Table S1: List of Enrichr Signatures collections. Table S2: Cross validation performance. Table S3: List of DNA crosslink repair genes. Table S4: Cohort Characteristics and univariate analysis. Table S5: p-values for t-tests of model scores between different clinical variables; companion to Figure S4.
ARTICLE ABSTRACT
Head and neck squamous cell carcinoma (HNSCC) is characterized by the frequent manifestation of DNA crosslink repair defects. We established novel expression-based DNA repair defect markers to determine the clinical impact of such repair defects. Using hypersensitivity to the DNA crosslinking agents, mitomycin C and olaparib, as proxies for functional DNA repair defects in a panel of 25 HNSCC cell lines, we applied machine learning to define gene expression models that predict repair defects. The expression profiles established predicted hypersensitivity to DNA-damaging agents and were associated with mutations in crosslink repair genes, as well as downregulation of DNA damage response and repair genes, in two independent datasets. The prognostic value of the repair defect prediction profiles was assessed in two retrospective cohorts with a total of 180 patients with advanced HPV-negative HNSCC, who were treated with cisplatin-based chemoradiotherapy. DNA repair defects, as predicted by the profiles, were associated with poor outcome in both patient cohorts. The poor prognosis association was particularly strong in normoxic tumor samples and was linked to an increased risk of distant metastasis. In vitro, only crosslink repair–defective HNSCC cell lines are highly migratory and invasive. This phenotype could also be induced in cells by inhibiting rad51 in repair competent and reduced by DNA-PK inhibition. In conclusion, DNA crosslink repair prediction expression profiles reveal a poor prognosis association in HNSCC.
This study uses innovative machine learning-based approaches to derive models that predict the effect of DNA repair defects on treatment outcome in HNSCC.
