Predicting drug susceptibility of non-small cell lung cancers based on genetic lesions

Abstract

Somatic genetic alterations in cancers have been linked with response to targeted therapeutics by creation of specific dependency on activated oncogenic signaling pathways. However, no tools currently exist to systematically connect such genetic lesions to therapeutic vulnerability. We have therefore developed a genomics approach to identify lesions associated with therapeutically relevant oncogene dependency. Using integrated genomic profiling, we have demonstrated that the genomes of a large panel of human non-small cell lung cancer (NSCLC) cell lines are highly representative of those of primary NSCLC tumors. Using cell-based compound screening coupled with diverse computational approaches to integrate orthogonal genomic and biochemical data sets, we identified molecular and genomic predictors of therapeutic response to clinically relevant compounds. Using this approach, we showed that v-Ki-ras2 Kirsten rat sarcoma viral oncogene homolog (KRAS) mutations confer enhanced Hsp90 dependency and validated this finding in mice with KRAS-driven lung adenocarcinoma, as these mice exhibited dramatic tumor regression when treated with an Hsp90 inhibitor. In addition, we found that cells with copy number enhancement of v-abl Abelson murine leukemia viral oncogene homolog 2 (ABL2) and ephrin receptor kinase and v-src sarcoma (Schmidt-Ruppin A-2) viral oncogene homolog (avian) (SRC) kinase family genes were exquisitely sensitive to treatment with the SRC/ABL inhibitor dasatinib, both in vitro and when it xenografted into mice. Thus, genomically annotated cell-line collections may help translate cancer genomics information into clinical practice by defining critical pathway dependencies amenable to therapeutic inhibition.

Figures

Figure 1. Genomic validation of 84 NSCLC cell lines.

(A) Chromosomal copy number changes of NSCLC cell lines are plotted against those of 371 primary NSCLC tumors. The q values (false discovery rates) for each alteration (x axis) are plotted at each genome position (y axis). Left panel shows chromosomal losses (cell lines, purple; primary tumors, dark blue); right panel shows chromosomal gains (cell lines, red; primary tumors, blue). Genomic positions corresponding to even-numbered chromosomes are shaded; dotted lines indicate centromeres; green lines, q value cutoff (0.25) for significance. Genes represent known targets of mutation in lung adenocarcinomas. Putative targets near peaks are given in parentheses. Genes identified by GISTIC using stringent filtering criteria for peak border detection are marked by asterisks. (B) Oncogene mutations present in NSCLC cell lines (black bars) are plotted according to their relative frequencies in comparison with primary lung tumors (gray bars) (–25). (C) Transcriptional profiles of primary renal cell carcinomas (orange) and corresponding cell lines (red); primary lung tumors (dark green) and lung cancer cell lines (light green); primary lymphomas (blue) and lymphoma cell lines (purple) were analyzed by hierarchical clustering. To reduce noise, probe sets were filtered prior to clustering (coefficient of variation from 1.0 through –10.0, present call rate, 20%; absolute expression greater than 100 in more than 20% of samples).

Figure 2. Robustness of phenotypic properties of EGFR-mutant lung cancer cells in vivo.

(A) The first 2 principal components (PC1 and PC2) distinguish cell lines with mutated (mut) EGFR (red dots) and WT EGFR (blue dots) (n = 54). (B) The signature (fold change greater than 2; absolute difference, 100; P < 0.01) of EGFR-mutant cell lines (n = 8/54) was used for hierarchical clustering of 123 primary adenocarcinomas (35) annotated for the presence (EGFR mut, red bars) or absence (EGFRWT, dark blue bars) of EGFR mutations. (C) Probability of survival was estimated for all 123 primary adenocarcinomas with known EGFR mutation status following grouping according to relative abundance of 337 RNA transcripts identified as differentially expressed between EGFR-mutant and EGFR WT cell lines. EGFR-mutant tumors (n = 13) were excluded from survival analyses. Survival probabilities are depicted as Kaplan-Meier survival estimate curves. (D) The same analysis was performed using 86 lung tumors from Beer et al. (37) with available survival data. Two groups were formed according to relative abundance of the EGFR mutation–specific genes, and survival analysis was performed as in D. (E) The association between presence (amplification, green; mutation, red; deletion, yellow) of genetic lesions identified in the cell lines and sensitivity of the respective cell lines to treatment with the EGFR inhibitor erlotinib was analyzed by Welch’s t test and Fisher’s exact test. Significant lesions are marked by gray (P < 0.05) or black (P < 0.0001) boxes.

Figure 3. Sensitivity profiles of compounds determined by high-throughput cell-line screening.

GI50 values (y axes) for 12 compounds are shown for the successfully screened (Supplemental Table 5) cell lines (x axes show individual cell lines). Due to the fact that rapamycin typically fails to completely abrogate cellular proliferation (79), the 25% inhibitory concentration is shown for these compounds. Bars represent GI50 (GI25 values in the case of rapamycin, y axis) throughout the cell-line collection (x axis) ranked according to sensitivity. The maximum concentration is adapted to the GI50 value (GI25 values in the case of rapamycin; 10 μM for 17-AAG, erlotinib, vandetanib, lapatinib, sunitinib, rapamycin, and PD168393; 30 μM for SU-11274, dasatinib, and purvalanol; 60 μM for VX-680; 90 μM for UO126) of resistant cell lines. The 5 most sensitive cell lines for each compound are highlighted in table form.

Figure 4. Hierarchical clustering of compound activity uncovers mutated EGFR as a target for dasatinib activity.

(A) Displayed is a hierarchical cluster of cell lines and compounds, clustered according to GI50 values (red, high compound activity; white, low compound activity) after logarithmic transformation and normalization. 77 cells reached full compound coverage. The presence (black) or absence (gray) of selected lesions is annotated in the right panel. (B) Correlation of activity of compounds to presence of amplifications (red) and deletions (blue) as well as oncogene mutations (mut) was used for hierarchical clustering. Putative target genes inside and bordering (*) the region defined by GISTIC are annotated. (C) Upper panel shows that binding mode of erlotinib (white) to WT EGFR. Dasatinib (pink) is modeled into the ATP-binding site of EGFR. The 2-amino-thiazole forms 2 hydrogen bonds with the hinge region of the kinase. Lower panel shows that the chloro-methyl-phenyl ring of dasatinib binds to a hydrophobic pocket near the gatekeeper Thr790 and helix C and will clash with the Met side chain of the EGFR drug-resistance mutation T790M. (D) Upper panel shows that Ba/F3 cells ectopically expressing mutant EGFR with (delEx19 + T790M) or without (delEx19) the T790M mutation were treated for 12 hours with the either dasatinib or erlotinib, and phospho-EGFR and EGFR levels were detected by immunoblotting. Lower panel shows that the same cells were treated for 96 hours with either dasatinib or erlotinib and viability was assessed. Growth inhibition relative to untreated cells (y axis) is shown as a function of compound concentrations.

Figure 5. KRAS mutations predict response to inhibition of Hsp90 in vitro and in vivo.

(A) The sensitive and resistant cell lines were sorted according to their GI50 values and annotated for the presence of KRAS mutations (asterisks and black columns). Bar height represents the respective GI50 values. The association of KRAS mutations and 17-AAG sensitivity (GI50 < 0.07 μM = sensitive; GI50 > 0.83 μM = resistant; according to the lower and upper 25th percentiles) was calculated by Fisher’s exact test for the lung cancer data set (upper panel) and for the NCI60 data set (lower panel). (B) Upper panel shows that whole-cell lysates of the indicated KRAS WT and KRAS mutated cell lines treated with different concentrations of 17-AAG were analyzed for levels of c-RAF, KRAS, cyclin D1, and AKT by immunoblotting. Lower panel shows that extracts of the indicated cells treated with either control (C) or 0.5 μM (H322 and Calu-6) or 1 μM (H2122) of 17-AAG were subjected to coimmunoprecipitation with antibodies to either KRAS (top) or Hsp90 (bottom); immunoconjugates were analyzed for levels of Hsp90 (top) or KRAS (bottom) by immunoblotting. Noncontiguous bands run on the same gel are separated by a black line (H2122). WB, Western blot. (C) Displayed are coronal MRI scans of lox-stop-loxKRASG12D mice before and after 7 days of treatment with either 17-DMAG or vehicle. The areas of lung tumors were manually segmented and measured on each magnetic resonance slice, and total tumor volume reduction was calculated for all mice treated with 17-DMAG (n = 4) and placebo (n = 3). SD of tumor volume in the cohort of treated and untreated mice was calculated and is depicted as error bars.

Figure 6. Identification of functionally relevant targets for dasatinib activity.

(A) Left panel shows that cell lines with copy number gain involving at least 1 gene encoding dasatinib target are labeled with asterisks and black columns. The probability of these cells being dasatinib sensitive was calculated by Fisher’s exact test. In right panel, dasatinib GI50 values are shown as box plots (representing the 25th to 75th percentile; whisker representing the 95th percentile; dots representing outliers) for cell lines with (TESP+ 1 gene) and without (TESP– 1 gene) copy number gain of dasatinib target genes (Wilcoxon test). (B) H322M cells harboring amplified SRC were either left untreated or transduced with an empty vector control (H322Mcont) or with shRNA targeting SRC (H322MSRCkd). After puromycin selection, levels of SRC in H322M cells transduced with the indicated vectors were analyzed by immunoblotting (top). The H322MSRCkd lanes were run on the same gel but were noncontiguous, as indicated by the white line. Viability was quantified by cell counting. Error bars represent SD between different experiments. (C) H322M cells were transduced with vectors encoding either active SRC or active SRC with a gatekeeper mutation SRC (T341M). Stable cells were treated with dasatinib for 96 hours. Viability is shown as percentage of untreated controls. Error bars indicate SD of 3 independent experiments. (D) Dasatinib-sensitive (TESP+; H322M) or -resistant cells (TESP–; A549) were grown s.c. in nude mice. After 14 days of treatment (vehicle, dasatinib), tumor volumes were measured as diameters. SD of tumor volume in the cohort of treated and untreated mice was calculated and is depicted as error bars.