Introduction
Protein arginine methylation is a pervasive posttranslational modification (PTM) that plays an important role in cell signaling, but when deregulated, can be associated with different diseases. Indeed, the overexpression of protein arginine methyltransferases (PRMTs) is frequently seen in human cancers, and is generally associated with poor clinical outcome1,2, which has prompted the development of PRMT small molecule inhibitors3. There are nine enzymes in the PRMT family: PRMT1-9; PRMT4 is also referred to as CARM1. CARM1 together with PRMT1, 2, 3, 6, & 8, are classified as type I PRMTs that catalyze the deposition of asymmetric dimethylarginine (ADMA), while PRMT5 and 9 are grouped together as type II PRMTs and are responsible for symmetric dimethylarginine (SDMA)4.
CARM1 was the first PRMT to be implicated in transcriptional regulation through its ability to be recruited by nuclear receptors and then methylate histone H35. Subsequent studies have sought to unravel the biological roles of CARM1 by screening for its substrates using protein arrays6, and by the direct testing of candidate substrates7, small-pool screening8 and mass spectrometry approaches9,10. A host of CARM1 substrates have been identified, which clearly places this enzyme in a central position for coactivating transcription. It methylates and regulates the activity of other coactivators like p300/CBP11 and the H3K4 methyltransferases KMT2C and KMT2D9,10. It also stabilizes promoter/enhancer looping by modifying MED12 in the Mediator complex9,12. CARM1 modulates chromatin-remodeling by targeting Pontin13. Finally, CARM1 directly methylates a number of transcription factors including PAX714, SOX215, RUNX16, and YY17. Mechanistically, a CARM1 methylated substrate is often able to recruit an effector molecule, or “reader”, to the newly created methyl-motif. One such effector is TDRD318, which binds both the histone H3R17me2a and the MED12-R1899me2a marks9,19, and in-turn recruits TOP3B to chromatin to resolve R-loops and untangle RNA structures20. TDRD3 harbors a Tudor domain for the purpose of reading the ADMA marks deposited by CARM1 and PRMT1. Other Tudor domain-containing proteins like SND1, SMN, SPF30 and TDRD1, all selectively interact with SDMA marked motifs21. It is likely that additional methylarginine effector proteins exist.
To expand our understanding of how CARM1 regulates cellular processes, we performed a focused search of additional substrates of this enzyme and identified the nuclear factor I (NFI) family of transcription factors. The NFI family has four members: NFIA, NFIB, NFIC, and NFIX22, and all four harbor a conserved motif that can be modified by CARM1. Importantly, both Carm1 and Nfib null mice display very similar lung hyperplasia that causes death just after birth due to respiratory defects23,24,25, possibly indicative of a mechanistic link between CARM1 and NFIB in the lung. NFIB was reported as an oncogene that promotes metastasis of small cell lung cancer (SCLC) by increasing the chromatin accessibility in gene distal regions26,27,28. High levels of NFIB are associated with human SCLC metastases and poor overall survival28. Like NFIB, CARM1 is often overexpressed in human cancers, including lung cancer1,29. Furthermore, in a mouse model of overexpression, elevated CARM1 levels in Keratin 5 (K5) expressing tissues (skin and the mammary gland) present with a high incidence of spontaneous tumors30. These findings suggest that CARM1 inhibitors may have therapeutic value, and such compounds have been developed by Pharma31,32 and in an academic setting33. CARM1 inhibitors display anti-tumor activity, in a pre-clinical setting, against multiple myeloma31,32, acute myeloid leukemia34, breast cancer33 and diffuse large B-cell lymphoma35.
The recent study showing that CARM1 inhibition is a vulnerability for CREBBP/EP300 mutations carrying lymphomas35, prompted us to investigate if other tumor types that are driven by CARM1 substrates may also be responsive to CARM1 inhibitors. Here, we explored the consequence of NFIB methylation, and show that the TRIM29 protein interacts with NFIB in a CARM1-dependent manner. The NFIB methylation site (R388) is important for promoting the transcription of its target genes, and as an effector of this mark, TRIM29 functions as a transcriptional coactivator. These cell-based studies encouraged us to investigate the importance of NFIB methylation by CARM1 in vivo. Using a genetically engineered mouse model (GEMM) for SCLC, we find that both Carm1 conditional loss and NfibR388K knock-in results in a significant increase in survival, and the tumors derived from these two models display very similar patterns of open chromatin. Furthermore, SCLC PDX models are sensitive to CARM1 inhibitor treatment. These studies suggest that tumors that are driven by NFIB amplification are vulnerable to the targeting of CARM1 activity.
Results
NFIB is a CARM1 substrate
To identify substrates for CARM1, we immunoprecipitated ADMA-containing peptides using previously reported ADMA antibodies36 from proteolytically digested CARM1 wild-type (WT) and knockout (KO) mouse embryonic fibroblasts (MEFs). Both the antibodies (D4H5 and D6A8) used for the enrichment of ADMA methylated peptides have been shown to recognize a subset of CARM1 substrates, but they also recognize PRMT1 substrates36. Thus, the use of CARM1 WT and KO MEFs in this experiment will help us identify the CARM1-specific substrates. The ADMA-enriched peptides were then identified by LC-MS/MS. A total of 1062 peptides that contain di-methylated arginine were cataloged (Supplementary Data S1A). Among them, 270 peptides were identified from CARM1 WT, but not KO MEFs (Fig. 1a), suggesting that these are sites of CARM1-mediated methylation.
a ADMA-containing proteins were immunoprecipitated from CARM1 wild-type (WT) or knockout (Carm1−/−) MEFs, and identified by LC-MS/MS. b Identified peptides from the nuclear factor I family. c NFIB protein was immunoprecipitated from either CARM1 WT or KO MEFs, and immunoblotted with antibodies against ADMA (D4H5) or NFIB. d GFP vector control, GFP-NFIB-WT or -R388K, GFP-NFIC-WT or -R395K constructs were transfected into HeLa cells. GFP-tagged proteins were immunoprecipitated and immunoblotted with antibodies against ADMA (D4H5) or GFP. e Validation of NFIB methylation using the NFIBme2a antibody. Total cell lysates were analyzed from NFIB WT/KO HeLa cells, CARM1 WT/KO MEFs and WT MEFs treated with CARM1 inhibitor (CARM1i, TP-064). Actin shown as a loading control. f IP-MS was performed, in triplicate, on cell lysates from WT or Carm1−/− MEFs using the NFIBme2a antibody. Exclusive spectrum counts for all the four NFI family proteins were analyzed and plotted. Data present in mean ± SEM; P values determined by two-tailed student’s t-test. All blots are shown as representative data from three independent experiments.
Notably, all the nuclear factor I (NFI) family members (NFIA, NFIB, NFIC and NFIX) were identified as methylated in CARM1 WT MEFs only, and they harbor a conserved amino acid sequence around the methylated arginine site (Fig. 1b). The motif is proline-rich, which is indicative of a CARM1 methylation motif and consistent with the reported recognition motif for the D4H5 antibody used for the peptide-enrichment step10. The NFI peptides constitute about 8% of total peptides unique to CARM1 WT MEFs, indicating that they are dominant targets for CARM1. We are particularly interested in the methylation of NFIB, since both Nfib and Carm1 knockout mice die perinatally, due to a similar deficiency in lung development24,25,37,38. To confirm that NFIB is methylated by CARM1 in the cells, we immunoprecipitated (IPed) NFIB from both CARM1 WT MEFs and CARM1 KO MEFs, and detected the methylation of NFIB by immunoblotting with the specific ADMA antibody (D4H5) (Fig. 1c). The ADMA antibody only detected IPed NFIB from CARM1 WT, but not CARM1 KO MEFs. Furthermore, we then made GFP fusions of NFIB or NFIBR388K (where the arginine methylation site is mutated to lysine) and transfected them into HeLa cells. The GFP-tagged NFIB proteins were then IPed for the detection of ADMA by Western (Fig. 1d). Methylation on wild-type, but not on mutated NFIB, can be observed. Similarly, methylation of NFIC is abolished when R395 is mutated to lysine.
We also generated our own methyl-specific antibody that recognizes the NFIBR388me2a motif (Supplementary Fig. 1a) and produced NFIB KO HeLa cells to help characterize this antibody (Supplementary Fig. 1b). This purified antibody (NFIBme2a) was tested on total cell lysates from NFIB WT and KO HeLa cells, CARM1 WT and KO MEFs, as well as WT MEFs treated with vehicle or a specific CARM1 inhibitor31 (TP-064). This NFIBme2a antibody is totally CARM1-dependent as it does not recognize any proteins when CARM1 is absent or inhibited (Fig. 1e). The Western analysis of the NFIB knockout cells reveals that the NFIBme2a antibody does recognize additional CARM1 methylated proteins, apart from NFIB, and these might be other NFI family members (Fig. 1e). Indeed, mass spec. analysis was performed in triplicate on CARM1 WT and KO MEFs, using our own methyl-specific antibody, and we identified strong enrichment of the NFI family members (Fig. 1f), as well as a number of additional proteins that are putative CARM1 substrates (Supplementary Data S1B).
CARM1 is a transcriptional coactivator for NFIB
NFIB is a transcription factor that governs the expression of numerous genes critical for SCLC differentiation, hair follicle stem cell behavior, androgen receptor signaling, megakaryocyte cell maturation and neuron development39. We were interested in determining whether the methylation of NFIB by CARM1 is a way of promoting its transcriptional activity. To identify direct NFIB target genes, we performed chromatin immunoprecipitation followed by sequencing (ChIP-seq) to detect the distributions of NFIB binding sites across the genome of HeLa cells. ChIP-seq was performed using two independent NFIB antibodies, and also using a GFP antibody on a HeLa cell line that stably expresses GFP-NFIB (Supplementary Fig. 1c). By comparing all 3 sets of ChIP-seq data, we found 1696 high-confident loci were directly bound by NFIB, which we refer to as NFIBYYY (YYY = yes for all three Abs) (Supplementary Fig. 1d). We also performed a ChIP-seq experiment using an antibody that recognizes the histone H3R17me2a mark, which CARM1 deposits. Although this H3R17me2 antibody recognizes the methylation site on H3, it also cross-reacts with many other CARM1 substrates8, and methylated NFIB itself (Supplementary Fig. 1e). Thus, the H3R17me2a ChIP peak denotes regions of CARM1 activity, and not necessarily the H3R17me2a mark itself. We found that roughly one third (523) of the NFIBYYY peaks overlap with sites that harbor CARM1 activity (Supplementary Fig. 1f and Supplementary Data S2), and represent NFIB-bound loci that are candidates for CARM1 regulation. We selected six NFIB-bound loci from this set of 523 peaks and validated this binding by ChIP-qPCR (Supplementary Fig. 1g). To test whether methylation of NFIB affects transcription regulation, we compared the mRNA levels of these selected genes upon the overexpression of wild-type NFIB (NFIB-WT) or the methylation-deficient mutant (NFIB-R388K). Generally, overexpression of NFIB-WT significantly increased the mRNA levels of the selected genes, while overexpression of NFIB-R388K did not promote transcription (Supplementary Fig. 1h). Consistent with this finding, inhibition of NFIB methylation by the CARM1 inhibitor, TP-064, also significantly inhibited the expression of all the selected NFIB target genes, except FERMT1 (Supplementary Fig. 1i). These results indicate that the methylation of NFIB by CARM1 plays an important role in regulating the gene expression program driven by this transcription factor.
TRIM29 is an effector molecule for methylated NFIB
As a transcription factor, NFIB harbors DNA-binding domains towards its N-terminus, and the C-terminal region carries the transactivation and repression activities40. The CARM1 methylation site on NFIB (R388) is located in the C-terminal transcription modulation domain, and thus this methylation will likely not impact the DNA-binding properties of NFIB. In many cases, the posttranslational modification of TFs regulates their activator/repressor functions by facilitating the recruitment of effector proteins that co-regulate transcription. We thus sought to identify a “reader” for the NFIB methylation site. To do so, we ectopically expressed GFP-tagged NFIB-WT or its R388K mutant version in HeLa cells. We then purified the respective protein complexes using the GFP-Trap affinity approach and identified the interacting proteins by proteolytic digestion and LC-MS/MS (Fig. 2a and Supplementary Data S3). We were interested in identifying proteins that bound GFP-NFIB but not GFP-NFIBR388K, as these will be the candidate effector proteins for the CARM1 methylation site. We selected 15 proteins that were identified by two or more peptides in the GFP-NFIB complex and displayed zero peptides in the GFP-NFIBR388K complex. Most of the selected proteins had reported functions related to transcriptional regulators. These effector candidates were cloned into an expression vector with a tag and overexpressed in HeLa cells (in a few cases endogenous proteins were directly detected using commercial antibodies), and the cells were then treated with DMSO or CARM1 inhibitor. Proteins interacting with NFIB were co-immunoprecipitated and examined for FLAG tag signal (Fig. 2b). TRIM29 was shown to interact with NFIB in a CARM1-dependent manner. This methylation-dependent interaction was further validated by an in vitro peptide pull-down assay, where GST-tagged TRIM29 preferentially binds to methylated NFIB peptides (Fig. 2c), demonstrating that this is a direct interaction. TDRD3 is a methylation reader that binds to many proteins marked with ADMA, including histones H3 and H4, RNA polymerase II, and MED129,19,41. Here, we show that TDRD3 can also bind to the CARM1 methylation site on NFIB, and they can be co-IPed (Fig. 2c and Supplementary Fig. 2a). Thus, the methylated form of NFIB may be coregulated by either TRIM29 or TDRD3, which was further investigated.
a Workflow of the screening for proteins that bind wild-type but not mutant NFIB. b HEK293T cells were transfected with indicated transcription regulators. The cells were cultured in the presence or absence of CARM1 inhibitor. The NFIB protein complex was immunoprecipitated and the association of these candidates was tested by Western blot, using an anti-FLAG antibody. c GST-tagged full-length TRIM29 was incubated with biotin-labeled unmethylated or methylated NFIB peptides. Peptide pull-down was performed using streptavidin-conjugated beads. d Scheme showing the domains of TRIM29 that were fused to FLAG and GST for the experiments performed in (e) and (f). e Full length and truncated TRIM29 constructs were transfected into HEK293T cells. The cells were cultured in the presence or absence of CARM1 inhibitor. NFIB was IPed and the interaction of these TRIM29 mutants was tested by Western blot, using an anti-FLAG antibody. f GST-tagged full-length and truncated TRIM29 were incubated with biotin-labeled unmethylated or methylated NFIB peptides. Peptide pull-down was performed using streptavidin-conjugated beads. g ChIP-seq tracks showing the distribution of NFIB in promoter regions of GFAP gene. Three different antibodies were used to generate the tracks: an Abcam antibody, a Bethyl antibody, and a GST antibody that detects ectopically expressed GFP-NFIB. h HEK293T cells were co-transfected with GFAP-luc, together with GFP control construct, NFIB-WT, NFIB-R388K, TRIM29, TRIM29 plus NFIB-WT, or TRIM29 plus NFIB-R388K. The luciferase activities were detected by dual luciferase assay. The relative luciferase activities are normalized to the GFP control. Experiments were performed in biological triplicate. Data present in mean ± SEM; P values determined by two-tailed student’s t-test. i HEK293T cells were co-transfected with GFAP-luc, together with GFP control construct, NFIB, NFIB plus TDRD3, or NFIB plus TRIM29. The luciferase activities were detected by dual luciferase assay. The relative luciferase activities are normalized to the GFP control. Experiments were performed in triplicate. Data present in mean ± SEM; P values determined by two-tailed student’s t-test. j TRIM29-interacting proteins are identified by GFP trap in H69 SCLC cell line, followed by MS analysis as in (a). List of high-confidence candidate interacting proteins from MS analysis of GFP-tagged TRIM29 and GFP (control). Asterisk denotes TRIM29-binding proteins that have previously been identified by MS42. All blots are shown as representative data from three independent experiments.
We next mapped the region of TRIM29 that directly interacts with methylated NFIB. To do so, we generated four truncated forms of TRIM29 (Fig. 2d) and subjected these FLAG-tagged proteins to a co-IP with endogenous NFIB. For this experiment, HeLa cells were also treated with CARM1 inhibitor. We observed that both the N-terminal domain and the C-terminal domain of TRIM29 interact with NFIB (Fig. 2e). When NFIB methylation is inhibited, the interaction between the C-terminus and NFIB is significantly reduced, while interaction between the N-terminus and NFIB remained unchanged, suggesting that the C-terminal domain of TRIM29 is responsible for the methylation-dependent interaction. To confirm this, a peptide pull-down assay was performed using GST fusion of full-length TRIM29, as well as fusions of the C-terminal domain only, and a fusion of TRIM29 without its C-terminus (Fig. 2d). Clearly, the C-terminal domain specifically interacts directly with methylated NFIB (Fig. 2f). Using a co-IP assay, we also show that endogenous NFIB and TRIM29 interact, and that this interaction is prevented by CARM1 inhibitors (Supplementary Fig. 2b).
The C-terminal domain of TRIM29 was reported to mediate interactions with many proteins42. This region is an orphan domain that has no homologous regions in other proteins. To further characterize TRIM29 as an arginine methylation reader, we incubated the C-terminal domain of TRIM29 with a methylated GAR motif from fibrillarin. TRIM29 interacted strongly with the fibrillarin GAR motif that has the ADMA marks, but not the SDMA form (Supplementary Fig. 2c). As expected, the Tudor domain of TDRD3 showed stronger binding to ADMA marks, while the Tudor domain of SMN showed stronger binding to SDMA marks. Thus, TRIM29 likely binds additional methylated proteins apart from just NFIB.
TRIM29 promotes the transcription of NFIB target genes
Effectors that specifically bind to arginine methylated transcription regulators can shape the downstream transcription program13,19,43. GFAP is a well-studied target gene of both NFIA and NFIB in astrocytes44,45,46, and a direct NFIB target identified by our ChIP-seq. Indeed, both NFIB antibodies (Abcam and Bethyl) display GFAP promoter peaks, which are elevated in ectopic GFP-NFIB expression (Fig. 2g). To establish if TRIM29 can coactivate the expression of NFIB regulated GFAP, we performed an in vitro luciferase reporter assay. We found that TRIM29 can significantly promote transcription driven off the GFAP promoter. This coactivator activity is seen when wild-type NFIB is overexpressed with TRIM29, but not when the methylation deficient NFIB mutant is overexpressed (Fig. 2h). Next, we compared the coactivator activity of TRIM29 and TDRD3, as we have shown that both these proteins can function as readers of the methylated NFIB peptide (Fig. 2c). Using the GFAP luciferase reporter assay, we see that TRIM29, but not TDRD3, functions with NFIB to coactivate expression of this promoter (Fig. 2i and Supplementary Fig. 2d). We thus focused our attention on TRIM29 as a novel effector molecule for methylated NFIB. Next, we confirmed TRIM29 protein expression in the cell lines used in this study (Supplementary Fig. 2e) and observed that expression of the majority of the identified NFIB-regulated genes (see Supplementary Fig. 1g) is promoted upon overexpression of TRIM29 (Supplementary Fig. 2f).
Overexpression of NFIB is correlated with advanced tumor stages and metastasis in small cell lung cancers27,28 therefore, we hypothesize that methylation of NFIB may also regulate the progression of SCLC. Since the mechanistic studies described above were performed in HeLa and HEK293T cells, we sought to confirm the NFIB/TRIM29 interaction in SCLC cells. To that end, we performed the GFP-trap experiment with GFP-TRIM29 in H69 SCLC cell line and found that the top-ranked binding protein is NFIB (Fig. 2j and Supplementary Data S4). Thus, reciprocal GFP-trap experiments in HeLa and H69 cell lines identified the NFIB/TRIM29 interaction (Fig. 2a, j).
CARM1-mediated methylation of NFIB is required for tumor growth of SCLC in vivo
We examined the CARM1-NFIB regulatory module in SCLC by immunostaining human tumor samples for CARM1, NFIB and NFIBme2a. We found that high levels of CARM1 and NFIB are independently predictors of poor patient survival (Fig. 3a) and a combination of CARM1high/NFIBhigh predicts worse survival (Fig. 3a). Next, we analyzed tumors harvested from TKO mouse model of SCLC which carries conditional knockout of three tumor suppressors (Rb1, Rbl2, Trp53)47 commonly mutated in human cancer. We observed increased levels of CARM1, NFIB, and NFIBme2a as cancer progresses into malignant disease (Fig. 3b). In addition, we noted that NFIB expression correlates with human SCLC cell lines sensitivity to CARM1 inhibitor (CARM1i, TP-064) (Supplementary Fig. 3a). To investigate the importance of CARM1’s enzymatic activity in driving SCLC development, we performed a xenograft using modified H69 (Fig. 3c–f) and CORL47 (Supplementary Fig. 3b, c) cells. These cells harbored a CRISPR-mediated knockout of CARM1 and were rescued with either WT or enzyme-dead (R168A) CARM1. Tumor volume was dramatically impaired when CARM1 was ablated, and this phenotype was not rescued when the KO cells were complemented with mutant CARM1 (Fig. 3c, d and Supplementary Fig. 3b). Our results implicate CARM1 methylation of NFIB in SCLC tumorigenesis. Thus, we wanted to independently test the role of NFIB R388 methylation in the regulation of cancer expansion. Depletion of NFIB led to attenuated growth of SCLC H69 cells in vivo (Fig. 3e, f). Complementation with wild-type NFIB, but not NFIB harboring an R388K substitution, was able to fully restore cancerous growth in vivo (Fig. 3e, f and Supplementary Fig. 3c). We further confirmed that CARM1i phenocopies genetic ablation of CARM1 and NFIB methylation in H69 and CORL47 cells and that inhibitor does not exert an unspecific effect on cancer cells growth (Supplementary Fig. 3d, e). Similar to genetic or pharmacological inhibition of CARM1 and ablation of NFIB methylation, we observed reduced tumorigenic capacity of SCLC cells upon TRIM29 deletion (Supplementary Fig. 3f, g). Together, these data argue that CARM1 regulates SCLC tumor growth via methylation of NFIB R388 and the subsequent recruitment of the methylation reader TRIM29.
a High expression of CARM1 and NFIB predicts poor SCLC patient survival. Analysis of SCLC tissue array IHC for CARM1 and NFIB. P values by Log-rank test, n = 32 (16 ♂ / 16 ♀, no sex difference noted). b Immunoblot analysis of CARM1, NFIB and NFIBR388me2a in three independent and representative tissue biopsies lysates (of six analyzed) from SCLC mouse model (TKO) at early and late stages of tumorigenesis with non-metastatic and metastatic disease. c H69 SCLC cells xenograft growth in NSG mice modified to express sgRNA CARM1 or sgRNA control and overexpressing WT or catalytically deficient R168A CARM1 (n = 5; P values by two-way ANOVA with Tukey’s testing for multiple comparisons, data represent mean ± SEM). d Immunoblots with indicated antibodies of cell lysates as in (c). e Complementation of sgRNA NFIB depleted H69 SCLC cells with WT or R388K mutant NFIB shows that NFIB methylation is required for SCLC cell growth in vivo (n = 5; P values determined as in (c)). sgControl as in (c). f Immunoblots with indicated antibodies of cell lysates as in (e). g Schematic of the Rb1; Rbl2; Trp53 (TKO) mouse SCLC models harboring either a conditional Carm1L/L knockout allele or a Nfib R388K mutant. h Representative HE and IHC of phospho-H3 staining, scale bars, 100 µm. i Quantification of tumor burden in indicated mutant mice (n = 6, balanced ♂/♀, no sex difference noted). P values by two-way ANOVA with Tukey’s testing for multiple comparisons; box plots, the line indicates the median, the box marks the 75th and 25th percentiles and the whiskers indicate the minimum and maximum values. j Quantification of proliferation marker (phospho-H3+ cells) in indicated mutant mice as in (i) (n = 6). k Kaplan–Meier survival curves of TKO control (n = 9, median survival 193d) and TKO;Carm1KO (n = 7, 297d) and TKO;NfibR388K mutant mice (n = 6, 288d). P values by Log-rank test, balanced ♂/♀, no sex difference noted. l Liver metastasis incidence in the indicated mouse models as in (i). m Immunoblots with indicated antibodies of two independent and representative tumor biopsy lysates from indicated mutant mice. In all panels, Vinculin is a loading control.
The CARM1/NFIB partnership promotes cancer in SCLC GEMMs
To directly test the role of the CARM1-NFIB axis in SCLC pathogenesis, we generated a NfibR388K mutant mouse model using a CRISPR/Cas9 strategy. The donor DNA for the homologous recombination-driven mutation was designed to replace R388 with a lysine residue, and at the same time, disrupt an EcoRV site in the Nfib gene (Supplementary Fig. 3h). The loss of the EcoRV site made for efficient genotyping (Supplementary Fig. 3i). NfibR388K homozygous mice were viable and displayed no overt phenotypes. To further validate the establishment of a mutant NFIB mouse model, we immunoprecipitated NFIB from lung and spleen tissue of wild-type or NfibR388K homozygous mice and performed Western analysis using the anti-NFIBme2a antibody we developed (Fig. 1e). NFIB IPed from wild-type mice, but not NfibR388K homozygous mice, can be detected with the anti-NFIBme2a antibody, thus confirming that the R388 site is mutated in the mouse model (Supplementary Fig. 3j). We then crossed the NfibR388K mouse with the TKO GEMM of SCLC to obtained TKO;NfibR388K mice (Fig. 3g). In addition, we have previously generated a conditional knockout allele for CARM125, and this mouse was also crossed onto the TKO line to generate TKO;Carm1KO mice (Fig. 3g). Adenoviral-Cre intratracheal lavage was used to induce the conditional alleles recombination, and initiate cancer development. Both the loss of CARM1 and the mutation of the CARM1 methylation site on NFIB in the TKO mouse model resulted in dramatically reduced lung tumor burden, as well as the levels of the tumor cell proliferation biomarker, phospho-H3 (Fig. 3h–j). Both TKO;NfibR388K and TKO;Carm1KO mutant mice displayed significantly prolonged survival over the TKO control (Fig. 3k). Both models also display a striking decrease in liver metastasis (Fig. 3l). Loss of NFIB methylation was confirmed in both mouse models by Western blotting with the NFIBme2a antibody (Fig. 3m). These results support a role for CARM1-mediated arginine methylation of NFIB in the development of SCLC.
CARM1 and NFIB share the ability to promote an open chromatin state
To gain a molecular understanding of the NFIB/CARM1 pathway, we perform ATAC-seq and RNA-seq on the same tumor biopsies obtained for TKO, TKO;NfibR388K and TKO;Carm1KO mutant mice. ATAC-seq revealed massive changes in chromatin accessibility. Mapping of the accessibility in different genomic regions reveals that TKO;NfibR388K and TKO;Carm1KO show similar patterns compared to TKO (Supplementary Fig. 4a). A more detailed analysis of all changes in chromatin accessibility revealed that loss of CARM1 or mutation of NFIB cause over 30,000 sites of up- and downregulation (Fig. 4a, b). Importantly, when changes between TKO;NfibR388K and TKO;Carm1KO are compared, there is almost no difference in accessibility (Fig. 4c), which implies that the chromatin changes observed in TKO;NfibR388K and TKO;Carm1KO are matching. This is also true for the RNA-seq changes observed between tumor samples (Supplementary Fig. 4b). The similarity of the accessibility and expression changes, between TKO;NfibR388K and TKO;Carm KO tumors, is further validated by measurement of the linear correlation of these two data sets (Fig. 4d, e), the apparent grouping of TKO;NfibR388K and TKO;Carm KO when gene cluster analysis is performed on the transcriptome changes (Fig. 4f) and the principal component analyses (Supplementary Fig. 4c, d). As would be expected from these parallels, TKO;NfibR388K and TKO;Carm KO tumors display overlap in the Venn diagram depicting differentially expressed genes and chromatin accessibility changes (Supplementary Fig. 4e) and share GSEA signatures (Fig. 4g and Supplementary Fig. 5a). In addition, we found that previously published27 NFIB ChIP-seq peaks obtained from murine KP1 cells significantly overlapped with the down-regulated ATAC-seq peaks in TKO;NfibR388K and TKO;Carm1KO versus TKO control of which over 93% shared peaks between TKO;NfibR388K and TKO;Carm1KO (Supplementary Fig. 4f). As an example of some of the commonly regulated targets, which require both CARM1 and the NFIB methylation site, we show the browser tracks for LINGO1, SOX1, EPHB3, FOXA and MYCL (Supplementary Fig. 5b) and validate that these changes also occur at the protein level (Fig. 4h). Thus, CARM1 methylates NFIB and together they cooperate to reshape chromatin and coregulate gene expression.
a–c ATAC-seq analysis of regions of differential chromatin accessibility in TKO (control) vs TKO;NfibR388K vs TKO;Carm1KO. d Correlation of differential chromatin accessibility between control, NFIB mutant and CARM1 knockout tumors collected from n = 3 independent mice for each group. Pearson’s correlation coefficient (r) was calculated. e Correlation of differential gene expression between control, NFIB mutant and CARM1 knockout tumors collected from n = 3 independent mice for each group. Pearson’s correlation coefficient (r) was calculated. f Gene clusters analysis of the expression changes detected by RNA-seq. g The CARM1-NFIB axis regulates multiple oncogenic programs. Examples of top gene set enrichment analysis (GSEA) signatures associated with CARM1 and NFIB K388 methylation deficiency in cancer cells isolated from TKO (control) vs. TKO;NfibR388K vs. TKO;Carm1KO mutant mice. Normalized enrichment scores (NES) and false discovery rate (FDR) are provided (detailed statistics description in “Methods”). h Immunoblots with the indicated antibodies of tumor biopsy lysates from TKO (control), TKO;NfibR388K, TKO;Carm1KO mutant mice. Two independent samples are shown for each genotype. Vinculin shown as a loading control. i–k CARM1 inhibition attenuates the growth of three SCLC PDX models and in combination with Etoposide and Cisplatin (E/P) lead to tumor regression in NSG mice. Tumor volume quantification and immunoblots of the PDX were treated as indicated (n = 6 mice per group; P values were determined by two-way ANOVA with Tukey’s testing for multiple comparisons, data represent mean ± SEM).
CARM1 small molecule inhibitors have therapeutic potential for the treatment of SCLC
Importantly, transcription factors like NFIB are considered “undruggable” targets48. However, as an enzyme, CARM1 is the therapeutically tractable target, and several potent and specific CARM1 inhibitors (CARM1i) have recently become available31,32,33; these inhibitors are active in in vivo mouse models. We thus tested the hypothesis that CARM1i may be of therapeutic value for SCLC by comparing it to the standard of care, which is etoposide and cisplatin (EP) chemotherapy. Three independent SCLC patient-derived xenograft (PDX) models were tested with EP, CARM1i, and EP + CARM1i. Monotherapy with CARM1i or EP could significantly attenuate tumor growth in all three models but was insufficient to lead to cancer regression (Fig. 4i–k). Strikingly, the EP + CARM1i combination led to the most dramatic effect on tumor volume and led to partial regression in two out of three PDXs. Thus, pre-clinical studies have provided four independent lines of evidence (complementation assays using xenografts; a NfibR388K knock-in mouse, a CARM1 conditional knockout mouse, and CARM1 inhibitors) supporting CARM1 as a potential therapeutic target for SCLC patients.
Discussion
NFIB functions as an oncogene in small cell lung cancer26, by promoting SCLC metastasis and regulating chromatin accessibility27,28,49,50. There are also NFIB-independent mechanisms for SCLC metastases which involve the loss of the histone methyltransferase KMT2C51. Our study shows that loss of NFIB methylation inhibits the proliferation of SCLC cell lines as well as the progression of SCLC in mice (Fig. 3). These data further establish that the methylation of NFIB by CARM1 promotes the development of SCLC. Besides SCLC, NFIB was also reported to be amplified in ER-negative breast cancer and esophageal squamous cell carcinoma52,53,54, with possible oncogenic roles in these cancers. Furthermore, NFIB is reported to interact with transcription regulators to modulate its downstream transcription events. It was first shown to bind adjacent to the glucocorticoid receptor (GR) docking sequence on the promoter region of WAP gene and synergistically regulate the transcription of WAP55. Subsequent studies confirmed the association between NFIB and GR in gene regulation56,57. An interaction between NFIB and RFX1 was reported, and these transcription factors work together to modulate the expression of human growth hormone58. NFIB can regulate the transcription of androgen-dependent
genes as well as estrogen responsive genes, by interacting with FOXA159,60. Interestingly, tumor samples from TKO;NfibR388K and TKO;Carm1KO mice display reduced FOXA1 expression (Fig. 4h and Supplementary Fig. 4e), suggesting the existence of a possible feedback loop. Moreover, NFIB can interact with FOXP2 to activate neuronal maturation genes61, and with KDM4D to regulate the expression of adipogenic regulators62. In this study, we report the arginine methylation-dependent interaction between NFIB and TRIM29 (Fig. 2), adding to our mechanistic understanding of how NFIB transcription programs are regulated.
A functional link between NFIB and TRIM29 has been made before. Indeed, a comparison of transcription regulatory networks in breast cancer, identified an overlapping TRIM29/NFIB risk-associated regulon63. Within a network, regulons overlap because different transcription factors and co-factors can regulate similar gene sets, which suggests that components of overlapping regulons work together. TRIM29 can regulate other transcriptional networks apart from NFIB. Our data shows that TRIM29 is a positive regulator of the transcription of NFIB target genes. This coactivator function of TRIM29 has been reported for MMP9 and p63, in non-small cell lung cancer and cervical cancer cells, respectively64,65. How TRIM29 functions as a coactivator is not clear. However, the ability of TRIM29 to interact with DNA repair proteins may provide mechanistic hint. Indeed, DNA-PK, XRCC6 and RUVBL1/2 were identified as highly ranked TRIM29 binders by both us and others42 (Fig. 2j). DNA-PK has numerous functions in the regulation of transcription, including (1) the phosphorylation of the general transcription factors TBP and TFIIB that stimulates basal transcription66; (2) the regulation of transcriptionally poised RNAPII67; (3) its presence at active sites of transcription68; and (4) its ability to co-regulator the androgen receptor69. The DNA-PK Ku regulatory proteins (XRCC6/5) were also identified as TRIM29 binders, and early studies using extracts from either DNA-PK- or XRCC6/5-null cells displayed decreased transcription of multiple promoters70. Moreover, RUVBL1/2, also known as Pontin and Reptin, have numerous roles in the control of transcription71. We also found that TRIM29 binds PRMT1 and the SWI/SNF complex (Fig. 2j), which both have clear roles in transcriptional regulation.
SCLC accounts for approximately 15% of lung cancer and is a highly malignant and nearly uniformly fatal disease. To date, no targeted therapy has been approved for SCLC and the disease remains commonly treated with conventional chemotherapy, inevitably leading to acquired resistance and relapse. Lack of novel therapeutics is in part due to the unique biology of SCLC, which is driven by mutations in tumor suppressors and amplifications of transcription factors like NFIB and MYC. Indeed, dysregulated transcriptional programs can drive transformation, giving rise to what has been termed “transcriptional addiction” in cancer72. Transcription factors are difficult proteins to target with small molecule inhibitors because of their lack of enzymatic activity for chemical intervention. However, transcription factors can be modified by enzymes to regulate their activity, and they can recruit proteins with enzymatic activity to function as transcriptional coregulators. These enzymes could serve as therapeutic targets as a workaround for this issue. Currently, the therapeutic standard of care for SCLC is etoposide and cisplatin (EP) chemotherapy. We have shown that arginine methylation of NFIB is important for the progression of small cell lung cancer (Figs. 3 and 4), and elimination of this methylation either by removal of CARM1 or by mutation of the NFIB methylation site, significantly prolongs the survival of the SCLC mouse model (Fig. 3). Unlike NFIB, CARM1 activity can be efficiently inhibited by a number of small molecule compounds. Some of these compounds have exhibited anti-tumor activity in preclinical studies of leukemia and lymphoma34,35. In the future, the treatment of an SCLC GEMM with CARM1 inhibitors will further determine their therapeutic values of targeting this pathway for this cancer type (Fig. 5).
NFIB is methylated by CARM1, which subsequently recruits TRIM29 and possibly other effectors proteins. TRIM29 itself likely part of a protein complex that promotes the transcriptional activity of NFIB. CARM1 inhibitors (CARM1i) will attenuate the oncogenic properties of NFIB amplification.
