Identification of hnRNP-A1 as a pharmacodynamic biomarker of type I PRMT inhibition in blood and tumor tissues

Arginine methylation has been recognized as a post-translational modification with pleiotropic effects that span from regulation of transcription to metabolic processes that contribute to aberrant cell proliferation and tumorigenesis. This has brought significant attention to the development of therapeutic strategies aimed at blocking the activity of protein arginine methyltransferases (PRMTs), which catalyze the formation of various methylated arginine products on a wide variety of cellular substrates. GSK3368715 is a small molecule inhibitor of type I PRMTs currently in clinical development. Here, we evaluate the effect of type I PRMT inhibition on arginine methylation in normal human peripheral blood mononuclear cells and utilize a broad proteomic approach to identify type I PRMT substrates. This work identified heterogenous nuclear ribonucleoprotein A1 (hnRNP-A1) as a pharmacodynamic biomarker of type I PRMT inhibition. Utilizing targeted mass spectrometry (MS), methods were developed to detect and quantitate changes in methylation of specific arginine residues on hnRNP-A1. This resulted in the development and validation of novel MS and immune assays useful for the assessment of GSK3368715 induced pharmacodynamic effects in blood and tumors that can be applied to GSK3368715 clinical trials.


Results
Pharmacology of type I PRMT inhibition. To explore changes to arginine methylation in proliferating and resting human PBMCs, cells cultured with and without T Cell Receptor (TCR)-activation, respectively, were treated for 72 h with GSK3368712, a type PRMT I inhibitor with similar chemical properties to GSK3368715 (8). Treatment led to changes in the levels of global MMA, SDMA and ADMA ( Fig. 2A). While a decrease in ADMA and an increase in MMA was observed in non-stimulated PBMCs, upon TCR-activation, treatment with GSK3368712 resulted in a robust decrease in ADMA and an increase in both MMA and SDMA. Evaluation of type I PRMT and PRMT5 gene expression demonstrated increased expression of PRMT5 and the type I PRMTs (PRMT1, PRMT3, PRMT4 and PRMT6) in stimulated PBMCs (Fig. 2B). PRMT1 protein, but not PRMT5 protein increased upon TCR activation [PRMT3 and PRMT4 protein levels were inconclusive (data not shown)], suggesting that increased levels MMA and SDMA with GSK3368712 treatment may be, in part, a consequence of increased type I PRMT levels ( Fig. 2A,B).

Identification of type I PRMT substrates.
To overcome technical limitations associated with quantitating global arginine methylation changes upon type I PRMT inhibition, methylated arginine enrichment proteomics (MethylScan) 15 was used to identify specific proteins that undergo changes in MMA, ADMA and SDMA upon GSK3368712 treatment. Human PBMCs obtained from four healthy donors were treated individually for 72 h with GSK3368712 in the presence of a TCR-activating agent. The purpose of T cell activation was twofold: (a) augmentation of the arginine methylation machinery to maximize identification of sites methylated by type I PRMTs and (b) generate, via clonal expansion, a high number of cells to yield sufficient total protein for enrichment of arginine methylated proteins. Total protein was isolated from the cultured PBMCs, digested with trypsin and immunoprecipitated with antibodies specific to each arginine methylation mark. Immunopurified peptides were resolved and sequenced by liquid chromatography-tandem mass spectra (LC-MS/MS), and fold-changes were calculated relative to DMSO-treated PBMCs. Applying a 2.5-fold change as a cutoff, a decrease of ADMA on 72 proteins common to all four donors was identified (Fig. 2C, Data file 1). In a similar fashion, GSK3368712 treatment resulted in an increase of MMA and SDMA on 198 and 64 proteins, respectively. Interestingly, of the 334 unique proteins that underwent any change in arginine methylation, 22 were identified as bearing a decrease in ADMA with a concomitant increase in MMA and SDMA at one or multiple arginine sites (Fig. 2D), and the www.nature.com/scientificreports/ majority of arginine methylation changes were detected on proteins involved in mRNA splicing, and RNA processing and metabolism (Fig. 2E). These findings were consistent with results observed in cancer cell lines (8) and suggest that many substrates are conserved across normal PBMCs and cancer.
Characterization of hnRNP-A1 as a pharmacodynamic biomarker of type I PRMT inhibition. To confirm the primary findings from the MethylScan study in TCR-activated PBMCs, we focused on the 22 proteins that underwent decreases in ADMA and increases in MMA and SDMA upon GSK3368712 treatment, consistent with the pharmacology of type I PRMT inhibition. Considering expression of each protein across tissues, access to commercially-available mouse monoclonal antibodies and relative abundance of tryptic peptides identified by mass spectrometry (MS), 9 proteins were selected for further evaluation (ALYREF, DHX9, EIF4H, EWSR1, G3BP1, HNRNPA1, KHDRSB1, SFPQ and TPR; Fig. 2D). Using the same human PBMC lysates analyzed by MethylScan and custom sandwich immunoassays (AlphaLISA) that allow indirect detection of ADMA on each target protein, only a reduction in ADMA on ALYREF and hnRNP-A1 could be confirmed (Fig. 3A). Given the robust signal detected with hnRNP-A1 compared to ALYREF, hnRNP-A1 was selected for further characterization. A number of factors, including, but not limited to, antibody quality or low protein expression may have contributed to the inability to confirm changes in the other potential substrates. HnRNP-A1 is a protein that is involved in several stages of RNA metabolism, from nascent mRNA assembly to RNA processing and splicing 13 , and has been previously described as a substrate of PRMT1 14 . These results were confirmed by analyzing the levels of ADMA-hnRNP-A1 in non-stimulated human PBMCs and two cancer cell lines treated with GSK3368712 for 72 and 48 h, respectively (Fig. 3B).
Arginine methylation analysis of hnRNP-A1 by mass spectrometry. The results obtained from the MethylScan and AlphaLISA studies from the in vitro culture of human cancer cells and healthy human PBMCs suggested that hnRNP-A1 could be utilized as a clinical biomarker of type I PRMT activity in human blood; therefore, modulation of arginine methylation on hnRNP-A1 by the type I PRMT inhibitor was further characterized. The MethylScan data demonstrated that treatment of TCR-activated human PBMCs with GSK3368712 modulated arginine methylation at various sites on hnRNPA1 (Data File 1), specifically in the GAR-region encompassing amino acids 194-232. This region contains six arginine residues all found to be mono-and/or di-methylated to various extents in the MethylScan data. Also, most of the hnRNP-A1 tryptic peptides identified contained multiple methylated arginine residues, precluding quantitation of methylation on specific residues and making the quantitative interpretation of the data challenging. To overcome this, we carried out an in-depth MS analysis of immunoprecipitated hnRNP-A1 from control and GSK3368712 treated Toledo cells ( Figure S1). We identified 12 distinct hnRNP-A1 methylated arginines, six of which are contained www.nature.com/scientificreports/ in the RG-rich region (R194, R196, R206, R218, R225 and R232; Fig. 4). Based on precursor ion intensity and the quality of the MS/MS spectra, we chose to focus on dimethyl arginine (DMA) sites R194, R206, and R225 found in peptides hnRNP-A1_180-196, hnRNP-A1_197-218 and hnRNP-A1_219-232, respectively. For all three sites, a reduction in DMA following treatment with GSK3368712 was observed ( Fig. 4B-D). An increase in the unmethylated forms of R194 and R206 was observed. While a modest decrease in the monomethylated form of R194 was noted, treatment with GSK3368712 led to an increase in monomethylated R206 and R225. To determine whether dimethylation at specific sites was symmetric, asymmetric or a combination, we monitored the generation of diagnostic neutral loss fragment ions specific for either ADMA or SDMA 16 using parallel reaction monitoring (PRM) LC-MS/MS. Monitoring of the diagnostic neutral loss fragment ions indicate R194 and R225 to be exclusively asymmetrically dimethylated suggesting these residues are solely type I PRMT substrates, while R206 was observed as a combination of ADMA and SDMA, suggesting methylation at this residue is the effect of both type I and type II PRMT activity in these cells (Fig. 4E-G). Additionally, it was important to investigate the effect of TCR activation on methylation of R225 on hnRNP-A1 in PBMCs. This residue is nearly 100% asymmetrically dimethylated in both non-stimulated and TCR-activated PBMCs with no detectable symmetrical dimethylation ( Figure S2). These findings demonstrate that TCR activation does not alter methylation of R225 on hnRNP-A1 in human PBMCs and supports further evaluation of ADM-R225-hnRNPA1 as a pharmacodynamic biomarker of type I PRMT inhibition. www.nature.com/scientificreports/

Development of LC-MS/MS for hnRNP-A1 detection and quantitation in
PBMCs from human blood. The lack of R225 in its SDMA form in both PBMCs and cancer cells, and the robust decrease in ADM-R225 observed with GSK3368712 treatment, together with the desire to select a site that would be amenable to both MS and immune-based assay development (see below for details), prompted the selection of R225 for further assessment. A targeted peptide mass spectrometry approach was chosen as it could offer rapid throughput and adequate sensitivity with complete specificity for the dimethylated form of R225 on hnRNP-A1 in human PBMCs. To this end, we developed a novel LC-MS/MS method for the determination of dimethyl (DM)-R225 in human PBMCs and mouse leukocytes protein lysates using the chymotrypsin-derived peptide 'SG-[R(Me)2]-GGF' spanning amino acids 223-228. Additionally, to normalize dimethyl-R225 levels to total hnRNP-A1 protein, this method included the quantification of another chymotryptically derived peptide spanning amino acids 154-167 (DDHDSVDKIVIQKY) of human hnRNP-A1 that lacked potential arginine methylation sites. The endoproteinase chymotrypsin, which cleaves at aromatic amino acids, was chosen as the digestion enzyme, instead of the more commonly used enzyme trypsin, which cleaves at arginine and lysine residues. The digestion efficiency of trypsin decreases at arginine sites that are dimethylated 17 , posing a significant challenge for the accurate quantitation of changes in methylated arginine forms. LC-MS/MS analysis of a mixture of synthetic peptides corresponding to hnRNP-A1 223-228 bearing unmethylated, mono-methylated, or asymmetricallydimethylated arginine 225 with the hnRNP-A1 154-167 peptide showed that these modified forms could be distinguished both by mass and chromatographic retention time (Fig. 5A). While all methylation states of R225 can be discriminated with this method, only the asymmetrically dimethylated form of R225 could be identified in human PBMC samples at appreciable levels ( Fig. 5B and S3A). This finding indicates that this residue is a high affinity substrate of type I PRMTs and suggests that methylation occurs in a processive fashion. Additionally, the www.nature.com/scientificreports/ removal of serine protease inhibitors from the lysis buffer significantly improved the sensitivity of the assay by improving chymotrypsin cleavage efficiency (Fig. 5C). Using 12.5 µg of PBMC derived protein lysate, the lower limit of quantification (LLOQ) for both DM-R225-hnRNP-A1 and hnRNP-A1 154-167 peptides in this assay was determined to be 50 ng/mL ( Figure S4). At all quality control sample concentrations examined, the intra-and inter-run precision expressed as % CV were less than or equal to 20% (25% at the LLOQ) (Table S1). Pre-analytical variables associated with both the processing and lysis of PBMCs and stability of hnRNP-A1 peptides over time were also assessed. As shown in Table S2, timing of lysate generation from collection of PBMC isolation and length of storage at -70ºC for up to 8 months do not significantly affect the levels of both hnRNP-A1 peptides.

Reduction of DM-R225-hnRNP-A1 in human PBMCs and mouse leukocytes by LC-MS/ MS. The optimized LC-MS/MS method was used to assess the levels of DM-R225-hnRNP-A1 in both human
PBMCs and mouse leukocytes treated with the type I PRMT inhibitors in vitro and in vivo, respectively. The effects of GSK3368712 were investigated in both non-stimulated and TCR-activated PBMCs. While treatment of non-stimulated human PBMCs led to minimal reduction of DM-R225-hnRNP-A1, a dose-dependent decrease of the biomarker was observed in TCR-activated PBMCs (Fig. 6A). This is consistent with the lack of de novo arginine methylation inhibition in non-proliferating cells, such as non-stimulated PBMCs that have high levels of dimethylated R225-hnRNP-A1. While these findings may appear discordant with the AlphaLISA results shown in Fig. 3B, where treatment with GSK3368712 led to a robust decrease in levels of ADMA-hnRNP-A1 in non-stimulated human PBMCs, it is worth noting that the buffer composition of the AlphaLISA allows proteinprotein interactions to be maintained and thus can detect any ADMA on hnRNP-A1 and any of its interacting proteins. Therefore, the decrease in assay signal is directly proportional to any decrease in ADMA on either hnRNP-A1 and/or interacting proteins that bear ADMA residues modulated by type I PRMTs. Given these results, we wanted to demonstrate the utility of the LC-MS/MS method for the assessment of pharmacodynamic www.nature.com/scientificreports/ changes on DM-R225-hnRNP-A1 in blood from mice treated with the type I PRMT inhibitor. For this in vivo study, and others herein, where quantitative assessment of pharmacodynamic changes are evaluated, the clinical compound, GSK3368715, is used. Treatment with compound once daily for 15 days led to a significant dosedependent reduction of DM-R225-hnRNP-A1. The highest dose evaluated, 600 mg/kg, yielded an approximate 88% decrease in DM-R225-hnRNP-A1 compared to vehicle-treated animals (Fig. 6B).

Levels of hnRNP-A1 across blood cell populations.
To identify immune cell subtypes with this modification and assess PD changes in DM-R225-hnRNP-A1, we determined the levels of both dimethylated-R225-hnRNP-A1 and hnRNP-A1 154-167 peptides in monocytes, neutrophils, PBMCs and all white blood cells (WBCs) from four healthy donors using our novel optimized LC-MS/MS method. Upon confirmation of cell purity and relative abundance by flow cytometry analysis ( Figure S5 and Table 1), LC-MS/MS analysis of both peptides revealed the highest levels in monocytes and PBMCs, and the lowest levels in neutrophils (Fig. 7A). This finding is consistent with the low levels of hnRNP-A1 detected in all leukocytes, in which granulocytes represent the most abundant cell population (40-70% of all white blood cells; 18 ). While this finding was initially hypothesized to be attributable to the high levels of digestive enzymes in neutrophils, addition of neutrophil elastase and collagenase inhibitors failed to improve the detection of hnRNP-A1 (data not shown), suggesting that hnRNP-A1 is likely to be expressed at low levels in neutrophils. This observation is also supported by the work published by Song et al. 19 showing that the levels of hnRNP-A1 decrease significantly upon granulocytic differentiation. These findings identified PBMCs as the optimal blood compartment to be analyzed for levels of hnRNP-A1 and its PD changes upon treatment with the type I PRMT inhibitor in humans. To evaluate levels of both hnRNP-A1 peptides in human PBMCs, cell preparation tubes (CPT) were used to isolate PBMCs from human blood. As shown in Fig. 7B, the levels of the two peptides in PBMCs obtained from 10 donors range from ~ 300 ng/mL to ~ 1,240 ng/mL. Given that the LLOQ for DM-R225-hnRNP-A1 is 50 ng/mL, this assay allows detection of at least a sixfold-reduction (or ~ 87%) in DMA in cases with the lowest expression of hnRNP-A1.
Generation of a monoclonal antibody specific to ADM-R225-hnRNP-A1. To assess the potential use of hnRNP-A1 as a clinical biomarker of target engagement in human tissues, we explored the opportunity to develop an antibody that is specific to asymmetrically dimethylated R225 of human hnRNP-A1. The choice of R225 over R194 and R206 was based on several factors, such as, but not limited to, lack of symmetric dimethylation, higher immunogenicity potential and higher likelihood of selectivity for the asymmetrically dimethyl form over non-methylated, mono-methylated and symmetrically dimethylated arginine. A synthetic peptide corresponding to amino acids 220-231 of hnRNP-A1 (GNFSGRGGFGGS) bearing asymmetrically dimethylated R225 was used to immunize mice. A custom electrochemiluminescence method (MesoScale Discovery, MSD) was used to determine the selectivity of hybridoma-derived supernatants for the ADMA form of R225 by employing biotinylated peptides bearing all forms of arginine methylation on residue R225, including the non-methylated form as a control. One clone (26H3) was selected from the primary screen (Table S3) based on both affinity and selectivity for the ADM-R225-hnRNP-A1 peptide over non-methylated, MM-and SDM-R225 hnRNP-A1 peptides. Purified antibody from the 26H3 clone was further characterized against several concentrations of peptides bearing all three methylation modifications of R225 in a custom AlphaLISA (Fig. 8A). Clone 26H3 showed complete selectivity for the ADMA over the non-methylated and symmetrically dimethylated forms of R225-hnRNP-A1 while retaining some affinity for the mono-methylated form (K d > 500 nM). By Western blot, 26H3 detects a single band at the expected molecular weight (34-38 Kd) for hnRNP-A1 in PBMCs and Toledo cells ( Figure S6). Additionally, the use of recombinant hnRNP-A1 expressed in prokaryotic and eukaryotic cells further demonstrates the specificity of these antibodies for the dimethylated form of the protein as shown by the detection of ADMA on hnRNPA1 in C-myc/DDK-tagged hnRNP-A1 expressed in HEK293 human kidney cells, but not in hnRNPA1 expressed in E. coli where arginine methyltransferases do not exist. The presence of multiple bands in the HEK-expressed material can be attributed to the presence of multiple recombinant hnRNPA1 degradation products. Together, these data demonstrate that 26H3 is highly specific for ADMA-hnRNPA1.
Development of hnRNP-A1 IHC assays. The preliminary characterization of the novel antibody introduced the opportunity to develop an immunohistochemistry (IHC) method to monitor the pharmacodynamic effects of the type I PRMT inhibitor on hnRNP-A1 in human tumor tissues. Clone 26H3 was tested at sev- www.nature.com/scientificreports/ eral concentrations and conditions in human normal and tumor tissues along with a commercially available antibody for the detection of total hnRNP-A1 for use in normalization (Fig. 8B). The levels of both asymmetrically-dimethylated-R225 and total hnRNP-A1 in the nuclear and cytoplasmic compartments of tumor cells were quantitated using the H-score methodology. IHC results across eight tumor types revealed localization of hnRNP-A1 to the nucleus, with very little detection in the cytoplasm (Data file 2) while mouse and rabbit IgG isotype control antibodies revealed no stain in the histologies examined ( Figure S7). The overlapping patterns of expression (Fig. 8B) and comparable levels of stain intensity (Fig. 8C) of ADM-R225-and total hnRNP-A1 across all tumor types suggest that the majority of the hnRNP-A1 protein pool is nearly fully asymmetricallydimethylated at the R225 site. Only in breast cancer and melanoma did we observe a small, yet statistically-significant, difference in levels of ADMA and total hnRNP-A1, as indicated by the lower ADM-R225/Total ratios, suggesting that hnRNP-A1 may not be completely asymmetrically-dimethylated in these two tumor types. However, additional studies in a larger sample set would be required to determine whether the effect is indeed tumor type-specific or the result of small sample size. To evaluate the pharmacodynamic effects of the type I PRMT inhibitor, optimized IHC conditions for both ADM-R225-hnRNP-A1 and total hnRNP-A1 antibodies were employed on formalin-fixed tumor tissues from a mouse xenograft model using the Toledo human diffuse large B-cell lymphoma (DLBCL) cell line. Treatment of engrafted SCID mice with GSK3368715 administered once daily for 8 days led to a significant reduction of the asymmetric dimethyl mark on R225 of hnRNP-A1 in tumors tissues ( Fig. 8D-E). Importantly, the levels of total hnRNP-A1 protein were not affected by compound treatment at doses where tumor regression was observed, except at the highest dose tested (600 mg/kg) (Fig. 8G, Table S4, www.nature.com/scientificreports/ Figure S8). However, upon normalization of ADMA to total hnRNPA1 levels, a dose proportional decrease in ADMA-hnRNPA1 is still observed at the highest dose (600 mg/kg) (Fig. 8F). Finally, comparison of the reduction in ADM-R225-hnRNP-A1 in mouse blood (Fig. 6B) to that observed in tumor xenografts demonstrates similar reduction in ADMA-hnRNPA1 in tumor and circulation following multiple days of GSK3368715 treatment using these distinctly different assays (Fig. 8G). While these studies were conducted in different strains of mice, it is possible to compare pharmacodynamic changes across studies in this instance since the exposure of GSK3368715 was quite similar in each strain (Table S5), and ADMA-hnRNPA1 was assessed after many days of dosing where steady state levels of drug would be reached.

Discussion
Type I PRMTs catalyze the formation of asymmetric dimethyl arginine (ADMA) on a wide range of cellular substrates spanning histones, signaling molecules and RNA splicing factors. Pharmacological inhibition of type I PRMTs has been shown to yield anti-proliferative effects in multiple tumor types (8). The type I PRMT inhibitor GSK3368715 is currently being evaluated in a first time in human clinical trial to investigate the safety, pharmacokinetics, pharmacodynamics and clinical activity in participants with solid tumors and DLBCL (NCT03666988). The complex bimodal pharmacology of type I PRMT inhibition, characterized by a decrease in ADMA with a concomitant increase in MMA and SDMA, posed a challenge for the development of clinical biomarker assays based on the detection of global methylated arginine levels. While Western blots can provide useful qualitative information in a pre-clinical setting, their use for clinical samples is suboptimal due to both procedural complexity and semi-quantitative nature associated with the lack of uniform signal linearity across broad molecular weight ranges. To overcome this challenge, we explored the effect of type I PRMT inhibition in human PBMCs employing a proteomic approach aimed at identifying substrates that could be monitored for the reduction in asymmetric dimethylation at specific arginine sites. Our study identified several candidate substrates as potential PD biomarkers of type I PRMT inhibition. Through custom immunoassays, hnRNP-A1 was identified as a promising biomarker of type I PRMT inhibition. HnRNPs are ubiquitously expressed across tissues with varying levels of abundance 20 , and hnRNP-A1 has been widely investigated for its involvement in all key steps of RNA metabolism 21 with a focus on how PTMs, such as phosphorylation [22][23][24] , sumoylation 25 and methylation 26,27 , modulate its function as a key regulator of gene expression. The reduction of ADMA on hnRNP-A1 was demonstrated in both cancer lines and human PBMCs and this prompted us to develop an LC-MS/MS method aimed at detecting and quantifying dimethylarginine 225 levels on hnRNP-A1 in human PBMCs. The advantage of this method over immunodetection assays is the ability to fully discriminate all forms of arginine methylation on R225, avoiding potential cross-reactivity with MMA which increases upon type I PRMT inhibition. The novelty of this LC-MS/MS method is in the use of chymotrypsin to digest proteins. Typically, trypsin is used; however, trypsin cleaves at arginine residues, and the efficiency of cleavage is diminished at post-translationally modified arginines. Furthermore, removal of serine protease inhibitors from cell lysates resulted in a significant improvement of assay robustness and linearity, making the method more amenable for use in a clinical setting. Using this LC-MS/MS assay, decreases in DMA-R225-hnRNP-A1, but not hnRNP-A1, are observed in both human PBMCs and mouse leukocytes after type I PRMT inhibition (Fig. 6). We observed that PBMCs and monocytes have higher levels of hnRNP-A1 expression and dimethylated R225 than neutrophils and total white blood cells (Fig. 7A). This enabled identification of cell preparation tubes (CPT) as the optimal collection method to enrich for cell populations (PBMCs) expressing hnRNP-A1. Finally, given the LLOQ of the LC-MS/MS assay, the levels of DMA-hnRNP-A1 in PBMCs across donors fall in a range that allows detection of at least an 87% reduction in DMA-R225-hnRNP-A1 even in cases with the lowest levels of the protein. Importantly, the specificity of this assay allowed us to determine that R225-hnRNP-A1 is present in both human PBMCs and Toledo cells in its asymmetrically-dimethylated form only, suggesting that this residue is a high affinity substrate of type I PRMTs.
Establishing the utility of hnRNP-A1 as a biomarker of target engagement in surrogate tissue, such as PBMCs, prompted us to investigate the pharmacodynamic effects of type I PRMT inhibition on levels of ADMA-hnRNP-A1 in tumor tissues. Mass spectrometric characterization of hnRNP-A1 in cancer cells allowed identification of R220-R231 as a candidate epitope for the generation of a mouse monoclonal antibody specific to the asymmetric dimethylated form of R225 on hnRNP-A1. We were able to identify an antibody clone, 26H3, with high affinity and selectivity for ADM-R225 over unmethylated-, symmetrically dimethylated-and monomethylated-R225 (Fig. 8A). This antibody was used to develop a novel IHC assay for the detection of ADMA-hnRNP-A1 in tumor tissues (Fig. 8B). Additionally, an IHC assay for total hnRNP-A1 was developed in parallel to normalize for the changes in ADM-R225-hnRNP-A1. Both assays were used to assess the effects of GSK3368715 in Toledo tumor xenografts, and successfully demonstrated the dose-dependent decrease in levels of ADM-R225, but not total, hnRNP-A1 in tumor tissue. Importantly, since LC-MS/MS and IHC methods allowed us to quantitate the pharmacodynamic effect of the type I PRMT inhibitor on hnRNP-A1 in mouse PBMCs and Toledo xenografts, we were able to determine that the dose dependent reduction in ADM-R225-hnRNP-A1 in both blood and tumor tissues is similar in these two compartments following multiple days of dosing (Fig. 8G). In summary, this work led to the development of fit-for-purpose LC-MS/MS and IHC assays as two novel methodologies that allow for quantitation of changes in asymmetrically-dimethylated R225 on human hnRNP-A1 with high specificity and sensitivity in both surrogate and tumor tissues. These assays are currently being used to assess pharmacodynamic changes in patients treated with GSK3368715 (clinical trial NCT03666988).

Materials and methods
Culture of human PBMCs and cancer lines. Cryopreserved PBMCs (AllCells, PB005F) were cultured in AIM-V AlbuMAX medium (Thermo Fisher Scientific, #31035025) at 37 °C in 5% CO 2 . GSK3368712 was synthesized and purified as previously described (8)  Quantitative PCR. Non stimulated and TCR-activated human PBMCs were cultured with either 0.1% DMSO or 2 µM GSK3368712 in AIM-V AlbuMAX medium for 72 h and then harvested for isolation of total RNA using the RNeasy Mini Kit, according to the manufacturer's protocol (Qiagen, #74104). RNA was quantitated using NanoDrop (Thermo Fisher Scientific). All samples were then reverse-transcribed to cDNA and subjected to quantitative PCR using TaqMan RNA-to-CT 1-Step Kit (Thermo Fisher Scientific, #4392938) and triplicate reactions were run on ABI ViiA 7 (Applied Biosystems) according to the manufacture's protocol. PCR amplifications were performed with specific TaqMan primers for PRMT1, PRMT3, PRMT4, PRMT5 and PRMT6 (Thermo Fisher Scientific, #Hs01587651g1, Hs00411605m1, Hs00406354m1 Hs01047356m1, Hs00250803s1, respectively). mRNA expression of target genes was normalized to the endogenous reference 18S gene (Hs03003631_g1) and the relative fold change from non stimulated, DMSO-treated control PBMCs was calculated using the 2^Δ ΔCT method.  www.nature.com/scientificreports/ PAK C18 columns and split into 3-4 mg aliquots for enrichment with MMA motif [mme-RG] immunoaffinity beads (CST, #12235), a mixture immunoaffinity beads conjugated to ADMA motif [adme-R] and custom ADMA antibody (clone D10F7A10) (CST, #13474; modified method), and SDMA motif [sdme-RG] immunoaffinity beads (CST, #13563). Enriched peptides were loaded directly onto a PicoFrit capillary column packed with Magic C18 AQ reversed-phase resin. Two non-sequential replicate injections were run for each enrichment. Proteomic analysis was carried out using the MethylScan method as previously described 28 . Each enriched sample was analyzed by LC-MS/MS in a data-dependent manner on a Thermo Orbitrap Q Exactive mass spectrometer using a top-twenty MS/MS method with a dynamic repeat count of one, and a repeat duration of 30 s. Peptides were eluted using a 90-or 120-min linear gradient of acetonitrile in 0.125% formic acid delivered at 280 nL/min. Peptide sequences were identified by searching MS/MS spectra against the SwissProt Homo sapiens database using SEQUEST 29 with a mass accuracy of 5 ppm for precursor ions and 0.02 Da for product ions. Enzyme specificity was set to semi-trypsin with up to four mis-cleavages allowed. Cysteine carboamidomethylation was specified as a fixed modification, oxidation of methionine and mono-or di-methylation on arginine residues were allowed as variable modifications. Reverse decoy databases were included for all searches to estimate false discovery rates (FDR), and filtered using a 2.5% FDR. All quantitative results were generated using Skyline 30 to extract the integrated peak area of the corresponding peptide assignments. Accuracy of quantitative data was ensured by manual review in Skyline or in the ion chromatogram files. Fold changes were calculated for each treatment relative to the DMSO control. % CV values between replicates of each condition were calculated as a measure of analytical reproducibility. For each comparison, we set the fold change observed between two conditions to be at least 2.5 fold and for the methylated peptides identified to be present in all samples from the four donors. Protein lists were merged for all 3 methyl marks and duplicates were removed to obtain a master list of proteins with an increase in the monomethyl-and symmetric dimethyl marks (for MMA and SDMA immunoprecipitations) and a concomitant decrease in the dimethyl mark (for ADMA immunoprecipitations) on the detected peptide (Data File 1). Overlaps were generated using Venny 2.1 (https ://bioin fogp.cnb.csic.es/tools / venny /) to obtain a list of changed proteins that were common across PBMCs from all four donors. Analysis of overlap enrichment on the list of gene names of the common proteins was performed using the Reactome biological pathway webtool.

Quantitation of ADMA on hnRNP-A1 in human cells by AlphaLISA.
Lysates from Toledo and Jurkat cancer cells and PBMCs treated with GSK3368712 were generated as described above for Western blot analysis and protein concentrations were assessed using the BCA Protein Assay Kit. 1 µg of lysates, in triplicates, was incubated with a custom anti-ADMA antibody (clone D10F7A10; provided to GSK by CST as a beta product) and an antibody specific to hnRNP-A1 (Abcam, #ab5832), at a final concentration of 1 µg/mL in AlphaLISA universal buffer for 1 h at room temperature. Anti-rabbit-IgG-coated acceptor beads (Perkin Elmer, #AL104) were added at a final concentration of 20 µg/mL and incubated for 1 h at room temperature. Anti-mouse IgG donor-coated beads (Perkin Elmer, #AS104) were added at a final concentration of 40 µg/mL and incubated for an additional 2 h at room temperature. Microplates were shaken after each addition and centrifuged prior to luminescence detection on an Envision microplate reader.

Analysis of hnRNP-A1 arginine methylation in Toledo cells by LC-MS/MS. Toledo cells were cul-
tured with 0.1% DMSO or 1 μM GSK3368712 for 48 h, collected and lysed in CellLytic M buffer. hnRNP-A1 was immunoprecipitated with mouse anti-hnRNP-A1 antibody (Abcam, #ab5832) using the Pierce Classic Magnetic IP/Co-IP Kit (Thermo Fisher Scientific, #88804) per manufacturer's instructions. Immunoprecipitated eluates were precipitated with acetone, separated by SDS-PAGE and visualized by Coomassie staining. The hnRNP-A1 bands were excised, reduced using TCEP (Thermo, #77720), alkylated with iodoacetamide (Sigma, #I1149) and digested overnight with trypsin at 37 °C (Promega, #V5111). The following method was performed as previously described (8). After organic extraction, samples were injected on an Easy nLC1000 UHPLC system (Thermo Scientific). The nanoLC was interfaced to a Q-Exactive Hybrid Quadrupole-Orbitrap Mass Spectrometer (Thermo Scientific). Tryptic peptides were separated on a 25 cm × 75 μm ID, PepMap C18, 3 μm particle column (Thermo Scientific) using a 40 min gradient of 2-30% acetonitrile/0.2% formic acid and a flow of 300 nL/min. MS-based peptide sequencing was accomplished by tandem mass spectrometry using data dependent LC-MS/ MS. Uninterpreted tandem MS spectra were searched for peptide matches against the human UniProt protein sequence database using Mascot (Matrix Science). Carboamidomethylation was selected as a fixed modification www.nature.com/scientificreports/ on cysteine residues. Oxidation on methionine and methylation and dimethylation on arginine residues were selected as variable modifications. MS/MS spectra for methylated peptides were manually validated to confirm the site of mono or dimethylation. Integrated peak areas from Extracted Ion Chromatograms (XICs) from the MS scan were used to monitor un-, mono-and dimethylation in control and inhibitor treated samples. Identified hnRNP-A1 methylation sites were further interrogated using a parallel reaction monitoring (PRM) method targeting the dimethylated peptides. Diagnostic ions for either ADMA (neutral loss of 45.0578) or SDMA (neutral loss of 31.0422) were monitored and allowed the determination of the ADMA/SDMA status.  Generation of mouse monoclonal antibody against ADM-R225-hnRNP-A1. All antibody clones were generated and purified by GenScript (NJ, USA). BALB/C mice were immunized with a synthetic peptide corresponding to amino acids 220-231 of human hnRNP-A1, GNFSG(R ADMA )GGFGGSC, according to GenScript's proprietary rapid immunization MonoExpress protocol. Biotinylated peptides corresponding to all methylated forms of the R225-containing immunizing peptide were synthesized and purified at GenScript and used to determine the affinity and selectivity of antibodies for the asymmetrically-dimethylated form of R225-hnRNP-A1. Supernatants from the hybridoma cultures were screened against the biotinylated hnRNP-A1 peptides in a custom immunoassay based on electrochemiluminescent readout (MesoScale Discovery). Purified antibody from clone 26H3 was tested against biotinylated hnRNP-A1 peptides to determine affinity and selectivity using the custom AlphaLISA as described above.

Development of LC
In vivo studies. For the PD study in blood, naïve male CD-1 mice (Charles River, MA) were dosed orally once daily with either a saline solution (vehicle) or GSK3368715 ranging from 0.137 to 600 mg/kg (n = 3 per group) for 15 days. 4 h after the last dose, mice were euthanized by carbon dioxide inhalation and full blood volume was collected in heparinized vacutainer tubes (BD, #367671) by terminal cardiac puncture. Leukocytes were purified from whole blood, incubated with red blood cell lysis buffer ( www.nature.com/scientificreports/ GSK3368715 was chromatographically separated on an Acquity BEH Phenyl, 2.1 × 50 mm column using 10 mM ammonium formate and acetonitrile using a gradient mobile phase followed by multiple reaction monitoring employing Atmospheric Pressure Chemical Ionization (APCI) in positive-ion mode. Blood pharmacokinetic parameters were obtained from the concentration-time profiles using non-compartmental analysis with Phoenix WinNonlin v6.3 (Certara, Princeton, NJ). Individual mouse profiles were used for CD1 mice while mean composite (n = 3/time point) concentration time profiles were used for SCID blood pharmacokinetic parameter determination.
All studies were conducted in accordance with the GSK Policy on the Care, Welfare and Treatment of Laboratory Animals and were reviewed by the Institutional Animal Care and Use Committee at GSK.
Immunohistochemistry. Both total and ADM-R225-hnRNP-A1 IHC assays were developed and at QualTek Molecular Laboratories using anti-hnRNP-A1 (Abcam, #ab177152) and anti-ADM-R225-hnRNP-A1 (clone 26H3; custom development for GSK by GenScript). Human tissues were sourced from QualTek's tissue bank and a total of 4 cases per histology were evaluated. Human tissues and tumor xenografts were freshly cut at a thickness of 4-5 μm and mounted onto positively-charged, capillary gap glass slides prior to use. Briefly, slides were baked at 60 °C (dry heat) and subjected to antigen/epitope retrieval using a steam heat induced epitope recovery (SHIER) solution for 20 min in the capillary slide gap in the upper chamber of a Black and Decker Steamer, as previously described by Ladner et al. 31 . The SHIER solutions used for antigen retrieval of total hnRNP-A1 and ADM-R225-hnRNP-A1 were AR10 (pH 10) and citrate-based (pH 6.0-6.2), respectively. Following antigen retrieval, all process steps were automated using a TechMate Instrument (Roche Diagnostics) running QML workmate software v3.96. After blocking in 5% goat serum for 15 min, slides were incubated at room temperature either with the hnRNP-A1 antibody over night or with the ADM-R225-hnRNP-A1 antibody for 1 h. Slides were then washed in tris-buffered saline with 0.15% Tween-20 (TBS-T) and incubated with either mouse or rabbit Polink-2 Plus secondary antibodies (GBI Labs) for 25 min at room temperature. Following washes in TBS-T, slides were blocked in hydrogen peroxide (3 × 2.5-min incubations) and incubated with Polink-2 Plus Horseradish Peroxidase (HRP, GBI Labs) for 25 min at room temperature. Slides were then exposed to a chromogen substrate (DAB) (3 × 2.5-min incubations) and counterstained with hematoxylin. Slides were rinsed in distilled water, dehydrated in both an alcohol series (95%, 100% ethanol) and in an organic solvent (100% xylene, four changes), then permanently coverslipped using CytoSeal 60 (Thermo Fisher scientific) and reviewed on an Olympus BX60F5 microscope using 20-40X objectives. Mouse IgG1 and rabbit IgG isotype-match controls corresponding to the IHC assay conditions for total hnRNP-A1 and ADMA-R225 were included on each tissue tested. All tissues analyzed were scored by a board-certified pathologist and the levels of hnRNP-A1 protein, both total and ADM-R225, are reported as H-Scores, and calculated by summing the percentage of cells with intensity of expression (brown staining) multiplied by their corresponding differential intensity on a four-point semi-quantitative scale (0, 1 + , 2 + , 3 +). H-Scores range from 0 to 300 based on the following equation: Representative regions of interest were identified visually, and high-resolution images were taken from a live microscope view using Amscope software.
Quantification and statistical analysis. Graphing of results and statistical analyses were performed using Microsoft Excel. Sample sizes are indicated in the figure legends and data are expressed as the mean ± standard deviation (SD). Where applicable, statistical significance was evaluated using either a two-tailed Student's t-test (Microsoft Excel) or the Mann-Whitney test (Study Director software). IC 50 s were calculated in Xlfit (IDBS) using a 4 parameter logistic model (sigmoidal dose-response). Figure 1 was created using ChemDraw (PerkinElmer) and Microsoft PowerPoint.

Use of human biological samples.
For all studies involving human biological samples, methods were performed in accordance with the relevant guidelines and regulations, and the use of these samples was reviewed and approved by GSK Research & Development Compliance (RDC) Human Biological Sample Use Committee. Written informed consents and/or a waiver of consent approved by an Institutional Review Board (IRB) in accordance with all applicable laws were provided for all study participants.