In the past few decades, it has become clear that tumors from the same anatomic site aren't a single disease, but rather come in many different forms, much like how a respiratory virus, a bacterial sinus infection, and seasonal allergies can all produce the same runny nose and congestion. Treating the symptoms without knowing the underlying cause means some patients get the appropriate medicine and others don't. Characterizing what's driving a tumor's growth underneath the surface is essential to knowing how to treat it.
Head and neck squamous cell carcinoma (HNSCC) is a common and deadly cancer with roughly 600,000 new cases a year. Some HNSCC tumors are caused by human papillomavirus (HPV) infection, while others arise independently. The latter have worse prognosis, with a five-year survival of roughly 50% despite aggressive multimodal therapy. Patients can be grouped into broad molecular subtypes like "basal" or "classical," but those labels are mostly descriptive and don’t inform physicians on therapeutic directions. A new study in npj Precision Oncology led by Vaz and colleagues in the Gujral lab at Fred Hutch tackles this gap by further subcategorizing HNSCC tumors and investigating whether these updated categorizations can inform tumor vulnerabilities.
Lead author Joel Vaz explains the importance of this work: "Head and neck cancers that look similar histologically can be driven by very different biological programs, which may explain why patients with the same diagnosis respond differently to treatment. Our work moves beyond simply describing these differences by connecting distinct tumor states to their specific dependencies and therapeutic vulnerabilities."
The team combined RNA-sequencing data from five independent cohorts, assessing a total of 727 tumors, one of the largest unified HNSCC transcriptomic datasets assembled to date. After correcting for batch effects across datasets, unsupervised clustering split the tumors into nine molecular groups: two HPV-positive, six HPV-negative, and 1 non-tumor cluster. Layering on mutation data and clinical outcomes confirmed these clusters tracked known biology, such as TP53 mutations, loss of the tumor suppressor CDKN2A, and survival differences. The six HPV-negative clusters grouped into four functional subtypes: cell cycle/proliferative (characterized by high expression of the oncogene MYC and MET/FAK signaling), well-differentiated epithelial, mitochondrial metabolism (oxidative phosphorylation-driven), and epithelial-mesenchymal-transition (EMT)/mesenchymal (with immunosuppressive and invasive characteristics). Notably, the classic "basal" subtype separated into two distinct subsets: the cell cycle/proliferative and the well-differentiated epithelial state. The "classical" subtype split into two different mitochondrial metabolism states. The mesenchymal class corresponded with the EMT/mesenchymal state.
To assess whether these refined transcriptional subtypes are linked to specific genetic vulnerabilities, the team assessed genome-scale CRISPR dependency data from the DepMap database for HNSCC cell lines representative of each subtype. DepMap quantifies the necessity of genes for cell survival. This analysis revealed that tumor state relies on a distinct set of essential genes: the cell cycle/proliferative subtype depends on mitosis and autophagy related genes; the well-differentiated epithelial subtype on ERBB2/PI3K signaling and chromatin organization; the mitochondrial metabolism subtype on oxidative phosphorylation; and the EMT/mesenchymal subtype on G2/M checkpoint control, glycolysis, and integrin/Notch signaling.
To determine drug sensitivities for the subtypes, they next assessed published pharmacologic screening data. They found differing pharmacologic dependencies for each subtype, largely aligning with the genetic dependencies observed. For instance, mitochondrial metabolism tumors were selectively vulnerable to inhibition of metabolic stress and redox pathways such as with CHK1 and TOP1 inhibitors. While the EMT, cell cycle/proliferative, and well-differentiated epithelial subtypes were all sensitive to inhibitors of the cell surface receptor epidermal growth factor receptor (EGFR), each responded most strongly to a different specific EGFR inhibitor.
EGFR is a clinically actionable target that is commonly expressed in HNSCC. Thus, the team then set out to build a model to predict which patients would respond to EGFR inhibitors. They built a support vector regression machine-learning model trained on response data to the EGFR inhibitor erlotinib from HNSCC cell lines with available gene expression. They narrowed down to a 13-gene signature, combining genes associated with sensitivity (like CCND2 and TEAD3) and resistance (like TYRP1 and TPM2). Biologically, the signature mapped cleanly onto tumor state: sensitivity genes were enriched in the well-differentiated subtype, resistance genes in the metabolic subtype.
The researchers then put their model to the test on patient-derived 3D "microtumors," lab-grown mini-tumors built directly from fresh surgical tissue that preserve the native tumor microenvironment. The model accurately predicted erlotinib response in these six independent patient samples, outperforming EGFR expression level alone as a predictor.
Vaz frames the work as a foundation rather than a finish line: "This study gives us a roadmap, but it also raises new questions. Can these tumor states help us predict which treatment will work best for each patient? And can we stop tumors from switching into more aggressive or treatment-resistant states? That's what we're working on next."
By linking transcriptomic subtype, functional genetic dependencies and preclinical drug responses validation, this study offers a template not just for HNSCC but for other cancers where molecular subtypes exist but haven't yet translated into treatment decisions.
Fred Hutch/University of Washington/Seattle Children’s Cancer Consortium Members Drs. Lauran Shih, Emily Marchiano, Slobodan Beronja, Bruce Clurman, Cristina Rodriguez, Brittany Barber and Taran Gujral contributed to this research.
The spotlighted research was funded by the Kuni Foundation, the National Cancer Institute, National Institutes of Health support for the Fred Hutch/University of Washington/Seattle Children’s Cancer Consortium, and M.J. Murdock Charitable Trust support for the Experimental Histopathology Shared Resource.
Vaz JM, Zhu S, Useche M, Shih L, Marchiano E, Beronja S, Clurman BE, Rodriguez C, Barber B, Chan M, Gujral TS. 2026. Transcriptional states define dependencies and therapeutic vulnerabilities in head and neck cancer. npj Precis. Onc. https://doi.org/10.1038/s41698-026-01509-8.
Science Spotlight writer Kelly Mitchell is a postdoctoral fellow in the Paddison Lab at Fred Hutch Cancer Center. She utilizes live cell reporters and CRISPR screening to study how glioblastoma cancer cells resist chemotherapy and radiation treatment. She obtained her PhD in cellular biology from Albert Einstein College of Medicine.