Identification and validation of a machine learning model of complete response to radiation in rectal cancer reveals immune infiltrate and TGFβ as key predictors

Abstract

Background: It is uncertain which biological features underpin the response of rectal cancer (RC) to radiotherapy. No biomarker is currently in clinical use to select patients for treatment modifications.

Methods: We identified two cohorts of patients (total N = 249) with RC treated with neoadjuvant radiotherapy (45Gy/25) plus fluoropyrimidine. This discovery set included 57 cases with pathological complete response (pCR) to chemoradiotherapy (23%). Pre-treatment cancer biopsies were assessed using transcriptome-wide mRNA expression and targeted DNA sequencing for copy number and driver mutations. Biological candidate and machine learning (ML) approaches were used to identify predictors of pCR to radiotherapy independent of tumour stage. Findings were assessed in 107 cases from an independent validation set (GSE87211).

Findings: Three gene expression sets showed significant independent associations with pCR: Fibroblast-TGFβ Response Signature (F-TBRS) with radioresistance; and cytotoxic lymphocyte (CL) expression signature and consensus molecular subtype CMS1 with radiosensitivity. These associations were replicated in the validation cohort. In parallel, a gradient boosting machine model comprising the expression of 33 genes generated in the discovery cohort showed high performance in GSE87211 with 90% sensitivity, 86% specificity. Biological and ML signatures indicated similar mechanisms underlying radiation response, and showed better AUC and p-values than published transcriptomic signatures of radiation response in RC.

Interpretation: RCs responding completely to chemoradiotherapy (CRT) have biological characteristics of immune response and absence of immune inhibitory TGFβ signalling. These tumours may be identified with a potential biomarker based on a 33 gene expression signature. This could help select patients likely to respond to treatment with a primary radiotherapy approach as for anal cancer. Conversely, those with predicted radioresistance may be candidates for clinical trials evaluating addition of immune-oncology agents and stromal TGFβ signalling inhibition.

Funding: The Stratification in Colorectal Cancer Consortium (S:CORT) was funded by the Medical Research Council and Cancer Research UK (MR/M016587/1).

Keywords: Genes; Immune response; Precision medicine; Prediction; Radiotherapy; Rectal neoplasms; TGFβ.

Fig. 1

Clinical and molecular data in Grampian, Aristotle and GSE87211. a. Overlap of samples profiled for RNA and DNA platforms. b. Comparison of the main clinical and molecular profiles by cohort. c. Most common driver mutations by gene. d. Copy number alterations by chromosome arm. (note RNA data only available from GSE87211).

Fig. 2

Candidate analysis. a. Univariable regression adjusted for T and N stage for prediction of pCR in candidate biological features in discovery cohort (Grampian and Aristotle combined). b. Multivariable model adjusted for T and N stage after stepwise backwards regression in discovery cohort. c. ROC curve applying the 3 variables (F-TBRS, Cytotoxic lymphocytes and CMS1) as one compound variable in discovery cohort. d. Univariable model of the 3 variables (F-TBRS, Cytotoxic lymphocytes and CMS1) adjusted for T and N stage in GSE87211. e. ROC curve applying the 3 BRSC variables combined in GSE87211.

Fig. 3

Machine learning. a. Analytical pipeline used to derive and validate a new signature to predict pCR. b. 10-fold cross validation accuracy and number of genes for each ML method tested. The one with highest accuracy was selected. EN: elastic net, GBM: gradient boosting machine, LR: lasso regression, NN: neural net; RF: random Forest; SVM: support-vector machine. The prefix F- refers to functional. c. RSS predictive scores from Gradient boosting machine model on GSE87211. d. ROC curve for Gradient boosting machine model on GSE87211. e. Confusion matrix for the new RSS signature in GSE87211.

Fig. 4

RSS biology a. GSEA for cancer hallmarks from differential expression analyses based on pCR and RSS prediction in GSE87211. b. Association of BRSC molecular features and RSS genes by meta-analysis for all genes in discovery cohort, GSE87211 and TCGA separately (see Supplementary Figure S8 for full analysis).

Fig. 5

Comparison of our signatures with published ones. a. Number of overlapping entrez genes in published signatures and ours. b. Pearson correlation across different signatures in GSE87211. c. Heatmap of samples ranked by each signature sorted by biological score of RSS in GSE87211. d. Prediction to pCR in GSE87211 by each signature sorted by AUC.

Fig. 6

Clinical implementation of RSS. NCCN guidelines define clinical options for patients with rectal cancer. Adding a certified test based on the 33 gene signature described here would inform decision making, resulting in increased organ preservation, reduced treatment morbidity, and has potential to improve outcomes.