A Three-Gene Immune Signature Including IDO1, BIN1 and PLXNC1 Predicts Survival in Acute Myeloid Leukemia

Introduction A large body of evidence has increasingly demonstrated that the immune tumor microenvironment (TME) critically contributes to acute myeloid leukemia (AML) development. However, current AML prognostic classifications only rely on leukemia cell-intrinsic alterations and do not incorporate immunological markers referring to TME. Indoleamine 2,3-dioxygenase 1 (IDO1), which is negatively regulated by the BIN1 proto-oncogene, is a central mediator of immune tolerance in the AML TME. This study aimed to identify IDO1-interacting genes in the AML TME and to develop a prognostic immune gene signature.

Methods Biological and clinical data of 732 patients with de novo AML treated with curative intent were retrieved from public TCGA and HOVON datasets. Patients >= 65 years were excluded from survival analyses. Co-expression analysis was performed through cBioPortal on TCGA data aiming at discovering new IDO1-interacting genes. Cox regression analysis was used to identify most survival-predicting genes in order to generate a prognostic score. Differential expression (DE) analysis was performed using the nSolver software package (NanoString Technologies, Seattle, WA).

Results BIN1 and IDO1 expression were negatively correlated in HOVON cases (P<0.0001, Figure 1A). Further co-expression analyses of RNA-sequencing data from TCGA allowed us to identify PLXNC1, a semaphorin receptor involved in inflammation and immune response, as an IDO1-interacting gene and a strong predictor of survival (P<0.05, Figure 1B). PLXNC1 was negatively correlated with IDO1 in the HOVON dataset (P<0.0001, Figure 1C). The IDO1-BIN1-PLXNC1 immune gene signature predicted AML survival both in HOVON (P<0.0001, Figure 1D) and in TCGA (P=0.001) cases. We, then, sought to identify commonalities between DE genes in IDO1^low versus IDO1^high TCGA cases and in PLXNC1^low versus PLXNC1^high TCGA cases. Interestingly, CXCR2 was the only shared gene, suggesting the IDO1 and PLXNC1 overexpression largely reflected non-redundant gene expression programs. Results from enrichment analysis based on the top DE genes between PLXNC1^low and PLXNC1^high cases revealed the major involvement of PLXNC1 in viral and immune response-related pathways whereas the same analysis performed using the top DE genes between IDO1^low and IDO1^high cases confirmed the IDO1 key role in interferon signaling pathways. Furthermore, we identified IKBKB, FOSL1 and TLR9 as the top DE genes between IDO1^low and IDO1^high cases, whereas GZMH, GNLY, IFIT2 and IFIT3 were the top DE genes between PLXNC1^low and PLXNC1^high cases. We then identified TCGA cases with amplification, up-regulation or down-regulation in at least one of the previously reported seven genes. The group of TCGA cases with alterations showed higher frequency of both TP53 (18% vs 5%, P<0,05) and KRAS mutations (11% vs 2%, P<0,05) compared to cases without alterations (Figure 1E). Conversely, CEBPA mutations were prevalent in TCGA patients without alterations than in ones with alterations (12% vs 0%, P<0,05). Finally, the top DE genes between IDO1^low and IDO1^high and between PLXNC1^low and PLXNC1^high cases, when considered in aggregate, provided a significant survival prediction (P<0.05, HR=1.6, Figure 1F).

Conclusions Our data identify PLXNC1 as a novel IDO1-correlated gene. A three-gene immune signature that includes PLXCN1, IDO1 and BIN1 strongly predicted clinical outcome in large AML cohorts. Moreover, IDO1 and PLXNC1 expression-based DE analysis generated an immunological signature highly predictive of prognosis. In light of the emerging role of immunotherapies for AML, our findings support the incorporation of TME-associated immune biomarkers into current AML classification and prognostication algorithms.