bioRxiv2020
Although high clinical response rates are seen for immune checkpoint blockade (ICB) treatment of metastatic melanomas, both intrinsic and acquired ICB resistance remain considerable clinical challenges1. Combination ICB (anti-PD-1 + anti-CTLA-4) shows improved patient benefit2-5, but is associated with severe adverse events and exceedingly high cost. Therefore, there is a dire need to stratify individual patients for their likelihood of responding to either anti-PD-1 or anti-CTLA-4 monotherapy, or the combination. Since it is conceivable that ICB responses are influenced by both tumor cell-intrinsic and -extrinsic factors6-9, we hypothesized that a predictive genetic classifier ought to mirror both these features. In a panel of patient-derived melanoma xenografts10 (PDX), we noted that cells derived from the human tumor microenvironment (TME) that were co-grafted with the tumor cells were naturally replaced by murine cells after the first passage. Taking advantage of the XenofilteR11 algorithm we recently developed to deconvolute human from murine RNA sequence reads from PDX10, we obtained curated human melanoma tumor cell RNA reads. These expression signals were computationally subtracted from the total RNA profiles in bulk (tumor cell + TME) melanomas from patients. We thus derived one genetic signature that is purely tumor cell-intrinsic ("InTumor"), and one that comprises tumor cell-extrinsic RNA profiles ("ExTumor"). Here we report that the InTumor signature predicts patient response to anti-PD-1, but not anti-CTLA-4 treatment. This was validated in melanoma PDX and cell lines, which confirmed that InTumorLO tumors were effectively eliminated by adoptive cell transfer of T-Cell Receptor (TCR)-matched cytotoxic T cells, whereas InTumorHI melanomas were refractory and grew out as fast as tumors challenged with unmatched T cells. In contrast, the ExTumor signature predicts patient response to anti-CTLA-4 but not anti-PD-1. Most importantly, we used the InTumor and ExTumor signatures in conjunction to generate an ICB response quadrant, which predicts clinical benefit for five independent melanoma patient cohorts treated with either mono- or combination ICB. Specifically, these signatures enable identification of patients who have a much higher chance of responding to the combination treatment than to either monotherapy (p < 0.05), as well as patients who are likely to experience little benefit from receiving anti-CTLA-4 on top of anti-PD-1 (p < 0.05). These signatures may be clinically exploited to distinguish patients who need combined PD-1 + CTLA-4 blockade from those who are likely to benefit from either anti-CTLA-4 or anti-PD-1 monotherapy.