remove adapter sequences. The cleaned HiFi reads, together with ONT and \nHi-C data when available, were assembled into contigs using hifiasm 53. Haplotypic duplications \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n15 \n \nwere removed using the purge_dups package 54. Potential contaminant sequences from \nmicroorganisms and organelle genomes were identified and removed by comparing the contig \nsequences against the NCBI nt/nr database55. For accessions with Hi-C data, pseudochromosomes \nwere constructed using the 3D-DNA software56. For the remaining accessions, chromosome-level \nassemblies were generated using RagTag57, guided by previously published reference genomes2,6. \nBUSCO completeness of the assembled genomes was estimated using the \nembryophyta_odb10 database15. Base accuracy was assessed using Merqury 16. The presence of \ntelomeres was determined by quarTeT58 and alignments to telomere-related repeat unit59. \n \nGenome annotation \nRepeat sequences were predicted in each assembly using the EDTA pipeline 60. These repeat \nsequences, along with previously generated custom repeat libraries for different Citrullus species2, \nwere combined and processed to remove redundancy using the cleanup_nested.pl script from the \nEDTA package. The resulting non-redundant Citrullus repeat library was then used to mask repeats \nin the assembled genomes. Protein -coding genes were predicted in each genome using the \nMAKER pipeline61, incorporating evidence from  transcript mapping, protein homology, and ab \ninitio gene prediction s. To prepare transcript evidence, de novo  transcript assemblies were \ngenerated with Trinity62 for each species using RNA-seq data from various tissues obtained from \nNCBI SRA database (Supplementary Table 18). Furthermore, PacBio Iso-Seq full length cDNA \nsequences from our previous study6 and coding sequences of protein-coding genes from published \nwatermelon genomes, were also incorporated as transcript evidence. Proteome sequences from \ncucumber63, melon64, pumpkin65, chayote66, snake gourd67, wax gourd68, and Arabidopsis69, as well \nas proteins from the Swiss -Prot database 70 were used as protein homology evidence.  \nAUGUSTUS71 and SNAP72 were used for ab initio gene predictions. \nTo further improve gene predictions across previously published and newly developed \nwatermelon genomes, predicted genes from each genome were mapped to other genomes using \nLiftoff73. For each genome, coding sequences from all other genomes were projected onto it \nrequiring at least 90% sequence identity and coverage. The best gene models were then selected \nfrom the original and projected sets using EVidenceModeler74. For functional annotation, protein \nsequences of predicted genes in each genome were aligned against the Swiss-Prot, TrEMBL, and \nTAIR10 databases using DIAMOND 75, followed by assigning human readable functional \ndescriptions using AHRD ( https://github.com/groupschoof/AHRD). The Blast2GO suite 76 was \nutilized for GO annotation and functional enrichment analysis. Putative nucleotide -binding site \n(NB-ARC) domain-containing genes were identified using the RGAugury pipeline77 (v 2.2). \n \nChromosomal rearrangement and gene flow detection \nEach genome assembly was aligned to the ‘97103’ reference genome using AnchorWave 78 to \nidentify collinear blocks. Additionally, gene -based syntenic blocks were detected using the R \npackage GENESPACE79. Syntenic blocks identified by both programs were used to infer inter -\nchromosomal translocations and large inversions (≥ 100 kb), followed by manual inspection based \non HiFi read and genome alignments. \nGene flow from C. mucosospermus  to cultivated watermelon was detected using a \ncomposite-likelihood approach implemented in TreeMix 80, with C. colocynthis  used as the \noutgroup. Subsequently, genomic regions in cultivated watermelon that were introgressed from C. \nmucosospermus were identified using the ABBA -BABA test (D -statistic), as previously \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n16 \n \ndescribed31. Specifically, non-overlapping 100-kb windows across the genome with the top 5% fd \nvalues (degree of introgression) were defined as introgressed regions. \n \nGene-based super-pangenome construction and phylogenetic analysis \nGene families across the 138 high-quality Citrullus genomes were identified using OrthoFinder81 \n(v2.5.5). Synteny information was then used to separate paralogous genes that were not located \nwithin the same syntenic regions. Syntenic gene blocks between each pair of genomes were \ndetected using MCScanX82, and the synonymous substitution rates ( Ks) between syntenic genes \nwere calculated using the Yang-Nielsen algorithm implemented in the PAML package83. Syntenic \northologous gene blocks were used to refine the OrthoFinder-defined orthologous groups, dividing \nthem into syntenic orthologous gene families and singletons. Based on their presence across the \n138 genomes, syntenic orthologous gene families were classified into four categories: core (present \nin all 138 genomes), softcore (present in 137 genomes), shell (present in 2 -136 genomes), and \nprivate (present in only one genome). \nTo reconstruct the species tree, single-copy orthologous genes (SCOs) were identified by \nclustering predicted proteins from 13 species, including seven Citrullus species, as well as bottle \ngourd84, cucumber63, melon64, snake gourd67, bitter gourd85, and walnut86. Protein-guided multiple \ncoding sequence alignments of SCOs were obtained using TranslatorX87. Divergence times among \nspecies were estimated using BEAST2 with the ‘Optimised Relaxed Clock’ model, calibrated with \nknown divergence times for Fagales -Cucurbitales (85.6-109 million years ago [Mya]), Cucumis-\nCitrullus (16.4-24.2 Mya), and cucumber -melon (5.96-13.1 Mya)88. The time of domestication, \nrepresented by the divergence between C. lanatus subsp. cordophanus and cultivated watermelon, \nwas estimated using the SMC++ program89. \n \nGenome resequencing of watermelon core accessions \nA watermelon core collection comprising 323 representative wild and domesticated accessions \nwas constructed from ~1,400 accessions using GenoCore 90, based on SNPs derived from our \npreviously reported GBS data91, eight of which were also included in de novo genome assemblies. \nAdditionally, accessions harboring important breeding traits were also included in the core \ncollection. Genomic DNA was extracted from young leaf samples of the core accessions using the \nQiagen DNeasy Plant Kit. Shotgun DNA libraries we re constructed from the extracted DNA and \nsequenced on the BGISEQ -500 platform to generate 150 -bp paired -end reads. Genomic \nresequencing data from an additional 463 watermelon accessions were retrieved from previous \nstudies2,6,91. Raw sequencing reads were processed to remove adapter sequences and low -quality \nbases using Trimmomatic (v0.38)92. \n \nGraph pangenome construction and SV genotyping \nEach of the 137 genomes was aligned to the ‘97103’ reference genome using AnchorWave 78. \nStructural variants (SVs), including large insertions, deletions, and inversions, as well as SNPs and \nsmall indels, were identified based on the alignments using NucDiff 93. Deletions longer than 10 \nkb were further validated based on HiFi read coverage. SVs identified from the 137 accessions \nwere merged using bcftools 94 (v1.14), and redundant variants were removed using the script \nfindDup.R ( https://github.com/vgteam/giraffe-sv-paper/tree/master/scripts/sv/remap-to-dedup-\nmerged-svs). A pangenome graph was constructed using PanGenie 95 (v3.0.1) from the identified \nSVs, SNPs, and small indels, with the ‘97103’ genome used as the reference. SVs in the graph \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n17 \n \npangenome were then genotyped in the resequenced accessions using PanGenie with the cleaned \nresequencing reads. \n \nGenome-wide association studies \nSNPs in all resequenced Citrullus accessions were called using the Sentieon package \n(https://www.sentieon.com/), followed by hard filtering with recommended parameters 96. A total \nof 20,210,332 bi-allelic SNPs with both missing rates and heterozygous rates below 20% were \nretained. These SNPs were then combined with all identified SVs from the graph pangenome and \nused for GWAS. Five accessions from C. naudinianus, C. rehmii, and C. ecirrhosus were excluded \nbecause of small sample sizes and lack of phenotypic data. For each trait, variants with a high \nmissing rate (>30%) or low minor allele counts (<10) among accessions with phenotypic data were \nremoved. To account for population structure, a kinship matrix was generated using FaST-LMM97 \n(v2.07), and GWAS was performed using the linear mixed model implemented in FaST -LMM. \nGenome-wide significance thresholds were determined by calculating the effective number of \nindependent variants using the Genetic type 1 Error Calculator98 (GEC v0.2). \n \nGenomic prediction \nTo generate marker panels for genomic prediction of each trait, we applied CropGBM 43 (v1.1.2) \nto perform feature selection using SNP-only and SNP+SV variant sets. Prior to feature selection, \nvariants with a minor allele frequency (MAF) < 0.05 or missing rate greater than 20% were \nexcluded to retain high -confidence variants. Additionally, linkage disequilibrium (LD) pruning \nwas performed to remove highly correlated variants, using an r2 threshold of 0.999. The resulting \nmarker panels were used to generate genomic prediction models with CropGBM. Five-fold cross-\nvalidation was used with five repeats to evaluate model performance. \n \nGenetic mapping of flesh color intensity trait \nTo map loci controlling flesh color intensity, two independent F2 populations were developed from \ncrosses of cultivars ‘Ming 58’ (scarlet red flesh) × ‘JX2’ (pink flesh) and ‘JLM’ (yellow flesh) × \n‘Cream of Saskatchewan’ (pale yellow flesh; hereafter ‘CS’). Flesh color intensity was determined \nat 34 days after pollinati on (DAP). For bulk segregant analysis (BSA), two DNA pools were \nconstructed from the ‘Ming 58’ and ‘JX2’ F2 population: one comprising 20 individuals with the \nlowest chroma values and the oth er comprising 20 individuals with the highest chroma values. \nGenomic DNA was extracted from each individual using the CTAB method, mixed in equal \namount within each pool, and subjected to library construction and whole-genome sequencing on \nthe Illumina HiS eq platform. Variant calling was performed using the Sentieon package \n(https://www.sentieon.com/), followed by hard filtering with recommended parameters 96. BSA \nwas performed using the R package QTLseqr 99 to identify QTLs of flesh color intensity . To \nperform fine mapping of the candidate region, larger F 2 segregating populations were generated, \nconsisting of 636 individuals from the cross of ‘JLM’ × ‘CS’ and 141 individuals from the cross \nof ‘Ming 58’ × ‘JX2’. Based on SNPs identified between parental lines, Kompetitive Allele \nSpecific PCR (KASP) markers (Supplementary Table 19) were developed. The linkage map was \nconstructed from KASP markers in each population using QTL IciMapping100 (v4.2). \n \nQuantitative RT-PCR and transient dual-luciferase activity assay \nTotal RNA was extracted using the Quick RNA isolation kit (Huayueyang Biotechnologies Co., \nLtd.). First-strand cDNA was synthesized from 1 ug of total RNA using SuperScriptTM III Reverse \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n18 \n \nTranscriptase (Invitrogen , Carlsbad, CA, USA ) with oligo(dT)18 primers, according to the \nmanufacturer’s instructions. Gene expression levels were quantified using SYBR Green -based \nqPCR on a Roche LightCycler® 480 system  (Roche, Basel, Switzerland) . Three biological \nreplicates were performed for each gene , with watermelon ClActin1 gene used as the internal \nreference. Relative expression was calculated using the 2−ΔΔCt method after normalization against \nClActin1. \nTo compare the promoter activity associated with different copy numbers of the 1 ,258-bp \nsequence upstream of ClFCI1, one, two, and three copies of this sequence were PCR -amplified \nfrom ‘CS’ and ‘JLM’ using primers FCI -SV-PstI/BamHI (Supplementary Table 2 0). The \nresulting PCR fragments were cloned into the PstI and BamHI sites of the pGreenII 0800‐LUC \nvector. The constructs were then transformed into watermelon fruit protoplasts following the \nmethod described previously101. Luciferase activity was measured u sing the Dual -Luciferase \nReporter Assay Kit following the manufacturer’s instructions (Vazyme Biotech, China). Ten \nbiological replicates were performed for each construct. \n \nAgrobacterium-mediated transformation \nThe full -length cDNA and a partial cDNA fragment of ClFCI1 were amplified using FCI -\nPacI/AscI and FCI -AscI/PacI primers, respectively (Supplementary Table 20 ). These PCR \nproducts were cloned into the PacI/AscI sites of pMDC85  (ref. 102) to generate ClFCI1 \noverexpression and  knockdown constructs, respectively. The resulting constructs were then \nintroduced into  Agrobacterium tumefaciens strain C58/ATCC 33970. Plant transformation was \nperformed as previously described101. Transgene insertion in the transformed watermelon lines  \nwas confirmed by PCR using the AS013 PAT/bar Kit (Envirologix Inc., Portland, ME, USA). \n \nCharacterization of ClFCI1 transgenic lines \nCarotenoids were extracted from mature fruit flesh (5 g; 34 DAP) of ClFCI1 overexpression and \nknockdown lines using a hexane:acetone:ethanol (50:25:25, v/v/v) mixture. Carotenoid \ncomposition and content were then determined using a Nexera HPLC system (Shimadzu). \nFor transcriptome analyses, total RNA was extracted from the fruit flesh of ClFCI1 \nknockdown and overexpression lines, as well as the wild -type line (‘ZZJM’). RNA-Seq libraries \nwere constructed from total RNA using the TruSeqTM RNA Sample Prep Kit (Illumina, USA) and \nsequenced on the Illumina HiSeq 4000 platform to generate paired -end 150 -bp reads. Three \nbiological replicates were conducted for each sample. Raw RNA -Seq reads were cleaned using \nTrimmomatic92, and the cleaned reads were aligned to the ‘97103’ reference genome using \nHISAT2 (ref. 103). Raw counts for each protein-coding gene were calculated using featureCounts104 \nand then normalized to transcripts per million (TPM). Differentially expressed genes (DEGs) were \nidentified using DESeq2 (ref. 105) by comparing ClFCI1-knockdown and ClFCI1-overexpression \nlines to wild -type fruits. The Benjamini -Hochberg method 106 was used to control the false \ndiscovery rate (FDR ). Genes with FDR < 0.05 and |log₂(fold change)|  ≥ 1 were considered \nsignificantly differentially expressed. \n \nData availability \nRaw HiFi, ONT , Hi-C, and genome resequencing  reads have been deposited in the NCBI \nBioProject database under accession number PRJNA1272048. Genome assemblies and \nannotations, and variant files  in VCF format are available at CuGenDBv2 \n(http://cucurbitgenomics.org/v2/ftp/pan-genome/watermelon/graph_pangenome/). \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n19 \n \n \nAuthor contributions \nZ.F. and Y .X. designed and supervised the project. S.G., S.A.H., C.M., R.J., S.E.B., P.W., C.K., \nA.L. and R.G. contributed to sample collection and DNA extraction. S.G., H.S., S.Liao, J.Zhang, \nR.J. and Z.F. coordinated genome sequencing. S.G., S.Liao, J.Zhang, G.G., J.W., Y .Y ., Y .R., S.T., \nS.Li and H.Z. performed phenotyping for fruit-quality traits. H.S., Z.Z., X.Z. and S.W. contributed \nto genome assembly and annotation, as well as pangenome and population genetic analyses. H.S. \nand J.Zhao conducted the  genomic prediction analysis. J.Zhang, H.S. and S.Liao contributed to \ngenetic mapping and gene functional characterization. H.S., Z.Z., J.Zhang., X.Z, and S.W. wrote \nthe manuscript. Z.F. and Y .X. revised the manuscript. \n \nConflict of interest \nThe authors declare no conflict of interest. \n \nAcknowledgements \nWe thank Susanne S. Renner for providing seeds of C. lanatus subsp. cordophanus. This research \nwas supported by grants from Beijing Rural Revitalization Agricultural Science and Technology \nProject (NY2401130025), National Natural Science Foundation of China (Grant No. 32172592, \n32330093), the Scientific and Technological Innovation Capacity Building Project of BAAFS \n(KJCX20251008), the Scientist Training Program of BAAFS (JKZX202401),  Ministry of \nAgriculture of China (CARS-25), USDA National Institute of Food and Agriculture Specialty Crop \nResearch Initiative (2015-51181-24285 and 2020-51181-32139).  \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n20 \n \nReferences \n1. Renner, S. S. How Russian breeders discovered Citrullus mucosospermus and Citrullus lanatus var. \ncordophanus, the likely closest relatives of domesticated watermelon. Rep. Cucurbit Genet. Coop. \n42, 10–12 (2019). \n2. Wu, S. et al. A Citrullus genus super‐pangenome reveals extensive variations in wild and cultivated \nwatermelons and sheds light on watermelon evolution and domestication. Plant Biotechnol. J. 21, \n1926–1928 (2023). \n3. Paris, H. S. Origin and emergence of the sweet dessert watermelon, Citrullus lanatus. Ann. Bot. 116, \n133–148 (2015). \n4. Zhang, Y . et al. Telomere-to-telomere Citrullus super-pangenome provides direction for watermelon \nbreeding. Nat. Genet. 56, 1750–1761 (2024). \n5. Levi, A. et al. Genetic resources of watermelon. In Genetics and Genomics of Cucurbitaceae (eds. \nGrumet, R., Katzir, N. & Garcia-Mas, J.) 87–110 (Springer International Publishing, Cham, 2017). \n6. Guo, S. et al. Resequencing of 414 cultivated and wild watermelon accessions identifies selection \nfor fruit quality traits. Nat. Genet. 51, 1616–1623 (2019). \n7. Renner, S. S., Sousa, A. & Chomicki, G. Chromosome numbers, Sudanese wild forms, and \nclassification of the watermelon genus Citrullus, with 50 names allocated to seven biological species. \nTaxon 66, 1393–1405 (2017). \n8. Zhang, L. et al. An update on genomics and molecular breeding in watermelon. Acta Hortic. 1411, \n305–318 (2024). \n9. Zhou, Y . et al. Graph pangenome captures missing heritability and empowers tomato breeding. \nNature 606, 527–534 (2022). \n10. Groza, C. et al. Pangenome graphs improve the analysis of structural variants in rare genetic diseases. \nNat. Commun. 15, 657 (2024). \n11. Khan, A. W. et al. Super-pangenome by integrating the wild side of a species for accelerated crop \nimprovement. Trends Plant. Sci. 25, 148–158 (2020). \n12. Yang, J. et al. Genetic relationship and pedigree of Chinese watermelon varieties based on diversity \nof perfect SNPs. Hortic. Plant J. 8, 489–498 (2022). \n13. Zhang, H. et al. Genetic diversity, population structure, and formation of a core collection of 1197 \nCitrullus accessions. HortScience 51, 23–29 (2016). \n14. Igolkina, A. A. et al. Towards an unbiased characterization of genetic polymorphism: a comparison \nof 27 A. thaliana genomes. Preprint at https://doi.org/10.1101/2024.05.30.596703 (2024). \n15. Manni, M., Berkeley, M. R., Seppey, M., Simão, F. A. & Zdobnov, E. M. BUSCO update: novel and \nstreamlined workflows along with broader and deeper phylogenetic coverage for scoring of \neukaryotic, prokaryotic, and viral genomes. Mol. Biol. Evol. 38, 4647–4654 (2021). \n16. Rhie, A., Walenz, B. P., Koren, S. & Phillippy, A. M. Merqury: reference-free quality, completeness, \nand phasing assessment for genome assemblies. Genome Biol. 21, 245 (2020). \n17. Levin, D. A. The Role of Chromosomal Change in Plant Evolution . (Oxford Univ. Press, Oxford, \n2002). \n18. Wellenreuther, M. & Bernatchez, L. Eco-evolutionary genomics of chromosomal inversions. Trends \nEcol. Evol. 33, 427–440 (2018). \n19. Deng, Y . et al. A telomere-to-telomere gap-free reference genome of watermelon and its mutation \nlibrary provide important resources for gene discovery and breeding. Mol. Plant. 15, 1268–1284 \n(2022). \n20. Renner, S. S. et al. A chromosome -level genome of a Kordofan melon illuminates the origin of \ndomesticated watermelons. Proc. Natl. Acad. Sci. U S A 118, e2101486118 (2021). \n21. Zhang, J. et al. A unique chromosome translocation disrupting ClWIP1 leads to gynoecy  in \nwatermelon. Plant J. 101, 265–277 (2020). \n22. Jiao, D. et al. Identification of allelic relationship and translocation region among chromosomal \ntranslocation lines that leads to less-seed watermelon. Hortic. Res. 11, uhae087 (2024). \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n21 \n \n23. Branham, S. E. et al. QTL mapping of resistance to Fusarium oxysporum f. sp. niveum race 2 and \nPapaya ringspot virus in Citrullus amarus. Theor. Appl. Genet. 133, 677–687 (2020). \n24. Gao, L. Transcriptome analysis and fine mapping of major genes controlling flesh firmness and sour \nflesh in watermelon. Ph.D. Thesis. Chinese Academy of Agricultural Sciences (2018). \n25. Ren, Y . et al. An integrated genetic map based on four mapping populations and quantitative trait \nloci associated with economically important traits in watermelon ( Citrullus lanatus). BMC Plant \nBiol. 14, 33 (2014). \n26. Maples, B. K., Gravel, S., Kenny, E. E. & Bustamante, C. D. RFMix: a discriminative modeling \napproach for rapid and robust local-ancestry inference. Am. J. Hum. Genet. 93, 278–288 (2013). \n27. Connallon, T. & Olito, C. Natural selection and the distribution of chromosomal inversion lengths. \nMol. Ecol. 31, 3627–3641 (2022). \n28. Schaefer, H., Heibl, C. & Renner, S. S. Gourds afloat: a dated phylogeny reveals an Asian origin of \nthe gourd family (Cucurbitaceae) and numerous oversea dispersal events. Proc. R. Soc. B: Biol. Sci. \n276, 843–851 (2009). \n29. Meyer, R. S. & Purugganan, M. D. Evolution of crop species: genetics of domestication and \ndiversification. Nat. Rev. Genet. 14, 840–852 (2013). \n30. Pérez-Escobar, O. A. et al. Genome sequencing of up to 6,000-year-old Citrullus seeds reveals use \nof a bitter-fleshed species prior to watermelon domestication. Mol. Biol. Evol. 39, msac168 (2022). \n31. Martin, S. H., Davey, J. W. & Jiggins, C. D. Evaluating the use of ABBA–BABA statistics to locate \nintrogressed loci. Mol. Biol. Evol. 32, 244–257 (2015). \n32. Andolfo, G., Sánchez, C. S., Cañizares, J., Pico, M. B. & Ercolano, M. R. Large -scale gene gains \nand losses molded the NLR defense arsenal during the Cucurbita evolution. Planta 254, 82 (2021). \n33. Schemberger, M. O. et al. Transcriptome profiling of non -climacteric ‘yellow’ melon during \nripening: insights on sugar metabolism. BMC Genomics 21, 262 (2020). \n34. Dou, J. L. et al. Genetic mapping reveals a candidate gene ( ClFS1) for fruit shape in watermelon \n(Citrullus lanatus L.). Theor. Appl. Genet. 131, 947–958 (2018). \n35. Katuuramu, D. N., Levi, A. & Wechter, W. P. Genome‐wide association study of soluble solids \ncontent, flesh color, and fruit shape in citron watermelon. Plant Genome 16, e20391 (2023). \n36. Wang, J. et al. A natural variant of NON-RIPENING promotes fruit ripening in watermelon. Plant \nCell 37, koae313 (2024). \n37. Yin, J. et al. A single amino acid substitution in the AAA -type ATPase LRD6-6 activates immune \nresponses but decreases grain quality in rice. Front. Plant Sci. 15, 1451897 (2024). \n38. Zhang, B. et al. The mitochondrial outer membrane AAA ATPase AtOM66 affects cell death and \npathogen resistance in Arabidopsis thaliana. Plant J. 80, 709–727 (2014). \n39. Waldo, B. D., Branham, S. E., Levi, A., Patrick Wechter, W. & Rutter, W. B. Distinct genomic loci \nunderlie quantitative resistance to Meloidogyne enterolobii galling and reproduction in Citrullus \namarus. Plant Dis. 107, 2126–2132 (2023). \n40. Sun, T. et al. ChIP -seq reveals broad roles of SARD1 and CBP60g in regulating plant immunity. \nNat. Commun. 6, 10159 (2015). \n41. Larkin, R. M. et al. REDUCED CHLOROPLAST COVERAGE  genes from Arabidopsis thaliana \nhelp to establish the size of the chloroplast compartment. Proc. Natl. Acad. Sci. U S A 113, E1116–\nE1125 (2016). \n42. Crossa, J. et al. Genomic selection in plant breeding: methods, models, and perspectives. Trends \nPlant Sci. 22, 961–975 (2017). \n43. Yan, J. et al. LightGBM: accelerated genomically designed crop breeding through ensemble learning. \nGenome Biol. 22, 271 (2021).  \n44. Liu, S. et al. Nucleotide variation in the phytoene synthase (ClPsy1) gene contributes to golden flesh \nin watermelon (Citrullus lanatus L.). Theor. Appl. Genet. 135, 185–200 (2022). \n45. Branham, S. et al. Genetic mapping of a major codominant QTL associated with β-carotene \naccumulation in watermelon. Mol. Breed. 37, 146 (2017). \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n22 \n \n46. Nie, H. et al. High -quality genome assembly and genetic mapping reveal a gene regulating flesh \ncolor in watermelon (Citrullus lanatus). Front. Plant Sci. 14, 1142856 (2023). \n47. Zhang, J. et al. Decreased protein abundance of lycopene β-cyclase contributes to red flesh in \ndomesticated watermelon. Plant Physiol. 183, 1171–1183 (2020). \n48. Fang, X. F. et al. Clorf encodes carotenoid isomerase and regulates orange flesh color in watermelon \n(Citrullus lanatus L.). J. Agric. Food Chem. 71, 15445–15455 (2023). \n49. Liu, S. et al. Identification of chromosome region and candidate genes for canary-yellow flesh (Cyf) \nlocus in watermelon (Citrullus lanatus). Plant Sci. 329, 111594 (2023). \n50. Zhang, W. et al. Clpf encodes pentatricopeptide repeat protein (PPR5) and regulates pink flesh color \nin watermelon (Citrullus lanatus L.). Theor. Appl. Genet. 137, 126 (2024). \n51. Doyle, J. J. & Doyle, J. L. A rapid DNA isolation procedure for small quantities of fresh leaf tissue. \nPhytochem. Bull. 19, 11–15 (1987). \n52. Sim, S. B., Corpuz, R. L., Simmonds, T. J. & Geib, S. M. HiFiAdapterFilt, a memory efficient read \nprocessing pipeline, prevents occurrence of adapter sequence in PacBio HiFi reads and their \nnegative impacts on genome assembly. BMC Genomics 23, 157 (2022). \n53. Cheng, H., Concepcion, G. T., Feng, X., Zhang, H. & Li, H. Haplotype-resolved de novo assembly \nusing phased assembly graphs with hifiasm. Nat. Methods 18, 170–175 (2021). \n54. Guan, D. et al. Identifying and removing haplotypic duplication in primary genome assemblies. \nBioinformatics 36, 2896–2898 (2020). \n55. Camacho, C. et al. BLAST+: architecture and applications. BMC Bioinform. 10, 421 (2009). \n56. Dudchenko, O. et al. De novo assembly of the Aedes aegypti  genome using Hi -C yields \nchromosome-length scaffolds. Science (1979) 356, 92–95 (2017). \n57. Alonge, M. et al. Automated assembly scaffolding using RagTag elevates a new tomato system for \nhigh-throughput genome editing. Genome Biol. 23, 258 (2022). \n58. Lin, Y . et al. quarTeT: a telomere-to-telomere toolkit for gap-free genome assembly and centromeric \nrepeat identification. Hortic. Res. 10, uhad127 (2023). \n59. Guo, S. G. et al. The draft genome of watermelon (Citrullus lanatus) and resequencing of 20 diverse \naccessions. Nat. Genet. 45, 51–58 (2013). \n60. Ou, S. et al. Benchmarking transposable element annotation methods for creation of a streamlined, \ncomprehensive pipeline. Genome Biol. 20, 275 (2019). \n61. Cantarel, B. L. et al. MAKER: an easy -to-use annotation pipeline designed for emerging model \norganism genomes. Genome Res. 18, 188–196 (2008). \n62. Grabherr, M. G. et al. Full-length transcriptome assembly from RNA-seq data without a reference \ngenome. Nat. Biotechnol. 29, 644–652 (2011). \n63. Li, Q. et al. A chromosome-scale genome assembly of cucumber (Cucumis sativus L.). Gigascience \n8, giz072 (2019). \n64. Castanera, R., Ruggieri, V ., Pujol, M., Garcia -Mas, J. & Casacuberta, J. M. An improved melon \nreference genome with single-molecule sequencing uncovers a recent burst of transposable elements \nwith potential impact on genes. Front. Plant Sci. 10, 1815 (2020). \n65. Sun, H. et al. Karyotype stability and unbiased fractionation in the paleo -allotetraploid Cucurbita \ngenomes. Mol. Plant. 10, 1293–1306 (2017). \n66. Fu, A. et al. Combined genomic, transcriptomic, and metabolomic analyses provide insights into \nchayote (Sechium edule) evolution and fruit development. Hortic. Res. 8, 35 (2021). \n67. Ma, L. et al. The genome and transcriptome analysis of snake gourd provide insights into its \nevolution and fruit development and ripening. Hortic. Res. 7, 199 (2020). \n68. Xie, D. et al. The wax gourd genomes offer insights into the genetic diversity and ancestral cucurbit \nkaryotype. Nat. Commun. 10, 5158 (2019). \n69. Cheng, C. Y . et al. Araport11: a complete reannotation of the Arabidopsis thaliana reference genome. \nPlant J. 89, 789–804 (2017). \n70. The UniProt Consortium. UniProt: the universal protein knowledgebase in 2021. Nucleic Acids Res. \n49, D480–D489 (2021). \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n23 \n \n71. Stanke, M., Tzvetkova, A. & Morgenstern, B. AUGUSTUS at EGASP: using EST, protein and \ngenomic alignments for improved gene prediction in the human genome. Genome Biol . 7, S11 \n(2006). \n72. Korf, I. Gene finding in novel genomes. BMC Bioinform. 5, 59 (2004). \n73. Shumate, A. & Salzberg, S. L. Liftoff: accurate mapping of gene annotations. Bioinformatics 37, \n1639–1643 (2021). \n74. Haas, B. J. et al. Automated eukaryotic gene structure annotation using EVidenceModeler and the \nprogram to assemble spliced alignments. Genome Biol. 9, R7 (2008). \n75. Buchfink, B., Reuter, K. & Drost, H. -G. Sensitive protein alignments at tree -of-life scale using \nDIAMOND. Nat. Methods 18, 366–368 (2021). \n76. Gotz, S. et al. High -throughput functional annotation and data mining with the Blast2GO suite. \nNucleic Acids Res. 36, 3420–3435 (2008). \n77. Li, P. et al. RGAugury: a pipeline for genome-wide prediction of resistance gene analogs (RGAs) in \nplants. BMC Genomics 17, 852 (2016). \n78. Song, B. et al. AnchorWave: sensitive alignment of genomes with high sequence diversity, extensive \nstructural polymorphism, and whole -genome duplication. Proc. Nat. Acad. Sci. U S A  119, \ne2113075119 (2022). \n79. Lovell, J. T. et al. GENESPACE tracks regions of interest and gene copy number variation across \nmultiple genomes. eLife 11, e78526 (2022). \n80. Pickrell, J. K. & Pritchard, J. K. Inference of population splits and mixtures from genome -wide \nallele frequency data. PLoS Genet. 8, e1002967 (2012). \n81. Emms, D. M. & Kelly, S. OrthoFinder: phylogenetic orthology inference for comparative genomics. \nGenome Biol. 20, 238 (2019). \n82. Wang, Y. et al. MCScanX: a toolkit for detection and evolutionary analysis of gene synteny and \ncollinearity. Nucleic Acids Res. 40, e49 (2012). \n83. Yang, Z. PAML 4: phylogenetic analysis by maximum likelihood. Mol. Biol. Evol. 24, 1586–1591 \n(2007). \n84. Wu, S. et al. The bottle gourd genome provides insights into Cucurbitaceae evolution and facilitates \nmapping of a Papaya ring‐spot virus resistance locus. Plant J. 92, 963–975 (2017). \n85. Matsumura, H. et al. Long -read bitter gourd ( Momordica charantia) genome and the genomic \narchitecture of nonclassic domestication. Proc. Nat. Acad. Sci. U S A 117, 14543–14551 (2020). \n86. Marrano, A. et al. High -quality chromosome -scale assembly of the walnut ( Juglans regia  L.) \nreference genome. Gigascience 9, giaa050 (2020). \n87. Abascal, F., Zardoya, R. & Telford, M. J. TranslatorX: multiple alignment of nucleotide sequences \nguided by amino acid translations. Nucleic Acids Res. 38, W7–W13 (2010). \n88. Bouckaert, R. et al. BEAST 2: a software platform for Bayesian evolutionary analysis. PLoS Comput. \nBiol. 10, e1003537 (2014). \n89. Terhorst, J., Kamm, J. A. & Song, Y. S. Robust and scalable inference of population history from \nhundreds of unphased whole genomes. Nat. Genet. 49, 303–309 (2017). \n90. Jeong, S. et al. GenoCore: a simple and fast algorithm for core subset selection from large genotype \ndatasets. PLoS ONE 12, e0181420 (2017). \n91. Wu, S. et al. Genome of ‘Charleston Gray’, the principal American watermelon cultivar, and genetic \ncharacterization of 1,365 accessions in the U.S. National Plant Germplasm System watermelon \ncollection. Plant Biotechnol. J. 17, 2246–2258 (2019). \n92. Bolger, A. M., Lohse, M. & Usadel, B. Trimmomatic: a flexible trimmer for Illumina sequence data. \nBioinformatics 30, 2114–2120 (2014). \n93. Khelik, K., Lagesen, K., Sandve, G. K., Rognes, T. & Nederbragt, A. J. NucDiff: in -depth \ncharacterization and annotation of differences between two sets of DNA sequences. BMC Bioinform. \n18, 338 (2017). \n94. Danecek, P. et al. Twelve years of SAMtools and BCFtools. Gigascience 10, giab008 (2021). \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint \n\n \n \n24 \n \n95. Ebler, J. et al. Pangenome-based genome inference allows efficient and accurate genotyping across \na wide spectrum of variant classes. Nat. Genet. 54, 518–525 (2022). \n96. McKenna, A. et al. The Genome Analysis Toolkit: A MapReduce framework for analyzing next -\ngeneration DNA sequencing data. Genome Res. 20, 1297–1303 (2010). \n97. Lippert, C. et al. FaST linear mixed models for genome-wide association studies. Nat. Methods 8, \n833–835 (2011). \n98. Li, M. -X., Yeung, J. M. Y ., Cherny, S. S. & Sham, P. C. Evaluating the effective numbers of \nindependent tests and significant p-value thresholds in commercial genotyping arrays and public \nimputation reference datasets. Hum. Genet. 131, 747–756 (2012). \n99. Mansfeld, B. N. & Grumet, R. QTLseqr: an R package for bulk segregant analysis with next -\ngeneration sequencing. Plant Genome 11, 180006 (2018). \n100. Meng, L., Li, H., Zhang, L. & Wang, J. QTL IciMapping: integrated software for genetic linkage \nmap construction and quantitative trait locus mapping in biparental populations. Crop J. 3, 269–283 \n(2015). \n101. Zhang, J. et al. High -level expression of a novel chromoplast phosphate transporter ClPHT4;2 is \nrequired for flesh color development in watermelon. New Phytol. 213, 1208–1221 (2017). \n102. Curtis, M. D. & Grossniklaus, U. A gateway cloning vector set for high -throughput functional \nanalysis of genes in planta. Plant Physiol. 133, 462–469 (2003). \n103. Kim, D., Paggi, J. M., Park, C., Bennett, C. & Salzberg, S. L. Graph -based genome alignment and \ngenotyping with HISAT2 and HISAT-genotype. Nat. Biotechnol. 37, 907–915 (2019). \n104. Liao, Y ., Smyth, G. K. & Shi, W. featureCounts: an efficient general purpose program for assigning \nsequence reads to genomic features. Bioinformatics 30, 923–930 (2014). \n105. Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-\nseq data with DESeq2. Genome Biol. 15, 550 (2014). \n106. Benjamini, Y . & Hochberg, Y . Controlling the false discovery rate: a practical and powerful approach \nto multiple testing. J. R. Stat. Soc. Series B 57, 289–300 (1995). \n \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted July 27, 2025. ; https://doi.org/10.1101/2025.07.25.666869doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}