Abstract
The epidermal growth factor receptor (EGFR) inhibitor cetuximab is the only FDA-approved oncogene-targeting therapy for head and neck squamous cell carcinoma (HNSCC). Despite variable treatment response, no biomarkers exist to stratify patients for cetuximab therapy in HNSCC. Here, we applied unbiased hierarchical clustering to reverse-phase protein array molecular profiles from patient-derived xenograft (PDX) tumors and revealed 2 PDX clusters defined by protein networks associated with EGFR inhibitor resistance. In vivo validation revealed unbiased clustering to classify PDX tumors according to cetuximab response with 88% accuracy. Next, a support vector machine classifier algorithm identified a minimalist biomarker signature consisting of 8 proteins - caveolin-1, Sox-2, AXL, STING, Brd4, claudin-7, connexin-43, and fibronectin - with expression that strongly predicted cetuximab response in PDXs using either protein or mRNA. A combination of caveolin-1 and Sox-2 protein levels was sufficient to maintain high predictive accuracy, which we validated in tumor samples from patients with HNSCC with known clinical response to cetuximab. These results support further investigation into the combined use of caveolin-1 and Sox-2 as predictive biomarkers for cetuximab response in the clinic.