Somatic mutation patterns and compound response in cancers

Abstract

The use of various cancer cell lines can recapitulate known tumor-associated mutations and genetically define cancer subsets. This approach also enables comparative surveys of associations between cancer mutations and drug responses. Here, we analyzed the effects of ~40,000 compounds on cancer cell lines that showed diverse mutation-dependent sensitivity profiles. Over 1,000 compounds exhibited unique sensitivity on cell lines with specific mutational genotypes, and these compounds were clustered into six different classes of mutation-oriented sensitivity. The present analysis provides new insights into the relationship between somatic mutations and selectivity response of chemicals, and these results should have applications related to predicting and optimizing thera-peutic windows for anti-cancer agents.

INTRODUCTION

Identifying the effects and mechanisms of known drugs provides perspectives for developing new cancer therapies. Approaches from systems biology and bioinformatics have been widely applied to discover new drug candidates with specific cellular activities and mechanisms, and these approaches have mainly focused on the lineage-based classification of cancer cell lines (1). Somatic mutations are important contributors to cancer progression and drug responses (2). A genotype-oriented analysis of compound response should thus be carried out using a wide variety of cancer cell lines. Accordingly, we used a new statistical method, termed Cell Line Enrichment Analysis (CLEA), to quantitatively analyze associations between genotype and drug sensitivity in cancer cell lines (2). Furthermore, this approach enabled us to measure the correlation between differentially expressed genes and mutational genotypes.

Anticancer compound screening of 60 cell lines by the National Cancer Institute (USA) (NCI60) was initiated in the late 1980s as a way to discover new drugs for leukemia. The NCI60 representing nine distinct tumor types (3): leukemia, colon, lung, CNS, renal, melanoma, ovarian, breast and prostate. The response data quantified the GI50 values for more than 40,000 chemical compounds and the results have been made available in a public database (DTP, http://dtp.cancer.gov/). The GI50 represents the compound concentration required to inhibit the growth of exposed cells to 50% of that of untreated control cells. The NCI60 panel provides many opportunities for identifying the pathways and mechanisms related to cancer at both the molecular and genetic levels (4,5). Specifically, the NCI60 panel of human tumor cell lines has been characterized at the molecular level. The analysis of RNA expression (DNA microarray data) provides unique transcriptional features for each cell line (6), and single nucleotide polymorphism data have provided estimates of DNA copy number variation at ∼120,000 sites (7). Additional types of molecular characterization of these cell lines include microRNA expression (8), DNA mutations (9), protein analysis (10), DNA methylation (11), functional target analysis (12) and the reverse phase protein array (RPPA) analysis (13). These data have been used to discover valuable relationships between compound structure, mechanism of action, cell lineage and tumor mutations, among others.

CLEA is a valuable tool, particularly for the identification of genotype-dependent compound sensitivity. Using this analytical tool, we calculated the enrichment of each compound for each genotypic category of NCI60 cell line and then attempted to generate CLEA maps to select compounds with significant sensitivity againist one of the particular genotypes. Through hierarchical clustering analysis of GI50 data against various mutational genotypes, we attempted to confirm the existence of clusters of compounds that were restricted in terms of specific genotypes. The aim of this study was to systematically identify all potential compounds exhibiting specific sensitivity to a tumor genotype, and these findings should have applications for identifying potential compounds for genotype-oriented cancer therapies.

 

RESULTS AND DISCUSSION

The frequency of mutational genotypes in NCI60 cell lines

The NCI60 cell lines have been extensively characterized at the molecular level, and mutation information for the NCI60 cell lines is publicly available through the DTP website. A total 30 different mutations are annotated for these 60 cell lines, and we calculated the frequency of individual mutations in each cell line (Table 1). In summary, TP53 showed the highest mutation frequency, i.e., 44 of 60 cell lines (74.33%) harbored the TP53 mutation. CDKN2A also showed a relative high mutation frequency, as it was detected in 35 of 60 cell lines (58.33%). However, genes such as EGFR, BRCA2 and NF2 showed a low frequency of mutation (5%) in the NCI60 cell lines, and NOTCH1, HRAS, MSH6 and VHL only showed a mutation frequency of 3.33%. Mutations in FBXW7, FLT3, PDGFRA, MAP2K4, and BRCA1 and gene amplification in KRAS, AKT2, and EGFR were just observed in one cell line (1.67%). To ascertain statistical significance in the CLEA analysis (see Methods section for greater detail), genes having mutations or amplifications in more than 3 cell lines (TP53, CDKN2A, PTEN, KRAS, RAF, PIK3CA, APC, c-MYC-amp, STK11, CTNNB1, SMAD4, RB1, MLH1, NRAS, and TN_stromal) were selected for evaluating the association with compound response.

Hierarchical clustering of genotype-specific compounds

To identify patterns of genotype-specific compound responses, we selected subsets of compounds using the GI50 profile pattern on the CLEA map. First, the −logGI50 value of 5 (i.e., GI50=10 μM) was adopted as the bipartite cutoff to determine whether a compound was sensitive (＞5) or insensitive (＜5) to any of the cell lines in the NCI60 panel (14). Second, the enrichment score (AUC value) in the CLEA analysis was used to select genotype-specific compounds (see Methods section for greater detail). An AUC value of 0.85 and a P value of 0.01 were used as cutoff values to ensure that compounds had a significant sensitivity for a particular genotype. We only included compounds that demonstrated strong potency (i.e., −logGI50 value of ＞5) against at least one cell line in the NCI60 panel. As a result, a total of 1,161 non-redundant compounds were compiled, satisfying the above-mentioned filters. Hierarchical clustering was carried out for these selected compounds against 15 genotypic categories (Fig. 1). We identified six major groups of compounds that showed unique sensitivity based on mutational genotype; these compound clusters were sensitive to genetic mutations in the KRAS, STK11, MLH1, CTNNB1 and BRAF genes, or sensitive to TN-stromal genotype. This result provides direct clues for understanding common mechanisms of action (MOA) between diverse compounds with similar applications.

Table 1.Frequency of diverse mutations in the NCI60 cell lines. In this study, 15 mutational genotypes with an occurrence in ＞3 cell lines (left column) were selected for CLEA analysis to associate mutations with compound response (GI50). c-MYC-AMP, KRAS-Amp, AKT2-Amp and EGFR-Amp represent gene amplifications

Genotype-dependent sensitivity of compounds

From each of the six clusters shown in Fig. 1, we selected one representative compound and further analyzed its genotypespecific cellular response. Neuroblastoma RAS viral oncogene homolog (NRAS) is a member of the Ras gene family and encodes 21-kDa proteins that are members of the super family of small GTP-binding proteins. NRAS has diverse intracellular signaling functions that include control of cellular proliferation, growth, and apoptosis (15). Somatic activating mutations in RAS are present in up to 30% of all human cancers (16). We found that NSC639187 (Landomycin A) belonged to the cluster of NRAS sensitivity (Fig. 2A). Three cell lines harboring NRAS mutations demonstrated superior compound responses (i.e., GI50) as compared to wild-type cell lines. Landomycin A, a natural antibiotic, is known to induce the inhibition of DNA synthesis, interference with cellular processes critical for DNA synthesis and inhibit cell cycle progression from G1/S phase to S phase (17), and we found that the cytotoxicity of this compound was, on average, ＞10-fold higher in NRAS-mutant cell lines.

STK11 (LKB1) encodes a serine-threonine kinase that directly phosphorylates, and activates AMPK, a central metabolic sensor. AMPK regulates lipid, cholesterol and glucose metabolism in specialized metabolic tissues, such as the liver, muscle and adipose tissues (18). STK11 protein is involved in two biologically important pathways that lead to cancer. First, STK11 helps to maintain a polarized epithelium, and second, STK11 activates the AMP-dependent kinase (AMPK), which controls the cellular energy balance (19). These insights into STK11 function suggest that it may represent a target of novel therapeutic strategies via its regulation of AMPK activity. Furthermore, Metformin, a widely prescribed oral hypoglycemic for diabetes, is known to activate AMPK. In the present study, NSC650914 (Phenoxan) (Fig. 2B) was identified from cluster associated with STK11 mutations, as shown in Fig. 1, and five cell lines harboring STK11 mutations showed relatively high sensitivity to Phenoxan. This compound is known to affect the mitochondrial respiratory chain in human ovarian carcinoma cell lines treated with tumor necrosis factor-alpha (TNF-α) (20), and TNF-α is known to induce insulin resistance through the AMPK pathway (21).

Fig. 1.Hierarchical clustering of the 1,161 compounds. The significance level (P value) of the enrichment score (AUC value) was used against the 15 genotypic categories for clustering. Six major groups of compounds with unique genotype-specific cellular responses, are indicated by the corresponding mutated genes shown on the left. Red color represents sensitive responses (low GI50, AUC ＞ 50) to the genotype, while the green represents resistance responses (high GI50, AUC ＜ 50).

Fig. 2.Chemical structure and genotype-specific cellular response of various compounds. (A) An NRAS mutation-specific compound and its cellular response. (B) A STK11 mutation-specific compound and its cellular response. (C) A MLH1 mutation-specific compound and its cellular response. The enrichment of the mutant cell lines over the wild-type cell lines are displayed in a −logGI50 waterfall plot. Compound specificity for the 15 mutational classes is displayed in a bar graph.

MutL homolog 1 (MLH1) is a gene commonly associated with hereditary nonpolyposis colorectal cancers (22). NSC741896 (4-(1-benzofuran-2-yl)-5H-1,2,3-dithiazole-5-thione) was defined as having MLH1 mutation-dependent activity (Fig. 2C). Four cell lines harboring MLH1 mutations showed ＞10-fold sensitivity to this compound in comparison to MLH1 wild-type cell lines. The study by Konstantinova et al. provided the first evidence supporting the in vitro antiproliferative activity of 1,2,3-dithiazoles on human breast cancer cell lines (23). Here, we found that the MOA of this compound was more related to MLH1 genotype than cell type (i.e., breast cancer cell line).

CTNNB1 is a regulator of cell adhesion and a key downstream effector in the Wnt signaling pathway. CTNNB1 has also been implicated in tumorigenesis through the phosphorylation and destabilization induced by CK1 and GSK-3beta (24). We found that NSC731431 (Amarbellisine) possessed CTNNB1-dependent sensitivity against the NCI60 cell lines (Fig. 3A). Amarbellisine was previously reported as a strictly growth inhibitory and antiproliferative molecule in a general cancer drug discovery study (25), and its strong association with CTNNB1 mutations should help to further understanding this drug’s MOA and anticancer applications.

Fig. 3.Chemical structure and genotype-specific cellular response of various compounds. (A) A CTNNB1 mutation-specific compound and its cellular response. (B) A BRAF mutation-specific compound and its cellular response. (C) A TN stromal-specific compound and its cellular response. The enrichment of the mutant cell lines over wild-type cell lines are displayed in a −logGI50 waterfall plot. Compound specificity for the 15 mutational classes is displayed in a bar graph.

BRAF is an oncogene that encodes the B-Raf protein, which is involved in intracellular signaling and cell growth. BRAF was shown to be frequently mutated in human cancers (26), and the V600E mutation of the BRAF gene has been associated with hairy cell leukemia in numerous studies (27). BRAF mutations yielded the most statistically significant associations with compound activity. NSC46061 (butanedioic acid compound with 10-[3-(4-methyl-1-piperazinyl) propyl]-2-(trifluoromethyl)-10Hphenothiazine (2：1)) was selected as a representative of the BRAF-dependent compound cluster (Fig. 3B), as it has been reported to represent a signature compound associated with mutations in the RAS-BRAF pathway (28). Additional compounds in the same cluster provide a promising resource for the development of new BRAF-specific cancer therapies.

Tenascin (TN) was previously shown to be highly expressed in cancer cells (29), and we identified NSC127716 (2-deoxy-50-azacytidine; Decitabine, Dacogen) as displaying a TN-stromal-dependent cellular response (Fig. 3C). Decitabine reactivates unmethylated p21 and, in some cases, this effect is independent of wild-type p53. Decitabine was also shown to restore the expression of Apaf1 in primaryAML cells and increase the susceptibility of bladder TCC to cisplatin. Other reports have also demonstrated that this compound potentiates p53 inducibility of NOXA, activates reexpression of p73 in AML cells and mediates cell cycle arrest in the G2/M phase via the p38 MAP kinase pathway (30,31). Thus, we believe that TN-stromal can be used as a unique marker to predict the sensitivity of cancer cells to decitabine.

NCI60 chemical screening data have been used to identify new anticancer agents and understand the MOA of anti-cancer compounds (32). Although many studies have demonstrated correlations between chemical structure and MOA or cancer lineage, the importance of cancer genotype in compound responses has not been appropriately addressed. By taking advantage of publicly available genotype data regarding the NCI60 cell lines and CLEA technology, we identified a subset of compounds exhibiting significant associations with cancer genotype in terms of their cellular response. In the CLEA analysis, hierarchical clustering of selected potent compounds revealed a total of six clusters of compounds that were sensitive to mutations in KRAS, STK11, MLH1, CTNNB1, and BRAF genes and TN-stromal genotype. Many non-NCI60 compounds in previous studies were also retrospectively validated for their genotype-specific activity (33,34). In the present study, we identified a subset of compounds with specific activity against the major cancer genotypes present in the NCI60 cell line collection. These results provide a unique resource for optimizing anticancer therapies and new drug discovery.

 

MATERIALS AND METHODS

NCI60 response database

The NCI60 response database contains more than 40,000 compounds with negative log-transformed GI50 values (−logGI50), which can be used to characterize sensitivity across the 60 cancer cell lines. Briefly, the cell lines were grown in 96-well plates and exposed to the test compound for 48 hours. Growth inhibition was expressed in terms of the GI50, i.e., the concentration required to inhibit cell growth by 50% in comparison to that in untreated controls (3). We filtered out 9,075 compounds with missing data (−logGI50 values available for less than 45 cell lines) or those that possessed a minimal level of variance across the 60 cell lines (a standard deviation less than 0.1 across the available lines). The −logGI50 values of the remaining 34,921 compounds were used for further analysis.

Statistical analysis

Cell Line Enrichment Analysis (CLEA) was designed as a statistical analysis method to associate experimental data (compound response, gene/protein expression and protein phosphorylation) with cancer genotypes of gene mutations. We previously reported the use of CLEA for associating chemical activity with other cellular parameters (2). Briefly, the prioritization of cell lines of particular genotypes for a specific compound was analyzed on a Receiver Operating Characteristic curve (ROC) plot. The Area Under the Curve (AUC) of the ROC plot was used as a measure of "sensitivity" or "resistance". The AUC score will around 50 if random enrichment and near 100 if perfect enrichment. The statistical significance (P value) for the AUC score was assigned through 1,000 permutation tests.

Software support

The 2D structures and annotations of the NCI60 compounds were displayed using MarvinSchetch developed by ChemAxon (http://www.chemaxon.com/). Hierarchical clustering was carried out using Cluster3.0 (developed by Human Genome Center, June 2002, ((35). Tree Viewer (developed by Eisen's laboratory, (36)) was used to visualize the clustered data.