Fig. 6: Chemogenomic profiling of PDXs reveals actionable feature for difficult-to-treat tumors.

a Heatmap of actionable features for PDX models (n = 36) derived from clinical data (receptor and germline BRCA1/2 status), whole-genome sequencing (WGS; copy number alterations and mutations), outlier expression from RNA-sequencing and reverse-phase protein array (RPPA) and chemosensitivity from in vivo profiling. b Heatmap of FGFR1 copy number and RNA expression across PDX library (n = 33). c Waterfall plot of BestAvgResponse for PDX clinical trial in mice treated with BGJ398 (n = 5). d Tumor growth curve for mammary fat pad tumors of GCRC1971 PDX either untreated or BGJ398-treated (n = 1 per arm). e Tumor growth curve for spontaneous luciferase-tagged skull-base metastases of GCRC1971 PDX untreated and BGJ398-treated mice (n = 3 per arm). Mean ± SEM. f Heatmap of PI3K/mTOR combined RPPA signature score (above) and expression of individual probes (below) across PDXs (n = 36). Samples ranked on PI3K/mTOR score. g Waterfall plot of BestAvgResponse for PDX clinical trial in mice treated with everolimus (n = 20). h Correlation plot of PI3K/mTOR RPPA score and BestAvgResponse to everolimus for cohort in g. i Tumor growth curve for GCRC1979 PDX either vehicle or everolimus-treated (n = 3 per arm). Mean ± SEM. j Schematic of clinical history for patient GCRC1738, including treatments and tumor sizes. k Tumor growth curves for GCRC1738 PDX either untreated, doxorubicin, paclitaxel, gemcitabine or cisplatin treated (n = 1 per arm).