Combining imaging- and gene-based hypoxia biomarkers in cervical cancer improves prediction of treatment failure independent of intratumor heterogeneity

Emerging biomarkers based on medical images and molecular characterization of tumor biopsies open up for combining the two disciplines and exploiting their synergy in treatment planning. We compared pretreatment classification of cervical cancer patients by two previously validated imaging- and gene-based hypoxia biomarkers, evaluated the influence of intratumor heterogeneity, and investigated the benefit of combining them in prediction of treatment failure. The imaging-based biomarker was hypoxic fraction, determined from diagnostic dynamic contrast enhanced (DCE)-MR images. The gene-based biomarker was a hypoxia gene expression signature determined from tumor biopsies. Paired data were available for 118 patients. Intratumor heterogeneity was assessed by variance analysis of MR images and multiple biopsies from the same tumor. The two biomarkers were combined using a dimension-reduction procedure. The biomarkers classified 75% of the tumors with the same hypoxia status. Both intratumor heterogeneity and distribution pattern of hypoxia from imaging were unrelated to inconsistent classification by the two biomarkers, and the hypoxia status of the slice covering the biopsy region was representative of the whole tumor. Hypoxia by genes was independent on tumor cell fraction and showed minor heterogeneity across multiple biopsies in 9 tumors. This suggested that the two biomarkers could contain complementary biological information. Combination of the biomarkers into a composite score led to improved prediction of treatment failure (HR:7.3) compared to imaging (HR:3.8) and genes (HR:3.0) and prognostic impact in multivariate analysis with clinical variables. In conclusion, combining imaging- and gene-based biomarkers enables more precise and informative assessment of hypoxia-related treatment resistance in cervical cancer, independent of intratumor heterogeneity.

142

Fjeldbo CS, Hompland T, Hillestad T, Aarnes EK, Günther CC, Kristensen GB, Malinen E, Lyng H