Figure 3. Simulated genome-wide knockout screen identifies targets of redox metabolism in radiation-resistant cancers.

(A) Schematic comparing maximum total NADPH production between WT (left) and G6PD-knockout (right) models for individual patient tumors. Note that for reactions with more than one associated gene (e.g. GLUD1/2 reaction), only one gene is knocked out at a time. (B) Effect of simulated knockout of each individual gene in Recon3D on total NADPH production in TCGA tumors. Values are the ratio of total NADPH production after versus before knockout. Genes are rank ordered based on increasing mean KO/WT ratio (decreasing gene knockout effect) across all tumor models. Outset: KO/WT ratios are averaged across all tumor models. Inset: For the top 15 genes, KO/WT ratios from individual patient tumor models are shown, along with the comparison between radiation-sensitive and -resistant cohorts. (C) Volcano plot comparing the effect of each simulated gene knockout (individual dot) on total NADPH production between radiation-sensitive and -resistant tumors. X-axis: log₂(Resistant/Sensitive), where “Resistant” equals the mean (WT-KO)/WT ratio in radiation-resistant tumors, and “Sensitive” equals the mean (WT-KO)/WT ratio in radiation-sensitive tumors; values < 0 (green dot on the left of the dotted line) signify knockouts with greater effects on total NADPH production in radiation-sensitive tumors, whereas values > 0 (red dot on the right of the dotted line) signify knockouts with greater effects on total NADPH production in radiation-resistant tumors. Y-axis: statistical significance (false discovery rate-adjusted p-values based on the Benjamini-Hochberg procedure) comparing knockout effects between radiation-sensitive and -resistant tumors; values above the dotted line (FDR-adjusted p-value ≤ 0.05) are statistically-significant. The size of each dot is proportional to the overall effect size (mean (WT-KO)/WT ratio across all tumor models regardless of radiation sensitivity). (D) Volcano plot comparing the effect of each simulated gene knockout on total reduced GSH production between radiation-sensitive and -resistant tumors. Genes tested by experimental siRNA knockdown studies are bolded. (E) Schematic demonstrating the measurement of ΔE_hc GSH/GSSG (difference in glutathione half-cell potential between siRNA knockdown and negative control) in radiation-sensitive and -resistant cancer cell lines. (F) Comparison of model-predicted and experimentally-measured effects of gene knockdown on reduced GSH production. Top 3 rows: ΔE_hcRes - ΔE_hcSens in siRNA knockdowns across all three cell line pairs. ΔE_hcRes - ΔE_hcSens > 0 for gene knockdowns causing greater oxidation in the radiation-resistant cell line, corresponding to a model-predicted log₂(Resistant/Sensitive) > 0. Middle row: t-statistic from 1-sample t-test comparing the three experimentally-measured values of ΔE_hcRes - ΔE_hcSens to the null hypothesis population mean of zero (equal effect in radiation-sensitive and -resistant cell lines). Bottom row: model-predicted log fold change in gene knockout effect on reduced GSH production between radiation-resistant and -sensitive TCGA tumor models. ns: not significant, *: p ≤ 0.05, **: p ≤ 0.01, ***: p ≤ 0.001, ****: p ≤ 0.0001. Boxplots: box = 25^th, 50^th, and 75^th percentiles, whiskers = 1.5 times the interquartile range. See also Figures S9–S10, S11A, S12–S13, Tables S2–S3.