2.1 Plant materials and growth conditions

In this study, the watermelon (Citrullus lanatus) cultivar 97103 (W) was used as the scion in all grafting combinations. Information on the eight watermelon rootstocks used in this study can be found in Table S1. Watermelon cultivars 97103 (W) and wild watermelon Yongshi (Y) were the primary rootstocks used for further grafting experiments involving scion growth measurements, paraffin section observation, sugar content determination, and RNA-seq analysis after grafting. All plants were grown in a controlled environment with a day/night (14/10 h) cycle at 300 μmol m⁻²·s⁻¹, 24°C, and 60% relative humidity.

2.2 Plant grafting and the culturing of grafted seedlings

Grafting was performed 8–12 days after sowing, after the emergence of the first true leaf. Two grafting methods were used: grafting with one rootstock cotyledon (+C), following the method described by Xu et al. (2022), and grafting without rootstock cotyledons (-C), in which all cotyledons and growth points of the rootstock were completely removed (Figure 1A).

The grafted seedlings were immediately placed in a healing box with a transparent plastic cover to maintain high humidity (95%–100%) and kept in a growth room with a 14-hour light/10-hour dark cycle at 24°C. The lighting schedules for the seedlings were as follows: on 0 Days After Grafting (DAG), they were kept in darkness. On 1 DAG, they received supplemental lighting at 100 μmol m⁻²·s⁻¹ for 30 minutes in both the morning and afternoon. On 2 DAG, the lighting duration was extended to 1 hour in the morning and afternoon. On 3 DAG, the seedlings were exposed to supplemental lighting at 100 μmol m⁻²·s⁻¹ for 14 hours. On 4 DAG, the light intensity was raised to 300 μmol m⁻²·s⁻¹ for 30 minutes in both the morning and afternoon. On 5 DAG, the lighting at 300 μmol m⁻²·s⁻¹ was extended to 1 hour in both the morning and afternoon, with the cover removed for 30 minutes during each period. From 6 to 8 DAG, the seedlings received 14 hours of supplemental lighting at 300 μmol m⁻²·s⁻¹, with the cover removed for 1 hour in each period. On 9 to 11 DAG, they continued to receive 14 hours of lighting at 300 μmol m⁻²·s⁻¹, with the cover removed for 14 hours. On 12 DAG, the cover was permanently removed, and seedlings were transferred to normal growth conditions.

Three weeks after grafting, the surviving grafts were transferred to a greenhouse located at the Wuhan Academy of Agricultural Sciences (Wuhan, China). Between 15–20 plants were established, and manual pollination was performed, resulting in one to two fruits retained per plant. This cultivation phase extended from August to November 2021.

2.3 Survival rate, scion growth, adhesive force testing, and fruit yield measurement

The survival rate was calculated 14 days after grafting using the formula: Survival rate = (number of survived grafted seedlings / total number of grafted seedlings) × 100%. Grafted seedlings were considered successful if they showed clear growth of the scion and maintained an upright posture. Each graft combination was repeated three times, with 24 grafted seedlings per replicate. For scion fresh weight, 15–24 grafted seedlings were weighed 14 days after grafting, and the average weight was calculated. The true leaf area was determined by laying each leaf of every grafted seedling flat on a ruler, photographing it, and calculating the leaf area using ImageJ (https://imagej.nih.gov/ij/) (Figure S7).

When the watermelons matured, the weight and fruit Brix (%) of each watermelon fruit were measured. The ATAGO(PAL-1) Portable Digital Fruit Brix Meter (ATAGO U.S.A. Inc.) was used to measure the Brix (Brix, %) values of each watermelon fruit.

2.4 Sample preparation

At 0, 6, 24, 72, 120, and 168 hours after grafting (HAG), the scions and rootstocks were separated, and the graft union regions of rootstocks were sampled as depicted in Figure S2, frozen in liquid nitrogen and stored in a − 80°C freezer for RNA-seq, biochemical analyses and qRT-PCR. For each graft combination, six grafted seedlings were randomly selected, and the entire sampling procedure was repeated four times.

2.5 Microscopy

Graft union sites from the rootstock and scion were collected at 24, 72, and 168 hours after grafting (HAG). The graft unions were initially fixed in a 70% FAA solution (ethanol: formaldehyde: acetic acid = 18:1:1, v/v/v) for 24 hours and then transferred to 70% ethanol for long-term preservation at 4°C. Paraffin embedding and sectioning were performed according to Xiong et al. (2021). Sections were cut vertically at 10 μm thickness using a rotary microtome (Leica RM2255), then dewaxed, rehydrated, and stained with 1% safranin for 2 minutes. After dehydration, the sections were counterstained with 1% fast green in 90% alcohol for 15 seconds, cleared, and mounted with neutral balata. Imaging was performed using a fluorescence microscope (Leica DM6B).

To monitor the proliferation of callus tissue at the incision sites under various treatments, watermelon seedlings were incised as depicted in Figure S8. The development of callus tissue at these sites was imaged daily using a stereo microscope Leica M205FA (Leica Microsystems GmbH). The management of both cut and grafted watermelon seedlings was maintained consistently throughout the observation period.

2.6 Metabolite quantification

The plant tissues were sampled and prepared as described above. The following quantification kits were used in this experiment: Plant Sucrose Content Assay Kit, AKPL006M; Plant Fructose Content Assay Kit, AKPL007M; D-Glucose Content Assay Kit, AKSU001M; Fructokinase (FRK) Activity Assay Kit, AKSU084M (Beijing Boxbio Science & Technology Co., Ltd). Each sample was tested thrice with three technical replicates. Absorbance values for sucrose (wavelength: 480 nm), fructose (wavelength: 480 nm), glucose (wavelength: 505 nm), and fructokinase (wavelength: 340 nm) using a TECAN Spark® Multimode Microplate Reader (Tecan Group Ltd.) were recorded for calculating the concentration or enzyme activity.

2.7 Transcriptome analysis

RNA-seq was performed by Biomarker Technologies Corporation, Beijing (http://www.biomarker.com.cn/). Nucleic acids were extracted using the following kits: Tiangen DP411 (Tiangen Biotech Co.), CLB + Adlai RN40 (Aidlab Biotechnologies Co.), CTAB+Adlai RN40 (Aidlab Biotechnologies Co.), and Tiangen DP762-T1C/TRIzol (Tiangen Biotech Co.). The concentration of the extracted nucleic acids was determined using a Nanodrop2000 (Thermo Fisher Scientific Co.), and their integrity was assessed with an Agilent2100 LabChip GX (Agilent Technologies Co.). Library construction was carried out using the Hieff NGS Ultima Dual-mode mRNA Library Prep Kit for Illumina (Yeasen Biotechnology Co.), and product purification was performed using HieffNGS DNA Selection Beads (Yeasen Biotechnology Co.).

Significant differences in gene expression were identified using a false discovery rate (FDR) of ≤0.01 and an absolute log2(fold change) (|log2(FC)|) greater than 1 (where FC is the fold change calculated by FPKM values). FPKM stands for fragments per kilobase of exon per million reads mapped. Gene functions were annotated using the Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) databases. Functional enrichment analyses of DEGs were performed to identify significant enrichments in specific GO terms or KEGG pathways. GO terms and KEGG pathways with a corrected P-value ≤0.05 indicate significant enrichment. Protein sequences were extracted from the provided link and submitted to STRING database to generate a protein–protein interaction network for HUB gene identification, focusing on the model plant Arabidopsis thaliana, with an interaction score threshold of 0.40.

2.8 Identification of the Watermelon FRK gene family

Watermelon Citrullus lanatus subsp. vulgaris cv. 97103 v2 genome was obtain from CuGenDBv2 (http://cucurbitgenomics.org/v2/), while the Arabidopsis TAIR10_genome was sourced from the Arabidopsis Information Resource (https://www.arabidopsis.org/download/list?dir=Genes/TAIR10_genome_release). Using the Arabidopsis FRK gene family data from Riggs et al. (2017), we obtained gene and protein information for AtFRKs in Arabidopsis thaliana. An initial selection of watermelon ClFRKs was performed using Tbtools-blast with the AtFRKs protein sequences. Conserved domains within AtFRKs were identified using the NCBI's CD-Search Tool. Watermelon ClFRKs were screened based on conserved domains (pfkB: PF00294, bac_FRK: cd01167) and compared with AtFRK1-7. A phylogenetic tree was constructed using MEGA11 with the protein sequences, employing the Maximum Likelihood (ML) method.

2.9 Exogenous sugar treatments

Sucrose, fructose, and glucose solutions (Hefei Bomei Biotechnology Co.) were prepared by dissolving the sugars in deionized water to achieve the desired concentrations. Approximately 2 mL of the solution was sprayed evenly onto the watermelon rootstocks 12–16 hours prior to grafting, ensuring the liquid was nearly dripping off. This spraying process was repeated one hour after grafting.

2.10 RNA extraction, cDNA synthesis and qRT-PCR

RNA was extracted using the TranZol protocol (Code: ET101-01, Beijing TransGen Biotech Co.). cDNA synthesis followed the procedures outlined in the TransScript® One-Step gDNA Removal and cDNA Synthesis SuperMix Kit (Code: AT311-02, Beijing TransGen Biotech Co.). Primers were designed using Primer3Plus (https://www.primer3plus.com/index.html). Real-Time Quantitative Reverse Transcription PCR (qRT-PCR) was conducted using the 2 × TransStartTM TOP Green qPCR SuperMix Kit (Code: AQ131-01, Beijing TransGen Biotech Co.) on a QuantStudio 7 Flex Real-Time PCR System (Applied Biosystems). ClADP (Watermelon ADP/ATP transport protein) served as the reference gene, and relative gene expression levels were calculated using the 2^-ΔΔCt method (Livak & Schmittgen, 2001). Primer sequences are provided in Table S2.

2.11 Molecular cloning and transformation

The Cucumber Green Mottle Mosaic Virus (CGMMV) vector was used for VIGS experiments to silence the ClFRK1 (Cla97C01G005540) gene (Liu et al., 2020). The 431–686 bp region of the ClFRK1 gene was selected as the silencing fragment. This fragment was amplified by PCR and cloned into the CGMMV vector. Primer sequences are listed in Table S2. The vector was assembled through homologous recombination following digestion with BamHI. Agrobacterium strain GV3101 harboring the CGMMV vectors was cultured at 28°C in LB medium supplemented with appropriate antibiotics. After 24 hours, the Agrobacterium cells were harvested and resuspended in infiltration buffer (10 mmol/L MgCl₂, 10 mmol/L MES, 100 μmol/L acetosyringone), adjusting the optical density to OD600 = 0.4. The seed infection method was employed as follows: watermelon seeds were first sterilized in a 55°C water bath for 30 minutes. The seed coats were then removed, and the seeds were incubated at 30°C for 48 hours to promote germination. Subsequently, the germinated seeds were soaked in 40 mL of Agrobacterium inoculum, ensuring they were partially immersed, and incubated in the dark at 20–23°C for 24 hours. After incubation, the inoculated seeds were placed on sterile filter paper to absorb any excess inoculum before sowing.

3.1 Removal of rootstock cotyledon decreases the graft survival rate and delays the healing process

To explore the role of rootstock cotyledons in the graft recovery process, we utilized two grafting methods. In one method, we retained one rootstock cotyledon (+C), and in the other, we removed both cotyledons (-C) (Figure 1A). For these experiments, the watermelon cultivar 97103 (Citrullus lanatus) was grafted onto eight different watermelon rootstocks (Figure 1B, Table S1), and the graft survival rate was assessed 14 days after grafting (DAG). We observed that the survival rate significantly decreased following the removal of the rootstock cotyledons (Figure 1B, C). Notably, grafts without rootstock cotyledons exhibited higher survival rates in the four wild watermelon rootstocks compared to the four cultivated watermelon rootstocks, where the survival rates were lower (Figure 1C).

Next, we selected one wild rootstock, called “Yongshi” (Y), and one cultivar rootstock, known as “97103” (W) to further investigate cotyledon-related survival rates. We observed a slight decrease in survival rate for W/Y -C (from 100% to 66.4%) and a substantial decrease for W/W -C (from 97.1% to 27.8%) (Figure 1C). During the healing process, we monitored the total area of true leaves of the scion and found that the growth rate was fastest in W/Y + C, followed by W/Y -C, and slowest in W/W -C (Figure 1D). The fresh weight of the scion at 14 days after grafting (DAG) corroborated these observations, showing a significant decrease in scion fresh weight due to cotyledon removal, with a larger reduction in W/W -C (19.09%) compared to W/Y (51.29%) (Figure 1E).

To better understand the observed differences in graft survival rates and growth outcomes, we examined paraffin sections of the graft surface at 3, 5, and 7 DAG. At 3 DAG, no significant differences were observed among W/Y + C, W/Y -C, and W/W + C; however, W/W -C showed gaps at the graft surface that were not filled with cellular tissue (Figure S1). By 5 DAG, W/Y + C and W/W + C exhibited minimal necrotic tissue and showed signs of vascular reconnection, with W/Y + C demonstrating more pronounced reconnection. By contrast, W/Y -C and W/W -C still exhibited a distinct necrotic layer (Figure S1). At 7 DAG, differences between graft combinations became more evident. W/Y + C and W/W + C showed no necrotic layer, whereas W/Y -C and W/W -C had persistent necrotic layers (Figure 1F). Completed vascular reconnection was observed in either W/Y + C or W/Y -C, but in W/W, it only occurred in W/W + C but not in W/W -C (Figure 1F). These results suggest that the absence of rootstock cotyledons impaired graft healing, particularly in cultivated watermelon rootstocks, leading to increased graft failure.

3.2 Impact of cotyledon removal on gene expression at the graft junction

To better understand the effects of cotyledon removal, we conducted RNA-seq analyses on the rootstock portion of the graft junction at 0, 6, 72, 120, and 168 hours after grafting (HAG) (Figures 2A-E and S2). The Principal Component Analysis (PCA) clustering revealed distinct differences in transcriptome profiles not only between grafts with and without rootstock cotyledons but also between the W/Y and W/W graft combinations, with the greatest variation occurring at 6 HAG (Figure 2A). Next, differentially expressed genes (DEGs) were analyzed by DESeq2 (Fold Change≥2, FDR <0.01). Overall, 1947 genes were upregulated and 3110 genes were downregulated in W/W -C, whereas 2122 and 2494 were up- and downregulated in W/Y -C, respectively (Figure 2B). Interestingly, the highest number of DEGs (3200 genes) was observed at 168 HAG in W/W, while the highest number of DEGs (2944 genes) was observed at 6 HAG in W/Y (Figure 2B).

To further compare DEGs in W/W and W/Y following cotyledon removal, we generated Venn diagrams to visualize the proportion of shared and unique DEGs at each time point (Figure 2C). Next, KEGG and GO enrichment analyses were performed on DEGs shared between W/W and W/Y to obtain a broad picture of biological processes among different post-grafting stages (Figures 2D-E and S3). Among all time points, the downregulated DEGs were significantly enriched in KEGG categories, including ‘starch and sucrose metabolism’ (ko00500), ‘fructose and mannose metabolism’ (ko00051) and ‘other glycan degradation’ (ko00511). Meanwhile, the upregulated genes were significantly enriched in the metabolic pathways of various amino acids such as ‘valine, leucine and isoleucine degradation’ (ko00280) and ‘alanine, aspartate and glutamate metabolism’ (ko00250). Additionally, genes related to ‘propanoate metabolism’ (ko00640) and ‘glyoxylate and dicarboxylate metabolism’ (ko00630) were significantly upregulated (Figure 2D). Consistent with our hypothesis, these results suggest that cotyledon removal impacts key metabolic processes across sugars, amino acids, and fatty acids.

GO term enrichment analysis of the DEGs at different time points revealed that downregulated genes were significantly enriched in biological processes related to cell division and tissue development, such as ‘DNA replication’ (GO:0006260), ‘unidimensional cell growth’ (GO:0009826), and ‘microtubule cytoskeleton organization’ (GO:0000226). Processes related to cell wall component biogenesis and degradation, including ‘cell wall biogenesis’ (GO:0042546), ‘cellulose biosynthetic process’ (GO:0030244), ‘lignin catabolic process’ (GO:0046274), and ‘pectin catabolic process’ (GO:0045490) were also significantly enriched (Figure 2E, Figure S3). These results indicate that the removal of cotyledons led to the attenuation of these processes, consistent with the observed persistence of necrotic layers and gap formation at the graft junction in grafts without rootstock cotyledons at 168 HAG (7 DAG), compared to grafts with cotyledons where these issues were less apparent (Figure 1F). Moreover, downregulated genes were also enriched in processes related to reactive oxygen species (ROS) degradation, such as ‘hydrogen peroxide catabolic process’ (GO:0042744) and ‘peroxidase activity’ (GO:0004601). Conversely, upregulated genes were significantly enriched in stress- and defense-related pathways, including the ‘response to reactive oxygen species’ (GO:0000302), ‘abscisic acid-activated signaling pathway’ (GO:0009738), and ‘response to abiotic stimulus’ (GO:0009628) (Figure 2E). These findings suggest that the removal of cotyledons led to reduced ROS degradation at the graft junction, triggering a stronger stress and defense response. In terms of molecular function, downregulated genes were significantly enriched in ‘ATP binding’ (GO:0005524), reflecting the reduced energy supply due to impaired sugar metabolism. These findings align with the KEGG pathway enrichment results, reinforcing the conclusion that the removal of cotyledons weakens sugar metabolism, leading to an energy deficit that impairs the graft healing process.

3.3 Protein–protein interaction network analysis of graft phenotype-related genes

Grafts without rootstock cotyledons (-C) had lower survival rates as well as weaker growth and healing compared to grafts with rootstock cotyledons (+C). Further, we found that the W/W grafting without cotyledons (W/W-C) had lower survival rates, weaker growth and healing compared to the W/Y grafting without cotyledons (W/Y-C) (Figure 1). Expanding on this result, by comparing W/W and W/Y, we found that the W/W-C combination exhibited reduced healing (delayed about 3 days) and survival rates (58.18% lower) relative to W/Y-C (Figures 3A, 1C, and S1). To investigate the underlying genetic factors, we established four differential comparison groups: (1) graft-induced genes “0 HAG_vs_+C up” (encompassing all DEGs upregulated in the four graft combinations at 6, 72, 120, and 168 HAG compared to 0 HAG); (2) genes that responded to rootstock cotyledon removal, “W/W + C_vs_W/W-C down” (including all DEGs downregulated in W/W-C compared to W/W + C), and (3) “W/Y + C_vs_W/Y-C down” (including all DEGs downregulated in W/Y-C compared to W/Y + C); and (4) DEGs between different watermelon species grafted without cotyledons, “W/W-C_vs_W/Y-C up” (including all DEGs upregulated in W/Y-C compared to W/W-C) (Figure 3B). Venn analysis of these groups identified 290 hub genes associated with graft phenotype (Figure 3B).

Subsequent protein–protein interaction (PPI) network analysis of these 290 hub genes identified three primary interaction regions: Region I (carbohydrate metabolism), Region II (cell wall processes), and Region III (cell division) (Figure 3C). In Region I, the central enzyme fructokinase FRK7, which is involved in fructose metabolism, played a crucial role. Region II showed strong interactions among proteins involved in DNA replication, cell cycle, and cell division, including Cell Division Cycle proteins (CDC45, CDC6B), Kinesins (KIN5C, KIN10A), and Cyclins (CYCA1-1, CYCB2-4). In Region III, key cell wall enzymes such as Expansin (EXPB3) and Cellulose Synthase (CESA8) were identified (Figure 3D, E). Notably, the expression patterns of FRK7 aligned with the observed graft phenotypes (Figure 1B, D, E), suggesting a potential role for FRK7 in response to cotyledon removal and graft healing. Further characterization of the watermelon FRK gene family revealed that FRK7 is equivalent to ClFRK1, located on chromosome 1 (Figure S4).

3.4 Exogenous sugar treatment promotes graft healing in watermelon

To further evaluate the potential changes in sugar content at the graft junction after cotyledon removal, we measured sucrose, glucose, and fructose levels of the rootstock cotyledon and the graft junction. Interestingly, each sugar decreased during the first 24 hours after cotyledon removal, then continuously increased thereafter (Figure 4A). In grafts with cotyledons (W/W + C), sucrose and fructose levels at the graft junction were high and similar to those in the rootstock cotyledons, whereas the contents of these sugars were lower in grafts without cotyledons (W/W-C) (Figure 4A). Glucose content was highest in the rootstock cotyledons, followed by the graft junction in W/W + C, and was lowest in the graft junction of W/W-C (Figure 4A).

Moreover, the sugar content of the hypocotyls of the eight watermelon rootstocks was measured. The results showed that the sugar content (sucrose, glucose, and fructose) among the rootstocks showed a strong positive correlation (R² = 0.7071, p = 0.0089) with graft survival rates when these rootstocks were used without cotyledons (Figures 1C and 4B, C). To further investigate the relationship between sugar content and graft survival rate, we applied exogenous sucrose (Suc), glucose (Glu), and fructose (Fru) treatments to grafts without rootstock cotyledons. Three sugar treatments significantly improved graft survival rates and their optimal concentrations were 3%, 2%, and 2%, respectively (Figure 4D). Optimal sugar treatments also increased scion fresh weight, leaf number, leaf area at 14 DAG, and adhesive force at 72 HAG (Figure 4E, H). Taken together, these results provide strong evidence for high sugar content as a crucial determinant for the grafting success of a rootstock, and fructose shared the best performance in each character.

Next, we evaluated the effects of different sugar treatments during the graft healing period on the development of watermelon self-grafts (W/W) (Figures 4I, J and S5). Grafted plants were treated with various sugars during the healing period and then transplanted into the greenhouse 3 weeks after grafting. At 7 weeks after grafting, all sugar treatments increased the single fruit weight by approximately 33% in W/W-C, matching the weight observed in W/W + C (Figure 4I, J). Additionally, these treatments mitigated the losses in scion length caused by the removal of rootstock cotyledons (Figure S5 A, B).Bby contrast, the central and edge Brix values of the fruit were unaffected (Figure S5 C). Given the common use of pumpkin as a rootstock for watermelon in agricultural settings, we extended our sugar treatment tests to watermelon/pumpkin (W/P) grafts. Compared to W/P-C, the grafting survival rate increased by 48.1% in glucose treatment, 50.0% in sucrose treatment and 54.2% in fructose treatment. (Figure S6). Moreover, sugar treatments enhanced the growth of both the scion and rootstock in W/P-C. (Figure S6 A, B); Similar increases in single fruit weight (3%Suc: 29.6%, 2%Fru: 35.2%, and 2%Glu: 32.6%) and scion length (3%Suc: 8.2%, 2%Fru: 21.8%, and 2%Glu: 34.6%) were also observed (Figure S6 C-E).

3.5 Fructose regulates watermelon grafting healing through CIFRK1

Sugars played a crucial role in the graft healing processes, and the absence of cotyledons during grafting, which led to reduced sugar availability at the grafting interface, is a primary cause of grafting failure (Figure 4C). Protein–protein interaction (PPI) network analysis identified ClFRK1 as a key gene in the fructose metabolism pathway (Figure 3A-E). Based on our results, we hypothesized that ClFRK1 was activated by fructose and regulated the graft healing process.

To test this hypothesis, we first used a virus-induced gene silencing (VIGS) approach to silence the expression of ClFRK1 during the graft healing period (Figure 5). In grafts with rootstock cotyledons (+C), silencing of ClFRK1 led to a significantly reduced survival rate in the ClFRK1-silenced rootstock combination (CIFRK1-VIGS) compared to non-inoculated (NI) and empty vector-inoculated control rootstocks (VC) (Figure 5). This outcome was comparable to grafting without rootstock cotyledons (-C). Exogenous application of fructose to the hypocotyls of intact watermelon seedlings significantly increased ClFRK1 expression and fructokinase activity (Figure 5). Additionally, the PPI results linked several cell division-related genes with ClFRK1, supporting its involvement in cell division processes. Callus tissue observations revealed significantly less callus formation in ClFRK1-silenced grafts compared to NI grafts (W/W + C) and were similar to those observed in W/W-C (Figure 5). Exogenous fructose treatment mitigated these adverse effects caused by the removal of rootstock cotyledons and silencing of ClFRK1 (Figure 5). Microscopic observations at 168 hours after grafting (HAG) revealed no necrotic layers and extensive xylem reconnection between the scion and rootstock in both W/W + C and W/W-C treated with fructose. By contrast, W/W-C and ClFRK1-silenced combinations exhibited considerable necrotic layers and gaps, with no evident xylem reconnection (Figure 5). Importantly, these results suggest that high ClFRK1 expression is positively associated with callus proliferation at the grafting junction, indicating a critical role in promoting graft healing.