SAM-Competitive PRMT5 Inhibitor PF-06939999 Demonstrates Antitumor Activity in Splicing Dysregulated NSCLC with Decreased Liability of Drug Resistance

This article is featured in Highlights of This Issue, p. 1

PRMT5 is the major type II arginine methyltransferase that utilizes S-adenosylmethionine (SAM) to catalyze both mono- and symmetric dimethylation on protein substrates (1) involved in a variety of cellular processes including transcription, cell signaling, mRNA translation, DNA repair, protein stability, and pre-mRNA splicing (2–14). PRMT5 binds MEP50 and accessory proteins, forming a methylosome complex possessing symmetric dimethylation activity for proteins possessing RGG/RG motifs (15). Spliceosome proteins SmD1, SmD3, and SmB/B' are methylated by PRMT5, increasing their affinity for SMN1 protein, and facilitating assembly of small nuclear ribonucleoprotein (snRNP) complexes responsible for proper splice site recognition (3, 4). PRMT5 genetic inhibition leads to increased intron retention and exon skipping in pre-mRNAs, resulting in mRNA nonsense-mediated decay or alternatively spliced mRNAs (16), and proteins involved in pre-mRNA processing are highly enriched as cellular substrates of PRMT5 methylation (17, 18).

PRMT5 overexpression has been observed in multiple cancers, including non–small cell lung cancer (NSCLC), where enzyme function has been linked to enhanced tumor cell growth and poor patient survival (2, 19, 20). Cancer cells are dependent on PRMT5 enzyme activity for growth, as genetic inhibition or catalytic inhibition of PRMT5 blocks cancer cell proliferation and promotes apoptosis (19, 21–24). PRMT5 inhibitors have recently entered the clinic (NCT02783300, NCT03614728, NCT03573310, NCT03854227) targeting multiple hematopoietic and solid tumor indications.

Translational strategies around utilization of PRMT5 inhibitors in the clinic should take advantage of tumor-specific vulnerabilities that enable effective use of these drugs at therapeutic doses, such as targeting dependencies generated through PRMT5 function as a splicing regulator. Specific splicing events induced by PRMT5 inhibition have particular relevance for the therapeutic targeting of PRMT5 in cancer (16). For example, PRMT5 inhibition leads to exon 6 skipping in MDM4, targeting the transcript for nonsense mediated decay, and activation of the p53 pathway. Activation of the p53–p21 pathway was linked to the antiproliferative activity of PRMT5 inhibitors and TP53 mutation status correlated with response to PRMT5 inhibitors (23). In addition, MYC-driven upregulation of snRNPs and PRMT5 generates higher reliance on proper splicing fidelity, increasing dependence on PRMT5 activity, which can be therapeutically exploited in Eu-MYC–driven mouse lymphoma models (25).

Perturbations in mRNA splicing are common features of tumorigenesis, driven by mutations and altered expression of splicing regulators (26–28). Recently, analysis of The Cancer Genome Atlas (TCGA) identified recurrent somatic mutations and copy number alterations in 119 splicing factor genes with known roles in RNA splicing, including snRNP assembly and branch point recognition (29). PRMT5 methylation of splice factors perturbed in cancer, such as SRSF1 (18), raises the possibility of targeting PRMT5 in cancers enriched in certain splice factors mutations. Accordingly, PRMT5 inhibition reduces splicing fidelity, resulting in preferential killing of splice factor–mutant leukemias (30).

In this study, we describe the discovery of potent and selective SAM-competitive PRMT5 inhibitors exemplified by clinical candidate, PF-06939999. PF-06939999 demonstrates in vitro and in vivo antiproliferative activity across a panel of NSCLC cell models concomitant with reduction in cellular SDMA levels, and increased sensitivity in NSCLC cell lines harboring RBM10 inactivation. Acquired drug resistance models were used to demonstrate that SAM-competitive PRMT5 inhibitors are less susceptible to the development of mutations that confer complete drug resistance compared with previously described SAM-cooperative binding compounds. Overall, our data highlight targeting PRMT5 represents an opportunity for targeting vulnerabilities in NSCLC created by dependencies on alternative splicing pathways.

Additional methods are described in the Supplementary Methods.

Detailed description for synthesis of PF-06855800 (example 88) can be found in patent US2016244475A1 (31). Detailed description for synthesis of PF-06939999 (example 190) can be found in patent US2017348313A1 (32).

Crystallization of full-length human PRMT5/MEP50 complexed with cofactor site inhibitors was performed at 13°C by hanging-drop vapor-diffusion methods. A solution of 5:1 molar ratio (2.5 μL) of inhibitor compound to PRMT5/MEP50 complex (13 mg/mL) was mixed with 2.5 μL of reservoir solution containing 13% to 15% (w/v) PEG3350, 0.1M MES, pH 6.5 to 7.5, 0.25 mol/L NaCl, and 20% (v/v) ethylene glycol. Microseeding from initial crystals produced crystals suitable for data collection. Crystals for data collection were flash-frozen in liquid N₂ using 25% (v/v) ethylene glycol in the mother liquor as a cryoprotectant and shipped for to the Advanced Photon Source IMCA-CAT beamline 17-ID at Argonne National Labs for diffraction data collection. Diffraction data were processed with autoPROC from Global Phasing (33) and structure solution and refinement were done with BUSTER (34) using the published structure of human PRMT5/MEP50 (PDB ID 4GQB) as the initial model. Model building was done with COOT (35). Crystallographic statistics are included in Supplementary Table S1.

A2780 cells (Sigma #93112519), NCI-H441 (ATCC HTB-174), and NCI-H1975 (ATCC CRL-5908) were maintained in RPMI1640 + 10% FBS + 1× penicillin/streptomycin (Thermo Fisher Scientific #15070063). A427 (ATCC HTB-53) cells were maintained in EMEM + 10% FBS + 1× penicillin/streptomycin. Additional NSCLC cell lines used in the proliferation assays were grown in manufacturer-recommended growth media. Cell lines used in xenografts were validated via STR typing and tested to confirm absence of mycoplasma.

A2780 cells were seeded at 2.5 × 10⁵ cells in a T75 flask and treated with GI₅₀ doses of PF-06855800 (1 nmol/L) and EPZ015666 (450 nmol/L). Media was refreshed twice per week and cells counted once per week to monitor proliferation. Once proliferation reached a plateau, compound concentrations were increased incrementally for a total of 40 weeks. The final concentrations of each compound used at the final resistance stage was 14 nmol/L for PF-06855800 and 15 μmol/L for EPZ015666. At week 40, resistant cell lines were analyzed by whole-exome sequencing for acquired mutations at WuXi NextCode.

Cells were lysed in RIPA buffer + protease/phosphatase inhibitor, sonicated, and cleared by centrifugation. Histones were purified using the Histone Purification Kit (Active Motif). Protein lysates were loaded at 20 μg and histones at 500 ng. Antibody information is in the supplemental data. Blots were imaged on a LI-COR Odyssey CLx imager and bands quantified using the LI-COR Image Studio software.

NSCLC cells were seeded in 96-well plates in recommended culture media and incubated overnight at 37°C, 5% CO₂. The following day, fresh media with compound (diluted in DMSO) was added and cells incubated at 37°C, 5% CO₂ for 7 days with media/compound refreshed at day 3 to 4. Cells were lysed in Cell Titer Glo (Promega #G7570) reagent and read on a plate reader with luminescence filter. Alternatively, CyQuant reagent was added to the plates at day 7, the plate incubated at 37°C for 1 hour and read on plate reader with fluorescent filter.

Cells were seeded at appropriate densities to allow for growth of 80% to 90% confluency by day 5. Cells were treated in dose response with PF-06939999 with DMSO as a control for either 4 or 5 days, stained with propidium iodide and analyzed by flow cytometry (FlowJo software) to observe changes in the phases of the cell cycle.

A427, NCI-H441, and NCI-H1975 cells were seeded in triplicate in recommended growth media overnight at 37°C, 5% CO₂. The following day, plates were treated with either 0.1% DMSO or 30 nmol/L PF-06939999 for 72 hours at 37°C, 5% CO₂. After RNA isolation, the library was prepared using the TruSeq Stranded mRNA Library kit (Illumina) and sequenced at 100 MM read depth with the Illumina HiSeq X 10 platform. Sequencing reads (150 bp paired end) were mapped to hg19 genome using STAR and quantified using RSEM. The DESeq2 program was used for differential expression analysis and alternative splicing analysis was done using rMATS version 4.02 (36). A minimum number of reads mapping to a splice junction was imposed to filter low-coverage splice junctions from the analysis: the minimal number of junction reads for at least one sample group for rMAT testing is 10. Gene pathway analysis utilized a hypergeometric test with FDR correction against the MSigDB database (37).

Cells were dichotomized into two classes (sensitive and resistant) based on EC₅₀ values and curve shapes. Gene set enrichment analysis was performed as described previously (38).

All animal procedures were approved by Pfizer's Institutional Animal Care and Use Committee (IACUC) and Crown Bio's IACUC and completed in accordance with the guidelines and regulations of the IACUC, NIH, and Animal Welfare Act.

A427 xenograft experiments were conducted at Pfizer and performed using 6- to 8-week-old female NSG mice purchased from The Jackson Laboratory (strain name NOD.Cg-Prkdc^scidIL2rg^tm1Wjl/SzJ). In brief, 5 × 10⁶ A427 cells (0.2 mL in 50% Matrigel, Trevigen; and 50% serum-free RPMI1640, Gibco) were inoculated subcutaneously into the right flank of each mouse. NCI-H441 xenograft experiments were conducted at Crown Bio using 7- to 8-week-old male Nu/Nu mice purchased from Charles River Laboratories (strain name Crl:NU-Foxn1nu). NCI-H441 cells (5 × 10⁶; 0.1 mL in 50% Matrigel, 50% PBS) were injected subcutaneously into the flank of each mouse. Once tumors were palpable, tumor length and width were measured by calipers 2 to 3 times weekly. Animals were randomly assigned to experimental groups and treatment was initiated at day 0, when tumor volume reached 150 mm³ on average. Ten animals were enrolled in each treatment arm. Tumor volume was calculated by the standard formula L × W² × 0.5. For all lung xenograft experiments, PF-06939999 was administered daily by oral gavage at doses of 3, 10, and 30 mg/kg [vehicle: 0.5% methylcellulose (w/v) solution with 0.1% polysorbate 80 (w/v) in water]. In addition, the NCI-H441 in vivo experiment included an additional treatment arm of PF-06939999 dosed orally at 5 mg/kg twice daily.

Values are reported as the mean ± SD, unless noted as SEM in the figure, from triplicates. GraphPad Prism software was used for statistical analysis. For mouse studies, effects of drug treatment on tumor volume were analyzed by two-way ANOVA, with multiple comparisons using Tukey post hoc test. Statistics programs in R were used to perform one-sided t tests for genetic associations of sensitivity. A P value less than 0.05 was considered statistically significant.

Crystal structures are publicly available in the Protein Data Bank: 7MX7, 7MXA, 7MXC, 7MXG, 7MXN.

RNA sequencing (RNA-seq) data are publicly available at the NCBI Gene Expression Omnibus by GEO accession number: GSE174615. All other data are included in the text or Supplementary Material.

Our discovery effort began with the published crystal structure of PRMT5:MEP50 complexed with the nucleoside A9145C (39). To minimize the synthetic complexity associated with A9145C and AdoMet analogs, we employed a ligand truncation strategy to define a minimum pharmacophore. This effort led to a key finding of adenosine (Fig. 1A) as an efficient inhibitor of PRMT5 (K_i = 100 nmol/L; CI = 70–140 nmol/L). Determination of the PRMT5:MEP50:adenosine co-crystal structure revealed adenosine bound in the cofactor pocket, with the purine and the sugar binding similar to what is observed in the A9145C co-crystal structure (Fig. 1A). Specifically, we observed the purine donating an H-bond to Asp419 and accepting from Met420. In addition, both ribose hydroxyls were observed interacting with Glu392 and the 5′-hydroxyl appeared to interact with a network of crystallographic water molecules. Finally, a deep binding pocket was observed adjacent to the 5′-carbon, providing a convenient vector for establishing additional protein/ligand interactions. An iterative protein structure and property-based design effort eventually led to incorporation of a 3-fluoro-4-chlorobenzene at the 5′ position. Co-crystal structure determination showed that the dihalogenated phenyl ring efficiently filled space in the deep pocket and established an edge to face interaction with Tyr324 (Fig. 1A). Careful consideration around protein pocket complementarity and small molecule physical properties led to PF-06855800 (Fig. 1A), a high quality in vitro and in vivo tool compound (31).

Tool compound PF-06855800 was utilized to design a structurally orthogonal, cofactor competitive series of inhibitors with diverse physicochemical properties. Further examination of the deep binding pocket associated with the PF-06855800 co-crystal structure revealed opportunities to interact with Glu444. Glu444 is a catalytically important residue responsible for binding the substrate Arg guanidine terminus during the methyltransferase reaction. To achieve this interaction, a complete redesign of the PF-06855800 deep pocket binding region was required. Guided by the PF-06855800 co-crystal structure, we first removed the 5′ carbon and attached hydroxyl, and moved the ribose oxygen from within the sugar ring to the 5′ position. This modification created a phenyl ether and positioned the phenyl ring in the deep pocket such that appropriate vectors were made available to target Glu444 with basic amines. Second, we altered the overall substitution pattern around the dihalogenated phenyl ring. With a slightly different pose of the phenyl ring buried in the deep pocket, a redesign of phenyl substituents was required to preserve efficient pocket complementarity.

Targeting Glu444 from the phenyl ring led to design of a nitrogen-containing 6-membered ring, ultimately yielding a tetrahydroisoquinoline (THIQ) system. Modeling the THIQ suggested the basic nitrogen could potentially form a charge–charge interaction with Glu444 residue. Moreover, modeling predicted the THIQ nitrogen could establish H-bonds directly with Leu437 and Glu435 backbone carbonyl oxygens. Structure-based iterative design into the deepest region of the cofactor site binding pocket led to optimal THIQ substitution. A fluorine atom para to the ether link along with an adjacent difluoromethyl group ultimately gave PF-06939999. A co-crystal structure was solved with PF-06939999 bound in the cofactor site with extension into the edge of the Arg substrate binding site (Fig. 1A). As modeling suggested, the co-crystal structure confirmed interaction with Glu444, Leu437, and Glu435. Moreover, the deep pocket was efficiently complemented by THIQ halogen substitution. PF-06939999 is a small molecule possessing a desirable balance between tight enzyme binding and drug-like physicochemical properties.

PF-06855800 and PF-06939999 are tight binding reversible inhibitors with demonstrated biochemical and cellular activity against PRMT5. Enzyme assays using full length human PRMT5/MEP50 protein complex were used to estimate inhibition constants (K_i) and evaluate inhibitor mechanism of action. A K_i value of 11 pmol/L (95% CI = 7–17 pmol/L) was determined for PF-06855800; however, an accurate K_i for PF-06939999 could not be calculated because the potency was below the lower limit of quantitation for the assay (<5 pmol/L). The mechanism of inhibition for PF-06855800 was investigated by determining the IC₅₀ at multiple concentrations of SAM substrate (40). A linear increase in IC₅₀ was observed upon increasing SAM concentration, consistent with a competitive inhibition model (Supplementary Fig. S1A).

Direct binding studies were performed to quantify binding affinity and inhibitor off-rate for PF-06939999. Single-cycle kinetic analysis was performed by surface plasmon resonance (SPR; Fig. 1B; Table 1), yielding a K_D of 5.8 pmol/L and an off-rate of 1.87 × 10^–5 s^–1, corresponding to an enzyme-inhibitor complex half-life of over 600 minutes. Given the structural similarity to PF-06855800, PF-06939999 is also expected to be a SAM-competitive inhibitor; however, the long half-life and high potency of PF-06939999 make equilibrium binding studies difficult. Therefore, we attempted to evaluate SAM competition by preforming the enzyme–SAM complex using various concentrations of SAM and initiating the reaction in the presence of various concentrations of inhibitor. Resultant IC₅₀ values vary as a function of SAM concentration (Fig. 1C), suggestive of competitive inhibition. Superposition of PF-06939999 bound to PRMT5/MEP50 with the PRMT5/MEP50 S-adenosyl-homocysteine (SAH) structure (Fig. 1D) shows PF-06939999 occupying a large portion of the SAH binding pocket, further supporting a SAM-competitive mechanism for PF-06939999.

Selectivity studies across a panel of protein methyltransferases indicate that PF-06855800 is at least one million–fold selective for PRMT5 over the other enzymes in the panel and has a similar selectivity profile to adenosine (Supplementary Table S2). We also demonstrate that PF-06855800 is at least 250,000-fold selective across a panel of 40 diverse protein kinases (Supplementary Tables S3 and S4). Selectivity of PF-06939999 at 10 μmol/L was assessed in both protein methyltransferase and kinase selectivity panels and showed no activity above 20% inhibition (Supplementary Tables S2 and S3).

PF-06939999 showed moderate plasma clearance and steady-state volume of distribution (∼40 mL/min/kg and 3.8 L/kg, respectively) in male Wistar-Han rats following a single intravenous administration at the dose of 2 mg/kg (Supplementary Fig. S1B). Oral bioavailability was moderate (∼40%) in rats following a single oral administration at the dose of 10 mg/kg. Elimination half-lives for the intravenous and oral administration were 1.5 and 3.2 hours, respectively, suggesting flip-flop kinetics, although it might be due to the limited time points after 7 hours postdose. These results demonstrate that PF-06939999 is orally available in animals.

The emergence of drug resistance is common for many targeted and chemotherapies utilized in oncology. PF-06855800 demonstrates dose-dependent reduction of symmetric dimethyl arginine (SDMA) and an antiproliferative response in A2780 cells (Supplementary Fig. S2A and S2B). Drug-resistant cells were generated by treating cells with proliferation IC₅₀ doses of SAM-competitive inhibitor PF-06855800 or peptide substrate inhibitor EPZ015666 (21), and increasing the dose as cells became resistant (Fig. 2A). After 13 weeks, cells treated with EPZ015666 developed complete resistance up to 15 μmol/L. Interestingly, PF-06855800 cells developed partial resistance, evidenced by a 5-fold shift in IC₅₀ after 38 weeks of constant drug exposure. Once compound was removed, both resistant cell lines maintained drug resistance following extended culture in drug-free media, indicating a stable drug resistance mechanism.

Cells resistant to PF-06855800 (A2780–5800R) or EPZ015666 (A2780–5666R) were tested for sensitivity to both inhibitors. A2780–5666R cells were insensitive to EPZ015666 (IC₅₀ > 10 μmol/L) but remained sensitive to PF-06855800 (Fig. 2B). A2780–5800R cells retained sensitivity to PF-06855800 and EPZ015666, but the IC₅₀ of PF-06855800 shifted 5-fold compared with parental cells. SDMA levels in cells treated with both compounds associated with the antiproliferative activity of each compound, that is, resistant cells did not demonstrate SDMA reduction (Supplementary Fig. S2C). Overall, these data suggest emergence of compound-selective drug resistance mutations.

To identify genetic mechanisms of drug resistance, whole-exome sequencing was performed on A2780–5800R and A2780–5666R cells (Supplementary Data S1). Unique missense mutations were identified in the PRMT5 coding region. A2780–5666R cells acquired missense mutation 979T>C, converting phenylalanine 327 to leucine. This mutation resides in the binding site for EPZ015666, predicted to negatively impact compound binding (Fig. 2C). Attempts to generate co-crystal structures of EPZ015666 bound to PRMT5(Phe327Leu):MEP50 were unsuccessful. A2780–5800R cells acquired missense mutation 1259T>C, converting methionine 420 to threonine. Co-crystal structures obtained for PRMT5(Met420Thr):MEP50 bound to PF-06855800 (Fig. 2C) or PF-06939999 (Supplementary Fig. S2D) highlight that the compound interaction with the threonine in weaker than with methionine in wild type PRMT5:MEP50 due to a loss of van der Waals interactions between the side chain and the pyrrolopyrimidine ring.

To further characterize the impact of identified PRMT5 mutations on inhibitor sensitivity, biochemical K_i values were generated. PRMT5(Phe327Leu)/MEP50 enzyme complex demonstrated similar inhibition to PF-06855800 (Ki = 3 pmol/L) as wild-type PRMT5 (Ki = 11 pmol/L), although resistant to EPZ015666 treatment at 10 μmol/L. In contrast, PRMT5(Met420Thr)/MEP50 showed similar inhibition to EPZ015666 (Ki = 270 nmol/L) as wild-type PRMT5 (Ki = 220 nmol/L), and demonstrated partial resistance to PF-06855800 with a 22-fold potency shift in K_i (220 pmol/L). These data suggest SAM-competitive PRMT5 inhibitors may be less susceptible to developing mutations that confer complete drug resistance since only conservative amino acid changes may be tolerated in the cofactor binding pocket.

In A427, PF-06939999 treatment for 72 hours showed dose-dependent loss of SDMA on several proteins, with a cellular IC₅₀ of 1.1 nmol/L (Fig. 3A). Purified histones were also analyzed for levels of H3R8me2s, H4R3me2s, and H2AR3me2s upon inhibitor treatment. In concordance with results reported with other PRMT5 inhibitors (21, 41), neither PF-06939999 or PF-06855800 decreased histone methylation after 72-hour treatment (Supplementary Fig. S3A and S3B). Moreover, cellular fractionation and immunofluorescence show PRMT5 localization uniquely in the cytoplasm in several NSCLC cell lines (Supplementary Fig. S3C and S3D). Overall, these data suggest the primary mechanism of action of these PRMT5 inhibitors is independent of direct chromatin regulation.

PRMT5 overexpression is a driver of tumor cell growth and survival in NSCLC. PF-06939999 treatment led to dose-dependent antiproliferation in NSCLC cell lines (Fig. 3B). To better understand mechanisms driving growth arrest, cells were analyzed for phenotypic responses to PF-06939999 treatment. Cell-cycle analysis demonstrated mixed cell-cycle effects; A427 cells arrested in G₁, NCI-H1975 cells arrested in G₂–M, and NCI-H441 showed asynchronous cell-cycle arrest in response to PF-06939999 treatment (Fig. 3C; Supplementary Fig. S4A–S4C). Only A427 cells showed modest induction of apoptosis, as measured by increased cleaved PARP and cleaved caspase-3 (Fig. 3D; Supplementary Fig. S4D) in response to PF-06939999 treatment. Treatment of A549 cells with PF-06939999–induced senescence, evidenced by cell morphology changes and β-galactosidase staining ((Fig. 3E). Interestingly, A549 and A427 cells are wild type for p53, but do not respond with a strong apoptotic response as observed in p53 wild type lymphoma cells (23).

To understand the molecular mechanisms of PF-06939999 response, NSCLC cell lines were analyzed by RNA-seq following drug treatment. Differential gene expression was analyzed in A427, NCI-H441, and NCI-H1975 cells (Supplementary Fig. S5) after PF-06939999 treatment, and 277 genes were differentially regulated in all three cell lines (Supplementary Fig. S5A and S5E). Gene sets unique to individual cell lines suggest mechanisms of drug response may be dependent on cellular context (Supplementary Fig. S5B and S5C). However, common upregulated genes showed enrichment in several pathways involved in RNA splicing and processing, and common downregulated genes were enriched in metabolic pathways (Supplementary Fig. S5D).

PRMT5 is an important regulator of pre-mRNA splicing, so the RNA-seq data was further analyzed for alternative splicing changes (Fig. 4; Supplementary Fig. S6). PF-06939999 treatment increased exon skipping and intron retention with minimal impact on other types of alternative splicing events (Fig. 4A; Supplementary Fig. S6A). Interestingly, overlap of the skipped exons across the three cell lines suggests a set of pre-mRNA splicing events consistently dependent on PRMT5 function in NSCLC (Fig. 4B; Supplementary Fig. S6D), with a conserved set of genes showing 577 exon skipping events shared by all three cell lines, and 1671 shared by at least two, after PF-06939999 treatment. Gene ontology analysis of the skipped exons shows enrichment for microtubule organization, cell cycle, and DNA repair pathways (Fig. 4C; Supplementary Fig. S6B). The number of cassette exons increased upon PF-06939999 treatment was lower compared with skipped exons, with 130 events common to all three cell lines (Fig. 4D). Pathway analysis for higher cassette exon inclusion identified genes involved in cell cycle and RNA splicing pathways (Fig. 4E; Supplementary Fig. S6C). These results highlight the importance of PRMT5 in regulating pre-mRNA splicing and suggest a unique set of genes highly dependent on PRMT5 function for posttranscriptional regulation.

To further validate the antitumor effects of PF-06939999 treatment in vivo, tumor growth inhibition studies (TGI) were conducted in two mouse xenograft models of NSCLC. PF-06939999 showed significant tumor suppression as an orally administered single agent in two splicing factor mutant NSCLC xenograft models. PF-06939999 demonstrated dose dependent TGI in A427 (RBM10^I348N) tumors at day 44, of 52.1% at 3 mg/kg, 74.9% at 10 mg/kg, and 100.6% at 30 mg/kg once daily doses (Fig. 5A). Modulation of SDMA in A427 tumors was evaluated via ELISA at the end of the study (Fig. 5B). PF-06939999 was well tolerated with minimal body weight loss (Fig. 5C). In the NCI-H441 (U2AF1^S34F) model, PF-06939999 demonstrated dose-dependent TGI at day 36 of 55.5% at 5 mg/kg twice a day, 14.9% at 3 mg/kg once daily, 46.6% at 10 mg/kg once daily, and 87.2% at 30 mg/kg once daily doses (Fig. 5D). SDMA levels assessed in the NCI-H441 tumors were evaluated via ELISA at the end of the study (Fig. 5E). PF-06939999 was well tolerated with minimal body weight loss (Fig. 5F).

Following repeated oral administration of PF-06939999 to mice bearing xenograft tumors with NCI-H441 at the doses of 3, 10, and 30 mg/kg, PF-06939999 was rapidly absorbed with t_max of 1 to 2 hours (Supplementary Fig. S7A). Mean unbound C_max ranged from 28 to 333 nmol/L at the doses of 3 to 30 mg/kg, whereas mean unbound AUC were 165 to 1,266 nmol/L/h/mL. Increases in oral exposures were dose proportional. Plasma concentrations of PF-06939999 in mice bearing xenograft tumors with A427 at the doses of 3, 10, and 30 mg/kg were comparable with those in xenograft models with NCI-H441 (Supplementary Fig. S7B). Unbound C_ave (AUC divided by a dosing interval of 24 hours) in NCI-H441 was 2.3- to 17.7-fold higher than the lowest cytotoxic concentration of 3 nmol/L and unbound C_ave was 0.51- to 4.3-fold higher than the lowest cytotoxic concentration of 10 nmol/L in A427 (Supplementary Fig. S7C–S7F).

To identify molecular features influencing response to PF-06939999, a proliferation screen across a panel of NSCLC cell lines was conducted. Cell lines exhibited a range of sensitivities to PF-06939999 with 54% of the cell lines demonstrating GI₅₀ at ≤50 nmol/L (Fig. 6A), a dose representing IC₉₀ of SDMA modulation. To facilitate the analysis of biomarkers predictive of response to PF-06939999 in an unbiased manner, we analyzed screening data using multivariate elastic net analysis (Supplementary Data S2–S3). Molecular association analysis of genetic features, including mutations and copy number, did not identify any statistically significant combination of features associated with inhibitor response. Previous studies have identified several genetic factors influencing cellular response to PRMT5 genetic loss or inhibitors, including MTAP deletion (42–44), MYC expression (25), TP53 mutation (23), and CLSN1A expression (45). Although shRNA knockdown of PRMT5 has been shown to be synthetic lethal with MTAP deletion, there was no significant association with PF-06939999 sensitivity in NSCLC in this study (P = 0.97). In addition, our studies did not show significant association between p53 status and sensitivity to PRMT5 inhibitors in NSCLC (P = 0.26).

Cell line gene expression was analyzed for pathways associated with sensitivity and resistance to PF-06939999 treatment in NSCLC. Gene set enrichment analysis (GSEA; ref. 38) identified MYC, cell cycle, and DNA repair pathways positively associated with sensitivity to PRMT5 inhibition (Fig. 6B; Supplementary Fig. S8A and S8B). Target genes of MYC include many RNA processing proteins, and MYC overexpression leads to increased dependency on core splicing machinery (25). Because proteins important in splicing are known substrates of PRMT5 (4, 17, 18), and mutations in many splicing regulators are reported to influence tumorigenesis (28, 46), we analyzed global alternative splicing patterns associated with sensitivity to PF-06939999 in NSCLC. Basal alternative splicing patterns showed differences in total cassette exons, 5′ and 3′ splice site events, and retained introns in sensitive versus more resistant cells (Fig. 6C). RNA processing genes have been identified in tumors, but individual gene mutations are found at low frequency and are often lineage specific (29). In NSCLC, the most common splicing factor mutations are RBM10 and U2AF1, represented at approximately 11% and 6%, respectively (29). Using the CCLE mutation annotation for RBM10 and U2AF1, a t test showed significance with RBM10 mutations (P = 0.0016) in the NSCLC cell lines most sensitive to PF-06939999 inhibition (Fig. 6D), but not with U2AF1 hotspot mutations (P = 0.46).

To further investigate the association of RBM10 mutation with PF-06939999 response in NSCLC cell lines, we analyzed RBM10 protein expression by Western blot analysis. Cell lines with RBM10 frameshift mutations have undetectable amounts of protein compared to wild type cells, while cell lines containing RBM10 missense mutations show reduced protein expression compared to wild type cells (Fig. 6D). Because the frameshift mutation in NCI-H1975 causes loss of RBM10, these cells were used to evaluate the impact of RBM10 mutation on PF-06939999 sensitivity. NCI-H1975 cells were transfected with full-length RBM10 cDNA, and clones were identified that expressed RBM10 protein at levels comparable with a RBM10 wild-type cell line (NCI-H460; Fig. 6E). Analysis of alternative splicing comparing RBM10 cDNA expressing NCI-H1975 cells with parental shows significant increases in cassette exon inclusion with minimal impact to other types of alternative splicing, as is predicted for the function of RBM10 (Supplementary Fig. S8C; ref. 47). Re-expression of RBM10 in NCI-H1975 cells showed a 4- to 5-fold reduction in sensitivity to PF-06939999 compared with parental cells, suggesting that RBM10 loss functionally impacts response to PF-06939999 (Fig. 6E).

PF-06939999 is a potent and selective SAM-competitive PRMT5 inhibitor, discovered through structure-based design. PF-06939999 displays in vitro and in vivo antiproliferative activity in cancer cells concomitant with loss of SDMA levels. Utilizing models of acquired drug resistance, we demonstrate that SAM-competitive PRMT5 inhibitors similar to PF-06939999 have a lower propensity for the development of mutations that confer complete drug resistance compared to SAM-cooperative substrate site inhibitors such as EPZ015666. Structure-based analysis highlights the requirement of key amino acids residues in the PRMT5 cofactor binding pocket similarly required for binding of SAM or PF-06939999. Indeed, only conservative amino acid changes that slightly shift the binding potency of PF-06855800 were identified in drug-resistant cell lines. Drug-resistant cell lines developed using the SAM-cooperative binding compound EPZ015666 displayed complete drug resistance consistent with the acquisition of a nonconservative amino acid change in the substrate pocket that reduces inhibitor binding. The location of this mutation is unlikely to interfere with either the substrate recruitment or PRMT5 enzymatic function, while greatly interfering with binding of EPZ015666. The maintained inhibitory activity of PF-06855800 in EPZ015666-resistant cell lines and EPZ015666 in PF-06855800–resistant cell lines highlights the specificity of these mutations for the specific binding mode of each inhibitor.

While PRMT5 inhibitors have demonstrated broad antiproliferative activity across hematologic and solid tumor cell lines (23, 30, 45), this is the first in-depth characterization of PRMT5 inhibition in NSCLC. PRMT5 is overexpressed in lung cancer, conferring oncogenic function leading to hyperproliferation and accelerated metastasis. Consistent with several recent reports, we have shown reduced protein SDMA levels on numerous cellular proteins following PF-06939999 treatment. Mass spectrometry analysis has identified many proteins involved in pre-mRNA splicing including SmB, D1, and D3 as targets of PRMT5 (17, 18). The cytoplasmic localization of the PRMT5/MEP50 complex and lack of changes on histone SDMA following PF-06939999 treatment contradicts a role for PRMT5 as a direct chromatin regulator.

RNA-seq analysis of PF-06939999–treated NSCLC cell lines has identified alternative splicing changes, both exon skipping and intron retention events, consistent with the role of PRMT5 in pre-mRNA regulation. Interestingly, comparative analysis of splicing changes in 3 cell lines identified a subset of shared splicing changes induced by PRMT5 inhibition, suggesting that regulation of specific exon/intron features may be more dependent on protein SDMA. One example was induction of the p53–p21 pathway, which has previously been described as a consequence of MDM4 splicing changes induced by PRMT5 inhibition in cell lines with wild-type TP53, although TP53 mutational status did not stratify with PF-06939999 response in our NSCLC cell line panel. These data highlight the complexity of additional splicing changes in genes involved in additional pathways impacting cell proliferation.

The strongest molecular features associated with sensitivity of NSCLC cell lines to PF-06939999 were MYC, cell cycle, and splicing pathways. The relationship between MYC pathway activation and splicing fidelity has been reported previously (25). Splicing dysregulation is a well-characterized molecular feature of NSCLC. Indeed, the incidence of 119 splicing factor gene mutations in LUAD is greater than 60%, one of the highest percentages of all tumor types (29). RBM10 mutation or deletion is the most common splicing factor mutation in NSCLC, identified at a frequency of 8% (48). RBM10 loss in NSCLC has been associated with exon inclusion, including cancer-relevant genes such as NUMB and CREBB (47, 49). Experimental re-expression of RBM10 in RBM10-mutant NSCLC cells normalized splicing patterns and attenuated the cellular response to PF-06939999. These results are consistent with recent reports that additional splicing factor mutations such as SF3B1 and SRSF2 in AML may sensitize cells to PRMT5 inhibition (30). Identification of predictive biomarkers of response at lower doses of PF-06939999 may be important for increasing the therapeutic index of PRMT5 inhibitors in the clinic. Overall, targeting dysregulation in splicing pathways represent an opportunity for PRMT5 inhibitors, such as clinical candidate PF-06939999, in the treatment of cancer.

K. Jensen-Pergakes reports other support from Pfizer outside the submitted work, as well as a patent for PC072700 pending to Pfizer. I.J. McAlpine reports a patent for US2017348313 AA issued and a patent for US2016244475 AA issued. M.A. McTigue reports a patent for 2016244475A1 issued and a patent for 2017348313A1 issued, and is an employee and stock holder of Pfizer, Inc. T. Xie reports other support from Pfizer outside the submitted work, as well as a patent for PC072700 pending. C.P. Dillon is full-time employee of Pfizer. Y. Wang reports a patent for PC072700 pending. S. Yamazaki reports other support from Pfizer Inc. and Pfizer Inc. outside the submitted work. E. Hendrickson reports other support from Pfizer, Inc. during the conduct of the study and other support from Pfizer, Inc. outside the submitted work. C. Chung reports personal fees from Pfizer Inc. during the conduct of the study. R.A. Kumpf reports a patent for US 2016/0244475 Al pending to Pfizer and a patent for US 2017/348313 A1 pending to Pfizer. R.L. Patman reports a patent for 2016244475A1 issued and a patent for 2017348313A1 issued. M. Tran-Dube reports a patent for US2017/0348313 A1 issued. M. Wythes reports a patent for US2017348313 issued to Pfizer owned. T.A. Paul reports personal fees from Pfizer, Inc. during the conduct of the study and personal fees from Pfizer, Inc., outside the submitted work. No disclosures were reported by the other authors.

K. Jensen-Pergakes: Conceptualization, supervision, investigation, methodology, writing–original draft, writing–review and editing. J. Tatlock: Supervision, investigation, methodology, writing–review and editing. K.A. Maegley: Supervision, validation, investigation, visualization, methodology, writing–original draft. I.J. McAlpine: Conceptualization, resources, supervision, validation, investigation, visualization, methodology, writing–original draft. M. McTigue: Conceptualization, supervision, investigation, visualization, methodology. T. Xie: Data curation, software, formal analysis, visualization. C.P. Dillon: Conceptualization, supervision, investigation, methodology. Y. Wang: Supervision, investigation. S. Yamazaki: Data curation, investigation. N. Spiegel: Investigation. M. Shi: Investigation. A. Nemeth: Investigation. N. Miller: Investigation. E. Hendrickson: Investigation, visualization. H. Lam: Investigation. J. Sherrill: Data curation, investigation, visualization. C. Chung: Data curation, formal analysis, visualization. E.A. McMillan: Data curation, software, formal analysis. S.K. Bryant: Investigation. P. Palde: Supervision, investigation, visualization, methodology, writing–review and editing. J. Braganza: Investigation. A. Brooun: Supervision, validation, investigation, visualization, methodology. Y. Deng: Investigation. V. Goshtasbi: Investigation. S.E. Kephart: Investigation. R.A. Kumpf: Resources, supervision, investigation. W. Liu: Investigation. R.L. Patman: Formal analysis, investigation. E. Rui: Conceptualization, formal analysis, supervision, validation. S. Scales: Investigation. M. Tran-Dube: Investigation. F. Wang: Conceptualization. M. Wythes: Conceptualization, supervision. T.A. Paul: Formal analysis, supervision, investigation, writing–original draft, writing–review and editing.

We would like to thank Dr. Zhengang Peng and his scientific team, at WuXi AppTec for their experimental contributions to these studies.

All studies were funded by Pfizer, Inc.