AML is a type of hematological malignancy with poor clinical outcome. Although the therapy success depends on patients’ age and leukemia type, the limited advances in treatment regime largely resulted in the unsatisfied outcome of AML patients. AML patients benefit from these non-targeted cytotoxic drugs, but chemotherapy intolerance also is a big problem . Besides, AML cells sheltered in the tumor microenvironment, where immune activities were dysregulated, resulting in cancer cells’ self-renewal and developing drug-resistant [18, 19, 20, 21]. Thus, early identify high-risk patients and provide them additional immune targeting therapies may have promising results in clinical management. So, in this study, we constructed two different immune-based prognostic signatures to sort out high-risk patients and identified their most survival-related checkpoint molecules. Those dysregulated expressed immune genes may have therapeutic potential in AML clinical practice.
In the current study, we included 415 AML patients’ mRNA expression and corresponded clinical information obtained from TCGA and GEO databases to construct prognostic predicting signatures. Firstly, based on the TCGA cohort, we constructed a five-immune genes survival signature, including CALCRL, CLEC11A, CRLF3, PLXNB1, and SOCS1. A majority of these genes are suggested to contribute to leukemia pathogenesis [22, 23]. For example, CALCRL expression was positively associated with the engraftment capacity of primary AML patient samples in mice and higher SOCS1 expression was an independent poor prognostic factor for non-M3 and CN-AML patients [24, 25].
With this immune-genes related signature, we divided TCGA cohort patients into high risk and low risk groups, who showed significantly different OS and DFS. What’s more, we found patients’ Immune Risk score is significantly and positively associated with their cytogenetic risk classification. Under different cytogenetic risk conditions, the Immune Risk signature statistically correlated with patients’ survival. However, when we validated this signature in the GEO cohort, ROC curves demonstrated this signature has very low predictive efficiency. And there is no survival difference between patients who were classified into different risk groups by the signature.
The above results indicated that analyzing AML patients’ survival sorely relies on dysregulated immune genes expression and subsequently impaired immune activation is not that convincing. The role immunity system played in the AML tumor microenvironment is too complicated to be simplified into different gene [26, 27, 28]. So, we took different infiltration of various immune cells into concern as many studies have reported that the fraction of infiltrated immune cells in the tumor microenvironment is so unique that could be regarded as patients’ instinct feature to predict tumor progression [29, 30].
Thus, we used the CIBERSORT method to evaluate the fraction of infiltrated immune cells and generated an immunotype-related signature, which included five immune cells: T cells CD4 memory activated, Eosinophils, Macrophages M2, Mast cell resting and T cells regulatory (Tregs). With the immunotype-related signature, we classified patients into high risk and low risk. Different risk group patients, both in the TCGA cohort and the GEO cohort, showed significantly different outcomes.
For comprehensively understanding the role of immunity in predicting AML survival, we further classified patients into Favorable Risk and Poor Risk by combining analysis of dysregulated expressed immune genes and infiltrated immune cells. Patients’ OS and DFS showed significant differences between Favorable Risk group and Poor Risk group, which indicates that the combined analysis of these two prognostic signatures optimized their predicting ability.
Finally, we compared the different expression of checkpoint molecules in Favorable Risk and Poor Risk groups. We found only CD274 and PDCD1 expression significantly correlated with AML patients’ OS in the TCGA cohort, while in the GEO cohort, neither of those dysregulated expressed checkpoint molecules was associated with patients’ survival. Based on the fact that the pathogenesis of AML is very complicated, we considered that combined analysis of different risk factors to predict patients’ prognosis is more reliable and all those differently expressed checkpoint molecules may serve better as potential therapeutic targets.
Although predicting AML patients’ survival with these constructed signatures is promising, there are also some limitations for implying them into clinical management. As those signatures are generated from two retrospective cohorts and enrolled amount is limited as well, larger prospective cohorts are needed to validate the prognostic value of signatures. What’s more, whether these identified checkpoint molecules could act as potential targets need to be further explored by mechanism studies.