Abstract
Background:
Immune checkpoint inhibition (ICI) has become a standard treatment to re-invigorate tumor-attacking T cell responses in multiple cancer indications, yet a patient's response is unpredictable even with a confirmed expression of the relevant targets such as PD-1 or PD-L1. Previously identified biomarkers of response have relatively low accuracy, making it difficult to reliably employ them as predictors of clinical response.
Methods:
We comprehensively phenotyped peripheral blood CD8+ T cells from patients with non-small cell lung cancer by analyzing surface marker expression, transcriptome, and TCR repertoire with single-cell sequencing technology. The cohorts were comprised of patients who (a) responded to anti-PD(L)1 treatment for a prolonged period of time (b) were new-on-treatment responders, and (c) were new-on-treatment nonresponders. Using various bioinformatics analyses, we defined the signatures of ICI response and evaluated their performance on external scRNA-seq datasets.
Results:
We identified response-specific signals in cell type and cell state proportions as well as in TCR repertoire diversity and TCR inter-donor similarity. The enrichment analysis revealed several pathways and regulatory modules enriched in different response groups. Using machine learning, we identified cell-type-specific signatures that predicted the ICI response with an accuracy between 66% and 93% at the single cell level and up to 94% at the patient level. Effector memory CD8+ T cells in long-term responders were most predictive of response, and the inferred effector memory signature could be successfully applied to two related scRNA-seq datasets. CD44, GIMAP4, CD69, and CCL4L2 were among the most relevant contributing markers defining the predictive ML signatures on lung cancer samples.
Conclusion:
Our findings suggest that CD8+ T cell subset-specific models reach an accuracy that possesses the potential to inform treatment decisions in a clinical setting.