Purpose: To identify a predictive biomarker for durvalumab, an anti–programmed death ligand 1 (PD-L1) mAb.

Experimental Design: RNA sequencing of 97 advanced-stage non–small cell lung carcinoma (NSCLC) biopsies from a nonrandomized phase Ib/II clinical trial (1108/NCT01693562) were profiled to identify a predictive signature; 62 locally advanced or metastatic urothelial cancer tumors from the same study were profiled to confirm predictive utility of the signature. Thirty NSCLC patients provided pre- and posttreatment tumors for messenger RNA (mRNA) analysis. NSCLC with ≥25% tumor cells and urothelial cancer with ≥25% tumor or immune cells stained for PD-L1 at any intensity were scored PD-L1 positive (PD-L1⁺). Kaplan–Meier and Cox proportional hazards analyses were used to adjust for gender, age, prior therapies, histology, ECOG status, liver metastasis, and smoking. Tumor mutation burden (TMB) was calculated using data from The Cancer Genome Atlas (TCGA).

Results: In the NSCLC discovery set, a four-gene IFNγ-positive (IFNγ⁺) signature comprising IFNγ, CD274, LAG3, and CXCL9 was associated with higher overall response rates, longer median progression-free survival, and overall survival compared with signature-low patients. IFNγ⁺-signature NSCLC patients had improved survival regardless of IHC PD-L1 status. These associations were replicated in a urothelial cancer cohort. The IFNγ⁺ signature was induced 2-fold (P = 0.003) by durvalumab after 8 weeks of therapy in patients with NSCLC, and baseline signature was associated with TMB but not survival in TCGA data.

Conclusions: The IFNγ⁺ mRNA signature may assist in identifying patients with improved outcomes with durvalumab, independent of PD-L1 assessed by IHC. Clin Cancer Res; 24(16); 3857–66. ©2018 AACR.