Fig. 4: In vivo validation of cell-state-specific depletion by OncoTarget/OncoTreat-predicted drugs in a DIPG17 cell-line-derived subcutaneous xenograft model.
a, Bioluminescence imaging 1 month post-injection, confirming tumor implantation in the DIPG17 subcutaneous xenograft model. b, UMAP projection of VIPER-inferred protein activity from 14,176 single cells across four vehicle-treated tumors, colored by assigned human DMG cell state based on the strongest statistically significant enrichment of human state-specific MRs per cell (one-sided GSEA, FDR-adjusted). c, Heatmap of viperSimilarity NES quantifying concordance between human and mouse cell state MRs. Red indicates concordant enrichment and blue indicates inverse enrichment. d, OncoTarget and OncoTreat scores for drugs tested in vivo, shown as −log10(Bonferroni-adjusted P values) derived from one-sided aREA, demonstrating conservation of predicted cell-state-specific activity between human tumors and the model. e, Percent change in tumor volume 5 days post treatment across nine drug arms and vehicle control. Statistical significance between drug-treated and vehicle-treated tumors was assessed using two-sided Welch’s t-test. Representative ultrasound images before and after avapritinib treatment illustrate treatment-associated morphological changes (pink arrow). Each data point represents an individual mouse (biological replicate): vehicle (n = 5), trametinib (n = 4), mocetinostat (n = 3), napabucasin (n = 3), avapritinib (n = 2), dinaciclib (n = 2), etoposide (n = 2), larotrectinib (n = 2), ruxolitinib (n = 2) and venetoclax (n = 2). f,g, log2(fold change (FC)) in the fraction of cells assigned to each cell state following drug treatment relative to vehicle-treated tumors (14,176 single cells from four tumors), across 38,615 single cells from tumors treated with five OPC-targeting drugs (f) and 42,896 single cells from tumors treated with four AC-targeting drugs (g). Bar heights represent mean log2FC across biological replicates (each dot represents an independent tumor); error bars, s.e.m. Statistical significance of differences in the proportion of cells assigned to the predicted targeted cell state was assessed using χ2 tests for each treatment arm, with FDR correction applied within each drug across cell states: avapritinib (n = 2; POPC < 10−300), mocetinostat (n = 2; POPC = 6.94 × 10−298), dinaciclib (n = 2; POPC < 10−300), etoposide (n = 2; POPC < 10−300), trametinib (n = 2; POPC < 10−300); ruxolitinib (n = 2; PAC = 3.36 × 10−6), larotrectinib (n = 2; PAC = 6.14 × 10−13), venetoclax (n = 3; PAC = 4.64 × 10−2) and napabucasin (n = 3; PAC = 2.51 × 10−15; significant shift opposite to the predicted direction). Asterisks denote statistical significance: *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. h, Pharmacodynamic analysis of cell-state-specific tumorigenic MR activity inversion in vivo. Analytic aREA plots show enrichment of the top (red ticks) and bottom (blue ticks) 50 human DMG cell-state-specific tumorigenic MRs in drug-treated versus vehicle-treated protein activity signatures from the same cell state. Representative examples are shown: trametinib in OPC-like cells and venetoclax in AC-like cells. P values correspond to two-sided aREA for each individual drug-state comparison. i, Cell-state-specific differential protein activity analyses (NES, x axis) confirming cell-state-specific (y axis) drug-mediated inversion of OncoTarget-predicted dependencies: CDK2 activity inhibition by dinaciclib in OPC-like cells (P = 5.1 × 10−33, one-sided aREA) and JAK1 inhibition by ruxolitinib in AC-like cells (P = 1.8 × 10−9, one-sided aREA).
