IL2-STAT5 immune signatures to predict responses to PD-1 inhibition and azacitidine treatment in acute myeloid leukemia (AML): A subset analysis of a phase 2 study.

Abstract

7041 Background: Combining PD-1/PD-L1 blockade with hypomethylating agents (HMA) shows encouraging preliminary efficacy in AML, but immune features predictive of response are lacking. Methods: We treated 11 relapsed/refractory (R/R) AML patients with azacitidine (AZA) and nivolumab (Nivo) in a phase 2 clinical trial ( Daver N et al Cancer Discovery 2018). Patient characteristics: median (med) age 65 years (47-73), 63% adverse cytogenetics, 27% TP53 mutated. Pretreatment bone marrow aspirates had immune-phenotypic 17-color flow analysis and NanoString RNA quantification of 1469 immune-relevant genes. Results were correlated with clinical, pathological and molecular data. Results: The median courses of AZA+Nivo administered was 3 (range 1-17). The CR/CRi rate was 45% (including 2 CR, 1 CRi, 1 CRN and 1 CRP), with a median time to response of 1.8 months (range 0.8-4.9 months). The median overall survival was 13 months with 27% patients alive at 1 year. We found significant positive correlations between proportions of T-effector cells at baseline, and CD3+, CD8+, and T-regulatory cells at end of cycle (EOC) 1 (r > 0.75, p < 0.01 for all). At EOC2, these correlations were no longer significant. However, there was a significant positive correlation between T-effector cells at baseline and T-regulatory cells (r = 1, p < 0.001) at EOC4. Using NanoString analysis, we found 105 differentially expressed genes (fold change = 1.5, p < 0.05) between responders (5/11) and non-responders (6/11) at pretreatment. IL2-STAT5, TP53 and TNF Hallmark pathways and immune response from GO gene sets were highly enriched (q < 5x10-4) in responders. We then utilized z-score distribution analysis to quantify the degree of activation of known immunologic pathways. We found that signatures highly specific to neutrophils, NK cells, T-cells and eosinophils were significantly (p < 0.05) upregulated in patients with CR compared to non-responders at pretreatment. Conclusions: Our data demonstrates that a signature suggestive of lymphocyte activation in the pretreatment BM may be associated with augmented clinical response to PD-1 based therapies. Similar underlying pathways that have consistently predicted for responses to PD-1 inhibition in solid cancers, primarily IL2-STAT5 genes, may have predictive relevance in AML. Such pretreatment flow and NanoString signatures may help select AML patients most likely to benefit from PD-1 blockade plus HMA, further enhancing the benefit-risk ratio with such therapies.