Acute myeloid leukemia (AML) is one of the most common hematological cancers in adults, characterized by the accumulation of immature myeloblasts in the bone marrow and peripheral blood at the expense of normal blood components (1). Unlike many other cancers, AML has a low tumor mutation burden (TMB) with an average of 10–13 coding mutations per patient (2). Although we have understood the role of mutational genes in driving tumor progression along with an uncomplicated mutational landscape, the overall therapeutic strategy for AML patients has remained the same for the last 30 years (3, 4). The conventional treatment paradigm has a restricted contribution to improve overall survival (OS), especially in the elderly population (4). Although allogeneic hematopoietic cell transplantation (alloHCT) and chemotherapy regimens allowed a five-year survival rate of 40–70% in younger patients (<40 years of age), survival in the elderly remains poor (5, 6) with a risk of relapse within 5 years from diagnosis as high as 75% (7). Therefore, it is urgent to identify potential biomarkers to inform prognosis and treatment allocation in this setting.

It has been well proved that allogeneic hematopoietic cell transplantation (alloHCT) is successful in treating AML and AML is immune-responsive (8). However, immunosuppression could also be caused by AML blasts, leading to paradoxical immunosuppression in patients with AML (9). Therefore, further understanding of how immune cells battle with AML blasts could lead to more effective therapies for AML.

Previous studies (10–12) have stratified AML patients into low- and high-risk groups based on immune and stromal scores with the ESTIMATE algorithm (13), which is based on single sample Gene Set Enrichment Analysis and then generates stromal and immune scores to predict the infiltration of stromal and immune cells in tumors. The identified differentially expressed genes (DEGs) from low- and high-risk groups were then studied to look for potential prognostic value in AML patients. However, there is a lack of a comprehensive study on the utility of immune-related gene (IRG) expression in predicting AML prognosis and comparing AML with healthy samples. This study aimed to fill the gap by focusing on the relationship between IRG expression and AML patients’ prognosis.

Genetic and clinical factors play increasingly important roles in predicting OS and event-free survival (EFS) for AML patients (16). Patient-related factors, AML-related genetic factors and MRD monitoring are considered as key factors responsible for AML prognosis (17). In the past, immune factors have been largely ignored. This study constructed an immune prognostic signature for AML based on the TCGA-LAML and GTEx-whole blood cohort, which improved the accuracy of ELN2017 risk stratification system.

Among the 17 IRGs involved in the immune-17 signature, we roughly divided them into three categories according to genes’ function: 1) innate immunity-related genes; 2) specific immunity-related genes; 3) endocrine-related genes. APOBEC3G, MIF, MX1, ISG20, MPO, FGR, and IL1R2 are innate immunity-related genes. APOBEC3G is highly expressed in various cancers and plays an essential role in regulating tumor growth and innate immune responses (18, 19). MIF contributes to the immune escape, anti-inflammatory, and immune tolerance in either innate or adaptive immune cells (20). Primary AML highly expressed MIF and MIF drives the bone marrow mesenchymal stromal cells (BM-MSC) to express IL-8, which in turn assists AML cell survival and proliferation (21). Base on this finding, an anti-MIF monoclonal antibody (Imalumab) is being studied in a phase I study (NCT01765790) to assess the safety, pharmacokinetics, tolerability, and antitumor activity against solid cancers (22). FGR is a member of the Src family and contributes to the transition of signals from cell surface receptors and promotes inflammatory cytokines releases. FGR expression is restricted to myeloid lineage and is markedly highly expressed in a subset of AML (23). CALR, CCL4, and GZMB are considered genes of the adaptive immunity. CALR is found in myeloproliferative neoplasms (MPN) and represents an MPN-driver mutation. The presence of this gene is included in the current diagnostic criteria of Ph⁽⁻⁾ MPN (24). The AML patients converted from MPN had more CALR mutation rate frequency (25). Moreover, there are reports that the expression of CALR is remarkably higher than other hematologic malignancies, such as ambiguous lineage, ALL, MPN, MDS/MPN (26). CCL4 and GZMB are associated with T-cell immunity. CCL4 is a biomarker of multiple sclerosis and associated with inflammation and T-cell activation (27). AML patients with monocytic differentiation had increased serum levels of CCL4 along with CCL5 and CCL3. The three biomarkers promote an inflammatory state and participated in the progression of suppressing T cell-related immune response (28). GZMB encodes the preprotein secreted by NK cell and CTLs, which is related to the apoptosis of target cells (29). Hypermethylation of the enhancer upstream of GZMB might contribute to an inferior overall survival of AML (30). The mutation in these genes might cause obstacles to the elimination of abnormal cells. Finally, TRH and CLEC11A not only have functions in the endocrine system, but also play a role in immune system. Other genes included in the 17-immune signature have been reported to participate in immune response. These include HSPA1B, IGHD5-18, IGHV4-39, IGHV5-51, and PLXNB2, providing directions and clues for our future research.

Although the immune-17 signature has been proved effective in both training and validation cohorts, the AUC value for this model applied in the validation cohorts is not satisfying; it might be due to different data platforms. The training cohort is generated from RNA-sequencing, while the validation cohort is obtained from the microarray platform. In addition, the different ethnicity distribution may contribute to this result. Although the AUC value is lower in the validation cohort than that in the training cohort, the signature still exhibited predictive power in validation cohort evidenced by the Kaplan–Meier survival and ROC curve analysis.

ELN2017 risk stratification system (integrated cytogenetic and mutational status information) has been utilized in general practice (17). However, non-genetic potential mechanisms have been found to play essential roles in AML patients’ survival (31, 32). In this study, from the immunology perspective, we improved the accuracy of ELN2017 risk stratification system by incorporating the immune-17 signature. Different immune signatures in myeloid neoplasms need to be investigated, which will likely improve knowledge on disease pathogenesis and inform novel therapies. Several novel drugs such as anti-TIM3 and anti-CD47 antibodies are macrophage and lymphocyte immune checkpoint inhibitors, which are under active investigation and (33, 34). These drugs reactivate immune response against myeloid blasts, thus blocking leukemia immune escape.

In conclusion, we constructed an immune-related signature, which is a reliable and accurate model to predict AML patients’ OS. We also refined the ELN2017 risk stratification system after the incorporation of this model. The immune-17 signature may be implemented to refine AML prognosis and, in the future, to inform treatment with novel immunotherapies.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: (https://www.ncbi.nlm.nih.gov/), GSE37642.

RL, ZD, and PJ performed the research. RL designed the research study. SW, GJ, RX, WW, ZJ, and XL analyzed the data. RL, KX, XW, JL and PZ wrote and revised the paper. All authors contributed to the article and approved the submitted version.

This work was supported by the National Key Research and Development Program of China (2019YFA0905900), the Science and technology development fund project of Nanjing Medical University (NMUB2019224) and the Scientific Research Project of Shanghai Science and Technology Committee (17411952000).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.