Use of genomic signatures to predict response to histone deacetylase inhibitors and correlation with tumor subtypes and molecular pathways.

Abstract

e13596 Background: Histone deacetylase (HDAC) inhibitors are a promising class of cancer drugs. The ability to predict which tumors are likely to respond helps in planning clinical trials and delivering treatment. Methods: We developed gene expression signatures using cell lines before and after treatment with valproic acid (VPA), trichostatin A (TSA), and suberoylanilide hydroxamic acid (SAHA). Gene expression signatures for pathway activation were developed using induction of genes regulating epigenetic processes (HDAC1, HDAC4, SIRT1, DNMT2, and EZH2). Using logistic regression we predicted drug sensitivity and epigenetic signaling pathways in subtypes of cancers using data from 32 public datasets. Prediction of pathway activity and drug responsiveness was validated using human tumor cells grown in 3D culture or mouse xenografts. Results: Pathway activation patterns and drug sensitivity vary across tumor types and subtypes. Thyroid and GI cancers and mesothelioma are predicted to have highest response to HDAC inhibitors (p<0.0001). There were significant differences in HDAC pathway activity and drug response within tumor subtypes. For example, glioblastoma, particularly Freije class HC2A and HC2B, are more likely to be sensitive to HDAC inhibitors than lower grade gliomas, which have lower HDAC4 and EZH2 activation (p<0.0001). Basal and HER2-positive breast tumors have higher HDAC4 activity and sensitivity to HDAC inhibitors than ER-positive tumors (p<0.0005). We saw correlation between components of the epigenetic machinery. HDAC inhibitor sensitivity correlates with MYC and EZH2 activation, particularly in ovarian tumors, mesotheliomas, basal breast cancer, and gliomas (p<0.001). Preclinical drug studies in cell lines and patient tumors grown in 3D culture and orthotopic models validate our genomic analyses. Conclusions: Epigenetic pathway activity and sensitivity to epigenetic modifiers vary across tumor types and subtypes. High-throughput expression profiling can impact cancer drug selection. Pre-identification of patient response may improve response rates and assist in identifying optimal inclusion criteria for clinical trials. No significant financial relationships to disclose.