Function and mechanism of double‐stranded RNA‐binding protein 1 in small RNA metabolism in tomato

SUMMARY

Tomato is an important vegetable crop that is rich in genetic variation, but published high‐quality tomato reference genomes are limited. Small RNAs play ubiquitous roles in plants, but their biogenesis in tomatoes is not well studied. We assembled a high‐quality chromosome‐level reference genome for the tomato cultivar Ailsa Craig 57 (A57) and generated CRISPR mutants deficient in double‐stranded RNA‐binding protein 1 (SlDRB1) in this cultivar. Mutations in SlDRB1 led to severely abnormal development (e.g., loss of leaf polarity, pollen abortion, and infertility). High‐throughput sRNA sequencing analysis revealed that SlDRB1, similar to its Arabidopsis ortholog Hyponastic Leaves 1 (HYL1), is required for the biogenesis of most miRNAs. Interestingly, SlDRB1 is also required for the accumulation of a large population of small RNAs derived from rRNAs and negatively regulates miR6026, which is dependent on DCL2 instead of DCL1. Comprehensive protein–protein interaction studies between SlDRB1 and SlDCL1 uncovered detailed interaction mechanisms involving subdomains of these two proteins. Their potential role in miRNA biogenesis is discussed.

Significance Statement

This study provides a high‐quality chromosome‐level reference genome for tomato cultivar Ailsa Craig 57 and a user‐friendly webserver, which provides a crucial genomic resource for tomato research and improvement. The research also reveals that SlDRB1 plays a dual role in microRNA biogenesis and stabilization of rRNA‐derived small RNAs, and provides new mechanistic insights into the SlDRB1‐mediated regulatory mechanism of small RNA biogenesis by demonstrating interactions between SlDRB1 and different domains of SlDCL1.

INTRODUCTION

Plant small RNAs, including microRNAs (miRNAs) and small interfering RNAs (siRNAs), play important gene regulatory roles in plant development, responses to various environmental signals, and both biotic and abiotic stress tolerance (Chen, 2009; Khraiwesh et al., 2012; Liu et al., 2023; Song, Li, et al., 2019; Zhao et al., 2025; Zhou et al., 2024). Plant miRNAs and siRNAs are 20–24‐nucleotide (nt) sRNAs that are produced by Dicer‐like (DCL) proteins: miRNAs from single‐stranded primary miRNA (pri‐miRNA) with stem‐loop structures and siRNAs from double‐stranded RNA precursors (Bologna & Voinnet, 2014; Chapman & Carrington, 2007). Both miRNA and siRNA bind to the Argonaute (AGO) protein to form an RNA‐induced silencing complex (RISC). Based on the complementarity between the mature miRNA (siRNA) and the target RNA, RISC may catalyze the directed cleavage of target RNAs, inhibit the translation of mRNA, or contribute to the methylation of DNA (Fang & Qi, 2016). Plant genomes harbor multiple DCL and AGO genes, which form parallel pathways that are specialized for miRNA and siRNA (Bologna & Voinnet, 2014; Chapman & Carrington, 2007).

The biogenesis of plant miRNA has been studied extensively (Achkar et al., 2016; Yu et al., 2026). In plants, pri‐miRNAs are transcribed by RNA polymerase II (Pol II) and are subjected to splicing like pre‐messenger RNAs (pre‐mRNAs) (Szarzynska et al., 2009; Xie et al., 2005). Once transcribed, pri‐miRNAs are transferred to the dicing body (D‐body), which is a membraneless organelle formed by liquid–liquid phase separation (LLPS) (Xie et al., 2021). The D‐body contains the core miRNA processor Dicer‐like 1 (DCL1), Serrate (SE), and Hyponastic Leaves 1 (HYL1), which form the microprocessor that cleaves pri‐miRNA to yield 20–24‐nt miRNA/miRNA* duplexes with 2‐nt 3′ overhangs (Fang & Spector, 2007; Song et al., 2007). The majority of the miRNAs have a 5′ terminal U, which favors loading into AGO1‐RISC and functions at the posttranscriptional level. Mature miRNAs in the RISC complex are degraded by the small RNA degrading nuclease 1 (SDN1) and uridylation‐triggered degradation to maintain their homeostasis (Ramachandran & Chen, 2008; Zhao et al., 2012).

In addition to miRNAs, a novel class of small non‐coding RNAs derived from ribosomal RNA, termed rRNA‐derived small RNAs (rdsRNAs), has been identified in both animals and plants (Chen et al., 2017). These molecules are generated through the cleavage of ribosomal RNA, with the majority originating from the 28S rRNA, 18S rRNA, 5.8S rRNA, and 5S rRNA gene sequences, and a minority derived from internal transcribed spacer (ITS) regions (Chen et al., 2017; Lambert et al., 2019; Rosace et al., 2020). In the past, rdsRNAs were regarded as degradation by‐products and were routinely removed during data analysis filtering steps (Choy et al., 2015; Fowler et al., 2018). However, recent studies have revealed that rdsRNAs possess potential biological functions. For example, rdsRNAs can associate with AGO proteins and participate in RNA interference pathways (Guan & Grigoriev, 2021). Moreover, some rdsRNAs target RPS28 mRNA, which provides evidence for a role in regulating ribosomal protein biosynthesis (Guan & Grigoriev, 2021). Regarding their biogenesis, the production of rdsRNAs may not be entirely Dicer‐dependent. For instance, under conditions of Dicer deficiency, certain rdsRNAs retain their ability to bind AGO2 (Lambert et al., 2019). Furthermore, in Arabidopsis thaliana, the expression levels of 5.8S rRNA‐derived rdsRNAs are significantly elevated in both the dcl1 single mutant and the dcl234 triple mutant. These data further indicate that the biogenesis of rdsRNAs is negatively regulated by Dicer‐mediated pathways (Asha & Soniya, 2017).

In plants, HYL1 has been extensively studied because it is one of the key players in miRNA biogenesis. The first evidence that HYL1 contributes to miRNA biogenesis came from genetic screens for mutants that were phenotypically similar to the DCL1 mutant (Han et al., 2004; Vaucheret et al., 2004; Vazquez et al., 2004). Recent findings have shown that HYL1 affects the localization of Pol II to MIR genes, interacts with many transcription factors, and protects pri‐miRNAs from being attacked by the nuclear exosome (Bielewicz et al., 2023; Gao et al., 2020). These data suggest that HYL1 regulates the transcription of genes encoding pri‐miRNAs and the stability of pri‐miRNAs. It was reported that temperature influences the secondary structure of pri‐miRNAs and that some miRNAs, including miR156, miR161, miR167, miR171, and miR173, are HYL1‐dependent at high temperatures but not at low temperatures (Ré et al., 2019), which is consistent with HYL1 facilitating the folding of pri‐miRNA secondary structures that are necessary for correct processing. HYL1 was also shown to interact with HEN1 (Baranauske et al., 2015), with a miRNA/miRNA* duplex after its release from pre‐miRNA (Xie et al., 2021), and with AGO1 in the endoplasmic reticulum monitoring its distribution onto polysomes (Yang et al., 2021). These results indicate that HYL1 may also deliver miRNA/miRNA* duplexes for downstream modifications, RISC loading, and translational repression of target mRNA. The core function of HYL1 is to promote efficient and precise processing of pri‐ and pre‐miRNAs by interacting with DCL1 and SE to form the D‐body (Dong et al., 2008; Fang & Spector, 2007; Song et al., 2007). Several studies on the interactions between HYL1 and DCL1 have revealed the importance of dsRBD2 from both HYL1 and DCL1 in mediating their interactions (Liu et al., 2013; Yang et al., 2014). However, a comprehensive mapping of the domains required for interactions between HYL1 and DCL1 is not available yet, and the role of these interactions in the biogenesis of other small RNAs remains to be further studied.

Tomato is an important vegetable worldwide and a model plant for fruit biology. Although the involvement of miRNAs in tomato fruit ripening and leaf development was reported almost 20 years ago (Moxon et al., 2008; Ori et al., 2007), small RNA biogenesis is not well studied in tomato due to a lack of mutants. The tomato genome sequence was first reported in 2011 (Tomato Genome, 2012). The Sol Genomics Network (SGN) webserver has provided online access and easy‐to‐use tomato genome information since that time (Bombarely et al., 2011), offering key resources for the tomato research community. Tomato is rich in genetic variation, and recent advancements in long‐read sequencing technology now allow a single lab to build its own high‐quality reference genome for tomato research. In this study, we sequenced and assembled a high‐quality chromosome‐level reference genome for our working variety Ailsa Craig 57 (A57). We introduced mutations into the HYL1 homolog of double‐stranded RNA‐binding protein 1 from tomato (SlDRB1) using CRISPR/Cas9 technology. We sequenced small RNAs and mRNAs. Then we compared the sRNA and mRNA profiles. Alignments to the newly assembled and annotated genome revealed that SlDRB1 positively regulated the abundance of a large set of rRNA‐derived sRNAs (rdsRNAs), possibly by protecting sRNAs from tailing and degradation. However, we found that a small set of rdsRNAs was negatively regulated by an unknown mechanism. As in Arabidopsis, SlDRB1 was required for the biogenesis of most miRNAs in tomato, but a few miRNAs were up‐regulated indirectly. Comprehensive studies on the interactions between SlDRB1 and SlDCL1 revealed that SlDRB1 forms homodimers and heterodimers with SlDCL1 through their dsRBD2 domains. In addition, SlDRB1 and SlDCL1 interacted using the dsRBD1 and C‐terminal domains of SlDRB1 and the PAZ domain of SlDCL1.

RESULTS

High‐quality assembly of a chromosome‐level reference genome of tomato cultivar Ailsa Craig 57

To generate a chromosome‐level reference genome for the tomato cultivar Solanum lycopersicum Ailsa Craig 57 (A57), we first employed PacBio HiFi CCS sequencing to obtain ~29 Gb of high‐fidelity long reads (~34× coverage). Assembly with HiFiasm (Cheng et al., 2021) yielded 914 Mb of contigs. We then applied three rounds of Pilon (Walker et al., 2014) polishing to correct residual base errors using 54 Gb of Illumina short‐read data (~64×). To anchor contigs onto chromosomes, we performed Hi‐C sequencing, generating ~127 Gb of sequence data (~149×). After read mapping and stringent filtering, we anchored contigs to 12 pseudo‐chromosomes, totaling 807.5 Mb. By removing 93.33 Mb of redundant contigs, we obtained a final assembly of 820.96 Mb, with a chromosome anchoring rate of 98.31% (Figure 1a; Table 1). The contig N50 of the A57 final assembly is 36.04 Mb, which is markedly higher than those of SL2.5 and SL4.0, which are 0.09 and 6.01 Mb, respectively (Table 1). BUSCO analysis (Simão et al., 2015) against the eukaryotic single‐copy gene set showed that the A57 genome achieved 98.46% completeness (including CS and CD), slightly higher than those of SL2.5 (98.39%) and SL4.0 (98.33%) (Figure 1c; Table 1). These results indicate that the A57 assembly matches or exceeds existing tomato references in both contiguity and gene completeness.

Figure 1.

Assembly and quality assessment of the tomato A57 genome.

(a) Normalized inter‐chromosomal contact count heatmaps showing Hi‐C‐based interaction density among contigs from A57.

(b) Dot plot showing the relationship between the A57 chromosomes and SL4.0 reference chromosomes.

(c) Comparison of BUSCO completeness among different Solanum lycopersicum genome assemblies and their corresponding proteomes. CD, complete duplicate; CS, complete single copy; F, fragmented; M, missing BUSCO gene; * and ** denote number and percent, respectively, of genes detected in each category.

Table 1.

Comparison of different genome assemblies

	A57	SL 2.5	SL 4.0
Contig N50 (Mb)	36.04	0.09	6.01
Longest Contig (Mb)	66.78	2.49	26.29
Total Contig number	425	26 965	448
Scaffold N50 (Mb)	67.62	66.47	65.27
Longest scaffold (Mb)	94.85	98.54	90.86
Total scaffolds size (Mb)	820.96	823.94	782.52
% of assembled chromosome	98.31	97.35	98.77
Predicted gene number	33 689	34 725	34 688
Average CDS (bp)	1111	1036	1005
Average exon number	4.69	4.61	4.74

CDS, coding sequence.

Using the SL4.0 tomato reference as a scaffold, we performed chromosome‐level synteny mapping and reordered the A57 sequences to match SL4.0's linear chromosome layout (Figure 1b). Gene annotation combined ab initio prediction, homology‐based methods, and RNA‐seq evidence, resulting in 33 689 high‐confidence gene models that are comparable to SL4.0 (Table 1). The average coding sequence (CDS) length in A57 is 1111 bp, which is slightly longer than the 1005 bp average CDS length reported for SL4.0 (Table 1). A BUSCO assessment of the predicted proteome recovered 94.67% of conserved single‐copy genes in A57, outperforming ITAG2.4's 91.57% and ITAG4.1's 86.56% (Figure 1c), demonstrating a marked improvement in annotation completeness and accuracy.

To facilitate easy access to our newly sequenced A57 genome and annotation, we set up a webserver, named TomatoHub, and provide commonly used web‐based services, such as BLAST tools and JBrowse (Figure S1A,C). A gene ID converter module was also created for users to convert gene ID between different genome versions (Figure S1B). TomatoHub is accessible at http://lifenglab.hzau.edu.cn/TomatoHub.

Generation and phenotypic characterization of SlDRB1 mutants using CRISPR/Cas9

Genetic studies in the model plant A. thaliana have shown that HYL1 plays essential roles during the biosynthesis of several conserved miRNAs (Szarzynska et al., 2009). In order to study its function in tomato, the HYL1 gene homolog, SlDRB1, was edited in S. lycopersicum cv. A57 and three independent CRISPR mutant lines were obtained by targeting three sites in the second exon of the SlDRB1 gene (Methods). Since the development of the mutants was severely impaired and no progeny were produced, three T0 lines were analyzed using high‐throughput sequencing. In Line slydrb1‐1, 80.88% of sequencing reads contained mutations leading to amino acid substitutions or premature stop codons. Predominantly, a 305‐bp deletion between target sites 1 and 3 (accounting for 34.73%) resulted in an amino acid substitution at the ninth codon and a premature stop codon at the 10th codon. In Line slydrb1‐2, 66.75% of reads showed evidence of amino acid changes or premature stop codons. We observed two major editing types: a 3‐bp deletion at target site 2 leading to the deletion of the 33rd amino acid residue with no other changes and a 1‐bp insertion at target site 2 leading to a change at the 34th amino acid residue and the conversion of the 35th codon to a premature stop codon. In Line slydrb1‐3, 62.67% of reads showed evidence of amino acid substitutions or premature termination codons. We observed two major editing types: a 3‐bp deletion after nucleotide 2258 resulting in the deletion of the 70th amino acid residue with no other changes and a 2‐bp insertion at target site 2 leading to a premature stop at the 52nd codon (Figure 2a; Figure S2D; Data S7).

Figure 2.

Characterization of the SlDRB1 knockout lines.

(a) Schematic diagram of the gene structure and genomic sequences from the SlDRB1 knockout lines. White rectangles represent UTRs, black rectangles represent coding sequence (CDS), and lines represent introns. The gRNA sequence and its relative position are indicated by red letters and dashed lines, while the PAM (NGG) sequence is marked with blue letters. The sequence data from the wild‐type (WT) and three T0 knockout lines are shown in the boxes, with the edited sites highlighted in yellow (deletions are represented by dashed lines and insertions by bold bases). The red asterisks denote the termination of translation. The percentage of each editing type is indicated on the right.

(b) Phenotypic comparison between tomato SlDRB1 mutants and WT. Plant phenotypes of 45‐day‐old WT (left) and slydrb1‐2 mutant (right), scale bar = 5 cm; floral organ morphology at flowering stage and pollen viability assay (Alexander staining), left: WT, right: slydrb1, scale bar = 2 cm; leaf morphology comparison, top: WT, bottom: slydrb1, scale bar = 2 cm.

(c) Comparison of pollen traits between WT and slydrb1 mutants. Statistical parameters include total pollen count and pollen abortion rate. Sample sizes: n (WT) = 3, n (slydrb1) = 3. For pollen count, the P‐value was 0.0038, indicating a statistically significant difference (**P < 0.01). For pollen abortion rate, the P‐value was less than 0.0001 (****P < 0.0001), as determined by a t‐test.

(d) Quantitative traits characterized in WT and slydrb1 mutants. Statistical parameters include plant height and stem diameter. Sample sizes: n (WT) = 3, n (slydrb1) = 3 for both traits. The P‐value for plant height was 0.0444 (*P < 0.05) and for stem diameter was less than 0.0001 (****P < 0.0001). Statistically significant differences were analyzed using a t‐test.

(e) Phenotypes of 90‐day‐old tomato SlDRB1 mutants and wild‐type (WT) plants. Apical meristem phenotype of slydrb1‐1 and whole‐plant phenotypes of slydrb1‐2/3 and WT plants, scale bar = 5 cm.

Phenotypic characterization showed that mutations in SlDRB1 led to leaf deformations that varied from narrow to thread‐like leaves lacking any leaf polarity (Figure 2e). We found that wild‐type and mutant plants produced an average of 67 and 7 pollen grains per anther, respectively, and that the pollen abortion rates were 0.06 and 0.92, respectively (Figure 2c). Thus, the SlDRB1 mutation also led to severe reductions in fertility. The heights, stem diameters, and flower size were also reduced in mutants relative to wild‐type plants (Figure 2b,d). A previous study reported developmental defects in the Arabidopsis hyl1 mutant. These developmental defects included dwarf plants, narrow leaves, delayed flowering time, and reduced fertility (Lu & Fedoroff, 2000). The developmental defects are more severe in the tomato slydrb1 mutant than in the Arabidopsis hyl1 mutant.

SlDRB1 contributes to the accumulation of ribosomal RNA‐derived small RNAs

In order to determine the function of SlDRB1 in small RNA biogenesis, we sampled leaf, petiole, stem, and shoot tissues and compared the sRNA levels in A57 and slydrb1‐1 using sRNA‐seq. We obtained a list of 11 million sRNA sequences of 18–30‐nt with different numbers of reads in different libraries (Data S1). To simplify the analysis, we focused on the 20–24‐nt RNAs with relatively high‐expression levels, which we termed high‐expression sRNAs (hsRNAs); these were defined as sRNAs that had 10 transcripts per million (TPM) in at least one sample because functional sRNAs associated with various AGO proteins are usually 20–24‐nt RNAs (Mi et al., 2008). After the initial data processing and filtering, 19 605 such high‐level sRNAs were retrieved (Data S2), and 13 185 of them were mapped to the A57 genome (Data S3). Annotation of these mapped sRNAs revealed that 11.9% were from a gene region, 50.0% were from an rRNA region, 4.1% were from a repeat region, 12.2% were from regions associated with both genes and repeats, and the remaining 21.6% were miRNAs and other non‐annotated loci (Figure 3a). The high abundance of repeat sRNAs is mainly derived from those repeats in gene regions. Among the repeat sRNAs, 32.2 and 54.8% were from DNA transposons and retrotransposons, respectively (Figure 3a), and 86.3% of the retrotransposon sRNAs were mapped to long terminal repeat (LTR) retrotransposons (Figure 3a). This analysis revealed that the rRNA‐derived sRNAs (rdsRNAs) were the most abundant sRNA population in the sequencing library.

Figure 3.

Analysis of rRNA‐derived sRNAs.

(a) Classification of highly expressed sRNAs in the A57 genome. repeat: sRNA sequences located exclusively in repeat regions; gene: sequences located within gene regions; gene&repeat: sequences located on repetitive elements within genes (e.g., introns); others: sequences mapped to miRNAs or other unannotated regions.

(b) Regulation of high‐abundance rdsRNAs by SlDRB1. The regulatory classification of rdsRNAs was defined by the fold change (FC) in their expression levels between the slydrb1 and wild‐type (WT) libraries. Negative: All values ≥2.0 or a mix of 0.50–2.00 (excluding ≤0.5); Positive: All values ≤0.5 or a mix of 0.50–2.00 (no value ≥2.0); Insignificant: all values between 0.50 and 2.00; Opposite: A mixture of both ≥2.0 and ≤0.5 values.

(c) Comparison of the size distribution of rdsRNA2809 across different libraries.

(d) Profile of 3′‐truncated sRNA isoforms from rdsRNA2809 in slydrb1 and WT shoot apices. The y‐axis represents the number of tailed bases, and the x‐axis represents the length of small RNAs.

(e) Analysis of truncation and tailing frequencies of different rdsRNAs using 3′ RACE. The y‐axis represents the percentage of reads carrying 3′‐terminal mismatches (including both truncation and tailing events) relative to the total number of reads perfectly matching the target sequence. The x‐axis indicates the specific rdsRNAs analyzed: rdsRNA2511, rdsRNA2809, and rdsRNA3473.

In order to obtain a comprehensive understanding of these rdsRNAs, high‐abundance sRNAs of all lengths were aligned to the rRNA of the A57 genome, and 11 448 unique reads were retrieved. We found that 66 and 33% of the rdsRNAs were derived from 28S and 18S RNA, respectively (Figure S3A). A length analysis of these sRNAs indicated that the percentage of sRNAs declined as length increased (Figure S3B). Next, the impact of the slydrb1 mutation on the biogenesis of these rdsRNAs was tested by comparing the ratio of their abundance in the mutant relative to wild‐type plants in the four pairs of samples (Methods section). The results showed that 68% of the rdsRNAs were positively regulated by SlDRB1 while only 2% of them were negatively regulated by SlDRB1 and that rdsRNAs derived from different types of rRNA showed similar trends (Figure 3b; Figure S3C–E; Data S3). The rdsRNAs negatively regulated by SlDRB1 were mainly expressed in leaf tissue and their distribution pattern on rRNA largely overlapped with the rdsRNAs that were positively regulated by SlDRB1 (Figure S3F). We further looked into a cluster of rdsRNAs, designated rdsRNA2809, which were generated from the same locus of 28S rRNA with the same 5′ end and 3′ nucleotides that mapped to successive positions (Figure 3c; Data S4). The majority of rdsRNAs in this cluster were down‐regulated in the slydrb1 mutant, which was evident in the petiole and shoot tissue (Figure 3c). When the total sRNA database was queried with the shortest rdsRNA in this cluster, many low abundance sRNAs were retrieved that contained 1–7‐nt 3′ nucleotides that did not match the rRNA sequences (Data S4). To further validate this phenomenon, we performed a 3′ RACE assay targeting 28S rRNA to analyze the 3′ termini of small RNAs in shoot tissue. The results showed that, compared to the wild type, the abundance of small RNAs carrying 3′ end truncations and tails was higher in the slydrb1 mutant (Figure 3e). These results demonstrate that the 3′‐end modifications of the rdsRNAs, including both truncation and tailing, were significantly enhanced in the slydrb1 mutant, consistent with previous reports that 3′ tailing often leads to small RNA degradation (Song, Wang, et al., 2019; Zhai et al., 2013). When the tailed and fully matched rdsRNAs were plotted on 2‐D plots, the proportion of the tailed rdsRNAs was higher in the slydrb1 mutant than in the wild‐type sample (Figure 3d). These results provide evidence that SlDRB1 may protect the majority of rdsRNAs from tailing and degradation.

SlDRB1 may have opposite effects on different miRNAs

To investigate the impact of the slydrb1 mutation on the accumulation of miRNAs, miRNAs were annotated and their precursors were predicted (see Methods). We identified 120 known and new miRNAs and quantified their expression levels in the four pairs of wild‐type and mutant samples (Data S5). For each miRNA, the ratio of its TPM in slydrb1 to wild type was calculated for each pair of samples. The analysis showed that the expression of 80 out of the 120 miRNAs was reduced in the mutant, 14 were increased, 23 were differentially affected in different tissues, and the remaining 2 were not affected (Figure S4A). This result was in line with the finding that HYL1 is required for the biogenesis of most miRNAs in Arabidopsis. Further analysis of the leaf transcriptome data from the slydrb1 mutant revealed that 2025 mRNAs were up‐regulated, 1077 mRNAs were down‐regulated, and the abundance of 30 589 mRNAs was not significantly different (Figure S4E).

We further investigated miR162, miR168, and miR6026, which target DCL1, AGO1, and DCL2 mRNA, respectively. In the leaf samples, miR162 and miR168 were down‐regulated in the slydrb1 mutant while miR6026 was up‐regulated (Figure 4a). These data were independently validated using Northern blots (Figure 4d). Similar trends were observed for miR162 and miR168 in the three other pairs of samples and for miR6026 in shoot samples. By contrast, the difference for miR6026 in petiole and stem samples was not pronounced (Figure S4B). Data from mRNA‐seq experiments revealed that mRNA levels of DCL1 targeted by miR162 and DCL2b targeted by miR6026 were increased, whereas that of AGO1 targeted by miR168 did not change significantly (Figure 4b). These results were consistent with those from quantitative real‐time PCR analysis (Figure 4c). Previous studies showed that the expression of miR162 and miR168 was dependent on DCL1 and HYL1 (Vaucheret et al., 2004; Xie et al., 2003). Thus, it was not surprising to see their reduction and an increase in target levels in the slydrb1 mutant. It was reported that miR6026 is produced by DCL2 (mainly DCL2a and DCL2b) in tomato, and that the autoregulation of DCL2 expression occurs through phasiRNA production (Wang, Deng, et al., 2018; Wang, Hardcastle, et al., 2018). Small RNA mapping analysis showed that phasiRNAs triggered by miR6026 from DCL2b (A57g11g03502) were also markedly increased in four tissues (Figure 4e,f; Figure S4C,D). Furthermore, the impact of the slydrb1 mutation on the precision of miRNA precursor processing was less pronounced than its effect on miRNA expression levels (Figure 4g,h). These results indicate that the slydrb1 mutation may indirectly enhance DCL2b expression to increase miR6026 accumulation and that for the DRB1‐dependent miRNAs, the mutation has a more pronounced impact on accumulation than on the accuracy of mature miRNA sequences.

Figure 4.

Analysis of miRNAs and their targets.

(a) Relative abundance of miR162‐3p, miR168‐5p, and miR6026‐3p in leaves from slydrb1 and wild‐type (WT) plants.

(b) Relative mRNA levels of SlDCL1, SlAGO1, and SlDCL2b in leaves of the slydrb1 mutant and WT plants. Statistical analysis was performed using a moderated t‐test, with SlDCL1 (***P < 0.001), SlAGO1 (ns, P = 0.14), and SlDCL2b (***P < 0.001). Data are presented as mean values ± SD (n = 3).

(c) Relative expression levels of SlDCL1 (A57g10g02780), SlAGO1 (A57g06g20430), and SlDCL2b (A57g11g03502) in leaf tissues. Gene expression was detected using real‐time quantitative PCR (qPCR) (Table S1) and calculated using the 2^−ΔΔCt method (normalized to the control group as 1). Data are presented as mean values ± SD (n = 3). Statistical analysis was performed using Student's t‐test to compare the treatment group with the control group. The P‐values were 0.0011 for SlDCL1 (**P < 0.01), 0.4951 for SlAGO1 (P > 0.05, ns), and 0.0035 for SlDCL2b (**P < 0.01).

(d) Northern blot analysis of miR162‐3p, miR168‐5p, miR6026‐3p, miR6022, and miR159 small RNAs in leaves from WT and the slydrb1 mutant. The numbers above each lane indicate relative gray values normalized to U6. Each lane represents an independent biological replicate, with two replicates for WT and two for the slydrb1 mutant.

(e) Distribution of sRNAs on the SlDCL2b genomic locus in slydrb1 and WT leaves. The density of sRNAs derived from the positive and negative strands is indicated by blue and red tracks, respectively.

(f) Size distribution of sRNAs mapped to the SlDCL2b locus in slydrb1 and WT leaves.

(g) Analysis of miRNA processing efficiency. The bar chart illustrates the processing efficiency for each indicated miRNA in leaf tissues. Processing efficiency is defined as the ratio of the sum of reads corresponding to the mature miRNA and its associated miRNA‐5p/3p to the total reads of the precursor miRNA (pre‐miRNA).

(h) Analysis of miRNA expression levels in leaf tissues. The bar chart displays the expression profiles of each indicated miRNA in the leaf tissues of WT and mutant plants.

SlDRB1 interacts with SlDCL1 through both dsRBD and PAZ domains

A previous study showed that Arabidopsis HYL1 interacts with DCL1. To investigate these interactions in more detail, different DNA fragments encoding different domains of SlDCL1 were cloned into an AD vector and a sequence encoding SlDRB1 was cloned into a BD vector for yeast two‐hybrid (Y2H) assays (Figure 5a,b). The results showed that all yeast clones grew well on SD/‐Leu/‐Trp medium, whereas on selective medium, only yeast strains harboring positive control plasmids or harboring BD‐SlDRB1 and AD‐SlDCL1‐dsRBD12 grew well. However, very small colonies were visible for yeast strains harboring BD‐SlDRB1 and AD‐SlDCL1‐PAZ (Figure 5c), indicating that SlDRB1 interacts strongly with SlDCL1‐dsRBD12 and weakly with SlDCL1‐PAZ. To further test these interactions in vivo, their CDSs were fused with sequences encoding the N‐ and C‐terminal halves of luciferase, respectively (Figure 5d; Figure S5A), and luciferase complementation assays (LCA) were conducted. The results showed that clear luminescence signals were observed in patches co‐expressing SlDRB1‐nLUC and cLUC‐SlDCL1‐PAZ, as well as in those co‐expressing SlDRB1‐nLUC and cLUC‐SlDCL1‐dsRBD12 (Figure 5e; Figure S5B), indicating that SlDRB1 interacts with the PAZ and dsRBD12 domains of SlDCL1 in vivo.

Figure 5.

Interaction between intact SlDRB1 and different domains of SlDCL1.

(a) Schematic diagram of the domain organization in SlDCL1 and SlDRB1. SlDCL1, 1914 amino acids (aa) in length, is divided into seven segments according to its major domains. SlDRB1, with a full length of 349 aa, is divided into three segments based on its primary domains.

(b) Schematic diagram of the yeast two‐hybrid (Y2H) vector construction. All expression was driven by the ADH1 promoter. DNA fragments encoding the six domains of SlDCL1 were individually cloned into the activation domain (AD) vector. The full‐length coding sequence (CDS) of SlDRB1 was cloned into the binding domain (BD) vector.

(c) Y2H results. The top panel shows the growth of yeast on a quadruple dropout medium (SD‐Leu‐Trp‐His‐Ade). The bottom panel shows the growth of yeast on a double dropout medium (SD‐Leu‐Trp).

(d) Luciferase complementation assay (LCA) vector construction. All gene expression was driven by the CaMV 35S promoter. SlDRB1 was fused to the DNA fragments encoding the N‐terminus of luciferase (nLUC). DNA fragments encoding the PAZ domain and the dsRBD12 domain of SlDCL1 were fused to DNA fragments encoding the C‐terminus of luciferase (cLUC).

(e) LCA validating interactions between SlDRB1 and SlDCL1‐PAZ and between SlDRB1 and SlDCL1‐dsRBD12. The injection site at the lower right corner of the abaxial surface of the leaf represents the experimental group, while the other three injection sites serve as negative controls.

dsRBD2 domains promote interactions between SlDRB1 and SlDCL1

To further elucidate the interactions between SlDRB1 and the SlDCL1‐dsRBD12 domain, the dsRBD1, dsRBD2, and C‐terminal domain (CTD) of SlDRB1 and dsRBD1 and dsRBD2 from SlDCL1 were cloned into AD and BD vectors, respectively, for Y2H assays (Figure 6a,b). The results showed that while all strains grew well on the SD‐L‐T medium, on the SD‐L‐T‐H‐A medium, the yeast strains that grew similarly to the positive control were those containing AD‐SlDRB1‐dsRBD2 and BD‐SlDRB1‐dsRBD2, AD‐SlDCL1‐dsRBD2 and BD‐SlDRB1‐dsRBD2, AD‐SlDRB1‐dsRBD2 and BD‐SlDCL1‐dsRBD2, and AD‐SlDCL1‐dsRBD2 and BD‐SlDCL1‐dsRBD2 (Figure 6c). These data indicate that the dsRBD2 domains promote interactions between SlDRB1 and SlDCL1. The yeast strains harboring AD‐SlDRB1‐CTD and BD‐SlDRB1‐dsRBD1, AD‐SlDRB1‐CTD and BD‐SlDRB1‐CTD, and AD‐SlDRB1‐CTD and BD‐SlDCL1‐dsRBD1 also grew as dense colonies, albeit less dense than the positive control (Figure 6c), indicating that SlDRB1‐CTD was capable of interacting with SlDRB1‐dsRBD1 and that SlDRB1‐CTD was capable of interacting with SlDCL1‐dsRBD1. A few small colonies of yeast strains harboring AD‐SlDRB1‐dsRBD2 and BD‐SlDRB1‐dsRBD1 grew on the selective medium, indicating weak interactions between SlDRB1‐dsRBD2 and SlDRB1‐dsRBD1 in yeast cells.

Figure 6.

Fine mapping interactions among SlDCL1‐dsRBD and SlDRB1 subdomains.

(a) Diagram of the PAZ and dsRBD1/2 domains of the SlDCL1 protein, along with the structural domains of the SlDRB1 protein.

(b) Schematic representation of yeast two‐hybrid (Y2H) vector construction. The expression of all cassettes was driven by the ADH1 promoter, with the dsRBD1/2 domains of SlDCL1 and SlDRB1 cloned into activation domain (AD) and binding domain (BD) vectors, respectively.

(c) Validation of domain interactions via the Y2H assay. The left panel shows growth on the quadruple dropout medium (SD‐Leu‐Trp‐His‐Ade) and the right panel on double dropout medium (SD‐Leu‐Trp). Empty vector controls occupy the first row and the first column; experimental groups are indicated by red rectangles; the bottom row includes negative controls (CK−) and positive controls (CK+).

(d) Luciferase complementation assay (LCA) confirming protein–protein interactions. The lower right leaf position indicates the experimental group. The other three positions serve as negative controls. The seven different co‐injection combinations include: SlDRB1‐dsRBD2‐nLUC with cLUC‐SlDRB1‐dsRBD1 and cLUC‐SlDRB1‐dsRBD2; SlDRB1‐CTD‐nLUC with cLUC‐SlDRB1‐dsRBD1; cLUC‐SlDRB1‐CTD and cLUC‐SlDCL1‐dsRBD1 co‐injections; SlDRB1‐dsRBD2‐nLUC with cLUC‐SlDCL1‐dsRBD2; and SlDCL1‐dsRBD2‐nLUC with cLUC‐SlDCL1‐dsRBD2.

(e) Co‐IP assay validating protein interactions. HA‐tagged proteins are immunoprecipitated using HA magnetic beads; detection employs anti‐Flag and anti‐HA antibodies. Seven experimental groups feature Input (left) and IP‐HA (right) samples; ‘+’ indicates plasmid expression in Nicotiana benthamiana leaves; molecular weights are provided in kilodaltons (kDa), with specific protein positions marked; asterisks denote target bands.

(f) Bimolecular fluorescence complementation (BiFC) assay with H2B‐mCherry as a nuclear localization marker. The left label indicates the co‐injected protein pairs in N. benthamiana. Scale bar = 20 μm.

(g) Interaction network chart of the dsRBD1/2 domains from SlDCL1 and three structural domains from SlDRB1. Red circles signify interactions confirmed by all four methods (Y2H, LCA, BiFC, Co‐IP); gray circles indicate interactions confirmed using only LCA.

To validate the above interactions in plants, LCA, Co‐IP, and BiFC assays were conducted in Nicotiana benthamiana. The results showed that co‐expression of all tested protein domain pairs, including SlDRB1‐dsRBD2 and SlDRB1‐dsRBD1, SlDRB1‐dsRBD2 and SlDRB1‐dsRBD2, SlDRB1‐CTD and SlDRB1‐dsRBD1, SlDRB1‐CTD and SlDRB1‐CTD, SlDRB1‐CTD and SlDCL1‐dsRBD1, SlDRB1‐dsRBD2 and SlDCL1‐dsRBD2, and SlDCL1‐dsRBD2 and SlDCL1‐dsRBD2, produced unambiguous luminescence in LCA and fluorescence in bimolecular fluorescence complementation (BiFC) (Figure 6d,f), and unequivocal co‐immunoprecipitation in Co‐IP assays (Figure 6e), while their negative controls did not (Figure 6d,e; Figure S6).

In summary, with four different assays, we found that SlDRB1‐dsRBD2, SlDRB1‐CTD, and SlDCL1‐dsRBD2 can mediate interactions between SlDRB1 and SlDCL1. Moreover, SlDRB1‐dsRBD2 and SlDCL1‐dsRBD2 mediate interactions between SlDRB1 and SlDCL1. Additionally, SlDRB1‐CTD interacts with SlDCL1‐dsRBD1 and with SlDRB1‐dsRBD1 (Figure 6g).

SlDCL1‐PAZ differentially interacts with SlDRB1 subdomains

In an early study, we showed that SlDCL1‐PAZ interacts weakly with SlDRB1 in the Y2H assay but strongly in LCAs (Figure 5). To reconcile the conflicting data obtained from the two assays, we hypothesized that intramolecular interactions involving the SlDRB1 subdomains may inhibit its interaction with SlDCL1‐PAZ. To test this hypothesis, the interactions between SlDCL1‐PAZ and SlDRB1 subdomains were examined in four different assays.

The Y2H assay results showed that although the yeast strain containing AD‐SlDCL1‐PAZ and BD‐SlDRB1 produced very few small colonies on the selective medium (Figure 5c), strains harboring AD‐SlDCL1‐PAZ and BD‐SlDRB1‐dsRBD1 and harboring AD‐SlDCL1‐PAZ and BD‐SlDRB1‐CTD produced dense clones, similar to the positive control, and that strains harboring AD‐SlDCL1‐PAZ and BD‐SlDRB1‐dsRBD2 produced less dense colonies on the selective medium (Figure 7a, left). However, when the interacting domains were switched between the AD and BD vectors, the corresponding yeast strains no longer grew on the selective medium (Figure 7a, right). These results provide evidence that SlDCL1‐PAZ interacts with dsRBD1, dsRBD2, and the CTD of SlDRB1 in yeast but that these interactions may be sensitive to inhibition by adjacent fusion domains.

Figure 7.

Fine mapping interactions among SlDCL1‐PAZ and SlDRB1 subdomains.

(a) Y2H assay validating interactions between the domain SlDCL1‐PAZ and SlDRB1. The left panel shows the growth of yeast co‐transformed with AD‐SlDCL1‐PAZ and BD‐SlDRB1‐dsRBD1/dsRBD2/CTD on quadruple dropout (SD‐Leu‐Trp‐His‐Ade) and double dropout (SD‐Leu‐Trp) media. The right panel shows the growth of yeast co‐transformed with BD‐SlDCL1‐PAZ and AD‐SlDRB1‐dsRBD1/dsRBD2/CTD on the corresponding media.

(b) Luciferase complementation assay (LCA) for validating protein–protein interactions. The injection site at the lower right corner on the leaf surface represents the experimental group, while the other three sites serve as negative controls. The left panel shows results from co‐expression of SlDCL1‐PAZ‐nLUC with cLUC‐SlDRB1‐dsRBD1/dsRBD2/CTD. The right panel shows results from co‐expression of SlDRB1‐dsRBD1/dsRBD2/CTD‐nLUC with cLUC‐SlDCL1‐PAZ.

(c) Bimolecular fluorescence complementation (BiFC) assay detecting the interactions between the SlDCL1‐PAZ domain and SlDRB1. H2B‐mCherry indicates nuclear localization. The top three rows represent experimental groups. The remaining rows are controls. Scale bar = 20 μm.

(d) Co‐IP assay validating the interaction between the SlDCL1‐PAZ domain and SlDRB1. SlDRB1‐dsRBD1/dsRBD2/CTD‐HA fusion proteins were immunoprecipitated using HA nanomagnetic beads. SlDCL1‐PAZ‐Flag and SlDRB1 domains‐HA fusion proteins, along with SlMAP65‐HA, were detected using anti‐Flag and anti‐HA antibodies, respectively. Three experimental groups were set up. The left lane in each group represents the Input sample, and the right lane represents the IP‐HA sample. ‘+’ indicates the corresponding vector expressed in Nicotiana benthamiana. Protein molecular weights are marked in kDa, with numbers indicating the positions of target proteins and asterisks (*) indicating specific target bands.

These interactions were also tested in plants using LCA, Co‐IP, and BiFC assays. The LCA and BiFC assays produced unambiguous luminescence (for LCA) and fluorescence (for BiFC) signals when SlDCL1‐PAZ was co‐expressed with each subdomain of SlDRB1 (Figure 7b,c). However, in the Co‐IP assay, when SlDRB1‐dsRBD2 and SlDCL1‐PAZ were co‐expressed, the SlDCL1‐PAZ protein accumulated to low levels and no co‐immunoprecipitation was detected. By contrast, SlDCL1‐PAZ was detected in immunoprecipitated fractions when co‐expressed with SlDRB1‐dsRBD1 and SlDRB1‐CTD (Figure 7d). These results indicate that SlDCL1‐PAZ interacts with the dsRBD1 and CTD of SlDRB1 in vivo, but its interaction with SlDRB1‐dsRBD2 is weak.

DISCUSSION

We generated a high‐quality genome assembly for a newly sequenced tomato cultivar S. lycopersicum A57 using PacBio HiFi long reads combined with Hi‐C scaffolding. Although the proportion of sequences anchored to pseudo‐chromosomes was comparable to previously reported tomato assemblies, our final genome exhibited markedly improved structural continuity and gene annotation completeness. The assembly achieved a contig N50 of 36.04 Mb and a scaffold N50 of 67.62 Mb, representing a substantial enhancement in genome contiguity. In total, 33 689 protein‐coding genes were predicted, with more complete gene models, fewer fragmented annotations, and a longer average CDS length compared with the current reference genome. BUSCO evaluation of the annotated protein set revealed 96.4% completeness, reflecting the improved recovery of conserved single‐copy orthologs (Figure 1; Table 1). The higher contiguity also reduced gene redundancy and enabled more accurate reconstruction of exon–intron structures, ultimately yielding more reliable functional annotations. Together, these improvements demonstrate that, despite similar chromosome anchoring rates, our assembly provides a more comprehensive and biologically informative representation of the tomato genome. This improved genome assembly and the accompanying user‐friendly webserver (Figure S1) will serve as a valuable genomic resource for downstream applications, including transcriptional regulation studies, comparative genomics, evolutionary analyses, and molecular breeding.

The conserved dsRNA‐binding protein HYL1 has been well characterized for its role in miRNA biogenesis in the model plant Arabidopsis. However, its function remains poorly understood in other plant species. In this study, mutations were introduced into the tomato ortholog of HYL1 (i.e., SlDRB1) using CRISPR/Cas9 technology, and the resulting mutants were characterized at the phenotypic and molecular levels. These mutations led to pleiotropic phenotypes, including both vegetative and reproductive abnormalities (Figure 2; Figure S2). These phenotypes are much more severe than those of the hyl1 mutant in Arabidopsis, which exhibits dwarfism, abnormal leaf morphology, and significantly reduced seed set (Lu & Fedoroff, 2000). One plausible explanation is that most characterized Arabidopsis hyl1 alleles contain partial or weak loss‐of‐function mutations, whereas the CRISPR‐generated slydrb1 mutants obtained in this study are null alleles. Alternatively, the severe phenotypes may reflect additional, previously uncharacterized functions of SlDRB1 revealed in our work. We found that a substantial portion of TE siRNAs depends on SlDRB1 (Figure 3a), suggesting that SlDRB1 is involved in TE silencing and plant genome protection. The Arabidopsis genome contains far fewer TEs than the tomato genome (Lamesch et al., 2012); therefore, hyl1 mutation in Arabidopsis is less detrimental than slydrb1 mutation in tomato. It is also interesting to note that slydrb1 mutation increased the level of miR6026 (Figure 4a), which is produced by DCL2 (Wang, Hardcastle, et al., 2018). This result suggests potential competition for substrates between DCL1‐DRB1 and the DCL2 pathway, the mechanisms of which merit further investigation.

Comparisons of the small RNA profiles between the mutants and wild‐type plants revealed that a large number of high‐abundance small RNAs were derived from ribosomal RNAs, most of which accumulated to lower levels in the slydrb1 mutant in young shoots. Further analysis showed that the proportion of tailed rdsRNAs was increased in the mutant shoots, indicating that SlDRB1 protects rdsRNAs from degradation (Figure 3), a finding not previously reported in other plant species. Most of these rdsRNAs form clusters with continuous lengths ranging from 18 to 30 nt, and their secondary structures do not fold into a typical miRNA structure (Figure S3G). These results indicate that the pathway for their biogenesis is different from the DCL1‐dependent miRNA pathway and is worth further investigation.

Our finding that the SlDRB1 mutant exhibits reduced accumulation of most canonical miRNAs (Figure 4; Figure S4) supports that SlDRB1 functions as a key factor in miRNA biogenesis, analogous to HYL1 in Arabidopsis (Gonzalo et al., 2024). Interestingly, we found that the SlDRB1 mutation did not reduce the precision of the mature sequences of the SlDRB1‐dependent miRNAs (Figure 4g), which is different from that observed for the Arabidopsis hyl1 mutant as reported in vitro or in heterologous dicing systems (Dong et al., 2008; Zhang & Li, 2024). This discrepancy suggests that the precision of mature miRNA sequences is determined not only by Dicing but also by a downstream process. To further investigate the mechanism of SlDRB1 in miRNA biogenesis, a comprehensive SlDRB1‐SlDCL1 interaction study was conducted with multiple methods. The results showed that SlDCL1 and SlDRB1 interact with each other via their dsRBD2 domains (Figures 5 and 6), which is consistent with the findings from similar experiments conducted with their Arabidopsis homologs (Liu et al., 2013; Yang et al., 2014). In addition, our results also revealed that SlDCL1 and SlDRB1 interact with each other via SlDCL1‐PAZ and SlDRB1‐dsRBD1/CTD (Figure 7). It was reported that HYL1 and SE accelerate the rate of DCL1‐mediated cleavage of pri‐ and pre‐miR167b substrate and promote accurate processing (Dong et al., 2008). For both cleavage reactions, the PAZ domain of DCL1 interacts with the pri‐ and pre‐miRNA substrates and sets the starting point for the length determination of miRNAs (Wei et al., 2021). Our recent study showed that a mutation in the DCL1‐PAZ domain reduces the precision and efficiency of the second cleavage of particular miRNAs in Arabidopsis (Zhang & Li, 2024). These studies pointed out the importance of the PAZ domain in determining the accuracy and efficiency of miRNA processing. Interactions between SlDCL1‐PAZ and the SlDRB1‐dsRBD and the CTD may facilitate interactions between the PAZ domain and pri‐/pre‐miRNA substrates, thus contributing to their accurate processing.

MATERIALS AND METHODS

Plant materials, bacterial strains, and plasmids

Wild‐type N. benthamiana HZ1 was used for transient assays and has been described previously (Wang, Zhang, et al., 2024). Wild‐type S. lycopersicum Ailsa Craig 57 (A57) was used as the recipient material for transgenic experiments and has been described previously (Ma et al., 2023). For genome sequencing, fresh leaves were harvested from 2‐week‐old A57 plants grown in a growth room under a 14 h/10 h light/dark cycle and were immediately frozen in liquid nitrogen. The frozen samples were sent for third‐generation sequencing using the PacBio Sequel II platform (BeryGene, Beijing, China). A 15‐kb SMRT PacBio High‐Fidelity (HiFi) library was prepared for the A57 genome. One SMRT cell was used for library sequencing. Fresh leaves from 2‐week‐old plants were sampled and frozen immediately in liquid nitrogen for Hi‐C library construction. The library was then sequenced in a 150‐bp paired‐end mode on the Illumina NovaSeq 6000 platform (BeryGene, Beijing, China).

Escherichia coli strain DH5α (ToloBio, Wuxi, China) and Agrobacterium tumefaciens strain GV3101 (maintained in our laboratory) were used for cloning and transient expression, respectively. Yeast strain Y2HGold (Weidi, Shanghai, China) was employed for Y2H assays.

For SlDRB1 (corresponding to A57g04g20140 in the A57 genome and Solyc04g076420 in the SL4.0 genome) knockout, three CRISPR‐Cas9 target sites were selected within or adjacent to the second exon (Figure 2a) using the CRISPR‐P2.0 online tool (Liu et al., 2017). The gRNA cassette gRNA1‐tRNA‐gRNA2‐tRNA‐gRNA3 was synthesized using templates as previously described (Wang, Li, et al., 2024) and cloned into the pTX041 vector after digestion with Bsa I (NEB, R3733, Ipswich, MA, USA) to construct the recombinant plasmid pTX041‐SlDRB1. For Y2H assays, the CDS of the test protein or its domain was cloned into the pGADT7 and pGBKT7 vectors (Clontech/Takara Bio, San Jose, CA, USA). For LCAs, the CDS of the test protein (domain) was cloned into pCAMBIA1300‐nLUC and pCAMBIA1300‐cLUC (a kind gift from Prof. Bie Zhilong's laboratory, Huazhong Agricultural University). For BiFC assays, the CDS of the test protein (domain) was cloned into pH7LIC18 and pH7LIC20 (constructed in this study; Data S6). For co‐immunoprecipitation assays, the CDS of the test protein (domain) was cloned into pH7LIC4.1 and pH7LIC8.1 (constructed in this study; Data S6). All primers used in the above cloning experiments are listed in Table S1.

Bioinformatics analysis

Genome assembly

For genome contig assembly, the raw PacBio HiFi CCS data were assembled into 2281 A57 contigs using hifiasm (v0.14) (Cheng et al., 2021) (Table 1). To assemble scaffolds from contigs, raw Hi‐C data were first filtered using Trim Galore (v0.6.6, MINLEN:120) to yield clean data. The clean Hi‐C data were aligned to A57 contigs using Juicer (‐‐assembly ‐s Mbol) (Durand, Shamim, et al., 2016). Guided by the resulting alignment data, A57 contigs were assembled into scaffolds using the 3d‐dna software (‐I 8000 ‐r 2) (Dudchenko et al., 2017). Final genome scaffolds were obtained after correcting misassembled scaffolds using Juicebox (v1.11.08) (Durand, Robinson, et al., 2016).

Genome and functional annotation

For genome annotation, we integrated three complementary strategies: transcriptome‐guided prediction, de novo gene prediction, and homology‐based gene alignment. For transcriptome‐guided prediction, we collected 202 publicly available tomato RNA‐seq datasets from NCBI (accessions: PRJNA590146, PRJNA807664, PRJNA887075, PRJNA1030391, PRJNA1089146) and 12 in‐house tomato RNA‐seq datasets. All raw reads were processed using Trim Galore (v0.6.6). The resulting clean data were aligned to the A57 genome using HISAT2 (v2.2.1) (Kim et al., 2015). The alignment data were then processed using StringTie (v1.3.3b) (Pertea et al., 2015) to generate mRNA sequences. Finally, the mRNA sequences assembled from the RNA‐seq datasets were integrated to annotate coding gene structures using PASA (Program to Assemble Spliced Alignments v2.5.2) (Haas et al., 2003).

For de novo assemblies using AUGUSTUS (v3.4.0), a set of 4150 well‐annotated genes generated from a PASA analysis was selected as the gene model training dataset (with at least two exons, lengths of CDSs greater than 900 bp, ratios of CDS length over total exon length greater than 0.7, and no overlapping region between genes). The annotation file of these genes was processed using AUGUSTUS to generate species‐specific gene model training results and de novo gene predictions (Stanke et al., 2006). For annotation‐independent de novo annotation, GeneMark (v4.68_lic) was used (Lukashin & Borodovsky, 1998).

For homology‐dependent annotation, protein sequences from tomato (ITAG2.4) and plant Uniprot protein sequences were aligned to the A57 genome using blastp (Camacho et al., 2009). The alignment results were processed using GeneWise to generate homology‐dependent annotation.

The above four types of annotation results from the A57 genome were finally integrated into the A57 gene annotation file using EVM (EvidenceModeler v1.1.1) with different weight settings for different annotation results: 10 for PASA results, 4 for AUGUSTUS results, 3 for GeneWise results, and 1 for GeneMark results (Haas et al., 2008).

Tomato transformation and characterization

For SlDRB1 mutant construction, A. tumefaciens GV3101 harboring the pTX041‐SlDRB1 vector was used to stably transform A57 tomato lines as previously described (Ouyang et al., 2005). Tomato seeds were surface‐sterilized with 75% ethanol for 1 min, 50% sodium hypochlorite for 15 min, rinsed thoroughly with sterile water, and sown on 1/2 MS medium (containing 15 g L⁻¹ sucrose and 7.5 g L⁻¹ agar). The seeds were cultivated at 25°C in a tissue culture room under a 16 h light/8 h dark photoperiod. Cotyledons from 7‐ to 10‐day‐old seedlings were excised and pre‐cultured on solid pre‐inoculation medium (SPM; containing MS medium, 0.1 mg L⁻¹ myo‐inositol, 1.3 mg L⁻¹ thiamine‐HCl, 0.2 mg L⁻¹ 2,4‐D, 0.2 mg L⁻¹ KH₂PO₄, 0.1 mg L⁻¹ KT, 30 g L⁻¹ sucrose, and 7.5 g L⁻¹ agar, pH 5.8) for 1 day.

The A. tumefaciens strain GV3101 harboring the pTX041‐SlDRB1 vector was cultured overnight in LB medium supplemented with appropriate antibiotics, harvested by centrifugation, and resuspended in liquid pre‐inoculation medium (LPM; consisting of half‐strength MS medium, 50 mg L⁻¹ myo‐inositol, 0.2 mg L⁻¹ thiamine‐HCl, and 10 g L⁻¹ sucrose, pH 5.8) to an OD₆₀₀ of 0.3. Pre‐cultured cotyledon explants were immersed in the bacterial suspension for 3–5 min, blotted dry on sterile filter paper, and co‐cultivated on SPM for 2 days at 25°C in the dark.

Following co‐cultivation, explants were transferred to callus‐inducing medium (CIM; containing MS medium, 30 g L⁻¹ sucrose, 7.5 g L⁻¹ agar (pH 5.8), supplemented with 0.1 mg L⁻¹ IAA, 2.0 mg L⁻¹ zeatin, 100 mg L⁻¹ kanamycin, and 360 mg L⁻¹ timentin) for 2 weeks, and then were moved to shoot induction medium (SIM; containing MS medium, 30 g L⁻¹ sucrose, 7.5 g L⁻¹ agar, pH 5.8, supplemented with 0.2 mg L⁻¹ zeatin, 100 mg L⁻¹ kanamycin, and 300 mg L⁻¹ timentin). Adventitious shoots (approximately 1 cm) were excised and transferred to rooting medium (RM; containing MS medium, 30 g L⁻¹ sucrose, 7.5 g L⁻¹ agar, pH 5.8, supplemented with 2 mg L⁻¹ IBA, 50 mg L⁻¹ kanamycin, and 360 mg L⁻¹ timentin). Well‐rooted plantlets were transplanted to soil and grown under standard greenhouse conditions.

Primary transformants were screened by PCR using Cas9‐specific primers (Table S1). To characterize SlDRB1 editing, SlDRB1 fragments were amplified from positive transformants using primers flanking the target sites (Table S1) and subjected to Sanger sequencing. Sequence alignments were analyzed using SnapGene and NCBI ORFfinder to identify frameshift mutations and premature termination codons; only lines with confirmed loss‐of‐function mutations were retained for subsequent analyses.

Small RNA sequencing

Small RNA libraries were constructed using the VAHTS Small RNA Library Prep Kit for Illumina (Vazyme, Wuhan, China) according to the manufacturer's protocol and sequenced using the Illumina platform (Novogene, Beijing, China). Small RNA reads were processed and analyzed as previously described (Li et al., 2012). The total reads with lengths ranging from 18 to 30 nt were used for TPM calculations. Only the small RNAs with an expression level greater than 1 TPM in at least one library were retained for subsequent analysis. The remaining reads were mapped to the A57 genome, retaining only perfectly matched reads. Ribosomal RNAs (rRNAs) in the A57 genome were predicted using Barrnap (v0.9, https://github.com/tseemann/barrnap), and microRNAs (miRNAs) were identified against miRBase (Release 22, http://www.mirbase.org/).

mRNA sequencing

Raw sequencing data in FASTQ format were initially assessed for quality using FastQC (v0.11.9, Babraham Institute). Adapter sequences and low‐quality reads were trimmed using Trimmomatic (v0.39). The cleaned reads were then aligned to the A57 reference genome using HISAT2 (v2.2.1). SAMtools (v1.15.1) was employed to sort and convert the alignment files to the BAM format. Gene‐level read counts were quantified using featureCounts from the Subread package (v0.21.1). Differential expression analysis between the slydrb1 mutant and A57 wild‐type groups was performed using the limma‐voom pipeline (limma v3.62 and edgeR v4.4.2). Lowly expressed genes were filtered by retaining genes with CPM >1 in at least three samples. Library size normalization was conducted using the TMM method via calcNormFactors (edgeR v4.4.2). The voom function transformed count data into log2‐CPM values with precision weights, followed by linear modeling and empirical Bayes moderation to identify differentially expressed genes.

Small RNA Northern blot

Northern blot analysis was performed as previously described (Wang, Yu, et al., 2024). Briefly, 20–40 μg of total RNA were concentrated to complete dryness, then dissolved in 50% formamide (Sigma‐Aldrich, St. Louis, MO, USA) and 6× RNA loading dye, and denatured at 65°C for 10 min. Samples were separated by electrophoresis on a 15% denaturing polyacrylamide gel in 0.5× TBE buffer. RNA was then transferred onto a nylon membrane (GE Healthcare, Chicago, USA) and cross‐linked using EDC‐mediated chemical cross‐linking. The membrane was hybridized with a biotin‐labeled probe (Table S1) in PerfectHyb Plus Hybridization Buffer (Sigma‐Aldrich, USA) for 12 h. Signal detection was performed using a biotin nucleic acid detection kit (Beyotime, Shanghai, China).

Analysis of rdsRNA tailing by 3′ RACE

For 3′ RACE, some reagents from the VAHTS Small RNA Library Prep Kit (Vazyme, Wuhan, China) were used for amplification. Briefly, the 3′ adapter was ligated to the RNA samples, followed by reverse transcription using a primer complementary to the 3′ adapter (Table S1). Subsequently, PCR amplification was performed using small RNA‐specific primers paired with primers binding to the 3′ adapter (Table S1). The target PCR products were excised and purified from agarose gels, and then subjected to sequencing at Novogene Corporation. For data analysis, raw reads were first filtered to retain sequences of 18–50 nucleotides in length. These reads were then aligned to the target reference sequence using BLASTN, requiring a minimum of 18 nucleotides of perfect 100% sequence identity. Reads fully matching the reference across the aligned region were classified as unmodified. Reads with a perfect upstream match of ≥18 nt but carrying mismatches at the 3′ terminus were defined as putatively modified reads. The modification rate was calculated as the proportion of 3′‐terminus mismatched reads relative to all reads with a perfect match in the initial 18‐nt seed region.

Y2H assay

For the Y2H assay, the GAL4 system was employed. The Y2HGold yeast strain (Matchmaker, Takara Bio USA) was co‐transformed with pGADT7 and pGBKT7 constructs harboring the genes of interest for co‐expression. Co‐transformants were plated on SD minimal medium lacking Leu and Trp (SD/‐Leu/‐Trp) for selection. As controls, a yeast strain co‐transformed with pGBKT7‐53 and pGADT7‐T served as the positive control (CK+). Yeast co‐transformed with pGBKT7‐Lam and pGADT7‐T served as the negative control (CK−). To assess protein–protein interactions, the co‐transformed yeast strains were also plated on selective SD medium lacking Leu, Trp, His, and Ade (SD/‐Leu/‐Trp/‐His/‐Ade) and incubated at 30°C for 3 days, after which growth was observed. The entire experiment was performed with three independent biological replicates, each using freshly transformed yeast colonies.

LCA and BiFC

Agrobacterium strains were grown and induced as previously described (Zhang & Li, 2024). Before infiltration, agrobacteria were resuspended in infiltration buffer, diluted to working concentrations of OD₆₀₀ = 0.3 and OD₆₀₀ = 0.2 for LCA and BiFC, respectively, and mixed in equal volumes. Agro‐mixtures with different combinations of test protein pairs were infiltrated into N. benthamiana leaves. For LCA, 3 days after infiltration, luciferase substrate (0.1 μl 100 mM luciferin stock + 0.1 μl 10% Triton X‐100 + 9.8 ml ddH₂O) was rubbed onto the infiltrated patch, and luciferase activity was imaged after 5 min using the In Vivo Plant Imaging System NightShade LB 985 (Berthold, Bad Wildbad, Germany). The LCA experiment was performed with three independent biological replicates. For BiFC, infiltrated leaves were collected and imaged using a confocal laser scanning microscope TCS SP8 DLS (Leica, Wetzlar, Germany) 3 days after infiltration. Three independent biological replicates were conducted for the BiFC assay, and representative images are shown.

Co‐immunoprecipitation

For Co‐IP assays, proteins were transiently co‐expressed in N. benthamiana leaves by agroinfiltration (OD₆₀₀ = 0.5–1.0). Leaves were harvested 3 days post‐infiltration (dpi). Total proteins were extracted from approximately 1 g of ground leaf tissue using lysis buffer (50 mM Tris–HCl, pH 7.5, 150 mM NaCl, 5 mM EDTA, pH 8.0, 1% NP‐40, 2 mM NaF, 5 mM DTT, 1 mM PMSF). After centrifugation, the supernatant was incubated with pre‐washed magnetic beads (LABLEAD, Beijing, China) on a rotating mixer at 4°C for 4 h or overnight. The beads were then washed five times with 1× TBS buffer. Bound proteins were eluted by boiling in SDS‐PAGE sample buffer and analyzed using immunoblotting with specific antibodies. All Co‐IP experiments were independently repeated twice (two biological replicates) with similar results.

ACCESSION NUMBERS

Raw reads of PacBio, Hi‐C, and small RNA sequencing have been submitted to the BIG Data Center, Chinese Academy of Sciences (GSA, http://bigd.big.ac.cn/gsa), under accession numbers CRA032478 and CRA031868.

AUTHOR CONTRIBUTIONS

FL, HZ and JW designed the study. YW, HW and LZ performed the wet lab experiments. JW performed genome assembly and annotation. FK built the TomatoHub webserver. HW performed small RNA bioinformatics analysis. FL, JW, YW, and HW analyzed the data and wrote the manuscript.

CONFLICT OF INTEREST

The authors declare that they have no conflicts of interest.

Supporting information

Data S1. List of sRNA sequences and their read numbers in different libraries.

Data S2. List of 20–24‐nt sRNAs with high‐expression levels (19 605).

Data S3. List of high‐level sRNAs mapped to the A57 genome (showing annotation in the second to last column) with expression in different libraries.

Data S4. List of rdsRNA2809 clusters.

Data S5. List of miRNAs and their annotation and expression values.

Data S6. Genbank files of the vectors pH7LIC18, pH7LIC20, pH7LIC4.1, and pH7LIC8.1.

Data S7. High‐throughput sequencing data from three T0 generation mutants.

Table S1. List of primers and Northern blot probes.

Figure S1. Function modules on TomatoHub.

Figure S2. Characterization of slydrb1 mutants.

Figure S3. Bioinformatics analysis of rdsRNAs.

Figure S4. Overall miRNA expression profile and DCL2‐derived phasiRNA profiles.

Figure S5. Luciferase complementation assay to study interactions between SlDCL1 and SlDRB1.

Figure S6. Luciferase complementation assay.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are openly available in BIGD at http://bigd.big.ac.cn/gsa, reference number CRA032478 and CRA031868.