Skip to main content
eScholarship
Open Access Publications from the University of California

Identification of Genomic Predictors of Response to the CDK4/6 Inhibitor Palbociclib using the UCLATORL Panel of Human Cancer Cell Lines

  • Author(s): Conklin, Dylan Francis
  • Advisor(s): Zhang, Zuo-Feng
  • Finn, Richard
  • et al.
Abstract

The landscape of genetic alterations that can lead to cancer is vast and complex. Precancerous cells accumulate mutations that affect the various molecular pathways involved in cancer progression. In the last two decades, hundreds of novel therapeutics designed to inhibit different molecular targets within these oncogenic pathways have recently entered clinical development. The success or failure of these compounds will depend on the ability to correctly identify subpopulations of cancer patients likely to be sensitive or resistant to these therapies. Patient-specific sensitivity to treatment is likely to be determined, at least in part, to the underlying genomic alterations that occurred during the development of the patient's particular cancer.

Palbociclib is a highly specific small molecule inhibitor of cyclin-dependent kinases 4 and 6 currently in clinical development by Pfizer. It is designed to inhibit the cell cycle at the G1/S transition via preventing the phosphorylation of Rb by the Cyclin D/CDK4/6 enzyme complex. Our lab previously identified the estrogen receptor positive subpopulation of breast cancer patients as distinctively likely to benefit from treatment with palbociclib. This observation spurred the initiation of a Phase II clinical trial in this patient population where remarkable efficacy was observed. Given this translational success, we wished to investigate the molecular determinants of response to palbociclib in several additional cancer types.

To this end, we assayed the in vitro sensitivity to palbociclib across a panel of 416 cancer cell lines derived from 12 distinct cancer types. We observed highly differential response to treatment both within and between cancer types. IC50s (the concentration of palbociclib required to inhibit fifty percent of population doublings) ranged from the low nanomolar range to above the highest dose tested (1�M). This response distribution was much broader than was observed in similar screens performed by the Broad and Sanger institutes, where the vast majority of cell lines assayed were listed as having IC50s above 1uM. Our ability to identify a higher proportion of palbociclib -sensitive cell lines can likely be credited to methodological innovations aimed at optimizing our screening protocol for the detection of longer-term cytostatic effects, as opposed to shorter-term cytotoxic effects of treatment. The generation of this highly differential response dataset allowed for a unique opportunity to explore the possible genetic mechanisms underlying differential sensitivity to treatment with palbociclib in vitro.

We interrogated two large genomics datasets for genotype-response associations. The first of which consisted of whole exome point mutation data downloaded from the Cancer Cell Line Encyclopedia's hybrid capture sequencing database. This data was restricted by various criteria to enrich for functional, somatic point mutations that are known to be causally involved in carcinogenesis. From this restricted dataset we identified three proto-oncogenes (MLL, TSHR, SMO) where presumptive activating mutations were associated with resistance to treatment with palbociclib across our cell line panel. We further identified five recessive cancer genes where likely loss-of-function point mutations were significantly associated with palbociclib response. Mutations in two of these genes (CDH1, TOPBP1) associated with palbociclib sensitivity, while mutations in the other three (RB1, FANCA, NBN) associated with resistance.

The other dataset interrogated for genotype-response associations was a copy number alteration dataset derived from comparative genomic hybridization arrays. This dataset was organized by gene and also restricted by various criteria to enrich for amplifications or deletions of genes likely to be causally involved in carcinogenesis. From this dataset we identified three chromosomal regions (17q12-21, 11q13, 1q32) where amplification was associated with sensitivity to palbociclib. Two amplified regions (19q13, 8q13) were found to be associated with resistance. Homozygous deletions of the 13q14 chromosomal region were found to be strongly associated with resistance to palbociclib.

Following the identification of these candidate palbociclib response biomarkers from the crude, semi-supervised screens, several post-hoc analyses were performed to strengthen the argument for causation for each of the genotype-response associations. These analyses included: (1) a comprehensive literature search to investigate the possible causal mechanism of each biomarker, (2) Pearson correlation analysis to identify inter-biomarker associations followed by multiple regression to isolate the independent effect of each variable, (3) control of confounding by cell line growth rate and histology, and (4) analysis of misclassification in the genomics datasets.

Following these extensive post-hoc analyses, we pruned our original set of fourteen biomarkers down to the eight most likely to play a causal role in determining sensitivity or resistance to treatment with palbociclib. The final set of candidate sensitivity biomarkers included: loss-of-function point mutations in CDH1, loss-of-function point mutations in TOPBP1, chromosomal amplification of 17q12-21 (ERBB2) and chromosomal amplification of 11q13 (CCND1). The final set of candidate resistance biomarkers included: activating point mutations in SMO, chromosomal amplification of CCNE1, loss-of-function point mutations in RB1 and chromosomal deletion of 13q14 (RB1).

This final set of eight candidate biomarkers was analyzed by strata representing each of the 12 cancer types in our cell line panel. The four candidate sensitivity biomarkers were sufficiently frequent and associated with sensitivity in 6 of the 12 cancer types in our panel. These were the breast, colon, head/neck, lung, ovarian and upper gastrointestinal strata. The four candidate resistance biomarkers were sufficiently frequent and associated with resistance in 6 of the 12 cancer types in our panel. These were the breast, colon, kidney, lung, ovarian and upper gastrointestinal strata.

The identified palbociclib response biomarkers represent good candidates for clinical translation, as we have shown them to be independently associated with palbociclib response in vitro and have a strong biologic rationale for causality. Follow-up experiments can further elucidate the molecular biology behind these associations and further validate these biomarkers before they are applied to the clinical setting. Ultimately, these biomarkers may be clinically implemented across a wide range of cancer types to identify patient subpopulations most likely to benefit from treatment with palbociclib.

Main Content
Current View