An interactive resource to identify cancer genetic and lineage dependencies targeted by small molecules

Abstract

The high rate of clinical response to protein-kinase-targeting drugs matched to cancer patients with specific genomic alterations has prompted efforts to use cancer cell line (CCL) profiling to identify additional biomarkers of small-molecule sensitivities. We have quantitatively measured the sensitivity of 242 genomically characterized CCLs to an Informer Set of 354 small molecules that target many nodes in cell circuitry, uncovering protein dependencies that: (1) associate with specific cancer-genomic alterations and (2) can be targeted by small molecules. We have created the Cancer Therapeutics Response Portal (http://www.broadinstitute.org/ctrp) to enable users to correlate genetic features to sensitivity in individual lineages and control for confounding factors of CCL profiling. We report a candidate dependency, associating activating mutations in the oncogene β-catenin with sensitivity to the Bcl-2 family antagonist, navitoclax. The resource can be used to develop novel therapeutic hypotheses and to accelerate discovery of drugs matched to patients by their cancer genotype and lineage.

Figure 1. Response of CCLs to Informer Set

Sensitivity of 242 CCLs to small-molecule probes/drugs was assessed at dose (CellTiterGlo) and areas under the concentration-response curve (AUC) were computed. Data are shown as box plots indicating distributions of AUC values for each compound (A) and a heatmap of AUC values (scale represents AUC values ranging between 1 (sensitive; red) and 6 (unresponsive; blue)) (B) for single CCLs (columns) treated with single compounds (rows). Missing numerical values in heatmap were imputed using a k-nearest neighbors approach. AUC distributions were analyzed by incorporating context-dependent exclusions (C) of cell lines (grey bars represent excluded cell lines). See also Figure S1 and Table S1.

Figure 2. Genetic dependencies targeted by small molecules

The distribution of CCL response (AUC values) to compound treatment is represented as a heatmap denoting sensitivity (red) or unresponsiveness (blue) aligned with genomic alterations for corresponding CCLs (gray bars). The resource identified known clinically drug-targeted genetic dependencies (A) and known drug-resistance mechanisms (BRAF V600E outlier cell lines: *RKO; ^#SKMEL28). The resource also suggests dependencies with both mutation and copy number variation in MYC (B). Global analysis of the resource showed EGFR-mutated CCLs are unresponsive to NAMPT inhibitors (C). CNV-H: high-copy number (≥8 copies), TES: all targeted-exome sequencing mutant calls, TES-A: targeted-exome sequencing, non-neutral missense mutations; Onco: Oncomap mutant calls, MUT: any mutation call. See also Figure S2, Table S2, User Guide S1, and Table S3.

Figure 3. Lineage dependencies targeted by small molecules

Ovarian CCLs are highly sensitive to ML210 and RSL3 (A). An expanded panel of ovarian CCLs showed sensivity to ML210 (IC₅₀ of ~10 nM) independent of the BRCA1 status of the CCLs (B). See also Figure S3.

Figure 4. Mutations in β-catenin associate with sensitivity to navitoclax

Activating mutations in β-catenin (CTNNB1) or mutations in members of its destruction complex (AXIN1; CSNK1A1) correlate with sensitivity to navitoclax (A). Previous studies have linked the Wnt/β-catenin pathway to expression of Bcl2 family members (B). An elastic-net regression model (black circles: observed; red crosses: predicted; weighted root-mean-squared error: 1.45) predicts AUC sensitivity values across CCLs treated with navitoclax (C). Heatmap depicts model features (rows; e.g., mutation, copy number) sorted by descending weight (black bars) across all CCLs tested (columns). Scale represents range of normalized values between -3 and 3 (red: relative higher copy number, presence of mutation; blue: relative lower copy number). All model features (Table S4) were input to Ingenuity Pathway Analysis (Jimenez-Marin et al., 2009) and the highest-scoring network (D) contains β-catenin as a central node (p=10⁻³⁸). The network contains members of the β-catenin pathway present in the regression model (brown), other genes present in the model (dark grey), and molecular interactions with non-regression-model features (light grey). See also Table S4 and Table S5.

Figure 5. Confirmation experiments for navitoclax/β-catenin

Response to navitoclax observed in large-scale profiling was confirmed in 3 independent experiments (A) with the 7 most sensitive CTNNB1-mutant CCLs (gray bars) and 4 control CCLs lacking mutations in CTNNB1 (white bars). In parallel, caspase 3/7 activation after navitoclax treatment was measured (B), showing that loss of viability was due to induction of apotosis. The response of previously untested CTNNB1-mutant CCLs (red bars) to navitoclax was measured using the same conditions from our initial profiling experiments (C). Data are represented as mean +/- SD. See also Figure S4.

Figure 6. Small-molecule induction of β-catenin levels and sensitivity to navitoclax

Treatment with the GSK3β inhibitor CHIR-99021 led to increased levels of β-catenin in RKO and HT29 (non-mutant) cells (A) and relatively little change in HEC59 (non-mutant) and SW48 (S33Y CTNNB1 mutant) cells (D). Sensitivity to navitoclax was assessed after pre-treatment with CHIR-99021 (red) and compared to DMSO-pretreated controls (black). RKO (B) and HT29 (C) cells, which had increased levels of β-catenin, showed 4-fold increase in sensitivity to navitoclax, while HEC59 (E) and SW48 (F) cells, which had unchanged levels of β-catenin, demonstrate no significant change in sensitivity. Data are represented as mean +/− SD. See also Figure S5.