1 ERG signaling in prostate cancer is driven through PRMT5-dependent methylation of the 2 Androgen Receptor 3 4 Zineb Mounir1*, Joshua M. Korn1, Thomas Westerling5, Fallon Lin1, Christina A. Kirby3, Markus 5 Schirle2, Gregg McAllister2, Greg Hoffman2, Nadire Ramadan2, Anke Hartung4¥, Yan Feng2, D. 6 Randal Kipp, Christopher Quinn, Michelle Fodor, Jason Baird, Marie Schoumacher1§, Ronald 7 Meyer1, James Deeds1, Gilles Buchwalter5∫, Travis Stams3, Nicholas Keen1, William R Sellers1, 8 Myles Brown5, Raymond Pagliarini1£ 9 1 Department of Oncology, 2 Developmental and Molecular Pathways, and 3 Center for Proteomic 10 Chemistry, Novartis Institutes for BioMedical Research, Cambridge, MA, 02139, USA. 11 4 Genomics Institute of the Novartis Research Foundation, Novartis Institutes for BioMedical 12 Research, San Diego, 02139, USA. 13 5 Department of Medical Oncology and Center for Functional Cancer Epigenetics, Dana-Farber 14 Cancer Institute and Harvard Medical School, Boston, MA, 02115, USA. 15 *Current address: Genentech, South San Francisco, CA, 94080, USA 16 ¥ Current address: Organovo, San Diego, CA 92121, USA. 17 § Current address: Laboratoires Servier, Neuilly-sur-Seine, FRANCE 18 ∫ Current address: Celgene Avilomics Research, Bedford, MA, 01730 19 £ Correspondence: Raymond Pagliarini [email protected] ; 250 Massachusetts 20 Avenue, Cambridge, MA, 02139. (617)- 871-4307 21 22 Abstract 23 The TMPRSS2:ERG gene fusion is common in androgen receptor (AR) positive prostate 24 cancers, yet its function remains poorly understood. From a screen for functionally relevant ERG 25 interactors, we identify the arginine methyltransferase PRMT5. ERG recruits PRMT5 to AR- 26 target genes, where PRMT5 methylates AR on arginine 761. This attenuates AR recruitment and 27 transcription of genes expressed in differentiated prostate epithelium. The AR-inhibitory 28 function of PRMT5 is restricted to TMPRSS2:ERG-positive prostate cancer cells. Mutation of 29 this methylation site on AR results in a transcriptionally hyperactive AR, suggesting that the 30 proliferative effects of ERG and PRMT5 are mediated through attenuating AR’s ability to induce 31 genes normally involved in lineage differentiation. This provides a rationale for targeting 32 PRMT5 in TMPRSS2:ERG positive prostate cancers. Moreover, methylation of AR at arginine 33 761 highlights a mechanism for how the ERG oncogene may coax AR towards inducing 34 proliferation versus differentiation. 35 36 Introduction 37 Prostate cancer (PC) is highly prevalent and lethal (Siegel et al, 2015). Drugs targeting 38 the Androgen Receptor (AR), a “lineage driver” of PC (Garraway & Sellers, 2006), are an 39 important therapeutic approach. AR is an androgen (i.e. testosterone)-activated nuclear hormone 40 receptor that regulates normal prostate gland growth and differentiation. In PC however, AR 41 facilitates unregulated proliferation (Mills, 2014). While it is unclear how AR and other lineage 42 factors switch between promoting normal lineage differentiation vs. tumor growth, it is 43 hypothesized that somatic mutations in additional genes may facilitate such changes (Garraway 44 & Sellers, 2006). Many PCs bear chromosomal translocations resulting in aberrant expression 45 of the ETS transcription factor ERG, most commonly through the TMPRSS2:ERG fusion (Shah 46 & Chinnaiyan, 2009). TMPRSS2:ERG, alone or in combination with additional genetic 47 alterations, promotes prostate tumor formation in mice (Baena et al, 2013; Chen et al, 2013; King 48 et al, 2009; Klezovitch et al, 2008; Mounir et al, 2014; Tomlins et al, 2008). ERG is recruited to 49 many AR target genes and represses AR-dependent transcription (Yu et al, 2010), suggesting 50 ERG functions at least in part through attenuating AR target gene expression. However, ERG 51 also regulates the expression of AR-independent genes thought to drive oncogenic function 52 (Klezovitch et al, 2008; Mounir et al, 2014; Tomlins et al, 2008; Wang et al, 2008). We 53 explored whether a deeper mechanistic understanding of ERG proliferative function could yield 54 therapeutic insights into targeting this key PC oncogene. 55 56 Results and Discussion 57 To identify genes that selectively facilitate the growth of TMPRSS2:ERG positive PC 58 cells, we performed a pooled short hairpin RNA (shRNA) screen in TMPRSS2:ERG and AR- 59 positive VCaP prostate cancer cells, using ERG-negative 22Rv1 cells as a control (Materials and 60 Methods). The shRNA pool targets 648 genes involved in transcriptional and epigenetic 61 regulation (Supplementary file 1). While ERG shRNAs were not in the pool, AR shRNAs were 62 preferentially depleted from VCaP cells, underscoring AR dependence in this cell line. Thirty 63 two (32) genes showing VCaP-selective shRNA depletion (Materials and Methods) were 64 considered for further study (Figure 1A; Supplementary file 1). 65 We next narrowed the shRNA screen hit list by focusing on candidates more likely to be 66 ERG interacting proteins. We immunoprecipitated ERG from VCaP cells, and then identified 67 co-immunoprecipitated proteins by mass spectrometry (Materials and Methods). Identified 68 proteins (Supplementary file 2) included AR and DNA-PKcs, previously known ERG interactors 69 (Brenner et al, 2011; Yu et al, 2010). Eight of the VCaP-selective shRNA screen hits that also 70 co-immunoprecipitated with ERG were further validated by directed ERG co- 71 immunoprecipitation experiments in VCaP cells. Of these, AR and PRMT5 were the only 72 proteins that co-immunoprecipitated with ERG but not IgG control; these interactions were not 73 overtly influenced by exposure to an androgen analog (R1881, Figure 1B; Figure 1—figure 74 supplement 1A). We next tested whether the ERG/PRMT5 interaction is observed in other 75 models. PRMT5 co-immunoprecipitated with ERG in 22Rv1 cells ectopically expressing ERG. 76 This interaction was still observed upon expression of ERG bearing mutations in the DNA 77 binding domain (“Dx”, Figure 1C), suggesting DNA binding is not required for the 78 ERG/PRMT5 interaction. Reciprocal co-immunoprecipitation experiments using overexpressed 79 ERG and PRMT5 in AR-negative 293 and PC3 cells suggest the ERG/PRMT5 interaction can 80 occur in the absence of AR (Figure 1—figure supplement 1B). Further work in 293 cells using 81 truncated ERG constructs suggested that the conserved ETS DNA binding domain of ERG was 82 necessary for the observed co-immunoprecipitation with PRMT5 (Figure 1—figure supplement 83 1C). Given this evidence that ERG and PRMT5 co-exist in a protein complex, we focused 84 further efforts on PRMT5, as to our knowledge it has not been previously linked to ERG biology. 85 To validate the growth effects of PRMT5 knockdown, we transduced ERG-positive 86 VCaP cells, and ERG-negative 22Rv1 and LNCaP PC cells, with three independent doxycycline 87 (Dox)-inducible shRNA vectors targeting PRMT5 and a non-targeting control shRNA (NTC). 88 PRMT5 knockdown was robust in all cell lines (Figure 1—figure supplement 1D). Robust 89 growth inhibition was observed in VCaP cells; in contrast PRMT5 knockdown had no growth 90 effects in ERG-negative 22Rv1 cells, and only minor effects in ERG negative LNCaP cells 91 (Figure 1D-F). Deletion of methylthioadenosine phosphorylase (MTAP), which is common 92 across cancers, is a major determinant of sensitivity to PRMT5 inhibition (Kryukov et al, 2016; 93 Mavrakis et al, 2016); as VCaP, LNCaP, and 22Rv1 cells are all MTAP intact, the observed 94 sensitivity of VCaP to PRMT5 shRNA is not due to MTAP deletion. The project Achilles 95 shRNA screen dataset (Kryukov et al, 2016) contains three prostate cancer cell lines (VCaP, 96 22Rv1 and TMPRSS2:ERG positive NCI-H660) and one PRMT5 hairpin likely to have minimal 97 off-target effects. This shRNA shows a trend of sensitivity in ERG-positive lines, in agreement 98 with our findings (Figure 1—figure supplement 1E). 99 PRMT5 is a protein arginine methyltransferase that regulates multiple signaling pathways 100 through the mono- and symmetric di-methylation of arginines on its target proteins (Yang & 101 Bedford, 2013).To determine whether the antiproliferative effects of PRMT5 knockdown in 102 ERG positive VCaP cells were mediated through methyltransferase activity, we expressed 103 shRNA-resistant wild-type PRMT5, or a catalytically inactive G365A/R368A double mutant 104 (Materials and Methods) (Antonysamy et al, 2012) along with PRMT5 shRNA in VCaP cells. 105 WT PRMT5, but not the G365A/R368A mutant, rescued the effects of PRMT5 knockdown on 106 VCaP cell proliferation (Figure 1G), indicating a requirement for PRMT5 catalytic function to 107 support VCaP proliferation. 108 To understand pathways affected by PRMT5, we performed transcriptional profiling of 109 PRMT5 knockdown in VCaP cells, followed by the identification of significantly altered 110 pathways (Figure 2A; Supplementary file 3; see Materials and Methods). Among these, AR 111 activation was the second most significantly affected pathway, and is a key pathway in common 112 with previous reports of ERG knockdown in VCaP cells (Chen et al, 2013; Mounir et al, 2014; 113 Yu et al, 2010). AR pathway upregulation was apparent using multiple published AR gene 114 signatures (Figure 2—figure supplement 1). Using quantitative PCR of reverse transcribed RNA 115 (qRT-PCR), we confirmed that knockdown of either PRMT5 or ERG increased the expression of 116 the AR target genes PSA, NKX3-1 and SLC45A3 (Figure 2B). Expression of shRNA-resistant 117 WT PRMT5, but not the G365A/R368A mutant, rescued the effects of PRMT5 knockdown on 118 AR target gene expression (Figure 2C; Figure 2—figure supplement 2A), demonstrating that 119 PRMT5 methyltransferase activity is required for repression of AR target genes. The effect of 120 PRMT5 knockdown was restricted to genes co-regulated by both AR and ERG, as PRMT5 121 knockdown did not affect previously published (Mounir et al, 2014) AR-independent ERG target 122 genes in VCaP (Figure 2—figure supplement 2B). In addition, PRMT5 knockdown did not 123 induce AR target gene expression in ERG negative 22Rv1 or LNCaP PC cells (Figure 2—figure 124 supplement 2C).