The watermelon germplasms HSH-F, Sanbai, and CG149 are preserved by the Industrial Crops Institute of Hubei Academy of Agricultural Sciences, Wuhan. The flesh colors of these germplasms are canary yellow, white, and orange, respectively.

For genetic inheritance analysis and genetic mapping of the white flesh trait, the canary yellow flesh inbred line HSH-F (P₁) was crossed with the white flesh line Sanbai (P₂) to produce F₁ hybrids. Subsequently, the F₁ plants were backcrossed with HSH-F and Sanbai to produce the BC₁P₁ and BC₁P₂ populations and were self-pollinated to generate the F₂ segregation population. F₂ individuals were further self-pollinated and individually harvested to generate F₃ families. The F₂ segregation population was used for genetic mapping of the white flesh gene. To examine whether the white and canary yellow flesh traits are controlled by the same locus, the orange flesh line CG149 (P₃) was crossed with HSH-F and Sanbai to produce the F₁(P₃×P₁) and F₁(P₃×P₂) hybrids, respectively. The plant materials were creeping-cultured in a greenhouse under natural light conditions at the vegetable test base of Hubei Academy of Agricultural Sciences, Wuhan (30°36′ N, 104°18′ E). The greenhouse was maintained at daily temperatures between 20°C and 40°C and relative humidity between 50% and 85%. Female flowers were pollinated from 7:00 to 9:00 a.m. The flesh colors of the watermelon fruits were visually recorded at 34 days after pollination (DAP).

Genomic DNA was extracted from the leaves of HSH-F and Sanbai using the cetyltrimethyl-ammonium bromide (CTAB) method (Porebski et al., 1997). The integrity of genomic DNA was examined using 1% agarose gel, purity was assessed with a NanoDrop2000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA), and the concentrations were determined using Qubit 3.0 (Thermo Fisher Scientific). A DNA library was constructed using the NEB_Next Ultra II DNA Library Prep Kit for Illumina (New England Biolabs, Hitchin, UK) according to the manufacturer’s instructions and then sequenced using the HiSeq 4000 (Illumina, San Diego, CA, USA) platform with a paired-end run (2 × 150 bp). After removing the adaptors and low-quality reads, the clean reads were mapped to the watermelon reference genome 97103 (v2) (http://cucurbitgenomics.org/organism/21) using the Burrows–Wheeler Aligner (BWA, v0.7.15-r1140) software. The Genome Analysis ToolKit (GATK, v3.7) was used for single nucleotide polymorphism (SNP) and insertion–deletion (InDel) calling.

For genetic inheritance analysis, the F₂ (n = 137), BC₁P₁ (n = 39), and BC₁P₂ (n = 41) segregation populations derived from the cross between HSH-F and Sanbai were grown in the spring of 2019, and the flesh colors were recorded individually at 34 DAP. To validate the genetic inheritance of flesh color, 51 F₃ families were grown in the fall of 2019 for phenotyping and segregation analysis. For each F₃ family, the flesh colors of 15 individuals were recorded. The deviation from the expected segregation ratio of each population was analyzed using the chi-square test.

For preliminary mapping of the white flesh locus, an SNP panel containing 390 SNPs distributed evenly across 11 watermelon chromosomes was developed based on published watermelon resequencing data (ftp://cucurbitgenomics.org/pub/cucurbit/reseq/watermelon/v2/). The primer sequences and genomic positions of the 390 SNPs are listed in Supplementary Table S1 . The F₂ population with 137 individuals and the 51 F₃ families derived from the cross between HSH-F and Sanbai were firstly genotyped using the SNP panel. Based on the phenotypes of the F₂ and F₃ populations, QTL analysis was performed by composite interval mapping (CIM) using WinQTLCart2.5 software (Li et al., 2007). The logarithm of the odds (LOD) threshold was set at 2.5. To narrow the QTL region (Chr06:15.92–24.50 Mb), six InDel markers ( Supplementary Table S1 ) with an average density of approximately 1.2 Mb were designed and used to genotype the 137 F₂ individuals.

According to the preliminary mapping results, two flanking InDel markers were used to screen recombinant plants from the 1,054 F₂ individuals at the seedling stage in the spring of 2020. These recombinants were further transplanted in the greenhouse for phenotyping. Seven SNP markers ( Supplementary Table S1 ) with an average density of about 100 kb were designed and used to genotype the recombinant plants in order to fine-map the white flesh locus.

Functional annotation of the genes in the fine-mapped region was based on watermelon reference genome 97103 (v2) (ftp://cucurbitgenomics.org/pub/cucurbit/reseq/watermelon/v2/).

Flesh samples from the HSH-F and Sanbai lines were collected from the center of three parental fruits at 10, 18, 26, and 34 DAP. The flesh tissues were immediately frozen in liquid nitrogen and stored at −80°C for RNA extraction. Total RNA was extracted using an RNAprep Pure Plant Kit (BioTeKe, Beijing, China) according to the manufacturer’s instructions. The concentrations of RNA were determined using NanoDrop spectrometry. Reverse transcription was performed using a RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific). The expression levels of the candidate genes were determined by quantitative real-time PCR (qRT-PCR) assays using the MonAmp Fast SYBR Green qPCR Mixture (Monad, Wuhan, China) with the QuantStudio 5 system (ABI, Foster City, CA, USA). The ClACT (Cla97C02G026960) gene was used as an internal reference. The relative expression levels were calculated using the 2^{−(ΔΔ C t)} method (Livak and Schmittgen, 2001). Each test had three biological and technical replicates. The genes and primers used for qRT-PCR analysis are listed in Supplementary Table S1 . Student’s t-test was used to assess statistical differences, and a p-value of 0.05 was used to determine the significance of the differences between means.

Flesh samples from the HSH-F, Sanbai, and CG149 lines were collected from the center of three parental fruits at 34 DAP. The flesh tissues were cut into small pieces, immediately frozen in liquid nitrogen, and then stored at −80°C. Subsequently, the samples were freeze-dried, ground into powder (30 Hz, 1.5 min), and stored at −80°C until use. About 50 mg powder was weighed and extracted with 0.5 ml mixed solution of n-hexane/acetone/ethanol (1:1:1, v/v/v) and the extract vortexed for 20 min at room temperature. Supernatants were then collected after centrifugation at 12,000 rpm for 5 min at 4°C. The residue was re-extracted by repeating the above steps under the same conditions, evaporated to dryness, and reconstituted in a mixed solution of methyl alcohol/methyl tert-butyl ether (1:1, v/v). The solution was filtered using a 0.22-μm membrane filter for further liquid chromatography–tandem mass spectrometry (LC-MS/MS) analysis. The carotenoid content was determined using MetWare (http://www.metware.cn/) based on the AB SciexQTRAP 6500 LC-MS/MS platform. Three biological repeats were performed for each line.

Transmission electron microscopy (TEM) was used to observe the morphology and ultrastructure of the chromoplasts in the fruits of HSH-F and Sanbai at 34 DAP. Flesh samples from the central areas were cut into 2 mm × 2 mm pieces and immediately fixed with 2.5% glutaraldehyde. After embedding in Spurr’s resin, ultrathin sections of the samples were obtained with a Leica EM UC6 ultramicrotome (Leica Microsystems, Wetzlar, Germany). The sections were examined and photographed with an H-7650 transmission electron microscope (Hitachi, Tokyo, Japan) on the electron microscopy platform of Huazhong Agricultural University.

The flesh color of the two parents, HSH-F and Sanbai, was observed at 10, 18, 26, and 34 DAP ( Figure 1A ). Pigment accumulation could be visually observed in the canary yellow flesh parent at 18 DAP, and a distinguishable flesh color was observed in the canary yellow and white flesh parents at 26 DAP. The LC-MS/MS-based determination of the carotenoid content in the fruits of the two parents at 34 DAP showed that the canary yellow flesh contains abundant xanthophyll, whereas the white flesh contains trace amounts of carotenoids ( Figure 1B and Supplementary Table S2 ). In the canary yellow flesh, 30 types of carotenoids were identified, and most of these were xanthophyll. The five most abundant carotenoids in the canary yellow flesh were violaxanthin-myristate-laurate (7.68 ± 1.80 µg/g), violaxanthin palmitate (6.41 ± 2.50 µg/g), lutein (6.35 ± 1.49 µg/g), violaxanthin myristate (4.77 ± 1.32 µg/g), and violaxanthin-myristate-caprate (2.96 ± 1.24 µg/g) ( Supplementary Table S2 ). In white flesh, 19 types of carotenoids were identified, but the carotenoid content was extremely low ( Supplementary Table S2 ).

Previous studies have implied an interaction effect between the white flesh and yellow flesh loci (Henderson, 1989; Henderson et al., 1998). In this study, three parents with canary yellow (HSH-F, P₁), white (Sanbai, P₂), and orange (CG149, P₃) flesh colors ( Figure 2A ) were used for genetic analysis of the flesh color in watermelon. Determination of the carotenoid content revealed that carotenes were highly abundant in the orange flesh parent CG149, which lacked xanthophyll ( Figure 1B and Supplementary Table S2 ). The three most abundant carotenes in CG149 were (E/Z)-phytoene (42.94 ± 1.46 µg/g), lycopene (41.65 ± 3.74 µg/g), and β-carotene (37.00 ± 1.40 µg/g) ( Supplementary Table S2 ).

The canary yellow flesh trait was observed in the fruits of the F₁ hybrids produced from crossing the canary yellow with the white flesh parents (HSH-F × Sanbai) ( Figure 2B ). The F₂ segregation population derived from the cross between HSH-F and Sanbai showed pure yellow, asymmetrical yellow, and white flesh ( Figure 2C ). Both pure and asymmetrical canary yellow individuals were classified into the yellow flesh group. In the F₂ population of 137 individuals, 106 plants showed yellow flesh and 31 showed white flesh, confirming a 3:1 Mendelian ratio (χ ² = 0.199, p = 0.656) ( Table 1 ). In the F₃ population of 51 families, 40 families showed canary yellow or segregated flesh color and 11 families showed white flesh color, which also fitted the expected Mendelian ratio of 3:1 (χ ² = 0.154, p = 0.699). All 39 individuals in the BC₁P₁ population had canary yellow flesh. In the BC₁P₂ population of 41 individuals, 16 showed canary yellow flesh and 25 showed white flesh, fitting the Mendelian ratio of 1:1 (χ ² = 0.988, p = 0.32). Thus, the flesh color variation in this biparental population is controlled by a single nuclear locus.

To confirm whether the white flesh results from the malfunction of the canary yellow flesh locus, the canary yellow and white flesh parents were crossed with the orange flesh parent (CG149). Both F₁ hybrids (CG149 × HSH-F and CG149 × Sanbai) showed a canary yellow flesh color ( Figure 2B ). Thus, the white flesh line Sanbai complemented the malfunction in xanthophyll biosynthesis in the orange flesh line CG149. Hence, the genes involved in xanthophyll biosynthesis in the white flesh parent might function normally ( Figure 2D ). These results indicated that the white flesh trait in watermelon is controlled by a single recessive locus, i.e., Clwf2, and is epistatic to the canary yellow flesh.

For preliminary mapping of the white flesh locus, the SNP panel containing 390 SNP markers ( Supplementary Table S1 ) was used to identify the genotype of the 137 F₂ individuals and the 51 F₃ families. Based on the phenotypic data of the F₂ population and the F₃ families, a locus associated with variations in flesh color was mapped to the watermelon chromosome 6 in an 8.58-Mb region ( Figure 3A ) between the markers M401 (15.92 Mb) and M412 (24.50 Mb). The highest LOD scores for the 2019 F₂ and 2019 F₃ populations were 21.43 and 13.22, which explained 65.543% and 63.55% of the phenotypic variation, respectively ( Supplementary Table S3 ).

To fine-map the Clwf2 locus, the genomes of the parental lines HSH-F and Sanbai were resequenced using the Illumina HiSeq platform for marker development. A total of 96,524 variants were identified between the two parents, including 86,230 SNPs and 10,294 InDels ( Supplementary Table S4 ). Firstly, six InDel markers were designed and used to identify the genotype of the 137 F₂ individuals used for preliminary mapping. This narrowed the Clwf2 locus in the region of 1.02 Mb between markers ID_23467996 (23.47 Mb) and ID_24488295 (23.49 Mb) ( Figure 3B ). The markers ID_23467996 and ID_24488295 were then used to screen the recombinants in an F₂ population of 1,054 individuals at the seedling stage. Then, the recombinants were transplanted in the greenhouse for phenotyping and were further genotyped using seven SNP markers ( Figure 3B ). Finally, the Clwf2 locus was delimited between the markers S_21474486 and S_24306830, with a physical distance of 132.3 kb (Chr06:24,174,486–24,306,830) ( Figure 3B ).

According to the watermelon reference genome 97103 (v2), the 132.3-kb genome region contains 13 putative genes ( Figure 3C ; Supplementary Table S4 ), 11 of which have been annotated. However, no gene in this region was annotated in the CBP. The 11 annotated genes include two leucine-rich repeat receptor-like kinases (LRR-RLKs) (Cla97C06G121800 and Cla97C06G121810), four chloroplastic proteins (Cla97C06G121830, Cla97C06G121870, Cla97C06G121880, and Cla97C06G121890), one cyclin-dependent kinase (CDK)-activating kinase assembly factor (Cla97C06G121840), one ATP-dependent DNA helicase (Cla97C06G121850), one APETALA2-like ethylene-responsive factor (AP2/ERF) (Cla97C06G121910), one heptahelical transmembrane protein (Cla97C06G121900), and one histone-lysine N-methyltransferase (Cla97C06G121920). For the four chloroplastic protein genes, Cla97C06G121830 encodes a pentatricopeptide repeat-containing protein (PPR), Cla97C06G121870 encodes a 50S ribosomal protein, Cla97C06G121880 encodes a DNA-directed RNA polymerase subunit beta, and Cla97C06G121890 encodes a tetratricopeptide repeat (TPR)-like superfamily protein.

No variant was identified in the candidate region in the comparison of the assembled genome resequencing data of the two parents. The expression levels of the 13 genes in the flesh samples from the two parents at 10, 18, 26, and 34 DAP were then analyzed using qRT-PCR. Among the 13 genes, four (Cla97C06G121800, Cla97C06G121810, Cla97C06G121820, and Cla97C06G121870) were only expressed in white flesh at 34 DAP ( Supplementary Figure S2 ). Most of the other nine genes showed low or no expression at 10 DAP in both parents, but their expression levels progressively increased during fruit ripening ( Figure 4 ). The expression levels of Cla97C06G121860, Cla97C06G121880, Cla97C06G121890, and Cla97C06G121900 were significantly higher in the canary yellow than in the white flesh at 18 and 26 DAP. Notably, Cla97C06G121890 and Cla97C06G121900 showed almost no expression in immature white flesh fruits (10, 18, and 26 DAP). Cla97C06G121890 is closely related to the Arabidopsis (Arabidopsis thaliana) REDUCED CHLOROPLAST COVERAGE (REC) proteins, which contribute to establishing the size of the chloroplast compartment and are essential for pigmentation (Larkin et al., 2016). Thus, the white flesh trait in watermelon might be related to the downregulation of the chloroplastic protein.

Chromoplasts are plastids that accumulate the carotenoid pigments in plants (Egea et al., 2010; Li and Yuan, 2013; Sun et al., 2018). Studies have demonstrated that chloroplasts and chromoplasts are interconvertible (Egea et al., 2010; Li and Yuan, 2013). RCP2, a homolog of Cla97C06G121890 in Mimulus, was required for chromoplast development (Stanley et al., 2020). To investigate whether the repressed expression of the chloroplast protein gene affects the development of chromoplasts in watermelon, we examined the ultrastructure of the chromoplasts in flesh cells by TEM at 34 DAP. Notably, rounded and regularly shaped chromoplasts were observed in both canary yellow and white flesh cells ( Figures 5A–D ).

The chromoplast number and size have been reported to affect the accumulation of carotenoids (Sun et al., 2020). We subsequently assessed chromoplast proliferation in white and canary yellow flesh. Because the flesh cell of watermelon is too large to calculate the accurate number of chromoplasts, we then evaluated the relative expression levels of the plastid division genes in fruits of the white and canary yellow flesh parents, including ClFtsZ1, ClFtsZ2, ClARC1, ClARC6, ClCDP, ClPDV1, and ClPDV2 (Osteryoung and Pyke, 2014; Chen et al., 2019). Most of the chloroplast proliferation genes were significantly downregulated during white flesh fruit development, particularly the FtsZ genes ( Figure 5E ). Thus, chromoplast proliferation may be repressed in white flesh watermelon.

The C20 GGPP molecule is the direct precursor for carotenoid biosynthesis. We first examined the relative expression level of GGPPS (Cla97C02G039100) in the fruits of the canary yellow and white flesh parents. The expression level of GGPPS in canary yellow flesh was significantly higher than that in white flesh during the whole fruit development stage ( Figure 6 ). PSY is the primary bottleneck gene in carotenoid biosynthesis (Nisar et al., 2015; Yuan et al., 2015). Three PSY genes have been annotated in the watermelon genome: PSY1 (Cla97C01G008760), PSY2 (Cla97C02G050140), and PSY3 (Cla97C07G137500). PSY1 had a much higher expression level in watermelon fruits compared to PSY2 and PSY3 ( Figure 6 ). The expression level of PSY1 gradually increased with fruit ripening, and PSY1 showed a significantly higher expression level in canary yellow flesh than in white flesh at all developmental stages. Similarly, the PDS (Cla97C07G142100) and Z-ISO (Cla97C07G142750) genes showed significantly higher expression levels in canary yellow flesh than in white flesh at all developmental stages. We also examined the expression levels of ZDS (Cla97C06G118930), CRTISO (Cla97C10G200950), LCYB (Cla97C04G070940 and Cla97C10G193770), CHYB (Cla97C05G090480), and ZEP (Cla97C02G038200). These genes exhibited increased expression levels with fruit ripening in both parental lines; however, all genes showed significantly higher expression levels in the canary yellow flesh than in the white flesh at 18 or 26 DAP, the vital stages for pigmentation ( Figure 6 ). Thus, the entire CBP is repressed in white flesh.