Abstract Background: Acute myeloid leukemia (AML) is a highly aggressive and heterogeneous hematologic malignancy. Despite advances in precision medicine and the development of various targeted therapies, a subset of high-risk patients still face poor prognosis and recurrence/refractory issues. The immune microenvironment plays a key role in the development of AML, and the role of plasmacytoid dendritic cells (PDCs)，a kind of immune cell with multiple subtypes, in the AML microenvironment remains controversial. This study integrates transcriptomics data with single-cell RNA sequencing (scRNA-seq) data and employs multiple machine learning (ML) methods to elucidate the role of PDCs in predicting survival prognosis and therapy response to induction treatment in AML patients. Materials and Methods: Single-cell sequencing data from 8 AML patients at our institution was integrated with public scRNA-seq data (GSE198681). These 8 AML patients received ABC-14 regimen (Azacitidine, Venetoclax, and Chidamide, ClinicalTrials.gov NCT06451861.) induction regimens. GSE198681 shows “3+7” induction regimens results. Five ML methods – Lasso regression, Random Survival Forest (RSF) algorithm, Support Vector Machine Recursive Feature Elimination (SVM-RFE) algorithm, Elastic Net model, and Boruta model – were used to screen for genes associated with AML prognosis and to construct a predictive model. This model was subsequently applied to public databases to assess patient prognosis, drug response prediction, enrichment analyses, immune checkpoint expression, and immune cell infiltration in AML. Results: This study identified 7 key genes significantly associated with AML prognosis: SPINT2, TCF4, STIM1, LSP1, KDM3B, ADA2 (CECR1), and CST3. Among these, TCF4 emerged as a common adverse prognostic marker for both the ABC-14 regimen and “3+7” induction regimens. The prognostic model based on these genes effectively stratified patients into high-risk and low-risk groups (P 0.0001) and also demonstrated predictive power for chemotherapy response in the BEAT-AML database, with LSP1 contributing nearly 100% importance in this prediction. The immune microenvironment exhibited significant heterogeneity between high-risk and low-risk AML patients: the low-risk group appeared to possess stronger anti-tumor immunity, while the high-risk group displayed features of an immunosuppressive state. Molecules PD-1, TIGIT, and LAG3 could potentially serve as therapeutic targets to reverse immune dysfunction in high-risk AML. Analysis of the interaction between transplantation status and risk score in 363 patients from the BEAT-AML database revealed transplantation as a protective factor and the risk score as a hazardous factor, both with significant P-values. No interaction effect was found between them (P = 0.654), indicating that the “effect of transplantation” does not vary according to the risk score, meaning high-risk patients can similarly benefit from transplantation. Conclusion: This study is the first to construct a prognostic and induction therapy efficacy prediction model for PDCs in AML. It provides novel insights for AML treatment and prognosis assessment, and the genes incorporated into the model hold potential as new biological markers and therapeutic targets.
Li et al. (Mon,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: