MTAP deficiency creates an exploitable target for antifolate therapy in 9p21-loss cancers

Methylthioadenosine phosphorylase, an essential enzyme for the adenine salvage pathway, is often deficient (MTAPdef) in tumors with 9p21 loss and hypothetically renders tumors susceptible to synthetic lethality by antifolates targeting de novo purine synthesis. Here we report our single arm phase II trial (NCT02693717) that assesses pemetrexed in MTAPdef urothelial carcinoma (UC) with the primary endpoint of overall response rate (ORR). Three of 7 enrolled MTAPdef patients show response to pemetrexed (ORR 43%). Furthermore, a historic cohort shows 4 of 4 MTAPdef patients respond to pemetrexed as compared to 1 of 10 MTAP-proficient patients. In vitro and in vivo preclinical data using UC cell lines demonstrate increased sensitivity to pemetrexed by inducing DNA damage, and distorting nucleotide pools. In addition, MTAP-knockdown increases sensitivity to pemetrexed. Furthermore, in a lung adenocarcinoma retrospective cohort (N = 72) from the published BATTLE2 clinical trial (NCT01248247), MTAPdef associates with an improved response rate to pemetrexed. Our data demonstrate a synthetic lethal interaction between MTAPdef and de novo purine inhibition, which represents a promising therapeutic strategy for larger prospective trials.

Adenine nucleotides are synthesized either by de novo biosynthesis, which is supported by folate-mediated single-carbon metabolism, or by salvage biosynthesis controlled by MTAP [24][25][26][27] . Therefore, we hypothesize that MTAP loss produces a synthetic lethal vulnerability to de novo adenine synthesis inhibition thus rendering MTAP-deficient (MTAP def ) tumor cells susceptible to antifolate agents (Fig. 1a) including pemetrexed, which is a potent antifolate agent that suppresses de novo purine synthesis (as well as thymidine synthesis) by inhibiting three of the key enzymes involved in folate metabolism, namely dihydrofolate reductase, thymidylate synthase, and glycinamide ribonucleotide formyl transferase 28 . As such, tumor MTAP deficiency may provide a metabolic vulnerability for the use of antifolate agents such as pemetrexed to effectively treat UC with 9p21 loss.
In this work, we test our hypothesis and analyze tumor tissues for MTAP expression status in correlation with clinical responses to pemetrexed in patients with metastatic UC who either enrolled on our phase II clinical trial (NCT02693717) or were historically treated with pemetrexed. We further corroborate our findings mechanistically using preclinical models and clinically using an independent clinical cohort of patients with metastatic lung adenocarcinoma.

Results
MTAP loss is prevalent in UC. Based upon genomic data 13 , more than 99% of MTAP def UC also contains CDKN2A loss. Since we already established a CLIA certified immunohistochemistry (IHC) test to determine MTAP gene expression status, we decided to use MTAP def as a surrogate biomarker for 9p21 loss. To evaluate the prevalence of MTAP loss in UC, we assessed The Cancer Genome Atlas (TCGA) database for MTAP somatic copy number alterations (SCNAs) and found that 106 of 408 (26%) muscle-invasive UC cases harbored a homozygous deletion of the MTAP gene 29 (Fig. 1b). Consistent with TCGA data, IHC analysis of a tumor tissue microarray from 151 UC patients demonstrated that MTAP loss occurred in 27.8% of tumor specimens (Fig. 1c). Of note, we observed no significant difference in the rate of MTAP loss between early-stage and advanced-stage tumors suggesting an early onset of this genomic event ( Fig. 1d and Supplementary Table 1).
Patient characteristics. In a cohort of 21 patients, including a historical cohort of 14 metastatic UC patients (4 with MTAP def and 10 with MTAP prof , respectively) treated with pemetrexed beyond first line of therapy ( Supplementary Fig. 1) and our current trial cohort of 7 patients (all with MTAP def ) enrolled in clinical trial NCT02693717, the median age was 70 (Table 1). Approximately two thirds of the patients were male. More than half (52%) of the patients had a limited Eastern Cooperative Oncology Group (ECOG) performance status (1 or 2). Prior cisplatin-based chemotherapy was given in 16 of 21 (76%) of patients and prior ICT was given in 6 of 21 (28%) of patients. Patients enrolled in NCT02693717 (09/2017-2/2019) were more likely to have received prior ICT (71%) as compared to patients included in our historical cohort (7%) as patients in the historical cohort were treated before the approval of ICT for UC (9/2014-2/ 2016). Furthermore, 47% of patients in our entire cohort had received ≥2 lines of prior therapy, as detailed in Table 1.
Pemetrexed is clinically effective against MTAP def UC. In our historical cohort analysis, all MTAP def UC patients (4 of 4) had an objective response to pemetrexed (ORR 100%), while only one of the 10 MTAP-proficient (MTAP prof ) UC had an objective response (ORR 10%). Notably, 8/10 patients with MTAP prof UC had an increase in tumor volume from baseline as best response (Fig. 1e).
Our prospective study NCT02693717 (Fig. 1f) enrolled 7 patients with previously treated MTAP def UC but was intentionally terminated early in favor of another trial with pemetrexed plus anti-PD-L1 combination therapy in the same population of patients. The primary endpoint was overall response rate (ORR) to pemetrexed, which was 42% (3 of 7 patients) (Fig. 1g). The secondary endpoints were survival and safety of pemetrexed. After a median follow up of 7.1 months, the estimated median progression-free survival (PFS) is 5.3 months and median overall survival (OS) is 7.7 months. All and treatment-related adverse events (AEs) experienced by patients treated with pemetrexed in our phase II trial are summarized in Supplementary Tables 2 and  3. There were no grade 4 or 5 treatment-related AEs. Most common treatment-related AE was anemia and includes grade 3 anemia in 3 of 7 (43%) of patients. Figure 1h shows an imaging example of a patient with metastatic MTAP def UC who had a near-complete response to pemetrexed (after progressing through first-line gemcitabine and cisplatin) as compared to a patient with metastatic MTAP prof UC who progressed on pemetrexed (after progressing through first-line gemcitabine and cisplatin).
Antifolates are effective in human MTAP def UC cell lines and xenograft models by distorting nucleotide pools and inducing DNA double-strand breaks. To further test our hypothesis that MTAP deficiency may create a vulnerability to antifolate agents, we assessed the sensitivity of UC cell lines to pemetrexed. For this purpose, we screened the available human UC cell lines, confirmed MTAP expression status using Western blot, and identified four MTAP prof UC cell lines (HT-1376, T24, HT-1197, and J82) and four MTAP def UC cell lines (RT112, RT4, UM-UC-3, and 253 J) (Fig. 2a). The T24 cell line was identified as MTAP prof but p16 (product of CDKN2A) deficient ( Supplementary Fig. 2a). We also confirmed the functional consequences of MTAP loss by detecting a >30-fold increase in the concentrations of its substrate, MTA, in the culture medium of MTAP def UC cell lines as compared to the culture medium of MTAP prof UC cell lines ( Fig. 2a and Supplementary Fig. 2b), consistent with previous reports 14,15,30 . We found that the viability of MTAP def UC cell lines in response to pemetrexed treatment was much lower (IC 50 < 0.25 μM) as compared to MTAP prof UC cell lines (IC 50 > 20 μM) (Fig. 2b). In addition, we ruled out that accumulating MTA has any impact on the MTAP def and MTAP prof cell lines viability in the absence or presence of pemetrexed (Supplementary Fig. 3a, b). On the other hand, we did not observe a difference in the viability of MTAP def cell lines in comparison to MTAP prof cell lines when treated with the antimetabolite gemcitabine, which is commonly used in urothelial cancer (Supplementary Fig. 4a). Notably, the T24 cell line did not demonstrate high levels of sensitivity to pemetrexed (Fig. 2b)  these findings, a much higher proportion of MTAP def UC cells underwent apoptosis in response to pemetrexed treatment as compared to MTAP prof UC cells, as measured by proportion of cells in sub-G1 phase (Fig. 2c) and poly (ADP-ribose) polymerase-1 (PARP1) cleavage ( Supplementary Fig. 2c). Of note, these striking differences in apoptosis were observed even though MTAP def UC cells were treated with pemetrexed at a 40-fold lower concentration (0.5 μM) as compared to MTAP prof UC cells (20 μM) (Fig. 2c), suggesting a wide therapeutic window of pemetrexed for MTAP def UC. Next, we assessed whether pemetrexed treatment led to increased DNA double-strand breaks (DSBs) in correlation with MTAP def . DSBs were evaluated via the detection of phosphorylated histone H2A, X variant (γH2AX) and foci of p53-binding protein 1 (53BP1). The 4 MTAP def cell lines had significantly higher γH2AX and 53BP1 signals as compared to the 4 MTAP prof cell lines (Fig. 3a-d). As a positive control, the γH2AX response to treatment with gemcitabine was not different between MTAP def and MTAP prof cell lines ( Supplementary Fig. 4b, c). Furthermore, we assessed the in vitro distortion of nucleotide monophosphate (NMPs) pools upon exposure to pemetrexed. At baseline, we observed that MTAP def cell lines had a trend for lower levels of AMP, CMP, UMP and GMP as compared to MTAP prof cell lines. However, the difference was not significant. Upon treatment with pemetrexed (5 um), NMPs increased by several folds. The fold rise of NMPs had a higher trend in MTAP def as compared to MTAP prof cell lines. However, the difference was not significant between the MTAP def and MTAP prof cell lines ( Supplementary Fig. 5a, b).
Additionally, we analyzed an independent large-scale cancer cell line drug sensitivity database, the Genomics of Drug Sensitivity in Cancer (GDSC) 31 , We found that MTAP def UC cell lines from the GDSC database (n = 5) were significantly more susceptible to methotrexate with a mean IC 50 of 0.25 μM as compared to a mean IC 50 of 1.3 μM for the MTAP prof cell lines (n = 6) (P = 0.036) ( Supplementary Fig. 6a).
To further confirm our hypothesis that tumor MTAP deficiency leads to antifolate sensitivity, we carried out in vivo experiments in xenograft models of UC, which demonstrated that the MTAP prof HT-1376 tumor was resistant to pemetrexed ( Fig. 2d) whereas the MTAP def UM-UC-3 tumor was sensitive to pemetrexed (Fig. 2e). As expected, in vivo pemetrexed treatment also resulted in significantly higher levels of DNA damage (as shown by 53BP1 staining) in MTAP def UM-UC-3 tumors as compared to the MTAP prof HT-1376 tumors (Supplementary Fig. 7a-f).
Pemetrexed is effective against MTAP knockdown UC cells. To verify that pemetrexed sensitivity is attributed to loss of MTAP instead of other co-deleted genes in the 9p21.3 region 32,33 , we selectively knocked down the MTAP gene (MTAP KD ) in HT-1376 (a cell line with intact P16 expression), which led to substantial loss of MTAP protein expression and significant accumulation of its substrate MTA in the culture medium ( Fig. 2f) in two different colonies: shMTAP2 and shMTAP3. Viability of the MTAP KD HT-1376 cell lines (shMTAP2 and shMTAP3) in response to pemetrexed treatment was much lower than parental and scramble control shRNA (Fig. 2g). Increased sensitivity to pemetrexed with MTAP KD was also reproducible using a mixture of shRNA to knock down the MTAP gene in HT-1376 (Supplementary Fig. 2d, e).
L-alanosine, inhibitor of de novo adenine synthesis, is effective against human MTAP def UC cell lines. To further verify that the Nodal only 1 (25%) 6 (60%) 3 (43%) 10 (48%) Visceral (lung, bone, liver, or others) a One MTAP prof retrospective patient with MVAC as metastatic therapy was included. All other MVAC therapy was adjuvant or neo-adjuvant therapy. b Other prior therapies include Gemcitabine and Cyclophosphamide (n = 1), carboplatin and paclitaxel (n = 2), and paclitaxel (n = 1). MTAP def MTAP deficient, MTAP prof MTAP proficient, ECOG PS Eastern Cooperative Oncology Group performance status, Gem/cis gemcitabine/cisplatin, GTA gemcitabine, taxotere, and adriamycin, MVAC methotrexate, vinblastine, adriamycin, and cisplatin, CGI cisplatin, gemcitabine, and ifosfamide, IAGEM ifosfamide, adriamycin, and gemcitabine.   synthetic lethal interaction between inhibition of de novo adenine synthesis and tumor MTAP deficiency is a pathway-specific interaction, we tested the sensitivity of MTAP def and MTAP prof cell lines to L-alanosine, a well-known amino acid analog and a potent inhibitor of de novo adenosine monophosphate (AMP) synthesis. Our data indicate that MTAP def cell lines were significantly more susceptible to cytotoxicity from L-alanosine treatment as compared to MTAP prof cell lines ( Supplementary Fig. 8a).
Pemetrexed-containing chemotherapy is clinically effective against CDKN2A lo /MTAP lo lung adenocarcinoma. To further validate the correlation between tumor MTAP loss and sensitivity to antifolate agents, we took advantage of the BATTLE-2 trial 34 that enrolled patients with metastatic nonsmall cell lung cancer, some of which were treated with pemetrexed-containing chemotherapy. Among a total of 200 patients, we identified a cohort of 72 patients with lung adenocarcinoma (LUAD) who were   Eight human bladder cancer cell lines were treated with no treatment (a) or 5 μM of PEM (b) for 24 h. Cells were then fixed and stained with γ-H2AX (red), and 53BP1 (green) antibodies as well as DAPI stain (blue). Images were quantified with ImageJ for γ-H2AX (c) and 53BP1 (d). Data represent mean ± SEM of four fields and analyzed with Welch T test.
treated with pemetrexed plus carboplatin and had tumor RNA expression data on both the MTAP and CDKN2A genes (Fig. 4a). Based on the co-expression of MTAP and CDKN2A, patients were divided into four groups: CDKN2A hi /MTAP hi , CDKN2A hi / MTAP lo , CDKN2A lo /MTAP hi , and CDKN2A lo /MTAP lo (Fig. 4b).
Given the limitation of having only RNA expression data without copy number data, we chose CDKN2A lo /MTAP lo as a representative of 9p21 loss and compared it to all other groups (presumably without 9p21 loss). We noted that objective responses (54%) were significantly higher in CDKN2A lo /MTAP lo patients compared with that (24%) in the rest of patients (p = 0.018 by two-sided Fisher's exact test) (Fig. 4c and Supplementary Table 4). Of note, among the baseline characteristics of this cohort of patients with LUAD (Table 2), we observed that epidermal growth factor receptor (EGFR) alterations were more prevalent among CDKN2A lo /MTAP lo patients as compared to the rest (42% vs. 16%, P = 0.04). To assess the impact of frequently altered genes in lung cancer, we performed a generalized linear model. However, none of the assessed genes showed a significant positive nor negative association with response to therapy except MTAP (Fig. 4d, Supplementary Table 5). An example of positron emission tomography-computerized tomography of a responding patient with a CDKN2A lo /MTAP lo tumor and a non-responding patient with a CDKN2A hi /MTAP hi tumor is shown in Fig. 4e.

Discussion
Immunotherapy has revolutionized the therapeutic landscape across different cancer types including UC 35 . However, in the foreseen future, chemotherapy agents either alone or in combination with immunotherapy will continue to be one of the main therapeutic modalities for most cancers. In the era of precision medicine, predictive biomarkers, such as genomic alterations for targeted therapy 36 and PD-L1 expression and tumor mutation burden for immunotherapy 8,37 , are playing an increasingly important role to guide selection of therapeutic strategies. However, there are currently no reliable biomarkers other than tumor histology to guide selection of chemotherapy regimens. In this study, we evaluated the clinical outcomes of patients with UC who were treated with pemetrexed stratified by tumor MTAP protein expression as a surrogate marker for 9p21 status. The sensitivity to antifolates created by MTAP loss was further supported by in vitro and in vivo experiments using human UC cell lines. This sensitivity seemed to be unique to MTAP loss rather than CDKN2A loss, based upon the data that MTAP knockdown in the MTAP prof cell line HT-1376 rendered it sensitive to pemetrexed cytotoxicity whereas loss of the CDKN2A gene product p16 in MTAP prof cell line T24 did not result in sensitivity to pemetrexed. Mechanistically, antifolates are known to disrupt nucleotide biosynthesis and DNA replication, leading to DNA breakage and programmed cell death 38,39 . Our findings support this mechanism in MTAP def and a future direction of ours is to capitalize on this pathway and test combinations that target folates and DNA repair.
Furthermore, in an independent cohort of patients with LUAD, where pemetrexed is commonly used in frontline platinum-based chemotherapy combinations, we found that CDKN2A lo /MTAP lo (surrogate marker for 9p21 loss) patients had significantly higher clinical response rates to such therapy regimens as compared to patients with likely 9p21 intact tumors. Collectively, our data support the hypothesis that tumor MTAP deficiency creates a metabolic vulnerability for therapy with antifolate agents such as pemetrexed. To optimize the clinical utility of our findings, we aim to investigate MTAP among a combination of biomarkers that have been associated with responsiveness to pemetrexed such as TS and DHFR 40 .
We report an observation from clinical data, which we confirmed by performing mechanistic studies and our investigation on MTAP deficiency and antifolate sensitivity is, to our best knowledge, the first in urothelial cancer. One of the few common features among the 21 analyzed patients was their receipt of pemetrexed as second-line therapy or beyond, although their exposure to prior lines of therapy varied. Despite these clinical heterogeneities, our study represents the first effort to assess clinical response to pemetrexed in correlation with a frequent tumor genomic alteration in UC. Our data demonstrate that clinical responses to pemetrexed are enriched in patients with UC containing MTAP loss. These data compare favorably to the previously published data on pemetrexed as a second-line therapy, which showed a modest RR of 28% (13 of 47 patients) in UC without biomarker correlation 41 . In addition, these clinical data in UC are supported by preclinical evidence using human UC cell lines and by an independent cohort of patients with LUAD. Given the association between low MTAP expression and EGFR alterations, it is plausible that antifolates and an EGFR-directed oral tyrosine kinase inhibitor (TKI) could demonstrate synergy. In fact, a recently published phase III randomized trial in patients with advanced LUAD harboring an EGFR-sensitizing mutation showed that adding pemetrexed and carboplatin chemotherapy to gefitinib significantly prolonged PFS and OS 42 . In UC, an association was reported between MTAP loss and FGFR alterations 43,44 . Therefore, our findings support testing the addition of antifolates to the FDA-approved fibroblast growth factor receptor (FGFR)-directed oral TKI in FGFR-mutant UC 36 .
Pemetrexed has been reported to augment intra-tumor immune responses through increased T cell infiltration/activation along with modulation of innate immune pathways and as a result, enhanced the anti-tumor activity of anti-PD-L1 in pre-clinical models 45 . Based upon the immune-modulatory ability of pemetrexed and its pronounced cytotoxicity in MTAP def UC, it appears that pemetrexed can be combined with anti-PD-L1 to provide a highly effective treatment for patients with MTAP def UC. In fact, we have launched a phase II trial assessing a sequential combination of pemetrexed and avelumab (anti-PD-L1 antibody) in patients with MTAP def UC (NCT03744793). Although this trial led to an intentional, early termination of the pemetrexed monotherapy trial in MTAP def UC, we expect this tissue-rich, sequential combination trial will shed light on the interaction between pemetrexed and ICT and provide superior therapeutic benefits to patients with MTAP def UC as part of 9p21 loss. Although the term synthetic lethality classically refers to an interaction between two genetic events, synthetic lethality can also refer to cases in which the combination of a mutation and the action of a chemical compound causes lethality, whereas the mutation or compound alone is non-lethal 46 . Our data reveal a synthetic lethality strategy to exploit MTAP loss in UC, and potentially in LUAD, by inhibiting de novo adenine nucleotide synthesis, inducing DNA damage, and potentially distorting nucleotide pools using antifolate agents such as pemetrexed. Loss of MTAP along with adjacent genes such as CDKN2A in the chromosome region 9p21.3 represents one of the most frequent genomic defects existing in 14% of all malignancies and is associated with poor clinical outcomes [47][48][49] . Therefore, efforts to develop novel strategies to target MTAP def tumors have increased in recent years, resulting in the identification of a number of synthetic lethal inhibitors, including methionine adenosyltransferase II alpha (MAT2A) and type I and type II protein arginine N-methyltransferase (PRMT) inhibitors 14,15,50,51 . In this report, we identify an opportunity for a synthetic lethal strategy using a safe agent 52 , pemetrexed, in MTAP def tumors. These data may be extrapolated to other malignancies that harbor MTAP loss as part of chromosome region 9p21.3 deletion.

Methods
Urothelial carcinoma study population Trial design, patients, and treatment. NCT02693717 is a single-arm nonrandomized phase II clinical trial to evaluate pemetrexed disodium in previously treated metastatic MTAP def UC. MD Anderson Cancer Center (MDACC) and the National Cancer Institute (NCI) sponsored this trial. This trial was approved by the Institutional Review Boards (IRB) of MDACC (MDACC protocol 2015-0592). The primary objective of this trial was to evaluate the overall RR defined by Response Evaluation Criteria in Solid Tumors (RECIST) 1.1. The secondary objectives included: 1) evaluating PFS, 2) evaluating OS, 3) evaluating the safety of pemetrexed therapy, 4) collecting blood, urine, and tissue for future translational studies. All primary and secondary outcomes are reported in the manuscript. Enrollment was done between 9/1/2017 and 1/25/2019. Patients provided written consent and the signed consent form was scanned into the electronic medical record. Patients also received a copy of the signed consent form. Patients were not compensated for their participation. Patients were required to have histological confirmation of metastatic UC and sufficient tumor tissues for MTAP IHC testing. Patients who had received any non-antifolate-containing systemic therapy (including immunotherapy) were eligible. The total estimated accrual for this trial was 25 patients. However, only seven patients were enrolled due to competing protocols and the opening of NCT03744793. Patients received pemetrexed disodium IV over 10 min on day 1. Courses repeated every 21 days in the absence of disease progression or unacceptable toxicity. Pre-and post-treatment blood and urine samples were collected for correlative biological analyses per an IRB-approved laboratory protocol MDACC PA17-0577.
Historic cohort. We retrospectively evaluated 14 metastatic UC patients treated with pemetrexed beyond first line at MDACC between 1/1/2014 and 7/1/2018 (MDACC PA17-0577). Four patients were MTAP def and 10 were MTAP prof upon IHC staining of tumor tissue. Details about this cohort are described in the "patients characteristics" paragraph in the Results section and in Table 1.
Objectives and statistical plan. We analyzed tumor responses to pemetrexed based on MTAP status. Patient characteristics were tabulated. The response was considered to be complete (CR) or partial (PR) per RECIST 1.1. All other responses, including not evaluated were considered non-responses. The proportion of responses for MTAP def vs. MTAP prof were compared with Fisher's exact test to accommodate the small number of patients. AEs graded according to the National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE) version 4.03. OS was defined as the time from the first day of pemetrexed treatment until death or last contact. PFS was defined as the time from the first day of pemetrexed treatment until disease progression, death, or last disease assessment. Patients alive and without progression at their last assessment were censored on their last date of assessment before starting a new treatment. Kaplan-Meier curves are presented for OS, and PFS. Only descriptive results are provided due to sample sizes being too small for comparison. Analyses were performed in SAS 9.4 (The SAS Institute Inc, Cary, NC). Kaplan-Meier curves were created in Stata 14.1 (StatCorp, College Station, TX).

Lung adenocarcinoma study population
Patients and treatment. Among the 101 patients with LUAD enrolled in the BATTLE-2 study 34 and having enough tissue for gene expression analysis, we analyzed the distribution of CDKN2A and MTAP genes to separate patients with 9p21 deletion. Gene expression analysis was done by messenger RNA GeneChip Human Gene 1.0 ST Array from Affymetrix (Santa Clara, CA). CDKN2A and MTAP expression was divided into high (hi) or low (lo) at the median cutoff ( Supplementary Fig. 5,A, B). Patients then were divided into four groups based on the co-expression of MTAP and CDKN2A: CDKN2A hi /MTAP hi , CDKN2A hi / MTAP lo , CDKN2A lo /MTAP hi and CDKN2A lo /MTAP lo (Fig. 3b).
Objectives and statistical plan. Next, we evaluated 72 patients with metastatic LUAD treated with pemetrexed-based chemotherapy that had available imaging for response assessment based on RECIST 1.1. The response was considered to be complete (CR) or partial (PR) per RECIST version 1.1. All other responses, including not evaluated were considered nonresponders. The proportion of responses for CDKN2A lo /MTAP lo vs others were compared with Fisher's exact test to accommodate the small number of patients. OS was defined as the time from the first day of pemetrexed treatment until death or last contact. PFS was defined as the time from the first day of pemetrexed treatment until disease progression, death, or last disease assessment. Patients alive and without progression at their last assessment were censored on their last date of assessment before starting a new treatment. Kaplan-Meier curves are presented for OS, and PFS.
Generalized linear model. To evaluate the association of the 10 most altered genes in lung cancer with drug response, generalized linear model (GLM) was used to estimate the odds ratio and p-value for each gene in the lung cancer cohort (n = 72) independently, with 1 indicating a responder and 0 a non-responder. Genes with an odds ratio >1 (log (odds ratio) >0) and a p value <0.05 are considered to be positively associated with responders. Genes with an odds ratio <1 (log (odds ratio) <0) and a p value <0.05 are considered to be negatively associated with responders.
Patient samples for tissue microarray IHC. Patients eligible for enrollment were those seen at Mayo Clinic (Scottsdale, Arizona) who were >18 years old, able to provide informed consent, and undergoing evaluation for genitourinary diseases. Patients were contacted during routine clinical visits or in preoperative settings within Mayo Clinic departments and divisions, including urology, radiation oncology, pathology, and medical oncology. Patients were excluded if they declined to participate or if the banking of their biospecimens would compromise the availability of tissue for diagnosis and standard clinical care. The protocol for collecting biospecimens, the process for consenting patients, and the current informed consent form were approved by the Mayo Clinic IRB (protocol no. 08-000980). Patients were enrolled from June 1, 2010, through January 1, 2013. Patients provided written consent and the signed consent form was scanned into the electronic medical record. Patients also received a copy of the signed consent form.
Genomics of drug sensitivity in cancer (GDSC) data. We analyzed an independent large-scale cancer cell line drug sensitivity dataset, the GDSC 31 . Analysis was performed using IC 50 values (drug concentration that reduces viability by 50%). MTAP deficiency in cell lines was determined from copy number variation (CNV) in MTAP from the Cancer Cell Line Encyclopedia (CCLE) 53 . Log2 ratio cutoff values <−3 were used to detect deep deletions for each CCLE pan-cancer cell line. Only cell lines with CNV status consistent with low expression of MTAP were maintained. UM-UC3 was excluded as the IC 50 value in the public database was not reproducible with our repeated experiments using methotrexate. 5637 was excluded as it contained an amplification of the MTAP gene. However, the MTAP protein was only expressed in the nucleus based upon our IHC study. Therefore, the enzymatic activity of MTAP was likely restricted and functionally represented an MTAP def cell line. We tested the difference in IC 50 between UC cell lines with or without deep deletion in MTAP by using a one-sided Wilcoxon rank-sum test.
Immunohistochemistry and immunofluorescent staining. MTAP immunohistochemical staining was performed on a 4 μm unstained slide of a formalin-fixed paraffin-embedded (FFPE) tissue microarray using an automated immunohistochemical stainer (Leica Bond III, Leica Biosystems, Buffalo Grove IL). Following deparaffinization and rehydration, antigen retrieval was performed at 95°C for 30 min with citrate buffer, pH 6.0. Endogenous peroxidase was blocked with 3% peroxide for 5 min. Primary anti-methylthioadenosine phosphorylase antibody (ProteinTech, Rosemont, IL) was applied at 1:1200 dilution for 15 min. Primary antibody detection was carried out using a commercial polymer system (Bond Polymer Refine Detection, Leica Biosystems, Buffalo Grove IL), and stain development was achieved by incubation with DAB and DAB Enhancer (Leica Biosystems, Buffalo Grove IL). Tumor cells were assessed for the presence or absence of labeling, defined as the presence of any staining or complete loss of staining, respectively.
γ-H2AX and 53BP1 immunofluorescent staining for 8 cell lines was performed with Nunc Lab-Tek II 8-well chamber slides system (ThermoFisher, Waltham, MA). Briefly, cells were plated in individual wells at about 70-80% confluence. Five μM pemetrexed or 5 μM gemcitabine was then added into medium, and cells were cultured for additional 24 h. At the end of treatment, anti-γ-H2AX (1:200, MilliporeSigma) or anti-53BP1 (1:200, Cell Signaling) were added to slides after fixation and permeabilization. For 53BP1 staining on FFPE sides, deparaffinization, rehydration, and antigen retrieval were processed the same as immunohistochemistry. Antibody (1:200) was then placed to tissues and incubated for 30 min at room temperature. Positive signals were captured with EVOS M5000 fluorescent microscope (ThermoFisher, Waltham, MA). At least 3 areas at 20x magnification were taken from each slide for further data analysis. Fluorescence processing and quantification were performed with ImageJ 55 .
Cell viability assay. Cell viability was determined by MTT assay. Briefly, cells were plated at a density of 2,000 cells per well in 96-well cell culture plates. Cells were then treated with PEM or L-alanosine at various doses for 72 h. Each dose was tested at least in triplicate. At the end of treatment, 0.5 mg/mL MTT was added into the cultures for 2 h. Insoluble MTT metabolites were dissolved with DMSO and absorbance was measured at 530 nm with background subtraction. IC 50 values were then calculated with Graphpad Prism 7.0 software.
MTA and nucleotide pool quantification using ultra-high-performance liquid chromatography-electrospray ionization (UHPLC-ESI)-triple quadrupole mass spectrometry. For MTA measurement, 10 6 cells were plated in 10-cm dishes in triplicate and cultured for 48 h. Culture media and cell pellets were collected separately and kept at −80°C until measurement. At the day of quantitation, cell pellet was vortexed with 0.6 mL of methanol containing 25 ng of MTA/adenosine internal standard, centrifuged at 12,000 rpm (15300 × g) for 5 min. The supernatant was decanted and evaporated under nitrogen. Each sample was reconstituted in 100 µL water. Cell line media (250 µL) was processed as cell pellets. An Agilent UHPLC Infinity II system was combined with a Chromolith reverse phase column (100 × 2 mm, 1.5 µm, MilliporeSigma, Billerica, MA). Mobile phases used were (A) 0.1% formic acid and water and (B) 0.1% formic acid and methanol at a flow rate of 0.3 mL/min. The isocratic elution was held at 20% A for 2 min, and the gradient was ramped up to 95% B in 5 min then held at 95% B for 2 min before being reduced to 80% A and then re-equilibrated for 2 mins. The column temperature was 30°C and the injection volume was 5.0 μL. Molecules were introduced through a Jet Stream electrospray ionization source to an Agilent 6495 triple quadrupole mass spectrometer (Agilent, Santa Clara, CA). The source was operated in a positive ion mode under optimum conditions. Nucleosides were identified by retention time and parent product ion transitions for MTA quantitative ion (m/z 297.81-36.1) and qualitative ion (m/z 297.81-119.0), and quantified using MassHunter software (Agilent, Santa Clara, CA).
A negative ionization method was developed on the Agilent 6495 LC-MS-MS to determine mono and tri-phosphate nucleotide levels. A Phenomenex SEPAX column (2.1 × 50 mm, 3.5 μm) was used with a gradient of Buffer A: 5 mM ammonium acetate and Buffer B: Acetonitrile. The cell pellets were harvested and extracted using 150 ul methanol/water (85/15) spiked with 25 ng/mL C13-CTP, C13-ATP, and C13-dGTP, the vortexed for 1 min and placed on ice for 10 min. After centrifugation at 11,000 rpm (12,800 × g) at 4°C for 15 min, the supernatant was diluted with 1350 μl 60% Methanol before loading onto an OASIS HLB cartridge (Waters Corp., Milford, MA) that had previously been activated with methanol and water according to the SPE protocol. The final eluent (200 ul) was derivatized with MBSTFA (75 μl) and the derivatization reaction was completed over 5 min with consistent vortex. The samples were then centrifuged at 11,000 rpm for 15 min and 10 μl of the supernatant was injected in triplicate. Under optimal conditions, the lower limit of quantitation (LLOQ) was 0.1 ng/ mL 56 .
MTAP knockdown. Lentiviral particles of human short hairpin RNAs (shRNAs) specific to the MTAP gene and control scramble shRNA were purchased from Santa Cruz Biotechnology (Dallas, TX). Transduction of the HT-1376 cell line was carried out according to the manufacturer's instructions. Briefly, cells were cultured in 12-well plates to reach 50% confluence, transduced with lentiviral particles, split into 100 mm plates, and then selected with 2 μg/mL puromycin-containing culture medium. Single-cell colonies were selected and expanded in puromycin-containing culture medium. RT-PCR and Western blot were then used to confirm the extent of individual MTAP gene knockdown.
Determination of cell cycle distribution by flow cytometry. Cells were plated in 6-well plates and treated with pemetrexed for 72 h. At the end of treatment, cell cycle distribution was determined by propidium iodide (PI) staining (Milli-poreSigma, Burlington, MA). Cells were collected and adjusted to a concentration of 10 5 cells/mL, fixed by 75% ethanol for more than 1 h, and then treated with RNase A for 30 min at room temperature. After 1 h incubation with PI staining buffer at room temperature, FACSCanto II flow cytometry (BD Biosciences, San Jose, CA) was used to detect cell cycle 57 . When apoptotic cells are stained with PI and analyzed with a flow cytometer, they display a broad G1 (hypodiploid) peak, which can be easily discriminated from the narrow peak of cells with normal (diploid) DNA content in the red fluorescence channels. Each experiment was repeated at least three times.
Xenograft animal models and in vivo toxicity studies. Animal experiments were carried out under conditions adhering to approved protocols from the Institutional Animal Care and Use Committee at MDACC. Briefly, anesthetized mice were injected in the right flank subcutaneously with 106 UM-UC-3, HT-1376, HT-1376/ shCtrl, or HT-1376/shMTAP cells at day 0. Tumor growth was monitored by measuring tumor size with a caliper twice a week, and pemetrexed treatment was initiated once tumor size reached about 50-100 mm 3 . The dose of pemetrexed for both models was 200 mg/kg with a frequency of three times a week. Tumor volume was calculated using a caliber. Mice were euthanized when tumor size reached 1.5 cm in diameter, ulceration reached 0.2 cm, or moribund occurred, and were recorded as deaths.
Statistical analysis of in vitro and in vivo data. All measurement data were expressed as mean ± standard deviation. If not stated otherwise, comparisons between groups were made using the t test. P values less than 0.05 were considered significant.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data generated or analyzed during this study are included in this published article, its Supplementary Information and Source Data files. For TCGA BLCA cohort shown in Fig. 1b, the genomic data can be retrieved from NCI Genomic Data Commons (NCI-GDC: https://gdc.cancer.gov). Publicly available datasets pertaining to Supplementary  Fig. 2a can be downloaded from https://depmap.org/portal/download/. Tissue microarray data pertaining to Fig. 1c, d is included in Source Data file. Clinical data pertaining Fig. 1e, g, h is included in Source Data file. Uncropped Western Blot gels and in vitro data pertaining to Fig. 2a-g is included in Source Data file. Clinical data and gene expression data pertaining to Fig. 4a-e is included in Source Data file. Additional data related to the current study, including the study protocol of the clinical trial NCT02693717, are available from the corresponding author (Jianjun Gao) on reasonable request that does not include confidential patient information. Source data are provided with this paper.