Pharmacological inhibition of PRMT7 links arginine monomethylation to the cellular stress responses

Protein arginine methyltransferases (PRMTs) regulate diverse biological processes and are increasingly being recognized for their potential as drug targets. Here we report the discovery of a potent, selective and cell active chemical probe for PRMT7. SGC3027 is a cell permeable prodrug, which in cells, is converted to SGC8158, a potent, SAM-competitive PRMT7 inhibitor. Inhibition or knockout of cellular PRMT7 resulted in drastically reduced levels of arginine monomethylation of HSP70 family members and other stress-associated proteins. Structural and biochemical analysis revealed that PRMT7-driven in vitro methylation of HSP70 at R469 requires an ATP-bound, open conformation of HSP70. In cells, SGC3027 inhibited methylation of both constitutive and inducible forms of HSP70, and led to decreased tolerance for perturbations of proteostasis including heat shock and proteasome inhibitors. These results demonstrate a role for PRMT7 and arginine methylation in stress response.

P rotein arginine methyltransferases (PRMTs) methylate arginine residues in histone and non-histone proteins in a mono, and symmetric or asymmetric dimethyl manner 1,2 . Arginine methylation of both histone and non-histone substrates has major roles in transcription and chromatin regulation, cell signaling, DNA damage response, and RNA and protein metabolism 3,4 . PRMT7, a member of the PRMT family, has been functionally implicated in a wide range of cellular processes including DNA damage signaling [5][6][7] , imprinting 8 , and regulation of pluripotency [9][10][11] . Recently several elegant studies using Prmt7 knockout mouse models also revealed the role of this methyltransferase in maintenance of muscle satellite cell quiescence, muscle oxidative metabolism, and B cell biology [12][13][14] . Although these studies have greatly expanded our understanding of PRMT7 biology, it remains an understudied member of the PRMT family with poor understanding of its cellular substrates.
PRMT enzymes display methylation preference for RGG/RG motifs enriched at protein-protein interfaces, whereas PRMT7 has been reported to target RXR motifs in arginine and lysinerich regions 15,16 . PRMT7 is the sole evolutionary conserved class III PRMT enzyme, the subfamily which carries out only monomethylation of arginine [17][18][19] . Other PRMT family members such as PRMT1 or PRMT5 catalyze arginine dimethylation in an asymmetric or symmetric manner, respectively, playing distinctly different downstream biological roles 1 . Remarkably, PRMT7mediated monomethylation of histone H4R17 allosterically potentiates PRMT5 activity on H4R3 20 . Thus, possible overlap between substrates for PRMT7 and other PRMT enzymes and their interplay is complex and for most part still largely unknown. The best-characterized PRMT7 substrates are histone proteins, such as H3, H4, H2B, and H2A 1,3,6,18 . Additional non-histone PRMT7 substrates such as DVL3 21 , G3BP2 22 , and eukaryotic translation initiation factor 2 alpha (eIF2α) 23 have also been described. Proteomics studies have identified an abundance of cellular monomethyl arginine-containing proteins [24][25][26][27] , however as other PRMT family members may be responsible for this methylation, it is not clear which of these substrates are dependent on PRMT7 as systematic studies of PRMT7 cellular substrates are lacking.
To enable further discovery of PRMT7 biology and to better explore its potential as a therapeutic target, here, we report a chemical probe of PRMT7 methyltransferase activity. SGC8158 is a potent, selective, and SAM-competitive inhibitor of PRMT7. To achieve cell permeability, we utilize a prodrug strategy where upon conversion of SGC3027 by cellular reductases, the active component, SGC8158, potently and specifically inhibits PRMT7driven methylation of cellular substrates. A systematic screen of arginine monomethylated proteins dependent on PRMT7 in cells identifies several RG, RGG, and RXR motif proteins. HSP70 family members involved in stress response, apoptosis, and proteostasis are PRMT7 substrates in vitro and in cells. Our data shows that PRMT7 methylates HSPA8 (Hsc70) and HSPA1 (Hsp70) on R469, which resides in a highly conserved sequence in the substrate-binding domain. SGC3027 inhibits the PRMT7driven methylation impacting the thermotolerance and proteostatic stress response in cells.

Results
PRMT7 chemical probe compound characterization. S-adenosyl methionine (SAM) is a co-factor required by all methyltransferases. To identify potential PRMT7 inhibitors, a library of SAM-mimetic small molecules was assembled and screened by a scintillation proximity assay (SPA) against PRMT7 using 3 H-SAM and a histone H2B peptide (residues 23-37) as a substrate ( Supplementary Fig. 1a). This screening resulted in the identification of SGC0911 with IC 50 of 1 µM (Fig. 1a). Further derivatization of this hit yielded SGC8172, which displayed improved potency (IC 50 < 2.5 nM) but was not selective for PRMT7 (Supplementary Table 1). Additional optimization of the methylene linker length between the terminal amine moiety of SGC0911 and the adenosine core structure resulted in SGC8158, a potent PRMT7 inhibitor (IC 50 < 2.5 nM) (Fig. 1a, b), which showed good selectivity over a panel of 35 methyltransferases including PRMTs ( Fig. 1c; Supplementary Tables 2 and 3) and kinases ( Supplementary Fig. 2). Importantly, we also developed a negative control compound (SGC8158N) which was markedly less potent (14 ± 2 µM) against PRMT7 (Fig. 1a, c) and other protein methyltransferases (Supplementary Table 2). Binding of SGC8158 to PRMT7 was also confirmed by surface plasmon resonance (SPR; Supplementary Fig. 1b, c). From kinetic fitting, a K D value of 6.4 ± 1.2 nM, k on of 4.4 ± 1.1 × 10 6 M −1 s −1 and k off of 2.6 ± 0.5 × 10 −2 s −1 were calculated from triplicate experiments. We next investigated the mechanism of action (MOA) of SGC8158 by determining the IC 50 values at various concentrations of substrate (SAM and peptide). With increasing concentrations of SAM in the presence of a constant peptide concentration, we observed higher IC 50 values, which indicated a SAM-competitive pattern of inhibition ( Supplementary Fig. 3). However, no change in IC 50 value was observed as the concentration of peptide was increased at fixed concentration of SAM indicating a noncompetitive pattern of inhibition with respect to peptide substrate ( Supplementary Fig. 3).
Human PRMT7 shares 93% sequence identity with Mus musculus PRMT7 (MmPRMT7), and the key residues involved in SAM and substrate-binding sites are conserved between the two species. To provide further insight into the mechanism of action of SGC8158, we solved the crystal structure of full-length MmPRMT7 in complex with SGC8158 refined to 2.4 Å resolution (referred to here as MmPRMT7_SGC8158). PRMT7 is a pseudodimer in nature, composed of catalytically active (N-terminal) and inactive (C-terminal) methyltransferase domains 28 . The structure showed that SGC8158, interacted only with the catalytically active N-terminal methyltransferase domain of PRMT7 in MmPRMT7_SGC8158 (Fig. 1d). Clear electron density was observed for the ribosyl and biphenylmethylamine moieties of SGC8158 ( Supplementary Fig. 4). The ribosyl moiety of SGC8158 is almost identical in position to that of SAH in SAHbound MmPRMT7 (MmPRMT7_SAH), (PDB ID: 4C4A), explaining the SAM-competitive kinetics observed in our assays ( Supplementary Fig. 2). The biphenylmethylamine moiety of SGC8158 extends toward the conserved THW loop region (residues Arg311 to Met315) and displaces the Trp314 side chain, to occupy a hydrophobic pocket composed of Trp282, Phe146, Tyr48, Met296, Arg44, and Arg311 side chains, and forms an edge-to-face π-stacking interaction with Trp282 (Fig. 1d). Compared to MmPRMT7_SAH (PDB ID: 4C4A), the His313 and Trp314 residues in THW loop were distorted in SGC8158-bound MmPRMT7 and could not be modeled (Fig. 1d, e). The PRMT-conserved THW loop is part of the activesite pocket and involved in substrate binding 29 , but interestingly SGC8158 did not affect the peptide substrate binding (Supplementary Fig. 3). Comparison of the THW loop region in MmPRMT7 with all other PRMTs showed that this loop has a variable length and conformation (Fig. 1f), suggesting that it likely has a role in the selectivity of SGC8158 for PRMT7. Taken together, these data indicate that SGC8158 is a potent and selective inhibitor in vitro that binds in the adenosine region of the SAM-binding pocket of PRMT7.
Identification of PRMT7 substrates. In order to evaluate pharmacological inhibition of PRMT7 in cells, we first sought to identify a cellular biomarker of its methylation activity. PRMT7 has a distinct substrate preference for RXR motifs surrounded by basic residues 17 , and although RGG and RXR motifs are abundant in the proteome 16 , very few have been validated in the cellular context. The fact that PRMT7 localization is mostly cytoplasmic (Fig. 2a, b), and most of the previously investigated substrates are histone proteins (i.e. nuclear), led us to undertake a proteomics-based exploration of potential substrates of PRMT7. Wild-type (WT) and PRMT7 knockout (KO) HCT116 cells were subjected to SILAC (stable isotope labeling by/with amino acids in cell culture) and monomethyl arginine immunoprecipitation followed by mass spectrometry analysis that included a targeted list of HSPA8 peptides (to ensure MS2 quantitation) within the data-dependent acquisition (DDA) cycle. significantly differentially methylated peptides representing 24 unique proteins were identified. Twenty-one peptides (from 18 proteins) were previously reported as arginine methylated 30 (highlighted in Fig. 2c, Supplementary Table 4). The analysis of total protein levels in PRMT7 KO and WT cells showed no significant change in protein abundance for the differentially methylated peptides indicating that the observed reduction in methylation was due to reduced monomethlation activity as opposed to perturbation of total protein levels (Supplementary Table 4). Most of the identified methylated proteins were  associated with RNA metabolism (Fig. 2d). For several proteins such as HSPA8, HSPA6/1A/B no detectable levels of R469 methylated peptides were found in the immunoprecipitated samples originating from the PRMT7 KO cells, thus we performed validation and quantified their methylation in the input samples ( Supplementary Fig. 5). This analysis showed that HSPA8 peptide FELTGIPPAPR-469 is highly methylated in a PRMT7-dependent manner in HCT116 cells. Sequences surrounding R469 are highly conserved among HSP70 family members (Fig. 2e), including the constitutively expressed HSPA8, and stress inducible HSPA6/HSPA1 proteins. The PRMT7dependent HSP70 monomethylation was also detected using pan-monomethyl arginine antibody in HCT116 cell lysates ( Supplementary Fig. 6a, b).
PRMT7 methylates HSP70 proteins in cells. In order to determine HSPA8 methylation in other cell systems, we investigated the endogenous regulation of HSP70 methylation in PRMT7 KO (HCT116, C2C12 cells), siRNA knockdown (HEK293 and MCF7 cells) and mouse embryonic fibroblasts (MEFs) derived from WT or Prmt7 KO mice. The PRMT7 genetic knockout or knockdown resulted in decreased methylation associated with monomethyl arginine signal coinciding with the HSP70 specific signal (Fig. 3a, Supplementary Fig. 7). As HSP70 proteins are induced and have a role in heat shock response, we investigated whether the induced forms of HSP70, such as HSPA6, and HSPA1, are methylated by PRMT7. Heatshock exposure resulted in increased levels of the inducible HSP70 forms, coinciding with increased abundance of the arginine monomethylated signal indicating a matched rapid methylation by PRMT7 (Fig. 3b).
The re-expression of WT PRMT7 or the catalytic-dead mutant R44A in HCT116 PRMT7 KO cells indicated that only WT PRMT7 was able to methylate HSP70 proteins (Fig. 3c). Moreover, introduction of WT or R469A mutant HSP70 into HCT116 PRMT7 KO cells confirmed the methylation site of HSPA8/ HSPA1 as identified in the proteomic analysis dataset (Fig. 3d). R469 has previously been identified as a methylation site for PRMT4 and PRMT7 31 ; however, in MCF7 cells, we observed a major effect on HSP70 methylation driven by PRMT7 but not PRMT4 ( Supplementary Fig. 7) possibly indicating cell-typespecific effects. These results demonstrate robust PRMT7dependent methylation of both steady state and inducible HSP70 proteins in cells.
PRMT7 methylates the open form of HSP70. HSP70 proteins have dynamic structures with nucleotide and substrate-binding domains undergoing marked conformational changes upon nucleotide or substrate binding 32,33 . Posttranslational modification of various HSP70 proteins has been shown to modulate their activity and can alter the conformational landscape of modified chaperones [34][35][36] . As R469 resides in the substrate-binding domain, we examined the 3D protein structure to determine the accessibility of the R469 side chain. Available full-length structures of HSPA5 (65% sequence identity to HSPA8) and HSPA1A (86% identity) were analyzed for the positioning of R469. The structures of HSPA5 in both the ATP-bound open state 37 as well as the ADP-bound closed lid state 37,38 for the substrate binding and nucleotide binding domains, respectively, and HSPA1A 39 substrate-binding domain in the ADP-bound state were used. Analysis of these HSP70 structures indicated that R469 resides in a loop of the substrate-binding domain which is likely of limited accessibility in the ADP-bound form of HSP70 when the substrate-binding domain is in a closed conformation (Fig. 4a-c).
In order to establish whether PRMT7 methylates HSP70 in vitro, we performed full characterization of PRMT7 enzyme kinetics with HSPA8 as substrate. In agreement with structural analysis, our data indicated that PRMT7 methylates HSPA8 in the presence of ATP (Fig. 4d, e). Apparent kinetic parameters were then determined for HSPA8 methylation yielding a K m for SAM of 1.6 ± 0.1 µM (Fig. 4d), and K m for HSPA8 of 10.6 ± 0.1 µM (Fig. 4e). To test the specificity of this activity, we also tested the activity of PRMT7 on HSPA8 with a single arginine to lysine mutation (R469K). Consistent with previous findings, PRMT7 was completely inactive with the HSPA8-R469K mutant as substrate in the presence or absence of ATP (Fig. 4f). We further tested the effect of SGC8158 and SGC8158N on PRMT7dependent methylation of HSPA8. SGC8158 inhibited PRMT7 methylation of full-length HSPA8 in vitro with IC 50 = 294 ± 26 nM under balanced conditions. As expected, SGC8158N showed very poor inhibition with an IC 50 value estimated to be higher than 100 µM (Fig. 5a).
SGC3027, a prodrug form of SGC8158, inhibits PRMT7 in cells. Having established that PRMT7 methylates cellular HSP70, we returned to SGC8158 in order to determine its cellular activity. We observed no inhibition of cellular HSP70 methylation by this compound, likely due to its SAM-like structure and low cell permeability often associated with SAM analogs 40,4142 . To increase the cellular permeability, we employed the Trimethyl Lock prodrug strategy in which SGC8158 is derivatized with a quinonebutanoic acid that masks a secondary amine group and increases lipophilicity 43 . The resulting derivative, SGC3027, undergoes reduction in cells followed by rapid lactonization, releasing the active component SGC8158 (Fig. 5b). The same strategy was employed for negative control prodrug compound SGC3027N ( Supplementary Fig. 8c) SGC3027 and SGC3027N prodrug compounds were efficiently converted into the active component, SGC8158 and SGC8158N, respectively, in cells ( Supplementary Fig. 8). SGC3027 inhibited HSP70 methylation with IC 50 of 2.4 ± 0.1 μM (Fig. 5c, d) in C2C12 cells, and the inactive compound had no effect at 5 μM, the cellular IC 90 of SGC3027, and had a minimal effect at 10 μM (Fig. 5e). In addition, SGC3027, but not SGC3027N, was effective at reducing HSP70 methylation in several commonly used cell lines ( Supplementary Fig. 9). SGC3027 selectively inhibited PRMT7 but not PRMT1, 4, 5, 6, 9, and DOT1L (closest in vitro hits) at 5 μM exposure ( Supplementary Fig. 10). In Prmt7 KO MEFs, SGC3027 does not affect the methylation of HSP70 ( Supplementary Fig. 11), further confirming the on-target activity for PRMT7. Taken together these data demonstrate that SGC3027 is a selective and cell-active inhibitor of PRMT7.
PRMT7-driven methylation is cytoprotective in proteostasis disruption. HSP70 family members have key roles in nascent protein folding and refolding, as well as distinct roles in antiapoptotic responses 44 . To determine whether PRMT7-driven methylation contributes to protection from cellular toxic insults, we investigated cell survival in response to thermal and proteasome stress using the genetic Prmt7 KO models and a chemical biology approach employing the SGC3027 chemical probe. Prmt7 KO MEF cells were more sensitive to acute heat stress with fewer cells surviving the treatment (Fig. 6a) and more cells undergoing apoptosis (Fig. 6b). SGC3027, but not the negative control SGC3027N, showed a similar response (Fig. 6c, Supplementary  Fig. 12a), indicating that SGC3027 phenocopies the genetic ablation of Prmt7. We also investigated whether PRMT7 stress protection extends to proteasome stress using proteasome inhibitor bortezomib. Compared to WT cells, Prmt7 KO MEF cells were more sensitive to acute bortezomib-induced cell death, having a poorer recovery after 4 or 20 h exposure to bortezomib (Fig. 6d, Supplementary Fig 12b). SGC3027, but not SGC3027N, sensitized the WT Prmt7 MEFs to bortezomib, whereas neither compound affected Prmt7 KO cells (Fig. 6e, f). These results indicate that PRMT7-driven methylation has a cytoprotective role in stress response, whereas inhibition of PRMT7 catalytic function can sensitize cells to toxic stimuli. To gain further insight into how R469 methylation may regulate HSP70 function, we investigated the effects of mutation (R469K) or PRMT7 inhibition on the functional properties of HSP70. R469K mutation did not affect HSP70 (HSPA8) driven ATPase activity ( Supplementary  Fig. 13) or the ability of HSPA8 to bind to the co-chaperones STIP1 (HOP) or STUB1 (CHIP) (Supplementary Fig. 14). The R469K mutation, however, reduced the ability of HSP70 to refold heat-denatured luciferase in cells ( Supplementary Fig. 15) as well as diminished the extent of stress granule prevention by HSP70 overexpression (Fig. 6g, h). SGC3027 consistently phenocopied the luciferase refolding and stress granule prevention effects of the mutation ( Supplementary Fig. 15, Fig. 6g, h), further indicating selective modulation of cellular PRMT7 function. Thus, PRMT7 methylation of HSP70 proteins impacts the function of HSP70 in the cellular stress response.

Discussion
PRMT7 is a monomethyl arginine methyltransferase that has a role in muscle physiology and stem cell biology 3,10-13 . However, the PRMT7 substrates that mediate this biology are not well understood, and tools to selectively and temporally modulate PRMT7 catalytic activity are lacking. Here we report the selective PRMT7 chemical probe, SGC3027, and a closely related but inactive compound, SGC3027N, for use as a specificity control for  biological experiments. We also identified PRMT7 substrates using proteomics and further validated HSPA8 and related HSP70 family members as PRMT7 substrates by employing in vitro assays, genetic methods, and chemical biology. Our data suggest that PRMT7 activity has a role in HSP70 protein function, cellular thermotolerance, and proteasomal stress response. Therefore, SGC3027 may be a useful modulator of cellular proteostasis in stress response under physiological and pathological conditions. SGC8158 is a structural derivative of SAM, acts as a SAMcompetitive inhibitor of PRMT7, and occupies the adenosine pocket in the SAM-binding site of PRMT7. Other potent and selective SAM-like inhibitors of methyltransferases including EPZ004777 45 and SGC0946 for DOT1L 46 , and LLY-283 for PRMT5 47 feature extensively modified adenosine and methionine moieties that likely enhance the cell permeability of otherwise cell impermeable SAM. We employed an alternative prodrug strategy in which adding the quinonebutanoic moiety increased cell permeability and allowed for cellular reductases to generate the active compound, SGC8158. Notably, the cellular reductasedriven activation of SGC3027 to the active SGC8158 may vary among cell types depending on the abundance or activity of reductase enzymes. Thus, we recommend that the appropriate concentration range for use of these probes should be empirically evaluated in each experimental setting by monitoring a biomarker such as HSP70 methylation before evaluation of specific biological readouts.
SGC3027 sensitized cells to heat and proteasomal stresses indicating that PRMT7 catalytic function is required for normal physiological response to these stimuli. Prmt7 KO cells were also more sensitive to these stresses, indicating that the chemical probe phenocopies the genetic knockout effects. Heat stress and proteasome inhibition elicits orchestrated protective responses, with the central goal of maintenance of cell protein homeostasis, proteostasis. Failure to maintain proteostasis results in numerous diseases. For example, altered proteostatic balance due to rewiring of chaperome complexes has been observed in cancer cells, contributing to drug resistance 48,49 .   50,51 . Interestingly, PRMT7 was identified in a screen for sensitization to topoisomerase inhibitors in cancer cells 5 , suggesting a wider range of stressors against which PRMT7 may be protective. Our findings of PRMT7 inhibition leading to the sensitization of cells to bortezomib-induced cell death indicate potential therapeutic applications in cancers such as multiple myeloma and some lymphomas. It is possible that PRMT7 inhibition may lead to overcoming resistance to bortezomib or other therapies. Further work is needed to determine whether PRMT7 inhibition can indeed synergize with proteasome inhibitor drugs in therapeutically relevant cell systems, and these efforts will be aided by the use of SGC3027 as a chemical probe for PRMT7. The majority of PRMT7 substrates identified in this study were functionally implicated in RNA metabolism and splicing. Previous arginine methylome studies have also noted the abundance of mono and dimethyl arginine posttranslational modifications (PTMs) in proteins involved in splicing, transcription, and RNA metabolism [24][25][26][27]52,53 . Although about half of the PRMT7dependent methylarginine peptides have been described in various studies [24][25][26][27]54 , several identified arginine methylation sites have not been previously reported, for example, the putative acetyltransferase NAT16, however, the related NAT10 and NAT6 are monomethylated 54 . Highlighting the complex relationship between PRMT enzyme activities, the PRMT7 methylation sites identified in several proteins, EIF4G1, ALYREF, RBM3, SF3B2, HSPA8, EWSR1, and SNRPB, were also reported as PRMT1, PRMT4, and PRMT5 sites 30,31,[55][56][57] . We have chosen to focus on high confidence methylation of the HSP70 family as cellular substrates of PRMT7 leading us to further investigate the role of PRMT7 in stress response.
HSP70 R469 monomethylation by PRMT4 and PRMT7 was recently reported as contributing to transcriptional activation by retinoic acid receptor RAR 31 . Interestingly, PRMT4 methylation of HSP70 did not seem to require an ATP-driven conformational change 31 , while we found that PRMT7-driven methylation occurs when HSP70 adopts an open, ATP-bound conformation that promotes R469 accessibility. In all of the examined structures of the substrate and ADP-bound forms of HSP70, R469 packs into the lid subdomain of the substrate-binding domain to ensure flexible, yet stable, interaction with the client protein 39 , but rendering this residue inaccessible to modifying enzymes such as PRMT7. Although the substrate-binding domains and lid    Loss of PRMT7 decreases cell survival and increases apoptosis levels after heat shock. MEF cells were heat-shocked for 20 min at 44°C. Apoptosis was monitored immediately after the heat shock and cell number was determined 24 h later. The results shown are mean ± SEM of two biological replicates, each technical triplicate (a) and three technical replicates (b). Statistical significance was determined with unpaired Student t-test (two-tailed). c SGC3027 inhibition of PRMT7 activity decreases cell survival and increases apoptosis levels after heat shock. Cells were pretreated with 3 µM compounds for 2 days before heat shock. Cell number was determined as in a. The results shown are mean ± SEM of two biological replicates, each technical n = 5. Statistical significance was determined with oneway ANOVA with Tukey's post-hoc test. subdomains are more variable among HSP70 proteins than other regions, possibly due to the need of accommodating a large number of substrates 39 , the region surrounding R469 is highly conserved among HSP70 family members and species. PTMs of the HSP70 lid and tail domains have been reported to disrupt the interaction with co-chaperone CHIP TPR 58 . Although the R469containing loop resides in close proximity to the CHIP binding region, the R469K mutation did not affect the binding of CHIP.
The cytoprotective function of HSP70 proteins has been attributed to the modulation of protein refolding and transport of client proteins, as well as direct regulation of apoptotic signaling pathways 48,59 . Several HSP70 inhibitors have been reported and their utility is being explored in counteracting bortezomib resistance in multiple myeloma as well as therapeutic applications in other cancers [60][61][62][63] . PTMs of the HSP70 family members such as, lysine methylation, ubiquitination, acetylation, and phosphorylation have been reported 64 , and some of the PTMs stabilize the HSP70/HSP90/HSP40 and client protein complexes allowing the formation of antiparallel HSP70 dimers 65 . K561 trimethylation by METTL21A and dimethylation by SETD1A have been reported to result in the modulation of HSP70 affinity for client proteins or potentiation of AURKB activity, respectively [66][67][68] . Here, using genetic and pharmacological means, we identify R469 monomethylation as an abundant cellular modification that depends on PRMT7 catalytic function and correlates with the cytoprotective properties of PRMT7. Recently, it was reported that PRMT7 interacts and methylates eukaryotic translation factor eIF2α and regulates stress granule formation in response to various stresses 23 . Interestingly, EIF4G1 that is methylated by PRMT7 and PRMT1 56 also has a role in translation and stress granules [69][70][71] . Stresses such as proteasome inhibition induce the stress granules in an eIF2α phosphorylation-dependent manner 72 , whereas HSP70 proteins that modulate and prevent the stress granule formation have important roles in their dynamic regulation 72,73 .
Our data indicate that HSP70 R469 methylation by PRMT7 is important for stress granule response upon proteasome inhibition and provides additional links to the complex interplay of proteasome, translation, and chaperone systems acting on stress granules to ensure cell survival.
We uncovered aspects of PRMT7 biology associated with proteostasis, identified PRMT7 cellular substrates and validated HSP70 family members HSPA1/6/8 as PRMT7 substrates, whose methylation is likely to contribute to the cytoprotective and stress response function of PRMT7. SGC3027, together with its negative control SGC3027N, will be useful tools for further understanding PRMT7 function in physiological and disease states.

Methods
Protein expression and purification. Full-length PRMT7 was expressed in Sf9 cells grown in HyQ® SFX Insect serum-free medium (ThermoScientific). Cells were harvested and lysed, and the cleared lysate was incubated with 5 mL anti-FLAG M2-Agarose (Sigma) in 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 10% glycerol and washed with the same buffer with 500 mM NaCl. The pure recombinant protein was eluted from the column using the same buffer with 0.1 mg/mL FLAG peptide (Sigma). Pure PRMT7 was flash frozen and stored at −80°C. Full-length HSPA8 was overexpressed in Escherichia coli strain BL21(DE3) V2R-pRARE2 during an overnight induction with 0.5 mM isopropyl 1-thio-D-galactopyranoside at 18°C. Cells were suspended in 20 mM Tris-HCl (pH 8.0, 1 mM DTT, 300 mM NaCl). The clarified lysate was loaded onto a Hispur TM nickel-nitrilotriacetic acid column (ThermoScientific) and washed with buffer. Then, protein was eluted and concentrated. Protein purity was determined by SDS-PAGE and liquid chromatography-mass spectrometry (LC-MS).
Radioactive activity assay in vitro. Assays using biotinylated H2B (23-37) as a substrate were performed in buffer (20 mM Tris-HCl, pH 8.5, 0.01% Tween-20, and 5 mM DTT) containing 5 nM PRMT7, 1.1 µM 3 H-SAM (Cat.# NET155-V250UC; Perkin Elmer; www.perkinelmer.com) and 0.3 µM H2B (23-37). The reaction mixtures were incubated for 60 min at 23°C. To stop the enzymatic reactions, 10 µL of 7.5 M guanidine hydrochloride was added, followed by 180 µL of buffer (20 mM Tris, pH 8.0), mixed and then transferred to a 96-well FlashPlate (Cat.# SMP103; Perkin Elmer; www.perkinelmer.com). After mixing, the reaction mixtures in Flash plates were incubated for 1 h and the CPM were measured using Topcount plate reader (Perkin Elmer, www.perkinelmer.com). The CPM counts in the absence of compound for each dataset were defined as 100% activity. In the absence of the enzyme, the CPM counts in each dataset were defined as background (0%). The IC 50 values were determined using GraphPad Prism 7 software. For the kinetic analysis of HSPA8 methylation by PRMT7, the assay mixture contained 20 mM Tris-HCl, pH 8.5, 0.01% Tween-20, and 5 mM dithiothreitol (DTT), 1 mM MgCl 2 , 1 mM ATP where indicated, 250 nM PRMT7, fixed concentration (20 µM) of SAM, various concentrations (up to 40 µM) of HSPA8; or fixed concentration (10 µM) of HSPA8 with different concentrations of SAM (up to 20 µM). IC 50 determinations of SGC8158 and SGC8158N were performed at 150 nM PRMT7, close to K m values of both SAM (2 µM) and substrate HSPA8 (7 µM). To determine the mode of action, the experiments were performed in the presence of fixed biotinylated H2B peptide (residues 23-37) or HSPA8 substrate concentration and increasing SAM concentration or at fixed concentration of SAM and varying concentration of the substrate. Twenty µL of reaction mixtures were incubated at 23°C for 60 min. To stop reactions, 100 µL of 10% trichloroacetic acid (TCA) was added, mixed and transferred to filter-plates (Millipore; cat.# MSFBN6B10; www.millipore.com). Plates were centrifuged at 930 × g (Allegra X-15R-Beckman Coulter, Inc.) for 2 min followed by two additional 10% TCA washes and one ethanol wash followed by centrifugation. Plates were dried and 30 µL MicroScint-O (Perkin Elmer) was added to each well, centrifuged and removed. 50 µL of MicroScint-O was added again and CPM was measured using Topcount plate reader. The IC 50 values were determined using GraphPad Prism 7 software. , and BCDIN3D the incorporation of a tritium-labeled methyl group into biotinylated substrate was monitored using scintillation proximity assay (SPA). Briefly, a 10 μL reaction containing 3 H-SAM and substrate at concentrations close to the apparent K m values for each enzyme (balanced conditions) was prepared. The reactions were quenched with 10 μL of 7.5 M guanidine hydrochloride; 180 μL of 20 mM Tris buffer (pH 8.0) were added, and the mixture was transferred to a 96-well FlashPlate and incubated for 1 h. The counts per minute (CPM) was measured on a TopCount plate reader. The CPM in the absence of compound or enzyme was defined as 100% activity and background (0%), respectively, for each dataset [74][75][76] . Selectivity of SGC8158 against 342 kinases was evaluated in a kinase panel TR-FRET assay at 1 and 0.1 μM compound in duplicates as described before 77 . Briefly, tagged kinases were incubated with compound and staurosporine-based fluorescent probe where the binding was detected using Tb conjugated anti-tag antibody energy transfer to the probe. Excitation wavelength was 337 nm and fluorescent emission signal was measured for Tb (486 nm) and fluorescent probe (BODIPY, 515 nm). Specific and non-specific binding was determined in the absence and presence of unlabeled staurosporine.
PRMT4 assay: C2C12 cells were seeded in 12 well plates and next day treated with compounds for 48 h. After 48 h cells were lysed in lysis buffer and analyzed in western blot for BAF155R1064me2a levels normalized to BAF155 82 .
PRMT5 assay: C2C12 cells were seeded in 12 well plates and next day treated with compounds for 48 h. After 48 h cells were lysed in lysis buffer and analyzed in western blot for SmBB'-Rme2s levels normalized to SmBB' 47 .
PRMT1 assay: MCF7 cells were seeded in 12 well plates and next day treated with compounds for 48 h. After 48 h cells were lysed in lysis buffer and analyzed in western blot for histone H4R3me2a levels normalized to H4 76 .
PRMT9 assay: HEK293T cells were seeded in 12 well plates and co-transfected with 0.9 µg FLAG-tagged PRMT9 and 0.1 µg GFP-tagged SAP145 WT or R508K mutant using jetPRIME ® transfection reagent, following manufacturer instructions. After 4 h compounds were added and after 20 h cells were lysed in lysis buffer and analyzed in western blot for SAP145-Rme2s levels normalized to GFP 83 . DOT1L assay: THP-1 cells were seeded in 12 well plates and next day treated with compounds for 48 h. After 48 h cells were lysed in lysis buffer and analyzed in western blot for histone H3K79me2 levels normalized to H3 46 .
Western blot. Total cell lysates or cellular fractions (as indicated) were resolved in 4-12% Bis-Tris Protein Gels (Invitrogen) and transferred in for 1.5 h (80 V) onto PVDF membrane (Millipore) in Tris-Glycine transfer buffer containing 20% MeOH and 0.05% SDS. Blots were blocked for 1 h in blocking buffer (5% milk in PBS) and incubated with primary antibodies in blocking buffer (5% BSA in PBST: 0.1% Tween-20 PBS) overnight at 4°C. After five washes with PBST the blots were incubated with goat-anti-rabbit (IR800) and donkey anti-mouse (IR680) antibodies in Odyssey Blocking Buffer (LiCor) for 1 h at RT and washed five times with PBST. The signal was read on an Odyssey scanner (LiCor) at 800 and 700 nm. Band intensities for western blot analysis were determined using Image Studio Ver 5.2 (Licor). The uncropped blots are provided in the Source Data file.
Knockdown. Cells were transfected with 15 nM of either non-targeting siRNA or siRNA against PRMT7, PRMT4, or PRMT5 (Dharmacon) using Lipofectamine™ RNAiMAX, following manufacturer instructions. After 3 days, the protein levels were measured by western blot as described above.
Cell growth and apoptosis assay after heat shock. MEF WT and Prmt7 KO cells were heat-shocked in a water bath for 20 min at 44°C. For experiments with SGC3027 or SGC3027N, cells were pretreated with 3 µM compounds for 2 days before heat shock. Cell number was determined with Vybrant® DyeCycle™ Green, following manufacturer's instructions, 24 h after heat shock and apoptosis levels were determined with IncuCyte® Caspase-3/7 Reagent within 12 h after heat shock using IncuCyte™ ZOOM live cell imaging device (Essen Bioscience) and analyzed with IncuCyte™ ZOOM (2015A) software. Apoptosis levels and cell confluency were analyzed with IncuCyte™ ZOOM (2015A) software.
Cell growth after bortezomib treatment. MEF WT and Prmt7 KO cells were treated with bortezomib for 4 or 24 h, and the confluency monitoring was started at 4 h after bortezomib (30 nM) treatment. For experiments with SGC3027 or SGC3027N, cells were pretreated with 4 µM compounds for 2 days before bortezomib treatment (30 nM). After 20 h bortezomib, SGC3027 or SGC3027N were removed and SGC3027 or SGC3027N were replaced. Cell confluency was monitored right after bortezomib addition using IncuCyte™ ZOOM live cell imaging device (Essen Bioscience) and analyzed with IncuCyte™ ZOOM (2015A) software.
Cellular fractionation. Cells were trypsinized and 1 × 10 6 cells were centrifuged at 400×g for 5 min at 4°C. Cell pellets were resuspended in 200 µL of hypotonic buffer (10 mM HEPES pH 7.5, 10 mM KCl, 1.5 mM MgCl 2 , 0.3 M Sucrose, 1 mM TCEP, 0.1% Triton X-100 and protease inhibitors). The cell suspensions were incubated on ice for 15 min followed by centrifugation at 1300×g for 5 min at 4°C. The supernatants were collected and cleared by centrifugation at 18,000×g to produce the cytoplasmic fraction. The pellets were then washed in hypotonic buffer, centrifuged again, and resuspended in an equal volume of lysis buffer (as described above in PRMT7 cellular assay) to produce nuclear fraction. The fractions were analyzed by western blot as described above.
For the stress granule formation assessment, the immunofluorescence was performed in HeLa cells with cherry-G3BP1 84 as above using anti-TIAR and secondary anti-rabbit IgG Alexa647 antibodies. Cells were transfected with GFP-HSPA1 WT or R469K, and next day treated with 20 µM MG132 for 6 h. Stress granules were quantified on a percentage per cells basis counting the number of cells with at least one discrete G3BP1 foci that were also positive for TIAR from >100 cells. Three independent experiments were used for microscopy analysis performed with Quorum Spinning Disk Confocal microscope equipped with 405, 491, 561, and 642-nm lasers. Statistical significance was assessed using GraphPad Prism 7 software via Student's t-test (unpaired, 95% confidence interval) and oneway ANOVA with Tukey's post-hoc test. p-values < 0.05 were considered statistically significant.
LC-MS measurement of intracellular compounds concentration. C2C12 cells were plated in 6-well plates (2 × 10 6 per well). Next day 3 µM of SGC3027 or SGC3027N was added to the cells and incubated for indicated times. After incubation, cells were washed with PBS, trypsinized, and cell pellets were collected by centrifugation at 500 × g for 2 min. Pellets were mixed with 20 µL of acetonitrile, centrifuged for 1 min at 18,000 × g and supernatants were collected and analyzed by LC-MS. To generate the standard curves, SGC8158 and SGC8158N compounds in two-fold dilution series from 0.025 to 25 µM were utilized. SGC3027 and SGC3027N compounds were also run to ensure the separation of the peaks and sufficient difference in the retention times. Standard curves were prepared in PBS. We spiked the PBS with 10 mM DMSO stock then did 2-fold dilution series for the calibration curve. The compounds were extracted using two volumes of acetonitrile (i.e. for each 20 μL of solution 40 μL acetonitrile was used). In the preliminary experiment during method development, we observed no ion suppression due to PBS; that means the area under the chromatographic peak for same concentration was similar from water or PBS after acetonitrile extraction. However, we did not evaluate the extraction efficiency of these compounds from the cell matrix, hence the concentration reported were relative concentrations, not absolute. Chromatographic separations were carried out on an ACQUITY UPLC BEH C18 (2.1 × 50 mm, 1.7 µm) column. The mobile phase was 0.1% formic acid in water (solvent A) and 0.1% formic acid in acetonitrile (solvent B) at a flow rate of 0.4 mL min −1 . A gradient starting at 95% solvent A going to 5% in 4.5 min, holding for 0.5 min, going back to 95% in 0.5 min and equilibrating the column for 1 min was employed. A Waters SYNAPT G2-S MS equipped with an atmospheric pressure ionization source was used for MS analysis. In a typical MS acquisition setting, we used capillary voltage at 2.0 kV, sampling cone voltage at 20 V and the trap collision energy at 30.0 eV (detailed settings can be found in the Supplementary Table 7). MassLynx 4.1 software from Waters was used for data analysis with the QuanLynx module for quantification. Standard curves were generated by using the linear fit of mass peak areas and the known concentrations of SGC8158 and SGC8158N.
HSPA8 ATPase assay and luciferase refolding assay. ATPase assay was performed according to Cheng et al. 85 Briefly, purified HSPA8 or mutant HSPA8 (R469K) (10 µM) was incubated for 3 h at RT with or without purified full-length PRMT7 (0.5 µM) and with or without SAM (cold, 5 µM), as indicated, in buffer containing: 20 mM Tris-HCl (pH = 8.5), 5 mM DTT, 1 mM MgCl 2 , 1 mM ATP, 0.01% Triton X-100. Before the ATPase assay, the samples were tested for monomethylation levels in western blot. Processed HSPA8WT/MUT samples were diluted to 1 µM in assay buffer containing: 1.7 µM Hsp40 (Enzo), 0.017% Triton X-100, 100 mM Tris-HCl pH 7.4, 20 mM KCl and 6 mM MgCl 2 . Fifteen µL of this mixture was added into each well of a 96-well plate and 10 µL of 2.5 mM ATP was added to start the reaction. The reaction was stopped after 0 and 1 h incubation at 37°C with 50 µL of BIOMOL GREEN™ Reagent (Enzo Life Sciences). After 30 min incubation at RT the absorbance was measured at 620 nm. The phosphate concentration was calculated from standard curve, prepared following manufacturer's instructions. As ATP had to be present for HSPA8 methylation reaction, the initial phosphate concentration (time 0) was measured for background subtraction.
Luciferase refolding assay 86 was performed in HEK293T cells transfected with HSPA1 WT or R469K. The next day cells were treated with 50 μg mL −1 cyclohexamide and heated for 60 min at 45°C to inactivate luciferase and, after 2 h recovery at 37°C, luciferase activity was measured by using luciferase assay (Promega). Luciferase activity was normalized to unheated control samples.
CRISPR/Cas9-mediated genome editing. HCT116 cells were seeded onto 6-well plates at a density of 40,000 cells per well, 24 h before transfection. Cells were transfected using Lipofectamine 2000 (Life Technologies) according to the manufacturer's instruction. A total of 3 µg Cas9 expression vector, 1 µg of gRNA expression vector, and 0.4 µg of pEBMultipuro (Wako Chemicals) as a transfection marker were transfected. After 48 h of transfection, 1 µg mL −1 puromycin was added for selection. The colonies were isolated by limiting dilution. PRMT7 destruction for the isolated clones was confirmed by Sanger sequencing and western blotting.
CRISPR for mouse C2C12 cells. C2C12 PRMT7 KO clones were generated by cloning guide RNA (caccgGTCATGTAGCATGTCGGCAT/aaacATGCCGACA TGCTACATGAC) into LentiGuide-Puro vector (from Zhang lab, obtained from Addgene), following Zhang laboratory protocols. Lentivirus was produced using standard protocols. 48 h post transfection the supernatant was collected, filtered through 0.5 µm filter and used to infect C2C12 cells in presence of polybrene to final conc. of 8 μg mL −1 . After 24 h media was changed and after 2 days puromycin was added to final concentration of 2 μg mL −1 . 5 × 10 4 puromycin selected cells were electroporated with 0.5 μg Cas9-GFP plasmid using Neon transfection system (Life Technologies) (1650 V, 10 ms, 3 pulses). Next day, GFP-positive cells were sorted and plated in 96-well plates (1 cell per well). The clones were analyzed for PRMT7 expression and HSP70 monomethylation in western blot. The PRMT7 KO and PRMT7 catalytic mutant clones were genotyped by Sanger sequencing PCRamplified TA-cloned genomic DNA. The following primers were used: forward CCA TCC AAT TGA GGT CAG CG and reverse TGG ACA TTC TTG AGC ACC TTA GT. The premature stop codon in exon 3 or 4 resulted in PRMT7 KO. The mutation in one of the alleles delY35, A35S in addition to premature codons in exon 3 or 4 of other alleles resulted in expression of catalytically inactive PRMT7 protein (Supplementary Table 5).
Enrichment of monomethylarginine peptides. The samples were enriched for monomethylated peptides with PTMScan® Mono-Methyl Arginine Motif [mme-RG] Kit (Cell Signaling) according to manufacturer's instructions. Briefly, cells were washed three times with PBS, lysed with 9 M urea buffer and centrifuged for 15 min at 20,000 × g to remove cellular debris. The recovered proteins were quantified with a BCA Protein Assay Kit (ThermoFisher Scientific). For each replicate 5 mg of light and heavy amino acid-labeled cell lysate was combined, reduced with DTT (RT for 1 h), alkylated with iodoacetamide (RT for 15 min), digested with 100 μg of Lys-C (Wako, RT for 2 h), followed by 4-fold dilution with 20 mM HEPES pH = 8 and digested with 100 μg of Trypsin Gold (Promega) overnight at RT. The protein digests were acidified with trifluoroacetic acid to final concentration of 1%, purified using Sep-Pak C18 classic cartridge (Waters, #WAT051910) and lyophilized. Peptides were resuspended in IP buffer, centrifuged 5 min at 10,000 × g at 4°C and supernatant was incubated with antimonomethyl arginine beads for 2 h at 4°C. After 2 washes with 1× IP buffer, two washes with 10× diluted IP buffer and one wash with HPLC water, the peptides were eluted using 50 µL of 0.15% trifluoroacetic acid. The eluate was purified, concentrated using C18 spin tips (Pierce, #84850) and lyophilized. The peptides were digested again with 250 ng of Trypsin Gold (Promega) and purified using C18 spin tips.
Mass spectrometric analysis of monomethylarginine peptides. Monomethylarginine peptides were analyzed by nanoLC-MS using a home-packed spray tip formed on a fused silica capillary column (0.75 μm internal diameter, 350 μm outer diameter) using a laser puller (Sutter Instrument Co., model P-2000). C18 reversed-phase material (Reprosil-Pur 120 C18-AQ, 3 μm, Dr. Maisch) in methanol was packed [15 (±1) cm] into the column using a pressure injection cell. An Eksigent 425 nano HPLC system (Sciex, Framingham, MA) was coupled to an Orbitrap Fusion Lumos (ThermoFisher Scientific, Waltham, MA). The LC gradient was delivered at 200 nL min −1 and consisted of a ramp of 2-35% acetonitrile (0.1% formic acid) over 116 min, 35-80% acetonitrile (0.1% formic acid) over 19 min, 80% acetonitrile (0.1% formic acid) for 30 min, and then 7.5% acetonitrile for 29 min. The Orbitrap Fusion Lumos (Tune version 3.3) with Xcalibur (version 4.4) was operated in data-dependent acquisition (DDA) mode with survey scans performed at 120,000 resolution, AGC target of 5 × 10 5 , with a maximum fill time of 50 ms, 400-1500 m/z range. Top speed mode was used with a 1 s cycle time and 10 s dynamic exclusion. Fragment ions from MS/MS were detected in the orbitrap with 15,000 resolution, with an AGC target of 2 × 10 5 , max fill time 35 ms. Charge states 2-6 were included with higher collisional dissociation (HCD) energy set at 32%. In addition to the 1 s data-dependent MS/MS, in every cycle a targeted list of precursors was collected with the same settings used in DDA, except the AGC target was 1 × 10 5 (targeted masses; 599. 33 Monomethyl arginine data analysis. Raw files were searched using MaxQuant version 1.6.2.1 and the UP000005640 UniProt Release 2018_08 human database (Swiss-Prot reference containing 20,352 protein entries, downloaded on 24 October, 2018). PTM scores for monomethyl arginine were generated using the Max-Quant platform as previously described where site level occupancy was calculated by the ratio of modified peptide in two samples, the unmodified peptide version and the protein ratio 89 . Cysteine residues were searched as a fixed modification of +57.0215 Da, oxidized methionine residues as a variable modification of +15.9949 Da and deamidated asparagine residues as a variable modification of +0.9840 Da, and methylation of lysine or arginine residues as a variable modification of +14.0266 Da. Heavy SILAC labeling of lysine (K) and arginine (R) residues were set as variable modifications of +10 Da for heavy R and +8 Da for heavy K. All default settings for the 'Orbitrap' instrument type were used. This included mass tolerances of 20 ppm and 0.5 Da for MS1 and MS2 searches, respectively. Requantify was enabled and peptides were queried using trypsin/P cleavage constraints with a maximum of two missed cleavages sites. Match between runs was enabled. The peptide and protein false-discovery rate was set to 0.01 (1% FDR).
Peptide-level mean normalized H/L ratios were first filtered for arginine monomethylated peptides occurring in at least two biological replicates, followed by significance testing using the limma package (v3.38.3) in R 90 . Significant hits were called as H/L ratio of <−1 (knockout cells (H) relative to control (L)) and a Benjamini-Hochberg adjusted p-value of <0.01 (n = 4). Gene ontology enrichment analysis was performed using clusterProfiler (ver. 3.10.1) 91 . p-values from four independent replicates calculated by empirical Bayes moderated t-tests and adjusted using the Benjamini-Hochberg procedure as implemented in the Bioconductor package limma (v3.38.3) 90 .
PRMT7 chemical probe synthesis. Experimental procedures and characterization data for chemical probe synthesis is described in Supplementary Methods section and illustrated in Supplementary Fig. 16.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
All mass spectrometry raw files been deposited in the MassIVE repository housed at UCSD (https://massive.ucsd.edu/) with the accession number MSV000084773 [https:// doi.org/doi:10.25345/C5GQ37]. Direct ftp download is available here ftp://massive.ucsd. edu/MSV000084773/. Raw input files are found within the updates folder. Known methylation sites were referenced from PhosphoSitePlus® v6.5.8 for Fig. 2c and Supplementary Table 4. The mPRMT7_SGC8158 structure has been deposited under the accession code PDB 6OGN. Data relating to Fig. 2c, Supplementary Fig. 2