Introduction

Dwarfism is a valuable and economically important plant architecture trait in crop breeding, and has positive effects on improving yield and high efficiency in labor reduction in management and harvesting strategies. Various dwarf mutants discovered in different plant species have been widely used in crop breeding, such as the Green Revolution genes sd1 in rice and Rht-D1b and Rht-B1b in wheat1,2. In cucurbits, dwarf or compact plant types have attracted much attention from plant breeders because of the higher planting densities to improve crop production. To date, several recessive genes conferring short internodes or bushy phenotypes have been reported in cucumber. For example, the truncated F-box protein CsaVBF1 is strongly associated with dwarfism in cucumber mutant si3. A CLAVATA1-type receptor-like protein, CsCLAVATA1, in cucumber was considered the best possible causal gene for the dwarf phenotype in the EMS-induced mutagenesis Csdw4. Additionally, a putative cytokinin oxidase gene CKX identified in the cp locus seemed to be responsible for the compact habit in cucumber line PI3089155, while the BR-C6 oxidase-encoding gene CsCYP85A1 and steroid 5α-reductase encoding gene CsDET2 from cucumber dwarf mutants scp-1 and scp-2, respectively, were confirmed to be functionally involved in brassinosteroid (BR) biosynthesis6,7. Compact growth habits in melon, including short internode and short lateral branching, are regulated by recessive or incomplete dominant genes, such as si-1, si-2, si-3, mdw1, and slb, which have not yet been cloned8,9,10. In squash, bushy plant habit is a dominant phenotype and controlled by the Bu locus, which also lacks functional characterization11.

Genes underlying the dwarf mutations were mainly involved in biosynthesis or the signal transduction pathway of plant hormones, such as gibberellins (GAs)1,2, cytokinin5, and BRs6,7, which regulate cell elongation and division. GAs, a class of important plant growth-promoting hormones, have been reported to play critical roles in controlling plant growth and development1,12,13. Diverse mutations in the biosynthetic and metabolic pathways of GAs producing bush types enable the elucidation of the underlying genetic basis of dwarfism. For example, Green Revolution genes sd1 in rice and Rht-D1b and Rht-B1b in wheat were reported to be involved in GA metabolism and signaling response pathways, which encode nonfunctional GA20 oxidase (GA20ox) and DELLA proteins without functional DELLA domains, respectively1,2. As the hub repressors in the GA signaling transduction pathway, DELLA proteins belonging to the GRAS gene family contain both N-terminal DELLA and VHYNP domains13,14. In wheat, the aforementioned Rht-D1b and Rht-B1b mutations in the N-terminal motif lead to reduced responsiveness to GA and dwarfism2. Additionally, the deletion of the 17 amino acid residue segment in the DELLA domain of GAI reduces plant height in Arabidopsis15. In the GA biosynthetic pathway, the CPS encoding enzyme is involved in an early step, which converts the GGDP to CDP in plastids, while KAO in the endoplasmic reticulum catalyzes the conversion of ent-kaurene acid GA1212,16,17. GA3 oxidase (GA3ox), as well as GA2 oxidase (GA2ox) and GA20ox, is important for the production of biologically active GAs in the final steps13,18. In monocots, mutations in GA3ox (GA 3β-hydroxylase), such as Dwarf1 (D1) from maize and OsGA3ox1 and OsGA3ox2 (Dwarf18 or D18) from rice, exhibit dwarfism19,20. To date, several GA 3β-hydroxylase genes have also been characterized in dicot species21,22,23,24. The GA4 gene encoding a 3β-hydroxylase was reported to be involved in GA biosynthesis in Arabidopsis21. The GA4-related protein Le with GA 3β-hydroxylation activity is able to convert GA20 to bioactive GA1, and its mutant allele le leads to a dwarf phenotype in pea22. Moreover, some GA3ox genes have also been functionally characterized in cucurbit crops, such as watermelon, cucumber, and pumpkin25,26,27. Notably, unlike the DELLA GA signaling mutants, the dwarf phenotype of GA biosynthetic mutants can be rescued, in some cases, by the application of exogenous GAs28.

Watermelon (Citrullus lanatus L.) is an economically important cucurbit crop, which accounts for 7% of the vegetable production area worldwide29. In watermelon, four genes conferring dwarfism have been reported, including gene dw-1 and its allele dw-1s, and two independent loci dw-2 and dw-330,31,32,33. Recently, a recessive locus named dsh has been located on chromosome 7, and the gene Cla010726 encoding a GA20ox-like protein is recognized as the most possible candidate gene34,35. In this study, we fine mapped a new dwarf locus, Cldf (Citrullus lanatus dwarfism), and gene Cla015407, encoding a GA 3β-hydroxylase, was recognized as the best possible causal gene. Sequence analysis revealed that the fourth polymorphism site (a G to A transition) at the 3′ AG splice receptor site of the intron leads to a 13 bp deletion in the coding sequence of Cldf in dwarf line N21 and thus results in a truncated protein lacking the conserved domain for binding of 2-oxoglutarate. Examination of exogenous GA3 application confirmed that Cldf is a GA-deficient mutant. Phylogenetic analysis suggested that there may be three ancient GA3ox lineages in the common ancestor of Cucurbitaceae. This new dwarf mutant line as well as the cosegregated marker will be helpful for breeding new watermelon cultivars with a dwarfism phenotype.

Materials and methods

Plant materials and morphological characterization

Two watermelon inbred lines used as parents in this study, M08 and N21, were grown in a greenhouse on the campus of Northwest A&F University, Yangling, China. M08 is an ordinary inbred material with normal vines, while line N21 with short internodes shows a dwarfism phenotype. For inheritance analysis and causal gene identification, two distinct F1 generations (N21 × M08 F1 and M08 × N21 F1) were generated by bidirectional crossing with two parental lines N21 and M08. Then, ten plants for each F1 generation were self-pollinated and individually harvested to produce F2 segregating populations. Subsequently, seven M08 × N21 F2 populations with a total of 1474 plants and three N21 × M08 F2 populations with 618 individuals were used for linkage analysis and identification of candidate genes for Cldf. Germinated seeds of two parental lines, as well as the F1 and F2 progenies, were directly sown in plastic pots and transferred to greenhouses under natural conditions at the third-leaf stage.

The phenotypes were visually recorded twice at seedling and mature stages and then classified as dwarfism or normal. The deviation from the expected 3:1 segregation ratio in the F2 population was tested using the χ2 test. To investigate the plant height of parental lines and F1 progeny, five individuals for each generation were randomly selected and measured with an ordinary steel ruler. The length of 22 internodes for each plant was also recorded. Using the SPSS 21.0 software, Duncan’s test was used to evaluate the significance of statistical data.

Whole-genome re-sequencing of two parental lines

Genomic DNA from young leaves of two parental lines was extracted using the CTAB (cetyl trimethylammonium bromide) method36. The quality of extracted genomic DNA was examined on a 1% agarose gel, and the purity was checked by a Nanodrop2000 spectrophotometer (Thermo Scientific, Wilmington, DE). Using the Illumina HiSeq X Ten platform, the genomes of two parental lines were re-sequenced to generate 150 bp paired-end reads by BioMarker Co. (Beijing, China).

Data analysis and marker development

After removing the adaptors, reads with more than 10% unknown bases, and low-quality reads, the clean data were mapped onto the reference genome of watermelon 97103 (http://www.icugi.org/) using the BWA software37. Raw single-nucleotide polymorphism (SNP) and indel calling was carried out via SAMtools software38. Then, after discarding the low-quality SNPs with read depths <20, high-confident SNPs and indels were obtained and used to develop corresponding CAPS (cleaved amplified polymorphic sequence) markers with the Primer Premier 5 software (http://www.premierbiosoft.com/). The sequencing data are accessible in the NCBI database under accession numbers SAMN11080422 and SAMN11080423. To validate the genomic polymorphism sites, all the reads mapped on candidate genes were visually investigated and compared between two parental lines using the JBrowse software39.

Molecular mapping of the Cldf locus

To preliminarily locate the Cldf locus, we designed one polymorphic marker for each chromosome based on the high-confidence SNPs identified above. Then, these 11 markers were used to screen a small F2 segregation population with 96 individuals. After initial chromosome anchoring of the Cldf locus, new flanking markers were developed to genotype the small population. Subsequent to delimiting the dwarf locus to a primary mapping interval, a larger population was used to identify recombinants. Then, four new polymorphic markers in the primary mapping region were designed and used to screen the recombinants to narrow down the mapping interval. Primer information of all the polymorphic markers is listed in Supplementary Table S1.

Candidate gene prediction and pathway-related gene identification

The annotated genes in the final mapping interval were analyzed according to the reference genome 97103. The genomic and coding sequences of the candidate gene were independently amplified from M08, N21, and 97103 and were then sent for sequencing. The software Geneious (http://www.geneious.com) was used to perform sequence analysis.

To date, the GA biosynthesis and signaling transduction pathways have been well characterized, in which genes encoding different functional enzymes have been cloned13,16,18. We retrieved amino acid sequences of one CPS1 (At4g02780) gene, two KAOs (KAO1, At1g05160; KAO2, At2g32440), five GA20oxs (AtGA20ox1, At4g25420; AtGA20ox2, At5g51810; AtGA20ox3, At5g07200; AtGA20ox4, At1g60980; AtGA20ox5, At1g44090), three GID1s (GID1a, At3g05120; GID1b, At3g63010; GID1c, At5g27320), and five DELLAs (RGA, At2g01570; GAI, At1g14920; RGL1, At1g66350; RGL2, At3g03450; RGL3, At5g17490) from Arabidopsis via the TAIR database (www.arabidopsis.org). Using protein sequences as queries, the respective homologs were identified in watermelon via the Blastp program (E value setting of 1.0 × 10−5).

RNA extraction and qRT-PCR analysis

For tissue-specific analysis, the roots, leaves, stems, tendrils, and both male and female flowers were independently sampled from lines M08 and N21. To analyze the expression levels of pathway-related genes, two adjacent unexpanded internodes from apical shoots were also independently harvested from two parental lines.

Using the RNA Simple Total RNA Kit (Tiangen, China), total RNA was extracted from the harvested samples, and the first strand complementary DNA (cDNA) was synthesized via the FastKing RT Kit with gDNase (Tiangen, China). Amplification was performed in a 20 µL reaction volume containing 10.0 µL of SYBR Green Premix (TaKaRa), 1.0 µL of cDNA template (80 ng/µL), 0.8 µL of each primer (10 µM), and 7.4 µL of ddH2O. Using a StepOnePlus Real-Time PCR System (Applied Biosystems, Foster, USA), the PCR amplification conditions included pre-denaturation for 5 min at 95 °C, followed by 40 cycles of 95 °C for 10 s and 60 °C for 30 s. The housekeeping gene Cla007792 was used as an internal reference40, and the relative expression level for each gene (three biological and three technical replicates) was calculated using the 2−∆∆Ct method41. All gene-specific primers used in quantitative reverse transcription PCR (qRT-PCR) experiments are listed in Supplementary Table S1.

Measurement of endogenous GA3 application

Homozygous recessive individuals at the four-leave stage were selected from the F2 population and used to treat endogenous GA3. GA3 powder was first dissolved in a small amount of ethanol and then diluted with ddH2O to the final concentration (200 mg/L). Seedlings sprayed with an equal volume of the corresponding mixture (ethanol and ddH2O) without GA3 were used as a control. For each treatment, nine seedlings were chosen and sprayed with endogenous GA3 or the corresponding mixture at 4-day intervals six times. Then, the plant height was measured with ordinary tapeline. After removing two maximum and two minimum values, the statistical data (five for each treatment) were analyzed with Student’s t test to evaluate the significance.

GA3ox homolog identification and phylogenetic analysis

For genome-wide identification of GA3oxs in watermelon, the amino acid sequence of the Cla015407 gene was used as a query to blast against the predicted protein file (v1, reference genome 97103) using the Blastp program (E value cutoff of 1.0 × 10−10). The reliability of candidate ClGA3oxs was validated through searching against the TAIR and NCBI databases. Then, using the reliable ClGA3ox candidates as queries, GA3ox homologs were genome widely identified from cucumber (Cucumis sativus, v3), melon (C. melo, v4), pumpkin (C. maxima, v1.1), and bottle gourd (L. siceraria, v1). All the predicted protein files of cucurbit species were downloaded from CuGenDB (http://cucurbitgenomics.org/). The amino acid sequences of published GA3oxs in different species, such as watermelon (ClGA3ox1)27, cucumber (CsGA3ox1 to CsGA3ox4)25, pumpkin (CmGA3ox1 to CmGA3ox4)26, tomato (SlGA3ox1 and SlGA3ox2)42, grape (VvGA3ox1 to VvGA3ox3)43, Arabidopsis (AtGA3ox1 to AtGA3ox4)44, soybean (GmGA3ox1 to GmGA3ox6)45, maize (ZmGA3ox1 and ZmGA3ox2)20, and rice (OsGA3ox1 and OsGA3ox2)19, were retrieved according their GenBank accession number or gene ID (Supplementary Table S2).

Multiple sequence alignment of full-length proteins was constructed via the Muscle software46. A neighbor-joining tree was generated with 1000 bootstrap replicates using MEGA 6.047.

Results

Phenotypic characterization and inheritance of the dwarfism trait

Compared with the normal line M08, the dwarf inbred line N21 with smaller leaves showed compact plant architecture (Fig. 1a). The objective phenotype can be visibly distinguished at the seedling stage (Supplementary Fig. S1a) and obviously classified as dwarfism or normal throughout the whole development stage. In addition, other morphological traits are also different between the two lines, such as leaf size, shape index, and trichome density of ovaries, and petals of male flowers (Supplementary Fig. S1b–d). Notably, the margin of the young leaf was slightly curled in line N21, and its growth vigor was much weaker than M08 (Fig. 1a). To compare the internode length and plant height among two parents and their F1 progenies, five individuals for each line were randomly selected. The internode length of N21 (4.0 ± 0.8 cm, 22 internodes) was much shorter than that of M08 (9.6 ± 1.7 cm, 22 internodes), and the plant height of the former (88.0 ± 6.1 cm) was also significantly less than that of the latter (211.6 ± 8.9 cm) (Fig. 1b). The plant height of reciprocally crossed F1 plants, as well as internode length, was also significantly higher than the N21 dwarf line, but significantly less than the ordinary line M08, indicating that the normal vine phenotype is dominant to dwarfism.

Fig. 1: Phenotypic characterization and statistical data analysis of watermelon dwarf line N21, normal line M08, and reciprocal crossing of F1 hybrids.
figure 1

a Morphological characterization of four watermelon lines. Magnified views of unexpanded leaves in two parental lines show different morphological types of leaf margins. The numbers in red represent internodes in the main stems. Bar = 5 cm. b The plant height and internode length of four watermelon lines. The blue and green bars represent plant height (five individuals for each line) and average internode length (22 internodes for each individual) data, respectively. The data are presented as the mean ± SD. Duncan’s test was conducted for statistical analysis. Different letters refer to significance at p < 0.05. c The number of dwarf and normal phenotype individuals in reciprocally crossed F2 segregation populations.

Full size image

To analyze the inheritance of the dwarfism phenotype, we collected phenotypic data from two reciprocal F2 segregation populations. As shown in Fig. 1c, there were 315 normal and 106 dwarf plants in the M08 × N21 F2 population (total 421 individuals), fitting a 3:1 Mendelian ratio (χ2 = 0.0008, p = 0.93). Moreover, 364 N21 × M08 F2 individuals contained 266 normal and 98 dwarf vine plants, which was also consistent with the Mendelian ratio of 3:1 (χ2 = 0.62, p = 0.40). Taken together, these data suggest that the dwarfism phenotype in N21 is controlled by a single recessive gene and is hereafter designated Cldf.

Genome-wide identification of high-confidence SNPs and indels

To obtain enough SNPs and indels for developing polymorphic marks, genomes of two parental lines were re-sequenced. After removing low-quality reads, we obtained a total of 39.0 and 36.6 million clean reads for lines M08 and N21, respectively, with ~11.7 and 11.0 GB of data and Q30 values above 93.0%, respectively (Table 1). Then, 98.59% and 98.25% of these clean reads for M08 and N21, respectively, could be successfully mapped on the reference genome, resulting in a total of 936,540 SNPs and 165,962 indels between two genomes. After discarding the low-quality sites with read counts <20, 152,894 high-confidence SNPs and 4018 indels were obtained and utilized to develop CAPS markers in the mapping strategy (Supplementary Table S3).

Table 1 Detailed characteristics of the DNA-seq data of M08 and N21.
Full size table

Linkage mapping of the dwarfism locus Cldf

A recent study reported that the Cla010726 gene on chromosome 7 encoding a GA20ox-like protein functions as the most possible candidate gene in watermelon mutant dsh35. To validate whether Cla010726 is the causal gene in line N21, a nearby polymorphic marker W12181814 was designed to screen a small M08 × N21 F2 population (96 individuals: 70 normal and 26 dwarf plants, p = 0.64 in χ2 test against 3:1 segregation ratio). Linkage analysis indicated that marker W12181814 was not linked with the dwarfism trait, inferring that the underlying gene Cldf in N21 is not Cla010726. To locate the Cldf gene on the chromosome, polymorphic markers were designed for the other ten chromosomes (data not shown) and then used to genotype individuals in the small population mentioned above. As a result, the marker W12181817 on chromosome 9 was confirmed to be linked with the dwarfism locus (Fig. 2a). Then, another three polymorphic markers (W0102191, W0114197, and W12181818) were designed to screen this small population. Subsequent linkage analysis implied that the dwarfism locus was delimited to a 5.97 Mb genomic region between markers W0102191 and W0114197, with 1 and 26 recombinants, respectively. Then, to narrow down this mapping interval, three new polymorphic markers (W1222182, W1221186, and W0114196) were designed to screen the 27 recombinants. Finally, the dwarfism trait was delimited within a 235.67 kb region between markers W1222182 and W1221186 (Fig. 2a), with 1 (1.04 cM) and 3 (3.13 cM) recombinants, respectively.

Fig. 2: Map-based cloning of the dwarfism gene Cldf.
figure 2

a Primary mapping of Cldf. The Cldf gene was preliminarily located between markers W1222182 and W1221186 on chromosome 09. The numbers within brackets indicate the number of recombinants between markers and phenotypes. b Fine mapping of the Cldf gene. Gene Cldf was finally delimited in a 32.88 kb region between markers W1222183 and W1221185. Marker W1221184 cosegregates with the phenotypes. c Schematic diagram of predicted genes. Three genes were annotated in the mapping interval, and the GA3ox homolog Cla015407 (in red) encoding a gibberellin 3β-hydroxylase was considered the most possible candidate gene. Another flanking GA3ox homolog outside the mapping region, Cla015408, was marked in blue.

Full size image

To further narrow down the initial mapping interval, the remaining larger segregating populations, including 325 M08 × N21 F2 and 364 N21 × M08 F2 individuals, were subjected to genotype with the primary flanking markers W1222182 and W1221186. Another eight new recombinants were identified from the M08 × N21 F2 generation, while only one recombinant was from the F2 offspring of N21 × M08. A total of 13 recombinants were used for further mapping of the dwarfism locus. Four new polymorphic markers (W1222183, W1222184, W1222185, and W0308192) were developed to genotype the 13 recombinants (Fig. 2b). Finally, the Cldf gene was delimited between markers W1222183 and W0308192, with one and seven recombinants, respectively. Two markers, W1221184 and W1221185, cosegregated with the phenotype. The physical distance of the fine mapping region was 106.72 kb, with two boundary markers W1222183 and W0308192 at 0.24 and 1.66 cM away from the dwarf locus, respectively.

To further precisely map the Cldf gene, additional individuals, including 1053 from the M08 × N21 F2 population and 254 from N21 × M08 F2, were genotyped with the above primary flanking markers W01222182 and W1221186. An additional 17 recombinants were obtained, including 15 from the M08 × N21 F2 population and 2 from N21 × M08 F2, which were also subjected to genotype with the four markers (W1222183, W1222184, W1222185, and W0308192). As a result, the Cldf locus was finally narrowed down to a 32.88 kb region between markers W1222183 and W1222185, with two recombinants for each marker (Fig. 2b). One cosegregated marker, W1221184, was obtained, which can be used for marker selection breeding programs.

Identification of candidate genes for the dwarfism gene Cldf

According to the annotated version of the reference genome, only three genes were annotated in the fine mapping region (Fig. 2c). Two homologous genes, Cla015405 and Cla015406, encode 2-oxoglutarate-dependent dioxygenase protein. The third Cla015407 gene is predicted to encode a GA 3β-hydroxylase (also named GA3ox), which is predicted to be involved in the final step of GA biosynthesis13,16,18. It is worth noting that another GA 3β-hydroxylase coding gene, Cla015408, is located outside the mapping interval, sharing 83.28% amino acid similarity with Cla015407. According to our re-sequencing data, we first analyzed the genomic polymorphisms of these four genes between two parents (Supplementary Fig. S2). As a result, no polymorphic sites were found in three genes (Cla015408, Cla015405, and Cla015406), while four SNP mutations and one indel mutation were identified in Cla015407 between the two parental lines. Hence, we proposed that the Cla015407 gene is the most likely causal gene underlying the dwarfism phenotype in the N21 line.

According to the genome annotation, the total nucleotide length of Cla015407 is 1257 bp and contains two exons (503 and 631 bp) and a 123 bp intron (Fig. 3a). To confirm the genomic variations observed above, we cloned the genomic sequence of this candidate gene from N21 and M08 and then compared them with the reference sequence from genome 97103, which is an East Asia watermelon cultivar with normal vines29. Undoubtedly, a total of five SNPs/indels were obtained, with four in the intron and one existing in the second exon (Fig. 3a). The first three polymorphisms in the intron, as well as that in the exon, were predicted with no effect on the gene structure or amino acid sequence changes.

Fig. 3: Sequence analysis of candidate gene Cla015407 among N21, M08, and reference genotype 97103.
figure 3

a Schematic diagram of genomic variations of Cla015407 among three genotypes. The physical positions of four SNPs and one indel between two parental lines are presented. The third SNP (G to A) in line N21 is predicted to lead to a 13 bp deletion in the coding sequence of Cldf. b Sequence alignment of the coding sequence of Cla015407 among the three genotypes. The 13 bp deletion was confirmed in the cDNA in line N21. c Alignment of predicted amino acid sequences of Cla015407 among three genotypes. The translation frameshift started at the 168 residue (C to W). A premature stop codon in Cldf resulted in a truncated protein with only 173 amino acid (aa) residues. The synonymous mutation in the exon with no aa conversion is also represented.

Full size image

Numerous studies have confirmed that introns probably possess a dinucleotide GT (splice donor site) at the 5′ boundary and an AG dinucleotide pair (splice receptor site) at the 3′ end48. The fourth mutation in the intron (G to A in line N21) may affect the original splicing of the intron and result in a 13 bp deletion in CDS of Cldf (Fig. 3a). To validate this assumption, we cloned the coding sequences of ClDF and Cldf alleles from two parents and the reference 97103 genotype. Sequence alignment showed that the cDNA sequences of ClDF in both M08 and 97103 are 1134 bp long and predicted to encode 377 amino acid residues, while a 13 bp deletion was found in Cldf occurring exactly at the fourth point mutation mentioned above (Fig. 3b). Moreover, this deletion could lead to frameshift translation and a premature stop codon, producing a truncated protein with only 173 amino acid residues (Fig. 3c). It is worth noting that the premature stop codon caused the lack of the conserved motif NyYPXCXXP (Supplementary Fig. S3) in Cldf, which is involved in the binding of 2-oxoglutarate43. Hence, we inferred that the fourth SNP in the intron is the causal mutation, which changes the function of Cldf in the dwarf line.

Expression analysis of ClDF/Cldf alleles and pathway-related genes

We examined the transcript abundance of Cla015407 by qRT-PCR in roots, leaves, stems, tendrils, and male and female flowers (Fig. 4). Compared with roots, the expression of ClDF in M08 was upregulated in stems and male flowers, with the highest transcript accumulation in male flowers. In dwarf line N21, the mutant allele Cldf was increased in stems, male flowers, and tendrils, of which the latter showed the highest transcriptional abundance. Similar expression patterns were observed in the five organs between two parental lines, except for tissue tendrils (Fig. 4).

Fig. 4: Expression analysis of ClDF and Cldf in different tissues of two parental lines.
figure 4

The transcriptional level of the respective gene in roots (M08) was set to a value of 1 and used as a reference. The data are shown as the mean ± SD. Different letters refer to significance at p < 0.05 (Duncan’s test). FF = female flowers; MF = male flowers.

Full size image

To date, the GA biosynthesis and signaling transduction pathways have been well characterized, and genes encoding different functional enzymes at each step in the pathways have been cloned (Supplementary Fig. S4)13,16,18. Using the amino acid sequences of the CPS1 gene, two KAOs (KAO1 and KAO2) and five GA20oxs (AtGA20ox1, AtGA20ox2, AtGA20ox3, AtGA20ox4, and AtGA20ox5) from Arabidopsis as queries45,49,50, we identified one CPS1 homolog (Cla006048), three KAO homologs (Cla021351, Cla006992, and Cla016164), and five GA20ox homologs (Cla002362, Cla006227, Cla008413, Cla013892, and Cla006941) in watermelon genome 97103 (Supplementary Table S4). The expression level of the dwarf candidate gene Cla010726 was also analyzed, which was published recently and predicted to encode a GA20ox-like protein35. Compared with the expression pattern in line M08, all four genes involved in the GA biosynthesis pathway were upregulated in unexpanded internodes of N21 (Fig. 5a). Similarly, three GA20ox homologs (Cla002362, Cla006227, and Cla010726) were significantly upregulated in dwarf line N21, while only the Cla013892 gene was downregulated compared to that in line M08. The GA receptor GID1 that was first identified in rice contains three orthologous copies (AtGID1a, AtGID1b, and AtGID1c) in Arabidopsis, which were confirmed with some overlapping but also distinct functions in plant developmental processes51. The active GA-GID1 complex could trigger the rapid degradation of DELLA proteins via the 26S proteasome pathway, which act as GA signaling repressors and contain five members (RGA, GAI, RGL1, RGL2, and RGL3) in Arabidopsis16,52,53,54. In watermelon, we identified two GID1 (Cla014721 and Cla011311) and five DELLA (Cla003932, Cla019759, Cla013228, Cla012302, and Cla011849) homologous genes (Supplementary Table S4). Interestingly, the transcription levels of GID1 homolog Cla014721 and DELLA gene Cla013228 were induced in dwarf line N21, while the expression of genes Cla011311 (GID1) and Cla019759 (DELLA) was significantly reduced in N21 compared to that in normal line M08 (Fig. 5b). Additionally, the transcription levels of the other three DELLA genes were not obviously different between the two parental lines.

Fig. 5
figure 5

Expression analysis of GA biosynthesis (a) and signaling transduction (b) pathways related genes in the stems of two parental lines. The transcriptional level of the respective genes in stems (M08) was set to a value of 1 and used as a reference. The data are shown as the mean ± SD. *,** represent significant differences in expression levels at p < 0.05 and p < 0.01, respectively (Student’s t test).

Full size image

Recovery of dwarfism phenotype by exogenous GA3 application

It has been reported that the dwarf phenotype of GA biosynthetic mutants can be rescued, in some cases, by the application of exogenous GA34,28. Therefore, we investigated the phenotypes of homozygous recessive individuals from the F2 population, which were treated with exogenous GA3 application (200 mg/L, detailed in Materials and methods section). As shown in Fig. 6, plant heights could be rescued by the application of GA3. In addition, the leaf color and leaf margin treated with exogenous GA3 were also restored to M08 appearance (Supplementary Fig. S1e).

Fig. 6: Recovery of the Cldf mutant by exogenous GA3.
figure 6

a Phenotypes of Cldf mutant seedlings that were treated with exogenous GA3 (200 mg/L). Red arrows indicate 20 cm (bottom) and 70 cm (top). b Plant heights of Cldf mutant seedlings treated with exogenous GA3 (200 mg/L). Plants sprayed with an equal volume of the corresponding mixture without GA3 were used as a control in this experiment. Nine seedlings were used for each treatment (details in “Materials and methods”). After removing two maximum and two minimum values, the statistical data (five for each treatment) were analyzed with Student’s t test to evaluate the significance. ** indicates a statistically significant difference in expression level at p < 0.01.

Full size image

Phylogenetic analysis of candidate gene Cldf

Increasing evidence has shown that enzymes involved in the final steps of the GA biosynthesis pathway are encoded by small multigene families19,20,42,43,44. Recently, GA3ox genes have been identified and cloned from several species, such as tomato, grape, Arabidopsis, soybean, maize, and rice19,20,42,43,44,45. Using Cla015407 as a query, we identified four ClGA3ox homologs in watermelon, which displayed two exons according to the gene annotation (Supplementary Table S5). A phylogenetic tree was constructed with protein sequences of four ClGA3oxs and homologs from tomato (two SlGA3oxs), grape (three VvGA3oxs), Arabidopsis (four AtGA3oxs), soybean (six GmGA3oxs), maize (two ZmGA3oxs), and rice (two OsGA3oxs) (Fig. 7). Homologs from monocot species (maize and rice) formed an independent lineage (group III) in the distance tree, while those from dicot genomes could be divided into two groups (I and II), which is similar to the topologies observed in published studies43,55. In the dicot lineage, group I contained homologs from five dicotyledon species, including the target gene Cla015407 and its flanking homolog Cla015408, while members in group II were only from three species, inferring their ancient origination in dicot plants.

Fig. 7: Phylogenetic analysis of the GA3ox gene family in watermelon and six other species.
figure 7

Three groups can be found in the distance tree. The candidate gene Cla015407 is highlighted in red. Numbers on nodes represent bootstrap values.

Full size image

To further investigate the evolutionary history of the GA3ox family in Cucurbitaceae, we identified 24 homologs genome-wide from four other cucurbit species, including C. sativus, C. melo, C. maxima, and L. siceraria (Supplementary Table S5). Notably, all the cloned GA3oxs in C. lanatas, C. sativus, and C. maxima were included in our identification25,26,27. Then, a total of 28 GA3oxs from Cucurbitaceae together with 19 homologs from other families were aligned to generate a distance tree, which could also be divided into three groups (Supplementary Fig. S5). Interestingly, members from cucurbit genomes mingled together in group I, while those from other species formed independent clades, which is similar to subgroup IIb. In subgroup IIa, GA3ox homologs were only from cucurbit species, inferring that this lineage may be specific to Cucurbitaceae.

Discussion

Watermelon is an important cucurbit crop worldwide, which accounts for 7% of the global vegetable production area29. Plant height in watermelon is a vital agronomic architecture trait that can increase fruit yield and reduce labor costs in crop cultivation and pruning. There are four reported genes conferring dwarfism in watermelon, including gene dw-1 and its allele dw-1s, as well as independent loci dw-2 and dw-330,31,32,33. In a previous study, a new recessive locus named dsh was located on chromosome 7, and the Cla010726 gene encoding a GA20ox-like protein was recognized as the best possible candidate gene34,35. Linkage analysis indicated that gene Cla010726 is not the causal gene in dwarf line N21, although similar morphological traits were observed in both compact materials, such as numerous branches and small and curled leaves (Fig. 1a)34,35. Moreover, our work presented herein suggested that a GA 3β-hydroxylase encoding gene Cla015407 was recognized as the best possible candidate gene leading to the dwarfism phenotype in line N21 (Figs. 2 and 3). Sequence analysis identified four SNPs and one indel in the genomic sequences of Cla015407 between two parental lines (Fig. 3a and Supplementary Fig. S2). Numerous studies have confirmed that introns probably contain the canonical splice model possessing the consensus 5′ GT splice donor site and the 3′ AG splice receptor site48. The fourth polymorphic site (G to A) disrupted the original splicing site of the intron in Cldf, resulting in a 13 bp deletion in the coding sequence (Fig. 3b). The mutant allele Cldf, which carries a premature stop codon that produces a truncated protein with only 173 amino acid residues, lacks the conserved motif NyYPXCXXP (Supplementary Fig. S3), which is considered to be involved in the binding of 2-oxoglutarate43. In addition, the dwarf phenotype could be recovered by the application of exogenous GA3 (Fig. 6). Overall, it is reasonable to speculate that the GA3ox homolog Cla015407, which is involved in the final step of the GA3 biosynthesis pathway, is the causal gene for the dwarfism phenotype in watermelon line N21.

The biosynthesis of active GAs is a complex and multistep process that recruits different functional enzymes to catalyze diverse intermediates (Supplementary Fig. S4). Gene CPS functioning in an early step of the GA biosynthetic pathway can convert GGDP to CDP in plastids, while KAO in the endoplasmic reticulum catalyzes the conversion of ent-kaurene acid GA1212,16,17. GA3ox and GA20ox play important roles in the final steps of the GA biosynthesis pathway13,18. In our study, the transcription levels of both CPS and KAO homologs, as well as three GA20ox members, were upregulated in dwarf line N21 compared to that in M08 (Fig. 5a), suggesting possible feedback regulation. Consistent with our conclusion, the levels of GA3ox and GA20ox transcripts increased in the kao1 kao2 double mutant of Arabidopsis49. The dynamic balance of active GAs in plants is maintained by DELLA-dependent feedback regulation of GA biosynthesis genes12. The active GA-GID1 complex could trigger rapid degradation of the master GA signaling repressor DELLA proteins16, while increasing DELLA activity obviously results in the accumulation of GA3ox1 and GA20ox1 transcripts49,56. The expression of GA4 (AtGA3ox1) was reduced 26% by GA3 treatment in the Arabidopsis rga (DELLA) mutant56. In the present study, five DELLA homolog genes were identified in watermelon (Supplementary Table S4), of which only one gene, Cla019759, was repressed in dwarf mutant Cldf (Fig. 5b). As the GA receptor, three GID1 orthologous copies (GID1A, GID1B, and GID1C) were confirmed to have overlapping but also functional specificity in regulating different developmental processes51,57. Moreover, GA treatment resulted in feedback inhibition of all three AtGID1 genes57. In this study, expression analyses revealed that the two ClGID1s exhibited distinct expression patterns between two parental lines (Fig. 5b), suggesting their possible distinct functions.

Numerous studies have shown that enzymes involved in the final steps of the GA biosynthesis pathway are encoded by small multigene families19,20,42,43,44. In plants, GA3oxs convert GA12 to bioactive GAs in the final step of the biosynthesis pathway13,16,18. In Arabidopsis, four GA3ox homologs have been identified and designated AtGA3ox1 to AtGA3ox4, which exhibit organ-specific expression patterns and some degree of functional redundancy44. Similarly, two GA3ox members in rice, OsGA3ox1 and OsGA3ox2, also showed different expression patterns19. In cucumber and pumpkin, four GA3ox genes have been characterized in each genome with possible redundant and specific functions25,26,55. In addition, a previous study reported that gene Cv3h (a GA3ox homolog and identified as Cla022286 in this study) may function in the developing seeds of watermelon27. Here, we infer that another GA3ox homologous gene, Cla015407, is responsible for internode elongation in dwarf line N21. Moreover, the germination rate of N21 seeds is much lower than that of M08 (data not shown), inferring that these two ClGA3oxs may have overlapping and specific functions. To further recover the evolution of the GA3ox family in plants, a phylogenetic tree was constructed with 28 GA3oxs from Cucurbitaceae and 19 homologs from other families. As shown in Supplementary Fig. S5, homologs prefer to gather together at the family level in different groups/subgroups, which is consistent with the observations in previous studies43,45,55. Additionally, it seems that there are three ancient lineages in the common ancestor of Cucurbitaceae, and one of them (subgroup IIa) is specific to cucurbit crops.