CTCF enhances pancreatic cancer progression via FLG-AS1-dependent epigenetic regulation and macrophage polarization

Abstract

CCCTC-binding factor (CTCF) regulates chromatin organization and is upregulated in pancreatic ductal adenocarcinoma (PDAC). We found that CTCF interacts with HNRNPU through a FLG-AS1-dependent mechanism, facilitating the recruitment of EP300 and activation of the m6A reader IGF2BP2. This activation promotes histone lactylation at the promoter region of IGF2BP2 stimulating the proliferation of PDAC cells. IGF2BP2 enhanced the mRNA stability of CSF1 and MYC. Moreover, FLG-AS1 directly interacts with HNRNPU to modulate alternative splicing of CSF1, thus promoting the M2 polarization of tumor associated macrophages (TAMs) in PDAC. The results indicated that CTCF-induced oncogenic modification of histone lactylation, m6A and alternative spilcing as multi-regulation modes of TAMs reprogramming in PDAC and identifies CTCF as a potential therapeutic target for PDAC immunotherapy whose inhibition M2 polarization through the IGF2BP2/CSF1/CSF1R axis. Curaxin combined with gemcitabine treatment has shown promising antitumor efficacy against PDAC.

Introduction

Pancreatic cancer is one of the most fatal types of alimentary tract malignancies and accounts for high cancer-related morbidity and mortality, with a 5-year survival rate of only 11% [1]. PDAC is a main type of pancreatic cancer characterized by a massive infiltration of tumor-associated macrophages (TAMs) [2]. Currently, the primary treatments for patients with PDAC are surgery and chemotherapy [3]. Nevertheless, the chances of curative resection are slim for patients diagnosed with PDAC at an advanced stage, and combination therapy based on gemcitabine provides only a marginal survival benefit [4]. Although immunotherapy appears to be one of the most promising strategies for carcinoma treatment [5], no suitable immunotherapy has shown effects against PDAC. This underlines the critical need for in-depth research on the role of immune response in promoting tumor cell endogenous events in PDAC.

The malignant progression of PDAC involves a range of epigenetic and epigenetic transcriptomic alterations [6]. In most cases, chromatin remodeling factors regulate gene expression via DNA methylation, histone modifications and chromatin insulators [7, 8]. CCCTC binding factor (CTCF) directly interacts with the promoters, insulators and silencers of oncogenes to enhance tumor proliferation, differentiation, and apoptosis [9, 10]. Our recent study showed that CTCF regulates the transcription of downregulated genes through altered histone modifications by binding to the promoter regions [11]. CTCF is abnormally expressed in multiple cancers [12, 13]. N6-methyladenosine (m6A) is one of the most common human RNA modifications in humans that controls mRNA splicing, translation and degradation [14]. M6A is dynamically modulated by a set of m6A enzymes, comprising writers (METTL3/METTL14), erasers (FTO and ALKBH5) and readers (IGF2BP1-3) [15]. Many studies have clarified the functional meanings of m6A regulators in malignant progression [16, 17]. In PDAC, we demonstrated the pro-tumor-intrinsic effect of CTCF-IGF2BP2-m6A axis on pancreatic cancer tumorigenesis.

Immunosuppression is an important hallmark of neoplasms [18]. The immunosuppressive tumor environment is characterized by the build-up of immunosuppressive cells, including TAMs and tumor-derived cytokines [19]. TAMs directly stimulate the proliferation of tumor cells by secreting cytokines and enhance tumor vascularization by secreting angiogenic stimulators [20]. TAMs also facilitate tumor progression by secreting regulatory cytokines such as TGFβ, IL-10 and ARG1 [21]. Higher infiltration of TAMs is correlated with worse prognosis of PDAC [22]. TAMs have been found to induce resistance to gemcitabine, the first-line chemotherapeutic drug for PDAC [23, 24]. Therefore, suppression of TAMs is a potential therapeutic strategy for PDAC treatment.

In our study, we revealed for the first time the connection between tumor intrinsic CTCF and polarization of TAMs in the PDAC microenvironment using in vitro and in vivo models of PDAC. We clarified the CTCF molecular mechanism of inducing the polarization of M2 macrophages by CTCF-IGF2BP2-CSF1 axis in PDAC to sustain tumor growth by combining ChIP-seq, m6A sequencing (m6A-seq), RIP-seq and RNA sequencing (RNA-seq). Finally, we emphasize that targeting CTCF (Curaxin) could improve the curative efficacy of gemcitabine chemotherapy in PDAC patients.

Results

CTCF is a tumor-promoting transcription factor and correlates with worse prognosis in PDAC

To characterize potential onco-transcription factors (oTFs) driving the progression of PDAC, we used three datasets: TCGA, single-cell RNA sequencing (CRA001160) and single-cell ATAC sequencing (GSE137069) for multi-omics data analysis [25, 26]. The genes with enriched binding motifs in accessible chromatin regions, upregulated in a subcluster of pancreatic ductal cells and associated with poor prognosis were considered to be candidate oTFs (Fig. 1A and Supplementary Fig. 1A). Based on this selection strategy, 8740 differentially expressed genes upregulated in PDAC were identified in the TCGA dataset, of which 739 upregulated differentially expressed genes had prognostic value (Fig. 1A, B). In addition, we screened 2519 genes up-regulated in PDAC tumor ductal cell clusters by single-cell RNA sequencing and scATAC-seq datasets, of which 746 transcription factors were notably enriched at highly accessible regions of PDAC ductal cell clusters and CTCF was the only oTF identified (Fig. 1A, C–G and Supplementary Fig. 1B–E). CTCF has been reported to be a multi-functional transcription factor involved in the regulation of chromatin structures in multi-cancers. However, its role in PDAC has not been fully defined [27, 28]. Therefore, we explored the clinical value of CTCF and elucidated its role in PDAC. Subsequently, we selected 46 paired PDAC tissues and para-cancerous tissues (Cohort 1) (Supplementary Table 1) for qPCR analysis, tissue microarrays (TMAs) from additional 110 PDAC patients (Cohort 2) (Supplementary Table 2) for IHC and IF and 6 paired PDAC tissues for WB. The analysis revealed that CTCF was upregulated in PDAC tissues compared with normal pancreatic tissues (Fig. 1H–J). To further clarify whether the higher expression of CTCF had any clinical relevance in PDAC, we divided patients from cohort 2 into the CTCF-High group and CTCF-Low group based on the IHC score of CTCF in PDAC tissues. Subsequent analysis of clinical follow-up data of patients between both groups indicated that upregulation of CTCF was correlated with worse prognosis in PDAC patients (Fig. 1K). In a nutshell, we validated that CTCF is overexpressed in PDAC as an oTF and is disadvantageous to patient survival.

Fig. 1. CTCF promotes the progression of pancreatic cancer.

A Schematic illustration of strategies for screening the key genes involved in the tumor progression of PDAC. B RNA-seq analysis of PDAC tissues and normal tissues. C scRNA-seq analysis of pancreatic ductal cells from PDAC revealed the differentially expressed genes. D, E Kaplan–Meier for CTCF based on the log-rank statistic test (p < 0.05). OS overall survival, DSS disease specific survival. F The expression of CTCF in PDAC pancreatic ductal cells and normal pancreatic ductal cells. G Distribution probability of CTCF binding motifs around PDAC pancreatic ductal cells scATAC-seq peak summits in Differential Accessible Regions (DARs). H Expression of CTCF in 46 pairs of tumor tissues and para-cancerous tissues from PDAC. I The CTCF level in PC tumor tissues and matched normal tissues verified by IHC and IF. Scale bar (IHC-left) = 500 μm. Scale bar (IHC-right) = 250 μm. Scale bar (IF) = 100 μm. J WB analysis demonstrated that CTCF is upregulated in clinical tumor samples (n = 6). K Prognostic analysis of CTCF in 110 cases of PC patients. L, M CCK8 was used to detect the cell viability of PCs at the indicated timepoint. N, O PCs were cultured in 6-well plates. Scale bar: 5 mm. P CTCF knockdown organoids of equal amounts of cells were taken pictures. Scale bar: 200 μm. Q The cell viability detected by the luminescence signal intensity. R Images of BALB/c nude mice which were PCs subcutaneously. S, T Tumor weights and volumes of the subcutaneous xenografts. U Kaplan–Meier survival curve presenting the overall survival of BALB/c nude mice (n = 6). V–X Representative images of IHC and IHC scores of CTCF and Ki-67 in tumor tissues from BALB/c nude mice. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

CTCF is a previously unidentified factor facilitating the proliferation of pancreatic cancer cells

Although CTCF is a prominent oncogene in many tumors, little is known about its function in PDAC progression. To investigate the pro-tumorigenic function of CTCF, PCs with CTCF knock down were prepared, and CCK-8 and colony formation assays were conducted to identify the role of CTCF in promoting PCs proliferation (Fig. 1L–O). To further confirm this observation, we constructed two organoids derived from separate surgical samples of PDAC patients with CTCF knock down (Supplementary Fig. 1F). Knockdown of CTCF decreased the viability of the organoids (Fig. 1P, Q). In nude subcutaneous and orthotopic xenograft mice models, we found that silencing CTCF diminished the tumor burden in PDAC and prolonged the median survival, whereas CTCF overexpression increased tumor burden and reduced median survival as indicated by results from nude subcutaneous and orthotopic xenograft mice models (Fig. 1R–U and Supplementary Fig. 1G–I). In addition, we performed WB and IHC staining assays using tumor tissues with CTCF antibody to verify the changes in CTCF levels in the respective groups (Fig. 1V, W and Supplementary Fig. 1J–L). CTCF depletion decreased PCs proliferation as indicated by Ki-67 staining of harvested tumors and vice-versa, suggesting that CTCF regulates the proliferation of PCs (Fig. 1V, X and Supplementary Fig. 1L). Taken together, these results indicate that CTCF acts as a carcinogenic factor to augment the growth of PCs.

Non-canonical function of CTCF in activating IGF2BP2 transcription through recruitment of HNRNPU

After establishing that CTCF promotes the proliferation of PCs, our subsequent objective was to delve into the mechanism of CTCF in PDAC. Given that previous studies have highlighted CTCF’s pivotal role in organizing chromatin into TAD structures and its potential as a transcription factor [29], we aimed to identify specific CTCF binding sites through transcriptome sequencing and Chromatin Immunoprecipitation sequencing (Fig. 2A). Through RNA-seq and CTCF ChIP-seq assays, we identified 766 differentially expressed genes (308 up-regulated and 458 down-regulated) between PANC-1 shCTCF and PANC-1 shNC and 3235 genes in the near CTCF binding sites. In the combined analysis, we identified IGF2BP2, NES, JAG1, TGFBI and ANKRD1 as potential downstream genes of CTCF (Fig. 2A, B). Subsequently, to further investigate the potential regulation of these genes by CTCF, we revealed that these genes were down-regulated in CTCF-deactivated PCs and up-regulated in CTCF-high-expressing PCs by qPCR and WB assays. (Fig. 2C, D and Supplementary Fig. 2A–D). Notably, knockdown of IGF2BP2, but not NES, JAG1, TGFBI or ANKRD1, blocked the CTCF-KD-mediated proliferation of PCs (Supplementary Figs. 2E–O and 3A–E), indicating that IGF2BP2 might be involved in CTCF-regulated PDAC proliferation. Further, we will explore the regulatory role of CTCF on IGF2BP2 expression. To investigate the role of CTCF in transactivation of IGF2BP2, CTCF was found to notably bind to the promoter region of IGF2BP2 in PCs (Fig. 2E) and CTCF augmented the luciferase activity of the IGF2BP2 promoter reporter (Fig. 2F). Furthermore, we investigated the precise sites of CTCF binding to the IGF2BP2 promoter. Through deletion analysis, we discovered that CTCF failed to activate the IGF2BP2 promoter when the −1550 to −1498 region was removed (Fig. 2G). Subsequently, we constructed three mutant reporters with altered binding motifs (Mutation of Site 1, Mutation of Site 2, and Mutation of both 2 sites) for luciferase detection to elucidate that Site 1 is a key site for CTCF binding to the promoter of IGF2BP2 (Fig. 2H). Previous studies have reported that CTCF is often involved in the transcriptional regulation of downstream genes as a member of the transcription complex. Therefore, we investigated whether CTCF forms a complex with other transcriptional regulatory proteins to regulate the IGF2BP2 expression. Hence, we explored the mechanism of CTCF activation of IGF2BP2 transcription by analyzing proteins interacting with CTCF using co-immunoprecipitation (co-IP) and mass spectrometry. Following a comprehensive screening of mass spectrometry results, we selected HNRNPU, a well-known RNA-binding protein, that can be involved in alternative splicing of RNA as a potential CTCF-interacting protein and further verified this conjecture by co-IP assays in the nucleus (Fig. 2I–K). Meanwhile, the confirmation was supported by ChIP-qPCR and dual-luciferase reporter assays (Fig. 2L, M). To evaluate the key structural domains of CTCF involved in binding with HNRNPU, we conducted molecular dynamics simulations, which revealed strong interactions (102 combinations) between the DBR + SR2 structural domain of CTCF and the SAP domain of HNRNPU (Fig. 2N). These simulations showed the spatial proximity of the amino acids involved in the CTCF-HNRNPU interaction (highlighted in red). Subsequently, through co-IP assays with truncated versions of each protein’s structural domains, we found that the interaction between CTCF and HNRNPU exclusively involved the DBR + SR2 domain of CTCF and the SAP domain of HNRNPU (Fig. 2O, P). Thus, it is curious whether the binding of HNRNPU to CTCF would affect CTCF to enhance the malignant progression of PDAC. Knockdown of HNRNPU in PCs overexpressing CTCF downregulated IGF2BP2, reducing the proliferation of CTCF-OE+si-HNRNPU PCs (Supplementary Fig. 4A–D). We further found that CTCF had a high co-localization relationship with HNRNPU in PDAC tissues (Supplementary Fig. 4E, F). Mechanistically, we found that CTCF was positively correlated with the HNRNPU and IGF2BP2 was associated with worse prognosis in PDAC based on RNA expression levels in TCGA-PAAD cohort and validation cohort 1 (Supplementary Fig. 4G–N). Taken together, these data indicate that the DBR + SR2 structural domain of CTCF recognizes and interacts with the SAP structural domain of HNRNPU to facilitate IGF2BP2 transcription.

Fig. 2. CTCF recruits HNRNPU to activate IGF2BP2 transcription.

A Searching for a potential CTCF-guided promoter locus. Left: RNA-seq screening of downregulated genes in PCs (shNC/sh-CTCF); Right: Locating the wild-type CTCF-binding site in PCs. Genes selected should meet: (1) downregulated in PCs; (2) with CTCF bind in the promoter. B Top five related CTCF downstream target genes in the heatmap of mRNA-seq. C CTCF binding sites at the region of IGF2BP2 were tracked with the IGV browser. D IGF2BP2 protein expression in PCs (shNC and shCTCF), examined by Western blot. E Association of CTCF with the promoter region of IGF2BP2 in PANC-1 (NC/CTCF-OE /sh1 CTCF/sh2 CTCF) analyzed by ChIP-qPCR. F, G HEK293T cells were co-transfected with IGF2BP2 promoter–luciferase truncations and CTCF or CTCF-del plasmids, and the luciferase activity was determined using a dual luciferase reporter assay after 48 h. H Dual luciferase assay of HEK293T cells co-transfected with firefly luciferase constructs containing the wild-type or mutant CTCF potential binding sites of IGF2BP2 promoter and CTCF plasmids were performed. I Visualization of protein bands by CTCF antibody incubated with total protein extracts from PANC-1 cells. J The sequence of the HNRNPU peptide after mass spectrometry. K Results of coimmunoprecipitation (Co-IP) in PANC-1. Normal rabbit IgG was used as a negative control. L Association of HNRNPU with the promoter region of IGF2BP2 in PANC-1 (NC/CTCF-OE /sh1 CTCF/sh2 CTCF) analyzed by ChIP-qPCR. M HEK293T cells were co-transfected with IGF2BP2 promoter–luciferase truncations and HNRNPU or HNRNPU-del plasmids, and the luciferase activity was determined using a dual luciferase reporter assay. N The simulated interaction diagram of CTCF and HNRNPU. Ability of the indicated Flag-labeled CTCF (O) or Flag-labeled HNRNPU (P) derivatives to co-immunoprecipitate HNRNPU (O) or CTCF (P) defined by immunoblotting in HEK293T cells.

CTCF activates histone lactylation modifications through FLG-AS1-mediated recruitment of EP300

Many reports have shown that long non-coding RNAs play a crucial role in CTCF-mediated gene transcription [30, 31]. Therefore, we postulated that some LncRNAs might be involved in the mechanism by which CTCF enhances the transcription of IGF2BP2 in PDAC. To search for the LncRNAs interacting with CTCF/HNRNPU complex, we performed RIP-seq assays using CTCF and HNRNPU antibodies and determined that FLG-AS1 was substantially bound to the CTCF/HNRNPU complex (Fig. 3A). RIP-qPCR further confirmed that CTCF and HNRNPU were enriched by FLG-AS1. And peaks of CTCF and HNRNPU were detected in the IGF2BP2 promoter in PCs (Fig. 3B). To determine the reliability of this binding mechanism, we found that CTCF and HNRNPU enrichment was detected in the complexes pulled down by FLG-AS1 but not in the negative control samples pulled down by FLG-AS1 beads only by using RNA pull-down assay (Fig. 3C). To gain a comprehensive understanding of the CTCF/HNRNPU/FLG-AS1 complex, we first separately predicted the binding regions of FLG-AS1-CTCF and FLG-AS1-HNRNPU using catRAPID [32] (Fig. 3D, E), and constructed plasmids with the respective regions truncated (Fig. 3F). Thereafter, deletion of the mapped biotin-labeled RNA pull-down showed that the region 1# of FLG-AS1 was necessary for its interaction with CTCF/HNRNPU complex (Fig. 3G, H). We further explored the functional domains of CTCF/HNRNPU that binds to FLG-AS1. The results indicated that the RBR region of the FLAG-tagged CTCF protein, and the RGG region of the FLAG-tagged HNRNPU protein played crucial role in interactions with FLG-AS1 as confirmed by RIP-qPCR (Fig. 3I, J). Concomitantly, we explored the intrinsic connection between FLG-AS1 and CTCF/HNRNPU complex. Co-IP and WB assay showed that knockdown of FLG-AS1 attenuated the binding of CTCF to HNRNPU in PCs, whereas overexpression of FLG-AS1 enhanced their interaction (Fig. 3K, L).

Fig. 3. CTCF activates histone lactylation modifications via FLG-AS1-mediated recruitment of EP300.

A RIP-seq screening of long non-coding RNAs which interact with CTCF and HNRNPU in PANC-1 cells. B RIP-qPCR analysis of FLG-AS1 enriched by CTCF and HNRNPU in PANC-1 cells (top). Genome browser snapshots of RIP‐seq signals for the genomic regions near FLG-AS1 in PANC-1 cells (bottom). C Immunoblotting to determine the specific association of HNRNPU and CTCF with biotinylated FLG-AS1. The prediction binding sites between FLG-AS1 and HNRNPU (D) and CTCF (E) by catRAID. F Secondary structure of FLG-AS1 analyzed by RNAfold web server and deletion mapping of biotinylated FLG-AS1 motifs, as indicated. The red boxes represent the remaining fragments of FLG-AS1, with the corresponding number label in the corner. Immunoblot showing the association of CTCF (G) or HNRNPU (H) with biotinylated FLG-AS1 RNA strands and the above-mentioned biotinylated FLG-AS1 motifs. RIP analysis for FLG-AS1 enrichment in HEK293T cells transfected with the FLAG-tagged full-length or truncated CTCF (I) and HNRNPU (J) constructs (n = 3). aa: amino acid. K, L Co-IP and Western blot assays indicating the interaction among HNRNPU, CTCF and EP300 in PANC-1 cells stably transfected with mock, FLG-AS1, shFLG-AS1. Association of EP300 with the promoter region of PANC-1 cells (shNC/shCTCF/shCTCF + OE-HNRNPU/shHNRNPU/shHNRNPU + OE-CTCF) (M) and PANC-1 cells (NC/OE-FLG-AS1/sh-FLG-AS1) (N) analyzed by ChIP-qPCR. Association of H3K27ac (O) and H3K18la (P) with the promoter region of PANC-1 cells (shNC/shCTCF/OE-CTCF/OE-EP300) analyzed by ChIP-qPCR.

Interestingly, HNRNPU has an EP300 binding domain [33], and many studies have reported that CTCF can act as a transcriptional activator by recruiting EP300 [34, 35]. Therefore, we investigated that the CTCF/HNRNPU complex could recruit EP300 in nucleus using co-IP (Fig. 3K, L). In addition, we speculated whether FLG-AS1 could affect the recruitment of EP300 to the CTCF/HNRNPU complex. Results of the Co-IP assay confirmed that FLG-AS1 could enhance EP300 enrichment in CTCF/HNRNPU complex (Fig. 3K, L). Subsequently, to explore whether CTCF/HNRNPU is required for the recruitment of EP300 to IGF2BP2 promoter, we conducted ChIP assays which indicated that EP300 level was notably decreased after knockdown of CTCF or HNRNPU but remained unchanged after overexpression of HNRNPU in CTCF-knockdown PCs or overexpression of CTCF in HNRNPU-knockdown PCs (Fig. 3M). Moreover, the enrichment of EP300 after the knockdown of FLG-AS1 was also decreased but increased by overexpression of FLG-AS1 (Fig. 3N), suggesting that CTCF/HNRNPU plays a role in the recruitment of EP300 to IGF2BP2 promoter and FLG-AS1 enhanced the enrichment of EP300 at this site.

EP300 was commonly thought to increase chromatin accessibility by enhancing histone acetylation modifications in the promoter region of IGF2BP2 [36]. However, ChIP-qPCR assays showed that overexpression of EP300 or disruption of the CTCF/HNRNPU complex did not change the level of histone acetylation modification at the promoter of IGF2BP2 (Fig. 3O). In contrast, recent related studies have reported that EP300 may be involved in the regulation of histone lactylation modifications to remodel chromatin accessibility, but few studies have been conducted in the oncology [37, 38]. Therefore, we hypothesized whether CTCF was involved in the transcription by altering the level of histone lactylation modification at the promoter of IGF2BP2. Accidentally, this mode did not change the protein levels of EP300, H3K27ac and H3K18la in PCs (Supplementary Fig. 5A–D). We observed that the level of H3K18la decreased disruption of the CTCF/HNRNPU complex but increased after overexpression of EP300 (Fig. 3P and Supplementary Fig. 5E, F). Furthermore, we found that the proliferation of CTCF-OE + siEP300 PCs and CTCF-OE + siFLG-AS1 PCs was lower compared with CTCF-OE PCs (Supplementary Fig. 5G–L). To highlight the importance of CTCF/HNRNPU/EP300-mediated transcription activation in PDAC, we engineered a mutant plasmid in the DBR + SR2 region and found that DBR + SR2 domain can impair the assembly of the CTCF/HNRNPU/EP300 complex, leading to decreasing expression and promoter lactylation of IGF2BP2 (Supplementary Fig. 6A–D). These results suggested that the CTCF/HNRNPU complex enhanced H3K18la level through FLG-AS1-mediated recruitment of EP300 to activate IGF2BP2 transcription.

IGF2BP2 strengthens the mRNA stability of CSF1 in an m6A-dependent manner

Since IGF2BP2 was proved to be transcriptionally regulated by the CTCF/HNRNPU/FLG-AS1 complex, we shall further discuss the mechanisms by which IGF2BP2 regulates downstream targets. As an m6A reader, IGF2BP2 recognizes and interacts with the “GGAC” sequence to increase RNA stability [39]. METTL3 and METTL14 are the main methyltransferases of m6A methylation [40]. IGF2BP2 enhanced RNA stability mainly by interacting with methylation sites written by METTL3 or METTL14 [41]. To observe the potential mRNA targets of IGF2BP2 regulated by CTCF, we first performed m6A-seq in PANC1-NC and PANC1-M3-M14-KD cells and detected 13034 and 9317 m6A peaks in PANC1-NC and PANC1-M3-M14-KD cells, which were primarily located in the coding sequence region and 3′ untranslated region, respectively (Fig. 4A–C). Since METTL3 and METTL14 majorly boost the m6A modification of RNA, our focus shifted to investigating down-regulated m6A peaks after knockdown of METTL3 and METTL14 in PANC-1 (Fig. 4D). We observed that the functions of transcription products were primarily enriched in macrophage polarization (Fig. 4E). Then, we conducted RNA-seq in PANC-1-NC and PANC-1-IGF2BP2 KD to discover that the differentially expressed genes were mainly involved in immune response and cytokine-mediated signaling pathways (Fig. 4F). To determine the critical targets of IGF2BP2 mediating anti-tumor immunity, we first clarified that the number of genes near the differential m6A modification site between PANC-1 shNC and PANC-1 shM3-M14 was 807. Combining the differentially expressed genes between PANC-1 shNC and PANC-1 shCTCF, and between PANC-1 shNC and PANC-1 shIGF2BP2, we screened a total of 5 potential downstream genes that may be involved in regulation by the CTCF-IGF2BP2 axis, and we finally chose CSF1 and MYC with the most substantial differences as candidate genes for further exploration (Fig. 4A). To further validate these targets, we evaluated the m6A, mRNA, and protein levels of CSF1 and MYC following shM3-shM14, shIGF2BP2, and shCTCF in PANC-1 (Fig. 4G and Supplementary Fig. 7A–N). We observed that the expression of CSF1 and MYC was significantly lower in the shCTCF group compared to the shNC group. Although these two genes were not among the top five most significantly altered genes identified in the previous analysis, their potential biological relevance warrants further investigation (Supplementary Fig. 7C). As anticipated, the m6A levels were downregulated in the shM3-shM14 group compared to the shNC group, and the expression of CSF1 and MYC diminished in the shIGF2BP2 group or shCTCF group compared to the shNC group (Fig. 4G and Supplementary Fig. 7A–K). Meanwhile, CTCF did not affect the expression of METTL3 or METTL14, nor did METTL3 and METTL14 influence the expression of CTCF. This suggested that CTCF may enhance the recognition of m6A sites primarily by upregulating IGF2BP2 expression (Supplementary Fig. 7D, K, L). To validate the m6A methylation status of CSF1 and MYC, we investigated the transcripts of CSF1 and MYC in shM3-shM14 PCs. We detected lower levels of CSF1 and MYC transcripts by examining m6A methylation in the CRD region of CSF1 and MYC (Fig. 4H) and revealed that IGF2BP2 knockdown in PCs abrogated the interaction of IGF2BP2 with the m6A modification region of CSF1 and MYC (Fig. 4I). Subsequently, to verify the interaction between IGF2BP2 and CSF1 or MYC, we constructed luciferase reporters for the CRD of CSF1 and MYC and confirmed that IGF2BP2 was enriched in wild-type reporter using RIP-qPCR assays (Fig. 4J, K). Notably, several studies have reported that IGF2BP2 could enhance its role in maintaining RNA stability [42, 43]. Then, to explore the mediatory role of IGF2BP2 in CSF1 and MYC mRNA stability, we found that IGF2BP2 knockdown diminished the stability of CSF1 and MYC by using actinomycin D (Fig. 4L, M). To confirm the tumor-promoting role of IGF2BP2 regulated by CTCF, we examined the stimulation of siCTCF + IGF2BP2-OE PCs proliferation following IGF2BP2 overexpression in CTCF-interfered cells (Supplementary Fig. 7O–Q). To further validate the important role of CTCF in pancreatic cancer cells, we analyzed scRNA-seq data. In addition, we classified CTCF-positive subcluster and CTCF-negative subcluster based on the expression level of CTCF in ductal cell clusters and regarded CTCF, HNRNPU and IGF2BP2 as a signature, and categorized them into signature-high subcluster and signature-low subcluster based on the expression level of signature (bounded by the median of the average expression level of the three markers) in the ductal cell clusters. Subsequently, GO analysis revealed that the functions of the CTCF-positive subcluster and the signature-high subcluster were mainly enriched in mRNA splicing, cell proliferation, and immune cell activation (Fig. 4N, O). Taken together, these data suggested that IGF2BP2, mediated by CTCF, reinforces the mRNA stability of CSF1 and MYC in PDAC.

Fig. 4. FLG-AS1 interacts with IGF2BP2 to enhance the mRNA stability of CSF1 in an m6A-dependent manner.

A Flow chart of m6A-seq in PANC-1 and RNA-seq of PANC-1 (shNC/shIGF2BP2) and PANC-1 (shNC/shCTCF). B The normalized distribution of m6A peaks and identified m6A motif. UTR untranslated region. C Overall m6A peaks identified by m6A-seq of PANC-1 with or without METTL3-METTL14 knockdown. D A total of 978 reduced m6A peaks was found in PANC-1 after knockdown of IGF2BP2. E GO analysis of these potential m6A related IGF2BP2 targets. F Pathway analysis of these potential IGF2BP2 targets. G m6A-seq reads along indicated mRNAs. Ranges of reads are indicated. H RIP showing the enrichment of m6A modification in the CSF1 and MYC in the METTL3 KD and METTL14 KD PCs (n = 3). I RIP detecting the enrichment of IGF2BP2 in the CSF1 and MYC in IGF2BP2 KD PCs (n = 3). J Schematic representation of WT and mutated CRD of the pmirGLO vector (CSF1 and MYC). K RIP detection of the enrichment of IGF2BP2 in the CRD WT and MUT luciferase reporters of CSF1 and MYC (n = 3). L, M Half-life of CSF1 and MYC after treatment with actinomycin D for the indicated times. GO analysis of potential CTCF related genes between CTCF+ and CTCF− (N) pancreatic ductal cells or signature score (CTCF, HNRNPU and IGF2BP2) high and signature score low (O) pancreatic ductal cells through scRNA-seq.

HNRNPU collaborates with FLG-AS1 to control exon skipping of CSF1 in PCs

We performed scRNA-seq for functional enrichment of differential genes between CTCF- positive ductal cells and CTCF- negative ductal cells to find that the CTCF/HNRNPU-IGF2BP2 axis may be potentially altered splicing levels during pro-carcinogenesis, and HNRNPU has been reported as a notable variable splicing regulatory factor [44], so we were further curious whether HNRNPU as a partner protein of CTCF might have an extra molecular mechanism in PDAC progression mediated by CTCF. To comprehensively analyze the mechanisms by which CTCF/HNRNPU complex mediates PCs proliferation, we identified 4505 distinct AS events and analyzed PSI transitions after transcriptome sequencing (Fig. 5A–D). Among them, exon skipping was the most frequent AS event (Fig. 5B, C). To speculate on the related functional changes induced by exon skipping caused by HNRNPU, we revealed that RNA methylation and cytokine regulation were enriched by GO analysis of genes with exon skipping (Fig. 5E). Notably, there were high PSI values for exon skipping of CSF1 in shHNRNPU group (Fig. 5F). Consequently, we delved deeper into the contribution of HNRNPU to PC progression by investigating the splicing patterns of CSF1 that could be altered by HNRNPU. Specifically, we found that HNRNPU influenced the splicing patterns, resulting in CSF1 exon 2 skipping, as depicted in Fig. 5G. We performed the expression of isoforms of CSF1 after knockdown of HNRNPU in PANC-1 and found that CSF1-L was down-regulated in shHNRNPU group (Supplementary Fig. 8A). To support our findings, we conducted RIP assays, demonstrating the cooperation between HNRNPU and CSF1 (Fig. 5H). Then, the CLIP-qPCR assays (Fig. 5I) indicated that HNRNPU could bind to the exon 2 of CSF1 (Fig. 5J). To provide more evidence of interaction between HNRNPU and CSF1, we generated decoy RNAs and observed that oligo 2 of CSF1 displayed the strongest binding to HNRNPU (Fig. 5K). Since FLG-AS1 was previously found to interact with HNRNPU, we were curious whether FLG-AS1 would be involved in HNRNPU regulation of AS. The binding of HNRNPU to CSF1 was abrogated after knockdown of FLG-AS1 in PCs by RIP-qPCR analysis (Fig. 5L). To validate this result, RT-PCR was performed to investigate the AS changes of CSF1 after knockdown of FLG-AS1 or HNRNPU in PCs (Fig. 5M). To explore FLG-AS1 as a pivotal factor in mediating the AS pattern of CSF1, we confirmed that skipping of CSF1 exon 2 was up-regulated after knockdown of FLG-AS1 in PCs. Similar results were obtained for the knockdown of HNRNPU, implying that FLG-AS1 may associate with HNRNPU to inhibit exon skipping in CSF1 (Fig. 5M). In addition, we transfected CTCF overexpression plasmid in PCs and observed that CTCF did not play a role in HNRNPU-FLG-AS1-mediated exon skipping of CSF1 (Fig. 5N). Further, we aimed to ascertain the pro-tumor function of CSF1-L. As expected, we showed that the CSF1-L group displayed a greater cell proliferative capability compared with the CSF1-S and control groups (Supplementary Fig. 8B–E). To observe the level of CSF1-L and CSF1-S, we detected the expression of CSF1-L and CSF1-S in 21 paired PDAC tissues from Ruijin Hospital and found that CSF1-L was highly expressed in tumor tissue, while level of CSF1-S was not different in tumor and paracancer tissues (Supplementary Fig. 8F, G). The higher expression of CSF1-L was associated with worse prognosis in PDAC patients (Supplementary Fig. 8H). In addition, the expression of CSF1-L was decreased in PANC-1 shCTCF or PANC-1 shIGF2BP2 compared to PANC-1 shNC (Supplementary Fig. 8I, J). Interestingly, we found that knockdown of HNRNPU in pancreatic cancer cell lines reduced the stability of CSF1, while overexpression of IGF2BP2 after knockdown of HNRNPU could improve the stability of CSF1 but not to the level of shNC (Fig. 5O). Similar experiments potentially illustrate that exon skipping of CSF1, which was regulated by HNRNPU, was recognized by IGF2BP2 to enhance its mRNA stability (Fig. 5P). To sum up, HNRNPU cooperates with FLG-AS1 to inhibit exon skipping of CSF1 in PCs to exert pro-oncogenic function.

Fig. 5. HNRNPU interacts with FLG-AS1 to control exon skipping of CSF1 in PCs.

A A schematic diagram shows the experimental design of transcriptome sequencing in PCs. B The constituent ratios of different AS events were shown in two groups. C Genes with different AS patterns were represented according to the type of AS events. D The changes of PSI of different AS events were shown in two groups. E GO analysis of genes with shift of exon skipping in two groups. F Analysis of △PSI in two groups: X axis represents Difference: Difference expression of Exon Inclusion Isoform between shNC and shHNRNPU. Y axis represents FDR. G Schematic diagrams show the alternative splicing of CSF1 by exon skipping. H, L RIP assays were performed to confirm the interaction between HNRNPU and pre-mRNA of CSF1. n = 3 for each group; data are shown as mean ± SD from three independent experiments. *p < 0.05; **p < 0.01; ***p < 0.001, between the indicated groups. I Schematic diagrams show the CLIP-qPCR. J CLIP assays were performed in PANC-1 cells to confirm the interaction between HNRNPU and the region of CSF1. K A series of bait-oligo were constructed, and RNA pull down assays were conducted to identify the site of CSF1 that HNRNPU interacted with in PANC-1 cells. M, N Representative images and quantification of the relationships between the HNRNPU/FLG-AS1 complex and AS patterns of CSF1 are presented (M). Gray value ratio of CSF1-L to CSF1-S in the same sample (N). O, P Half-life of CSF1 after treatment with actinomycin D for the indicated times in the shNC/sh-HNRNPU/sh-HNRNPU + IGF2BP2 OE PCs and NC/HNRNPU OE/ HNRNPU OE+sh-IGF2BP2 PCs.

CTCF drives M2 macrophage polarization via CSF1-CSF1R pathway to facilitate PDAC proliferation

Since we previously found that IGF2BP2 stabilizes CSF1 mRNA, and CSF1 is a secreted cytokine that can induce the polarization of M1-like macrophages into M2-like macrophages through binding to CSF1R, we hypothesized that CTCF in pancreatic cancer cells may alter the function of TAMs. To investigate the role of CSF1 in tumor cell proliferation, we established a co-culture model of tumor cells and PBMC-derived macrophages. We found that reducing CSF1 expression significantly inhibited tumor cell growth (Supplementary Fig. 9A–D); however, this effect was not observed in tumor cells cultured without macrophages (Supplementary Fig. 9E–H). Similarly, overexpression of CSF1 in tumor cells within the co-culture system markedly accelerated proliferation, an effect that was attenuated by the addition of a CSF1R inhibitor (Supplementary Fig. 9I–L). Notably, no such proliferation changes were detected in the absence of macrophages (Supplementary Fig. 9M–P). These findings suggest that the pro-proliferative effect of CSF1 on tumor cells is mediated through macrophages. Although recent studies have shown that macrophage infiltration in PDAC is not always detrimental, we found that the percentage of CD206-positive M2-type macrophages was significantly positively correlated with tumor stemness (Supplementary Fig. 10A, B), and IHC staining showed that patients with PDAC expressing high levels of CD68 had a worse prognosis (Supplementary Fig. 10C, D). We further confirmed that nearly all PDAC-associated TAMs are M2-like macrophages (Supplementary Fig. 10E, F). Then, we observed that CTCF-positive subcluster and signature-high subcluster had more abundant cell-cell interactions by CellPhone DB analysis, in which the interactions between CTCF-positive subcluster or signature-high subcluster and macrophages were particularly notable, mainly via the CSF1-CSF1R axis (Fig. 6A, B). We further investigated the ability of PCs in CTCF-expressing PDAC microenvironment to promote M2 macrophage polarization. In vitro, CTCF overexpression increased polarized M2 macrophages, but this effect could be blocked by CSF1 receptor inhibitors (CSF1Ri) (Pexidartnib) (Fig. 6C–I). To observe the formation of CTCF-induced pro-tumor macrophages, we revealed elevated CTCF expression in PCs, leading to an upregulation of the proportion of CD206⁺ and CD163⁺ cells and a downregulation of the ratio of CD80⁺ and CD86⁺ cells by flow cytometry. In contrast, the addition of CSF1Ri in PBMC-derived macrophages decreased the number of CD206⁺ and CD163⁺ cells but increased the amount of CD80 and CD86 macrophages (Fig. 6D, E and Supplementary Fig. 10G–J). Besides, IL-10, IL-6 and TGF-β detected by ELISA assay were enhanced in the CTCF-OE group, whereas IL-10, IL-6 and TGF-β were reduced in the CTCF-OE-CSF1Ri and CTCF-NC-CSF1Ri groups (Fig. 6F, G). Moreover, we demonstrated that CD163, CD206, TGF-β, Arginase-1, IL-6 and IL-10 were upregulated, whereas CD80 was downregulated after overexpression of CTCF in PCs by detecting RNA levels. The addition of CSF1Ri in PBMC-derived macrophages produced an opposite effect on, while the expression of these genes (Fig. 6H, I). To verify that the promotion of CTCF on polarization of macrophages relies on CSF1-CSF1R crosstalk, we constructed orthotopic xenograft mice and found that tumor weight and percentage of CD206⁺ and CD163⁺ cells were increased whereas the proportion of CD80⁺ and CD86⁺ cells was decreased in the CTCF-OE group, and mice in the CTCF-OE group had a worse prognosis. However, the CSF1Ri treatment reduced the tumor size and improved the prognosis of mice (Fig. 6J–L). In addition, there are the same changes for M1 and M2 macrophage marks in TAMs isolated from PDAC tissues in vivo (Supplementary Fig. 11A–G). In the co-culture model of PancO2 NC/CTCF-OE and BMDMs, the addition of a CSF1R inhibitor significantly suppressed M2 macrophage polarization (Supplementary Fig. 11H, I). Similarly, macrophage scavengers effectively abolished macrophages in mice, resulting in effects similar to those of CSF1Ri, which inhibited M2 macrophage polarization mediated by CTCF⁺-PCs (Fig. 6J, M, N and Supplementary Fig. 12A–C). Furthermore, to clarify that IGF2BP2, which is regulated by CTCF in PCs, could stimulate macrophage polarization through CSF1, we first revealed that the polarization of M2 macrophages was down-regulated in the sh-IGF2BP2 group and up-regulated in the IGF2BP2-OE group by co-culturing PCs knocking down or overexpressing IGF2BP2 with PBMC-derived macrophages (Supplementary Fig. 13A–G). Then, to confirm that the oncogenic function of IGF2BP2 largely relies on CSF1, we constructed seven groups of PCs (NC, shIGF2BP2, CSF1, shIGF2BP2 + CSF1, shCSF1, IGF2BP2, shCSF1 + IGF2BP2). After PBMC-derived macrophages co-cultured with PCs, the proliferation ability of PCs (Supplementary Fig. 14A–H) and M2 macrophage polarization of PBMC-derived macrophages (Supplementary Fig. 14I–T) were decreased in the shIGF2BP2 and shCSF1 groups, while the opposite was obvious in the CSF1 group and IGF2BP2 group. shIGF2BP2 + CSF1 group had similar proliferation (Supplementary Fig. 14C, D, G, H) and M2 macrophage polarization (Supplementary Fig. 14I, J, O–R) as the CSF1 group. And shCSF1 + IGF2BP2 group had comparable results with shCSF1 group (Supplementary Fig. 14A, B, E, F, K–N, S, T). Since our preliminary study found that the DBR + SR2 domain of CTCF is critical for its binding to HNRNPU and that HNRNPU can assist CTCF in recruiting EP300, we constructed PCs transfected with DBR + SR2 mutant plasmid and revealed that mutation of DBR + SR2 domain could make PCs impaired in promoting macrophage polarization (Supplementary Fig. 15A–F). Collectively, our finding suggests that CTCF in PCs could induce pro-tumor macrophage polarization via CSF1-CSF1R crosstalk.

Fig. 6. CTCF drives M2 macrophage polarization via the CSF1-CSF1R pathway to facilitate PDAC proliferation.

A The crosstalk among all cluster cells. B Dot plots showing mean strength of selected ligand-receptor pairs of macrophages and CTCF+ (up) or signature high (down) pancreatic ductal cells or CTCF− (up) or signature low (down) pancreatic ductal cells. Dot size indicates the mean strength level, and color shows the p value calculated by the Permutation test. C Flow chart of co-culture model of PCs and PBMC-derived macrophages. D, E Percentage of CD206+ or CD163+ cells in PBMC-derived TAMs. F, G ELISA analysis of IL-6, TGF-β and IL-10 in PBMC-derived TAMs. H, I qPCR analysis of the relative expression of M2 markers (Arginase-1, CD163, TGF-β, CD206, IL-10 and IL-6) and M1 marker (CD80) in PBMC-derived TAMs. J The indicated PancO2 cells were orthotopically transplanted into C57BL/6 mice that were treated with CSF1Ri or Clodronate Liposomes. Tumor weight (K) and overall survival (L) of PancO2 NC and PancO2-CTCF allografts treated with CSF1Ri. Tumor weight (M) and overall survival (N) of PancO2 NC and PancO2-CTCF allografts treated with Clodronate Liposomes.

A small molecule inhibitor, Curaxin, inhibits M2 macrophage polarization to reduce PDAC growth

Our findings demonstrate that CTCF can suppress antitumor immunity and possibly be a prospective treatment target for PDAC. To find a suitable CTCF inhibitor, we conducted screening tests using structure-based molecular docking and sequence-based deep learning [45] and selected the crystal structure of CTCF (PDB login code: 5KKQ) for molecular docking (Fig. 7A). Finally, we screened Curaxin as a promising CTCF inhibitor. To gain insight into the mechanism of CTCF inhibition by Curaxin, we revealed that Curaxin forms hydrogen bonds with the DBR structural groove of CTCF through R399, H340, Y343 and K344 by molecular docking analysis (Fig. 7B). The DARTS technique was performed to clarify that CTCF is protected from proteolysis by Curaxin binding (Fig. 7C). To further explore the treatment effects of Curaxin, we examined levels of CTCF, HNRNPU and IGF2BP2 in pancreatic cancer cell lines and observed that those with high CTCF levels were more sensitive to Curaxin treatment (Fig. 7D, E). Subsequently, we chose two cell lines (PANC-1 and PATU-8988) with higher CTCF expression and two cell lines (Mia-PACA2 and AsPC-1) with lower CTCF expression for further study. Silencing CTCF in PANC-1 and PATU-8988 decreased the sensitivity of these cells to Curaxin and decreased the colony formation capability of PANC-1 and PATU-8988, whereas overexpressing CTCF increased the sensitivity of Mia-PACA2 and AsPC-1 cells to Curaxin (Fig. 7F–I and Supplementary Fig. 16A–H). To further elucidate the oncogenic role of CTCF in PDAC, we employed syngeneic mice to establish subcutaneous and orthotopic PDAC models. We observed that the shCTCF PancO2 group exhibited significantly reduced tumor proliferation and M2 macrophage infiltration compared to the shNC PancO2 group. Moreover, mice in the shCTCF PancO2 group had markedly improved survival outcomes compared to the shNC PancO2 group (Supplementary Fig. 16I–P). To investigate the therapeutic value of Curaxin in PDAC, we further constructed an in vitro model and an orthotopic mouse model. In the co-culture system of PancO2 and BMDMs, M2 polarization of BMDMs increased with elevated CTCF expression in PancO2 cells. However, the addition of Curaxin to the co-culture significantly reduced M2 polarization levels (Supplementary Fig. 17A–D). The mice were then divided the mice into four groups: NC group, CTCF-OE group, NC + Curaxin group and CTCF-OE+Curaxin group. The NC + Curaxin group and CTCF-OE-Curaxin group were injected with Curaxin intraperitoneally (Fig. 7J, K). Intriguingly, the weight of tumor was largest in the CTCF-OE group and treatment with Curaxin caused tumor shrinkage in NC and CTCF-OE and improved the prognosis of mice (Fig. 7L, M). Thereafter, we wondered whether the TAMs-based TME was remodeled after treatment with Curaxin in PDAC. Flow cytometry showed that the CD163⁺ and CD206⁺ cells were downregulated whereas CD80⁺ cells were upregulated in the NC-Curaxin group (Fig. 7N, O and Supplementary Fig. 17E–G). A similar phenomenon was observed in the CTCF-OE+Curaxin and the CTCF-OE groups (Fig. 7N, O and Supplementary Fig. 17E–G). We comprehensively evaluated Curaxin’s therapeutic effects on PDAC to understand its potential benefits in detail. The treatment with Curaxin led to a notable reduction in the proliferation of pancreatic cancer cells (PCs), as demonstrated in Fig. 7P, Q. Additionally, the expression level of CSF1 in PDAC was reduced after the administration of Curaxin, as shown in Supplementary Fig. 17H, I. In short, we showed that Curaxin may hinder the proliferation of PCs by inhibiting CTCF.

Fig. 7. Curaxin inhibits M2 macrophage polarization to reduce PDAC growth.

A The scheme of the screening protocol for CTCF inhibitor. B The binding mode of Curaxin to the R399, H340, Y343 and K344 sites of CTCF in Zinc-finger double domain. C DARTS analysis of the interaction of Curaxin to CTCF. D Immunoblot of CTCF, HNRNPU and IGF2BP2 in different PDAC cell lines. GAPDH was used as a loading control. E CCK8 assay to evaluate the proliferation of the indicated four PDAC cell lines in response to Curaxin for 48 h. CCK8 assay to evaluate the proliferation of the indicated PANC-1 (NC/si-CTCF) (F) and PATU-8988 (NC/si-CTCF) (G) in response to Curaxin for 48 h. CCK8 assay to evaluate the proliferation of the indicated Mia-PACA2 (NC/CTCF-OE) (H) and AsPC-1 (NC/CTCF-OE) (I). J, K The indicated PancO2 cells were orthotopically transplanted into C57BL/6 mice that were treated with Curaxin. Tumor weight (L) and overall survival (M) of C57BL/6 mice among four groups. (n = 6). N Flow cytometry analysis of CD163⁺ macrophages and CD206⁺ macrophages from four groups. O Percentage of CD163⁺ macrophages or CD206⁺ macrophages in F4/80⁺ cells from four groups. IHC staining of Ki-67 (P) and IHC score of Ki-67 (Q) were shown in the four groups. Scale bar = 20 μm.

Curaxin sensitizes PDAC to gemcitabine therapy

Considering the role of gemcitabine as a first-line treatment option in PDAC chemotherapy and the potential role of TAMs in its resistance mechanisms [46], we were concerned to explore whether CTCF could affect the efficacy of gemcitabine in treating PDAC. Interestingly, we discovered that AsPC-1 and Mia-PACA2 were more sensitive to gemcitabine (Supplementary Fig. 18A). Knockdown of CTCF resulted in increased sensitivity to gemcitabine, whereas overexpression of CTCF caused decreased sensitivity to gemcitabine (Supplementary Fig. 18B–I). In the co-culture system of PancO2 and BMDMs, the combination of Curaxin and gemcitabine strongly suppressed M2 macrophage polarization (Supplementary Fig. 18J, K). Furthermore, our research focused on developing a previously unknown combination therapy for PDAC patients who have developed resistance to gemcitabine. To achieve this, we utilized the KPC model to assess the therapeutic potential of combining gemcitabine with Curaxin. Our assessment involved four distinct groups of mice in the study: control group, Curaxin group, gemcitabine group, and gemcitabine plus Curaxin group (Fig. 8A, B). Gem + Curaxin reduces PDAC weight and prolongs KPC mice survival without significant toxicity to Mice (Fig. 8C, D and Supplementary Fig. 18L, M). To gain deeper insights into the mechanism behind the combined treatment with Curaxin and Gemcitabine, we showed that the combination of Gem and Curaxin diminished the proliferation of PCs by using the IHC of Ki-67 (Fig. 8E). Moreover, we confirmed that Curaxin group and Curaxin + Gem group could suppress M2 markers but promote the expression of M1 markers by suppressing the level of CSF1 (Fig. 8F–H). We also observed a similar phenomenon in the PDOX model (Fig. 8I–O). Considering these results, it can be inferred that Curaxin may be a previously unidentified therapeutic approach in the clinical practice of PDAC treatment.

Fig. 8. Curaxin sensitizes PDAC to gemcitabine therapy in mice models.

A Experimental design program. B Representative macroscopic images of pancreatic cancer in KPC mice treated with vehicle, Gem, Curaxin, Gemcitabine and Gem + Curaxin after sacrifice. Statistical analysis for tumor weight (C) and survival (D) of KPC mice from different groups. E IHC scores of Ki-67, CTCF and CSF1 in pancreatic cancer tissues from KPC mice. Percentage of CD163+ macrophages (F) and CD206+ macrophages (F) and CD80+ macrophages (G) in F4/80+ cells from different groups. H CSF1 level of PDAC tissues from KPC mice by ELISA. Experimental design program (I) and representative image (J) of pancreatic cancer in PDOX mice treated with vehicle, Curaxin, Gemcitabine and Gem + Curaxin after sacrifice. Statistical analysis for tumor weight (K) and survival (L) of PDOX mice from different groups. M IHC scores of Ki-67, CTCF and CSF1 in pancreatic cancer tissues from PDOX mice. Percentage of CD163+ macrophages (N) and CD206+ macrophages (N) and CD80+ macrophages (O) in F4/80+ cells from PDOX mice.

Discussion

Malignant tumor patients’ poor survival is strongly linked to the upregulation of genes that promote tumor growth [47]. Our study showed for the first time that CTCF was highly expressed in PDAC tissues compared with paired normal tissues and had specific substantial oncogenic properties. Inhibition of CTCF could hinder the proliferation of PCs cells. These findings were in accordance with its clinicopathological characteristics and endorse our filtering process. To elucidate the molecular mechanisms, we investigated the function of CTCF in enhancing the transcriptional process of downstream target genes. CTCF cooperated with HNRNPU to form a complex that then bound to the promoter of IGF2BP2, which further recruited EP300 to enhance histone lactylation using FLG-AS1 as a scaffold. IGF2BP2 acts as an m6A reader to enhance the mRNA stability of CSF1. Meanwhile HNRNPU interacted with FLG-AS1 to suppress exon skipping of CSF1. Ultimately, PCs induced tumor immunosuppression by releasing CSF1 to enhance the polarization of M2 macrophages (Supplementary Fig. 19).

CTCF, known for its role in chromatin organization [29, 48], interacts with HNRNPU to recruit EP300 and activate the m6A reader IGF2BP2. This interaction enhances histone lactylation at the IGF2BP2 promoter, promoting PDAC cell proliferation. Upregulation of IGF2BP2 stabilizes CSF1 mRNA, contributing to the M2 polarization of tumor-associated macrophages (TAMs), which support tumor growth and immune evasion.

FLG-AS1 mediates the interaction between CTCF and HNRNPU, underscoring the importance of non-coding RNAs in cancer biology. Acting as a scaffold, FLG-AS1 facilitates the assembly of transcriptional complexes, similar to other contexts where lncRNAs modulate transcriptional activity and influence tumor microenvironment dynamics [49, 50].

Histone lactylation, primarily associated with metabolic regulation, also plays a critical role in cancer progression by altering chromatin accessibility and gene expression patterns that favor tumor growth and survival [51, 52]. Understanding how lactylation interacts with other histone modifications can provide deeper insights into transcriptional regulation. EP300, the most common histone acetyltransferase, has been reported to be a putative histone lysine lactylation writer protein [37]. However, the function of EP300-catalyzed histone lactylation and its potential specific mechanism are unclear in PDAC. Our findings show that EP300 plays a pivotal role by facilitating histone lactylation, which alters chromatin accessibility and gene expression patterns that favor tumor growth and survival. This highlights a critical aspect of EP300’s function in cancer progression, beyond its known role in acetylation.

The interplay between CTCF and the IGF2BP2/CSF1/CSF1R axis explains the immunosuppressive environment in PDAC. By stabilizing CSF1 mRNA, IGF2BP2 enhances CSF1 secretion, which binds to CSF1R on macrophages, promoting M2 polarization. This aligns with existing literature on M2 macrophages fostering tumor progression and therapy resistance, highlighting the potential for targeting macrophage polarization in cancer therapy [53, 54].

Our results suggest that targeting CTCF could enhance the efficacy of existing treatments. Combining curaxin with gemcitabine shows promising antitumor activity. This combination therapy could disrupt the CTCF-mediated oncogenic pathways, representing a potential avenue for PDAC treatment.

Future research should focus on dissecting the molecular interactions within this axis and exploring its clinical applicability. Understanding the cross-talk between CTCF and other transcription factors, as well as the broader impact of histone modifications on gene expression in PDAC, will be crucial.

In summary, our study elucidates a complex regulatory axis involving CTCF, IGF2BP2, and CSF1, driving PDAC progression through macrophage polarization and histone modification. These findings deepen our understanding of PDAC biology and open possibilities for therapeutic intervention. By integrating multi-omics data and leveraging advanced sequencing technologies, we can continue to uncover the complexities of cancer biology and develop more effective strategies for combating this devastating disease.

Materials and methods

Cell culture and reagents

PANC-1 and PATU-8988 were purchased from the Cell Resource Centre, Shanghai Institute of Biotechnology, Chinese Academy of Sciences. PancO2 cell line (Cat. NO. NM-D01) was purchased from Shanghai Model Organisms Center, Inc. PancO2, PANC-1, PATU-8988, Mia-PACA2, PancO2 and BMDMs were cultured in Dulbecco’s Modified Eagle Medium (DMEM) (BioChannel Biological Technology Co., Ltd.) supplemented with 10% FBS and penicillin/streptomycin. AsPC-1 and PBMC were cultured in 10% FBS RPMI 1640 medium.

Obtaining of PBMC-derived macrophages and BMDMs

PBMCs were isolated from fresh peripheral blood using Ficoll density gradient centrifugation. The collected PBMCs were washed with PBS and resuspended in complete RPMI 1640 medium. Cells were seeded in 6-well plates at a density of 1–2 × 10⁶ cells/mL and cultured in the presence of M-CSF (50 ng/mL) to induce macrophage differentiation. The medium was refreshed every 2–3 days with fresh M-CSF-supplemented medium. After 5–7 days, adherent macrophages were obtained.

BMDMs were obtained by isolating bone marrow cells from mouse femurs and tibias. Cells were cultured in complete DMEM supplemented with M-CSF (50 ng/mL), with medium refreshed every 2–3 days. After 7 days, adherent bone marrow-derived macrophages (BMDMs) were collected.

Plasmids and stable cell lines

The CTCF, HNRNPU, FLG-AS1, and IGF2BP2 silenced and overexpressed lentivirus was purchased from (Shanghai Bioegene Co., Ltd), and the related sequences were as Supplementary Table 3.

The CTCF, HNRNPU, and FLG-AS1 interference plasmids were purchased from Gene Chem (Shanghai, China), and the related sequences are presented in Supplementary Table 4.

Western blot analysis

RIPA Lysis Buffer (Strong, without inhibitors) (K1120, APExBIO, Houston, USA) was used to lyse cells. PAGE Gel Kits (P0105 LABLEAD lnc.) were used to prepare the gel for electrophoresis. The details of antibodies are shown in Supplementary Table 5. The stripes were captured by Tanon-5200 Chemiluminescent Imaging System (Tanon, China, Shanghai).

Quantitative real-time PCR analysis

SPARKeasy Improved Tissue/Cell RNA Kit (AC0202, Shandong Sparkjade Biotechnology Co., Ltd.) was used to extract RNA from PCs. Evo M-MLV RT Kit with gDNA Clean for qPCR II AG11711 (Accurate Biotechnology (Human) Co., Ltd.) was used to synthesis cDNA. NovoStar SYBR qPCR SuperMix plus E096-01B (Novopretein Scientific Inc.) was used for qPCR assays. HiScript III RT SuperMix for qPCR (+gDNA wiper) (Vazyme) was used to reverse transcription. The details of primers were presented in Supplementary Table 6.

Cell proliferation assay

The CCK-8 (Cell count kit-8) assay (C6050, New Cell & Molecular Biotech) was conducted to determine cell proliferation. Transfected Pancreatic cancer cells (2 × 10³) were collected at 48 h post-transfection and plated in a 96-well plate (701301, NEST Biotechnology). Subsequently, 100 µL of CCK-8 assay reagent was added at specific times and incubated for another 2 h, before measuring the absorbance at 450 nm. For the colony formation assay, 2 × 10³ cells/well were plated in 6-well plates and used to assess the proliferative capacity of pancreatic cancer cells. After 2 weeks, cells were fixed using 1% crystal violet stain solution at room temperature for 20 min, and the colonies were counted manually.

In vivo assays

A total of 5 × 10⁶ PATU-8988 or organoids derived from PDAC patients (PDO) were orthotopically or subcutaneously injected into per BALB/c nude mice. For the mouse model of CSF1Ri application, 5 × 10⁶ PancO2 cells were orthotopically or subcutaneously injected into per C57BL/6 nude mice. After 1 week, 20 mg/kg Pexidartnib 20 mg/kg Pexidartnib was injected into the peritoneal cavity of mice every other day for 3 weeks. For models of macrophage clearance in mice, before implanting PancO2 cells in mice, 200 μL of clodronate liposomes was injected intraperitoneally 1 week in advance. The volume and weight of the tumor were measured after 4 weeks. Volume = 0.5*length*width². LSL-KrasG12D/+, LSL-Trp53R172H/+, and Pdx1-Cre mouse models were generated in-house. Primers for genotyping were listed in the online Supplementary Table 7. Preclinical studies were performed with a KPC mouse model. For animal studies, the mice were randomly allocated into experimental groups. Investigators were also blinded during the in vivo study.

Co-IP

Exactly 500 μL of IP lysis buffer (Beyotime) with Proteinase Inhibitor (NCM Biotech) was used to lyse 1× 10⁷ cells for 10 min on ice. Nuclear extracts were centrifuged at 14,000 rpm, 4 °C for 20 min. After 15 min, 1 μg of anti-CTCF/anti-HNRNPU/anti-EP300/anti-IgG suspended in lysis buffer was added to the supernatant, which was collected in an EP tube. Then, this EP tube was incubated on a sky wheel for 1 h. After 1 h, the reaction mixtures were overnight incubated with 25 μL of protein A/G Dynabeads (Invitrogen) at 4 °C. The products of immunoprecipitation were washed six times with IP lysis buffer. The products were added to the loading buffer and put in a metal water bath at 100 °C for 15 min. The beads were then washed and recovered using magnets.

RNA-FISH for FLG-AS1

FLG-AS1 oligos were obtained from Servicebio Technology. Tumor tissues were immobilized with 4% paraformaldehyde for over 12 h. Thereafter, cells were dehydrated by an alcohol gradient and embedded in paraffin. Then, approximately 10 μm thick sections from tissues were successively deparaffinized, dehydrated and treated with 4% PFA and protease K. Next, slides were incubated with prehybridized in hybridization solution at room temperature for 2 h, followed by hybridization with Fluorescence-labeled single-strand probes overnight at 37 °C. Nuclei were stained with DAPI. Finally, images were captured under the fluorescence microscope (Zeiss).

The probe sequence of FLG-AS1 was

5‘-AAGTTAAGTATTAGTTTCCCTAGGCCAA -3’.

Immunohistochemistry (IHC) and immunofluorescence (IF) staining

Immunohistochemistry (IHC) staining was performed based on protocols from a previous study [17]. The antibodies used were listed in Supplementary Table 5. Sections were visualized, and images were acquired using the Olympus FSX100 microscope (Olympus, Japan). Further protein expression levels analysis was performed using the image-pro Plus 6.0 software by calculating the integrated optical density per stained area (IOD/area). To visualize the co-localization of CTCF and HNRNPU, PDAC tissues were fixed in 4% paraformaldehyde (PFA) for 30 min and embedded in the optimal cutting temperature (OCT) compound. Thereafter, the tissues were incubated with CTCF primary antibodies and anti-rabbit secondary antibodies. Nuclei were counterstained with DAPI, respectively. Eventually, the SP‐8 confocal Microscope was used for further visualization.

Dual-luciferase reporter assay

A dual luciferase reporter assay was performed as described in previous protocols [17]. HEK-293 T cells were cultured in 6-well plates (TCP011006, Guangzhou Jet Bio-Filtration Co., Ltd.) at 2 × 10⁵ cells per well for 24 h. HEK-293 T was co-transfected with the promoter of IGF2BP2 and CTCF/HNRNPU reporter plasmids (Bioegene). Dual luciferase reporter assay kit DL101-01(Vazyme Biotech Co., Ltd) was used to measure the luciferase activity.

ELISA

PBMC-derived macrophages (1 × 10⁶ cells) or BMDM (1 × 10⁶ cells) were co-cultured with PCs with or without CTCF overexpression (1 × 10⁶ cells) in 6-well plates for 48 h. Culture supernatants were collected and passed via 0.47-mm filters (Whatman, 6809-5012, SORFA Life Science). IL-10, IL-6, and TGFβ levels were detected using the mouse IL-10 ELISA kit (E-EL-H6154, Elabscience Biotechnology Co., Ltd), IL-6 ELISA kit (LCSCM17037, AMOY LUNCHANGSHUO BIOTECH, CO., LTD) and TGF-β ELISA kit (ELK1185, ELK Biotechnology).

Chromatin immunoprecipitation (ChIP)

The chromatin immunoprecipitation (ChIP) assay was performed as described in previous studies [34]. Antibodies used for ChIP are presented in Table S5. Additionally, the library for ChIP-seq was made with VAHTS Universal DNA Library Prep Kit for Illumina V3 (ND607, Vazyme Biotech Co., Ltd) following the manufacturer’s instructions. The generated DNA Library was sequenced with Illumina HiSeq X Ten using the paired-end module. For ChIP‐qPCR, ChIP products purified by Quick PCR Purification and Gel Extraction Kit (K004, Wuhan Fine Biotech Co., Ltd.) were used to amplify the PCR products for 45 cycles. The information on antibodies and related primers is separately presented in Supplementary Tables 5 and 6.

RNA immunoprecipitation (RIP)

Antibodies of CTCF, HNRNPU and IGF2BP2 were used for RIP assay. Magana RIP Kit was used to detect CSF1, CTCF and FLG-AS1 based on protocol of manufacturer. Immunoprecipitated RNA was extracted by phenol, chloroform and isoamyl alcohol (125:24:1 pH = 4.3) and Evo M-MLV RT Kit with gDNA Clean for qPCR II AG11711 (Accurate Biotechnology (Human) Co., Ltd.) was used for reverse transcription RNA. The levels of related RNA were showed by qPCR or RNA sequencing.

RNA pulldown assay

RNA pulldown assay was performed as described in previous studies [55]. RNA baits of CTCF and CSF1 were gotten from Bioegene Company. MagCapture™ RNA Pull Down kit (Millipore Corporation, USA) was used for this assay. We used WB assay to detect the level of RNA binding protein.

m6A sequencing

PANC1-NC and PANC1-METTL3-METTL14-KD cells were collected to perform m6A-seq. Firstly, total RNA was directly lysed with TRIzol (Thermo Fisher Scientific). The RNA integrity was assessed by Bioanalyzer 2100 (Agilent) and confirmed by electrophoresis with denaturing agarose gel. Dynabeads Oligo (dT) 25-61005 (Thermo Fisher) was used for the purification of Poly (A) RNA, which was then fragmented into pieces using Magnesium RNA Fragmentation Module (NEB). Afterwards, the cleaved RNA fragments were incubated with m6A-specific antibody (Synaptic Systems) for 2 h at 4 °C in IP buffer (50 mM Tris-HCl, 750 mM NaCl and 0.5% Igepal CA-630). Then the IP RNA was reverse transcribed to cDNA by SuperScript™ II Reverse Transcriptase (Invitrogen) and construct final cDNA libraries. At last, we performed the 2*150 bp paired-end sequencing.

AS analysis and M6A-seq data analysis

The initial step involved processing the raw FASTQ file with fastp v0.20.1 for quality control, serving to filter and trim sequencing reads while checking for contamination [56]. Then, we aligned the quality-assured reads to a reference genome using HISAT2 v2.2.1 [57].

Subsequently, the RNA-seq data using featureCounts v2.0.1 for read summarization and quantification [58]. Differentially expressed genes were detected by DESeq2 v1.30 [59].

To calculate PSI, we used the rMATS software, which estimates PSI based on read coverage across exon-exon junctions and exon inclusion levels. Alternative splicing events identified by rMATS with |PSI| > 0.1 and FDR threshold of 0.05 [60].

For the m6A-seq data analysis, MACS2 v2.2.7.1 was employed for peak calling using a p value threshold of 0.05, followed by the utilization of exomePeak v1.8.0 for differential analysis, identifying enriched modification sites with a FDR threshold of 0.05 [61].

GO biological process and pathway enrichment analyses were performed using clusterProfiler v4.8.1 [62].

Crosslinking and immunoprecipitation (CLIP)

Briefly, approximately 5 × 10⁶ cells were treated with 200 μM 4-thiouridine (4-SU) for 12 h and washed with cold PBS twice. Then, cells were crosslinked with 150 mJ/cm² of 365 nm ultraviolet light and lysed directly with lysis buffer and protease inhibitor. After that, anti-HNRNPU antibody and anti-IgG antibody was used to immunoprecipitated with RNA. Subsequently, protein-RNA complexes ran on the SDS-PAGE gel and recovered and de-crosslinked with proteinase K. Finally, targeted RNA was reverse transcribed into cDNA and used for further real-time qPCR analysis.

Flow cytometry

Cells were resuspended in 50 μL of staining buffer (PBS: FBS = 1000:1). Then, 1.5 μL of anti-CD163, anti-CD206, anti-CD80, anti-CD86, anti-F4/80 and anti-CD45 were added to the reaction for 30 min at 4 °C. After 30 min, PBMC-derived macrophages, BMDM or single cells from tumor tissues were washed twice with staining buffer, and 1% formaldehyde was used for fixation at 4 °C. The data were detected by an FCM (Beckman Coulter).

RNA stability assay

Plasmids were transfected in PANC-1 or PATU-8988 cells over 48 h. Further, we used actinomycin D (5 ng/μL) to treat these cells for 12 h or 24 h. Finally, RNA stability of CSF1 and MYC was detected by qPCR.

Molecular dynamics (MD) simulation

To perform the molecular dynamics (MD) simulations of CTCF and HNRHPU, we obtained the crystallographic structure of CTCF (PDB ID: 5K5H) from the RCSB Protein Data Bank. The 3D structure of HNRHPU was predicted using the I-TASSER Server [63]. The complexes formed between CTCF and HNRHPU were predicted using the Zdock program [64]. The MD simulations were conducted using Gromacs 5.1.5 with the Amber99sb force field. The binding free energy of CTCF with HNRHPU was evaluated using the molecular mechanics energies combined with the Poisson-Boltzmann and surface area continuum solvation (MM/PBSA) methods [65]. The interaction residues between CTCF and HNRHPU were analyzed using LIGPLOT (v.4.4.2) [66].

Grilling for small molecule inhibitor of CTCF

The chemical library was used to grille the small molecule inhibitor of CTCF exactly. At first, the Schrödinger suite of molecular modeling software was performed to dock CTCF(PDB:5KKQ) and the prepared Specs Library via the high-throughput virtual screening mode. Then, in accordance with the deep learning model based on TransformerCPI, the top 3000 compounds were chosen to further molecular docking. Finally, curaxin was selected among the top 3000 compounds through combining the deep learning model and the merging and clustering of the molecules docking poses.

DARTS analysis

Drug affinity responsive target stability (DARTS) were performed based on previously documented protocols [67]. PANC-1 cells were lysed by M-PER buffer (Pierce) supplemented with protease and phosphatase inhibitors on ice for 10 min. After centrifugation, protein lysates were mixed with TNC buffer (50 mM Tris-HCl, 50 mM NaCl, 10 mM CaCl2). Different concentration of Curaxin were added to equal volume of lysates and mixed well, incubating at room temperature for an hour. Pronase solution was used for digestion for 30 min and stopped by Protease inhibitor solution. After stopping the proteolysis, the expression of CTCF in samples were analyzed by Western Blot, while GAPDH was used as a negative control.

CatRAPID

The binding regions of FLG-AS1 and CTCF or HNRNPU were predicted by an online website CatRAPID (http://service.tartaglialab.com/page/catrapid_group). Specifically, by filling the sequence of corresponding proteins and RNA, we can directly obtain the predicted information of their binding regions.

Analysis of public scRNA-seq data and scATAC-seq data

All public database scRNA-seq data were downloaded from the National Omics Data Encyclopedia (NODE) with the accession code OEP003254 and the Genome Sequence Archive (GSA) database with the accession code CRA001160 [25, 68]. scATAC-seq data were downloaded from the Gene Expression Omnibus (GEO) database with the accession code GSE137069 [26]. For each set of data, quality control steps, dimensionality reduction, clustering, and cell annotation were strictly performed based on the parameters provided in the original publications.

For scRNA-seq data, ductal cells were first extracted, and gene expression in ductal cells was compared between the normal and the tumor groups using the FindMarkers() function in the R package Seurat (version 4.3.0). Thereafter, CTCF-positive ductal cells were defined as ductal cells with CTCF mRNA counts >0. After identifying the subgroups of CTCF-positive ductal cells, differential gene expression analysis was performed with the FindMarkers() function between CTCF-positive and CTCF-negative ductal cells. Only those with adjusted p values < 0.05 and |log2FC | > 0.25 were identified as differentially expressed genes (DEGs). GO and pathway enrichment analysis of CTCF-related DEGs was performed by Enrichr [69]. To illustrate the cell-cell interaction potential of CTCF-positive and CTCF-negative ductal cells with macrophages, we used CSOmap to construct a 3D pseudo space and calculate the interaction [70].

For scATAC-seq data, Motif-activity analysis in pancreatic ductal adenocarcinoma cancerous tissues compared to normal healthy tissue (normal) was performed with Signac’s wrapper of ChromVAR (version 1.9.0) [71].

Statistical analysis

The R platform and GraphPad Prism 9 were used for statistical analyses. We performed the repeated assays in corresponding figures. Data were presented as the mean ± SD. Paired two-tailed Student’s t test, one-way ANOVA, and Chi-square test were used to calculate the difference between the two groups. The variance is similar between the groups being statistically compared.

Author contributions

YH Liu designed experiments and drafted manuscripts; YH Liu, PY Liu, JY Lin, WX Qi, Y Liu, Y Jiang, ZW Yu, X Gao, XQ Sun, J Liu, JW Lin, SY Zhai, YZ Cao, K Qin and JW Li conducted experiments; YH Liu, PY Liu, JY Lin, Y Jiang, MM Chen, SY Zou, CL Wen, HL Bao and KY Sun were responsible for sample collection; SQ Duan, D Fu and JC Wang was responsible for data analysis; J Wang, Y Jiang, KY Sun, HL Bao and BY Shen discussed and revised the manuscript. All authors approved the manuscript.

Funding

This work was sponsored by The National Natural Science Foundation of China (81871906 and 82073326), Basic Research Program of Shanghai (20JC1412200), the National Key Research and Development Program of China (2020YFA0113000) and PostGraduate Innovation Fund of Interdiscipline and New Medicine from School of Medicine of Shanghai University.

Data availability

Data are available in a public, open access repository. All sequencing data generated in this study are deposited at Mendeley Data, 10.17632/hzjwhm46 and 10.17632/7fkktknnm8.1. All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Information.

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

Sample collections were approved by Ethnics Committee of Ruijin Hospital, Shanghai Jiaotong University School of Medicine (2021 Clinical Ethnics Review No. 161). Also, the animal research was authorized by the Shanghai Municipal Science and Technology Commission of Shanghai, China (SYXK-2018-0027). The informed consents were obtained from patients or their guardians, as appropriate. All methods were performed in accordance with the relevant guidelines and regulations.

Supplementary information

The online version contains supplementary material available at 10.1038/s41418-024-01423-1.