Leveraging big data of immune checkpoint blockade response identifies novel potential targets

Abstract

Background: The development of immune checkpoint blockade (ICB) has changed the way we treat various cancers. While ICB produces durable survival benefits in a number of malignancies, a large proportion of treated patients do not derive clinical benefit. Recent clinical profiling studies have shed light on molecular features and mechanisms that modulate response to ICB. Nevertheless, none of these identified molecular features were investigated in large enough cohorts to be of clinical value.

Materials and methods: Literature review was carried out to identify relevant studies including clinical dataset of patients treated with ICB [anti-programmed cell death protein 1 (PD-1)/programmed death-ligand 1 (PD-L1), anti-cytotoxic T-lymphocyte antigen 4 (CTLA-4) or the combination] and available sequencing data. Tumor mutational burden (TMB) and 37 previously reported gene expression (GE) signatures were computed with respect to the original publication. Biomarker association with ICB response (IR) and survival (progression-free survival/overall survival) was investigated separately within each study and combined together for meta-analysis.

Results: We carried out a comparative meta-analysis of genomic and transcriptomic biomarkers of IRs in over 3600 patients across 12 tumor types and implemented an open-source web application (predictIO.ca) for exploration. TMB and 21/37 gene signatures were predictive of IRs across tumor types. We next developed a de novo GE signature (PredictIO) from our pan-cancer analysis and demonstrated its superior predictive value over other biomarkers. To identify novel targets, we computed the T-cell dysfunction score for each gene within PredictIO and their ability to predict dual PD-1/CTLA-4 blockade in mice. Two genes, F2RL1 (encoding protease-activated receptor-2) and RBFOX2 (encoding RNA-binding motif protein 9), were concurrently associated with worse ICB clinical outcomes, T-cell dysfunction in ICB-naive patients and resistance to dual PD-1/CTLA-4 blockade in preclinical models.

Conclusion: Our study highlights the potential of large-scale meta-analyses in identifying novel biomarkers and potential therapeutic targets for cancer immunotherapy.

Keywords: biomarker; immunotherapy; machine learning; meta-analysis; scientific software; transcriptomic.