Identification and Validation of a Prognostic Risk-Scoring Model Based on Ferroptosis-Associated Cluster in Acute Myeloid Leukemia

Jinghua Wang, Zewei Zhuo, Yanjun Wang, Shuo Yang, Jierong Chen, Yulian Wang, Suxia Geng, Minming Li, Xin Du, Peilong Lai, Jianyu WengFrontiers in Cell and Developmental Biology 2022 · DOI

TL;DR

This study introduces a novel 18-gene prognostic risk-scoring model for Acute Myeloid Leukemia (AML) based on ferroptosis-associated genes. This model effectively stratifies patients into high- and low-risk groups, predicting survival outcomes and showing potential for guiding chemotherapy response and immune-based therapies.

What Problem Does This Paper Address?

Acute Myeloid Leukemia (AML) is a serious blood cancer with a high mortality rate and a significant challenge in predicting patient prognosis. While molecular genetic abnormalities are known to influence AML outcomes, individual patient responses can still vary. Existing risk stratification systems may not fully capture this variability. Ferroptosis, a specific type of programmed cell death, has emerged as a crucial player in AML development and its response to treatment. However, the precise role of ferroptosis-associated genes in predicting AML patient prognosis remained unclear, creating a need for a more refined prognostic tool.

Methods

The researchers employed a multi-stage approach:

  1. Data Acquisition and Preprocessing: They obtained gene expression and clinical data from three large public datasets: TCGA (training set), GEO (GSE71014), and TARGET (validation sets).
  2. Ferroptosis Gene Clustering: Using 261 known ferroptosis-related genes (FRGs), they performed consensus clustering on the TCGA AML samples to identify distinct molecular subtypes.
  3. Differential Gene Expression Analysis: They identified differentially expressed genes (DEGs) between these ferroptosis-related clusters.
  4. Prognostic Gene Selection: Univariate Cox regression and LASSO regression analyses were used to pinpoint genes significantly associated with patient survival from the DEGs.
  5. Risk-Scoring Model Construction: An 18-gene signature was selected to build a prognostic risk-scoring model, where a risk score was calculated based on the expression levels and coefficients of these genes.
  6. Model Validation: The model's predictive performance was validated using Kaplan-Meier survival analysis and Receiver Operating Characteristic (ROC) curves on both the training and independent validation datasets.
  7. Nomogram Development: A nomogram was constructed to integrate the risk score with key clinical factors (like age and AML risk category) for a more comprehensive prognostic prediction.
  8. Correlation Analyses: The study explored the relationship between the risk score and immune cell infiltration using the CIBERSORT algorithm, as well as mutation patterns and chemotherapy response.
  9. Experimental Validation: Quantitative Reverse Transcription Polymerase Chain Reaction (qRT-PCR) was used to validate the expression levels of selected risk genes in primary AML samples compared to normal samples.

Datasets

  • TCGA (The Cancer Genome Atlas) - LAML cohort: Used as the primary training set (n=151).
  • GEO (Gene Expression Omnibus) - GSE71014: Used as a validation set (n=104).
  • TARGET (Therapeutically Applicable Research to Generate Effective Treatments) - AML cohort: Used as another validation set (n=155).
  • IMvigor210 cohort: A bladder cancer immunotherapy dataset used to predict chemotherapy response due to missing drug data in the AML cohorts.

Key Insights and Ideas

  • Ferroptosis Subtypes in AML: The study successfully identified distinct AML patient clusters based on ferroptosis-related gene expression, revealing that these subtypes have different survival outcomes. This suggests ferroptosis plays a significant, previously uncharacterized role in AML stratification.
  • An 18-Gene Prognostic Signature: The development of a robust 18-gene signature that can accurately predict AML patient survival is a major contribution. This signature goes beyond traditional risk factors.
  • Independent Prognostic Power: The risk score derived from this signature was identified as an independent prognostic factor, even when considering established clinical parameters like age and AML risk category.
  • Nomogram for Clinical Application: The creation of a nomogram that integrates the risk score with clinical features offers a practical tool for clinicians to estimate individual patient survival more accurately.
  • Link to Immune Microenvironment: The study highlights a significant correlation between the risk score and immune cell infiltration, suggesting that the ferroptosis-related signature might influence the tumor's immune microenvironment, potentially impacting immunotherapy effectiveness.
  • Chemotherapy Response Prediction: The model showed promise in predicting chemotherapy response, indicating its potential to guide treatment decisions.

Key Results

  • Survival Prediction: The 18-gene prognostic model demonstrated strong predictive power, with Area Under the Curve (AUC) values of 0.81, 0.827, and 0.786 for 1-, 3-, and 5-year survival in the training set, respectively. These results were validated in independent cohorts.
  • Independent Factors: Age and the risk score were identified as independent prognostic factors in multivariate Cox regression analysis.
  • Nomogram Accuracy: The constructed nomogram showed good predictive accuracy, with a C-index of 0.789.
  • Immune Correlations: LGALS1 was found to be most closely associated with immune infiltration. The risk score showed negative correlations with immune-active cells (naïve B cells, CD4+ T cells, resting mast cells) and a positive correlation with monocytes.
  • Chemotherapy Response: Low-risk patients exhibited a higher chemotherapy response rate compared to high-risk patients.
  • Gene Expression Validation: qRT-PCR confirmed that LGALS1 and TMEM56 were significantly underexpressed in AML samples compared to normal samples.

Limitations and Future Work

  • Data Heterogeneity: The authors acknowledge that differences in data preprocessing and the background of patients across the TCGA, GEO, and TARGET datasets might have introduced some bias in prediction efficacy.
  • Limited Drug Data: The lack of comprehensive drug response data in the AML cohorts necessitated the use of a bladder cancer dataset for chemotherapy response prediction, highlighting the need for more AML-specific drug data.
  • Experimental Validation: While some genes were validated, more extensive experimental validation is required to fully elucidate the exact mechanisms of the 18 genes in AML progression, immune therapy, and drug resistance.
  • Prospective Studies: The authors emphasize the need for more prospective studies to confirm the reliability of the nomogram and the prognostic model.
  • RNA-Seq Data Integration: Integrating RNA-sequencing data from various sources would further enhance the robustness of the risk-scoring model.

Summary updated

Abstract

<b>Background:</b> Emerging evidence has proven that ferroptosis plays an important role in the development of acute myeloid leukemia (AML), whereas the exact role of ferroptosis-associated genes in AML patients' prognosis remained unclear. <b>Materials and Methods:</b> Gene expression profiles and corresponding clinical information of AML cases were obtained from the TCGA (TCGA-LAML), GEO (GSE71014), and TARGET databases (TARGET-AML). Patients in the TCGA cohort were well-grouped into two clusters based on ferroptosis-related genes, and differentially expressed genes were screened between the two clusters. Univariate Cox and LASSO regression analyses were applied to select prognosis-related genes for the construction of a prognostic risk-scoring model. Survival analysis was analyzed by Kaplan-Meier and receiver operator characteristic curves. Furthermore, we explored the correlation of the prognostic risk-scoring model with immune infiltration and chemotherapy response. Risk gene expression level was detected by quantitative reverse transcription polymerase chain reaction. <b>Results:</b> Eighteen signature genes, including ZSCAN4, ASTN1, CCL23, DLL3, EFNB3, FAM155B, FOXL1, HMX2, HRASLS, LGALS1, LHX6, MXRA5, PCDHB12, PRINS, TMEM56, TWIST1, ZFPM2, and ZNF560, were developed to construct a prognostic risk-scoring model. AML patients could be grouped into high- and low-risk groups, and low-risk patients showed better survival than high-risk patients. Area under the curve values of 1, 3, and 5 years were 0.81, 0.827, and 0.786 in the training set, respectively, indicating a good predictive efficacy. In addition, age and risk score were the independent prognostic factors after univariate and multivariate Cox regression analyses. A nomogram containing clinical factors and prognostic risk-scoring model was constructed to better estimate individual survival. Further analyses demonstrated that risk score was associated with the immune infiltration and response to chemotherapy. Our experiment data revealed that LGALS1 and TMEM56 showed notably decreased expression in AML samples than that of the normal samples. <b>Conclusion:</b> Our study shows that the prognostic risk-scoring model and key risk gene may provide potential prognostic biomarkers and therapeutic option for AML patients.
BibTeX
@inproceedings{wang2022identification,
  author = {Jinghua Wang and Zewei Zhuo and Yanjun Wang and Shuo Yang and Jierong Chen and Yulian Wang and Suxia Geng and Minming Li and Xin Du and Peilong Lai and Jianyu Weng},
  title = {{Identification and Validation of a Prognostic Risk-Scoring Model Based on Ferroptosis-Associated Cluster in Acute Myeloid Leukemia}},
  booktitle = {Frontiers in Cell and Developmental Biology},
  year = {2022},
  doi = {10.3389/fcell.2021.800267},
}
Figure