# Constructing a tumor immune microenvironment-driven prognostic model in acute myeloid leukemia using bioinformatics and validation data

## TL;DR

This study creates a model to predict outcomes in acute myeloid leukemia by analyzing the tumor immune environment and identifies key genes for risk stratification and treatment.

## Contribution

A novel immune prognostic model for AML using immune-related genes and validated with clinical and experimental data.

## Key findings

- Immune scores correlated with AML classification and survival, identifying 680 immune-related differentially expressed genes.
- Four key genes (CD163, IL10, MRC1, FCGR2B) were selected to build a risk score model with significant survival differences.
- CD163 was experimentally confirmed to be elevated in AML, while MRC1 showed no differential expression.

## Abstract

The tumor immune microenvironment (TIME) is a critical determinant of prognosis in acute myeloid leukemia (AML). This study aimed to develop a prognostic model based on immune-related hub differentially expressed genes (hub-DEGs) to refine risk stratification and identify therapeutic targets. Transcriptomic and clinical data from 149 TCGA-AML patients were analyzed using ESTIMATE and xCell algorithms to infer immune scores. Differentially expressed genes (DEGs) between high/low immune score groups were identified, followed by functional enrichment, protein-protein interaction (PPI) network analysis for selecting the hub-DEGs with the highest degree scores, and univariate Cox regression to pinpoint prognostic genes. External validation was performed on 562 GEO-AML patients. The final genes were selected by intersecting the prognostic DEGs and hub-DEGs. Next the immune prognostic model (IPM) was created using these genes. xCell and CIBERSORT algorithm were used to assess the correlation of IPM and different immune cells. Finally, Experimental validation of key genes (CD163, MRC1) was conducted via RT-PCR in 40 AML and 10 control samples. Immune scores correlated with FAB classification (ESTIMATE: p-value = 1.4e − 8; xCell: p-value = 3.7e − 9) and overall survival (ESTIMATE: v = 0.041). Analysis identified 680 immune-related DEGs enriched in immune response pathways. Intersection of prognostic DEGs (n = 34) and hub-DEGs (n = 30) yielded four genes (CD163, IL10, MRC1, FCGR2B). A risk score model stratified patients into high/low-risk groups with divergent survival (p-value = 0.00072). ROC analysis demonstrated predictive accuracy (AUC: 63.38–68.5% for 1–5-year survival). TIME analysis revealed associations between high-risk scores and immunosuppressive cell subsets, including Tregs and M2 macrophages. RT-qPCR confirmed elevated CD163 in AML (p < 0.001), while MRC1 showed no differential expression. This study establishes a TIME-centric prognostic model with clinical utility for risk stratification and therapeutic targeting in AML. Prospective validation and integration of advanced genomic technologies are warranted to refine its translational applicability.

The online version contains supplementary material available at 10.1038/s41598-025-03557-9.