Pervasive RNA-binding protein enrichment on TAD boundaries regulates TAD organization

Abstract

Mammalian genome is hierarchically organized by CTCF and cohesin through loop extrusion mechanism to facilitate the organization of topologically associating domains (TADs). Mounting evidence suggests additional factors/mechanisms exist to orchestrate TAD formation and maintenance. In this study, we investigate the potential role of RNA-binding proteins (RBPs) in TAD organization. By integrated analyses of global RBP binding and 3D genome mapping profiles from both K562 and HepG2 cells, our study unveils the prevalent enrichment of RBPs on TAD boundaries and define boundary-associated RBPs (baRBPs). We found that baRBP binding is correlated with enhanced TAD insulation strength and in a CTCF-independent manner. Moreover, baRBP binding is associated with nascent promoter transcription. Additional experimental testing was performed using RBFox2 as a paradigm. Knockdown of RBFox2 in K562 cells causes mild TAD reorganization. Moreover, RBFox2 enrichment on TAD boundaries is a conserved phenomenon in C2C12 myoblast (MB) cells. RBFox2 is downregulated and its bound boundaries are remodeled during MB differentiation into myotubes. Finally, transcriptional inhibition indeed decreases RBFox2 binding and disrupts TAD boundary insulation. Altogether, our findings demonstrate that RBPs can play an active role in modulating TAD organization through co-transcriptional association and synergistic actions with nascent promoter transcripts.

Graphical Abstract

Figure 1.

TAD boundaries are hotspots for RBP binding. (A) Schematic illustration of the analysis pipeline. Various datasets were collected from both K562 and HepG2 cells and integrated to elucidate the role of RBPs in 3D genome organization. (B, C) Bar plot showing the percentages of RBP ChIP-seq peaks overlapping with TAD boundaries (left y-axis) or TADs (right y-axis) in K562 and HepG2 cells. CTCF, cohesin subunit RAD21 and SMC3 ChIP-seq peaks are served as positive controls. (D, E) RBP occupancy on specific states of ENCODE-annotated genome segmentation of TAD boundaries in K562 and HepG2 cells. R, repressive regions; PF, promoter flanking regions; T, transcribed regions; CTCF, CTCF-binding sites; WE, weak enhancers; E, enhancers; TSSs, transcription start sites/promoters. For each RBP, the relative distribution of its occupied sites on individual segments (vertical comparison) is color-coded (key on the left). For each class of segment annotation, the relative distribution of individual RBPs is represented by bubble size (horizontal comparison), as indicated by relative RBP occupancy. Right: summed percentage of individual segment annotations covered by the surveyed RBPs. (F, G) Meta-gene plots of representative RBPs (POLR2G, RBFox2, TOE1 in K562 and RBFox2, POLR2G and HNRNPLL in HepG2 cells) ChIP-seq signals around TAD boundaries. B, boundary. (H, I) Representative heatmaps showing the enrichment of RBP ChIP-seq signals on TAD boundaries in K562 and HepG2 cells. (J, K) Ranking of RBP ChIP-seq signals at TAD boundaries in K562 and HepG2 cells.

Figure 2.

Network interaction of baRBPs and TFs at TAD boundaries. (A) Schematic illustration of the analysis pipeline. baRBP and TF ChIP-seq datasets were collected from both K562 and HepG2 cells and integrated to perform clustering analysis by using NMF method. (B, C) Segregation of baRBP peaks residing at TAD boundaries into 5 and 4 groups by NMF-inferred coefficient matrixes in K562 and HepG2 cells. (D, E) Representative NMF-segregated groups in K562 and HepG2 cells. Highlighted line: known physical interactions between members in each group annotated by GeneMANIA. (F, G) Segregation of baRBPs and TF peaks residing at boundaries into 5 and 4 groups by NMF-inferred coefficient matrixes in K562 and HepG2 cells. (H, I) Meta-gene plot of the ChIP-seq signals around TAD boundaries of three representative TFs in K562 (ATF1, MAX, TFDP1) and HepG2 cells (ASH2L, KDM5A, PHF8). B, boundary.

Figure 3.

baRBP enrichment correlates with increased insulation strength of TAD boundaries. (A) ADA of TADs with (+) or without (−) representative baRBPs (RBFox2 and TOE1) boundaries in K562 cells. (B) Comparison of ISs of +baRBP and −baRBP boundaries. (C) Comparison of domain scores of TADs with +baRBP or −baRBP boundaries in K562 cells. (D) Aggregate analysis of interaction frequency around +baRBP and −baRBP boundaries in K562 cells. (E) Quantification of interaction frequency around +baRBP and −baRBP boundaries in (D). (F–  J) The above analyses were performed in HepG2 cells for two representative baRBPs (RBFox2 and HNRNPLL). (K, L) Representative heatmaps and ‌University of California Santa Cruz (UCSC) Genome Browser tracks showing the baRBP (RBFox2 in K562 and HNRNPLL in HepG2) ChIP-seq signals around TADs with −baRBP boundaries (left) or +baRBP boundaries (right).

Figure 4.

baRBPs may facilitate TAD organization independent of CTCF and cohesin binding. (A, B) Proportions of different types of boundaries in K562 and HepG2 cells. (C, D) Overlapping of CTCF and representative baRBPs (RBFox2 and TOE1 in K562; RBFox2 and HNRNPLL in HepG2) ChIP-seq peaks at boundary regions. (E) Comparison of ISs of +baRBP and −baRBP boundaries at CTCF present boundaries. (F) ADA of TADs with or without baRBP at CTCF present boundaries. (G) Comparison of domain scores between TADs with or without baRBP at CTCF present boundaries. (H, J) The above analyses were performed at CTCF absent boundaries. (K, L) Representative heatmaps and UCSC tracks showing baRBP (RBFox2), CTCF and cohesin ChIP-seq signals around CTCF present or CTCF absent boundaries in K562 and HepG2 cells.

Figure 5.

baRBP enrichment at TAD boundaries is correlated with active transcription. (A) Heatmaps showing the GRO-seq signals around TADs with (+) or without (−) baRBP (RBFox2 and TOE1 in K562, RBFox2 and HNRNPLL in HepG2) boundaries. (B) Comparison of GRO-seq signals at +baRBP and random regions (shuffled) in K562 and HepG2 cells. (C) Comparison of GRO-seq signals at TAD boundaries with low (L), medium (M) or high (H) levels of +baRBP signals in K562 and HepG2 cells. (D) Comparison of ISs of +baRBP boundaries with (+) or without (−) GRO-seq signals in K562 and HepG2 cells. (E) Comparison of ISs of +baRBP boundaries with different levels of GRO-seq signals in K562 and HepG2 cells. (F) Representative heatmaps and tracks showing TADs with +baRBP (RBFox2) boundaries with different levels of GRO-seq signals in K562 cells. (G) Pie charts showing the proportions of ChIP-seq peaks of boundary-located baRBP (RBFox2 and TOE1 in K562, RBFox2 and HNRNPLL in HepG2) overlapping with promoters (+Promoter). (H) Heatmap showing the distribution of GRO-seq signals around +baRBP (RBFox2 and TOE1 in K562, RBFox2 and HNRNPLL in HepG2) or randomly shuffled regions at + promoter boundaries. (I) Comparison of GRO-seq signals of baRBP binding sites and random region at + promoter boundaries in K562 and HepG2 cells. (J) Heatmaps showing the distribution of RBFox2 CLIP-seq signals around baRBP binding sites (RBFox2 in K562) or randomly shuffled region. (K) Representative tracks showing the overlapping of RBFox2 ChIP-seq and CLIP-seq signals in K562 cells. (L, M) The above analyses were performed on RBFox2 in HepG2 cells.

Figure 6.

RBFox2 binding promotes TAD organization in K562. (A) Schematic illustration of the experimental testing of RBFox2 knockdown effect on 3D genome in K562 cells. (B) RBFox2 knockdown efficiency was detected by RT-qPCR. (C) Percentage of +RBFox2 boundaries in siNC sample (left) and the percentage of disappeared or conserved boundaries in siRBFox2 cells (right). (D) Aggregate analysis of interactions around +RBFox2 boundaries. (E) Comparison of inter- and intra-TAD interactions around +RBFox2 boundaries in siNC and siRBFox2 cells. (F) Aggregate analysis of interactions around +RBFox2 −CTCF boundaries in siNC and siRBFox2 cells. (G) Comparison of inter- and intra-TAD interactions around +RBFox2 −CTCF boundaries in siNC and siRBFox2 cells. (H, I) The above analyses were performed on +RBFox2 +CTCF boundaries. (J, K) Two representative heatmaps displaying the decreased intra-TAD interactions upon RBFox2 depletion. (L) Schematic illustration of the role of RBFox2 on TAD boundary organization in K562 cells.

Figure 7.

RBFox2 regulation of 3D genome is relevant in mouse MB differentiation. (A) Schematic illustration of experimental testing of RBFox2 regulation of 3D genome during C2C12 MBs differentiation into MTs. (B) UCSC track showing the expression of RBFox2 in MB and MT. (C) Meta-gene analysis showing the enrichment of RBFox2 ChIP-seq signals at TAD boundaries in MBs. (D) Comparison of ISs of +RBFox2 and −RBFox2 boundaries in MBs. (E) Comparison of expression of genes residing in TADs with +RBFox2 or −RBFox2 boundaries. (F) Dynamic changes of +RBFox2 boundaries during MB to MT differentiation. (G) Comparison of ISs of +RBFox2 boundaries in MB versus MT. (H) Representative heatmap displaying the increased interaction around +RBFox2 boundaries in MT versus MB. (I) An RBFox2 KO C2C12 MB was generated by CRISPR-Cas9 mediated genome editing and the cells were cultured in differentiation medium for 0, 1, 3, 5 days, followed by RT-qPCR detection of the mRNA levels of Myogenin, MyHC, MCK and Tnni2. (J, K) IF staining of MyHC protein was performed on day 5 and the quantification of MyHC + MTs is shown. (L) Schematic illustration of the experimental testing of transcription effect on RBFox2 binding in MBs. (M) Comparison of ISs of +RBFox2 boundaries with or without GRO-seq signals in MB. (N) Heatmaps showing RBFox2 ChIP-seq signals at +GRO-seq and −GRO-seq sites. (O) Two representative heatmaps displaying the interaction and RBFox2 ChIP-seq signals at +GRO-seq and −GRO-seq sites. (P) Heatmaps showing RBFox2 ChIP-seq signals on chromatin in Ctrl- or ActD-treated MB. (Q) Pie chart showing the decreased number of RBFox2 ChIP-seq peaks in ActD versus Ctrl MB. (R) Representative tracks showing the significant decreased RBFox2 binding upon ActD treatment. (S) Pie chart showing the decreased number of +RBFox2 boundaries in ActD versus Ctrl MB. (T) Comparison of ISs of +RBFox2 boundaries in ActD versus Ctrl MB. (U) Heatmap showing the decreased RBFox2 binding and decreased interaction at RBFox2 binding site upon ActD treatment. (V) Schematic illustrating the role of RBFox2 on TAD organization during C2C12 MB differentiation into MT.