Prediction of pathological response to neoadjuvant treatment in rectal cancer with a two-protein immunohistochemical score derived from stromal gene-profiling

Ann Oncol. 2017 Sep 1;28(9):2160-2168. doi: 10.1093/annonc/mdx293.

ABSTRACT

BACKGROUND: Preoperative chemoradiotherapy followed by surgical mesorectal resection is the standard of care for locally advanced rectal carcinomas. Yet, predicting that patients will respond to treatment remains an unmet clinical challenge.

EXPERIMENTAL DESIGN: Using laser-capture microdissection we isolated RNA from stroma and tumour glands from prospective pre-treatment samples (n = 15). Transcriptomic profiles were obtained hybridising PrimeView Affymetrix arrays. We modelled a carcinoma-associated fibroblast-specific genes filtering data using GSE39396.

RESULTS: The analysis of differentially expressed genes of stroma/tumour glands from responder and non-responder patients shows that most changes were associated with the stromal compartment; codifying mainly for extracellular matrix and ribosomal components. We built a carcinoma-associated fibroblast (CAF) specific classifier with genes showing changes in expression according to the tumour regression grade (FN1, COL3A1, COL1A1, MMP2 and IGFBP5). We assessed these five genes at the protein level by means of immunohistochemical staining in a patient’s cohort (n = 38). For predictive purposes we used a leave-one-out cross-validated model with a positive predictive value (PPV) of 83.3%. Random Forest identified FN1 and COL3A1 as the best predictors. Rebuilding the leave-one-out cross-validated regression model improved the classification performance with a PPV of 93.3%. An independent cohort was used for classifier validation (n = 36), achieving a PPV of 88.2%. In a multivariate analysis, the two-protein classifier proved to be the only independent predictor of response.

CONCLUSION: We developed a two-protein immunohistochemical classifier that performs well at predicting the non-response to neoadjuvant treatment in rectal cancer.

PMID:28911071 | DOI:10.1093/annonc/mdx293