Global atlas of enhancer-promoter interactome in cotton genome revealed by profiling RNA-RNA spatial interactions

Abstract

Background

RNA-RNA spatial interactions play crucial roles in various cellular processes, including gene transcriptional and post-transcriptional regulations. However, research on this area of RNA regulation remains limited in plants.

Results

Here, we adapt the global RNA-RNA interaction mapping method for plants and develop plant RNA in situ conformation sequencing, pRIC-seq, to generate comprehensive RNA-RNA spatial interaction maps for diploid and tetraploid cotton. We also perform global nuclear run-on followed by cap-selection assay, GRO-cap, and integrate multi-omics data to construct enhancer landscapes in these cotton species. Focusing on enhancer-promoter (E-P) RNA interactions, we find that tetraploid cotton, following polyploidy, innovates numerous novel E-P RNA interactions, thereby increasing its genomic regulatory complexity. Comparative analyses between wild-type and mutant fuzzless/lintless in tetraploid cotton reveal that RNA-RNA interactions, including E-P RNA interactions, play pivotal roles in fiber development. Our study also identifies short tandem repeats and transposable elements as potential mediators of E-P RNA interactions through base pairing within the cotton genome. Finally, integrating with genome-wide association studies (GWAS) and eQTLs from previous studies, we observe that our RNA-RNA interactions are significantly enriched near those functional mutation sites. Importantly, by using RAP-qPCR, we confirm that GWAS related enhancers interact with the promoters of protein-coding genes, explaining their regulatory mechanisms in fiber trait control.

Conclusions

Our results provide the first genome-wide RNA-RNA interaction map in higher plants and offer valuable insights into the enhancer-regulated pathway and targets for future breeding studies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13059-025-03907-w.

Background

RNA-RNA interactions are pivotal in various cellular processes, including regulation, RNA structure and function, splicing, localization, and the formation of ribonucleoprotein complexes [1–4]. Understanding these interactions is crucial for elucidating regulatory networks and mechanisms within living organisms.

Recent experimental techniques advancements, such as RIC-seq (RNA in situ conformation sequencing) [5] developed in human HeLa cells, have been proven to not only identify inter- and intra-molecular RNA-RNA interactions to elucidate RNA structure on a genome-wide scale, but also to decipher RNA-RNA interactions to reveal enhancer-promoter (E-P) interactions [6].

Enhancer-promoter (E-P) interactions are pivotal in regulating gene expression and have been extensively characterized in animal and plant genomes [7, 8]. However, our understanding of these interactions in plant genomes, particularly in economically significant crops such as cotton (Gossypium), remains limited. Cotton fiber is a highly specialized single-celled structure and serves as an excellent cell model for investigating cell differentiation, polar elongation, and cell wall biogenesis in plants [9]. The Gossypium genus has undergone complex evolutionary processes, including allopolyploidization, resulting in diverse genome types between diploid and tetraploid species [10]. These unique features raise fundamental questions about how E-P interactions function during fiber development and how genome evolution shapes and rewires E-P regulatory networks in Gossypium.

Here, we adapted the global RNA-RNA in vivo interaction mapping method, Plant RNA in situ conformation sequencing (pRIC-seq), to create comprehensive RNA-RNA spatial interaction maps for cotton. This study aims to compare these RNA interaction maps in diploid and tetraploid cotton, revealing the novel E-P interaction pairs that arise with polyploidization. Additionally, by comparing wild type and fuzzless/lintless (fl) mutant, which lacks normal fiber initiation and elongation, we explore the critical regulatory roles of RNA-RNA interactions, including E-P interaction pairs, during fiber development (Fig. 1a). Our findings present the first genome-wide RNA-RNA interaction map in higher plants and provide crucial insights into the regulatory mechanisms of RNA-RNA interactions in controlling key agronomic traits in crops.

Fig. 1.

RNA-RNA interactions in cotton identified by pRIC-seq. a Schematic view of integrated analysis of multi-omics data in cotton ovules. b The percentages of cytosine around the junction of chimeric reads in tetraploid cotton. c IGV visualization of pre-miRNA ghr-MIR390a secondary structure with sRNA-seq signals and pRIC-seq chimeric reads. Purple line indicates the junction located pCp. d Global RNA-RNA interaction map in tetraploid cotton. e The local interaction region at Ghir_D04:11,060,000–11110000, magnified from d. f Classification of total chimeric reads identified in tetraploid. g IGV snapshot of pRIC-seq interaction signals between Ghir_D04G006600 and Ghir_D04G006610, with corresponding RNA-seq, DNase-seq, and H3K27ac and H3K4me3 ChIP-seq tracks. h Conserved interaction signals in collinear regions among A genome of diploid, At, and Dt subgenomes of tetraploid

Results

pRIC-seq captures global RNA-RNA interactions in cotton genomes

To investigate RNA-RNA interactions in the cotton genome during fiber cell development, we developed the pRIC-seq method for plants (see Methods; Additional file 1: Fig. S1a), an adaptation of RIC-seq originally designed for mammalian cells [5]. This adaptation addresses the unique structural and compositional challenges posed by plant cells. The protocol begins with the preparation of protoplasts from the outer integument (OI) of cotton ovules, which serve as the fiber cell initiation layer. Based on our previous work on preparation of cotton protoplasts [11], we further refined the protocol for plant systems in several procedures on protoplast purification, RNA-RNA interaction library construction (see Methods; Additional file 1: Fig. S1b-e).

Using pRIC-seq, we analyzed global RNA-RNA interactions in diploid cotton (AA genome; 2n = 26) cultivar Shixiya1 (Gossypium arboreum), generating approximately 24 million reads for rRNA + and 67 million reads for rRNA- samples in two replicates. The chimeric reads, indicative of RNA-RNA interactions, constituted about 8.6% of input reads on the genome for the rRNA- sample (Additional file 2: Table S1). For allotetraploid cotton (AADD genome; 2n = 52), we generated rRNA- samples for Xuzhou 142 (Gossypium hirsutum) and its fuzzless/lintless mutant (fl). The WT samples yielded around 111 million reads in two replicates, while the fl sample yielded approximately 80 million reads, with an average of 11.1% and 11.9% of reads identified as chimeric reads, respectively (Additional file 2: Table S1). The percentages of nucleotide analog pCp-biotin around the junction of chimeric reads reached approximately 90% for all samples, indicating effective capture and identification of RNA-RNA interactions mediated by pCp (Fig. 1b, Additional file 1: Fig. S2a).

The reproducibility of pRIC-seq was confirmed with a correlation value greater than 0.9 for the relative abundance of chimeric reads between two replicates in both diploid (Shixiya1) and tetraploid (WT) (Additional file 1: Fig. S2b-c). Notably, a positive correlation was observed between RNA levels and the strength of intra-gene RNA chimeric read signals (Additional file 1: Fig. S2d). This pattern was consistent across different genes in WT or fl, as well as between paired differential expressed genes in both WT and fl (Additional file 1: Fig. S2e-f), further supporting the reproducibility of pRIC-seq data.

To validate the accuracy of the RNA structural information reflected by the pRIC-seq, we first analyzed the rRNA + sample from diploid cotton. The data supported the conserved loop structures of 18S rRNA, confirming the reliability of pRIC-seq (Additional file 1: Fig. S2g). Additionally, we assessed the support of pRIC-seq data for the stem-loop structure of miRNA precursors (pre-miRNAs). The distribution of pRIC-seq chimeric reads at the pre-miRNA stem-loop structures showed a concentration in paired regions (Additional file 1: Fig. S2h). For instance, in the two precursors of ghr-MIR390a and ghr-MIR160, pRIC-seq captured signals consistent with predicted structure (Fig. 1c, Additional file 1: Fig. S2i), demonstrating the accuracy of pRIC-seq in representing RNA secondary structures.

Based on the chimeric RNA information revealed by pRIC-seq, we constructed a global RNA-RNA interactions map for the allotetraploid cotton (Fig. 1d). There are strong interaction signals were observed between homologous chromosomes, with distinct interaction signals for At and Dt subgenome homologous chromosomes flanking both sides of the diagonal signal, indicating significant homologous chromosome RNA interactions in addition to those within the same chromosome. The similar interaction signals between subgenomes were also present in Hi-C data of allotetraploid cotton [12, 13]. But in contrast to the Hi-C map, pRIC-seq signals were predominantly concentrated in the transcribed regions, due to the technique’s characteristic of capturing transcriptional RNA-RNA interactions (Fig. 1e-h, Additional file 1: Fig. S2j). The distribution analysis of the chimeric reads showed that 43.0% were completely located in the intergenic regions, with the remainder linked to gene region (Fig. 1f). We amplified a local region of RNA-RNA interactions at Ghir_D04:11,060,000–11110000, exhibiting a ~ 50 kb distal RNA interaction between Ghir_D04G006600 and Ghir_D04G006610. Their interaction signals of chimeric reads from pRIC-seq were distinct from those obtained from RNA-seq and Histone ChIP-seq visualized using Integrative Genomics Viewer (IGV) (Fig. 1e and g).

The diploid RNA-RNA interaction map displayed a typical strong diagonal signal, similar to the DNA interaction map observed in Hi-C [14] (Additional file 1: Fig. S2k). Statistical analysis of the chimeric read distribution revealed fewer gene–gene interactions in the diploid, while the proportion of other types of interactions was similar to those observed in the tetraploid (Additional file 1: Fig. S2l). Figure 1h shows a conserved collinear region of the three homologous chromosomes in the diploid A genome and tetraploid At and Dt subgenomes, where chimeric signals reflecting intragenic interactions among three genes and intergenic interactions between ABCG40 and adjacent genes were highly conserved. We quantified intragenic interactions for 6,632 homologous genes across A, At and Dt genome/subgenomes, revealing significant correlation (Additional file 1: Fig. S2m). Notably, the correlations of intragenic chimeric reads between homologous genes in the tetraploid subgenomes (At and Dt) were significantly higher than those between the At subgenome and the diploid A genome, as well as between the Dt subgenome and the diploid A genome. This pattern suggests enhanced coordination of RNA-RNA interactions between these homologous chromosomes within the tetraploid species, potentially reflecting evolutionary adaptation or functional divergence.

In summary, pRIC-seq enabled the construction of intra- and intermolecular RNA-RNA interactions in diploid and tetraploid cotton, precisely capturing known RNA interactions. The high proportion of chimeric reads, particularly those containing pCp, underscores the robustness of our pRIC-seq method in capturing RNA-RNA interactions in plants. This comprehensive dataset provides a solid foundation for further analysis of the RNA interactome during cotton fiber cell development in different genetic backgrounds.

Identification of enhancers in cotton

Enhancers and promoters are fundamental regulatory elements in transcriptional regulation. While mammalian enhancers are typically characterized as bidirectionally transcribed and unstable RNAs, plant enhancers exhibit distinct transcriptional and chromatin features [15]. In plants, enhancer activity is primarily associated with enhancer transcription, chromatin accessibility, and histone modifications, particularly H3K27ac and H3K9ac [8, 16]. Based on these features, we integrated global nuclear run-on followed by cap-selection assay (GRO-cap), chromatin accessibility data (DNase-seq/ATAC-seq), and H3K27ac histone modification data to systematically identify and classify putative enhancers in cotton.

GRO-cap, a refined version of GRO-seq, is highly sensitive for detecting transcription start sites (TSSs) of nascent transcripts, enabling precise identification of active enhancers [17]. In this study, we conducted the first GRO-cap assay in cotton (Additional file 1: Fig. S3a-c). For both tetraploid and diploid cotton, two biological replicates were sequenced, yielding 124 million and 117 million uniquely mapped reads, respectively. Reproducibility analysis revealed a correlation of ~ 0.98 between replicates (Additional file 1: Fig. S3d, Additional file 2: Table S1). Nascent RNA signals were significantly enriched near active TSSs of annotated genes, with minor antisense transcription observed upstream of TSSs (Additional file 1: Fig. S3e-f).

Using MACS2, we identified 127,347 and 144,909 active TSSs of transcription units across the tetraploid and diploid cotton genomes, respectively. As anticipated, a higher proportion of TSSs were located proximal to annotated genes in both genomes (Fig. 2a). Genome-wide profiling further confirmed that GRO-cap signals were significantly enriched at gene TSS regions, which also showed co-enrichment of canonical promoter histone marks, H3K4me3 and H3K27ac (Additional file 1: Fig. S3g).

Fig. 2.

Global enhancer landscapes of diploid and tetraploid cotton. a Distribution of transcription units identified by GRO-cap across genomic regions. The 5’UTR region indicates the 2 kb upstream region of the translation start site. The 3’UTR region indicates the 2 kb downstream region of the translation end site. NTs: Nascent Transcripts. b Schematics for defining enhancers and promoters. c,d The heatmap shows the signal of enhancer features in diploid and tetraploid. e Bidirectional stable (BS) enhancer E5674 and unidirectional unstable (UU) enhancer E2773 were demonstrated using IGV. f Fluorescence scanning to examine the transcriptional activation potential of predicted enhancers based on the fluorescence intensity of reporter genes (top). Quantitative analysis of the transcriptional activation potential of predicted enhancers based on the intensity ratio of firefly luciferase and Renilla luciferase (fLUC/rLUC) (bottom). Data from three biological replicates. The p-values are from a two-sided Student’s t-test (***: p-value < 0.001)

In total, GRO-cap detected 24,186 and 16,456 promoters in the tetraploid and diploid cotton genomes, respectively, with 1,460 (6%) and 1,160 (7%) promoters exhibiting bidirectional transcription initiation. We also identified a substantial number of transcription units in the intergenic regions of tetraploid and diploid cotton, accounting for 32.7% and 27.6% of total transcription units, respectively (Fig. 2a). These intergenic transcription events suggest the presence of widespread non-coding regulatory elements, including putative enhancers, consistent with the enhancer-associated transcription captured by GRO-cap. Using an integrated multi-omics approach, we thus systematically identified potential enhancers in diploid and tetraploid cotton. The candidate enhancers were defined as regions outside promoters with H3K27ac binding, chromatin accessibility, and nascent RNA transcription units (see Methods; Fig. 2b). ChIP-seq and DNase-seq signals in enhancer regions were normalized to calculate z-scores, and the Rank Ordering of Super-enhancer (ROSE) algorithm [18] was used to classify super-enhancers (SEs) and typical enhancers (Es). Ultimately, 3,597 and 9,686 enhancers were identified in diploid and tetraploid, respectively (Fig. 2c-d, Additional file 1: Fig. S4a-b, Additional file 3: Table S2).

In the tetraploid cotton genome, 31% of enhancers were classified as unstable transcriptional enhancers (Additional file 1: Fig. S4c). Stable transcriptional enhancers exhibited higher levels of histone modifications (H3K4me3 and H3K27ac) and chromatin accessibility compared to unstable enhancers (Additional file 1: Fig. S4d). Further classification based on transcription direction and stability revealed four enhancer types: 2,173 (22.4%) bidirectional stable (BS) enhancers, 347 (3.6%) bidirectional unstable (BU) enhancers, 4,506 (46.5%) unidirectional stable (US) enhancers, and 2,660 (27.5%) unidirectional unstable (UU) enhancers (Fig. 2b). Representative examples of a BS enhancer (E5674) and a UU enhancer (E2773) are shown in Fig. 2e, while additional examples are presented in Additional file 1: Fig. S4e.

To validate our findings, we performed enhancer reporter assays in Nicotiana benthamiana to test the transcriptional activation ability of 30 randomly selected putative enhancers. Among the 27 successfully cloned enhancers, 19 exhibited significant enhancer activity compared to controls, yielding a success rate of ~ 70% (Fig. 2f, Additional file 1: Fig. S5). These results underscore the robustness of our enhancer identification approach.

In summary, we systematically identified and classified enhancers in cotton using a combination of GRO-cap, chromatin accessibility, and histone modification data, all of which can transcribe eRNA. These findings not only provide a comprehensive resource for studying transcriptional regulation in cotton but also establish a crucial foundation for subsequent enhancer-promoter interaction analyses, paving the way for a deeper understanding of regulatory networks driving cotton gene expression.

Polyploidy-driven innovations in RNA-RNA and enhancer-promoter interactions

In our data, it is noteworthy that the proportion of RNA-RNA interactions associated with intergenic regions is high in both tetraploid and diploid cotton (Fig. 1f, Additional file 1: Fig. S2l). We speculate that this may be associated with RNA-RNA interactions associated with regulatory elements in the intergenic region. Our analysis found that the chimeric reads in the intergenic and genetic regions respectively enriched in the peaks of ChIP-seq for H3K27ac and H3K4me3 (Fig. 3a), which are markers of enhancers and promoters, and there is coupled with the presence of attached RNA in both the promoter and enhancer regions (Fig. 2c-d, Additional file 1: Fig. S4f-g). Thus, analyzing the RNA-RNA interaction information of cotton is a feasible route to analyze E-P interactions (Fig. 3b). As shown in Fig. 3c, the chimeric reads signals of pRIC-seq can successfully exhibit the regulatory relation between enhancers and promoters.

Fig. 3.

Global E-P interaction maps in diploid and tetraploid cottons. a The enrichment of H3K27ac and H3K4me3 in intergenic and genic portions of the gene-intergenic chimeric reads. b Schematic representation of enhancer and promoter interactions via their RNAs. c IGV view showing the interaction between enhancer E5674 and promoter of Ghir_D05G015160. d Pie charts displaying the number of enhancer-promoter (E-P), enhancer-enhancer (E-E), and promoter-promoter (P-P) RNA-RNA interactions identified by pRIC-seq. e The percentage of enhancers that interact with proximal or distal promoters. f Schematic diagram showing the formation of E-P interactions in diploid and tetraploid. g Number of E-P interactions in different genome/subgenomes. h The interaction genes for a conserved enhancer named as E176 and E684 in diploid and tetraploid, respectively. Blue lines indicate conserved E-P interactions between diploid and tetraploid

From our pRIC-seq analysis, we identified the RNA-RNA interactions from 8,497 enhancer-promoter (E-P), 2,223 enhancer-enhancer (E-E), and 25,455 promoter-promoter (P-P) interactions in tetraploid cotton, compared to 2,024 E-P, 260 E-E, and 8,896 P-P interactions in diploid cotton (Fig. 3d). These interactions enabled us to construct E-P interaction maps for both tetraploid and diploid cotton (Additional file 1: Fig. S6b).

The global E-P interaction maps showed that about 83% of E-P interactions are cross-chromosomal. Consistent with published studies in animals [6], only about 8% of enhancers interact with the nearest promoter, while 92% interact with distal promoters in cotton (Fig. 3e). While there is substantial evidence for long-distance E-P interactions in plants [8, 19], studies on trans-chromosomal E-P interactions remain scarce. The distance of E-P interacting on the same chromosome peaked at approximately 2.3 kb, 8 kb and 34 kb in tetraploid and peaked at approximately 5 kb in diploid (Additional file 1: Fig. S6c). We also observed that enhancers in cotton exhibited a higher tendency to interact with multiple promoters, whereas most promoters were associated with a single enhancer (Additional file 1: Fig. S6d-e).

Polyploidization serves as a major evolutionary force for plant genomes, leading to extensive changes in gene regulatory networks. The allotetraploid AD genome is composed of Dt and At subgenomes. Following hybridization and polyploidization, the At subgenome created numerous novel interactions with the Dt subgenome (Fig. 3f-g, Additional file 1: Fig. S6b). Our analysis indicates that polyploidization from A to AD resulted in approximately 5,914 new E-P interactions arising from cross interactions between the two subgenomes, contributing nearly 71% of all interactions in the tetraploid genome (Fig. 3g).

Using BLAST analysis, we identified 1,255 conserved enhancers between diploid and tetraploid genomes, of which 194 were detected to interact with promoters. Among these, 65 enhancers retained conserved E-P interactions in both diploid and tetraploid genomes (Additional file 1: Fig. S6f). For instance, enhancer E176 interacts with the promoters of six genes in the diploid genome, while its tetraploid counterpart E684 interacts with the promoters of six genes, including conserved promoters of Ga08g00859 and Ghir_A08G008170 (Fig. 3h).

Collectively, our findings demonstrate that polyploidization has generated a large number of novel E-P interactions in cotton genomes. These newly formed E-P interactions may contribute significantly to the emergence of novel gene regulations and associated phenotypes in the allotetraploid cotton genome.

Sequence features underlying E-P RNA interactions in cotton

Enhancers regulate promoters across long distances to establish the spatiotemporal specificity of gene expression, regardless of distance and orientation [7, 19]. However, the mechanisms by which enhancers recognize their target promoters in plant genomes remain poorly understood. Our pRIC-seq data have established numerous E-P interaction connections, providing a robust dataset to uncover the underlying sequence features mediating these interactions.

Using MEME suite, we identified enriched motifs in the 250 bp region flanking the enhancer and promoter interaction sites. Many of these motifs exhibited complementary base pairing in their respective regions. Notably, the top 20 significant motifs were present in nearly 40% of enhancers and promoters, some of which showed strong complementary base pairing (Fig. 4a, Additional file 1: Fig. S7a). The minimum free energy (MFE) value of the 250 bp flanking regions for enhancer and promoter pairing was significantly lower than those of randomly shuffled sequences with the same nucleotide composition (Fig. 4b). Furthermore, increased chimeric reads supports that pairing correlates with lower MFE density (Additional file 1: Fig. S7b).

Fig. 4.

E-P interactions formed by complementary base pairing. a Motifs enriched in pairwise interacting RNA fragments from enhancers and promoters. b Comparison of MFE values between chimeric RNA fragments of E-P and shuffled fragments with the same nucleotide composition. The p-values are from a two-sided Student’s t-test (***: p-value < 0.001). c The Pie chart showing the statistics of TEs and STRs in E-P interaction pairs. d Percentages of TEs and STRs elements within 250 nucleotides flanking the E-P interaction site. Chi-square test was used for p-values (***: p-value < 0.001, **: p-value < 0.01). e IGV view showing the interaction between enhancer E3577 and promoter of Ghir_D09G024030. f Schematic of the RNA duplex formed between E3577 RNA and promoter RNA of Ghir_D09G024030

We next performed the sequence feature analysis and identified numerous repetitive sequences within the enhancer- and promoter-enriched motifs, including single-base repeats, as well as short tandem repeats (STRs) such as CG, AU and GAA (Fig. 4a, Additional file 1: Fig. S7a). Given that repetitive sequences and transposons (TEs) account for about 74% of Gossypium genomes, we explored their potential contributions to the observed E-P interactions. Statistical analysis revealed that, in tetraploid cotton, 45.8% of E-P interactions contained TEs or STRs on both sides, 37.1% of E-P interactions contained TEs or STRs in the enhancer region, and 9.2% of E-P interactions contained TEs or STRs in the promoter region (Fig. 4c), with diploid cotton displaying a similar pattern (Additional file 1: Fig. S7c). Further investigation showed that STRs and LINE/L1 elements were enriched in both promoter and enhancer interaction sites, and LTR/Gypsy elements were notably higher in enhancers compared to random controls (Fig. 4d). Additionally, we found that E-P interactions containing STR and TE exhibited lower MFE values, suggesting their involvement in stabilizing E-P interactions (Additional file 1: Fig. S7d). For example, enhancer E3577 with CCCCAA-repeat interacted with the promoter of gene Ghir_D09G024030, forming a long pairing stem with a MFE of −160.6 kcal/mol (Fig. 4e, f).

Taken together, the above results suggest that selective enhancer-promoter interactions might be primarily mediated through base complementary pairing with repetitive sequences and transposable elements, particularly STRs and Class I TEs, playing a supporting role. These results underscore the potential regulatory significance of STR sequences and TEs in modulating plant gene expression dynamics, warranting further investigation into their specific functions in E-P interactions.

Specific E-P RNA interactions regulate fiber cell development

Xuzhou 142 (WT) and its fuzzless/lintless mutant (fl) are commonly used as a pair of comparative materials for studying fiber development. The causal gene MYB25-like has been identified in previous studies [20, 21]. Previous comparative studies have focused on transcriptome, proteome and epigenetic modifications [11, 22, 23]. Here, we have the opportunity to compare their differences in E-P interactions uncovered by RNA-RNA interactions.

We first compared the gene expression levels between genes with enhancer interactions and those without such interactions at the whole genome level. We observed that genes with enhancer interactions exhibited significantly higher expression levels in both WT and fl (Fig. 5a, Additional file 1: Fig. S8a). To investigate the regulatory role of enhancers, we selected the super enhancer SE968, a 16 kb tandemly organized region actively expressed in ovule and leaf tissues (Fig. 5b). Based on E-P interaction data from our pRIC-seq, SE968 was predicted to regulate 15 potential targets in ovule tissue, including 14 protein-coding genes and one long non-coding RNA (lncRNA). To validate its role, we used a virus-induced gene silencing (VIGS) system [24–26] to silence SE968 in cotton seedling leaf tissues and examined the expression of its target genes. The RNA-seq analysis of SE968-silenced plants demonstrated effective and specific silencing of the enhancer, as the expression of neighboring coding genes was unaffected (Additional file 1: Fig. S8b). The silencing efficiency was further confirmed by qRT-PCR (Fig. 5c). The RNA-seq revealed many commonly differentially expressed genes in two independent VIGS-treated plants, which enriched in stimulus-related biological processes, aligning with the general responses induced by viral vectors (Additional file 1: Fig. S8c-e). Further examination of SE968’s target genes revealed that 12 targets were expressed in leaf tissue. Among them, Ghir_A11G006130, Ghir_A05G031770, Ghir_D04G004030, Ghir_D11G024060 and Ghir_D11G0039 showed significantly down-regulated expression, as confirmed by both RNA-seq and qRT-PCR. The remaining 7 target genes showed no significant expression changes (Fig. 5d, Additional file 1: Fig. S8f-g), which may be attributed to the tissue-specific nature of the E-P interactions. This case provides valuable insights into the downstream effects of enhancer silencing and underscores the reliability of our experimental design in studying enhancer-mediated gene regulation.

Fig. 5.

E-P interactions involved in fiber development. a Cumulative frequency of RNA expression for genes linked with or without enhancers in WT. T, genes promoters linked with enhancers; F, genes promoters linked no enhancers. The p-value was calculated by Kolmogorov–Smirnov test. b IGV view showing SE968 interacting with the promoters of Ghir_A05G031770 and Ghir_A12G020020. The red arc lines show the counts of chimeric reads in pRIC-seq. Purple read coverage shows the RNA level in tetraploid cotton leaf tissue from MaGenDB [27]. c The qRT-PCR analysis for VIGS knockdown effect on enhancer SE968. Data are mean ± sd, with three biological replicates. The p-values are from a two-sided ANOVA (****: p-value < 0.0001). d The expression analysis of enhancer-linked promoter genes with qRT-PCR. Data are mean ± sd, with three biological replicates. The p-values are from a two-sided ANOVA (****: p-value < 0.0001, ns: p-value > 0.05). e Venn diagram showing the intersection of E-P interactions identified in WT and fl samples. f Comparison of RNA expression fold changes (WT/fl, log2) across three gene groups classified by E-P interaction: WT-specific, Common, and fl-specific. The p-values are from two-sided Student’s t-test (***: p-value < 0.001). g Gene ontology analysis for the genes with WT specific E-P interactions. h IGV view showing SE470 interacting with the promoters of Ghir_A05G018650 (SMT2)

By quantifying the interaction with more than three chimeric reads, we identified 1,627 and 1,471 E-P interactions in WT and fl, respectively (Fig. 5e, Additional file 1: Fig. S9a, Additional file 4: Table S3). The gene expression associated with WT-specific E-P pairs was significantly higher than those of fl-specific and common E-P pairs (Fig. 5f). Gene ontology (GO) analysis showed that genes linked to WT-specific E-P pairs were particularly enriched in seed trichome development (Fig. 5g). In IGV tracks, we observed E-P interaction between enhancer SE470 and the promoter of Ghir_A05G018650 (SMT2), and an E-P interaction between enhancer E2154 and Ghir_D08G011000 (PEL3) in WT (Fig. 5h, Additional file 1: Fig. S9b). The expression of these genes was significantly higher in WT than fl (Fig. 5h, Additional file 1: Fig. S9b). Both SMT2 and PEL3 are implicated in trichome development, with PEL3 known to be involved in fiber trait [28], suggesting that the specific E-P interactions in WT are involved in fiber development.

Interestingly, we also identified higher expression of genes regulated by fl-specific E-P pairs in fl. For example, the enhancer E2773 specifically interacted with the promoter of Ghir_D10G007800 (NTT1) in fl, which exhibited higher expression in fl than in WT (Additional file 1: Fig. S9c). We analyzed the top 30 genes by expression fold change targeted by WT-specific and fl-specific E-P interactions, respectively. Among the WT-specific genes, fourteen were previously identified as specifically expressed in fiber cells [11, 28, 29] (Additional file 1: Fig. S9d).

These findings highlight the importance of specific E-P RNA interactions in regulating fiber cell development, providing insights into the molecular mechanisms underpinning fiber traits in cotton.

E-P interactions enrich within GWAS and eQTL sites

To investigate the impact of RNA-RNA interactions and E-P interactions on gene expression and phenotypic traits in cotton, we integrated expression quantitative trait locus (eQTL) and GWAS data from the published resource [30–35] (Fig. 6a, Additional file 1: Fig. S10a-c, Additional file 5: Table S4) with our high-resolution RNA-RNA interactions and E-P interactions maps of tetraploid cotton.

Fig. 6.

E-P interactions enrich within GWAS and eQTL sites. a Schematic diagram showing the eQTLs and GWAS loci. b Percentage of eQTL flanking regions (0–2 kb) showing target gene associations supported by chimeric reads. c GWAS loci are enriched in the RNA-RNA interaction sites from pRIC-seq data. Randomly shuffled regions were taken as control. d IGV snapshot of pRIC-seq chimeric reads supporting interaction between cis-eQTL and its target gene Ghir_A05G011060. The cis-eQTL is surrounded by GWAS loci controlling cotton fiber elongation. e GWAS-associated E-P interactions. f IGV snapshot of enhancer E5761 interacting with the promoter of Ghir_A06G016070 (APR3). A GWAS locus associated with fiber length were located in enhancer E5761. RAP-qPCR was used to verify their interaction. Data are mean ± sd, with three biological replicates. g RAP-qPCR validation for the interaction of hub enhancer E6609 to its multiple genes. Data are mean ± sd, with three biological replicates. The p-values are from two-sided Student’s t-test for g-h (***: p-value < 0.001, **: p-value < 0.01, *: p-value < 0.05, ns: p-value > 0.05)

Our analysis revealed that RNA-RNA interaction chimeric reads captured a higher proportion of connections between eQTLs and their target genes compared to random sites (Fig. 6b). Similarly, GWAS functional mutation sites are enriched in the interaction regions detected by pRIC-seq (Fig. 6c). We quantified cis- and trans-eQTL captured by pRIC-seq within varying flanking regions of the eQTLs, finding that wider flanking regions enhance the detection of relevant eQTLs involved in these interactions (Additional file 1: Fig. S10d). For instance, Fig. 6d shows a cis-eQTL and its regulated gene Ghir_A05G011060, surrounded by GWAS loci associated with cotton fiber elongation (Fig. 6d, Additional file 1: Fig. S10e). Importantly, our pRIC-seq captured the interaction between this cis-eQTL and its downstream-regulated gene, providing direct experimental evidence that this cis-eQTL functions as an enhancer, potentially influencing the expression of Ghir_A05G011060 and affecting cotton fiber elongation [34] (Fig. 6d, Additional file 1: Fig. S10e).

Next, we performed an association analysis between 17,006 GWAS sites associated with 17 cotton traits and the 8,497 E-P interactions identified in this study, generating a cotton GWAS sites interaction map that includes 75 enhancers and 151 promoters in the identified E-P interactions (Fig. 6e, Additional file 6: Table S5). For example, our E-P interaction map shows that fiber length GWAS site located in enhancer E5761 might remotely regulate the transcription of Ghir_A06G016070 (APR3) to control fiber length (Fig. 6f). These interactions were validated using RNA antisense purification coupled with quantitative PCR (RAP-qPCR) (Fig. 6f). APR3 is known to be involved in actin reorganization during polarized trichome growth in Arabidopsis [36]. Thus, we speculated that E5761 might regulate the expression of APR3 in cotton to control the elongation of fiber cells, similar to polarized trichome growth. Notably, we discovered a segment of the transposon LTR/Caulimovirus sequence within E5761. Chimeric reads from pRIC-seq data revealed that this transposon possibly mediates the interaction with the target gene, further supporting the E-P interactions pattern in the cotton genome mentioned above.

Interestingly, our analysis revealed that certain enhancers or promoters formed prominent interaction hubs on the GWAS sites interaction map. Specifically, we identified six “hub” enhancer sites, each with more than five connections. We validated the multiple interactions for hub enhancer E6609 using RAP-qPCR. In protoplast RNA lysis samples derived from cotton ovules, we selected six out of nine predicted target genes for validation. Remarkably, all six target genes were confirmed to interact with hub enhancer E6609, demonstrating the robustness of these interactions (Fig. 6g).

In summary, our analysis demonstrates that functional mutations, including GWAS and eQTL loci, are enriched in RNA-RNA and E-P interaction regions such as E5761 and E6609. These interactions significantly regulate fiber elongation genes, highlighting the crucial E-P regulatory networks that control the key cotton traits.

Discussion

RNA-RNA spatial interactions are vital for various cellular processes, including transcriptional and post-transcriptional regulation. Leveraging the Plant RNA in situ Conformation Sequencing (pRIC-seq) method developed in this study, we have pioneered the construction of comprehensive RNA-RNA interaction maps in plants. This study marks the first instance of generating global RNA-RNA interaction maps for both diploid and tetraploid cotton, thereby inferring E-P interactions from these RNA-RNA interactions.

Identification of enhancers in plants

The definition of enhancers in plants remains challenging, as several promoter elements have been shown to function as enhancers in mammalian studies [18, 37, 38]. In our study, distinguishing enhancers from promoters solely based on histone H3K27ac and H3K4me3 modification signals and chromatin accessibility proved difficult (Fig. 2c-d, Additional file 1: Fig. S4f-g). We therefore defined enriched elements at transcription start sites (TSSs) of genes as promoters, while those enriched elsewhere were classified as enhancers. However, this approach may overlook nuanced regulatory features shared by promoters and enhancers. Given the complexity of the cotton genome, it cannot be ruled out that some enhancers may represent unannotated genes or regulatory elements associated with nearby unannotated transcripts.

Given the necessity of enhancer-promoter interactions for enhancer function, incorporating interaction-based technologies such as RIC-seq and Hi-ChIP as reference data holds promise for improving the accuracy and positive identification rates of plant enhancers. These techniques provide direct evidence of functional interactions, complementing traditional chromatin and transcriptomic markers. Thus, the definition and characterization of enhancers in plants require further systematic investigation.

The utility of pRIC-seq in RNA-RNA and E-P interaction studies in plants

In mammals, E-P interactions are often mediated by RNA-RNA pairing between enhancer RNAs (eRNAs) and upstream antisense RNAs (uaRNAs) transcribed from bidirectional promoters [5, 6]. However, our study found that bidirectional transcription of promoters that produce uaRNAs is rarely found in cotton, consistent with previous observations in plants [39, 40]. It raises the question of whether similar mechanisms operate in plants.

Among the 9,686 enhancers identified in tetraploid cotton, 3,021 (31.2%) were involved in E-P interactions, all of which transcribed nascent eRNAs, as supported by GRO-cap, with 93.6% also exhibiting stable RNA-seq signals. Meanwhile, only 5.54% of the 5,476 promoters involved in E-P interactions showed bidirectional transcription. Strikingly, the majority of E-P RNA-RNA interactions occurred between eRNAs and unidirectional promoter transcripts with sense strands, accounting for over 93.6% of all E-P interactions. These findings suggest that, in contrast to mammalian systems, bidirectional promoter activity is not a prerequisite for RNA-mediated E-P interactions in plants. Instead, cotton enhancers can interact flexibly with promoter RNAs regardless of orientation, supporting a plant-specific model of E-P interaction. It should be noted that this study did not elucidate the precise molecular mechanisms governing the interactions between eRNAs and promoter RNAs. Furthermore, as the research was confined to cotton, the generalizability of our findings to other plant species requires further investigation.

Currently, high-throughput methods for validating RNA-RNA and E-P interactions in plant genomes are limited, posing a barrier to research in plant development and genetics. The development of pRIC-seq addresses this gap by providing a robust tool for such studies. Our findings demonstrate that pRIC-seq is highly effective in elucidating E-P interactions, which are crucial for mining potentially functional regulatory elements and understanding gene regulation.

Mechanisms of E-P interactions in plants

This pilot report explores how enhancers recognize promoters in plants. The mechanisms by which transposons regulate gene functions and phenotypes in plants are largely unknown. Our study reveals that many E-P interactions are mediated through RNA complementary base pairing. Additionally, we identified that transposable elements are crucial in establishing these interactions. Recent studies have reported that SINE transposon elements mediate selective interactions between enhancers and promoters in the human genome [6]. Our research also reveals a significant role for transposons and short tandem repeats (STRs) in establishing E-P interactions. However, the types of transposons involved in these interactions are more diverse, encompassing both LTR and non-LTR RNA transposons, as well as DNA transposons (Fig. 4c,d). This discovery addresses a significant question in plant genome transcription regulation. However, it is important to note that this does not cover all E-P interaction mechanisms. Many E-P interactions not mediated by transposable elements require further investigation to understand their formation and regulatory mechanisms.

Additionally, SNP mutations in enhancers like E5671 and E6609 may not significantly affect complementary pairing. We hypothesize that transcription factor (TF) interactions are crucial in these contexts. Analysis of TF binding sites from plantTFDB [41] revealed enrichment of factors such as ERF, C2H2, LBD, GATA, MYB, Dof, and BBR-BPC at chimeric-read contact sites, supporting our hypothesis (Additional file 1: Fig. S10f-g). Notably, transcription factor binding regions exhibited more feature mutations than random regions (Additional file 1: Fig. S10h). Previous studies have reported that pairs of interacting TFs bind to enhancers and promoters and contribute to their physical interactions [18].

Implications for agronomic trait regulation in cotton

Previous cotton population genetics studies, including genome-wide association studies (GWAS) and expression quantitative trait locus (eQTL) analyses, have identified numerous associations between agronomic trait phenotypes, gene expression differences, and genomic variations in cotton [30, 31]. However, a major challenge with GWAS approaches is that many of the identified GWAS loci reside in intergenic regions, which is also the case in cotton (Additional file 1: Fig. S10c), making it difficult to understand the underlying genetic mechanisms. Similarly, while eQTL analyses have revealed connections between gene expression differences and genetic variations, the regulatory mechanisms linking expression changes to sequence variations remain unclear in many cases [35, 42].

Our research indicates that numerous agronomic traits in cotton are regulated through enhancers. Notably, by constructing a global map of E-P interactions, we created a relationship map among GWAS loci. This map revealed internal connections between many GWAS loci that control the same traits. Additionally, our map revealed connections between different traits, suggesting the existence of pleiotropic genes or enhancer elements capable of controlling multiple traits. This insight provides a valuable framework for understanding the genetic architecture of complex traits and improving breeding strategies.

In conclusion, our study underscores the utility of pRIC-seq in advancing the understanding of RNA-RNA and E-P interactions in plants. It sheds light on the mechanisms underlying these interactions and highlights their significance in regulating important agronomic traits. The findings pave the way for future research to further unravel the complexities of gene regulation and to leverage this knowledge for crop improvement.

Conclusions

In this study, we developed the pRIC-seq technique and, for the first time, constructed a genome-wide RNA-RNA interactome and established an E-P interaction map in plants. Our work reveals the potential functions of RNA interactions during cotton polyploidization and development. These findings provide a resource for cotton research and breeding, and lays the groundwork for studying RNA interactions in other plant species.

Material and methods

Plant materials and growth conditions

The diploid cotton G. arboreum cv. Shixiya1 (SXY1), the upland cotton G. hirsutum cv. Xuzhou 142 (WT) and its mutant Xuzhou 142 fuzzless/lintless (fl) were cultivated in an automated greenhouse with mimicked natural conditions in setting 16 h light and 8 h dark for a day-night cycle.

pRIC-seq for protoplasts of cotton ovule outer integument

1. Protoplast preparation.

Protoplasts from the ovule outer integument at five stages (−2, −1, 0, 1, and 2 DPA) were prepared using partial tissue enzymatic digestion (PTED) [11]. After filtering through a 40 μm strainer and centrifugation (100 × g, 10 min) to remove undigested tissues and other large impurities, the protoplasts were washed with 20% sucrose and resuspended in 8% mannitol buffer. Viable protoplasts (> 90%, confirmed by Trypan Blue staining) were counted with a hemocytometer.

• (2)Crosslinking, permeabilization and MNase digestion.

Protoplasts (~ 107 cells) were crosslinked with 1% formaldehyde in W5 buffer at room temperature for 10 min, and then quenched with 0.125 M glycine. After washing three times with ice-cold W5 buffer, protoplasts were scraped into 1.5-mL tubes and collected by centrifugation (585 × g, 10 min at 4 °C). Protoplasts were resuspended in 1 mL of permeabilization buffer, mixed briefly and incubated on ice for 15 min. Protoplasts were spun down at 585 × g for 5 min at 4 °C. After washing three times with 1 × PNK buffer, the pellet was briefly incubated with 200 μL of 1 × MN mixture and dilute micrococcal nuclease 1:5000 for 10 min at 37 °C. The reaction was stopped by washing twice with 1 × PNK + EGTA buffer and twice with 1 × PNK buffer. Protoplasts were pelleted by centrifugation with 585 × g.

• (3)pCp-Biotin labelling and proximity ligation.

Protoplasts were gently resuspended in 100 μL of 1 × FastAP buffer plus 10 U of FastAP alkaline phosphatase and incubated at 37 °C for 15 min. The pellet was washed twice in 1 × PNK + EGTA buffer, twice in 1 × high-salt wash buffer and three times in 1 × PNK buffer. To deposit pCp-biotin at the 3’ ends, pellet was resuspended in 100 μL of ligation mixture 1 and incubated overnight at 16 °C. The pCp-labelled protoplasts were pelleted at 585 × g for 5 min at 4 °C and washed three times with 1 × PNK buffer. Protoplasts were further treated with 10 U of FastAP alkaline phosphatase in 100 μL of 1 × FastAP buffer for 15 min. After thorough washing, the pellet was resuspended in a PNK mixture and incubated at 37 °C for 45 min. The PNK treatment was stopped by washing twice with 1 × PNK + EGTA buffer and twice with 1 × PNK buffer (0.05% NP-40). For in situ proximity ligation, the ligation mixture 2 was applied directly to the pellet and incubated overnight at 16 °C. Protoplasts were pelleted at 1150 × g for 5 min at 4 °C and washed three times with 1 × PNK buffer.

• (4)RNA purification and DNase I treatment.

Protoplasts were pelleted at 585 × g. 200 μL of proteinase K buffer and 50 μL of proteinase K were applied and sequentially incubated at 65 °C for 3 h. The RNA was extracted with 750 μL of TRIzol, and 220 μL of chloroform following the manufacturer’s instructions. After centrifugation at 13,500 × g for 15 min at 4 °C, the supernatant was transferred to a 1.5 mL Eppendorf tube and precipitated overnight by mixing with 600 μL of isopropanol and 1.5 μL of GlycoBlueat at −80 °C. RNA was pelleted at 13,500 × g for 60 min at 4 °C and washed twice with 75% ethanol. The RNA pellet was dissolved in 15 μL of nuclease-free water. Total RNA with 20 μg was treated with a mixture containing 10 μL of 10 × RQ1 DNase I buffer, 3 μL of RNase inhibitor and 5 μL of DNase I at 37 °C for 30 min. RNA was then purified by TRIzol again and the RNA pellet was dissolved in 20 μL of nuclease-free water.

• (5)rRNA depletion.

DNase I treated RNA (5 μg) was then subjected to rRNA removal using the Ribo-off rRNA Depletion Kit. Purified RNA was resuspended in 16 μL of nuclease-free water.

• (6)RNA fragmentation and biotin enrichment.

The 16 μL RNA was mixed with 4 μL 5 × first-strand buffer and fragmented by heating at 94 °C for 5 min and immediately chilled on ice. The 20 μL fragmented RNA was incubated with preblocked MyOne™ Streptavidin C1 beads, 30 μL of nuclease-free water and 50 μL of 2 × TWB buffer for 30 min by rotating at room temperature. After washing 4 times with 1 × TWB buffer, the enriched RNAs were extracted with TRIzol and resuspended in 8 μL of nuclease-free water.

• (7)Strand-specific library preparation.

Strand-specific library was constructed using the VAHTS Universal V8 RNA-seq Library Prep Kit for Illumina (Vazyme, NR605), but skipped the fragment step. After size and quality check with the Agilent 2100 Bioanalyzer, the libraries were sequenced using the MGISEQ-2000 (BGI, China) in PE150 mode.

The detailed compositions of all buffers used in this pRIC-seq are summarized in Additional file 9: Table S8.

GRO-cap assay

The powder of ovule tissue at five stages (−2, −1, 0, 1, and 2 DPA), prepared by grinding in liquid nitrogen, was solubilised in 20 mL nuclear extraction buffer and filtered through a 40 μM Cell strainer. Triton X-100 was added to a final concentration of 1.2%, and the lysates were incubated on ice for 15 min. Nuclei were isolated using a 30–80% Percoll gradient, collected at the interface, and pelleted by centrifugation at 1,000 × g for 11 min at 4 °C. The nuclei were then washed twice with nuclear wash buffer. To assess the purity of isolated nuclei, western blot was performed using H3 (nuclear marker; ABclonal, A17562, 1:2,000), Actin (present in both nucleus and cytoplasm; ABclonal, A23959, 1:5,000), and UGPase (cytoplasmic marker; Agrisera, AS05-086, 1:5,000).

The purified nuclei were washed twice with 1 × transcription buffer and stored in 100 μL nuclei storage buffer. For transcription assays, 100 μL of 2 × reaction buffer was added to the nuclei, mixed and incubated at 30 °C for 5 min. Nuclear RNA was extracted using TRIzol and resuspended in 8 μL of nuclease-free water. To enrich for nascent RNA with cap structures, the RNA underwent enzymatic treatments: (1) Dephosphorylate tri- and di-phosphorylated RNA molecules while avoiding 5’-capped and mono-phosphorylated RNAs by using RNA 5’ Polyphosphatase. (2) Digest RNA with 5’-monophosphate ends by using Terminator 5’-Phosphate-Dependent Exonuclease. (3) Dephosphorylation of RNA 5’-ends by using Quick CIP. (4) Decap m⁷G-capped mRNA by using Tobacco acid pyrophosphatase (TAP). Finally, the uncapped RNA was used for library preparation with the VAHTS Small RNA Library Prep Kit, V2 for Illumina (Vazyme, NR811), following the manufacturer’s instructions.

The detailed compositions of all buffers used in this GRO-cap are summarized in Additional file 9: Table S8.

Dual luciferase reporter assay

The dual luciferase reporter assay was conducted to validate the function of enhancers in transcriptional activation. Briefly, the predicted enhancers elements were sent to companies for gene synthesis. The synthesis fragments were cloned into PGX-5dual [43] vector, which contains two luciferase reporter systems: a Firefly luciferase (fLUC) expression system driven by the mini 35S promoter and a Renilla luciferase (rLUC) expression system driven by a strong constitutive promoter from cauliflower mosaic virus. The enhancers were inserted upstream of the mini 35S promoter. Next, the vector was transformed into GV3101 chemically competent cell, and the empty vectors with 35S promoter alone were transformed in parallel as controls. Applying a needleless syringe, the mixture was co-infiltrated on the abaxial side of N. benthamiana leaves. Two days after infiltration, the leaves were subjected to quantitative analysis the ratio of fLUC/rLUC using a Dual Luciferase Reporter Gene Assay Kit (Vazyme, DL101) according to the manufacturer’s guidance. Sequences of enhancer elements used for gene synthesis are listed in Additional file 8: Table S7.

Plasmid construction and VIGS assay

Plasmids pTRV1, pTRV2 and pTRV2-GrCLA were utilized in the VIGS assay. A 478-bp fragment of enhancer SE968 was amplified from Gossypium hirsutum genomic DNA and inserted into pTRV2 between Sac I and Kpn I enzyme digestion sites. The PCR primers are listed in Additional file 7: Table S6.

For the VIGS assay, the plasmids pTRV2 (negative control), pTRV2-GrCLA (positive control), pTRV2-SE968 and associated plasmids were individually introduced into Agrobacterium strain GV3101. Transformed GV3101 cultures were grown overnight at 28 °C in an LB medium containing rifampicin (50 mg/L) and kanamycin (50 mg/L). The Agrobacterium cultures were then pelleted and resuspended (OD 600 = 1) in buffer with 10 mM MgCl₂, 10 mM MES, and 125 μM acetosyringone. After 3 h incubation at room temperature, the GV3101 strains harboring pTRV2, pTRV2-GrCLA or pTRV2-SE968 were mixed with an equal volume of GV3101 strain harboring pTRV1. The mixed Agrobacterium solutions were infiltrated into the abaxial side of cotyledons of 2-week-old cotton seedlings using needleless syringes. The cotton seedlings were grown for 2–3 weeks post-infiltration to allow the spread of the virus and the onset of gene silencing. Monitor the cotton plants for phenotypic changes. The enhancer silencing effects and the expression of targeting genes were validated by qRT-PCR.

qRT-PCR and RAP-qPCR

The qRT-PCR was performed using the SYBR Green I Master kit (Accurate Biology, AG11701). The primers for target genes are listed in Additional file 7: Table S6. The cotton Ubiquitin gene (GenBank accession number: DQ116441) was used as an internal control for the assays.

For RNA antisense purification (RAP) coupled quantitative PCR (RAP-qPCR), the 5’-biotin labeled and unlabeled antisense RNA probes with 37–50 bp lengths specifically targeting enhancers E5761, and E6609 were designed. RAP assays were performed using the BersinBio™ RNA Antisense Purification (RAP) Kit (BersinBio, China). Reverse transcription qPCR (qRT–qPCR) was performed as previously described [5]. The sequences of all RAP probes and RAP-qPCR primers are listed in Additional file 7: Table S6.

pRIC-seq data analysis

The pRIC-seq data were computationally processed with reference to the previous description [5]. Raw data from all the pRIC-seq libraries were deduplicated, and the retained reads were initially mapped to rRNA using STAR (v2.7.10) [44]. The remaining reads were separately mapped to the corresponding reference genome using STAR with parameters ‘–outSAMattributes All –alignIntronMin 1 –scoreGapNoncan −4 –chimSegmentMin 15 –chimJunctionOverhangMin 15 –outFilterMatchNminOver-Lread 0.33 –outFilterScoreMinOverLread 0.33 –alignSJoverhangMin 15 –alignSJDBoverhangMin 10 –alignSJstitchMismatchNmax 5 −1 5 5’. Reads that were not mapped using STAR were subsequently remapped with BWA-MEM (v0.7.17) [45] with default parameters. The diploid pRIC-seq data were aligned to the Gossypium arboreum (A2) reference genome [14], and the tetraploid pRIC-seq data were aligned to the TM-1 (Gossypium hirsutum) reference genome [46]. The low-quality and secondary mapping results were filtered using the SAMtools package [47] with “-q 30 -F 256” and the tag with “NH:i:1”. Next, chimeric reads were extracted from the alignment results of STAR and BWA-MEM, and those reads generated by RNA splicing events were discarded.

GRO-cap data analysis

The raw data were used to remove the universal adapter and filter reads less than 15nt using TrimGalore [48] with parameter ‘–paired –quality 20 –length 15 –fastqc –small_rna’. Then clean reads were aligned to the rRNA using STAR and remained reads were mapped to their corresponding genomes, Gossypium hirsutum and Gossypium arboreum, respectively, with the following parameters: ‘–outFilterMatchNminOverLread 0.33 –outFilterScore-MinOverLread 0.33 –outSAMattributes All’. The uniquely mapped reads were extracted and removed PCR duplicates using Picard (v3.1.0) [49]with default parameters.

Transcript levels of annotated genes were quantified using featureCounts [50] and calculated correlations between two replicates in the GRO-cap data. At the same time, we used the default parameters of deepTools [51] to calculate Spearman correlation coefficients between the two replicates, and these results all indicate the reliability of the data.

The documentation referred to deepTools was used to classify the strand-specific data into bam files for transcripts from the forward strand and bam files for transcripts from the reverse strand using SAMtools. Transcription units in the forward and reverse strands were identified directly from GRO-cap data using the MACS2 [52] program with the parameters ‘-f BAM –nomodel -p 0.05’ and quantified read counts for each transcription unit with featureCounts. We normalized reads to 10 million per sample and selected regions with at least 10 reads per transcription unit for subsequent analysis.

ChIP-seq data analysis

Cotton ovules were cross-linked for 15 min at room temperature with 1% formaldehyde, and cross-linking was stopped by PBS-glycine (0.125 M final). After washing twice with ice-cold PBS, they were immediately frozen in liquid nitrogen. Nuclear isolation, chromatin fragmentation, antibody incubation, and ChIP-seq libraries were prepared using our previously established method. After size and quality check with the Agilent 2100 Bioanalyzer, the libraries were sequenced on MGISEQ-2000 (BGI, China) in PE150 mode.

The raw ChIP-seq data of tetraploid cotton were obtained from published data [53], which were generated in our laboratory. The raw ChIP-seq reads were trimmed adapters using Fastp (v0.12.4) [54] and then the remaining reads were respectively mapped to the Gossypium arboreum and Gossypium hirsutum genomes using STAR with the following parameters ‘–alignEndsType Local –outFilterMismatchNmax 2 –outFilterMultimapNmax 1 –outFilterMultimap-ScoreRange 1’. After removal of PCR duplicates using Picard (v3.1.0) with default parameters, the two replicate BAM files were merged for peak calling. H3K4me3 and H3K27ac enriched peaks were identified by MACS2 [52] software with parameters ‘-nomodel -q 0.01’.

DNase data analysis

DNase-seq data for diploid and tetraploid cotton were downloaded from public data [55]. Trim Galore was used to remove the adapter in the raw data with parameters ‘–paired –quality 20 –length 15’. The clean reads were mapped to the Gossypium hirsutum and Gossypium arboreum genome using STAR with the following parameters ‘–alignEndsType Local –outFilterMismatchNmax 2 –outFilterMultimapNmax 1 –outFilterMultimapScoreRange 1’. PCR duplicates were removed using Picard. Accessible chromatin regions marked by deoxyribonuclease (DNase) I hypersensitivity were identified using MACS2 with parameters ‘-f BAM –nomodel -p 0.05’.

Identification of promoters and enhancers in cotton

In the analysis above, we have identified histone H3K27ac, H3K4me3-enriched peaks using ChIP-seq data, transcriptional units using GRO-cap, and accessible chromatin regions using DNase-seq. Promoters were defined as H3K4me3-enriched peaks located within ± 2 kb of the transcription start sites. For enhancer identification, we first combined histone H3K27ac-enriched regions and chromatin open regions as candidate enhancer regions. We then filtered out candidate enhancers that coincided with the promoter and filtered out candidates that did not contain a transcription unit. Candidate enhancers less than 5 kb apart were finally merged. A total of 9,686 enhancers were identified in the tetraploid, and a total of 3,597 enhancers were identified in diploids. We normalized the counts of ChIP-seq for H3K27ac and DNase-seq in enhancer regions to calculate z-scores, respectively. Based on the average of z-scores, the Rank Ordering of Super-Enhancer (ROSE) algorithm [18] was then applied to classify super enhancer (SE) and typical enhancer (E) (Additional file 1: Fig. S4a, b).

To identify conserved enhancers between diploids and tetraploids, we conducted BLAST analysis with the parameters ‘-evalue 1e-6’ to compare tetraploid enhancer sequences against their diploid counterparts. We applied stringent criteria, retaining only sequences with: (1) alignment identity of at least 90%. (2) alignment length exceeding 60% of the enhancer sequence length. Using these filters, we identified 1,255 conserved enhancers shared between the diploid and tetraploid cotton genomes. Among these, 176 enhancers were found to interact with promoters, and 60 exhibited conserved E-P interactions.

RNA-seq data analysis

Raw RNA-seq reads were obtained from published data [11] and then aligned to the rRNA using STAR (v2.7.10). The reads that not mapped to rRNA were mapped to the Gossypium hirsutum genome [46] using STAR with the local model. After PCR duplicates were removed using Picard (v3.10), gene expression quantifications were computed using the featureCounts program (v2.0.3) with the following parameters: -s 2 -t exon -p -minOverlap 3.

The genes with FPKM greater than 2 were selected for reproducibility analysis. DESeq2 [56] was used for differential gene expression analysis using gene read counts between WT and fl. Up-regulated genes were defined as log₂(foldchange) > 1 and p-value < 0.05, and down-regulated genes were defined as log₂(foldchange) < −1 and p-value < 0.05. We utilized the GO annotation prediction of genes using BLAST2GO, as described in previous literature [11], and employed the R package topGO to identify enriched GO terms.

Sequence motif analysis of E-P interactions

Sequences of the 250 bp region flanking the contact site of the E-P interaction were used for motif analysis utilisingthe STREME tool in MEME suite [57]. The parameters were set as follows: -nmotifs 10 –evalue 0.05 -minw 4 -maxw 30 -rna. We also calculated the minimum free energy (MFE) for the nucleotide sequences of the 250 bp regions flanking the contact sites. Each pair-sequence collected was shuffled to generate a pair of random sequences with the same nucleotide content as a control. The MFE between paired sequences was calculated using the DuplexFold program from the RNAstructure [58] software suite. The VARNA package [59] was used to visualize the base pairing and RNA secondary structures.

To assess the enrichment of TEs and STRs elements in E-P interaction pairs, genomic sites with transcription signals above 0.1 RPM were randomly selected as controls, given that approximately 90% of E-P interaction sites exhibited transcription signals greater than 0.1 RPM. Four randomised datasets, each consisting of 10,000 site pairs, were generated for comparison.

Analysis of GWAS combined with E-P interaction

Mutation data of cotton were downloaded from published data [30–34]. To integrate data from different reference genomes, BEDtools [60] was used to extract the 1 kb sequence flanking the mutation sites. The sequences were mapped to the Gossypium hirsutum genome using minimap2 [61] to obtain the location of the mutation site in the Gossypium hirsutum genome (Additional file 5: Table S4).

According to the regulation of eQTL sites and target genes known in the public database, chimeric reads were mapped to different ranges flanking the eQTL site and we calculated the percentage of cis/trans sites with target gene interactions supported by chimeric reads. To minimize interference from intramolecular interactions of genes, we opted to analyze only the 2 kb region extending in the direction away from the target gene body. To generate randomized control data, we first excluded target genes from the published data and randomly selected control genes from the remaining genes. Secondly, used the observed distribution range of eQTL loci around genes with eQTL associations to assign hypothetical eQTL loci to these control genes. To evaluate the enrichment of chimeric reads near GWAS loci, the percentage of mutations within a non-overlapping 50 nt window of a chimeric read contact site of ± 1 kb was calculated, and randomly selected loci in the GWAS locus deletion genomic region were used as random controls.

Transcription factor analysis

The information of transcriptional factors was downloaded from the PlantTFDB database [41], and potential binding sites of TFs on the reference genome were identified by the MCAST tool [57]. The enrichment of TF families was calculated by superimposing their signals at the E-P interaction site. The SNP number per kb in each transcription factor binding region was calculated and compared to the SNP number per kb in random regions of the genome.

Authors’ contributions

K.W., Y.Z., and M.W. conceived the work and designed the study. X.L., L.F., and K.S. performed the experiments. M.W. and Y.W. performed the bioinformatic analyses. K.W., M.W., and Y.Z. wrote the manuscript with input from other authors. All authors read and approved the final manuscript.

Funding

This work was supported by grants from National Key R&D Program of China (2022YFF1001400), National Natural Science Foundation of China (32070207), Natural Science Foundation of Hubei province (2024AFE007), Foundation of Hubei Hongshan Laboratory (2021hszd014), Hubei Provincial Natural Science Foundation of China (2024AFE008), Hubei Provincial Research Center of Basic Discipline of Biology to K.W.; China NSFC projects (82341023), the National Key R&D Program of China (2023YFC2307802), and the Fundamental Research Funds for the Central Universities (2042022dx0003) to Y.Z.

Data availability

The raw sequencing data from this study are deposited in the National Genomics Data Center (NGDC, https://ngdc.cncb.ac.cn/gsa) under the accession number CRA015576 of project PRJCA024429 [62]. DNase-seq data for both diploid and tetraploid cotton were downloaded from the European Nucleotide Archive (ENA) under BioProject accession number PRJEB47222 [55, 63]. RNA-seq and ATAC-seq data for tetraploid cotton were sourced from the NCBI Sequence Read Archive (SRA) under project PRJNA847210 [11, 64]. The source code for analyzing pRIC-seq data are available at Github (https://github.com/zhouyulab/pric) [65] under MIT license and at https://zenodo.org/records/17590989 in Zenodo [66].

Declarations

Ethics approval and consent to participate

Not applicable.

Competing interests

The authors declare no competing interests.