Identification of Informative Gene Expression Signatures Indicative of Vorinostat (Suberoylanilide Hydroxamic Acid, SAHA) Exposure and Clinical Efficacy in Patients with Advanced Cutaneous T-Cell Lymphoma.

Vorinostat is a histone deacetylase inhibitor currently under evaluation in numerous oncology clinical trials. In a Phase IIb trial, oral vorinostat resulted in a 29.7% overall objective response rate in patients (pts) with advanced cutaneous T-cell lymphoma (CTCL) and had an acceptable safety profile. These results prompted efforts to identify gene expression patterns that could elucidate the molecular mechanism of action (MOA), assess exposure to vorinostat and enrich for pts who are likely to respond. In the Phase IIb trial, gene expression profiles were obtained from 24 predose and 30 postdose (2 hr postdose on Day 15) PBMC samples. The gene expression associated with Sezary burden was easily identified in predose samples and consistent with published results. Although the power of this dataset was limited for development of a predose predictor of response, we identified three biologically-relevant pathways that correlated with response and deserve further validation. First, we found a coherent cluster of proliferation/cell cycle genes to be associated with resistance to therapy. This may imply that tumor aggressiveness is an important factor for clinical response. Second, a set of antioxidant genes was upregulated in non-responders. The generation of reactive oxygen species (ROS) is a component of the vorinostat MOA and increased ROS scavenging ability may confer resistance. Finally, cytotoxic cell markers were upregulated in responders and may represent another factor associated with contribution of T and NK cells to response. Each of these 3 patterns, if confirmed, would allow for 20–50% responder enrichment. We observed robust postdose gene expression changes in which ~942 genes exhibited significant regulation (fold-change>2, P<0.01 by paired t-test between predose and postdose samples) regardless of clinical outcome. Treated samples were discriminated from untreated with 87.5% accuracy based on leave one-out-cross-validation (LOOCV) using penalized analysis of microarrays (PAM). To understand the biology, we projected the preclinical postdose signatures derived from acute postdose changes in a panel of human lymphoid cell lines. Overall, 85% of genes significantly regulated by vorinostat in lymphoid cell lines were also regulated in the same direction in PBMC samples from CTCL pts. Thus, most of the observed postdose changes result from acute vorinostat effects on gene expression. The average preclinical postdose signature can be used to predict proximal vorinostat exposure with 90% accuracy. Among the gene expression signatures observed in clinical samples but not in cell lines, two deserve special attention. First, proliferation-associated genes are downregulated postdose and are differentially expressed between responders and non-responders. It may serve as an efficacy biomarker and would allow for 80% accurate discrimination of responders from non-responders in postdose samples based on LOOCV using PAM. Second, cytokines and genes associated with the humoral immune response were downregulated at the same time genes and cytokines associated with a cytotoxic immune response were upregulated. Such changes in the Th1-Th2 balance may reflect part of the MOA for vorinostat, and may be particularly relevant to CTCL, a disease caused by Th2 type skin-homing lymphocytes. Further evaluation of vorinostat in CTCL, including additional validation of gene expression signatures that may predict response, is warranted.