Using a stem cell-based signature to guide therapeutic selection in cancer

Abstract

Given the very substantial heterogeneity of most human cancers, it is likely that most cancer therapeutics will be active in only a small fraction of any population of patients. As such, the development of new therapeutics, coupled with methods to match a therapy with the individual patient, will be critical to achieving significant gains in disease outcome. One such opportunity is the use of expression signatures to identify key oncogenic phenotypes that can serve not only as biomarkers but also as a means of identifying therapeutic compounds that might specifically target these phenotypes. Given the potential importance of targeting tumors exhibiting a stem-like phenotype, we have developed an expression signature that reflects common biological aspects of various stem-like characteristics. The consensus stemness ranking (CSR) signature is upregulated in cancer stem cell-enriched samples at advanced tumor stages and is associated with poor prognosis in multiple cancer types. Using two independent computational approaches we utilized the CSR signature to identify clinically useful compounds that could target the CSR phenotype. In vitro assays confirmed selectivity of several predicted compounds including topoisomerase inhibitors and resveratrol towards breast cancer cell lines that exhibit a high-CSR phenotype. Importantly, the CSR signature could predict clinical response of breast cancer patients to a neoadjuvant regimen that included a CSR-specific agent. Collectively, these results suggest therapeutic opportunities to target the CSR phenotype in a relevant cohort of cancer patients.

Figure 1. Development of CSR signature

A. 1143 breast cancer samples were ranked using microarray expression data to measure the degree to which a tumor sample exhibits stem-like characteristics using four independent methods: “core ES”(4), “ES exp1” (2), “iPS”(6), and “IGS”-(3).See text for details. Low rankings (high stemness) shown in blue and high rankings (low stemness) shown in red. Samples that were ranked high (or low) by the consensus of all four ranking methods (boxed) represent the CSR training set. B. Image intensity display of the expression levels of genes comprising the CSR signature. Expression levels are standardized to zero mean and unit variance across samples, displayed with genes as rows and samples as columns, and color coded to indicate high (red) or low (blue) expression levels. C. CSR signature predictions in separated CD133+ and CD133-fractions from two human glioblastoma xenograft tumors. Separated cells were grown with or without laminin (Lam). D. Percentage of CD44+/CD24− /ESA+cells in 14 breast cancer cell lines was determined by flow cytometry. Each dot represents mean of at least three independent experiments. Cell lines are grouped as a function of CSR signature predictions following microarray expression profiling of unseparated cultures. Mann-Whitney test was used to calculate the p-value for the significance of difference between high-and low-CSR groups using values from all individual experiments.

Figure 2. High-CSR phenotype associates with poor survival in diverse cancer types

CSR phenotype was determined using binary regression with the CSR signature. Samples were divided into two cohorts using CSR probability of P=0.5 as a cutoff. Survival curve of patients, tumors of which displayed high CSR probability (P>0.5) is shown in red and survival curve of patients, tumors of which displayed low CSR probability (P<0.5) is shown in green. A. Survival analysis (censored at 10 years)in a dataset of 256 patients with estrogen receptor positive breast tumors (16). B. Survival analysis (censored at 10 years) of 60patients with grade 1 estrogen receptor positive breast tumors (16). C. Survival analysis of 274 lung adenocarcinoma patients (censored at 60 months) (7). D. Survival analysis of 47 medulloblastoma patients (censored at 6years)(17).

Figure 3. Specificity of predicted drugs towards high-CSR breast cancer cell lines

Twelve breast cancer cell lines were treated with the indicated drugs for three days. The viability of cells was assessed by MTS colorimetric assay. Cell lines were divided to two categories of high CSR (CSR>0.5-dark grey) and low CSR (CSR<0.5- light grey) as determined by expression microarrays. Cytotoxic effect was expressed as a percentage of the reading taken from a parallel plate at the time of drug application (% of day 0). Averages of duplicate wells from 2–4 independent experiments were pooled together according to the CSR category. The box shows 25^th percentile to 75^th percentile with a line at the median. The whiskers indicate the highest and the lowest values.

Figure 4. Clinical significance of CSR signature

CSR signature predictions in 132 breast cancer patients treated with a combination of Taxol, 5FU, doxorubicin (or epirubicin)and cyclophosphamide (TFAC) in the neoadjuvant setting (19) stratified by clinical endpoint. The box shows 25^th percentile to 75^th percentile with a line at the median. The whiskers show the highest and the lowest values. Patients that achieved a pathological complete response (pCR) to TFAC had a significantly higher CSR signature score than patients that did not respond to TFAC (p<0.0001 by two tailed Mann Whitney test).