Predictive Immune-Checkpoint Blockade Classifiers Identify Tumors Responding to Inhibition of PD-1 and/or CTLA-4

Abstract Purpose: Combining anti–PD-1 + anti–CTLA-4 immune-checkpoint blockade (ICB) shows improved patient benefit, but it is associated with severe immune-related adverse events and exceedingly high cost. Therefore, there is a dire need to predict which patients respond to monotherapy and which require combination ICB treatment. Experimental Design: In patient-derived melanoma xenografts (PDX), human tumor microenvironment (TME) cells were swiftly replaced by murine cells upon transplantation. Using our XenofilteR deconvolution algorithm we curated human tumor cell RNA reads, which were subsequently subtracted in silico from bulk (tumor cell + TME) patients' melanoma RNA. This produced a purely tumor cell–intrinsic signature (“InTumor”) and a signature comprising tumor cell–extrinsic RNA reads (“ExTumor”). Results: We show that whereas the InTumor signature predicts response to anti–PD-1, the ExTumor predicts anti–CTLA-4 benefit. In PDX, InTumorLO, but not InTumorHI, tumors are effectively eliminated by cytotoxic T cells. When used in conjunction, the InTumor and ExTumor signatures identify not only patients who have a substantially higher chance of responding to combination treatment than to either monotherapy, but also those who are likely to benefit little from anti–CTLA-4 on top of anti–PD-1. Conclusions: These signatures may be exploited to distinguish melanoma patients who need combination ICB blockade from those who likely benefit from either monotherapy.