Slug interacts with PRMT5 and LSD1
To gain the mechanistic insight into the Slug mediated transcriptional regulation, we used the affinity purification and mass spectrometry coupled approach to survey the interactome of Slug. Whole cell extracts were prepared from HEK293T cells expressing Flag-Slug and subjected to purification using an anti-Flag affinity gel. Mass spectrometric analysis indicates that Slug immunoprecipitates contain LSD1 and PRMT5 (Fig. 1A, 1B). The presence of LSD1 and PRMT5 in the Slug interactome was validated by western blotting analysis of the column eluates with LSD1 and PRMT5 antibodies (Fig. 1C).
To ascertain the interactions among Slug, PRMT5 and LSD1, we performed the co-immunoprecipitation experiments with endogenous and exogenous proteins. We found that immunoprecipitation (IP) of endogenous Slug from MDA-MB-231, SUM159 and Hs578T cells brought down LSD1 and PRMT5, suggesting that Slug interacts with these proteins (Fig. 1D). Reciprocally, IP with antibodies against endogenous LSD1 or PRMT5, the remaining two proteins can also be detected (Fig. 1D). Moreover, we carried out the co-immunoprecipitation experiments in HEK293T cells expressing tagged Slug, PRMT5 or LSD1 as indicated in Fig. 1E, and detected the association among Slug, PRMT5 and LSD1. Consistent with these results, the Glutathione-S-transferase (GST) pull-down assay further supported the interactions among Slug, PRMT5 and LSD1 (Fig. 1F). Combined, these experiments indicate that Slug specifically interacts with PRMT5 and LSD1 in vivo.
Molecular basis for the interactions among Slug, PRMT5 and LSD1
To further delineate the interactions among Slug, PRMT5 and LSD1, a series of truncated mutants of these proteins were generated and transfected into HEK293T cells to map the domains critical for their association. The N-terminal Slug (1–127 aa) includes the SNAG domain of Slug, and the C-terminal Slug (128–268 aa) contains the conserved zinc finger motif (Fig. 2A). Expressing these two Slug deletions with full length PRMT5 (PRMT5-FL) or LSD1 (LSD1-FL) in HEK293T cells, we found that the C-terminal region of Slug is responsible for its interaction with PRMT5 (Fig. 2B), whereas the N-terminal region of Slug associates with LSD1 (Fig. 2C).
PRMT5 contains three functional domains: the N-terminal TIM barrel region (1–303 aa), the middle Rossmann-fold segment (304–460 aa) and the C-terminal β-barrel domain (461–637 aa) (Fig. 2A). To identify the region responsible for the PRMT5 interaction with Slug or LSD1, we generated various PRMT5 domain-deletion mutants and expressed them with Slug or LSD1 in HEK293T cells. The N-terminal truncation of PRMT5 retained the ability to bind Slug (Fig. 2D). Interestingly, all PRMT5 mutants interact with LSD1, while the N-terminal-truncated PRMT5 are mostly responsible for interaction with LSD1 (Fig. 2E).
The N-terminal region of LSD1 comprises a SWIRM domain (1–276 aa) and the larger C-terminal segment of LSD1 includes a catalytic amine oxidase (AO) domain (277–876 aa) (Fig. 2A). The domain deletion mutants of LSD1 were co-transfected with Slug or PRMT5 into HEK293T cells. Co-IP analysis demonstrated that both the N-terminal and C-terminal region of LSD1 were able to interact with PRMT5 or Slug, whereas the interaction of the C terminus of LSD1 and PRMT5 (Fig. 2F) or Slug (Fig. 2G) was less significant. Collectively, our findings further support the specific associations among Slug, PRMT5 and LSD1, and provide the details of the molecular interactions relevant for the formation of the Slug-PRMT5-LSD1 complex, as schematically summarized in Fig. 2H.
Genome-wide analysis of transcriptional targets for Slug and its associated proteins LSD1 and PRMT5
To unravel the function and significance of the association among Slug, PRMT5 and LSD1, we analyzed the genome-wide transcriptional targets of the Slug-LSD1-PRMT5 complex by mining previously published ChIP-Seq datasets of Slug (GSE55421), PRMT5 (GSE130194) and LSD1 (GSE101150). We identified 7136 LSD1-specific binding peaks, 19,628 PRMT5-specific binding sites, and 73,702 Slug-specific binding sequences mostly residing in the promoter, intronic or intergenic regions (Fig. 3A, q (FDR) value cut off of 0.05). The data were then cross-analyzed for overlapping binding sites at the promoters for potential co-targets of Slug, PRMT5 and LSD1. A total of 87 specific promoters targeted by Slug, PRMT5 and LSD1 were identified (Fig. 3B). Gene ontology (GO) analysis with Metascape online analysis tool (https://metascape.org/) was applied to uncover various cellular events for the genes corresponding to these co-occupied promoters. These biological processes include cytoskeleton organization, cell morphogenesis, metabolism and development (Fig. 3C).
Importantly, Slug, PRMT5 and LSD1 exhibited similar peaks on the proximal promoter region of the EMT genes such as E-cadherin (CDH1) and vimentin (VIM) (Fig. 3D). Analysis of the genomic distributions of Slug, PRMT5 and LSD1 revealed similar binding motifs (Fig. 3E), suggesting that these proteins are functionally connected. Quantitative ChIP (qChIP) analysis in MDA-MB-231 cells with the antibodies against Slug, PRMT5, LSD1 on selected genes (e.g., CDH1 and VIM) showed that the promoters of these two genes were strongly enriched (Fig. 3F), validating the results derived from the public ChIP-seq datasets. In addition, qChIP analysis with the antibodies against H3K4me2 and H3K9me2 (two LSD1 substrates), H4R3me2s and H3R2me2s (two PRMT5 targets) revealed that the target promoters of CDH1 and VIM were specifically marked with H3K4me2, H3K9me2, H4R3me2s and H3R2me2s (Fig. 3G), further supporting the occupancy of these promoters by PRMT5 or LSD1.
Transcription regulation of CDH1 and VIM by the Slug-PRMT5-LSD1 complex
We then evaluated the regulation of E-cadherin and vimentin by the Slug-PRMT5-LSD1 complex. We found that Slug, PRMT5 and LSD1 co-occupied the promoters of E-cadherin and vimentin through the ChIP assays using antibodies against Slug, PRMT5 or LSD1 in MDA-MB-231 cells (Fig. 4A, upper panel). To further evaluated the conjecture that Slug, PRMT5 and LSD1 function in the same protein complex at the E-cadherin and vimentin promoters, ChIP/Re-ChIP experiments were carried out on the representative target gene CDH1 and VIM promoters in MDA-MB-231 cells. Soluble chromatin was first immunoprecipitated with antibodies against Slug, PRMT5 or LSD1. The immunoprecipitates were subsequently re-immunoprecipitated with indicated antibodies. The results demonstrated that the CDH1 and VIM promoters initially immunoprecipitated with antibodies against Slug could be re-immunoprecipitated with antibodies against LSD1 or PRMT5 (Fig. 4A, lower panel). Similar results were obtained when an initial ChIP was performed with antibodies against LSD1 or PRMT5 (Fig. 4A). These results support that Slug, PRMT5 and LSD1 occupy the target promoters together.
To further confirm the transcription regulation of E-cadherin/vimentin by the Slug/PRMT5/LSD1 complex, we performed dual-luciferase reporter assay. As shown, Slug was able to repress the E-cadherin (WT) or activate vimentin (WT) promoter activity (Figure S1A). The mutated promoters did not respond to Slug (Figure S1A). Consistently, Slug was no longer able to repress the E-cadherin or activate vimentin promoter activity when PRMT5 or LSD1 was silenced, further supporting the targeting of E-cadherin and vimentin by the Slug/PRMT5/LSD1 complex (Figure S1A). To test the binding specificity, we mutated the binding motif CCCAAA (WT) to TCTGAG (Mut) and performed biotinylated oligonucleotide pull-down assay. We determined that Slug, PRMT5 or LSD1 specifically binds to the wild-type, but not to the mutant E-cadherin or vimentin probes (Figure S1B). Taken together, we concluded that Slug/PRMT5/LSD1 binds directly to CCCAAA motif in the E-cadherin and vimentin promoters.
We wondered how Slug, PRMT5 and LSD1 are recruited to the target genes. MDA-MB-231 cells were infected with shRNAs targeted to Slug, PRMT5 and LSD1 mRNA along with a shNTC control. The knockdown effects of shRNAs were confirmed by Western blotting (Figure S2). Q-ChIP experiments indicate that the depletion of Slug, PRMT5 or LSD1 led to a drastic reduction of the recruitment of the corresponding protein to the target promoters of CDH1 and VIM (Fig. 4B). Interestingly, whereas the Slug knockdown was associated with a reduced recruitment of PRMT5 and LSD1 on the CDH1 and VIM promoters, the depletion of either PRMT5 or LSD1 had only negligible effect on the recruitment of Slug (Fig. 4B), suggesting that PRMT5 and LSD1 are recruited on target promoters by Slug to act as transcription regulators.
Among the identified target genes of the Slug-PMRT5-LSD1 complex, CDH1 and VIM are important molecular markers of EMT. Downregulation of epithelial cell markers, like E-cadherin, and enhanced expression of mesenchymal markers such as vimentin, have been characterized as hallmarks during EMT process. To examine the transcription repression of CDH1 and transcription activation of VIM by the Slug-PRMT5-LSD1 complex, Slug was overexpressed in MCF10A cells, leading to decreased expression of E-cadherin and increased expression of vimentin at both the transcriptional and protein levels (Fig. 4C, 4D and Figure S3). Significantly, the alterations of E-cadherin and vimentin upon Slug overexpression were offset when PRMT5 or LSD1 was depleted in MCF10A cells, and this weakening trend was even more pronounced when LSD1 and PRMT5 were simultaneously knocked down (Fig. 4C, 4D and Figure S3). Since the promoter recruitment of Slug, PRMT5 and LSD1 is consistent with the E-cadherin and vimentin expression patterns, it appears that Slug functions in a dual mode in modulating gene expression during the EMT process.
To further gain the molecular insights into the dual regulatory mode mediated by the Slug-PRMT5-LSD1 complex on the E-cadherin and vimentin promoters, the expression of Slug, PRMT5 or LSD1 was individually silenced by their corresponding shRNA in MDA-MB-231 cells. Subsequent qChIP experiments showed that a marked increase in H3K9me2 but largely unchanged H3K4me2 on the vimentin promoter upon the depletion of Slug or LSD1, as well as a significant decrease of H3R2me2s but unchanged H4R3me2s on the vimentin promoter upon Slug or PRMT5 depletion (Fig. 4E). On the other hand, the qChIP analysis revealed that the levels of H3K4me2 were markedly increased at the E-cadherin promoter upon the depletion of Slug or LSD1, and the levels of H4R3me2s were significantly reduced at the E-cadherin promoter upon Slug or PRMT5 knockdown, whereas the levels of H3K9me2 and H3R2me2s did not change much upon knockdown of Slug, LSD1, or PRMT5 individually (Fig. 4E). Collectively, these experiments indicate that PRMT5 and LSD1 were recruited by Slug to suppress E-cadherin expression and activate vimentin transcription. Moreover, co-silencing of PRMT5 and LSD1 resulted in more prominent changes than individual knockdowns in the expression of the two EMT markers in MDA-MB-231 (Fig. 4F and G and Figure S3) and SUM159 (Fig. 4H and I and Figure S3) cells. These results further support the notion that Slug coordinates with PRMT5 and LSD1 to orchestrate the transcription of E-cadherin and vimentin. Previous studies have revealed that Slug is able to transcriptionally inhibit Claudin1 and transcriptionally activate ZEB1 [11, 12]. We found that the Slug-PRMT5-LSD1 complex mediated dual regulatory mode is also adapted to Claudin-1 and ZEB1 genes modulation (Figure S4A-C).
Given the roles of Slug, LSD1 and PRMT5 in EMT and cancer progression, we explored the functional coordination of the Slug-PRMT5-LSD1 complex in cell invasion by the transwell assay. Consistent with aforementioned the functional link between Slug, LSD1 and PRMT5, the positive effect of Slug overexpression on the invasive ability of MCF10A cells was partially attenuated by LSD1 or PRMT5 knockdown, and more severely reduced upon simultaneous depletion of PRMT5 and LSD1 (Fig. 4J). In addition, in the highly invasive MDA-MB-231 and SUM159 cells, the depletion of PRMT5 or LSD1 separately resulted in decreased invasive potential of these cells, and co-knockdown of PRMT5 and LSD1 led to more pronounced reduction in the cell invasion (Fig. 4K, 4L). Moreover, MDA-MB-231 with Slug depletion led to a decrease in the invasive potential, whereas the inhibitory effect of Slug knockdown on the invasiveness was not significantly rescued when PRMT5 or LSD1 was ectopically expressed in MDA-MB-231 cells (Fig. 4M). Taken together, these data support a vital role for the Slug-PRMT5-LSD1 complex in the regulation of invasion.
Co-inhibition of PRMT5 and LSD1 synergistically suppresses breast cancer progression
Despite the crucial role of Slug in modulating EMT and breast cancer metastasis, there is no effective method to directly target Slug pharmaceutically. Since PRMT5 and LSD1 are epigenetic enzymes that are more druggable than Slug itself, combined targeting of PRMT5 and LSD1 may be a more effective therapeutic strategy for the treatment of metastatic breast cancer. To investigate whether the combined inhibition of PRMT5 and LSD1 would synergistically impede breast cancer progression, we employed a PRMT5 inhibitory compound EPZ015666 and a selective LSD1 inhibitor SP2509. We found that SP2509 treatment effectively increased the level of H3K4me2 and H3K9me2 in MDA-MB-231 and SUM159 cells, without affecting the levels of H4R3me2s and H3R2me2s (Fig. 5A, B). In contrast, EPZ015666 treatment effectively decreased the levels of H4R3me2s and H3R2me2s, but did not alter the level of H3K4me2 and H3K9me2 (Fig. 5A, B) in MDA-MB-231 and SUM159 cells. Moreover, the two inhibitors did not affect the Slug, PRMT5 and LSD1 protein levels overall (Fig. 5A, B) and the interaction among them (Figure S5A). These results suggest that SP2509 and EPZ015666 effectively impaired the enzymatic activity of LSD1 and PRMT5 respectively in breast cancer cells.
The effects of EPZ015666 and SP2509 separately or combined on EMT of breast cancer cells were assessed by western blotting and qPCR. The results revealed that either EPZ015666 or SP2509 resulted in the induction of E-cadherin and the reduction of vimentin, and strikingly the treatment of both EPZ015666 and SP2509 led to stronger changes than single inhibitor treatment in the expression of these two markers at both protein (Fig. 5A, B and Figure S5B) and mRNA (Fig. 5C, D and Figure S5B) levels in breast cancer cells. In line with these findings, the treatment of SP2509 or EPZ015666 alone decreased breast cancer cell invasion potential, whereas the double inhibitor treatment led to a significantly stronger inhibitory effect (Fig. 5E, F).
We next evaluated the effects of EPZ015666 and SP2509 on breast tumor growth and metastasis in vivo. MDA-MB-231 cells stably expressing firefly luciferase were orthotopically implanted onto female nude mice mammary fat pat or intravenously injected into female nude mice for the study of spontaneous metastasis or seeding lung metastasis, respectively. After 10 days, the mice xenografted with breast cancer tumors were then divided into a control group and various treatment groups, including the SP2509 group (25 mg/kg), the EPZ015666 (100 mg/kg) group, the SP2509 plus EPZ015666 group. Tumor volumes were measured at indicated time with calipers and harvested on day 41. Monotherapy with either SP2509 or EPZ015666 alone partially inhibited the growth of the breast tumors, interestingly, the combined addition of SP2509 and EPZ015666 caused obvious synergistic effects in reducing tumor volumes (Fig. 5G-I). Moreover, the results revealed that, in the orthotopically implanted groups, the combination therapy with SP2509 and EPZ015666 was significantly more effective in decreasing spontaneous lung metastasis than monotherapy with either SP2509 or EPZ015666 alone (Fig. 5J, K). In addition, in the intravenous groups, the treatment with both drugs together resulted in a more dramatic decrease in experimental lung metastasis than the treatment with either inhibitor alone (Fig. 5L, M). Furthermore, we did not observe obvious drug toxicity during the course of treatment on mice, suggesting dosage and therapeutic regimen were well tolerated in mice (Figure S5C). Taken together, these data demonstrate that targeting both PRMT5 and LSD1 for inhibition is a potential novel therapeutic option for metastatic breast cancer patients.
PRMT5 and LSD1 are coordinately expressed in breast tumor specimens and their high expression portends poor prognosis in breast cancer patients
As the combined targeting of PRMT5 and LSD1 presents an effective approach against metastatic breast cancer, we extended our analysis to a clinically and pathologically relevant context. We therefore surveyed publicly available gene-expression data in The Cancer Genome Atlas (TCGA) database to compare PRMT5 or LSD1 expression in normal human breast tissues and breast cancer specimens. We found that breast cancer samples expressed significantly higher PRMT5 or LSD1 levels than normal breast cells, (n = 1222, P < 0.01, Fig. 6A). Next, Kaplan–Meier survival analysis with online tool (http://kmplot.com/analysis/) demonstrated that both enhanced LSD1 expression and higher PRMT5 expression were associated with shorter relapse-free survival (RFS) and overall survival (OS) of breast cancer patients (Fig. 6B). These data suggest that the enhanced expression of PRMT5 and LSD1 is associated with adverse outcomes of breast cancer patients.
If the combined targeting of PRMT5 and LSD1 is clinically meaningful, then PRMT5 and LSD1 likely would exhibit similar expression pattern in breast cancer patients. To this end, we first test the correlation between the expression level of PRMT5 and LSD1 in 114 human breast cancer samples using immunohistochemical (IHC) analysis. Interestingly, the expression of PRMT5 positively correlated with LSD1 in breast tumor specimens (Fig. 6C, D). To further ascertain this finding, we analyzed publicly available gene-expression datasets that have larger sample size of breast cancer patients. The results showed that PRMT5 expression was significantly positively correlated with the level of LSD1 in both RNA-sequencing dataset (n = 4712, P < 0.001, Fig. 6E, left) and DNA microarray dataset (n = 9639, P < 0.001, Fig. 6E, right).
We further analyzed the PRMT5 and LSD1 co-expression in distinct breast cancer subtypes. We found that PRMT5 expression was positively correlated with the expression of LSD1 in both mRNA and protein level among all subtypes including luminal, HER2 + and Basal-like subtype (Figure S6A, B). Interestingly, we noticed that this kind of positive correlation trend seemed to be a little bit more pronounced in basal-like breast patients with higher malignancy (Figure S6A, B). Collectively, these data support the observation that expression pattern of PRMT5 and LSD1 is similar in breast cancer patients.