br Gene expression analysis br
Gene Fer1 analysis
To study the relationship between pretreatment expression levels of OCM-related genes and cancer cell response to an-titumor drugs, we used gene expression microarray data for 635 cell lines from the Cancer Cell Line Encyclopedia that were matched with drug response information for all agents available from the Genomics of Drug Sensitivity in Cancer dataset [22,25–27]. We further refer to this gene expression and drug response dataset as GDSC-CCLE dataset. The GDSC drug response data included the total of 251 drugs with sensitivity measures previously reported by Iorio et al. . Based on earlier reports [1,3,4,8–10,14,28], we selected 34 OCM-related genes that included AHCY, ALDH1L1, ALDH2, AMT, ATIC, BHMT, CBS, CTH, DHFR, FOLH1, FOLR1, FOLR2, FOLR3, FTCD, GART, MAT1A, MAT2A, MAT2B, MTHFD1, MTHFD2, MTHFD2L, MTHFR, MTHFS, MTR, MTRR, NNMT, PEMT, PHGDH, SHMT1, SHMT2, SLC19A1, SLC46A1, TCN2, and TYMS. Their log2-transformed expres-sion levels were downloaded from the from CCLE web re-source of the Broad Institute . These transcriptional mea-sures had been generated using Affymetrix Human Genome U133 Plus 2.0 microarrays and normalized using the Robust Multi-array Average (RMA) algorithm . For each gene, expression data from multiple microarray probes were aver-aged. The distribution of the log2-transformed gene expres-sion measures was examined using histogram plots.
Drug sensitivity measures
To explore possible associations of pretreatment OCM gene expression with tumor cell response to a broad range of anti-cancer drugs, we examined correlations between pre-treatment transcriptional levels of 34 OCM-related genes and drug sensitivity, measured as IC50 (the total drug inhibitor concentration that reduced cell activity by 50%) of 251 agents among all cancer cell lines . Chemosensitivity values from the GDSC resource , in the ln(IC50) format, were obtained from the Supplementary Table 4A of Iorio et. el.
 and converted to the log10(IC50) scale, to which we further refer as log(IC50). Fourteen agents (bicalutamide, UNC0638, JQ1, AZD6482, AZD6244, CHIR-99021, BMS-708163, GSK269962A, BMS-536924, RDEA119, GDC0941, olaparib, afatinib, and PLX4720) had duplicate IC50 mea-surements within the GDSC dataset. The concordance between duplicate cell line chemosensitivity measures within the GDSC dataset and their agreement with drug response measures independently generated by the CCLE and Genentech studies has been demonstrated and val-idated in previous studies [30,31]. Accordingly, all the 14 agents in the GDSC dataset had p ≤ 3.37 × 10−8 for Pearson correlations between duplicate cell line response in our analysis. For these 14 agents with biological duplicates of drug response measures, we used a combined average of their log(IC50) drug response measurements from separate experiments .
Statistical analysis of association between OCM gene expression and drug response
Identities of the cell lines present in both CCLE and GDSC datasets were verified using information from Cellosaurus (32). Correlation and regression analyses were performed using R packages Hmisc, ggpubr, and lm.beta v.1.5–1, R environment v. 3.2.3, 3.4.1, and 3.5.3, RStudio v. 1.0.153, Python v. 2.7.12 and 2.7.15, and rpy2 v. 2.8.5. Association be-tween pretreatment log2-transformed gene expression mea-sures and log(IC50) of drug response was examined using Pearson correlation analysis. The resulting p-values were ad-justed for multiple testing using the Benjamini and Hochberg’s method  of false discovery rate (FDR) adjustment. FDR adjustment of correlations between OCM gene expression levels and log(IC50) values accounted for all 251 cancer drugs in the GDSC dataset and 34 folate genes. We selected genes associated with drug response using a cutoff of the absolute value of Pearson correlation coefficient r > 0.3 and p < 0.05 for FDR-adjusted p-values. The gene-drug pairs satisfying these criteria included 51 distinct agents, 50 of which had defined genetic drug targets (Table 1). We examined the bi-ological roles of those OCM genes that satisfied these crite-ria and were associated with multiple antitumor agents, and of those cancer drugs which were associated with multiple genes. Analysis of overlapping agents associated with multi-ple folate genes was conducted using Venn diagrams inferred using Venny 2.1.0 .
To validate the robustness of the Pearson correlation re-sults, we examined the non-parametric Spearman correla-tion of log2-transformed OCM gene expression values with 21
log(IC50). The resulting p-values were adjusted for FDR while accounting for all 251 agents and 34 OCM genes. We exam-ined which of the agents with Pearson r > 0.3 and FDR adjusted p < 0.05 in Pearson correlation also satisfied the threshold for the absolute value of Spearman correlation co-efficient ρ > 0.3 and FDR adjusted p < 0.05 in Spearman correlation analysis.