A mutation in the intron splice acceptor site of a GA3ox gene confers dwarf architecture in watermelon (Citrulluslanatus L.)

Dwarf architecture is an important trait associated with plant yield, lodging resistance and labor cost. Here, we aimed to identify a gene causing dwarfism in watermelon. The ‘w106’ (dwarf) and ‘Charleston Gray’ (vine) were used as parents to construct F1 and F2 progeny. Dwarf architecture of ‘w106’ was mainly caused by longitudinal cell length reduction and was controlled by a single recessive gene. Whole-genome sequencing of two parents and two bulk DNAs of F2 population localized this gene to a 2.63-Mb region on chromosome 9; this was further narrowed to a 541-kb region. Within this region, Cla015407, encoding a gibberellin 3β-hydroxylase (GA3ox), was the candidate gene. Cla015407 had a SNP mutation (G → A) in the splice acceptor site of the intron, leading to altered splicing event and generating two splicing isoforms in dwarf plants. One splicing isoform retained the intron sequences, while the other had a 13-bp deletion in the second exon of GA3ox transcript, both resulting in truncated proteins and loss of the functional Fe2OG dioxygenase domain in dwarf plants. RNA-Seq analysis indicated that expression of Cla015407 and other GA biosynthetic and metabolic genes were mostly up-regulated in the shoots of dwarf plants compared with vine plants in F2 population. Measurement of endogenous GA levels indicated that bioactive GA4 was significantly decreased in the shoots of dwarf plants. Moreover, the dwarf phenotype can be rescued by exogenous applications of GA3 or GA4+7, with the latter having a more distinct effect than the former. Subcellular localization analyses of GA3ox proteins from two parents revealed their subcellular targeting in nucleus and cytosol. Here, a GA3ox gene controlling dwarf architecture was identified, and loss function of GA3ox leads to GA4 reduction and dwarfism phenotype in watermelon.

Bulk DnA construction and illumina sequencing. DNAs were extracted using the CTAB method from leaves of both parental lines and F 2 plants for BSA-Seq. Two bulk DNA samples, dwarf bulk (D-bulk) and vine bulk (V-bulk), were constructed by mixing equal amounts of DNAs from 25 dwarf plants and 25 vine plants from the F 2 population, respectively. Paired-end DNA libraries were prepared according to the manufacturer's instructions (Truseq Library Construction; Illumina, San Diego, CA, USA). First, genomic DNA was sheared into 350-bp fragments using a Covaris S220 sonicator (Woburn, MA, USA). Second, ends of the gDNA fragments were repaired, and 3′ ends were adenylated. Then, the size distributions and concentrations of the libraries were determined using an Agilent 2100 Bioanalyzer (Agilent Technologies, Waldbronn, Germany) and quantified by real-time PCR. Finally, DNA libraries were sequenced using an Illumina HiSeq X at Genepioneer (Nanjing, Jiangsu, China) according to the manufacturer's instructions for paired-end 150-bp reads.
The short reads from D-bulk and V-bulk were aligned to the '97103' reference genome 32 using BWA software 46 . Alignment files were converted to SAM/BAM files using SAM tools 47 . SNPs and Insertion/deletion polymorphisms (InDels) were also assessed.
Gene location association analysis. All samples underwent variant calling using the Unified Genotyper function of the GATK program 48 . The SNPs and InDels were filtered using the Variant Filtration parameter of GATK. ANNOVAR, which is an efficient software tool, was used to annotate the SNPs or InDels based on the GFF3 files for the reference genome 49 . The homozygous SNPs/InDels between two parental lines were extracted from the vcf files.
A SNP index was used to indicate the proportion of reads harboring SNPs that differed from reference sequences 50 2 ] 1/251 . Using these formulae, we assessed whether the measured values fell within the following ranges, − 1 ≤ Δ(SNP-index) ≤ 1 and 0 ≤ ED ≤ 1.414 51,52 . The greater of the ED value, the closer of the object site 52 . The Δ(InDel-index) and EDs of InDel sites were calculated using the InDel-index 53 as described above for calculating for SNP regions. Using a 1-kb sliding window, an average SNP/InDel-index was calculated over a 1-Mb interval.
Mapping of the candidate gene. To minimize the genetic interval and verify the accuracy of BSA-Seq, 161 simple sequence repeat (SSR) markers within the BSA-Seq region were developed based on the wholegenome sequencing of the two parental lines. These newly developed SSR markers were first screened for polymorphisms between the two bulk DNAs, then the polymorphic SSRs were used to screen for recombinants in the dwarf individuals of the F 2 population.

Transcriptome sequencing of dwarf and vine plants.
Transcriptome sequencing (RNA-Seq) was performed to analyze the expression of candidate genes and reveal the related pathways involved in dwarf architecture. Total RNAs were extracted from shoots of dwarf and vine plants in the F 2 population using TRIzol Reagent (Invitrogen, Carlsbad, CA, USA). Approximately 10 µg total RNA was subjected to Poly(A) mRNA isolation using poly-T oligo-attached magnetic beads (Invitrogen, Carlsbad, CA, USA). Following purification, the mRNA was fragmented into small pieces and the cleaved RNA fragments were reverse-transcribed to create the final cDNA library. The average insert size for the paired-end libraries was 300 bp (± 50 bp). Then, the paired-end sequencing was performed on an Illumina HiSeq 4000 platform at Genepioneer (Nanjing, Jiangsu, China) following the vendor's recommended protocol. Triplicates of each sample were carried out for Illumina sequencing.
Gene expression was assessed using the fragments per kilobase of transcript per million fragments mapped (FPKM) method 54 . The differentially expressed genes (DEGs) were determined using the criterion |log 2 (Fold Change)|≥ 1 and FDR < 0.05. The corresponding functions were revealed using the KEGG Automatic Annotation Server 55 .
Measurement of endogenous GA levels using internal standards. At the fourth-true leave stage, the shoots of vine and dwarf plants in two parents and F 2 population grown under the same condition were harvested. The endogenous levels of 18 kinds of GAs involved in GA biosynthetic and metabolic pathway (GA 1 , GA 3 , GA 4 , GA 5 , GA 6 , GA 7 , GA 8 , GA 9 , GA 12 , GA 15 , GA 19 , GA 20 , GA 24 , GA 29 , GA 34 , GA 44 , GA 51 and GA 53 ) were measured using 2H2-GA internal standards coupled with UPLC-MS/MS analyses at LC Sciences (Hangzhou, China). For the measurement, shoots from three seedlings of vine and dwarf plants were mixed and ground into fine powder in liquid nitrogen, respectively; weighed ~ 100 mg sample, added 1.0 ng/g corresponding 2H2-GA internal standards and 1 mL methanol/water (80/20, v/v) extracting solution at 4 °C for overnight, then centrifuged at 10,000×g for 20 min at 4 °C; the supernatant was taken and absorbed through the C18/SAX solid-phase extraction column; 2 mL methanol/water (20/80, v/v) was used to clean the SAX extraction column, and 3 mL ACN/FA (99/1, v/v) was used to desorb the target acid plant hormones retained on the SAX extraction column; the desorption solution was blow-dried with constant nitrogen flow at 40 °C and redissolved in 100 μL water, and 10 μL FA was added to the 100 μL solution and extracted twice with ether; the extracted organic phase was combined and blow-dried by nitrogen at room temperature, then dissolved in 100 μL ACN; adding 10 μL × 20 μmol/ mL TEA and 10 μL × 20 μmol/mL BTA to the ACN solution, swirl for 35 min at room temperature and then blow-dried with nitrogen; redissolved the solution in 200 μL H 2 O/ACN (90/10, v/v) for subsequent UPLC-MS/ MS analyses. The UPLC-MS/MS analyses were performed on Thermo Scientific Ultimate 3000-Thermo Scientific TSQ Quantiva. Three biological replicates were conducted for each measurement, and endogenous levels of GAs (ng/g fresh weight) were determined by means of three UPLC-MS/MS detection results. T-test was conducted for statistical analysis using SAS 8.0. The GAs with |log2Fold(FC)| ≥ 1 and statistical significance (p value < 0.05) were considered as significant difference. www.nature.com/scientificreports/ Exogenous treatments with GA 3 or GA 4+7 . In addition, the dwarf parental lines were employed to assess responses to exogenous application of GAs. The GA 3 (Ryon, Shanghai, China) and GA 4+7 (Ryon, Shanghai, China) are independently dissolved in a small amount of ethanol and then diluted with sterilized ddH 2 0 to the final concentration of 500 μM. The seedlings were sprayed independently with 500 μM GA 3 or GA 4+7 for four times at 3-day intervals. The control seedlings were sprayed with sterilized ddH 2 0. Three seedling were used for each treatment. The phenotypes of seedlings were analyzed and photographed 3-days after the last GA treatment.

Results
phenotypic and genetic analyses. The height of 'w106' was significant shorter than that of 'CG' (Fig. 1a).
Therefore, we conducted a microscopic observation of shoots of the 'CG' and 'w106' plants using paraffin sectioning. The cell sizes in transverse sections were not obviously different between the 'CG' and 'w106' plants ( Fig. 1b). However, the cell lengths in longitudinal sections were obviously shorter in 'w106' than in 'CG' plants ( Fig. 1c). Thus, the defective cell elongation appears to be the main cause for the reduced shoots and dwarf architecture in watermelon.
To assess the inheritance of the dwarf trait in watermelon, crosses were made between 'w106' and 'CG' . All the F 1 plants showed the vine phenotype. Among the 98 F 2 progeny, 72 individuals showed vine phenotype and 26 individuals showed dwarf phenotype, with the Chi-square test (χ 2 ) confirming the segregation ratio to be 3:1 (Table 1). These results indicated that the dwarf trait in 'w106' was controlled by a single recessive gene, designated as the short internode (Si) locus.
Si gene located on chromosome 9 (0.80-3.43 Mb) using BSA-seq. We sequenced the genomes of the two parental lines and two bulk DNAs using the Illumina HiSeq™ PE150 platform. The high-throughput sequencing results obtained 50.30, 61.89, 65.48 and 70.68 Mb clean reads for female parent, male parent, D-bulk and V-bulk, respectively (Table 2). A total of 33.65 Gb clean data were generated for the two parental lines, and 40.84 Gb clean data were generated for the two DNA bulks, with approximately 42-57 × sequencing depth and more than 99.00% 5 × coverage per sample (Table 2). Data were aligned to the reference genome of watermelon '97103' (https ://cucur bitge nomic s.org/organ ism/1), and 160,957 SNPs and 55,055 InDels, at a minimum of 5 reads, were identified between D-bulk and V-bulk. Each identified SNP or InDel was used to compute an SNP/ InDel-index. The graph for Δ(SNP/InDel-index) was plotted and computed against the genome positions by combining SNP/InDel-index of D-bulk and V-bulk (Fig. 2a,b). At the 99% significance level, the Δ(SNP-index)  Table S2). The results indicated that a candidate gene controlling the dwarf trait in 'w106' was located in the 0.80-3.43 Mb region of chromosome 9 (Fig. 2c).
further mapping analysis narrowed Si to a 541-kb interval and a candidate gene was predicted. To further narrow down the location of the Si locus detected by BSA-Seq, we selected 161 SSR markers from chromosome 9 (0.80-3.43 Mb) based on resequencing data of the two parental lines. All these SSR markers were first screened for polymorphisms between the two bulk DNA samples, and then, 16 polymorphic markers were applied to screen recombinants of the dwarf individuals in the F 2 population. Finally, two flanking markers, dw37 (Chr9:1620039) and dw134 (Chr9:2161629), obtaining one and four recombinants, respectively, placed the Si locus in a 541-kb region (Fig. 2d). Additionally, no recombinant was obtained using the marker dw128 (Chr9:1835342), indicating that the target gene neighbored dw128. All the polymorphic SSR markers used in this study and the obtained recombinants are listed in Supplementary Table S3.
According to the watermelon genome annotation, 66 putative genes (Cla015361-Cla015427) were detected within the 541-kb interval (Supplementary Table S4). Within this region, three SNPs and one InDel were identified among the two parental lines and two bulk DNAs according to the whole-genome sequencing data (Table 3). Among these four variations, two SNP variations (Chr9:1620753 and Chr9:1621230) occurred in the intergenic region of the genome and an InDel (Chr9:1996536) occurred in the upstream of Cla015387 (WD43). One SNP occurred in Chr9:1857472 (from 'G' in 'CG' to ' A' in 'w106'), locating at the splice-site acceptor in the intron of Cla015407 ( Fig. 2e; Table 3). Cla015407 encodes GA3ox, which is involved in the GA biosynthetic pathway. Additionally, the genome location of Cla015407 (Chr9:1856847-1858103) neighbored the SSR marker dw128 (Chr9:1835342), which did not identify any recombinants and was near the target gene. Therefore, Cla015407 was predicted to be the candidate gene conferring the dwarf architecture of 'w106' .

Sequences analyses of the candidate gene.
To verify the sequences of Cla015407 at the DNA and mRNA levels, we cloned the DNA and coding sequence (CDS) of Cla015407 from both parental lines. Fragments of 1,257 bp were amplified at the DNA level from both parental lines (Fig. 3a), and the sequencing analysis further verified the G → A variation at the 626th nucleotide of Cla015407 (Fig. 3c).
At the cDNA level, a fragment was amplified from vine parent 'CG' and two fragments were amplified from dwarf parent 'w106' , which indicated that this SNP mutation lead to altered splicing, generating two splicing isoforms in the dwarf plants (Fig. 3b). Sequence analyses of the splicing isoforms in 'w106' revealed that the full-length isoform (isoform 1) retained the intron sequences and contained the premature termination codon 'TAG' at the 505-507th nucleotides (Fig. 3d). Additionally, the truncated isoform (isoform 2) had a 13-bp deletion in the second exon compared with the CDS of 'CG' and contained the premature termination codon 'TGA' at the 520-522th nucleotides (Fig. 3d).
The proteins encoded by the transcripts of Cl015407 in vine and dwarf parents were also predicted. The transcript of Cl015407 in vine parent 'CG' encodes a protein with 377 aa (Fig. 3e). However, the full-length isoform (isoform 1) in the dwarf parent 'w106' contains an unspliced intron, introducing a stop codon (TAG) just after the splice donor site, thus translation of this full-length transcript is prematurely terminated and produces a protein with 168 aa (Fig. 3e). Moreover, the truncated isoform (isoform 2) has a 13-bp deletion in the second exon of Cla015407 in dwarf parent 'w106' and contains a premature termination codon 'TAG' at 520-522th nucleotides, leading to frame shift and premature termination, and resulted in a truncated protein with 173 www.nature.com/scientificreports/ aa residues (Fig. 3e). In summary, the two transcripts of Cl015407 from the dwarf plants resulted in truncated proteins and lost the functional Fe2OG dioxygenase domain. www.nature.com/scientificreports/  www.nature.com/scientificreports/ pared with vine plants (Fig. 4a). In addition, the heatmap generating by Cluster 3.0 clearly divided these 3,017 DEGs into two Clusters (I and II) (Fig. 4b).

DEGs identification between dwarf and vine plants and their KEGG
The KEGG pathway enrichment analyses were carried out for these DEGs (Fig. 4c,d). The up-regulated genes in dwarf plants were significantly enriched in KEGG pathways of 'plant hormone signal transduction' and 'photosynthesis' (Fig. 4c). Moreover, the down-regulated genes in dwarf plants were significantly enriched in KEGG pathways of 'pentose and glucuronate interconversions' , 'biosynthesis of secondary metabolites' , 'stilbenoid, diarylheptanoid and gingerol biosynthesis' , 'starch and sucrose metabolism' , 'metabolic pathways' , 'phenylpropanoid biosynthesis' , 'cyanoamino acid metabolism' , 'linoleic acid metabolism' , 'alanine, aspartate and glutamate metabolism' , 'ascorbate and aldarate metabolism' , and 'phenylalanine metabolism' (Fig. 4d). Table 4, a total of 20 genes involved in GA biosynthesis and metabolism were detected as expressed at least one library, including one gene for CPS, one gene for KS, one gene for KO, two genes for KAO, six genes for GA20ox, two genes for GA3ox and seven genes for GA2ox. Most of these GA biosynthetic and metabolic genes in dwarf plants showed higher expression levels than those of vine plants. The expression of our candidate gene, Cla015407 (GA3ox), was significantly increased in dwarf plants. Moreover, one KO gene (Cla020710) and three GA20ox genes (Cla002362, Cla006227 and Cla008413) involved in GA biosynthesis, and two GA2ox gene (Cla015162 and Cla019586) involved in GA metabolism, were also significantly up-regulated in dwarf plants compared with those in vine plants.

Endogenous levels of GAs were changed in the dwarf plants.
Except for GA 7 , the remaining 17 kinds of GAs were detected from the shoots of vine and dwarf plants in two parents and F 2 population using the 2H2-GA internal standards coupled with UPLC-MS/MS analyses. Among these GAs, endogenous level of GA 3 was significantly increased and GA 4 was significantly decreased in dwarf plants of two parents and F 2 population (Table 5). Additionally, endogenous levels of GA 9 and GA 29 were significantly increased in dwarf plants of F 2 population. The results indicated the reduced level of bioactive GA 4 might be the main cause for the dwarf phenotype in watermelon (Table 5). Moreover, we constructed the GA biosynthetic and metabolic pathway combining the endogenous levels of GAs and genes involved in this pathway (Fig. 5). Two major pathways, GA 53 -pathway (involving GA 53 , GA 44 , GA 19 , GA 20 , GA 1 , GA 8 , GA 29 , GA 5 , GA 3 and GA 6 ) and GA 12 -pathway (involving GA 12 , GA 15 , GA 24 , GA 9 , GA 4 , GA 34 , GA 51 and GA 7 ), were included. The GA 9 and GA 4 in the GA 12 -pathway showed different changes of endogenous levels between the vine and dwarf plants ( Fig. 5; Table 5). Endogenous level of GA 9 was increased in dwarf plants of F 2 population, which might be due to the increased expression level of GA20ox genes (Cla002362, Cla006227 and Cla008413) in dwarf plants as indicated in Table 4. Endogenous level of bioactive GA 4 was decreased in dwarf plants, suggesting the mutation of Cla015407 (GA3ox) impaired the biosynthesis of GA 4 in GA 12 -pathway. Endogenous level of GA 29 and GA 3 in the GA 53 -pathway were increased in dwarf plants ( Fig. 5; Table 5). The increased GA 29 content might be attributed to the increased expression level of GA2ox genes (Cla015162 and Cla019586) in dwarf plants as indicated in Table 4. Endogenous level of bioactive GA 3 was increased in dwarf plants, which might be due to the up-regulation of another GA3ox paralogue, Cla022285, as a result of feedback of reduced GA 4 content caused by the mutation of Cla015407 (GA3ox) ( Table 4).
Exogenous GA applications can rescue the dwarf phenotype. We investigated the responses of dwarf plants to 500 µM GA 3 and 500 µM GA 4+7 applications and found that the independent applications of GA 3 or GA 4+7 could both rescue the dwarf phenotype in watermelon (Fig. 6). Moreover, GA 4+7 treatments had a more distinct effect than GA 3 treatments, resulting in greater plant height (Fig. 6). These observations further verified that the Si gene is a GA biosynthetic gene.  www.nature.com/scientificreports/

Subcellular localization of GA3ox proteins. The subcellular localization of GA3ox proteins from 'CG'
and 'w106' , namely CG-Cla015407, w106-Cla015407-Iso1, and w106-Cla015407-Iso2, were analyzed by transient expression of the green fluorescent protein (GFP) fusion proteins in tobacco leaf epidermal cells. As shown in Fig. 7, all of the three GA3ox proteins were localized in the nucleus and cytosol.

Discussion
Plant height, as the key component of plant architecture, has been associated with both natural beauty and yield performance. Total cell number and cell size, resulting from cell division and expansion, determine the size of plant organs 56 . For example, the average cell size of the cucumber dwarf mutant scp2 was significantly smaller than that of wild type plants 57 . The cucumber Csdw mutant showed dwarfing phenotype and decreased internode length due to the reduced cell division in main stem 58 . The rice stemless dwarf 1 (std1) mutant showed severe dwarfing phenotype, abnormal cell morphology and reduced cell division rate 59 . Further analyses revealed that a large number of cells were blocked in the S and G2/M phases, with the presence of many multinucleate cells 59 .
In present study, microscopic observations of stem transverse and longitudinal sections revealed that the reduced plant height could be mainly attributed to the shorter longitudinal cell lengths (Fig. 1b,c). Genetic mapping and identification of dwarfism genes have occurred in cucurbits, such as scp-2 57 , Csdw 58 , cp 60 and scp-1 61 in cucumber; mdw1 62 in melon; and qCmB2 63 in pumpkin. In watermelon, the genes of dsh 37 , Cldf 38 , dw 39 and Cldw-1 40 , conferring dwarf phenotypes, were studied and identified. NGS-assisted BSA is an effective method to identify target genes controlling the dwarf traits by genotyping the bulked DNA samples having distinct or opposite extreme phenotypes in plants 58 . In this study, we employed the BSA-Seq to identify the candidate gene controlling the dwarf trait in watermelon and delimited the region to 0.80-3.43 Mb on chromosome 9 (Fig. 2a-c). A further mapping analysis finally located this gene in a 541-kb interval (Fig. 2d), with Cla015407 (GA3ox) being the candidate gene.
GA3ox catalyzes the last step of bioactive GA synthesis, which converts GA 20 to GA 1 , GA 5 to GA 3 , and GA 9 to GA 4 18 . Mutations in GA3ox genes, such as dwarf1 (d1) from maize 18 , dwarf18 (d18) from rice 26 , GA4 from Arabidopsis 27 , Msdwf1 from Medicago sativa 28 , and le from pea 64 , resulted in dwarfism. For example, the coding sequences of GA3ox2 isolated from d18 alleles were analyzed in rice. In the strong allele, d18-Id18 h , the deletion of a guanine base at 750 altered the reading frame, and in the weak allele, d18-dy, the 9-bp deletion deleted three amino acids 26 . The dwarf mutant Msdwf1 had an amino acid change in a conserved position of the GA3ox gene compared with the wild type in Medicago sativa 28 . Sequence alignment revealed a G-to-A transition conferring an alanine-to-threonine substitution at residue 229 in the le gene product in pea 64 . The same locus of GA3ox, Cla015407, conferring the dwarf phenotypes in watermelon were concurrently identified in this study and those of Wei et al 38 39 .
In the present study, the candidate gene, Cla015407 (GA3ox), significantly increased the expression level in dwarf plants compared with the vine plants (Table 4). Additionally, one KO gene (Cla020710), three GA20ox genes (Cla002362, Cla006227 and Cla008413) and two GA2ox genes (Cla015162 and Cla019586) were also significantly up-regulated in dwarf plants ( Table 4). The increased expression of GA related genes in this study were partially consistent with Wei et al 38 , in which the GA3ox (Cla015407), CPS (Cla006048), KAO (Cla021351, Cla006992 and Cla016164), and GA20ox (Cla002362, Cla006227 and Cla010726) were significantly up-regulated in dwarf line 'N21' . Different from our results, the expression of GA3ox (Cla015407) in the dwarf parent 'Duan125' was significantly reduced according to the report of Gebremeskel et al. 39 . The increase in expression of GA related genes in our study might be due to the feedback of low levels of bioactive GA 4 in dwarf plants as showing in Table 5. This is a well-known phenomenon whereby mutations or chemical intervention in GA biosynthesis or GA perception result in increases in GA20ox and GA3ox in a homeostatic mechanism 14 . Moreover, two transcripts of GA3ox, the intron retention isoform and 13-bp deletion isoform, were generated in the dwarf parent 'w106' in our study. Introns are often added to increase expression, although the mechanism through introns stimulate gene expression in plants remains unclear 65 . For instance, inserting the first intron from the UBQ10 gene into intronless genes markedly increased the latter's mRNA accumulation to over 150-fold in Arabidopsis 65 . Therefore, the up-regulated GA3ox gene might also be attributed to altered splicing event in dwarf plants.
Endogenous levels of GAs were measured using 2H2-GA internal standards and UPLC-MS/MS and we find that GA 3 , GA 9 and GA 29 were significantly increased and GA 4 was significantly decreased in dwarf plants ( Fig. 5; Table 5). The results were different from a previous study of Gebremeskel et al 39 , in which significantly lower GA 3 content was obtained in the dwarf parent 'Duan125' of watermelon. The increased content of GA 9 in GA 12 -pathway might be resulted from the up-regulation of GA20ox genes (Cla002362, Cla006227 and Cla010726) Figure 5. The predicted pathways of GA biosynthesis and metabolism in watermelon. The pathways were constructed according to the previous reports of GA biosynthetic and metabolic pathways in higher plants. Two major pathways, GA 53 -pathway and GA 12 -pathway were present in watermelon. The levels of GA 3 and GA 29 in GA 53 -pathway were increased in dwarf plants and marked with red color. The level of GA 9 in GA 12 -pathway was increased and marked with red color, and level of GA 4 in GA 12 -pathway was decreased in dwarf plants and marked with green color. www.nature.com/scientificreports/  www.nature.com/scientificreports/ in dwarf plants, and the reduced content of bioactive GA 4 indicated that loss function of GA3ox (Cla015407) impaired the GA 4 biosynthesis in GA 12 -pathway. Furthermore, GA 51 content in GA 12 -pathway was increased in dwarf plants, however, not to a significant level. The results indicate that mutation of GA3ox (Cla015407) promotes the biosynthesis of GA 51 and inhibits the biosynthesis of GA 4 from GA 9 in GA 12 -pathway. Moreover, endogenous level of GA 29 in GA 53 -pathway was increased, which might be attributed to the increased expression level of GA2ox genes (Cla015162 and Cla019586) in dwarf plants. Additionally, endogenous level of bioactive GA 3 in GA 53 -pathway was increased in dwarf plants, which aroused our interest. We speculated that the reduction in GA 4 level in GA 12 -pathway resulted in increased expression, through release of feedback, of another GA3ox paralogue, Cla022285, with the ability to produce GA 3 in GA 53 -pathway (Table 4; Fig. 5). In rice, two GA3ox genes, OsGA3ox1 and OsGA3ox2, have the activity for converting GA 20 to GA 1 , GA 5 to GA 3 , GA 44 to GA 38 , and GA 9 to GA 4 26 . Additionally, the maize dwarf-1 gene (putative GA 3β-hydroxylase) controls the three biosynthetic steps: GA 20 to GA 1 , GA 20 to GA 5 , and GA 5 to GA 3 66 . However, there is no evidence that Cucurbits contains such an GA3ox enzyme currently 67 . In cucumber, four GA 3-oxidases (CsGA3ox1, -2, -3, and -4) were identified and all these GA 3-oxidases oxidized the C 19 -GA GA 9 to GA 4 as the only product 67 . In this study, the presences of GA 1 , GA 5 , GA 6 , GA 3 and GA 4 indicate that the GA3ox proteins have the activity for catalyzing GA 20 to GA 1 , GA 20 to GA 5 , GA 5 to GA 6 , GA 5 to GA 3 , and GA 9 to GA 4 , and the absence of GA 7 suggests that the GA3ox proteins do not have 2,3-desaturation activity in watermelon.
The rescue of the dwarf phenotype has been reported in GA-deficient mutants in plants. For instance, the application of GA 3 could partially rescue the dwarf phenotype of the cucumber mutant Csdw 58 . In watermelon, the dwarf phenotypes of Cldf and dw could be rescued by exogenous GA 3 application 38,39 . In the present study, exogenous applications of GA 3 or GA 4+7 could rescue the height of dwarf plants, with the latter have a more distinct affect than the former (Fig. 6). The results further confirmed the Si gene is a GA biosynthetic gene and the dwarf phenotype might be attributed to the reduced GA 4 level. www.nature.com/scientificreports/