King's College London

Research portal

Spectral Clustering of Microarray Data Elucidates the Roles of Microenvironment Remodeling and Immune Responses in Survival of Head and Neck Squamous Cell Carcinoma

Research output: Contribution to journalArticlepeer-review

Johanna K. Thurlow, Claudia L. Pena Murillo, Keith D. Hunter, Francesca M. Buffa, Shalini Patiar, Guy Betts, Catharine M. L. West, Adrian L. Harris, Eric K. Parkinson, Paul R. Harrison, Bradford W. Ozanne, Max Partridge, Gabriela Kalna

Original languageEnglish
Pages (from-to)2881 - 2888
Number of pages8
JournalJournal of Clinical Oncology
Issue number17
Published10 Jun 2010

King's Authors


Purpose To identify functionally related prognostic gene sets for head and neck squamous cell carcinoma (HNSCC) by unsupervised statistical analysis of microarray data. Patients and Methods Microarray analysis was performed on 14 normal oral epithelium and 71 HNSCCs from patients with outcome data. Spectral clustering (SC) analysis of the data set identified multiple vectors representing distinct aspects of gene expression heterogeneity between samples. Gene ontology (GO) analysis of vector gene lists identified gene sets significantly enriched within defined biologic pathways. The prognostic significance of these was established by Cox survival analysis. Results The most influential SC vectors were V2 and V3. V2 separated normal from tumor samples. GO analysis of V2 gene lists identified pathways with heterogeneous expression between HNSCCs, notably focal adhesion (FA)/extracellular matrix remodeling and cytokine-cytokine receptor (CR) interactions. Similar analysis of V3 gene lists identified further heterogeneity in CR pathways. V2CR genes represent an innate immune response, whereas high expression of V3CR genes represented an adaptive immune response that was not dependent on human papillomavirus status. Survival analysis demonstrated that the FA gene set was prognostic of poor outcome, whereas classification for adaptive immune response by the CR gene set was prognostic of good outcome. A combined FA&CR model dramatically exceeded the performance of current clinical classifiers (P <.001 in our cohort and, importantly, P = .007 in an independent cohort of 60 HNSCCs). Conclusion The application of SC and GO algorithms to HNSCC microarray data identified gene sets highly significant for predicting patient outcome. Further large-scale studies will establish the usefulness of these gene sets in the clinical management of HNSCC.

View graph of relations

© 2020 King's College London | Strand | London WC2R 2LS | England | United Kingdom | Tel +44 (0)20 7836 5454