Introduction

Fruit shape is among one of vital characteristic traits for horticulture industry that exhibits a broad range of phenotypic variation, thus emphasizing its importance in breeding program. Many QTLs and genes related to fruit shape have been detected or cloned in different crops. In cucumber, various QTLs of fruit shape have been genetically mapped (Bo et al. 2015; Jiang et al. 2015; Weng et al. 2015). Using RIL populations, ten QTLs were identified for fruit length (FL), fruit diameter (FD), mature fruit length (MFL), and mature fruit diameter (MFD) (Miao et al. 2011). During the early cucumber development, seven kinesin genes (CsKF1 to CsKF7) were proved to play the key role in exponential cell production and enlargement in fruit (Yang et al. 2013). The sf1 gene was also proposed for controlling cucumber fruit length by regulating gibberellins and cytokinin biosynthesis, signal transduction, and auxin signaling (Wang et al. 2017). Another study identified 8 QTLs for mature and immature cucumber fruit length. A major-effect QTL fl3.2 was detected, which explained a maximum phenotypic variation of 38.87% (Wei et al. 2016). The CsSUM, the homologous gene of SUN in tomato, was the candidate gene for cucumber spherical fruit (Pan et al. 2017). In tomato, two categories of quantitative trait loci (QTLs) for fruit shape were suggested: One was locule number and flat shape controlled by lc (locule number) and fas (fasciated); the other was fruit shape controlled by SUN, OVATE, LOCULE NUMBER, and FASCIATED (Liu et al. 2002; Xiao et al. 2008; Munos et al. 2011; Rodriguez et al. 2011; van der Knaap et al. 2014). In addition, two suppressors of the ovate mutation (Sov1 and Sov2) were proved to be regulator of fruit shape. The self1 promotes fruit elongation by increasing cell layers in the ovary and the QTL fs8.1 together with SUN, OVATE also controls tomato fruit shape (Paran and van der Knaap 2007; Rodriguez et al. 2013; Chusreeaeom et al. 2014; Sun et al. 2015). In melon, the andromonoecious (a) was cloned and confirmed to control the fruit shape and carpel numbers (Boualem et al. 2008). Moreover, other QTLs associated with melon fruit shape (FS) were demonstrated by a number of QTL mapping (Paris et al. 2008; Fernandez-Silva et al. 2010; Tomason et al. 2013).

Watermelon [Citrullus lanatus (Thunb.) Matsum. & Nakai] belongs to Cucurbitaceae family and is one of the most important commercial crops worldwide having 11 chromosomes and a haploid genome of approximately 425 Mb (2n = 2x = 22). The genome sequence and the genes annotation of watermelon were published in 2013 following cucumber and melon (Huang et al. 2009; Garcia-Mas et al. 2012; Guo et al. 2013), providing the opportunity to apply next-generation sequencing to fine map genes function. It was considered that watermelon fruit shape was controlled by an incompletely dominant gene, resulting in elongate (OO), oval (Oo), and spherical (oo) fruits (Weetman 1937; Guner and Wehner 2004). Tanaka et al. (1995) also determined that spherical (Os) and oval (O+) watermelon fruits are controlled by a single allele which shows incomplete dominance through the cross between a spherical fruit inbred line and an oval fruit inbred line. Furthermore, Poole and Grimball (1945) also found the same inheritance pattern in F2 populations of ‘Peerless’ × ‘Baby Delight’ and ‘Northern Sweet’ × ‘Dove’. Allele ObE was proposed for elongate fruit, which was the dominant; allele ob for oblong fruit, which was the recessive, allele ObR (was not the same as the o gene for round) for the round fruit (Lou and Wehner 2016).

Application of next-generation sequencing (NGS) and release of draft genome assemblies for two diploid watermelon inbred lines (97103 and Charleston Gray) (http://cucurbitgenomics.org/) greatly facilitated the construction of many genetic maps of watermelon recently (Sandlin et al. 2012; Zhang et al. 2012; Ren et al. 2014; Cheng et al. 2016). However, only a limited number of QTLs or genes have been reported for watermelon fruit shape. Using a consensus map of three mapping populations in watermelon, the major QTLs for fruit length (FL), fruit width (FWD), fruit shape index (FSI), and fruit weight (FWT) were found to be stable across genetic backgrounds and environments. It was the first stable QTL for fruit size and shape in watermelon (Sandlin et al. 2012). Ren et al. (2014) constructed an integrated genetic map based on four mapping populations, and QTLs for fruit weight (FWT), fruit length (FL), fruit width (FWD), fruit shape index (FSI), and rind thickness (RTH) were detected and co-localized with QTL associated with Brix (BRX), suggesting the existence of pleiotropic effects on fruit maturity. The fruit shape index is the ratio of fruit length and width. There was positive correlation between fruit length, fruit width, and fruit shape index. Four QTLs associated with fruit shape index (FSI) were also detected through a genetic linkage map, and these QTLs explained the phenotypic variation of 31.9% (Sandlin et al. 2012; Cheng et al. 2016).

Information regarding inheritance pattern of fruit shape is elusive and candidate genes controlling the fruit shape in watermelon have not been reported yet. In this study, we investigated the inheritance of watermelon fruit shape gene in F2 population of ‘Duan125’ (elongate fruit) × ‘Zhengzhouzigua’ (spherical fruit). We identified a candidate gene on chromosome 3 associated with watermelon fruit shape (ClFS1) through genome-wide association studies (GWAS) of 315 watermelon accessions and bulked-segregant analysis (BSA) by genotyping a pair of bulked DNA samples from two sets of individuals with opposite extreme phenotypes (elongate individuals and spherical individuals). This is the first report about the candidate gene in controlling watermelon fruit shape. This study will open up the ways to shorten the breeding period and lead to selection of desired fruit shape in breeding program. This work will assist in selection of plants with desired fruit shape at the early stages of seedling or growth that will ultimately hasten the breeding process. This new allele provides a valuable tool in understanding the inheritance mechanisms of ClFS1 in watermelon.

Materials and methods

Plant materials

Seeds of two inbred lines elongate watermelon ‘Duan125’ (P1) and spherical watermelon ‘Zhengzhouzigua’ (P2) were obtained from National Watermelon and Melon Germplasm Resource Library (Zhengzhou, China) to derive F1 population by crossing them. Later, selfing was performed in F1 to raise F2 (Fig. 1). Backcross population was obtained by hybridizing F1 with each parent to create BC1P1 (F1 × Duan125) and BC1P2 (F1 × Zhengzhouzigua).

Fig. 1
figure 1

Materials of two different fruit shapes in watermelon. a Ovary of elongate watermelon ‘Duan125’. b Matured fruit of elongate watermelon ‘Duan125’. c Ovary of spherical watermelon ‘Zhengzhouzigua’. d Matured fruit of spherical watermelon ‘Zhengzhouzigua’

Full size image

For segregation analysis, F2 population was grown in two experiments conducted in 2016 winter, 2017 spring with 348 and 420 F2 individuals, respectively. The BC1P1 population (120 plants) and the BC1P2 (238 plants) population were investigated only in spring 2017 (Table S1). For each plant, the fruit length and width were recorded at mature stage. The fruit length and width were measured from ovary formation to fruit maturation to calculate shape index in two parents. At least ten fruits per genotype were evaluated. All the materials were grown in the experimental area of ‘Xinxiang’, Zhengzhou fruit research institute (Zhengzhou, China).

Bulked-segregant analysis of DNA

DNAs were isolated using the CTAB method (Porebski et al. 1997) from fresh leaves of F2 plants in the 2016 winter for BSA-seq and CAPS marker analysis. Two DNA pools, elongate watermelon fruit pool (E-pool) and spherical watermelon fruit pool (S-pool), were constructed by mixing an equal amount of DNAs from 30 elongate watermelon plants and 30 spherical watermelon plants. Pair-end sequencing libraries (read length 100 bp) with insert size of around 500 bp were prepared for sequencing with an Illumina Genome Analyzer IIx machine. The short reads from E-pool and S-pool were aligned to the ‘97103’ reference genome (Guo et al. 2013) with the BWA software (Li and Durbin 2009). Alignment files were converted to SAM/BAM files using SAM tools (Li et al. 2009), and then applied to the SNP-calling filter ‘Coval’ previously developed (Abe et al. 2012) to increase SNP-calling accuracy. SNP-index was calculated for all the SNP positions. We excluded SNP positions with SNP-index of < 0.6 and read depth < 6 from the two sequences, as these may represent spurious SNPs called due to genomic repeat sequence, sequencing, or alignment errors.

Two parameters, SNP-index and Δ (SNP-index) (Abe et al. 2012; Takagi et al. 2013) were calculated to identify candidate regions for watermelon fruit shape QTL. An SNP-index is the proportion of reads harboring the SNP that are different from the reference sequence. Δ (SNP-index) was obtained by subtraction of SNP-index of E-pool from that of S-pool. Thus, SNP-index = 0 if the entire short reads contain genomic fragments from mutation line; SNP-index = 1 if all the short reads are from reference genome ‘97103’. An average of SNP-index of SNPs located in a given genomic interval was calculated using a sliding window analysis with 1 Mb window size and 10 kb increment. The SNP-index graphs for E-pool and S-pool, as well as corresponding Δ (SNP-index) graph were plotted.

To generate confidence intervals of the SNP-index value under the null hypothesis of no QTL, we carried out computer simulation. We first made two pools of progeny with a given number of individuals by random sampling. From each pool, a given number of alleles corresponding to the read depth were sampled. We calculated SNP-index for each pool and derived the Δ (SNP-index). This process was repeated 10,000 times for each read depth and confidence intervals were generated. These intervals were plotted for all the genomic regions that have variable read depths.

GWAS

A total of 315 watermelon accessions which contained different fruit shape were re-sequenced and 2.3T data were obtained which had 85.42% average genome coverage and 9.24 × average sequence depth. The watermelon accessions were genotyped using 4,661,625 evenly spaced SNPs. The association between fruit shape and each SNP was tested using a unified mixed model (Yu et al. 2006; Zhang et al. 2010). This mixed linear model included principal components (Price et al. 2006) as a fixed effect to account for population structure, kinship matrix (Loiselle et al. 1995), as well as to explain the familial relatedness. Utilizing the Bayesian information criterion, a backward elimination procedure was implemented to determine the optimal number of principal components to include in the mixed model (Schwarz 1978). The false discovery rate was controlled at 5% using the Benjamini and Hochberg (1995) procedure. A likelihood ratio-based r2 statistic was used to assess the goodness of fit of each SNP (Sun et al. 2010). All analyses were conducted using the Genome Association and Prediction Integrated Tool software package (Lipka et al. 2012).

Fine mapping through CAPS markers

Watermelon genome sequence was obtained from the watermelon database (http://www.icugi.org); the sequence was compared with the re-sequenced data to identify SNPs via a filter pipeline (Takagi et al. 2013). To minimize the genetic interval for fine mapping and to verify the accuracy of BSA-Seq, The 298 cleaved amplified polymorphic sequences (CAPS) markers were developed based on SNPs generated from BSA-seq (Table S2).

The PCR was carried out in a total volume of 10 μL containing 5 μL 2× Power Taq PCR Master Mix (BioTeke, China), 0.5 μL of 10 μM per primer, 1 μL of 200 ng of genomic DNA, and 3 μL RNase-free water. All amplifications were performed on a EasyCycler (Analytik Jena, Germany) under the following conditions: 5 min at 95 °C; 28 cycles of 30 s at 94 °C, 30 s at 56 °C, 50 s at 72 °C, and a final extension step at 72 °C for 10 min. Amplified PCR products were digested using suitable restriction endonucleases according to the manufacturer’s instructions at 37 or 65 °C for 4–10 h. The digests were resolved by electrophoresis in 1.0% agarose gel or PAGE gel and visualized on a Versa Doc (Bio-Rad) after staining with ethidium bromide (EB).

Candidate gene prediction and testing marker for linkage to fruit shape of watermelon

The predicted genes in the mapping interval were downloaded from the watermelon database (http://www.icugi.org/). All predicted genes were sequenced and aligned to elongate and spherical watermelon to check the bases’ difference and to obtain the candidate gene. The candidate gene function was retrieved through NCBI (https://blast.ncbi.nlm.nih.gov/Blast.cgi). An indel marker (Table S3) was developed according to the candidate gene sequence and tested on 100 F2 individuals in 2017 spring. Marker was also tested to confirm the linkage on 105 watermelon progenies which contain 95 spherical watermelons and 10 elongate watermelons.

Real-time PCR analysis for candidate gene expression

The ovaries and fruits from different developmental stages and others tissues from flowering stage were collected for RNA extraction to analyse the gene expression. Total RNA was extracted using the plant total RNA purification kit (GeneMark, China) following the manufacturer’s instructions. The cDNA was synthesized with reverse transcriptase M-MLV (RNase H-) following the manufacturer’s instructions (Takara, Japan).

The primers of candidate gene and reference gene Actin (Kong et al. 2015) used in quantitative reverse transcription polymerase chain reaction (qRT-PCR) were designed based on Cucurbit Genomics Database (http://www.icugi.org) (Table S3). Expression levels of the target gene were evaluated by qRT-PCR using a LightCycler480 RT-PCR system (Roche, Swiss). All reactions were performed using the SYBR Green real-time PCR mix according to the manufacturer’s instructions. Each 20 μL RT-PCR reaction mixture containing 1 μL cDNA, 1 μL forward primer (10 μM), 1 μL reverse primer (10 μM), 10 μL 2× SYBR Green real-time PCR mix, and nuclease-free water to final volume of 20 μL was preheated at 95 °C for 5 min, followed by 45 cycles of 95, 60, and 72 °C for 30 s. High-resolution melting was performed under the following conditions: 1 min at 95 °C, 1 min at 40 °C, 1 s at 65 °C, and continuous at 95 °C. All experiments were performed in triplicate. The raw data of qRT-PCR were analyzed using LCS480 software 1.5.0.39 (Roche, Swiss) and the relative expression was determined using the 2−ΔΔCT method (Livak and Schmittgen 2001).

Results

Inheritance of fruit shape and changes of fruit shape index in watermelon

The watermelon fruit has three major kinds of shapes: elongate, oval, and spherical. We defined the fruit shape index of > 1.8, 1.4–1.6, and 1.0–1.1, for elongate, oval, and spherical fruit, respectively. Segregation ratio of watermelon fruit shape (ClFS1) between two F2 populations in 2 years is presented in Table S1. All fruits length and width were measured to calculate the fruit shape index at fruit ripening stage. The ClFS1 in the four segregating populations (F2-2016 Winter, F2-2017 Spring, BC1P1-2017 Spring, and BC1P2-2017 Spring) showed that the elongate fruit allele was incompletely dominant over the spherical fruit. The χ2 test showed the segregation ratio of two F2 populations was in accordance with 1:2:1 (Table S1), suggesting that ClFS1 was controlled by a single gene with incomplete dominance in this population.

To evaluate the effect of ClFS1 on development, fruit shape index was monitored during fruit development in two parents. The dynamics of the fruit shape index in elongate and spherical watermelon followed the similar pattern from 2 days before anthesis (DBA) in ovaries to fruit ripening with the highest value occurring at anthesis stage which was 3.15 and 1.43, respectively (Fig. 2). The fruit shape index of elongate and spherical watermelon was stable from 25 day post-anthesis (DPA) to fruit ripening, which was 1.83 and 1.02, respectively (Fig. 2). However, the final fruit shape index of both elongate and spherical watermelon was similar to the respective shape index recorded 2 days before anthesis.

Fig. 2
figure 2

Dynamic changes of fruit shape index between elongate and spherical watermelon during the ovary and fruit development. At least three fruits per stage were evaluated

Full size image

GWAS identifies ClFS1 gene located on chromosome 3

GWAS utilized a total of 315 watermelon accessions including elongate, oval, and spherical fruit shape to capture maximum genetic diversity, and was genotyped with a high-density, genome-wide coverage 4,661,625 and evenly spaced SNPs. To reduce the incidence of false-positive signals, a unified mixed linear model that controls population structure and familial relatedness was used (6PC + K (for 6 Principal Components and Kinship); Yu et al. 2006) to test associations between fruit shape index and 421,101 of the 4,661,625 SNPs with minor allele frequency > 5%.

The difference of tenfold was calculated through the analysis of natural variation of watermelon fruit shape across the diversity panel coupled with a 85% (broad-sense) heritability, revealing that the observed natural variation is largely dictated by genetic variation across the population rather than environmental factors (Harjes et al. 2008; Chandler et al. 2013). GWAS profiles among 315 varieties identified the major locus of 26.32–27.94 Mb designated on watermelon chromosome 3 which might control watermelon fruit shape (Fig. 3). All the SNPs information in this interval is listed in Table S4. The strongly associated SNPs were located at 26,783,440 and 26,847,336 bp and the two SNPs were within the coding region of Cla011249 and Cla011257, respectively (Fig. 3; Table S4).

Fig. 3
figure 3

Locus of watermelon fruit shape was identified through GWAS. Manhattan plot of genome-wide association for watermelon fruit shape showed that the region of 26.32–27.94 Mb on watermelon chromosome 3 controlled the fruit shape

Full size image

ClFS1 gene located on chromosome 3 through BSA-seq

We constructed F2 population by crossing elongate watermelon ‘Duan125’ and spherical watermelon ‘Zhengzhouzigua’ (Fig. 1). A total of 30 extreme elongate individuals (E-pool) and 30 extreme spherical individuals (S-pool) from the F2 population in 2016 winter were selected and mixed to generate two pools. We sequenced the two pools on the Illumina HiSeq™ PE150 platform. A total of 30.1 GB of raw data were generated for both pools, with approximately 30 × depth and more than 99% coverage for each. The high-throughput sequencing resulted in 102,832,531 and 108,242,182 short reads from E-pool and S-pool, respectively. Data were aligned to ‘97103’ watermelon reference genome (http://www.icugi.org) and 304,565 SNPs were identified between two pools. Each identified SNP was used to compute an SNP-index. Using 1 kb sliding window, an average SNP-index was calculated in a 1 Mb interval. SNP-index graphs of E-pool and S-pool were generated by plotting the average SNP-index against the position of each sliding window in the ‘97103’ watermelon genome assembly. Graph for Δ(SNP-index) was plotted and computed against the genome positions by combining SNP-index of E-pool and S-pool (Fig. 4a).

Fig. 4
figure 4

Genetic mapping of the fruit shape gene ClFS1 in watermelon. a Δ(SNP-index) graph of BSA-seq analysis. b Locus at the interval of 23.52–28.83 Mb on chromosome 3 was identified to control watermelon fruit shape. c Examination of recombinants in F2 refined the location of ClFS1 in an interval defined by two markers CAPS0158 and CAPS0165, which was a 46 kb interval

Full size image

The region on chromosome 3 from 23.52 to 28.83 Mb displayed a higher average than 0.5 Δ (SNP-index) (Fig. 4a, b) and was significantly different from 0 of Δ (SNP-index) value at 95% significance level. The results showed that there was a candidate gene controlling watermelon fruit shape in that region.

Analysis of the CAPS markers narrowed down ClFS1 to a 46 Kb interval

To confirm the watermelon fruit shape locus detected by BSA-seq and GWAS, 768 F2 individuals from the 2016 winter and 2017 spring (Table S1) were analyzed. We developed 298 CAPS markers from chromosome 3 and screened the F2 segregating population for polymorphic analysis (Table S2). Only six recombinant individuals were obtained between markers CAPS0158 and CAPS0165. Since no reliable SNPs were found in the region between the two markers, further fine mapping was unfeasible. Therefore, the ClFS1 gene was delimited in the region between markers CAPS0158 and CAPS0165, corresponding to 26.817 and 26.863 Mb (46 kb interval) on chromosome 3 (Fig. 4c).

Identification of ClFS1 gene

According to the watermelon genome database (http://www.icugi.org/), four putative genes were annotated in 46 kb interval (Table S5). DNA sequences of these four genes were obtained from the elongate fruit ‘Duan125’ and spherical fruit ‘Zhengzhouzigua’ parents. The CDS of Cla011257 represented a 159 bp indel (530–688 bp) between the two parents (Fig. 5a, b), while no sequence changes were found in the other three genes between the two parents. The analysis of Cla011257 protein sequence of ‘Duan125’ and ‘Zhengzhouzigua’ showed that 53 amino acids were deleted in elongate watermelon ‘Duan125’ (Fig. 5b). We proposed that Cla011257 is the candidate gene for fruit shape in watermelon.

Fig. 5
figure 5

Identification of watermelon fruit shape gene Cla011257. a Structure of Cla011257 gene. Open and gray boxes represent untranslated regions (UTRs) and exons, respectively, while lines denote introns. b Confirmation of the deletion in elongate watermelon by sequencing. The red dotted line indicates the 159 bp deletion from ACT to TCC. c Co-segregation of the fruit shapes phenotype and the 159 bp deletion of Cla011257 in F2 population. Genotyping by PCR revealed that 25 elongate fruit individuals were dominant homozygous (Hom) for the deletion and the 48 oval fruit individuals were heterozygous (Het), whereas 27 spherical fruit individuals were recessive homozygous (Null) (color figure online)

Full size image

To verify this hypothesis, we analyzed the nucleotide polymorphism of Cla011257 between two watermelon reference genomes spherical fruit ‘97103’ and elongate fruit ‘Charleston Gray’ (http://www.icugi.org/). A 159 bp deletion was also checked in ‘Charleston Gray’ compared with ‘97103’ (Fig. 6a), which was the same as above results. Furthermore, deletion of 53 amino acids was found in ‘Charleston Gray’ in the same region as above. The blast alignment to tair (https://www.arabidopsis.org/) showed that Cla011257 is highly homologous to AT3G16490 in Arabidopsis genome which encodes IQD26 proteins (Fig. 6). Network analysis of IQD26 protein in Arabidopsis thaliana (https://string-db.org/) showed that many IQD family members were associated with microtubule arrays and Ca2+-CaM signaling modules (Burstenbinder et al. 2017) and also had interaction with each other (Fig. 6b). Therefore, it further suggested that the Cla011257 might be the candidate gene controlling watermelon fruit shape.

Fig. 6
figure 6

Analysis of nucleotide polymorphisms and protein function prediction. a Nucleotide polymorphisms of Cla01257 among four watermelon genomes including two watermelon reference genomes ‘97103’ (spherical fruit) and ‘Charleston Gray’ (elongate fruit). The 159 bp deletion was also found in two reference genomes. b Network analysis of IQD26 protein which was blasted to high sequence consistency with Cla011257 in A. thaliana (https://string-db.org/) showed that many IQD family members had interaction each other

Full size image

Expression analysis of Cla011257 during watermelon fruit development

The expression pattern of Cla011257 was investigated using RT-PCR at flowering stage in different tissues that includes root, stem, young leaf, male flower, and flower bud; we also determined the expression level in ovary, pericarp, and flesh at different developmental stages of fruit in elongate and spherical plant parents. The primers for Cla0112557 and reference gene Actin are listed in Table S3. Cla011257 had the highest expression level in ovaries formation stage in two parents. The expression level of Cla011257 in elongate watermelon was almost threefold more than spherical watermelon at this stage (Fig. 7). During different developmental stages of fruit, transcript level of Cla011257 was significantly lower in other tissues as compared to ovaries in the formation stage (Fig. 7).

Fig. 7
figure 7

Relative expression level of Cla011257 during the different tissues and different development stages of ovary between elongate and spherical watermelons

Full size image

An indel marker was developed to check watermelon fruit shape

An indel marker was developed according to the 159 bp sequence of Cla011257. The primer for this marker is listed in Table S3. To verify this marker, 100 individuals including the six recombinant individuals were selected in F2 population to check the polymorphism. The result confirmed that 25 elongate individuals were homozygous dominant and 48 oval individuals were heterozygous, whereas 27 spherical individuals were homozygous recessive, consistent with the phenotype (Fig. 5c). This marker was used to screen 105 watermelon accessions, which contained 10 elongate and 95 spherical fruits (Table S6). As expected, the genotypes perfectly matched the phenotype in spherical and elongate watermelon (Table S6). These results indicated that the Cla011257 might control watermelon spherical fruit shape and the deletion of 159 bp in Cla011257 may result in the fruit elongation during evolutionary process.

Discussion

The conventional QTL mapping method is a laborious and time-consuming process that the molecular markers are developed and every individual is genotyped and phenotyped in a mapping population. NGS-assisted BSA is less laborious, much cheaper, and has no population size limitation for genotyping work, because it provides an effective and simple method to identify molecular markers linked to target genes/QTLs controlling the traits by genotyping only a pair of bulked DNA samples from two sets of individuals with distinct or opposite extreme phenotypes (Takagi et al. 2013; Yang et al. 2013). In this study, we employed this method together with GWAS to identify a candidate gene for fruit shape in watermelon.

As an important cucurbit crop throughout the world, watermelon presents different fruit shapes including elongate, oval, and spherical (Guo et al. 2013). However, the genetic pattern and the developmental mechanisms of watermelon fruit shape are not justified yet. Previously, the inheritance of the watermelon fruit shapes had been controversial, some researchers claimed that it was controlled by a single gene (O), and others thought that it was determined by a number of QTLs together (Tanaka et al. 1995; Guner and Wehner 2004; Sandlin et al. 2012; Zhang et al. 2012; Ren et al. 2014; Cheng et al. 2016). In our 2 year study, we found that watermelon fruit shape was controlled by a single, incompletely dominant gene resulting in fruit that are elongate (OO), oval (Oo), and spherical (oo).

Limited number of QTLs was identified in watermelon for fruit shape in the previous studies (Sandlin et al. 2012; Zhang et al. 2012; Ren et al. 2014; Cheng et al. 2016). Assuming that the QTLs belonged to the same QTL loci, the common genetic mechanisms could be shared underlying the fruit shape. Ren et al. (2014) identified three consensus QTLs for fruit length, fruit width, and fruit shape index on chromosome 3 of watermelon genome by constructing an integrated genetic map based on four mapping populations. Similarly, there were six QTLs for fruit length, fruit width, and fruit shape index on a genetic linkage map (Cheng et al. 2016). In this study, watermelon fruit shapes were divided into three distinct categories (elongate, oval, and spherical) by measuring fruit shape index and gene was located to a single locus on chromosome 3. To identify ClFS1 gene, we performed a combinatorial approach by integrating BSA-seq in an F2 population and GWAS in 315 watermelon accessions. The same locus on chromosomes 3 in watermelon genome was mapped using the GWAS- and NGS-assisted BSA approach (Figs. 3, 4). The BSA-seq was utilized for genome-wide identification of SNPs between two bulked pools, which had been used to develop molecular markers in gene mapping (Lee et al. 2014; Ramirez-Gonzalez et al. 2015). The ClFS1 gene was delimited in the region of 26.817–26.863 Mb on chromosome 3 using 298 CAPS markers (Fig. 3c; Table S2). Sequence annotation analysis showed that there are four putative genes in this region (Table S5). Sequences alignment between elongate plant and spherical plant showed that the CDS of Cla011257 represented a 159 bp deletion in elongate watermelon (Fig. 5a, b). The same 159 bp deletion was also checked in two watermelon reference genomes ‘Charleston Gray’ (elongate) and ‘97103’ (spherical) (Fig. 6a). Based on the deletion, an indel marker was developed to test the polymorphism among 105 watermelon accessions. Results identified that Cla011257 was the candidate gene for watermelon fruit shape (Table S6).

The critical role of cell division and cell enlargement in fruit shape has been demonstrated (Ando and Grumet 2010; Wang et al. 2017). In cucumber, the cell number and cell size in short cucumber fruit was lower than that of length, possibly due to abnormalities in auxin signaling in short fruit (Wang et al. 2017). Fruit development in many horticultural crops can be divided into three phases: ovary development, cell division, and cell expansion (Marcelis and Hofman-Eijer 1993). Cell division typically occurs about 3 day post-pollination and cell enlargement persists throughout the development of fruit, whereas ovary development occurred at pre-pollination (Marcelis and Hofman-Eijer 1993). In melon, it had been confirmed that the mature fruit shape exhibited a high correlation with the ovary and the fruit shape could be determined before pollination of ovary (Perin et al. 2002; Eduardo et al. 2007; Wei et al. 2016). In the present study, we noticed the similar mechanism that watermelon fruit shape can be predicted at the stage of ovary formation. The highest expression pattern of ClFS1 had been detected at the stage of ovary formation, and the expression level of elongate fruit was almost threefold higher than that of spherical fruit (Fig. 7), suggesting that the fruit shape was predominantly determined by ovary and the high expression of ClFS1 at the stage of ovary formation resulted in the watermelon fruit shape difference.

Genetic variants and genes mutation started from rare ones of very low frequency that occurred in one or few individuals in a given population, and these rare variants were turned into common ones through evolutionary forces such as selection, migration, and genetic drift. Even though there has been a great progress in several crops and model plants in generating comprehensive maps of genome variation (Lam et al. 2010; Chia et al. 2012; Qi et al. 2013; Lin et al. 2014). It had been reported that the watermelon genome speciation event occurred 15–23 million years ago (Guo et al. 2013). In the existing watermelon germplasm resources, most of the wild watermelons are spherical. In our study, the Cla011257 had the same sequence as the reference genome of ‘97103’ bearing spherical fruit and a 159 bp deletion was detected in another reference genome of ‘Charleston Gray’ which has elongate fruit. Analysis of 105 watermelon accessions showed that the consistency between genotype and phenotype of Cla011257 was 100% in spherical watermelon (Table S6). The most likely reason was that the ClFS1 genetic variant resulted in fruit shape variation during the process of watermelon evolution. When watermelon genome occurred, the ClFS1 determined spherical fruit and was recessively inherited. The 159 bp deletion of ClFS1 led to elongated fruit; inheritance was dominant and easy in descendant.

Phylogenetic analysis showed that Cla011257 designated as ClFS1 have high sequence identity with AT3G16490 in Arabidopsis genome which belongs to IQD protein family. Network analysis of IQD26 protein in A. thaliana (https://string-db.org/) showed that many IQD family members which have been proved to be associated with microtubule arrays and Ca2+-CaM signaling modules (Burstenbinder et al. 2017) have close interaction with each other (Fig. 6b). In tomato, the SUN controlling elongate fruit encodes a member of the IQD family of calmodulin-binding proteins which interacts with both calmodulin/calmodulin-like proteins (CaM/CMLs) and kinesin light-chain-related protein-1 (KLCR1); the IQD proteins can be recruited to microtubules through CaM/CMLs and KLCR1 (Abel et al. 2005; Xiao et al. 2008; Burstenbinder et al. 2013). The SUN gene led to elongate fruit by decreasing cell number in septum mediolateral and increasing cell number in proximal–distal direction of pericarp in tomato (Wu et al. 2011). Research in Arabidopsis showed that IQD domain families regulated cell shape and growth through the integrating CaM-dependent Ca2+ signaling and possibly other signal transduction pathways during plant development (Burstenbinder et al. 2017). IQD family may provide an assortment of versatile platform proteins that facilitate and specify CaM/CML dynamics during Ca2+ signaling at the cell periphery, on the cytoskeleton, and in the cell nucleus (Charpentier and Oldroyd 2013). Auxin could regulate the expression of IQDs, suggesting that plant growth and development regulation is the function of cellular auxin and calcium signaling (Cai et al. 2016). In this study, the different watermelon fruit shapes may be due to the ClFS1 that provide a bridge of Ca2+ signaling, resulting in the difference of auxin level in cells.

Discovery of watermelon fruit shape gene provides a good entry point to explain the genetic mechanisms of fruit development as well as providing fundamental insights into the domestication and selection history of watermelon. The results of this study demonstrated the candidate gene of watermelon fruit shape (ClFS1) Cla011257, an indel marker was developed which can be used to identify the fruit shape at the seedling stage. In addition, the current results will be useful in marker-assisted breeding for selection of desired traits at early stage of seedling.

Author contribution statement

HK and WL conceived the research and designed the experiments. XL and NH developed the plants population. LZ and SZ analyzed BSA and GWAS data. AA checked the manuscript. JD performed most of the experiment and wrote the manuscript. All authors reviewed and approved this submission.