Localization shift of a sugar transporter contributes to phloem unloading in sweet watermelons

Summary

• Unloading sugar from sink phloem by transporters is complex and much remains to be understood about this phenomenon in the watermelon fruit.
• Here, we report a novel vacuolar sugar transporter (ClVST1) identified through map-based cloning and association study, whose expression in fruit phloem is associated with accumulation of sucrose (Suc) in watermelon fruit. ClVST1⁹⁷ knockout lines show decreased sugar content and total biomass, whereas overexpression of ClVST1⁹⁷ increases Suc content.
• Population genomic and subcellular localization analyses strongly suggest a single-base change at the coding region of ClVST1⁹⁷ as a major molecular event during watermelon domestication, which results in the truncation of 45 amino acids and shifts the localization of ClVST1⁹⁷ to plasma membranes in sweet watermelons. Molecular, biochemical and phenotypic analyses indicate that ClVST1⁹⁷ is a novel sugar transporter for Suc and glucose efflux and unloading.
• Functional characterization of ClVST1 provides a novel strategy to increase sugar sink potency during watermelon domestication.

Introduction

Photosynthetically fixed carbon needs to be partitioned to different sink organs from the source in plants to meet the growth and development of sinks. This redistribution involves long-distance transport of sucrose (Suc) and glucose (Glc) in sieve elements and companion cell complex (SE/CC) from source to sink organs. However, little is known about the mechanisms of phloem unloading of sugars in sink tissues, which represents the starting point for transporting sugars from source tissues to sink organs where they are utilized in growth or storage (De Schepper et al., 2013; Ma et al., 2018).

Within the last decade, considerable research has been invested in the understanding of the various functions of the sugar transporters for sugar partition. AtSWEET11,12 and AtSWEET9 function in phloem loading (Chen et al., 2012) and nectar secretion (Lin et al., 2014) in Arabidopsis, and domesticated ZmSWEET4c and OsSWEET4 import hexose for seed filling in maize and rice (Sosso et al., 2015), respectively, while the recently reported Glc transporter ZmCST1 controls stomatal opening and carbon starvation in maize (Wang et al., 2019). Eventually, sugars are stored in vacuoles through tonoplast-localized sugar transporters such as tonoplast monosaccharide transporters in Arabidopsis (Wingenter et al., 2010), tonoplast sugar transporters in sugar beet (Bv, B. vulgaris; Jung et al., 2015), ClTST2 in watermelon (Citrullus lanatus) (Ren et al., 2018) and CmTST2 in melon (Cheng et al., 2018).

However, exporting sugars from SE/CC cells through the plasma membrane in crops is controlled in a sophisticated manner, involving the expression of several genes and their interactions with temperature and light responses, making it challenging to identify sugar unloading-related genes. In sweet sorghum, Bihmidine et al. (2015) reported that Suc accumulation in the stems follows an apoplasmic phloem unloading, but this process does not involve differential expression of Suc transporters (SUTs). However, Milne et al. (2017) reported that Suc unloading from stem de-energized protophloem SE in sorghum could be facilitated by reversal of sorghum bicolor sucrose transporter 1 and/or by uniporting sugars will eventually be exported transporters. Thus far, there is not sufficient understanding of the molecular mechanisms or the evolution of genes controlling sugars unloading in SE/CC cells.

In this study, we identify ClVST1⁹⁷, which encodes a novel vacuolar sugar transporter and is responsible for phloem unloading of Suc within the sweet watermelon fruit. Furthermore, we reveal that a single-base change in its coding region (C99A) results in the gain-of-function of unloading sugars from SE/CC. These combined results are discussed in the context of current models of source–sink interactions and indicate yet another important physiological role of subcellular localization shift in defining an important agronomic trait.

Materials and Methods

Carboxyffuorescein diacetate (CFDA) and esculin labeling

Phloem was labeled by CFDA and esculin solution (Sigma-Aldrich) as previously described (Zhang et al., 2004). We removed cortical cell layer pedicel to avoid any damages to the phloem. Afterwards, the pedicel was enlaced by a cotton thread at one end, and the other end of the cotton thread was immersed in a tube with 200 ml (1 mg ml^–1) CFDA or 200 ml (1 mM) esculin aqueous solution. Plants were allowed to translocate the carboxyffuorescein (CF) for 24 h and esculin for 30 min, and the fruit tissues were subsequently sectioned and examined using confocal laser scanning microscopy.

Plant materials and sugar content analysis

An F₈ population consisting of 96 recombinant inbred lines (RILs) derived from a cross between the cultivated high sugar content line 97 103 (C. lanatus) and the nonsweet wild accession U.S. Plant Introduction (PI) 296341-FR (Citrullus amarus) was used to develop the single nucleotide polymorphism (SNP) genetic map. Wild species watermelon PI 296341 was collected in southern Africa with nonsweet, thick, hard flesh, and round, medium-sized fruit that matures at c. 50 d after pollination (DAP). The 97 103 × PI 296 341 RIL population was grown in Beijing over 3 yr in a completely randomized block design with three replications. We used the center of the fruit for sugar content by high-performance liquid chromatography (HPLC) (LC-10A VP; Shimadzu Co. Ltd, Nakagyo-ku, Kyoto, Japan) or brix measurement.

Statistical analysis

We used proc glm in Sas 8.0 for analysis of variance (ANOVA). We calculated the broad-sense heritability (h²) for each trait basis as h² = σ_G²/σ_G² + σ_GE²/n + σ_e²/nr, where σ_G², σ_e² and σ_GE² are the variance estimates for genotype, experimental error, and genotype × environment interaction, respectively; n represents the number of environments and r represents the number of replications. We calculated the h² confidence intervals (CIs) according to reported method (Knapp et al., 1989). By using Sas proc corr (Littell et al., 1998), we calculated Pearson's phenotypic correlation coefficients among traits in all environments.

Mapping of sugar content quantitative trait locus (QTL) and association study

In our study, we performed the single-environment and multi-environment joint analyses based on a mixed-model-based composite interval mapping (MCIM) using QTLNetwork software. We defined the presence of a significant QTL by 1000 permutations at a significance threshold of P = 0.05. The confidence interval was calculated for each QTL as described by Yang & Zhu (2005). QTLs detected in different seasons or years for the same trait were considered the same QTL if their CIs overlapped. Juice from the centre of the watermelon fruit was used for Suc content detection with HPLC (LC-10A VP; Shimadzu). Brix was measured using a pocket refractometer (model pal-1; Atago Co. Ltd, Tokyo, Japan) using the same juice from each watermelon. Genome-wide association study was performed with the compressed mixed linear model implemented in the software Tassel (Zhang et al., 2010). The 326 representative watermelon accessions used for association study were collected worldwide, from 254 cultivars, 86 landraces, 22 semi-wild and 42 wild ancestor watermelon accessions representing the majority of cultivated varieties in breeding institutions and seed companies in recent decades.

RNA extraction and qRT-PCR

We used a Quick RNA isolation kit (Huayueyang Biotechnologies Co. Ltd, Haidian, Beijing, China) for total RNA extraction and SuperScript III transcriptase (Invitrogen) for cDNA synthesis. For quantitative reverse transcription polymerase chain reaction (qRT-PCR), experiments were run on a LightCycler 480 instrument (Roche), according to the manufacturer's instructions, following a thermal profle: 60 s hot start at 95°C followed by 40 cycles of 95°C for 10 s, 55°C for 10 s and 72°C for 30 s. Three replicates were performed for each gene. Primer efficiency was calculated using the following equation: effciency = 10^{(−1/slope)−1}. We used the watermelon ACTIN gene (ID Cla97C02G026960) and ClYLS8 (ID Cla97C02G038590) as internal controls in the analysis. The relative expression level of the small upstream open reading frame (uORF) and the N-terminal truncation main ORF of ClVST1 (ClVST1⁹⁷) was normalized to that of the ACTIN and ClYLS8 and averaged over the three biological replicates.

RACE and DNA constructs

We performed the 5′- and 3′-rapid amplification of cDNA ends (RACE) of the transcripts using the SMARTer RACE 5′/3′ KiT (Clontech Laboratories,–Takara Biomedical Technology, Haidian, Beijing, China). The Phusion-HF DNA Polymerase (New England Biolabs, Ipswich, MA, USA) was used for all PCR reactions. The p35S:ClVST1 eGFP fusion construct was generated using the gateway-specific destination vector pX-YFP_GW (Chen et al., 2010). For this construct, the full-length ClVST1 cDNA was amplified and the last stop codon removed by PCR using ClVST1-specific primers harboring the attB1 and attB2 sites (Supporting Information Table S1), and it was then cloned into pDONR221F1 (Invitrogen) via the BP reaction, followed by an LR reaction to transfer the entire ClVST1 sequence into yeast expression vector pDRF-GFP or ClVST1⁹⁷ into p35S driving plant expression vector pX-YFP_GW. The ClVST1 gene was inserted between the EcoRI and BamHI sites in the HEK293T cell expression vector pCMV6. gRNAs for the ClVST1 gene was assembled into the CRISPR/Cas9 vector pBSE401 using the golden gate method (Xing et al., 2014).

In situ hybridization

We performed RNA in situ hybridization using our standard protocols (Zhao et al., 2019), with slight modifications. Briefly, the 293 bp-specific fragments of the N-terminal truncation main ORF (ClVST1⁹⁷) were used as probes for hybridization. Primers for probe synthesis are listed in Table S1. Sense and antisense probes were synthesized by PCR amplification of cDNA with gene-specific primers containing T7 and SP6 RNA polymerase binding sites using SP6 and T7 RNA polymerases, respectively. The sense probe which cannot complement with ClVST1 mRNA was used as the negative control. Then in vitro transcription was carried out, followed by precipitation with ammonium acetate and ethyl alcohol, hydrolysis in carbonate buffer and then neutralization and resuspension in 50% formamide. At least six independent fruits at 1 DAP were dissected by hand, fixed in 3.7% formol-acetic-alcohol and vacuum-pumped for 20 min. They were then put through a series of alcohol gradients and embedded in paraffin for storage at 4°C. We divided the 1 DAP fruits into two sections. Sections hybridized to the sense probe of ClVST1⁹⁷ mRNA were used as negative controls, and samples from the neighboring sections were hybridized with an antisense probe of ClVST1 mRNA.

Generation of overexpression and mutant lines

Watermelon transformation, plant regeneration, and glasshouse care were performed as described previously (Tian et al., 2017). Agrobacterium tumefaciens strain GV3101, which harbored the DNA constructs, was used for overexpression and mutant transformation. The 35S-driving plant vector pX-YFP_GW (Chen et al., 2010) was used for overexpression. Briefly, the surface-sterilized watermelon seeds were cocultivated in the dark for 4 d. The regenerated buds were excised and transferred onto the selective medium, containing 0.01 mg l^–1 α-naphthlcetic acid, 0.1 mg l^–1 6-benzyl aminopurine and 2 mg l^–1 Basta. Constructs were designed to produce defined deletions within each target gene-coding sequence using two sgRNAs alongside the Cas9 endonuclease gene. The CRISPR/Cas9-positive lines were further genotyped for indel mutations using a forward primer to the left of the sgRNA and a reverse primer to the right of the sgRNA (Table S1). The Cas9-free T₂ population and homozygous mutants in clvst1 were used for phenotyping. Transgenic plants that did not show any Basta resistance and contained no editing in the same segregating population were used as negative controls. All lines were planted in the field for phenotyping at between 10 and 25°C in Beijing’s autumn season.

Starch staining of leaves

Three leaves closest to watermelon fruit were harvested in the morning. Starch staining was performed right after the harvest. Leaves were cleared in 80% (v/v) ethanol plus 5% (v/v) formic acid at room temperature, stained in KI₂ Lugol's iodine solution (1 g iodine and 2 g KI, in 100 ml water) for 10 min and washed three times in water.

Protoplast isolation and subcellular localization

To investigate the subcellular localization of ClVST1 in watermelon, leaf protoplasts were isolated from cells of 2-wk-old young leaves after seeding according to the method of Yoo et al. (2007). The p35S::ClVST1-eYFP fusion construct was purified and transformed into leaf protoplasts as previously described (Malnoy et al., 2016). A volume of 200 μl (2 × 10⁵ cells) protoplast and 20 μl plasmid (20 mg) was used for transformation. Protoplast and plasmid were pre-mixed and incubated at room temperature for 10 min, before the transformation. The protoplast and plasmid mix was mixed with an equal volume of polyethylene glycol (PEG) 4000 and incubated for 10 min at 25°C. A total of 440 μl of W5 solution was then added, mixed and incubated at room temperature for 10 min. The protoplast and plasmid were mixed in additional 1 ml W5 solution and incubated at room temperature for 10 min. The tube was centrifuged at 100 g for 5 min, the supernatant was discarded and 1 ml of W5 solution was added, followed by incubation at 25°C in dark for 16 h. Fluorescence of ClVST1-YFP (GFP) was analyzed using a Zeiss LSM700 confocal microscope at wavelengths of 488 and 530 nm for excitation and emission, respectively.

Preparation of yeast vesicles and sugar transport analysis

Yeast strains EBY4000 (for Glc transport assay) and INVScI (for Suc transport assay) were transformed with the pYESDEST52 vector harboring the ClVST1 ORF using the EasyComp transformation kit (Life Technologies, Carlsbad, CA, USA) according to the manufacturer’s instructions. Yeast cells transformed with the yeast expression empty vector pYESDEST52 (Life Technologies) were used as negative controls. Transformed yeast cells were grown in SD-Ura selection medium supplemented with 50mM maltose or Glc (Sigma Aldrich). The plates were incubated at 30°C for 3 d and positive clones were confirmed and sequenced. Microsomal membranes were isolated from 2 litres cultures. The yeast pellet was resuspended in 30 ml spheroplast buffer 1 M sorbitol, 50 mM Tris-HCl, pH 7.5, with 0.5-1U zymolyase-20T applied to 1ml yeast culture sediment. The suspension was incubated at 30°C with gentle shaking for 2 h to digest the cell wall. Spheroplasts were harvested at 2000 g for 10 min, and washed twice with spheroplast buffer. They were resuspended in 5 ml buffer A (lysis buffer) with 10 mM PIPES, 0.1 mM MgCl₂, 12% Percoll and protease inhibitor mixture (Sigma-Aldrich), pH 6.8, and homogenized with 60 strokes of a Dounce tissue homogenizer (Swedesboro, NJ, USA) with a tight pestle on ice. Homogenate was transferred to polyallomer tubes SW41, overlaid with 3 ml of 8% Percoll buffer A and 5 ml of reconstitution buffer (10 mM Hepes–NaOH, 100 mM NaCl and 0.1 mM KCl (pH 7.0)), and then centrifuged at 110 000 g for 30 min at 4°C to collect membranes floating at the top of the gradient. Membrane protein was diluted in 20 mM detergent octyl-beta-glucoside to form micelle. The method of Bradford (1976) was used to determined protein concentrations. ClVST1 protein was shown to present in isolated yeast vesicles by immunoblot analysis using HisProbe™-HRP Conjugate (Thermo Scientific, Waltham, MA, USA). For sugar uptake assay, yeast vesicles were incubated with 10 mM sugar (0.1 μCi μl^–1 [¹⁴C] sugar) solution with or without 4 mM adenosine triphosphate (ATP) as described in Rautengarten et al. (2014) at different time points, and then transferred to a Corning 2 ml filter (10 kd cutoff) to remove extra detergent and [¹⁴C] sugars, followed by centrifuging at 12 000 g for 5 min and washing (four times) with 300 μl reconstruction buffer before radioactivity detection by scintillation counting. For sugar uptake assays in oocytes, at 3 d after injection of the RNA, oocytes were incubated with 10 mM sugar (0.1 μCi μl^–1 [¹⁴C] sugar) solution as described by Chen et al. (2012) at different time points.

Sugar efflux analysis using patch-clamp in transformed HEK293T cells

We grew HEK293T cells in high-glucose DMEM medium (Invitrogen) containing 10% fetal calf serum, 50 μg ml^–1 streptomycin (Invitrogen) and 50 μg ml^–1 penicillin. The cells were transiently transfected at 50–70% confluence using Lipofectamine 2000 Reagent (Invitrogen) in Opti-MEM I reduced-serum medium (Gibco, Waltham, MA, USA). To distinguish positive transfected cells, we transformed ClVST1-GFP or empty GFP fusion protein for patch-clamp experiment. Only transformation with a > 80% positive rate was used for recording currents. After transfection, the cells were incubated for 48 h before electrophysiological measurement. The electrophysiological recordings were obtained under a microscope (IX71, Olympus, Shinjuku-ku, Tokyo, Japan) for visual control. Symmetric bath and pipette solutions in Hanks’ buffer with solutions were adjusted to pH 7.2 (HEPES/Tris) and pH 5.5 (MES/Tris), respectively. We perfused each concentration until the current reached a steady-state level. For efflux assays and monitoring of sugar-induced currents, sugar (50 mM) was injected into the pipette medium of the HEK293T cells. The electrophysiological signal was recorded using an amplifier (EPC10; HEKA, Reutlingen, Germany). Before formation of a seal, offset potentials were nulled directly. A HEK293T cell was made from whole-cell capacitance compensation (in pF). The data were analyzed using the Patchmaster software. All experiments were performed at room temperature.

Ratio of π diversity

The SNPs from 326 representative watermelon accessions (Guo et al., 2019) were used for nucleotide diversity (π) calculation using a 50 kb sliding window. We calculated the ratio of diversity (π_semi-wild/π_cultivated) for each window using Vcftools (Danecek et al., 2011) to calculate the pairwise genomic differentiation for semi-wild and cultivated populations of watermelon. The selective sweeps were defined as genome regions with the top 7% of the π ratio.

Accession numbers

Sequence data from this article can be found in the watermelon genome database (http://cucurbitgenomics.org) under the following accession numbers: the watermelon ClVST1 gene (Cla97C02G031010), the ACTIN gene (Cla97C02G026960) and ClYLS8 (Cla97C02G038590). Raw transcriptome sequencing reads have been deposited into the NCBI sequence read archive under accessions PRJNA564186.

Results

Phloem unloading follows an apoplasmic pathway in watermelon

Crops in the Cucurbitaceae family, consisting mainly of cucumber, watermelon, melon, and pumpkin, are widely consumed as fruits and vegetables. Unloading of photoassimilate in vascular tissue follows an apoplasmic pathway in cucumber fruit (Hu et al., 2011). To investigate whether apoplasmic or symplastic phloem unloading pathways play an essential role in watermelon, we loaded CF diacetate (DA) into leaf veins, which was degraded into fluorescent dye CF in cells. CF is membrane-impermeable and can only be transported via plasmodesmata and is often used as a fluorescent marker of symplastic phloem unloading to trace the behavior of photoassimilate unloading (Nie et al., 2010). Our experiments showed that CF was always restricted to the phloem strands in the fruit vascular bundles without obvious diffusion to the surrounding flesh parenchymal cells (Fig. S1a). This result indicates that the CF or photoassimilates were probably not unloaded via symplastic plasmodesmata in vascular bundles of watermelon fruit. Esculin is a fluorescent coumarin glucoside that as a phloem-transport tracer is unloaded by sugar transporter via an apoplasmic pathway in phloem (Knoblauch et al., 2015). When we loaded esculin to the stem, the probe entered the phloem vascular bundles within 30 min and was unloaded subsequently from the phloem to surrounding flesh parenchymal cells in watermelon (Fig. S1b). This result indicates that the phloem-transport tracer esculin or photoassimilates were unloaded via an apoplasmic pathway in watermelon vascular bundles.

Mapping and association study of sugar content QTL Qsuc2-1

Sugar unloading in phloem via the apoplasmic pathway contributes to sugar accumulation in sink tissues. However, reports concerning sugar transporters in the apoplasmic pathway unloading are lacking (Ma et al., 2018). One of the major Suc content QTL (Qbrix2-1 or Qsuc2-1) with R² = 22.14% was initially mapped to a 3.4 cM region on chromosome 2 (Ren et al., 2014). To further map this Suc content QTL, we constructed an ultra-high-density variation map containing 0.948 million high-quality SNPs in RILs from a segregating population derived from a cross between the elite sweet watermelon line 97 103 (Citrullus lanatus) and the nonsweet wild accession U.S. PI 296 341 (C. amarus) (Ren et al., 2018). Here, this ultra-high-density variation map was constructed using a bin-map in which the genetic distance is evaluated in ‘bins’ and not in cM units. A ‘bin’ is a genomic region on the genetic map with a distinctive segregation pattern. These adjacent bins were separated by recombination events in the RIL population. The skeleton bin map comprised 2492 recombination bins with an average interval physical length of recombination events (or bins) of 138.1 kb, ranging from 3.0 to 970 kb, meaning this SNP bin map reached a precision of between 3.0 and 970 kb for mapping genes and successfully mapped the QTL Qsuc2-2 in an interval of 797.4 kb for identifying the ClTST2 (Tonoplast Sugar Transporter) gene (Ren et al., 2018).

We measured the Suc content for RILs in three sequential years. The heritability (h²) of the Suc content trait was 68% and genotype × environment interactions were also observed. Using this high-density SNP bin map, we mapped the Suc content QTL Qsuc2-1 to an interval of two bins (1.2 cM) between 3919 and 4003 kb on chromosome 2, covering a physical position of 84 kb (Fig. 1a,b). This Qsuc2-1 region harbors eight ORFs (Fig. 1b) according to watermelon genome assembly version 2 (http://cucurbitgenomics.org) (Zheng et al., 2019). RNA-Seq data (Guo et al., 2015) showed ORF6 (ClVST1, Cla97C02G031010) was highly expressed, while the other seven ORFs were functionally irrelevant or showed nearly no expression in sweet watermelon line 97 103 (Fig. 1c; Table S2). Expressions of ORF6 were positively correlated with sugar accumulation during fruit developmental stages from 10 to 34 DAP, with high expression occurring in ripe fruit flesh of sweet-dessert watermelon line 97 103 (Fig. 1c) and nearly no expression in fruits of nonsweet PI 296 341 and 97 103 mesocarp, stem and carpopodium (Table S2).

Fig. 1

Mapping, association study and structure variation of ClVST1. (a, b) Sucrose (Suc) content of quantitative trait locus (QTL) Qsuc2-1 was mapped (a) to a physical position of 84 kb on chromosome 2 containing eight open reading frames (ORFs) (b). (c) Expression of the eight genes during watermelon fruit development from 10 to 34 d after pollination (DAP). (d) Single nucleotide polymorphism (SNP) C99A of ClVST1 is the locus most associated with sugar content in 326 watermelon accessions. (e) SNP C99A causes a premature stop codon (TAA) in the first exon of ClVST1⁹⁷, resulting in the truncation of 45 amino acids at the N-terminal of ClVST1⁹⁷ in cultivated watermelons.

All cultivars and landraces showed higher sugar content than all wild ancestor watermelons, while the semi-wild accessions contained both sweet and nonsweet watermelons. Furthermore, we performed an association study on 326 representative watermelon accessions collected on a worldwide basis (Table S3), using genomic resequencing data (Guo et al., 2019) covering a 0-10-Mb of Chr2 that contains the QTL Qsuc2-1. The most significant SNPs, out of 32 000 SNPs identified in this region, were located at C42T and C99A of the ClVST1 (ORF6) coding region in QTL Qsuc2-1 (Fig. 1d). To further identify association between ClVST1 alleles and sugar content, we examined its coding sequences in the 326 watermelon accessions. We identified nine SNPs in the coding sequence of ClVST1 in these accessions, including six synonymous polymorphisms at C42T, C246G, G1083A, C1245T, C1293T and G1332C, and three nonsynonymous polymorphisms at C99A, T1394C and C1512G (Table S3). We grouped the accessions based on sugar content and identified which allelic differences in ClVST1 were associated with sugar content. As expected, the variation C42T and C99A in the first exon were the most associated SNPs in 326 accessions (Fig. 1d; Table S3). The SNP C42T is a synonymous polymorphism that results in no amino acid change, while the nucleotide polymorphism of SNP C99A causes a premature stop codon (TAA) in the first exon, resulting in the generation of a small uORF and starting the N-termini truncated main ORF (defined as ClVST1⁹⁷) from the second initiation codon at 136 bp (Fig. 1e). Thus, variation SNP C99A causes N-terminal truncation of 45 amino acids in cultivars compared with nonsweet wild species. In the 326 germplasm collections, the SNP C99A causes truncation of 45 amino acids at N termini in sweet cultivars, landraces and sweet semi-wild accessions, while the full coding sequence existed in nonsweet semi-wild and wild species (defined as ClVST1^PI) (Fig. 1e). The other two nonsynonymous SNP sites (T1394C and C1512G) were not associated with sugar content.

ClVST1⁹⁷ is highly expressed in phloem

The Suc content increased greatly from 10 to 34 DAP in the fruit flesh of cultivar 97 103, while the Suc concentration remained consistently low in fruit of wild watermelon PI 296 341, as well as fruit mesocarp, leaf and root of 97 103 (Fig. 2a). RNA-Seq data (Guo et al., 2015) indicated that the expression of ClVST1 in these tissues displayed the same pattern as the Suc content (Fig. 2a). We then performed the quantitative PCR (qPCR) in 48 core germplasms sampled from 1197 worldwide accessions (Zhang et al., 2016). The low sugar content lines had nearly similar expression patterns to the wild accession PI 296 341, while the expression levels in high sugar content lines were 10- to 20-fold higher than those in nonsweet accessions (Fig. 2b). In the core collection, the ClVST1 mRNA level was highly correlated with the sugar content, with a correlation coefficient (r) of 0.75. Together these results suggest that the high expression of ClVST1⁹⁷ could contribute to the sweetness in cultivated watermelons, compared with no or low expression in wild or semi-wild accessions.

Fig. 2

Expression of ClVST1. (a) Transcript abundance of ClVST1⁹⁷ is positively correlated with sugar content during fruit ripening from 10 to 34 d after pollination (DAP) in the cultivar 97 103. Both sucrose (Suc) content and transcript abundances of ClVST1 remained consistently low in cultivar mesocarp leaf, root and wild fruits. Results of three biological replicates are used for calculating standard deviation (SD). (b) Positive correlation between sugar content and transcript abundance of ClVST1 in 48 represented germplasms. Means and SD of three biological replicates are shown. (c) In situ hybridization revealed that the red signals of ClVST1⁹⁷ were mostly restricted in external phloem (EP) and internal phloem (IP) in watermelon fruit, without obvious signals in surrounding flesh parenchymal (FP) cells, compared with the negative control (d). Bar, 50 μm.

To investigate the nature of the ClVST1⁹⁷ expression in different watermelon tissues, we examined its expression pattern by RNA in situ hybridization in 97 103 using the specific antisense probe of ClVST1⁹⁷. In cucurbit crops, the vascular bundles are bicollateral, with internal phloem located close to the fruit center and external phloem (EP) beyond the fruit center, separated by xylem (Hu et al., 2011). The result showed that ClVST1⁹⁷ was mainly expressed in both external phloem and internal phloem of the fruit (Fig. 2c), compared with the negative control using the sense probe which cannot complement with ClVST1⁹⁷ mRNA (Fig. 2d). RNA in situ hybridization suggested that the ClVST1⁹⁷ was localized in phloem of watermelon fruit vascular bundles, without obvious expression in surrounding flesh parenchymal cells (Fig. 2c).

ClVST1⁹⁷ contributes to sugar accumulation in watermelon

Our analyses suggest that ClVST1⁹⁷ is the highly plausible candidate underlying the Suc content QTL. To validate this, we first tested whether ClVST1⁹⁷ controls sugar content in watermelon by engineering null mutations in ClVST1⁹⁷ using CRISPR/Cas9 gene editing technology (Doudna & Charpentier, 2014) which contained two single-guide RNAs (sgRNAs) (Xing et al., 2014; Tian et al., 2017). We generated two CRISPR-clvst1 mutants with one harboring a 118 nt deletion between two protospacer-adjacent motif (PAM) sites and the other containing a C/T deletion in the first exon of the coding region (Fig. 3a), leading to a frameshift to generate a null mutant allele. Fewer branches and poor growth vigor were observed (Fig. 3b) in the T₂ populations of the clvst1⁹⁷-1 and clvst1⁹⁷-2 homozygous allele. In higher plants, photosynthate was transported from source to sink through phloem transporters. This prompted us to investigate whether the photosynthate was decreased in clvst1⁹⁷-1 and clvst1⁹⁷-2 owing to the change in source-to-sink Suc partitioning. Subsequently, we found that starch content in leaves was increased (Fig. 3c), probably as a result of insufficient unloading sugars to fruits from leaves in clvst1 mutants. Meanwhile, we observed smaller fruit size than the wild-type (WT) at 20 and 25 DAP in clvst1⁹⁷ mutants with delayed ripening (Fig. 3d). The photosynthetic efficiency of the clvst1⁹⁷-1 and clvst1⁹⁷-2 appeared lower than the WT (Fig. 3e). To this end, we treated the source leaves of WT and clvst1 mutants with [¹⁴C]-labeled CO₂ and investigated time-dependent uptake of labeled [¹⁴C] CO₂. Earlier studies in Arabidopsis showed that [¹³C] was incorporated into Suc to 45% in 60 min in leaves (Szecowka et al., 2013). Therefore, we set the transporting time of 3 h from 07:00 to 10:00 h to ensure that the photosynthesized sugars were fully unloaded. Fruit samples were harvested 6 and 24 h later to ensure that sugars were transported into watermelon fruit. The fixed C¹⁴-CO₂ in clvst1 mutants was decreased compared with the WT in fruit (Fig. 3f). Consistently, the ratio of fruit-fixed C¹⁴-CO₂ in 2-wk-old clvst1 mutants was reduced by 15% whereas the ratio of plant (leaf)-fixed C¹⁴-CO₂ increased (Fig. 3g). Fruit Suc content in clvst1 mutants decreased to 33 and 35 mg g^–1 fresh weight (FW) compared to 40 mg/g FW in the WT (Fig. 3h). Combining these results with less C¹⁴-CO₂ fixed in the wild PI 296 341 fruit (Fig. 3f) suggests that the ClVST1⁹⁷ may function as carbon repartition between watermelon source and sink.

Fig. 3

ClVST1⁹⁷ converts leaf starch content and fruit sugar partition in watermelon. (a) CRISPR/Cas9-generated clvst1⁹⁷ alleles harboring 118 nt or C/T deletions in the first exon of ClVST1⁹⁷. (b) Shorter main branches and fewer lateral branches in ClVST1⁹⁷-mutated plants. (c) Increased starch accumulation in clvst1⁹⁷ mutants. Three independent leaves close to fruit are shown for each. Bar, 5 cm. (d) Smaller fruit size and delayed ripening at 20 and 25 d after pollination (DAP) in clvst1⁹⁷ mutants. Bar, 5 cm. (e) Reduced photosynthetic efficiency of clvst1⁹⁷ mutants. (f) Fixed C¹⁴-CO₂ in fruit of clvst1⁹⁷ mutants decreased, compared with the wild-type (WT). Lower fixed C¹⁴-CO₂ in fruit of wild PI 296 341, compared with the cultivar WT. (g) Ratio of fruit and plant fixed dry biomass in 8-wk-old clvst1⁹⁷ mutants. (h) Decreased fruit sucrose (Suc) content in clvst1⁹⁷. (i) ClVST1⁹⁷ overexpressors with increased ClVST1⁹⁷ mRNA levels and Suc accumulation measured by FW in watermelon fruit. (j) Increased fruit dry biomass at 34 DAP in ClVST1⁹⁷-OEs, compared with the WT. (e–j) Thirty independent plants for mutants and 20 plants for overexpressor lines were planted for phenotyping. *, P < 0.05; **, P < 0.01 (t-test). Means and SD of n = 3 independent biological replicates are shown. P-values were calculated by two-tailed, unpaired t-tests.

We overexpressed ClVST1⁹⁷ from 97 103 in the white-fleshed semi-wild PI 179 878. A total of 15 independent overexpression (OE) transgenic T₀ plants were propagated, and two of the resulting T₂ lines with increased ClVST1⁹⁷ mRNA levels (ClVST1⁹⁷ OE-1 and ClVST1⁹⁷ OE-2) were selected for further analysis. Transgenic plants, which did not show Basta-resistance during seeding period in the same T₂ segregation population, were used as negative controls. In ClVST1⁹⁷ OE-1 and OE-2, fruit Suc content and ClVST1⁹⁷ mRNA levels increased significantly compared with the control (Fig. 3i), suggesting that the higher ClVST1⁹⁷ expression levels are one of the foundations for increasing flesh Suc accumulation in watermelon. Consistently, increased fruit fresh and dry biomass measured by FW and DW at 34 DAP were observed in ClVST1⁹⁷-OEs (Fig. 3j).

We then compared gene expression profiles between clvst1 and WT fruits using RNA-Seq, and identified 31 upregulated and 60 downregulated genes in clvst1 mutants (Table S4). As a result of reduced sugar accumulation in clvst1 mutant, key enzymes in Suc and ATP metabolism pathways were significantly downregulated, such as Suc synthase (Cla97C01G008790), fructose-1,6-bisphosphatase (Cla97C04G073970) fructose-bisphosphate aldolase (Cla97C09G183030, Cla97C11G210620) and ATP synthase (Cla97C03G055870, Cla97C08G146490). Furthermore, gene ontology term analysis revealed that downregulated genes were enriched with those in carbon utilization and ATP metabolism pathways (Table S5). These transcriptome results further indicated that the function of ClVST1⁹⁷ may associate with sugar and ATP metabolism.

The most associated SNP results in plasma membrane localization of ClVST1⁹⁷ in sweet watermelons

The most associated nonsynonymous SNP C99A causes the truncation of 45 amino acids at the N-terminal in sweet watermelons, compared with nonsweet semi-wild and wild species. This result indicates that the 45 amino acids truncated at the N termini of ClVST1⁹⁷ may be the causal structure variation for fruit sugar accumulation in sweet watermelons. To demonstrate the role of the 45 truncated amino acids, we identified a Tyr-based YXXXφ (where φ is a hydrophobic and X is a random residue) motif within these 45 amino acids (Fig. 4a). Both YXXφ and YXXXφN motifs were reported to be a sorting signal in endogenous proteins in human cells for protein localization (Nishimura & Balch, 1997; Kozik et al., 2010) but without any reports in plants as far as we know.

Fig. 4

Structure and localization of ClVST1 in cultivated and wild watermelons. (a) Transmembrane domains of ClVST1 and endogenous protein sorting signal YXXXφ (black arrow) in the N termini of ClVST1^PI in wild watermelons. The red arrow points to the truncated site. (b, c) Tonoplast localization of the wild accession ClVST1^PI-YFP fusion protein in yeast (b) and leaf protoplasts (c). (d) Outstretched figure of (c). Yellow and red fluorescence indicates tonoplast and chloroplast in leaf protoplasts in (c) and (d). (e, f) Plasma membrane and inner membrane localization of the cultivar ClVST1⁹⁷-GFP fusion protein in yeast (e) and leaf protoplasts (f). (g) Outstretched figure of (f). Targeting of cultivar ClVST1⁹⁷-YFP fusion protein to the plasma membrane of leaf protoplasts, where chloroplast was positioned on the cytoplasmic side of the YFP fluorescence (f, g). Yellow and red fluorescence indicates plasma membrane and chloroplast in leaf protoplasts, respectively (g). Representative cell for more than five protoplasts was shown for each localization.

To corroborate the localization of the transporter, we fused ClVST1^PI sequence from the wild species PI 296 341 with the GFP/YFP genes and expressed the fused protein in yeast and protoplasts made from watermelon leaf. The fluorescence of ClVST1^PI was observed in the vacuolar membrane of yeast (Fig. 4b), and leaf protoplasts (Fig. 4c,d). Subsequently, to investigate whether the N-terminal truncated 45 amino acids change the localization of ClVST1⁹⁷, the ClVST1⁹⁷ sequence from cultivar 97 103 was fused with the GFP or YFP genes and the fused protein was expressed in yeast and leaf protoplasts. The fluorescence of ClVST1⁹⁷ was observed on both the cytosol membrane and the inner membrane of yeast cells (Fig. 4e) and leaf protoplasts (Fig. 4d,f). Together these suggest the shift of localization of ClVST1 from the vacuolar membrane in wild species PI 296 341 to the plasma membrane in cultivar 97 103, caused by a single base change (C99A) in the coding region of ClVST1⁹⁷ which results in the truncation of 45 amino acids at the N termini of ClVST1⁹⁷. We hypothesize that the plasma membrane-localized ClVST1⁹⁷ may be responsible for sugar accumulation in cultivated watermelons.

ClVST1⁹⁷ functions as a novel sugar efflux in vitro

The protein ClVST1 is annotated as a member of the Nodulin-like subfamily, which is close to the tonoplast monosaccharide transporter AtTMT, the vacuolar glucose transporter AtVGT, and the sucrose transporter AtSUT subfamily within the major facilitator superfamily (MFS) (Fig. S2), containing 12 transmembrane domains (Fig. 4a). However, substrates transported by the closest proteins of ClVST1 in other plants have not been uncovered. To identify the substrate transported by ClVST1, we first expressed ClVST1⁹⁷ from cultivar 97 103 in Suc and Glc uptake mutant yeast strains SuSy7 and EBY4000 (Wieczorke et al., 1999), respectively. However, ClVST1⁹⁷ was unable to complement the yeast mutants on Suc or Glc selection plates (data not shown). Meanwhile, ClVST1⁹⁷ was unable to take up [¹⁴C]-Suc/[¹⁴C]-Glc in Xenopus oocytes (Fig. S3a,b). This result prompted us to hypothesize that ClVST1⁹⁷ may function as a sugar efflux. To verify this, cultivated ClVST1⁹⁷ from 97 103 and ClVST1^PI from wild PI 296 341 were heterozygously expressed in yeast, and vesicles were extracted. Subsequently, yeast vesicles were incubated with [¹⁴C]-Suc and [¹⁴C]-Glc. The cultivated ClVST1⁹⁷ protein demonstrated Suc/Glc transport activity with ATP by monitoring time-dependent uptake of [¹⁴C]-Suc/[¹⁴C]-Glc, while no radioactivity was detected when expressing wild ClVST1^PI or in the empty vector control (Fig. 5a,b). In vitro assays of ClVST1⁹⁷ protein activity in vesicles of yeast suggested a K_m value of 12.1 and 5.5 mM with V_max = 167.7 and 116.9 nmol min⁻¹ mg⁻¹ for transporting Suc and Glc (Fig. 5c,d). Moreover, we carried out Suc efflux assays in human embryonic kidney (HEK) 293T cells expressing the ClVST1⁹⁷-GFP protein using the patch-clamp technique. To mimic physiological proton gradients across the vacuolar membrane, the luminal (pipette) medium was buffered to pH 5.5 and the cytosolic (bath) medium was set to pH 7.2. When 5 mM Suc/Glc was applied to the cytosolic solution, cells harboring ClVST1⁹⁷-GFP responded with a strong inward current trace with respect to the GFP empty vector background current level (Fig. 5e,f). These Suc/Glc activated inward currents mean protons flow into cells while Suc/Glc flow out of cells. However, no inward current trace was detected when raffinose (Raf) was applied to the cytosolic solution and in GFP empty vector expressed control (Fig. 5e,f). Higher Suc/Glc-induced current response rates were determined at pH 5.5 than at pH 7.2 (Fig. 5f), indicating that ClVST1⁹⁷ may function as a pH-dependent sugar efflux. Higher Suc-induced than Glc-induced current response rates were determined at pH 5.5 when loading the same concentration sugars (Fig. 5f), possibly owing to the fact that HEK293T cells cannot digest Suc but have low endogenous Glc uptake activity (Takanaga et al., 2008; Takanaga & Frommer, 2010). These results indicate that the ClVST1⁹⁷ in sweet watermelons functions as a plasma membrane Suc/Glc efflux. Combined with its main expression in phloem (Fig. 2c), we suggest that plasma membrane-localized ClVST1⁹⁷ in cultivated watermelon serves as a phloem Suc/Glc efflux for unloading sugars to intercellular space.

Fig. 5

ClVST1⁹⁷ protein functions in sugar efflux. (a, b) ATP-dependent sucrose (Suc) (a) and glucose (Glc) (b) transport activity of the ClVST1⁹⁷ protein demonstrated by monitoring the uptake of [¹⁴C]-Suc and [¹⁴C]-Glc in yeast vesicles. Suc/Glc transporters AtSWEET12/ClSWEET3 and empty vector were used as positive and negative controls, respectively. Results of means and SD from three biological repeats were used. (c, d) K_m and V_max assays of Suc (c) and Glc (d) transport activity using ClVST1⁹⁷-expressed yeast vesicles. Means and SD results of three independent repeats. (e) Suc and Glc efflux assays in HEK293T cells expressing the ClVST1⁹⁷ protein from cultivar 97 103 using the patch-clamp technique. Cells harboring ClVST1⁹⁷ responded with a Suc/Glc-induced strong inward current trace compared with applying raffinose (Raf), and the empty vector expressed the background current level. These Suc/Glc-activated inward currents mean that Suc/Glc flow out of cells. Black arrows indicate the time point for loading sugars into HEK293T cells. (f) Suc/Glc-induced current response rates determined under different pH conditions in HEK293T cells. N indicates the number of independent cells used for current response rate analysis. Means and SD from two biological repeats were used. **, P < 0.01 (t-test).

Discussion

Rapid unloading of sugar is clearly a gain-of-function selection. Suc concentrations are very high in modern sweet watermelon C. lanatus (CL) cultivars, whereas Suc is nearly absent in the nonsweet wild watermelons, C. colocynthis (CC) and C. amarus (CA), indicating that Suc accumulation is a domesticated trait. However, the semi-wild species C. mucosospermus (CM) consists of both sweet and nonsweet watermelons and is supposed to be the closest ancestor of the cultivated watermelon (Guo et al., 2012). To evaluate the selection of ClVST1, we analyzed the haplotype of ClVST1 and identified that the SNP C99A is the nonsynonymous mutation most associated with sugar content in a collection of 326 wild and cultivated watermelon accessions (Guo et al., 2019), with all nonsweet watermelons harboring the allele CC while sweet watermelons are AA genotype (Figs 1d, 6a). Consistently, the phylogenetic tree of all 326 accessions using SNPs of ClVST1 showed the nonsweet semi-wild CM was grouped with wild ancestor CC and CA, while the sweet CM accessions clustered with CL landraces and cultivars (Fig. 6b). Moreover, the ratio of nucleotide diversity (π) noted that ClVST1 was located in the selective sweep regions during the domestication process between the closest ancestor CM and CL group (Fig. 6c). These results indicate the selection for ClVST1 during domestication, which aimed at increasing the sugars unloading within the watermelon fruit.

Fig. 6

ClVST1 is a selected gene. (a) Haplotype of ClVST1 gene shows that the single nucleotide polymorphisms (SNPs) C99A and C42T in the coding region are the most associated mutations among the watermelon population consisting of Citrullus amarus (CA), Citrullus colocynthis (CC), Citrullus mucosospermus (CM) and Citrullus lanatus (CL). (b) Phylogenetic tree of all 326 accessions using SNPs within ClVST1. (c) ClVST1 locates in Chr 02 selection sweep region by π ratio screening between CM and CL groups. The horizontal dashed line indicates the top 7% of selection signals. The black arrow denotes that the ClVST1 gene is located in the selection sweep region.

Protein isoforms with different N termini that display differences in localization contribute to the functional diversity by allowing protein to perform different functions in different cellular compartments (Ushijima et al., 2017). We found that ClVST1⁹⁷ is a newly evolved gene that only exists allelic variation at SNP C99A in watermelon. Besides contributing to localization diversity, the N termini of proteins could also form an uORF to produce a short polypeptide to regulate the expression of the main ORF (Wiese et al., 2004). Therefore, it is likely that the small uORF formed by the N-terminal truncation of ClVST1⁹⁷ mediates expression of the main ORF. Here, we found that knockout (Fig. 3g,h) or overexpression (Fig. 3i,j) of the main ORF ClVST1⁹⁷ alone can result in decreased or increased sugar content and biomass, supporting the sugar partitioning role of the main ORF ClVST1⁹⁷. However, although the small uORF was transcribed together with the main ORF ClVST1⁹⁷, the mRNA level of the small uORF was at a similar level to that of the main ORF (Fig. S4). Therefore, further study on the small uORF should help in determining whether and how it alternates sugar content in watermelons.

Given that the 45 amino acids truncated in the N termini of ClVST1⁹⁷, caused by SNP C99A, only existed in sweet watermelons (Fig. S5a), and the presence of full ClVST1^PI in low-sugar-content semi-wild and wild watermelons, as well as in other crops (Fig. S5b), it would be interesting to know whether ClVST1 plays a similar role in other crops, because wild ClVST1^PI has no activity for transporting Suc or Glc (Fig. 5a,b). Further research on single-base substitution by CRISPR/Cas9 at the SNP C99A could be interesting; however, it is not available at present because no PAM sequences were found near this SNP. Moreover, relative expression of ClVST1⁹⁷ in cultivar 97 103 fruit flesh increased 40-fold from the to 34 DAP, while remaining at a low level in the 97 103 mesocarp, leaf and root, and in PI 296 341 fruit and other nonsweet accessions (Fig. 2a). These results indicate that ClVST1⁹⁷ mRNA levels are positively correlated with Suc content during sugar accumulation. To illustrate the possible polymorphisms that may determine the mRNA levels during domestication, we analyzed the variations in the upstream 1500 bp promoter among 326 accessions and screening transcription factor binding sites in these sequences. In total, we identified six SNPs in the upstream 1500 bp promoter of ClVST1 at −52 C→T, −97 C→T, −99 T→C, −1051 T→G, −1354 C→T and −1386 A→G (Table S3). Interestingly, only SNP −1051 (T→G) was located in the consensus CANNTG core binding motif of basic region/helix–loop–helix (bHLH) transcription factor (Fisher & Goding, 1992); however, other variations have no predicted binding sites for transcription factors. Furthermore, we analyzed the coefficient of association with sugar content in the 326 accessions and found that the allelic difference at −1051 T→G in the ClVST1 promoter was more highly associated with sugar content than other SNPs in the upstream 1500 bp promoter (Table S3). These results indicate that the SNP −1051 (T→G) may be a critical variation for bHLH binding and regulation between sweet and nonsweet watermelons. However, there was less correlation with sugar content at the SNP −1051 (T→G) than at the SNP C99A in the 326 accessions (Table S3), indicating that the role of SNP C99A has to do with the key polymorphisms associated with sugar content. Thus, further study is needed to reveal the upstream regulatory factors of the ClVST1 in watermelon.

Sugar efflux from the SE/CC complexes is carrier-mediated, and energy coupled with protons returning down their electrochemical gradient is generated by a plasma membrane H⁺-ATPase (Patrick, 1997). The Suc/H⁺ symporter SUC was thought to be involved in phloem unloading in stems of sorghum (Milne et al., 2017). However, RNA-Seq data showed that SUC family genes were not expressed in watermelon fruits (Guo et al., 2015). The axial phloem unloading from the SE/CC complexes into lateral sinks are irreversibly committed to importing sugars into fruit sinks (Patrick, 1997). These clues indicate that the presence of proton motive force of sugar/proton antiporter may impose on sugar unloading from phloem in watermelon or other fruits. Watermelon accumulates a high concentration of sugars in fruit. The sugar/proton antiporter ClVST1⁹⁷, as an ATP- and pH-dependent sugar efflux, may function in sugar unloading from fruit phloem by accumulating a high concentration of sugars in intercellular space. Subsequently, plasma membrane-localized sugar transporters in watermelon flesh could take up sugars from intercellular space. However, as there was no dramatic decrease in fruit sugar content in ClVST1 mutants, it is probable that more sugar transporters are involved in the process of phloem unloading, because it follows an apoplasmic phloem unloading pathway in watermelon fruit.

For decades, the research focus on phloem transport has shifted from translocation to phloem loading and unloading (Patrick, 1997). However, investigation of phloem unloading has been much less intense because phloem unloading studies are hindered by the fact that the vascular tissue is buried deep within the sink, as well as the lack of stable and high-throughput technology for sugar efflux assay. The phloem unloading follows apoplasmic or symplasmic pathways; CFDA and esculin staining shows that watermelon (Fig. S1) and other cucurbits crops (Hu et al., 2011) undergo apoplasmic unloading in vascular tissue. These facts prompted us to use genetic mapping, biochemistry, CRISPR/Cas9 technology, as well as to combine population genetics and multi-omics datasets to assess that the ClVST1⁹⁷ functions in sugar unloading in phloem tissue of watermelon. Thus, our results provide not only a novel strategy for phloem sugar unloading, through manipulation of the N termini of sugar transporter proteins, but also a deeper understanding of fruit quality selection during breeding.

Acknowledgements

This research was supported by grants from the National Key Research and Development Program of China (2018YFD1000800), National Natural Science Foundation of China (31930096, 31772328), Beijing Natural Science Foundation (6182015), Beijing Agriculture Innovation Consortium (BAIC10-2020), Beijing Scholar Program (BSP026) and USDA National Institute of Food and Agriculture Specialty Crop Research Initiative (2015-51181-24285).

Author contributions

YR conducted most experiments and wrote the manuscript. HS performed the SNP calling and RNA-Seq data analysis. MZ generated overexpression and mutant transgenic lines in watermelon. SG performed the bioinformatics analysis. ZR performed sugar update assays in oocytes. JZhao conducted the RNA in situ hybridization. ML performed the phenotyping of transgenic plants. JZhang provided technical assistance. ST participated in the phenotype analysis. JW contributed to the metabolic analysis. YY conducted localization analysis. GG planted the population and managed this procedure in the glasshouse. HZ performed the transient expression assays in protoplasts. XZ designed the RNA in situ hybridization. HH provided the sugar content analysis. LL contributed to the interpretation of the experimental results. FL provided transgenic assistance. ZF contributed to the design of RIL resequencing and manuscript revision. YX managed and designed the study and revised the manuscript.