CRISPR-Drug Combinatorial Screening Identifies Effective Combination Treatments for MTAP-deleted cancer

Abstract

CDKN2A/MTAP co-deletion occurs frequently in non-small cell lung cancer and other solid tumors, including glioblastoma and pancreatic ductal adenocarcinoma. Lung cancer remains the leading cause of cancer-related mortality, and fewer than 15% of glioblastoma or pancreatic cancer patients survive 5 years, underscoring the need for more effective therapies[1], [2], [3], [4], [5]. PRMT5 is a synthetic-lethal dependency in MTAP-null tumors and an attractive therapeutic target for CDKN2A/MTAP-deleted cancers. A new revolutionary class of inhibitors, referred to as MTA-cooperative PRMT5 inhibitors, has shown promising results in ongoing early phase clinical trials. Nonetheless, effective cancer treatment typically requires therapeutic combinations to improve response rates and defeat emergent resistant clones. Thus, we sought to determine whether perturbation of other pathways could improve the efficacy of MTA-cooperative PRMT5 inhibitors. Using a paralog and single gene targeting CRISPR library we screened MTAP-deleted cancers in the presence or absence of MTA-cooperative PRMT5 inhibitors. Loss of several genes sensitized to PRMT5 inhibition, including members of the MAPK pathway. Chemical inhibition of MAPK pathway members using KRAS, MEK, ERK, and RAF inhibitors synergized with PRMT5 inhibition to kill CDKN2A/MTAP-null, RAS-active tumors. Further, MTA-cooperative PRMT5 inhibitors combined with either KRAS or RAF inhibitors led to complete responses in vivo, emphasizing the potential benefit for patients. Lastly, cell lines resistant to KRAS inhibition were not resistant to MTA-cooperative PRMT5 inhibitors and vice versa, suggesting non-cross-reactive mechanisms of resistance. Overall, this study identifies therapeutic combinations with MTA-cooperative PRMT5 inhibitors that may offer significant benefit to patients.

Introduction

Homozygous deletion of the cyclin-dependent kinase inhibitor 2A (CDKN2A) tumor suppressor locus is the most frequent focal deletion event in human cancer[6], [7], [8], [9], [10], occurring in 17% of all human cancers (TCGA), >50% of glioblastomas (GBM), 30% of pancreatic ductal adenocarcinoma (PDAC), and 15% of non-small cell lung cancer (NSCLC)[4], [10], [11], [12]—three cancers in dire need of new therapies. PDAC and GBM have 5-year survival rates below 15% and lung cancer remains the leading cause of cancer-related mortality, highlighting the need for more effective therapies in these tumors[3], [13], [14]. Deletion of CDKN2A frequently results in the co-deletion of methylthioadenosine phosphorylase (MTAP) as it lies within 50 Kb of the CDKN2A locus. In tumors lacking MTAP, its substrate, methylthioadenosine (MTA) accumulates and partially inhibits the protein arginine methyltransferase 5 (PRMT5) enzyme by competing with the PRMT5 cofactor S-adenosyl methionine (SAM) for binding to the PRMT5 active site, rendering tumor cells sensitive to further PRMT5 loss[6], [8], [9]. In keeping with this model, loss of methionine adenosyltransferase 2A (MAT2A), an enzyme required for the generation of SAM, is also synthetic lethal with MTAP loss; thus, the relative cellular abundance of SAM and MTA metabolites dictates the synthetic lethal relationships of PRMT5 and MAT2A with MTAP loss. This dependency makes PRMT5 a synthetic lethal therapeutic target in MTAP null cancers[7], [15].

PRMT5 regulates a host of important cellular functions including mRNA splicing, transcription, chromatin dynamics, and ribosome biogenesis[16], [17], [18], [19], [20], [21], [22]. Acting in co-complex with its requisite co-factor WDR77 and substrate adaptor proteins, the PRMT5 methylosome symmetrically dimethylates arginine residues (SDMA) within its target proteins to regulate cellular functions including chromatin organization, mRNA splicing, transcription, and cellular differentiation[16], [17], [18], [19], [20], [23], [24], [25], [26], [27]. Systematic SDMA substrate profiling and transcriptional analyses have demonstrated that PRMT5 is a key regulator of mRNA processing through SDMA modification of arginine and glycine-rich regions of substrate proteins including histones, ribosomal proteins, and splicing factors[24], [28], [29], [30]. Not surprisingly then, PRMT5 is a pan-essential gene in CRISPR knockout studies[15]. Yet the decreased activity seen in MTAP null tumor cells, creates a therapeutic window wherein one can inhibit the diminished PRMT5 activity in tumor cells while preserving viability in the MTAP/PRMT5 wildtype (WT) cells[6]. Previous PRMT5 inhibitors (PRMT5i) bound to the catalytic pocket of PRMT5 in a SAM cooperative and MTA-independent manner[31], [32], [33]. These compounds are insensitive to MTA accumulation and hence have non-selective cellular toxicity[6], [31], [32], [33]. Biochemical data suggests that therapeutics that target PRMT5 in a mechanism that cooperates with MTA could leverage the accumulation of MTA to increase the cellular selectivity of PRMT5 inhibition[6]. Indeed, recently developed MTA-cooperative PRMT5 inhibitors (MTAC-PRMT5i) including TNG462, TNG908, AMG-193, AZ-PRMT5–1, and MRTX1719 have been developed and are in the clinic[34], [35], [36], [37], [38]. In phase I and phase I/II studies of MTAC-PRMT5i, responses were seen in 27–33% of patients with treatment refractory tumors at well-tolerated doses[35], [39]. Notably, early data suggests that hematologic toxicity attendant to the first-generation PRMT5i is not a dose-limiting toxicity for the MTAC-PRMT5i. While these data are encouraging, a ~30% response rate among CDKN2A/MTAP null cancers suggests that the efficacy of MTAC-PRMT5i can be further improved. Moreover, it is likely that, with this level of response, progression free survival will be short. Thus, there will be a need to develop – potentially through rational studies – combinations with other anti-cancer drugs.

To advance towards this goal, we first evaluated the activity of three MTAC-PRMT5i across 900 barcoded cell lines and compared their selectivity profile to that of a first-generation PRMT5i and other selective or non-selective inhibitors finding that increased in vitro selectivity activity can be observed with this mode of action. Next, we employed functional drug-CRISPR combinatorial screens using a novel paralog and single gene targeting CRISPR library, to explore and identify combination targets that might increase the initial response rates and attain more durable responses. We found a number of individual genes and paralogous gene pairs that act as genetic sensitizers to MTAC-PRMT5i in CDKN2A/MTAP null cancer including the anti-apoptotic genes BCL2/BCL2L1 and RAS-MAP kinase genes: MAPK1 (ERK2), RAF1 (c-Raf), BRAF/RAF1 and KRAS. Notably, several members of the RAS MAP Kinase signaling pathway were discovered as sensitizers to MTAC-PRMT5i in our screen. MAPK signaling promotes tumorigenesis and is hyperactivated in many cancers, including 30% of NSCLC that harbor KRAS mutations[1], [2]. Importantly, the combination of adagrasib and MRTX1719 or the combination of belvarafenib and MRTX1719 were synergistic and led to greater combined activity leading to complete responses in KRAS, MTAP double-mutant NSCLC. Additionally, the drug combinations were well tolerated in mice. The RAS pathway targeting inhibitors as single agents also suffer from response rates in the ~10–30% range with modest increases in progression free survival[40], [41], [42], [43], [44], [45]. Thus, our demonstration of marked combination efficacy of RAS-MAPK targeting agents and MTAC-PRMT5i suggest these combinations may provide significant benefit to patients.

Materials and Methods

Cell culture

Cell culture Mammalian cell lines 293T (ATCC CRL-3216^™, RRID:CVCL_0063, acquired in 2022), MIA PaCa-2 (ATCC CRM-CRL-1420, RRID:CVCL_0428, acquired in 2022), and A172 (ATCC CRL1620, RRID:CVCL_0131, acquired in 2022) were maintained in high glucose Dulbecco’s Modified Eagle Medium (DMEM) (Fisher Scientific); LU99 (Riken RCB1900, RRID:CVCL_3015, acquired in 2019), NCI-H1648 (ATCC CRL-5882, RRID:CVCL_1482, acquired in 2021), NCI-H1666 (ATCC CRL-5885, RRID:CVCL_1485, acquired in 2023), PANC02.13 (ATCC CRL2554, RRID:CVCL_1634, acquired in 2024), and SNU410 (RRID: CVCL_5059, acquired in 2024) were maintained in RPMI-1640 Glutamax media (Fisher Scientific); KP4 (Riken RCB1005, RRID:CVCL_1338, acquired in 2022) were maintained in DMEM/F-12 media (Fisher Scientific), Calu-1 (ATCC HTB-54, RRID:CVCL_0608, acquired in 2023) were maintained in McCoy’s 5 A (Fisher Scientific), and SW1573 (ATCC CRL-2170, RRID:CVCL_1720, acquired in 2022) were maintained in Leibovitz’s L-15 media (Fisher Scientific). All cell lines were maintained in media containing 10% fetal bovine serum (FBS) (Sigma-Aldrich) and 1% Penicillin/streptomycin (Fisher Scientific) at 37°C in a high humidity 5% CO2 atmosphere incubator (Heracell 240i, Thermo Fisher Scientific) or 0% CO2 atmosphere for SW1573. SNU410-Cas9 stable cell line was acquired from the Genetic Perturbation Platform at the Broad Institute in 2024. To maintain high quality performance in the experiments, cell lines were not kept in culture after passage 25, STR profiling was completed by ATCC on all cultured cell lines regularly, and cell lines were continuously checked monthly for mycoplasma contamination by PCR.

Organoid Studies

Patient-derived organoid cultures were initiated and maintained as previously described [46], [47], [48]. Tissue was obtained from patients undergoing primary pancreatic resections or metastatic biopsies at the Dana-Farber Cancer Institute after obtaining written informed consent under IRB-approved protocols (14–408, 03–189). Studies were conducted in accordance with recognized ethical guidelines such as the Belmont Report. For compound testing, organoids were dissociated to single cells. 1000 viable cells were seeded into each well of ultra-low attachment 384-well plates with 30 uL of complete medium containing 10% Matrigel by volume. After 24 hours, compounds were added individually or in combinations to each well over 8-point dose curves along with DMSO controls. Cells were cultured for 7 days in the presence of compounds before assessing viability by adding 30 uL of CellTiter-Glo 3D (Promega) to each well, incubating for 1 hour at room temperature on a shaker, and measuring luminescence using a microplate reader (Envision). Each combination compound condition was performed in triplicate, and each dose point was normalized to DMSO controls to estimate relative viability. At least 2 independent experiments were performed for each compound and organoid condition.

Lentivirus generation

4–6 million 293T cells were seeded and transfected the subsequent day with 9 μg of packaging plasmid (psPAX2: Addgene# 12260), 0.9 μg of envelope plasmid (pMD2.G: Addgene# 12259), and 9 μg of the transfer plasmid containing the sequence of interest that should be integrated into the target cell’s genome. For transfection, Lipofectamine 2000 reagent (Thermo Fisher Scientific) and OPTI-MEM (Thermo Fisher Scientific) were used following manufacturer’s instructions. 48h and 72h after transfection, virus was harvested by pelleting cell debris at 150 × g for 5 minutes and straining the viral supernatant through a 45 μm filter.

Generation of stable Cas-9-expressing cell lines

For lentiviral infection of cell lines, 5e5–1e6 cells were seeded per well of a 6-well and 2 ml of lentivirus containing Cas9 with a blasticidin resistance gene (Cas9 cDNA in a lentiviral expression vector Addgene #125592) was added to the cells. To enhance infection efficacy, 10 μg/ml polybrene was used. After 48h (of infection), cells were split into blasticidin-containing growth media (10 μg/ml) and selected for at least 5 days. As a control, uninfected cells were included in the selection to ensure only cells that integrated the desired vector survived.

PRISM Screening

The PRISM screen for GSK3326595/pemrametostat, TNG908, MRTX1719, and MRTX9768 was performed in over 900 cancer cell lines and 866 cell lines passed quality control metrics for GSK3326595 and MRTX1719 and for 864 cell lines passed quality control metrics for MRTX9768. Screen data are available in Supplemental Table 1. In brief, these compounds were plated on 384 well plates by a Beckman Coulter Echo 655 by acoustic transfer at 8 doses in three-fold dilution. Two PRISM cell line collections were used in the assay: PR500 (including only solid tumor cell lines), and PR300+ (including solid and hematopoietic cell lines). Pools of 20–25 DNA barcoded cell lines are plated at 1250 cells per well in 384 well plates for five-day treatment with the drug. Cells were then lysed in Qiagen TCL mRNA lysis buffer, and then reverse transcription PCR was performed. This procedure was done as previously described [49], [50].

To process the data and calculate the AUC and IC50, we completed the following steps:

1. Each detection well contained 10 control barcodes in increasing abundances as spike-in controls. For each plate, we first create a reference profile by calculating the median of the log₂(MFI) values across negative control wells for each of these spiked-in barcodes.

2. For each well, a monotonic smooth p-spline was fit to map the spike in control levels to the reference profile. Next, we transform the log₂(MFI) for each cell barcode using the fitted spline to allow well-to-well comparisons by correcting for amplification and detection artifacts.

3. Next, the separability between negative and positive control treatments was assessed. In particular, we calculated the error rate of the optimum simple threshold classifier between the control samples for each cell line and plate combination. Error rate is a measure of overlap of the two control sets and was defined as:

   $$Error=\frac{FP+FN}{n}$$

   where FP is false positives, FN is false negatives, and n is the total number of controls. A threshold was set between the distributions of positive and negative control log₂(MFI) values (with everything below the threshold said to be positive and above said to be negative) such that this value is minimized. Additionally, we also calculated the dynamic range of each cell line. Dynamic range was defined as

   $$DR={\mu }_{-}-{\mu }_{+}$$

   where μ+/− stood for the median of the normalized logMFI values in positive/negative control samples.
4. We filtered out cell lines with error rate above 0.05 or a dynamic range less than 1.74 from the downstream analysis. Additionally, any cell line that had less than 2 passing replicates was also omitted for the sake of reproducibility. Finally, we computed viability by normalizing with respect to the median negative control for each plate. Log-fold-change viabilities were computed as:

   $$log-viability={log}_2\left(x\right)-{log}_2\left({\mu }_{-}\right)$$

   where ${\text{log}}_2\left(\text{x}\right)$ is the corrected log₂(MFI) value in the treatment and log₂(μ−) is the median corrected log₂(MFI) in the negative control wells in the same plate.

5. Log-viability scores were corrected for batch effects coming from pools and culture conditions using the ComBat algorithm as described in Johnson et al [51].

6. We fit a robust four-parameter logistic curve to the response of each cell line to the compound:

   $$f\left(x\right)=b+\frac{a-b}{1+e^{slog\frac{x}{EC50}}}$$

   With the following restrictions:

   1. We require that the upper asymptote of the curve be between 0.99 and 1.01
   2. We require that the lower asymptote of the curve be between 0 and 1.01
   3. We do not enforce decreasing curves
   4. We initialize the curve fitting algorithm to guess an upper asymptote of 1 and a lower asymptote of 0.5
   5. When the standard curve fit fails, we report the robust fits provided by the dr4pl R-package and computed AUC values for each dose-response curve and IC₅₀ values for curves that dropped below 50% viability.

7. Finally, the replicates were collapsed to a treatment-level profile by computing the median log-viability score for each cell line.

We then calculate univariate associations between the PRISM sensitivity profiles (each dose, log₂(AUC), and log₂(IC₅₀)) and genomic features or genetic dependencies. Specifically, univariate models were run on available feature sets from the Dependency Map (depmap.org) including CCLE genomic characterization data such as gene expression, cell lineage, mutation, copy number, metabolomics, and proteomics, as well as loss-of-function genetic perturbation (both RNAi and CRISPR) data. In addition to these datasets, viability data from the PRISM drug repurposing project was used as a feature set for univariate analysis. For discrete data, such as mutation and lineage, a t-test and associated p-values were calculated to determine differential sensitivities. For continuous feature sets, we computed the Pearson correlations and associated p-values. For both analysis types, q-values were computed from p-values using the Benjamini-Hochberg algorithm.

Predictive Modeling

Next, for each log-viability at a dose, log₂(AUC), or log₂(IC₅₀) we trained and fit multivariate models using the molecular characterizations and genetic dependencies of the PRISM cell lines. The resulting importance of various features could be used to suggest potential biomarkers of compound response or to inform potential hypotheses for mechanisms of action.

In particular, we trained random forest models using: i. CCLE features (copy number alterations, RNA expression, mutation status and lineage annotation) ii. CCLE features + reverse phase protein array (RPPA) + CRISPR + microRNA (miRNA) + metabolomics (MET) data. These data are shown in Figure 1 and Figure S1 and in Supplemental Table 1.

Figure 1. PRISM screening identifies MTAP status as top biomarker of MTA-cooperative PRMT5i response.

A. Cell density plots of the cell growth area under the curve (AUC) response in PRISM screen of 875 cell lines for MRTX1719, MRTX9768, TNG908; pemrametostat (type I PRMT5 inhibitor); paclitaxel, adavosertib (Wee1 inhibitor), encorafenib (BRAF inhibitor) and asciminib (allosteric BCR-ABL inhibitor). Lower two panels are density plots of the Chronos score for PRMT5 gene depletion by CRISPR Cas9 KO or the Demeter2 score for shRNA depletion from the cancer dependency map[15].

B. Scatterplot comparing AUC values for MRTX9768 and for MRTX1719 for all PRISM assayed cell lines, both at 1μM. 76 MTAP null cell lines are shown in blue; MTAP proficient cell lines are in red.

C. Scatterplot comparing AUC values MRTX9768 AUC and for pemrametostat, both at 1 μM for all PRISM assayed cell lines. 76 MTAP null cell lines are shown in blue; MTAP proficient cell lines are in red.

D. Top shRNA gene knockdowns correlated with the PRISM AUC response for MRTX1719 at 1μM.

E. Top shRNA gene knockdowns correlated with PRISM AUC response for pemrametostat at 1μM.

F. Top genes whose mRNA expression correlates with PRISM AUC response for MRTX1719 at 1μM.

G. Top genes whose mRNA expression correlates with PRISM AUC response to pemrametostat growth at 1μM.

H. Relationship between MTAP mRNA expression in transcripts per million (TPM) and the fraction of the MTAP transcription start site (TSS) that is methylated as detected in reduced representation bisulfite sequencing in CCLE.

I. Relationships between cell response to MRTX1719 in the PRISM screen (AUC) versus MTAP TSS methylation by cell line in CCLE.

For each model, we reported the cross-validated R-squared values and Pearson scores (the correlation between the model predictions and PRISM profiles) as the model performances. These performances described how accurate the model was.

For each feature of each model, the feature importances were computed after normalizing (the sum of the importances was set to 1 in each model) and tabulated along with the accuracy measures.

Code for PRISM analysis is at the github link: https://github.com/cmap/dockerized_mts.

Correlation of PRISM profiles

PRISM viability measurements in cell lines verified to be identical to Cancer Cell Line Encyclopedia lines using SNP fingerprint analysis were used as profiles to query previously reported genome-wide features (gene expression and copy number) of cell lines in the Cancer Cell Line Encyclopedia (http://www.broadinstitute.org/ccle ). Spearman's rank correlation was computed using the PRISM AUC measurement from each compound versus either gene expression or copy number, and the significance of correlation was calculated using permutation testing with 106 iterations. Moreover, linear models were fitted to test the association between the primary collapsed log fold change profiles of each drug and the DepMap Avana CRISPR knockout gene effect scores using the lmFit function from the limma R package (version 3.38.3) with default parameters [52]. P values were corrected within each dose profile for multiple hypotheses using the Benjamini–Hochberg method.

Digenic paralogous CRISPR-Cas9 sensitizer screen

Cas9-expressing cell lines were selected with blasticidin and transduced with the Digenic Paralog CRISPR library virus at low MOI (LU99=0.2, SW1573=0.27) to achieve 750–1,000 cells per sgRNA. LU99 was tested in duplicate; SW1573 in a single replicate. Cells (3×10⁶/well) were transduced in 12-well plates with polybrene (4–8 μg/mL), centrifuged at 2,000 rpm for 2 hours, and incubated for 18h. After three days of puromycin selection (2–3 μg/mL), cells were treated with DMSO, 10 nM MRTX9768 (LU99), or 20 nM MRTX1719 (SW1573). Cells were passaged for 21–25 days and harvested for gDNA isolation. Cells were treated at concentrations determined to be ~growth inhibition 30–40 (GI_30–40) for each cell line in a 7-day cell titer glo assay. Throughout the screen, cells were split and replated to maintain the representation at 1000x. Cell counts were taken at each passage to monitor growth. gDNA isolation using the NucleoSpin Blood XL (Takara Bio) was performed according to the manufacturer’s instructions.

PCR was performed using Titanium Taq DNA polymerase with 10 μg gDNA or 1 ng pDNA, P5/P7 primers, and master mix, followed by purification and sequencing on an Illumina NextSeq500 with 75 bp × 2 paired-end reads. PCR of gDNA and pDNA was performed in several 100 μl reactions (total volume) containing a maximum of 10 μg gDNA or 1 ng pDNA. DNA was PCR-amplified and barcoded with P5/ P7 primers using Titanium Taq DNA polymerase (Takara Bio) according to the manufacturer’s instructions. Briefly, per one reaction, a PCR master mix consisted of 1.5 μl of 50× Titanium Taq polymerase, 10 μl of 10× Titanium Taq reaction buffer, 8 μl of deoxynucleoside triphosphate, 5μL DMSO, 0.5 μl of P5 primer mix [100 μM] and 15.0 μl of water. Each well consisted of 50 μl of gDNA or pDNA plus water, 40 μl of PCR master mix and 10 μl of a uniquely barcoded P7 primer [5 μM]. PCR cycling conditions were: an initial 5 min at 95 °C followed by 30 s at 95 °C, 30 s at 55 °C, 20 s at 72 °C for 22 cycles and a final 10 min extension at 72 °C. Samples were purified with Agencourt AMPure XP SPRI beads (A63880; Beckman Coulter) according to the manufacturer’s instructions. On an Illumina NextSeq 500, 75 base pair (bp) × 2 paired-end sequencing was performed for all samples.

The raw/inferred log fold change (LFC) of each single gene and paralog is calculated using Chronos Python package[53]. In short, log fold change is calculated by taking the base-2 log of the fold change between the read counts (reads per million (RPM) + 1) and pDNA counts. The LFC difference between MTAC-PRMT5i (MRTX) and DMSO of the combined analysis of LU99 and SW1573 is ranked in Figure 2. Raw and calculated LFCs for each gene pair from each independent screen can be found in Supplemental Tables 2 and 3. The 20 paralogs with the most negative LFC and 5 paralogs with the most positive LFC are considered with a subset of them labeled. Single genes in each paralog pair are labeled to demonstrate potential true paralog effects. To further evaluate whether the paralog effects are true effects of the gene pair rather than driven by single gene effects, the more negative LFC out of the two single genes that makes up the paralog is compared with the LFC of the paralog. Paralogs are evaluated to have true paralog effects if the paralog LFC is smaller than −0.5 and the difference between paralog LFC and the minimum of single gene LFC is less than smaller than −0.2. Heatmaps showing the LFC difference of selected hits of single genes and paralogs are plotted. For some selected paralog hits, boxplots are shown comparing the LFC of screens in different drug conditions to show the synergy effect of paralogs (Figure 2).

Figure 2. Combinatorial CRISPR screening identifies potential sensitizers of MTA-cooperative PRMT5 inhibitors.

A. Schematic of drug-CRISPR combinatorial screening platform in MTAP and CDKN2A deleted non-small cell lung cancer cell lines.

B. Cell viability measured by CellTiter-Glo (CTG) is shown for a 12-days of treatment with MRTX9768 at day 1 and day 6 for 3 different cell lines. NCI-H1648 and LU99 are CDKN2A/MTAP co-deleted; CALU1 is MTAP proficient.

C. Single gene knockout results from the screen performed in (A) LU99 cells and SW1573 cells as a combined analysis of common hits from the two screens. Log fold changes between the PRMT5i drug (MRTX) arm vs DMSO arm: Red, significantly depleted in the screen. Green, significantly enriched in the screen.

D. Data from the screens in A, and C are analyzed for strongest paralogous/gene pair events and shown as waterfall plots. Points above the x-axis are genes that are enriched for KO in the screen. Points below the x-axis represent genes that are depleted for KO in the screen in the PRMT5 inhibitor treated cells compared to DMSO.

E. Scatterplot of each paralog gene pair comparing the log-fold change (LFC) between MRTX1719 and DMSO arms for each single knockout (SKO) or double knockout (DKO). Red dots represent gene pairs with statistically significant differences in the DKO than min SKO value of the pair.

F. Selected significantly enriched or significantly depleted paralogous gene pairs (left) and significantly enriched or significantly depleted single genes (right) in the MRTX/DMSO are shown as heatmaps for each cell line screened (LU99 and SW1573).

G. Box plots showing normalized sequence counts [LFC to pDNA] of each paralogous RAF gene pair, top MAPK1 gene pairs and top BCL2 gene pairs (6 sgRNAs for single genes and 18 sgRNAs for paralogous pairs) from the digenic CRISPR screen SW1573 in the DMSO and MRTX1719 arm of the assay shown in Figure 2. Zero represents baseline growth of the cells; a score of -1 is considered lethal.

For comparison of screen performance, the growth effects for loss of individual genes from the digenic screens performed herein are compared to the Chronos scores from prior DepMap single gene KO screens. For comparison of common CRISPR scoring methods, we also scored the two digenic CRISPR screens of each individual cell line screen (LU99, SW1573) using MAGeCK[54] to determine LFCs with FDR <0.05 labeled as significant.

Drug treatments

All drugs were dissolved in DMSO to 10 mM working stocks. The following compounds were acquired from MedChemExpress: MRTX1719 (HY-139611), MRTX9768 (HY-138684), TNG908 (HY-148419), Pemrametostat (HY-101563), RMC7977 (HY-156498), Ulixertinib (HY-15816), Venetoclax (ABT-199; HY-15531), Adagrasib (HY-130149), Trametinib (HY-10999), Belvarafenib (HY-109080), and MRTX1133 (HY-134813). Navitoclax (ABT-263) was acquired from Selleckchem (Cat No. S1001).

Cell viability and drug synergy scoring

Cells were seeded in 96-well plates at a density specified in the figure legends in 100 μl media. Subsequently, test compounds were added (all concentrations in technical triplicate) and DMSO content was normalized over all wells (DMSO content never exceeded 1 %). Drug dispensing was performed utilizing a Direct Digital Dispenser D300e (Tecan). Cells were treated for 7 days and cell viability assessed by Cell-tier Glo^™ luminescent viability assay (Promega Cat. No. G7571) and luminescence measured with a BioTek Cytation 7 cell imaging multi-mode reader (BioTek). Mean values +/− standard deviations (SD) were determined, and IC50 values calculated from dose response curves by GraphPad Prism (version 10, log(inhibitor) versus normalized response, variable slope).

For drug synergy scores, concentrations of MRTX1719 were combined with concentrations of the drug of interest and cell viability was determined using CTG. Cell viability was normalized to the DMSO control and analyzed via SynergyFinder[55] using the four-parametric logistic regression curve-fitting algorithm and synergy scores calculated with Loewe model. Scores less than −10 describe antagonistic interactions of the tested drugs, scores between −10 and 10 suggest additivity of the tested drugs, and scores larger than 10 suggest synergy between the tested drugs.

Colony formation assay

Cells were seeded in 12-well plates at a density specified in the figure legends in 1 ml media. Tested compounds were added with DMSO content normalized over all wells (DMSO content never exceeded 1%). Drug dispensing was performed using a Direct Digital Dispenser D300e (Tecan). Cells were treated for 14 days and drug(s) were redosed at day 7 before the cells were stained at day 14. For the staining, cells were washed with PBS and fixed in 10% formalin (Sigma-Aldrich) for 20 minutes. After washing with ddH20 0.5% crystal violet solution (10% methanol) was added to the cells for 20 minutes followed by washing twice with water. Wells were air-dried and imaged with an Epson Perfection V550 Scanner (Epson). For quantification, 10 % acetic acid (Thermo Fisher Scientific) was added to each well for de-staining and plates were left shaking on a Rocking Shaker for 20 min. Absorbance of the extracted crystal violet was measured at 590 nm for each well in triplicate in a 96-well plate with a BioTek Epoch 2 microplate reader (Agilent Technologies).

Western blotting

Cells were lysed in RIPA buffer supplemented with 1x protease inhibitor as well as 1x phosphatase inhibitor cocktail and incubated on ice for at least 15 minutes, sonicated, and centrifuged. Protein concentration was measured (Pierce BCA), and lysates were denatured by heating at 95°C. Proteins were separated by SDS-PAGE and transferred to nitrocellulose membrane. Membranes were blocked with Li-Cor PBS blocking buffer. After incubation with primary antibody overnight at 4 °C, secondary antibody (IRDye^® 680RD Goat anti-Mouse or IRDye^® 800CW Goat anti-Rabbit IgG Secondary Antibody) were incubated for 30–60 min at room temperature and detection was performed via Odyssey chemiluminescence. Bands were quantified using Fiji or Licor Software, with Vinculin as loading controls.

Antibodies used for experiments are as follows: IRDye 680RD Goat anti-Mouse IgG Secondary Antibody LI-COR Biosciences (926–68070) RRID:AB_10956588; IRDye 800CW Goat anti-Rabbit IgG Secondary Antibody LI-COR Biosciences (926–32211) RRID:AB_621843; Vinculin (Sigma Aldrich V9131–100UL) RRID:AB_477629; and the following antibodies from Cell Signaling Technology: Phospho-FRA1 (Ser 265) (5841S) RRID:AB_10835210, SDMA (13222S) RRID:AB_2714013, P21/Waf1/Cip1 (2947T) RRID:AB_823586, Phospho-Rb (8516S) RRID:AB_11178658, Total RAS (67648) RRID:AB_2910195, B-actin (3700) RRID:AB_2242334, GAPDH (2118) RRID:AB_561053, MYC (5605) RRID:AB_1903938, Phospho-ERK1/2 (9101S) RRID:AB_331646, and Total ERK1/2 (4696S) RRID:AB_390780.

RNA isolation, reverse transcription, and qPCR

For RNA isolation, the RNeasy Plus mini kit was used following manufacturer’s instructions. RNA quantity and quality were determined via NanoDrop One. 1–2 μg of RNA were further used for reverse transcription and cDNA synthesis performed via RevertAid First Strand cDNA synthesis Kit utilizing the oligo(dT) primers. Generated cDNA was used for qPCR utilizing Fast SYBR Green Master Mix following manufacturer’s instructions on a QuantStudio 3 Real-Time PCR System (Thermo Fisher Scientific). PCR reactions were performed in triplicates at 95°C for 20 s, followed by 40 cycles at 95°C for 1 s and 60°c for 20 s with one melt curve cycle. Differences in gene expression were calculated using the ΔΔC(T) method where GAPDH functioned as housekeeping gene to which the relative mRNA amounts were normalized to. Primers listed below were used for the experiments with a final concentration of 500 nM in each reaction. Statistical analysis was done in GraphPad PRISM (RRID: SCR_002798), where values were normalized to the DMSO control condition as expressing genes at 100%, and these normalized values were assessed by an Ordinary One-way ANOVA.

Primers used for qPCR analysis:

GAPDH

Forward: GGAGCGAGATCCCTCCAAAAT

Reverse: GGCTGTTGTCATACTTCTCATGG

CDKN1A_1

Forward: TGCCGAAGTCAGTTCCTTGT

Reverse: GTTCTGACATGGCGCCTCC

CDKN1A_2

Forward: AGTCAGTTCCTTGTGGAGCC

Reverse: GACATGGCGCCTCCTCTG

ANXA10

Forward: GCAATTCATGACTTTGGTTT

Reverse: TTTTCCATATCGCTCTTTGT

RELN

Forward: CATGGTTGCAAGTGTGACCC

Reverse: AAACCAGGGCCTTACCACTG

INHBA

Forward: CATTGCTCCCTCTGGCTATCAT

Reverse: GCACACAGCACGATTTGAGGTT

PTGS2

Forward: AGGAGGTCTTTGGTCTGGTG

Reverse: ACTGCTCATCACCCCATTCA

ITGA2

Forward: GGGAATCAGTATTACACAACGGG

Reverse: CCACAACATCTATGAGGGAAGGG

c-MYC

Forward: CCAGTAGCGACTCTGAGGAAG

Reverse: TGTGAGGAGGTTTGCTGTGG

FOSL1

Forward: ACCTACCCTCAGTACAGCCC

Reverse: TGCAGCCCAGATTTCTCATCT

ETV4

Forward: GGACTTCGCCTACGACTCAG

Reverse: CGCAGAGGTTTCTCATAGCC

ETV5

Forward: GTGTTGTGCCTGAGAGACTGGA

Reverse: CGACCTGTCCAGGCAATGAAGT

RNA seq analysis

RNA was isolated from A172 cells treated with 20nM MRTX1719 or DMSO control in 3x replicates for 5 days. RNA was isolated using the Qiagen RNeasy kit, according to the manufacturer’s instructions. cDNA library preparation and RNA sequencing were performed by NovoGene using the ABclonal mRNA-seq Lib Prep Kit for Illumina (poly A selection) to prep the libraries and the NovaSeq 6000 for sequencing. Illumina paired-end sequencing (150bp reads) was performed. Reads were aligned to the Ensembl homo sapiens grch38 using hisat2, and mapped reads were quantified using feature Counts (v2.0.3). Generated counts were imported into R (v4.3.2) for downstream analysis. Genes with less than 10 summation counts from all samples were removed. Differential gene expression analysis was performed using DESeq2 (v1.42.0). Genes with log₂-transformed foldchange of >1, or < −1 (respectively for up−, and down-regulation) and adjusted P-value <0.1 were considered as significant differentially expressed genes. The labeled top differentially expressed genes on the volcano plot were set at a more stringent log₂-transformed foldchange of > 2, or < −1.5 (respectively for top up−, and down-regulation) and adjusted P-value set to < 0.05.

RNA was isolated from LU99 cells treated with DMSO, PRMT5i (MRTX1719), RAFi (Belvarafenib), ERKi (Ulixertinib), or a combination of PRMT5i + RAFi or PRMT5i + ERKi in triplicate at doses described in the figure legends. RNA was isolated using the Qiagen RNeasy kit, according to the manufacturer’s instructions. cDNA library preparation and RNA sequencing were performed by NovoGene using the ABclonal mRNA-seq Lib Prep Kit for Illumina (poly A selection) to prep the libraries and the NovaSeq 6000 for sequencing. Illumina paired-end sequencing (150 bp reads) was conducted, and reads were aligned to the human genome (GRCh38.p13) using STAR (v2.5.2b)[56]. Raw counts were imported into R for downstream analysis. Genes with fewer than 10 counts across all samples were filtered out. Differential gene expression analysis was performed using DESeq2 (v1.34.0)[57]. Shrinkage of fold changes for lowly expressed genes was applied using the ashr package[58]. In addition, P values were adjusted for multiple testing using the Benjamini-Hochberg (BH) method and weighted by base mean expression using the IHW package[57]. Significant genes were retrieved using a threshold of 0.05 for the BH-adjusted p-value and a cutoff of 1 for the absolute value of the log₂FC. Gene set enrichment analysis (GSEA) was performed using ranked log2 fold change values and Hallmark gene sets available in the Molecular Signatures Database (MSigDB) using fGSEA (v1.20.0) and msigdbr (v7.5.1) R packages.

In Vivo Mouse Studies

Mice were treated and maintained on Dana-Farber Cancer Institute approval of IACUC protocol 08–023. Five million LU99 cells were injected subcutaneously into the right flank of 7-week-old female NCr nude mice (Taconic # NCRNU-F). Tumor volume was measured twice weekly, along with body weight. Mice were randomized into treatment groups when the average tumor volume reached ~180 mm³ (10-days post-injection) and treated with daily oral gavage of belvarafenib (30 mg/kg), MRTX1719 (50 mg/kg), adagrasib (50 mg/kg), drug combinations, or vehicle controls for 32 days, followed by 11 days (adagrasib) or 16 days (belvarafenib) of drug withdrawal. Drug doses used were the maximum tolerated dose combination of previously efficacious single agent treatments [59], [60], [61] that were shown to be well-tolerated and efficacious as single agent murine treatments. Prior to initiating the efficacy study, a 10-day toxicity body weight assessment was completed at the maximally efficacious doses for each drug individually and in combination, following the dosing schedule of the planned efficacy study. Body weight measurements for the toxicity study are provided in Supplemental Table 4.

Mice were euthanized at study endpoints or if tumor size exceeded 2.0 cm or body weight loss exceeded 15%. Drug doses were maximum tolerated, determined in prior studies, with a 10-day toxicity assessment before the efficacy study. Formulations included: Belvarafenib in 5% DMSO, 5% Cremophor EL, 0.5% methylcellulose; adagrasib in 10% captisol and 50 mmol/L citrate buffer (pH 5.0); MRTX1719 in 0.5% Hydroxypropyl methylcellulose and 0.2% Tween 80. Vehicle in the adagrasib study is the combination of the MRTX1719 vehicle and the adagrasib vehicle. N=8 mice per group were used in the study. The two studies were run concurrently and the MRTX1719 arm is common between the adagrasib and belvarafenib studies (MRTX1719 alone arm in the two figures are from the same 8 mice), all other arms are unique to the two figures. Progression free survival (PFS) was defined as tumor volume doubling from baseline and was represented using Kaplan–Meier plots. The log-rank test was employed for pairwise comparisons.

Generation of MTAC-PRMT5i-resistant and KRAS inhibitor-resistant cell lines

To generate MTAC-PRMT5i resistant cells, MiaPaca2 RAS active, CDKN2A/MTAP null tumor cells expressing Cas9 were continually treated with 10nM (IC₅₀ level) of the PRMT5 inhibitor MRTX1719 at 7 days for 10 weeks. An equal number of cells were seeded and maintained with the same volume of DMSO (DMSO content < 1%) in parallel as a control. When cell numbers were too low in the MRTX1719-treated cells, all remaining cells were replated. Cells were split and reseeded at 4 million cells in 15 cm plates with a final volume of 25mL of media twice a week. After 10 weeks, the dose of MRTX1719 was increased to 15nM for an additional 10 weeks. In total, cells were treated for 5 months before resistance was achieved, which was validated by >75-fold decreased sensitivity to MRTX1719 treatment by CTG.

To generate KRASi-resistant cell lines, SNU410 RAS-mutant, CDKN2A/MTAP null tumor cell lines were infected with ectopic wildtype KRAS or c-MYC expressing lentiviruses in the pLX314 backbone. Infected cell lines were selected with hygromycin for 7 days to generate stably expressing cell lines. Expression was confirmed by western blot and resistance was confirmed by cell growth despite KRASi treatment and KRASi IC₅₀ determination by CTG.

Network analysis of PRMT5 methyl substrates and ERK1/2-dependent phosphorylation substrates

PRMT5-dependent methylation targets were compiled from eight independent proteomics studies including our previous study[23], [24], [25], [28], [29], [31], [35], [62], [63]. Proteins reported as PRMT5 substrates in at least one study were included, and the number of supporting studies was recorded for each. ERK-dependent phospho-substrates were obtained from Klomp et al., 2024[64] and filtered using adjusted p-value < 0.01 and log₂ fold change < –0.25. A network was constructed using the igraph (v1.4.2) and ggraph (v2.1.0) R packages. Nodes represent protein-coding genes, and edges represent substrate-enzyme interactions (methylation by PRMT5 or phosphorylation by ERK1/2). Genes targeted by both enzymes were colored on a blue gradient based on the number of PRMT5 studies supporting methylation. Gene essentiality was determined using the "CRISPRInferredCommonEssentials" dataset from DepMap[15]. Node size reflects essentiality status, and labels were applied only to pan-essential genes that are dual substrates of PRMT5 and ERK.

RNA-seq analysis of published datasets

Raw counts of MiaPaca2 cells treated with MTAC-PRMT5i (AM9747 [39], 150nM for 6 days) or KRAS inhibitor (KRASi BI-2865 [65], 5 μM for 2 hours) were downloaded from GEO datasets GSE273376 and GSE228010, respectively. Genes with less than 10 summation counts from all samples were removed. Differential gene expression analysis was performed using DESeq2 (v1.34.0) [57]. Significant genes were retrieved using a threshold of 0.1 for the BH-adjusted p-value and a cutoff of 1 for the absolute value of the log2FC. Gene set enrichment analysis (GSEA) was performed using ranked log2 fold change values and Hallmark gene sets available in the Molecular Signatures Database (MSigDB) using fGSEA (v1.20.0) and msigdbr (v7.5.1) [66], [67] R packages.

Tumor sample datasets

Data on RAS/MAP kinase gene alterations in tumor samples were obtained from the Tumor Cancer Genome Atlas (TCGA) across 32 cancer types using the cBioPortal database (http://www.cbioportal.org ) [10]^. PanCancer Atlas TCGA samples with available copy number alteration and mutation data from the following cancer types were included: ACC (adrenocortical cancer), BLCA (bladder cancer), BRCA (breast cancer), CESC (cervical cancer), CHOL (cholangiocarcinoma), COADREAD (colorectal cancer), DLBC (large B-cell lymphoma), ESCA (esophageal cancer), GBM (glioblastoma), HNSC (head and neck cancer), KICH (kidney chromophobe), KIRC (kidney clear cell carcinoma), KIRP (kidney papillary cell carcinoma), LAML (acute myeloid leukemia), LGG (lower grade glioma), LIHC (liver cancer), LUAD (lung adenocarcinoma), LUSC (lung squamous cell carcinoma), MESO (mesothelioma), OV (ovarian cancer), PAAD (pancreatic cancer), PCPG (pheochromocytoma & paraganglioma), PRAD (prostate cancer), SARC (sarcoma), SKCM (melanoma), STAD (stomach cancer), TGCT (testicular cancer), THCA (thyroid cancer), THYM (thymoma), UCEC (uterine corpus endometrial carcinoma), UCS (uterine carcinosarcoma), UVM (ocular melanoma).GISTIC 2.0 [68] profiles were used to identify copy number alterations, and deep deletions or amplifications were recorded when the absolute GISTIC score was 2. All mutations were considered except for silent mutations. Co-occurring alterations between RAS/MAP kinase gene alterations (amplification, deep deletion, or mutation) and MTAP deep deletion were recorded for all tumor studies.

Statistical Analysis

Data are presented as mean ± SEM. Statistical analyses were performed using the GraphPad Prism (version 10) software. One-way ANOVA was used to determine statistical significance in colony formation assays and xenograft experiments, and the data are visualized using the listed symbols: ns: not significant, p-value >0.05, * p-value: 0.01–0.05, ** p-value: 0.001–0.01, *** p-value: >0.001, **** p-value: <0.0001.

Data availability statement

Publicly available data generated by others were used by the authors in this study; Mutation and copy number alteration data were obtained from TCGA at https://www.cbioportal.org/ . RNA sequencing data generated in this study are publicly available on GEO with accession numbers GSE282794 and GSE282795. All other raw data generated in this study are available upon request from the corresponding author.

Results

First generation PRMT5i suffered from toxicity due to lack of selectivity for tumor over normal tissue[32]. These first-generation inhibitors bound to the PRMT5 active site in a SAM-cooperative, or SAM-competitive manner. We previously advocated for a new class of PRMT5 inhibitors that would cooperate with the increased cellular levels of MTA to take advantage of the metabolic dysregulation enacted by deletion of MTAP[6]. Recently, several MTAC-PRMT5i have been developed and show promise in early pre-clinical and phase I/II clinical studies[35], [39]. To examine the selectivity of putative MTAC-PRMT5i in an unbiased manner we compared the cellular activity to the first generation PRMT5 inhibitors, using a Profiling Relative Inhibition Simultaneously in Mixtures (PRISM) screen (Figure 1)[49], [50]. Using pooled, barcoded cell lines .0+including 789 MTAP wildtype cell lines and 76 MTAP null cell lines, we profiled two structurally distinct Type I PRMT5 inhibitors, JNJ-64619178 and pemrametostat (GSK3326595), and identified sensitive cells either by fold-change in sensitivity by AUC (pemrametostat) from current PRISM screens and in prior drug repurposing screens (JNJ64619178 in Corsello, et al.; pemrametostat (GSK3326595) data in Table S1)[50]. In each case, broad, pan-cell line activity is observed with distributions that are similar to other pan-lethal therapeutics such as paclitaxel or adavosertib – a Wee1 inhibitor (Figure 1A). Similarly, we profiled the activity of three available MTA-cooperative PRMT5 inhibitors (MTAC-PRMT5i’s), MRTX1719, MRTX9768 and TNG908 and observed a distinct skewed distribution though with different levels of selectivity amongst the three (Figure 1A). These skewed distributions are similar to those observed with the BRAF inhibitor encorafenib[69] or the selective allosteric ABL inhibitor asciminib[70]. The distributions of the SAM-cooperative Type I PTMT5 inhibitors correspond well to the pan-lethal effects of CRISPR-mediated knockout of PRTM5 in the 1,104 cell lines screened in the DepMap project, while the MTA-cooperative inhibitors mirror the distribution of the shRNA mediated knockdown of PRMT5 which, we and others previously showed was synthetic lethal to the loss of MTAP[6], [7], [8], [9], [31](Fig. 1A bottom panels). Directly comparing the cell line specific log fold change at 1μM of two structurally related MTAC-PRMT5i’s MRTX9768 and MRTX1719 we observe a very high correlation of PRISM cell line sensitivity with a marked enrichment of cells lacking MTAP expression (Figure 1B, blue dots). In the same comparison between MRTX9768 and pemrametostat divergent cell activity between these mechanistically distinct PRMT5 inhibitors is observed (Figure 1C). In keeping with these data, the MTAC-PRMT5i MRTX1719 and MRTX9768 show significantly lower mean IC₅₀ and AUC values in the MTAP null setting compared to MTAP wildtype cells (Figure S1A–B) where MTAP null is defined as cell lines having low expression and low copy number (Fig. S1C). Next, we sought to identify the top shRNA genetic perturbations correlating with inhibitor sensitivity by correlating drug AUCs to the shRNA aggregated data in the Dependency Map[15]. The top shRNA correlate of MRTX1719 was PRMT5 (Fig.1D) while the top correlated gene knockdown for pemrametostat was GPR133 (Fig. 1E). We next examined the top gene expression correlates from the Cancer Cell Line Encyclopedia (CCLE) data[71], [72] to the same drug response data. Here, CDKN2A/B and MTAP were the most significant correlates of MRTX1719 AUCs, in this case with low expression correlating with lower- more sensitive - AUCs (Fig. 1F). On the other hand, the activity of the first generation PRMT5i pemrametostat (lower AUC values) was correlated with low expression ABCB1, a known drug transporter, and had no significant correlation with MTAP or CDKN2A or CDKN2B expression (Figure 1G). Relating inhibitor AUCs with copy-number alterations showed that loss of MTAP, CDKN2A and CDKN2B were highly correlated with cell line AUC values for MRTX9768, MRTX1719 and TNG908 (Figure S1D–F). Thus, the MTAC-PRMT5i’s recapitulate and correlate well with the PRMT5 shRNA data where the synthetic lethal relationship was originally discovered[6], [8], [9].

Interestingly, among our defined set of “MTAP-null” cells we observe cell lines where low gene expression of MTAP is not associated with gene deletion. In the cell lines for which we have paired promoter bisulfite sequencing methylation data, all of these cell lines showed methylation of the upstream promoter (Figure 1H) and were sensitive to MRTX1719 (Fig. 1I). Notably 5 of the 6 cell lines were derived from lymphoid malignancies. These data suggest loss of expression of MTAP through means other than gene deletion can render cells sensitive to MTAC-PRMT5i’s.

The in vitro activity of the MTAC-PRMT5i’s mirror the findings from the original genetic perturbation data and in early clinical trials tumor responses have been observed in patients with MTAP-deleted cancers with PR rates ranging from 21% to 33%, while at the same time sparing the hematologic toxicities seen with type I PRMT5 inhibitors[73], [74], [75]. While effective, single-agent MTAC-PRMT5i’s will almost certainly lead to mechanisms of secondary resistance[76], [77], [78], [79], [80] in responding patients in addition to the primary resistance that is already observable. Hence discovering combination therapies that will improve primary response rates, response duration, and can overcome emergent resistance will be required. To determine potential combination partners with MTAC-PRMT5i, we performed so-called “anchor” screens in which a given therapeutic is combined with CRISPR-Cas9 genetic screens. Herein, we chose to work with the MTAC-PRMT5i MRTX9768 and the clinical counterpart MRTX1719 as these showed a high degree of selectivity in PRISM screens (Figure 1, S1) and because early clinical data reported 33% of patients responded to this MRTX1719 drug in advanced, metastatic patient cohorts[35] while MRTX9786 is a non-clinical tool compound. Since initial trials have emphasized NSCLC patients, we started with this indication. Specifically, using the NSCLC cell lines LU99 and SW1573 both harboring CDKN2A/MTAP co-deletion and KRAS^G12C mutations[10], [81], we introduced a digenic CRISPR-Cas9 sgRNA library (Paralog v2.0) that targets genes and gene pairs encoding likely ‘druggable’ proteins and their nearest related paralogs and/or family members (Figures 2A–B) into these two NSCLC cell lines. To identify an optimal screening drug concentration, we performed 7 and 12-day viability CellTiter-Glo (CTG) assays to ascertain dose ranges where cell growth was inhibited 30–40% compared to DMSO control (Figure 2B – shown are 12-day assays; Figure S1G–H). The Paralog v2.0 CRISPR sgRNA library uses a single Cas9 and alternative tracers for sgRNA pairs and targets 4,547 single genes and 7,644 paralogous or homologous gene-pairs. SW1573 and LU99 CRISPR screens were conducted at 1,000X representation, and treatment arms showed robust depletion of PRMT5 SDMA activity (Figure S2A, D), an appropriate reduction in population doublings was observed in the drug treated arms (Figure S2B, E), and good separation between pan-essential and non-essential genes (Figure S2C, F) in comparison to Cancer Dependency screening[15]. Screens were scored as previously reported[82] using log fold-change drug over control and using MaGeCk[54] a common CRISPR scoring pipeline, with hits being consistent between these scoring methods (Figure 2C–D, S3A, Tables S2–3).

Our drug combinatorial CRISPR-Cas9 screen utilized a digenic, paralog sgRNA library to not only identify digenic vulnerabilities in addition to single gene hits as genetic sensitizers to MTAC-PRMT5i, but also to overcome limitations of single-gene perturbation screens due to gene redundancies[83]. The top paralogous pairs detected were RAF paralogous gene pairs and the pro-apoptotic gene pairs BCL2 and BCL2L1, and BCL2L1 and BCL2L2 (Figure 2D–F). The top single gene sensitizers of MTAC-PRMT5i included several key members of the RAS-MAPK signaling pathway: KRAS, RAF1 (c-RAF), and MAPK1 (ERK2) (Figure 2C–F). We also identified MAT2A, a regulator of the SAM biogenesis pathway, and previously identified as also synthetic lethal with MTAP loss, as showing increased depletion in combination with the MTAC-PRMT5i’s (Figure S3A), in-line with its role in regulating cellular responses to MTAP loss[6]. Indeed, a clinical trial is ongoing testing the combination of MAT2A inhibitors and MTAC-PRMT5i (Clinical Trial: NCT05975073). We next sought to examine single gene and paralog event concordance across the two screens for top combined hits and found good agreement between the two screens for BCL2 and MAPK repressors (Figure 2F). Combinatorial effects are visualized for each of the RAF A/C and RAF B/C pairs, the MAPK1 pairs, SHOC2 pairs, or BCL2 family pairs (Figure 2G), indicative of combinatorial digenic screening revealing pairs of genes that cooperatively sensitize to the MTAC-PRMT5i. Taken together, the CRISPR screens identified known and novel single genes and paralogous pairs of genes that regulate responses to PRMT5 inhibition in CDKN2A/MTAP null cancer.

To pharmacologically recapitulate the KO of the BCL2 and BCL2L1 paralogs, we compared inhibitors of BCL2 alone (venetoclax)[84], [85] or an inhibitor of BCL2, BCL2L1 and BCL2L2 (navitoclax)[86]. Consistent with the screen data, we found that the combination of MRTX1719 with navitoclax more effectively inhibited LU99 growth compared to the combination MRTX1719 with venetoclax (Figure S4A–G). These results highlight how the use of a digenic paralog sgRNA library resulted in the identification of a promising therapeutic target that might have been missed in a single gene perturbation sgRNA library.

The opportunity to target two functionally distinct driver events in cancer (e.g. KRAS mutation and CDKN2A/MTAP deletion) is appealing due to the likely truncal nature of both events. Moreover, if therapeutics against such events have non-overlapping mechanisms of resistance, then combinations should significantly reduce potential escape mechanisms. Therefore, we chose to focus on the RAS-MAPK pathway sensitizers KRAS, RAF1/BRAF/ARAF, and MAPK1 for our studies. To validate the role of MAPK pathway loss or inhibition in sensitizing cells to MTAC-PRMT5i, we first assessed the growth effects of combinatorial treatment of MRTX1719 with the ERK1/2 inhibitor ulixertinib (in phase II testing NCT03417739[45]) on a set of 5 NSCLC or PDAC cancer lines which are MTAP-null and harbor mutations that activate the RAS-MAPK pathway. In addition to LU99 and SW1573 cells (both NSCLC harboring a KRAS^G12C mutation) in which the screens were performed, the NSCLC cell line NCI-H1666 (BRAF^G466V), and PDAC cell lines MiaPaCa2 (KRAS^G12C), KP4 (KRAS^G12D), and PANC02.13 (KRAS^Q61R) were included (Fig. S5). Ulixertinib treatment inhibited MAPK signaling at the tested drug concentrations as evidenced by suppression of the downstream target phospho-Fra1 (Figure S6A) and combination treatments with MRTX1719 across this dose range resulted in significant impairment of colony formation compared to single treatments (Figure 3A–B, S7A–E). In addition, in vitro combination dose-matrix experiments showed synergistic effects (Loewe scores) for the combinations for all MTAP-null, RAS-MAPK mutant tested cell lines except SW1573 where additive scores were derived (Figure 3C). In contrast, CALU1 MTAP WT cells did not demonstrate additive or synergistic effects in the MTAC-PRMT5i + ERK1/2i combination (Figure 3C). Together, pharmacological inhibition of ERK1/2 sensitized NSCLC and PDAC MTAP-null/RAS- and BRAF-mutant cancer cell lines to MTAC-PRMT5i’s. Similar results were obtained with combinations of the MEK inhibitor trametinib and MRTX1719 in phospho-FRA1 suppression, colony formation, and proliferation (Figure S6B, S8 A–C, S9A–E). To further extend these findings, we tested patient-derived MTAP-null and MTAP-WT KRAS^G12D mutant PDAC organoid models, PANFR0368 (PF368) and PANFR0402 (PF402) respectively (Figure S8D), wherein synergy was observed between trametinib and MRTX1719 in the MTAP-null setting. Taken together, the data suggest combining MTAC-PRMT5i such as MRTX1719 with MAPK-pathway inhibitors results in greater impairment of cancer cell growth compared to single treatments. Notably, these data demonstrate that our findings of the MTAC-PRMT5i and RAS-MAPKi extend to other genotypes and cancer types beyond KRAS^G12C mutant NSCLC and suggest these drug combinations will work broadly on CDKN2A/MTAP null tumors harboring a myriad of RAS-MAPK activating mutations.

Figure 3. Validation of ERK2 loss as a synergistic combination with MTA-cooperative PRMT5 inhibitors.

A. Colony formation assays in LU99 (2500 cells seeded per well), SW1573 (1500 cells seeded per well), MiaPaca2 (500 cells seeded per well), NCI-H1666 (5000 cells seeded per well), and KP4 (500 cells seeded per well) cells. Cells were treated with DMSO, MRTX1719, and/or ulixertinib alone and in combination for 14 days with redosing of the drugs at day 7. Colonies were stained with 0.5% crystal violet. Representative image of 3 independent experiments (N=3).

B. Quantification of colony formation assay normalized to DMSO control. Shown are mean +/− SD and statistical significance represented using the following symbols: * p-value: 0.01–0.05, ** p-value: 0.001–0.01, *** p-value: >0.001, **** p-value: <0.0001. No symbol indicates no statistical significance (p-value >0.05).

C. Cell viability assessed via CTG of drug combination treatments of MRTX1719 and ulixertinib. LU99 (2500 cells/well), SW1573 (500 cells/well), and MiaPaca2 (1000 cells/well), NCI-H1666 (1500 cells/well), KP4 (1000 cells/well), PANC02.13 (2500 cells/well) CALU1 (1000 cells/well) were seeded per well and treated for 7 days with a concentration series of MRTX1719 and ulixertinib alone or combination. Cell viability was normalized to DMSO control and drug synergy scores were determined using SynergyFinder[55]. Synergy score heat map displayed for Loewe model performed in biological replicates (N=3).

The development of MAPK pathway inhibitors across various mutational backgrounds has been the subject of longstanding efforts to abrogate this pathway with effective inhibitors. Unfortunately, MEK1/2 and ERK1/2 inhibitors suffer from significant dose-limiting toxicity in patients. Importantly, RAS-mutant selective inhibitors have exhibited better tolerability with activity in the KRAS mutant setting, and BRAF/CRAF inhibitors such as belvarafenib and naporafenib have exhibited a lack of paradoxical activation and thus exhibited activity in certain RAS mutant settings, in keeping with genetic studies in mouse models[87], [88], [89], [90]. Given that knockdown of KRAS, and RAF1 and various digenic combinations thereof were cooperative with MRTX1719 in the anchor screens, we next tested these therapeutic combinations.

For these studies we used adagrasib, a selective KRAS^G12C inhibitor, and belvarafenib, a pan RAF inhibitor to assess efficacy of the MRTX1719 combinations by evaluating the in vitro growth effects in three KRAS^G12C mutated cell lines: LU99, SW1573, and MiaPaCa2. In agreement with the in vitro results obtained with MEK and ERK inhibitors, the combination of MRTX1719 and adagrasib also resulted in reduced pFRA1 (Fig. S6C) and significantly reduced cancer cell growth in LU99 and SW1573 (Figure 4A–D, S10A–C), highlighted by a synergistic Loewe CTG growth score in the three tested cell lines (Figure 4A).

Figure 4. KRAS inhibition shows synergistic combinatorial growth suppression with MTA-cooperative PRMT5i.

A. Cell viability of lung and pancreatic cell lines assessed via CTG treated with MRTX1719 and adagrasib combinations. LU99 (2,500 cells/well), SW1573 (500 cells/well) and MiaPaca2 (1,000 cells/well) cells were seeded per well and treated for 7 days over a concentration range of MRTX1719 and adagrasib alone or combination. Cell viability was normalized to DMSO control and drug synergy scores determined using Synergyfinder. Synergy score heat map displayed for Loewe model. N=3 biological replicates were analyzed per cell line.

B. Cell viability of pancreatic ductal adenocarcinoma organoids of the indicated genotypes were assessed via CTG after combination treatments with MRTX1719 and KRASi RMC-7977. Organoids were seeded and treated for 7 days with a concentration range of MRTX1719 and adagrasib alone or combination. Cell viability was normalized to DMSO control and plotted. Synergy scores were determined using Synergyfinder. Synergy score heat map displayed for Loewe model. N=2–3 biological replicates were analyzed per cell line.

C. Colony formation assay in LU99 (2,500 cells seeded per well), SW1573 (500 cells seeded per well), and MiaPaca2 (500 cells seeded per well) cells. Cells were treated with DMSO, MRTX1719, and/or adagrasib alone or in combination for 14 days with redosing at day 7. Colonies were stained with 0.5% crystal violet. Representative image of 3 independent experiments (n=3).

D. Quantification of colony formation assay normalized to DMSO control. Shown are mean +/− SD and statistical significance represented using the following symbols: * p-value: 0.01–0.05, ** p-value: 0.001–0.01, *** p-value: >0.001, **** p-value: <0.0001. No symbol indicates no statistical significance (p-value >0.05).

E. Subcutaneous flank LU99 xenograft tumors treated with either vehicle, adagrasib (50mg/kg), MRTX1719 (50mg/kg), or the combination (both at 50mg/kg) for 32 days. The drug treatments were withdrawn (arrow) and time to recurrence monitored. Data shown are the mean and SEM of tumor volume over time.

F. Quantification of the in vivo experiment shown in (E). N=8 mice per group at end of drug administration and end of drug withdrawal/recurrence period. ANOVA tests were performed for each (1) the mean tumor volume per group and (2) status of non-detectable disease/complete responses.

G. Co-occurence analysis from TCGA patient data derived from cBioPortal[10] for the indicated RAS/MAP kinases gene alterations (amplification or mutation in the listed MAPK gene) in MTAP-null (MTAPdeepdel) cancer across the listed TCGA cohorts. LUSC+LUAD represent the two combined NSCLC studies. All TCGA Pan studies where 3% or more of the patients harbored MTAP deep deletion and at least 1 patient harbored co-modification of MAPK and MTAP deep deletion are shown in the plot.

H. Progression-free survival Kaplan Meier Curve as in (G) for adagrasib single agent and MRTX1719/adagrasib combination experiment shown in Figure 4E–F.

Unless otherwise specified, data displayed as average +/− SD and statistical significance represented using the following symbols: * p-value: 0.01–0.05, ** p-value: 0.001–0.01, *** p-value: >0.001, **** p-value: <0.0001. No symbol indicates no statistical significance (p-value >0.05).

We next tested the combination of adagrasib and MRTX1719 in vivo. We identified well-tolerated combination doses in tolerability studies (Table S4) beginning with 50mg/kg MRTX1719 and 50mg/kg adagrasib, the typical single agent doses, finding these doses were well tolerated (Table S4). We then compared this combination to the typical single agent doses of 50mg/kg MRTX1719 and 50mg/kg adagrasib in tumor growth assays finding that treatment with the adagrasib and MRTX1719 combination showed improved tumor regression compared to the single agents (Figure 4E, 4F left). Notably, after drug discontinuation, we observed significant regrowth of tumors treated with the single agents but not with the combination (4E, black arrow). In addition, 4/8 of the mice in the combination treatment achieved a complete response (disease below the limit of detection), compared to 0/8 mice in the two single agent arms, (Figure 4E, 4F right). To extend these findings to the KRAS^G12D setting, we tested the aforementioned KRAS^G12D pancreas patient-derived organoid models. Here, we observed greater sensitivity to the combination of MRTX1719 with the pan-KRAS inhibitor RMC7977[91] (NCT05379985) in the MTAP-null but not MTAP-WT setting (Figure 4B). These data suggest that the combination effects of KRAS inhibition with MTAC-PRMT5i’s extends to additional KRAS genotypes and to cancer types beyond KRAS^G12C mutant NSCLC. To explore the potential relevance of dual pathway inhibition in patients, we determined the frequency of co-occurring MTAP deletion and RAS-MAPK pathway alterations across cancer. Each MAPK pathway member alone either passed significance or trended towards significance for co-occurrence (log2 odds ratio) with MTAP loss in pan cancer tumor sequencing analyses with: BRAF q=0.297, KRAS q=0.072, EGFR q<0.001, NF1 q<0.001, and NF2 q<0.001 (Figure 4G), with the highest degree of co-occurrence seen in GBM, PDAC (PAAD-TCGA study), NSCLC (LUAD+LUSC TCGA studies), melanoma (SKCM), and mesothelioma (MESO) (Figure 4G). These data suggest combinations of the two inhibitor classes are relevant across a broad range of patient indications.

Pan-RAF inhibitors have also shown therapeutic promise in pre-clinical studies[89], [90]. However, despite mouse genetic data, there has been limited single agent activity in the KRAS setting, with a greater focus on the NRAS and BRAF mutant settings. Notably, treatment with direct KRAS inhibitors has given rise to KRAS amplification as a common means of resistance[92]. As KRAS levels rise, it will likely become increasingly difficult to inhibit KRAS directly due to the adverse kinetic equilibrium. Thus, there is a strong interest in finding downstream inhibitors that can act in this setting. Based on the anchor screen data highlighting the role of dual RAF isoform inhibition (Fig. 2D–G) we next tested the combination activity of MRTX1719 with the pan-RAF inhibitor belvarafenib. Combinations of belvarafenib and MRTX1719 showed significantly improved suppression of colony formation in the lung and pancreas double-mutant models (Figure 5A–B; S11A–E), with pan-RAF inhibitors effectively repressing phospho-Fra1 (Figure S6D). Similarly, synergy was observed in double-mutant lines for the combination treatments when tested in short-term proliferation assays (Figure 5C). Correspondingly, using PDAC organoid models, we found that PANFR0368 (PF368, MTAP^del) was more sensitive to belvarafenib/MRTX1719 combination therapy than the MTAP-WT organoid PANFR0402 (PF402) (Figure 5D).

Figure 5. RAF inhibition shows synergistic growth combination with MTA-cooperative PRMT5i in vitro and in vivo.

A. Colony formation assay of NSCLC and PDAC cell lines treated with DMSO, MRTX1719, and/or belvarafenib alone or in combination for 14 days with redosing at day 7 day. Colonies were stained with 0.5% crystal violet. Representative image of 3 independent experiments (N=3).

B. Quantification of colony formation in (A) normalized to DMSO control. Shown are mean +/− SD and statistical significance represented using the following symbols: * p-value: 0.01–0.05, ** p-value: 0.001–0.01, *** p-value: >0.001, **** p-value: <0.0001. No symbol indicates no statistical significance (p-value >0.05).

C. Cell viability of lung and pancreatic cell lines assessed by CTG of drug combination treatments of MRTX1719 and belvarafenib. LU99 (2,500 cells/well), SW1573 (500 cells/well) and NCI-H1666 (1,500 cells/well) cells were seeded per well and treated for 7 days over a concentration range of MRTX1719 and belvarafenib alone or combination. Cell viability was normalized to DMSO control and drug synergy scores determined using Synergyfinder using the Loewe scoring model. N=3 biological replicates were analyzed per cell line.

D. Cell viability of pancreatic ductal adenocarcinoma organoids assessed via CTG of drug combination treatments of MRTX1719 and belvarafenib. Organoids were seeded and treated for 7 days with a concentration range of MRTX1719 and belvarafenib alone or combination. Cell viability was normalized to DMSO control and synergy calculated using the Loewe scoring model.

E. Subcutaneous flank LU99 xenografts were treated with either vehicle control, belvarafenib (30mg/kg), MRTX1719 (50mg/kg), or the drug combination of belvarafenib (30mg/kg) and MRTX1719 (50mg/kg) for 32 days. The drugs were withdrawn (arrow) and time to recurrence monitored. Data shown are the mean and SEM of tumor volume over time.

F. Quantification of in vivo experiment shown in (E). N=8 mice per group at end of drug administration and end of drug withdrawal/recurrence period. ANOVA tests were performed for each (1) the mean tumor volume per group and (2) status of non-detectable disease/complete responses.

G. Progression-free survival calculated by mRECIST[91] definition (time to tumor volume doubling from initial treatment volume) for belvarafenib single agent and MRTX1719/belvarafenib combination experiment that is shown in (E).

H. Progression-free survival Kaplan Meier Curve as in (G) for adagrasib single agent and MRTX1719/adagrasib combination experiment shown in Figure D-E.

To test this combination of RAFi/MTAC-PRMT5i in vivo, we again established well-tolerated combination doses and tested the single agents and combinations in the LU99 model. While both single agents are active the combination showed a trend towards lower tumor volumes during the initial treatment period (Figure 5E–F left panel). Importantly, there was a significant combinatorial growth effect in the number of complete responses observed in vivo that persisted after drug withdrawal (Figure 5F, right panel). Taken together, these data demonstrate that PRMT5 and RAS-MAPK signaling inhibitors show strong combinatorial growth effects in tumors harboring MAPK activity and MTAP loss. Interestingly, we note that while adagrasib had a strong effect at early time points in ablating tumor growth (Figure 4E–F; Days 1–17) on its own and in combination with MRTX1719, the pan-Raf inhibitor (belvarafenib) achieved more complete responses as a single agent than adagrasib (3/8 versus 0/8) and as a combination treatment with MRTX1719 (5/8 versus 4/8 CRs), and appeared more durable in the LU99 xenograft model after drug withdrawal (Figure 4D–E, 5E–H). This combination effect was also reflected in CRISPR screen results where paralog KO of RAF members resulted in stronger sensitization compared to single RAF loss whereas sensitization to PRMT5i in RAS genes was driven by single gene KRAS loss. Specifically, 100% of the adagrasib-only cohort had progressive disease by day 39, which was significantly less than in the belvarafenib-only or combination arms (Figure 4E–F, 5E–H) and this trend is also seen in the progression-free survival (PFS) models using published methods[92] (Figure 5G–H). Additionally, while both combinations (MRTX1719-adagrasib or MRTX-belvarafenib) were highly effective, MRTX+belvarafenib had a modestly better PFS in the pre-clinical LU99 NSCLC model (Figure 5G–H). Overall, our CRISPR-drug combinatorial screening identifies MAPK inhibitors (as well as other genes and pathways) as potentially actionable drug combinations for the treatment of CDKN2A/MTAP null cancer and the data demonstrate that paralogous gene pair depletion screens can identify candidate therapeutic targets beyond those observed in single gene KO screens.

To evaluate how therapeutics targeting the MAPK and PRMT5 pathways might be interacting to improve tumor responses, we performed RNAseq of individual drugs and combinations with RAFi belvarafenib, MTAC-PRMT5i MRTX1719, and ERK1/2i Ulixertinib compared to control (DMSO). After 5 days of treatment, we found significant enrichment of the KRAS signaling gene set among altered pathways in response to MTAC-PRMT5i; however, this effect is abrogated in both MAPKi + MTAC-PRMT5i combination treatments (Figure 6A; S12A; S13A–B), suggesting KRAS signaling is activated in response to PRMT5 inhibition but does not overcome concomitant inhibition of the RAS-MAPK pathway. While our study and Drizyte-Miller et al. (co-submitted article in this issue) are the first studies to investigate RNA changes in the combinations of MTAC-PRMT5i and RAS/RAF/MAPK inhibitors, several studies have looked at the single agents alone. We compared our single agent RNAseq results with those of published data in LU99 cells treated with MRTX1719 as well as of MiaPaca2 cells treated with either KRASi (BI-2865) or MTAC-PRMT5i (AM9747)[35], [39], [65]. In both the RAS-MAPK active PDAC and NSCLC cell lines used in this study, we find by GSEA analysis that RAS signaling was a significantly enriched pathway in response to MTAC-PRMT5i and validated this observation in qPCR assays (Figure 6A–B, S14A–B). This is in agreement with published single agent treatments for MTAC-PRMT5i[35], [39]. Next, we tested whether this observation might extend to a CDKN2A/MTAP null EGFR amplified glioblastoma cell line, A172, and again determined in this cellular context that KRAS signaling is upregulated in response to MTAC PRMT5i (Figure S15A), suggesting that MAPK pathway activation in response to PRMT5i is independent of RAS-mutation per se and can be observed in distinct lineages. Interestingly, we note that while the KRAS gene signature is upregulated by MTAC-PRMT5i, some key reported RAS-responsive genes including FOSL1 (encoding Fra1) and MYC were decreased upon MRTX1719 treatment (Figure S14A–B). We postulate this may be due to the cell cycle repressive effects of MTAC-PRMT5i which may hamper cell cycle-dependent RAS effectors such as E2Fs. Indeed, we see alterations in p53 response, increased CDKN1A expression, decreased phospho-RB, and decreased G2/M checkpoints, indicating that PRMT5 inhibition is modulating cell cycle progression (Figure 6A, S15–17). Finally, to test whether MTAC-PRMT5i enhances RAS-mediated transcriptional outputs in the absence of an activating mutation in the RAS-MAPK pathway, we treated the CDKN2A/MTAP null cell line NCI-H1648 with MTAC-PRMT5i and looked at candidate RAS target genes determined to be upregulated in the RAS-active, MTAP null lung lineages (shown in Figures 6A–B; S14A–B); neither RAS-responsive gene ITGA2 nor PTGS2 were upregulated in the MAPK wildtype cell line (Figure S14C), suggesting that enhancement of RAS-mediated transcriptional outputs in response to PRMT5 inhibition requires a RAS active genotype.

Figure 6. Protein and RNA level crosstalk observed across KRAS and PRMT5 signaling and non-cross resistance achieved in cell lines resistant to KRAS or MTAC-PRMT5 inhibitors.

A. GSEA of RNAseq data from LU99 cells treated for 5 days with DMSO control, 20nM MRTX1719, 300nM Belvarafenib, or the combination of 20nM MRTX1719 and 300nM Belvarafenib. Shown are significantly enriched Hallmark gene sets in treated LU99 cells compared to DMSO control (adjusted p-value < 0.01). Color scale indicates the normalized enrichment score (NES), and the node size represents the adjusted p-value.

B. GSEA identifies significantly enriched Hallmarks gene sets in published RNAseq data of MTAC-PRMT5i treated (AM9747[39]) or KRAS inhibitor treated MiaPaca2 cells for 6 days at 150nM compared with MiaPaca2 cells treated with KRASi[65] BI-2865 for 2 hours at 5 μM). Color scale indicates the normalized enrichment score (NES), and the node size represents the adjusted p-value.

C. Network showing distinct and shared PRMT5 methyl-substrates identified in at least one of eight mass spectrometry studies (see methods) and ERK1/2 phospho-substrates[64]. Node size reflects pan-essentiality across CRISPR screens in the Cancer Dependency Map[15]. Proteins post-translationally modified by both PRMT5 and ERK are colored, with a gradient indicating the number of studies supporting PRMT5 methylation. PRMT5 and ERK share 122 common substrates of which 39 are pan essential in cancer. Shared pan-essential substrates are labeled. Fisher’s exact test p<10⁻³ for all values tested: ERK1/2 and PRMT5 substrate commonality; ERK1/2 substrate pan-essentiality in DepMap; and PRMT5 substrate pan-essentiality in DepMap.

D. The MTAP null, KRAS^G12D mutant PDAC cell line SNU410 was engineered to ectopically express GFP, KRAS or c-MYC.

E. Western blot detection of KRAS or GAPDH as a loading control, in SNU410 cell lysates cells following infections and selection with lentivirus encoding GFP control or KRAS.

F. Cell viability of SNU410 cells expressing exogenous KRAS or GFP measured via CTG after 5 days of treatment with a range of concentrations of MRTX1133. Cell viability is normalized to DMSO controls.

G. Cell viability of SNU410 cells expressing exogenous KRAS or GFP measured via CTG after 7 days of treatment with a range of concentrations of MRTX1719. Cell viability is normalized to DMSO controls.

H. Schematic overview how MRTX1719-resistant MiaPaca2 cells were generated

I. Cell viability of MiaPaca2 control or MRTX1719-resistant cells measured via CTG. Cells were treated for 7 days with a concentration series of MRTX1719, adagrasib, ulixertinib, or belvarafenib and cell viability was normalized to DMSO control.

Comparing the combinations versus DMSO to the single agents versus DMSO, we note that there are 1,430 genes differentially regulated by the RAFi+MTAC-PRMT5i drug combination. Most of these combination-altered genes fall into the same gene sets noted as being altered in one or both of the single agents, suggesting that there is a change in the signal rather than a de novo gene set enrichment in the combination treatments (Figure 6A-RAFi, S12A-ERK1/2i). However, several gene sets - including P53 signaling and inflammation (Figure 6A, S12A) - are significantly regulated by MTAC-PRMT5i alone but do not pass significance in the combination treatments, suggesting loss of some signals from the PRMT5i when combined with RAS-MAPK inhibitors could play a role in the combination effect in repressing tumor growth.

To investigate how PRMT5 inhibition might enhance RAS-MAPK dependent transcriptional outputs, we examined treatment effects on RAS effectors. We found that MRTX-1719 treatment leads to induction of phospho-Fra1 in RAS-active, MTAP null cells (Figure S18A–C) in agreement with data from Drizyte-Miller, et al who found induction of phospho-ERK1/2 response to PRMT5 inhibition. To determine how PRMT5 and RAS signaling may intersect at the protein level, we next evaluated the potential substrate crosstalk between the two pathways. We compiled our own and all available PRMT5 substrate enrichment proteomics[23], [24], [25], [28], [29], [31], [35], [62], [63] and compared it to a recent ERK1/2-dependent phosphoproteomics study (Figure 6C)[64]. We identified 486 high-confidence methyl- substrates of PRMT5 and 1,105 high-confidence phospho-events responsive to ERK1/2 inhibition (Figure 6C; Table S5), many of which are pan-cancer essential genes in DepMap (enrichment of essential genes Fisher’s exact test p-value<10⁻³)[15]. PRMT5 and RAS-MAPK pathways share over 100 substrates of which 39 are also pan-essential genes regulating critical processes including mRNA transcription, protein synthesis, and cell cycle, they also drive hundreds of distinct cell functions. From our RNAseq and western blot data (Figure 6A, S16–17), it is clear that inhibition of PRMT5 activates p53, causes G1/S cell cycle halt, and activates inflammatory/cellular stress pathways. Taken together, we believe activation of RAS signaling may be a cell stress response following MTAC-PRMT5i rather than due to a single node of pathway convergence.

While there is overlap in RAS and PRMT5 regulated substrates, many such substrates were divergent suggesting that pathway inhibitors might act in a non-cross resistant manner. To test this, we generated isogenic lines of the KRAS^G12D, MTAP null pancreatic cancer cell line SNU410 (characterized in Figure 6D–E; S19A) ectopically expressing WT KRAS or MYC. These alterations represent two resistance mutations frequently observed in patients with acquired KRAS inhibitor resistance (KRAS or MYC amplification)[92], [93], [94]; Consistent with these data, KRAS overexpression induced resistance to the KRAS^G12D inhibitor MRTX1133, shifting the IC₅₀ (Figure 6F), while MYC overexpression led to increased cell proliferation versus GFP overexpression at all doses of MRTX1133 (Figure S19B–C). Despite genetically different mechanisms of resistance to KRAS inhibition, both the KRAS and MYC ectopically expressing cell lines were largely unchanged in their sensitivity to MRTX1719, demonstrating likely non-cross-reactive mechanisms of resistance to these two classes of agents and highlighting the potential of MTAC-PRMT5 inhibition to be efficacious in patients with acquired KRAS inhibitor resistance.

To test the reciprocal relationship, we treated MiaPaca2 cells to resistance with MRTX1719 and assessed the response of these cells to a panel of MAPKi (Figure 6H–I). These lines remained sensitive to RAS pathway inhibitors including RAS, RAF, and ERK inhibitors (Figure 6I), suggesting likely distinct mechanisms of resistance. These data formally demonstrate that despite some evidence of pathway cross-talk (e.g., phospho-FRA1 increases following MTAC-PRMT5i in Fig. S18A–B) the inhibitors show non-cross resistance consistent with the improved in vivo responses seen with the combination. Taken together, these data suggest that PRMT5 and RAS/RAF can be targeted in combination and that the significant combination efficacy may arise due to the independent suppression of resistant clones.

Overall, we find that exposure to MTAC-PRMT5i drives activation of a RAS gene signature and that inhibition of the RAS pathway in combination with MTAC-PRMT5i leads to increased cancer killing in the MTAP-null, RAS-active tumor setting. We identified known and novel regulators of the PRMT5 pathway and demonstrate effective combinations of MTAC-PRMT5 inhibition with RAS and RAF inhibitors for the treatment of CDKN2A/MTAP null, RAS active tumors across a range of tumor types. Further, we demonstrate non-cross-reactive mechanisms of resistance to MAPKi and MTAC-PRMT5i, suggesting that the combinations can lead to more durable responses and that MTAC-PRMT5i may be a reasonable second-line therapy for RAS-active, CDKN2A/MTAP deleted cancers that have developed resistance to KRAS inhibitors.

Discussion

Targeted therapies often show initial responses, but as with most cancer therapeutics, tumors recur due to the development of resistance to single agent therapy. Thus, we sought to identify single gene or paralogous gene pairs that could act in combination with the new class of MTAC-PRMT5i. Herein, we report combining a novel digenic targeting CRISPR library with co-treatment with MRTX1719, one of a new class of PRMT5 inhibitors, to define potential combination treatments for clinical trials. We also uncover a number of genes that interact with PRMT5 signaling in cancer that were not previously known.

To develop highly effective drug combinations, we need to discover therapeutics that have non-overlapping mechanisms of action such that no single genetic (or epigenetic) alteration can confer resistance to the combination. For example, we previously showed that two different inhibitors of BCR-ABL bearing non-cross resistant mechanisms of resistance, despite targeting the same protein, could eradicate CML tumor grafts in mice[70]. On the other hand, the combination of BRAF and MEK inhibitors can be undone by any single mechanism that reactivates ERK. There are a number of ways we can conceive of discovering combinations that behave more like the former case. One would be to find two truncal drivers in tumors that act by distinct pathways and mechanisms, and inhibitors of which would be unlikely to share mechanisms of resistance. Unfortunately, beyond co-mutation with TP53, there are few examples of such truncal driver pairs, and even fewer examples where drugs are available. Activation of the RAS-MAPK pathway due to mutation or amplification of pathway members co-occurs with MTAP gene perturbation in ~4–5% of all cancer patients, with notably higher percentages of patients harboring MTAP loss and MAPK activating gene alterations in NSCLC (7–8%), PDAC (14–15%), and GBM (35–37%)[10]. Thus, these early truncal events are amongst the most common pair of truncal driver events across human cancers.

Due to the high prevalence of cancers with hyperactivity of the RAS-MAPK pathway co-occurring with MTAP loss and its clinical relevance[95], together with the observation that several members of the RAS-MAPK pathway were strong hits in our screen, we focused our attention on assessing combination treatments of MTAC-PRMT5i with inhibitors of the RAS-MAPK pathway. While single agent therapeutics against both pathways were effective in regressing a double mutant lung cancer model, only the combinations led to complete regressions which in many cases were sustained. Creating sustained longer-term remissions in cancer requires non-cross resistant therapies. And indeed, we see evidence in both directions that the therapies are non-cross resistant, potentially explaining the durability in the responses we see (Figure 4, 6). Interestingly, we noted somewhat better therapeutic activity of the combination of MRTX1719 with the RAF inhibitor belvarafenib. Notably, resistance to treatment with G12C RAS inhibitors has been frequently mediated by KRAS amplification[92] which will create an unfavorable enzymatic equilibrium for direct KRAS inhibition. At the same time, KRAS amplification will drive continuous activation of downstream nodes, such as CRAF and BRAF, however, levels of those proteins will not necessarily be increased. Thus, there may be some advantage to the A/B/CRAF inhibitors in this regard.

While the therapeutic efficacy of these combinations might simply be due to the co-occurrent truncal mutations, it is also possible that there is cross talk between these two pathways. For example, mono-methyl arginine events have been observed for Raf and EGFR proteins[96], [97], but on the other hand, no symmetrically dimethylated members of the RAS-MAPK pathway have been identified in our or other unbiased proteomics enrichment studies for PRMT5 substrates[24], [28], [29]. Similarly, robust physical interactions between the PRMT5-WDR77 methylosome and core members of the RAS-MAPK pathway have not been well defined. In one unbiased proteomics study using biotin proximity labeling, MEK2 was found to interact with PRMT5 and several other PRMT family proteins[98]. This work will need to be validated and explored further to determine whether PRMT5 interacts transiently with MEK or other pathway regulator(s). In another study, DUSP14, a phosphatase that regulates MAPK phosphorylation, was identified as a T cell-specific substrate of PRMT5[99]. However, our SDMA IP-mass spectrometry in MiaPaca2 cells did not detect any DUSP proteins[29], these data are in agreement with findings by Mirati Therapeutics in LU99 cells where no canonical RAS-MAPK proteins were detected as PRMT5 methyl substrates[35]. Our transcriptomics data demonstrate that MTAC-PRMT5i treatment activates a number of cellular stress pathways (Figure 6A; S15A - inflammatory response, and p53, JAK/STAT, and NFkB pathways). Intriguingly, p53 and JNK have both been shown to regulate the transcription of DUSP proteins[99], [100], [101], key regulators of phospho-ERK1/2 and ERK1/2 effectors such as Fra1. Thus, a potential mechanism of MTAC-PRMT5i induction of RAS-mediated transcription is through a number of activated stress kinases acting directly or indirectly on ERK1/2 and its effector proteins.

In addition to RAS-MAPK, we observed the resistant or sensitizing effects of several other genes of interest outside the scope of this paper. CARM1 encoding PRMT4 which when knocked out scored as a strong mechanism of resistance/insensitivity to the MTAC-PRMT5i, with mildly repressive sgRNA scores in the DMSO arm and highly pro-growth scores in the +MRTX9786 arms. Little is known about PRMT4-PRMT5 crosstalk and no other methyltransferases tested showed interaction with MTA in previous assays[6]. PRMT4 places a distinct methyl mark compared to PRMT5 and has not previously been reported to alter a cellular response to PRMT5. Further work will be needed to understand the crosstalk between these two family members or to ascertain whether MRTX9786/MRTX1719 molecules can inhibit both CARM1 and PRMT5.

Another interesting question for follow-up study is how cancer cells die in response to PRMT5i in the TP53 WT and TP53 mutant settings. For example, the DNA damage-mediating E3 ligases in the SMURF family and the BCL2 family members being activated in the LU99 cells suggest that TP53 WT cells may be undergoing p53-dependent apoptosis in response to DNA damage caused by DNA-RNA R loops that are downstream of PRMT5 loss[27], [102], [103]. However, how a TP53-inactive, MTAP-null cell is decreasing in growth or dying in response to these inhibitors remains unresolved. It would be interesting to access specific responses in cohorts of patients or cell lines where TP53 is inactivated to determine the p53-independent cell death mechanism when PRMT5 activity is lost.

Some genes that notably did not hit in our screens include mitotic regulators and CDK4/6 inhibitors. Early PRMT5i that bound in a SAM-cooperative fashion (and lacked selectivity for the tumor) showed combinatorial effects together with mitotic spindle drugs (taxanes) in vitro[78], but we did not observe loss of mitotic regulators as strong hits in our screens. In this case, CRISPR screens would likely result in pan-lethal effects of mitotic regulators (as in Fig. 1A) and thus additional benefits of partial inhibition of such mitotic regulators (e.g., Wee1) in combination with PRMT5 inhibitors would be difficult to assess. CDK4/6i has been proposed as a rational combination with PRMT5i given the G1 checkpoint overlap between loss of CDKN2A and reliance of CDK4/6[104]. However, while CDK4 and CDK6 are independently and digenically disrupted in our screen, we did not observe obvious additional benefit of targeting these in the 21-day CRISPR screens. On the other hand, CDK4/6 inhibition often exerts more profound benefit in in vivo efficacy studies such as PDX models[105] thus, in vivo testing of combinations may still be worth exploring. In agreement with its known role as a regulator of cellular SAM concentration and thereby a synthetic lethal with MTAP deletion, we detect MAT2A as an additional sensitizer and potential combination to test with PRMT5; MAT2A and PRMT5 inhibitors are currently being tested as a rational combination in clinical trials for solid MTAP deleted cancers (NCT05975073).

In conclusion, using functional genomic screens we have identified combinations of PRMT5- and RAS-MAP kinase- inhibitors that are achieving complete responses in preclinical models. These combinations could be explored in clinical trials for patient benefit in 80,000–100,000 patients a year.

Significance:

Combining PRMT5 and MAPK pathway inhibitors leads to complete, durable responses in lung cancer models, providing an effective therapeutic strategy for the 4–5% of cancer patients harboring CDKN2A/MTAP deletion and MAPK alterations.