PDF
Citation
Print
Introduction
Acute myeloid leukemia (AML) is a highly diverse disease, with various subtypes characterized by distinct genetic profiles, clinical features, and prognoses [1,2]. AML is characterized by the clonal expansion of myeloid precursors that exhibit impaired differentiation. This condition results from the accumulation of genetic and epigenetic alterations in hematopoietic stem cells, which leads to dysregulated gene expression and chromosomal abnormalities [1].
Despite advancements in treatment protocols, the overall five-year survival rate for AML is estimated to be approximately 30% [2]; this rate has not significantly improved due to factors such as high relapse rates and resistance to standard chemotherapy. The challenges in improving patient outcomes highlight the urgent need for novel therapeutic strategies that can address the complex nature of the disease, thereby facilitating a more individualized treatment approach for AML patients.
Making treatment decisions and managing patients depend on identifying the key genes and pathways associated with cancer prognosis [3]. The widespread use of high-throughput sequencing technology has led to the accumulation of gene signature studies over time, resulting in large-scale, reusable gene expression profiles that are publicly accessible. Several prognostic studies have redundantly identified linked genes (or genes with similar functions) across multiple investigations, primarily because they utilize the same public dataset sources, such as The Cancer Genome Atlas (TCGA) [4].
The most common types of failure in both pediatric and adult AML remain resistant and relapsed disease. One of the primary reasons for AML relapse is the persistence of leukemic stem cells (LSCs) [5]. Drug resistance is also associated with LSCs. Therefore, to enhance risk classification in AML and predict prognosis, the gene expression signatures of LSCs have been identified and quantified. The 17-gene stemness score (LSC17) [6] and the six-gene LSC score in pediatric AML (pLSC6) [7] have the potential to redefine initial risk stratification and identify low-risk AML, which will aid in the development of new treatment strategies.
Pyroptosis is a form of programmed cell death that is characterized by inflammation and is mediated by the activation of inflammatory caspases, which are triggered by inflammasomes [8]. The six-gene pyroptosis-related signature has been closely linked to the prognosis of AML [8]. Additionally, the following gene signatures have been reported: IED172 (172-gene immune effector dysfunction) signature [9], PS29MRC signature [10], 24-gene signature [11], IDO1 (immune-related gene) signature [12], autophagy signature [13], 7-gene signature [14], IRG (immune-related gene) signature [15], hypoxia signature [16], and CXCR signature [17].
The literature-based strategy, which has gained popularity as an alternative to the data-driven approach, identifies and validates predictive biomarkers using prior research papers or databases [4]. In this study, we employed a literature-based methodology to develop a gene signature for AML prognosis using the penalized Cox regression technique, demonstrating its robustness. Additionally, we utilized the previously published Tumor Online Prognostic Analysis Platform (ToPP), a validated web- or code-based tool that ensures reproducibility, to build this gene signature model [18].
Materials and Methods
1. AML patient cohorts and gene expression data
To construct a reliable prognostic gene signature for AML, we utilized comprehensive gene expression datasets from various AML cohorts. Specifically, we sourced gene expression data for 151 adult AML patients from the Genomic Data Commons (GDC) TCGA-LAML (The Cancer Genome Atlas – Acute Myeloid Leukemia) dataset, which is accessible through the UCSC Xena browser (https://xenabrowser.net/datapages/). This dataset includes detailed expression profiles, point mutation information, and overall survival (OS) data. Additionally, we obtained gene expression data for 187 pediatric AML patients from the TARGET-AML (Therapeutically Applicable Research to Generate Effective Treatments – Acute Myeloid Leukemia) dataset, also hosted on the Xena browser platform (https://xenabrowser.net/datapages/) [19].
For further validation, we incorporated data from the BeatAML dataset, which includes 494 patients and is accessible through the Vizome portal (http://www.vizome.org/). This methodological approach ensures a comprehensive analysis of gene expression across various AML subtypes and patient demographics, aiming to enhance the prognostic accuracy of our gene signature.
The final analysis included three independent cohorts: TCGA-LAML (n=132), BeatAML (n=411), and TARGET-AML (n=187). An overview of the biological and clinical data from TCGA-LAML and BeatAML can be found in the previous publication [20] and is also summarized in the metadata (S1 and S2 Tables). The TCGA-LAML cohort consisted of adult patients with a median age of 55 years (range, 21 to 88 years), with 71 males and 61 females. The BeatAML cohort included adult patients with detailed clinical characteristics, including median white blood cell (WBC) count of 19.7×109/L (range, 0.1×109/L to 427.46×109/L), median bone marrow blast percentage of 70% (range, 0 to 98), and median peripheral blood blast percentage of 45.5% (range, 0% The TARGET-AML cohort comprised pediatric patients with a median age of 8 years (range, 0 to 22 years), with 91 males and 96 females. This cohort showed a median WBC count of 42.5×109/L (range, 1.3×109/L to 519×109/L), median bone marrow blast percentage of 73.6% (range, 14% to 100%), and median peripheral blood blast percentage of 59% (range, 0% to 97%).
Cases lacking survival status or relevant information were excluded from the analysis: 19 patients were excluded from the TCGA-LAML cohort, 83 patients from the BeatAML cohort, and no patients were excluded from the TARGET-AML cohort. For the purpose of this study, the TCGA-LAML cohort was designated as the primary analysis set. Additionally, the TARGET and BeatAML cohorts were utilized as validation sets to assess the generalizability of our findings across different datasets.
2. Candidate genes and literature for literature-based gene signatures
The 12 gene signatures referenced in the construction of the literature-based gene signature are detailed in Table 1. These include the IED172 signature [9], 29MRC signature [10], Pyroptosis signature [8], 24-gene signature [11], LSC17 signature [6], IDO1 signature [12], Autophagy signature [13], 7-gene signature [14], IRG signature [15], pLSC6 signature [7], Hypoxia signature [16], and CXCR signature. Collectively, these signatures comprise a total of 313 genes. After excluding 13 duplicates, we selected a final set of 300 genes for analysis (S3 Table).
3. Feature selection and model building
We applied the Least Absolute Shrinkage and Selection Operator (LASSO) Cox regression to refine gene selection and reduce the number of genes for constructing a gene signature. The LASSO-Cox regression was performed using the R glmnet package on the TCGA-LAML dataset.
We conducted univariate Cox regression and LASSO-Cox analysis on 300 genes using the TCGA dataset. To facilitate this analysis, we employed the online tool ToPP (http://www.biostatistics.online/topp/index.php) [18], which strea-mlines predictive modeling in survival analysis. Additionally, to analyze cohorts not available in the ToPP tool, we developed and utilized an in-house version of ToPP. Gene selection involved 10-fold cross-validation, repeated 100 times, to ensure robustness. Genes selected in more than 95 out of 100 iterations were included in the final prognostic model, thereby reducing the number of genes considered.
After constructing the prognostic model using the gene signature, patients were categorized into high-risk and low-risk groups based on the median literature-based signature (LBS) risk score as the cutoff point. Kaplan-Meier survival analysis, along with the log-rank test, was performed to compare the survival outcomes between the high-risk and low-risk groups based on the LBS scores.
4. Validation of the prognostic model
We conducted external validations using datasets from TARGET-AML and BeatAML, employing the median risk score as the criterion for grouping. For the TARGET-AML dataset, we utilized forest plots to visually represent the impact of the gene signature on event-free survival (EFS). For the TARGET-AML and BeatAML datasets, we used forest plots to illustrate the impact on OS.
5. Identification of differentially expressed genes between high-risk and low-risk groups
After stratifying patients into high-risk and low-risk groups based on the median LBS risk score as the cut-off point, differentially expressed genes (DEGs) between the two groups were identified using the ‘DESeq2’ package. This statistical framework is designed to assess differential expression in gene expression data through a model based on the negative binomial distribution. The TCGA-LAML raw count dataset was utilized for the DEG analysis. The criteria for determining DEGs included a Log2 absolute fold change greater than 1 and a false discovery rate of 0.05 or lower, ensuring that only genes with statistically significant and biologically relevant changes in expression were considered.
6. Enrichment analysis
In continuation of the DEG analysis, we conducted a functional enrichment analysis to identify the biological functions and pathways associated with the identified genes. The DEGs, previously identified through the DEG analysis, were categorized into highly expressed and lowly expressed genes, which served as the input for the subsequent functional enrichment analysis. This analysis was performed using gProfiler (https://biit.cs.ut.ee/gprofiler/gost), a robust tool that enables a comprehensive exploration of functional annotations and pathways related to a specific gene set.
7. Immune profile analysis
To investigate the immune status of the high- and low-risk groups based on the LBS score within the TCGA-LAML cohort, we employed CIBERSORTx [21]. The gene expression values used for the analysis were log2(FPKM+1), where FPKM (fragments per kilobase of transcript per million mapped reads) values served as parameters for the CIBERSORTx analysis. We utilized the LM22 signature matrix, which encompasses 22 distinct immune cell types in both the high- and low-risk groups. Additionally, we incorporated the LM22 Source GEP file as an optional resource. For significance analysis, we performed 1,000 permutations, and samples with p < 0.05 were chosen for further analysis.
8. Statistical analysis
Prior to conducting LASSO-Cox regression analysis using ToPP, the raw count data from the BeatAML dataset was converted into normalized expression measures, specifically FPKM. We obtained publicly accessible FPKM data from the TARGET database for the validation cohort. For the subsequent statistical analysis, we utilized log2 (FPKM+1) values.
The R ver. 4.3.3 (R Software, R Foundation for Statistical Computing) was used to analyze the data. We assessed survival outcomes using Kaplan-Meier curves and evaluated the statistical significance between two distinct patient groups through the log-rank p-value and hazard ratio. For paired data, the Wilcoxon matched-pairs signed-rank test was employed. To conduct hypothesis testing across all three cohorts, a one-way ANOVA followed by Tukey’s multiple comparison test was applied. A p-value of less than 0.05 was considered significant.
Results
1. Study workflow
The comprehensive workflow for integrating and validating gene signatures in pediatric AML is depicted in Fig. 1. The study utilized a literature-based gene list consisting of 300 genes. These genes underwent feature selection using LASSO-Cox regression, implemented via the R package glmnet, applying the lambda.min criterion. This process identified six key genes that contributed to the risk score, reducing the initial list from 300 genes.
Subsequent analyses included DEG analysis using DESeq2, which identified genes with significant expression differences between the risk groups based on the literature-based signature 6 (LBS6) risk score. Enrichment analysis conducted with gProfiler elucidated the biological pathways associated with the gene signature. Additionally, immune cell infiltration analysis using CIBERSORT offered insights into the immune landscape across different patient groups, further validating the robustness of the gene signature.
2. Construction of the LBS6 signature for prognostic classification of AML patients
The six genes finally selected were ETFB (electron transfer flavoprotein beta subunit) [11], ARL6IP5 [11], PTP4A3 [11], CSK [15], HS3ST3B1 [9], and PLA2G4A [11]. These genes were significantly correlated with the prognosis of AML and were utilized to construct the prognostic model. The formula for the risk score was as follows:
Risk score=0.9642×(ETFB)+0.4012×(ARL6IP5)+0.3433×(PTP4A3)+0.2832×(CSK)+0.1225×(PLA2G4A)–0.1552×(HS3ST3B1)
These selected genes were found to be significantly correlated with the prognosis of AML (OS) and were subsequently used as potential characteristics for constructing the prognostic model. Patients in the high-risk group exhibited significantly shorter OS compared to those in the low-risk group (p < 0.001; hazard ratio [HR], 4.2 [95% confidence interval (CI), 2.59 to 6.81]) (Fig. 2).
3. Validation of integrated 6-gene signature (LBS6) as a prognostic risk score
We applied the risk score estimated from the training data to two independent AML patient cohorts to validate the prognostic performance of the LBS6 gene signature. To assess the prognostic significance of LBS6 in pediatric AML, we calculated the median risk score of LBS6 values using an established equation in an independent cohort of pediatric AML patients from the TARGET database, which included available clinical outcomes and mRNA seq expression data. The distribution of LBS6 in the TARGET cohort was found to be comparable to that observed in the previous cohort. The LBS6 gene score identified a similar division into risk groups as in the earlier cohort, with patients categorized into the low-risk group (n=94) and the high-risk group (n=93). OS was also poorer in the high LBS6 group compared to the low LBS6 group within the TARGET cohort (HR, 2.05; 95% CI, 1.36 to 3.08; p < 0.001) (Fig. 3A). Additionally, EFS between the low and high-risk groups based on the LBS6 gene signature showed that the high-risk group exhibited a significantly poorer prognosis, with an HR of 2.09 (95% CI, 1.38 to 3.15; p < 0.001), indicating more than a twofold increase in the risk of adverse events compared to the low-risk group (Fig. 3B).
Fig. 4A shows a forest plot depicting the hazard ratios for various prognostic factors affecting OS in pediatric AML patients from the TARGET-AML cohort. The clinical trials that enrolled the TARGET patients used the traditional Children’s Oncology Group risk system. The presence of t(8;21) (q22;q22), inv(16)(p13.1q22), or t(16;16)(p13.1;q22) is classified as low risk, and the presence of monosomy 7, monosomy 5/5q deletion, or persistent disease at the end of the first course of induction therapy, or with FLT-3 internal tandem duplication high allelic ratio (> 0.4; FLT3-ITD high allelic ratio), is classified as high risk, and others as standard risk group [2].
Key variables demonstrating a significant impact included the LBS6 gene score (HR, 1.73; p=0.031), standard risk group (HR, 2.08; p=0.021), and age over 10 years (HR, 1.71; p=0.023). Likewise, a high LBS6 gene score was identified as a statistically significant prognostic factor for EFS, as shown in Fig. 4B. Significant variables for EFS in pediatric AML include the LBS6 gene score (LBS6_score_median: HR, 1.70; p=0.007), CEBPA mutation (HR, 0.3; p=0.049), and NPM mutation (HR, 0.33; p=0.038). Well-established prognostic factors, including specific mutations such as CEBPA, WT1, FLT3, and NPM1, demonstrated statistical significance in univariate survival analyses (S4 Table). However, when included in multivariate analysis alongside LBS6, these mutations did not exhibit statistically significant effects on OS or EFS (except CEBPA on EFS).
S5 Fig. showed that the LBS6_score_median was a significant prognostic factor influencing OS in patients with AML from the BeatAML cohort (HR, 1.52; p=0.0013), where higher scores are associated with an increased risk of lower OS. The multivariate forest plot showed that no significant effects were observed for other predictors, such as WBC count, CEBPA mutation status, FLT3 status, and NPM mutation status, suggesting that these variables do not significantly impact survival outcomes (S6 Fig.). Additionally, the plot illustrates the statistical power of the LBS6_score_median in predicting survival, highlighting its potential utility in patient risk stratification.
4. DEG analysis between high-risk and low-risk groups defined by the LBS6 gene signature
A DEG analysis was conducted to compare high-risk and low-risk groups defined by the LBS6 gene signature in AML. The analysis identified genes with an absolute log2 fold change > 1 and adjusted p-value (padj) < 0.05 as significantly different. This DEG analysis was based on 132 cases from the TCGA dataset. The volcano plot illustrates the distribution of genes, highlighting significantly upregulated and downregulated genes (S7 Fig.).
The analysis identified 77 genes with significant differences, comprising 17 upregulated and 60 downregulated genes (S8 Table). Notable upregulated genes included C1QL1, CAMKV, CREG2, DPEP3, DPYSL3, EGR4, H19, H2BC3, IGHV1OR21-1, IGLVI-70, LINC03000, PEG10, RHCG, RIMBP2, RNVU1-29, SCUBE1, and TLE6.
5. Enrichment analysis
The enrichment analysis revealed significant involvement of several molecular functions, particularly those associated with transcriptional activity (S9 Fig.). Notable terms include “DNA-binding transcription factor activity”, “RNA polymerase II-specific,” and “CXCR3 chemokine receptor binding.” These findings suggest a pivotal role for transcription regulation and chemokine signaling in the pathophysiology of the disease.
In terms of biological processes, there was a significant enrichment in processes related to calcium ion regulation, such as the “release of sequestered calcium ion into cytosol” and the “regulation of calcium ion sequestration,” suggesting that calcium signaling pathways may play critical roles in disease progression or response mechanisms in high-risk groups.
Cellular component analysis further supported these findings, revealing significant enrichment in components such as the “junctional sarcoplasmic reticulum membrane” and the “cell periphery,” highlighting the importance of specific cellular locales in the function and interaction of leukemic cells.
6. Immune cell population analysis: CIBERSORTx
To further explore the immunological characteristics of pediatric AML patients, we utilized CIBERSORTx with the LM22 matrix to analyze immune cell compositions between high-risk and low-risk groups, classified according to the LBS6 gene signature (S10 Fig.). We found a significantly higher presence of monocytes in patients classified as high-risk, suggesting increased immune activation in this group (p < 0.001). In contrast, the low-risk group exhibited a significantly higher abundance of resting mast cells (p < 0.001), which may reflect distinct immunological responses or interactions within the tumor microenvironment.
In addition to these noteworthy differences, the analysis of other immune cells, including naïve B cells, CD8 T cells, plasma cells, resting CD4 memory T cells, M2 macrophages, and eosinophils, did not show statistically significant differences between the high-risk and low-risk groups (S11 Table).
Discussion
We developed a predictive gene signature, termed LBS6, and an associated risk score using a LASSO-Cox regression model based on a literature-driven approach in the TCGA-LAML dataset. We validated LBS6 in two independent datasets: TARGET-AML and BeatAML. Our analysis utilized a curated set of 300 well-validated genes reported in the literature, which includes immune- and cell-death–related genes, across multiple AML datasets. Additionally, ToPP was employed to provide robust capabilities for prognostic analysis and model construction [18].
Notably, the risk score derived from the LBS6 gene signature demonstrated predictive significance in differentiating the survival status of AML patient cohorts. The LBS6 gene signature, which includes ETFB, ARL6IP5, PTP4A3, CSK, HS3ST3B1, and PLA2G4A, was significantly associated with lower OS rates across the TCGA dataset (HR, 4.2; 95% CI, 2.59 to 6.81; p < 0.0001), as well as in two independent datasets: BeatAML (HR, 1.52; 95% CI, 1.17 to 1.96; p=0.0013) and TARGET (HR, 2.04; 95% CI, 1.39 to 3.08; p < 0.001).
The comparison of our 6-gene signature with the well-established LSC17 gene signature revealed a similar decrease in survival rates between the high-risk and low-risk groups, with an HR of 4.2 (95% CI, 2.59 to 6.81) based on the LBS6 gene signature, compared to an HR of 4.67 (95% CI, 2.82 to 7.76) based on the LSC17 gene signature (S12 Fig.). This indicates that the LBS6 gene signature is comparable and effectively predicts the prognosis of patients with AML and that the risk score derived from the integrated gene signature of only six genes is a reliable predictor of AML outcomes.
This literature-based gene signature of LBS6 demonstrated consistent performance across the AML cohorts, despite the heterogeneity in therapies, age, and gene mutations among the patients. LBS6 signature emerged as a powerful prognostic tool in our study, demonstrating independence and significance in multivariate analyses. Notably, it retained predictive power for both OS and EFS across diverse AML patient cohorts, even after adjusting for well-established risk factors and genetic mutations, WBC count, and the status of key genes such as CEBPA, WT1, FLT3, and NPM1. The unique strength of the LBS6 signature in maintaining its prognostic relevance amidst these established factors underscores its potential to substantially significantly enhance AML risk stratification systems.
Among the six genes in the LBS6 signature, CSK (c-Src, C-terminal Src kinase), a regulator of Src family kinases, has emerged as a critical component linked to adverse clinical outcomes. High expression levels of CSK have been consistently associated with negative prognostic factors, such as advanced age, elevated WBC counts, and poor cytogenetics, underscoring its influence within the immune microenvironment [15]. Furthermore, the role of CSK in regulating Src family kinases, which are crucial for T-cell activation, aligns with its association with negative outcomes in AML, where pathways such as Akt/mammalian target of rapamycin are affected by alterations in c-Src activity [22]. This analysis of immune-related gene expression may reflect the significance of the tumor microenvironment in disease progression and patient outcomes in AML [12,15].
Four genes (ETFB, ARL6IP5, PTP4A3, and PLA2G4A), which comprised the LBS6 signature in our study, belong to the 24-gene prognostic signature [11] derived from a meta-analysis of Cox regression values across multiple training sets. Interestingly, ARL6IP5, PTP4A3, and PLA2G4A showed frequent interactions among the 24 genes within this signature [11].
Previous studies have indicated that ETFB is involved in mitochondrial function and energy metabolism, which are critical processes in cancer cell survival and proliferation [11]. Alterations in genes such as ETFB can lead to the disruption of host-cell homeostasis.
PTP4A3 (protein tyrosine phosphatase type IVA, member 3), also known as PRL-3, is a well-known oncogene that promotes cancer cell migration, invasion, and metastasis. High expression levels of PTP4A3 are associated with poor prognosis in various cancers, including AML [23]. Studies have shown that PTP4A3 is significantly upregulated in AML and correlates with adverse outcomes [23].
ARL6IP5 (ADP ribosylation factor like GTPase 6 interacting protein 5) is involved in various cellular processes, including intracellular transport [24]. ARL6IP5 expression has been correlated with chemotherapeutic response [24]. A 24-gene signature that includes ARL6IP5 and our LBS6 was associated with poor prognosis in AML, suggesting its contribution to the aggressive behavior of leukemia cells [11].
PLA2G4A (phospholipase A2 group IVA) has emerged as a significant gene in the progression and prognosis of AML. PLA2G4A serves as a potential biomarker for predicting shorter OS in patients with non-M3/NPM1 wild-type AML [25]. Additionally, PLA2G4A may function as a prognostic marker and a potential therapeutic target for specific subtypes of AML [26].
The PLA2G4A gene in our LBS6 signature is associated with necroptosis [27]. While our LBS6 signature included PLA2G4A, it did not contain any other genes linked to various forms of regulated cell death, such as necroptosis, pyroptosis, apoptosis, cuproptosis, and ferroptosis. Recent studies integrating lipid metabolism with immune-related genes have proposed new prognostic classifications for AML, highlighting the pivotal role of PLA2G4A in both metabolic and immune processes [28]. The collective evidence underscores the significance of PLA2G4A in the pathogenesis of AML, as well as its potential as a biomarker and therapeutic target.
HS3ST3B1 is a component of the IED172 signature, as reported by Rutella et al. [9]. Although this gene is part of a broader immune-related expression profile in AML, its specific clinical significance and individual contribution to leukemia pathogenesis have yet to be fully elucidated.
The enrichment analysis highlighted the significant involvement of several molecular functions, particularly those associated with transcriptional activity and chemokine signaling. Notably, “CXCR3 chemokine receptor binding” suggests that chemokine signaling plays a crucial role in the pathogenesis of AML, emphasizing its importance in immune activation and its potential as a therapeutic target. This signaling pathway recruits effector T cells to tumor sites, thereby enhancing anti-tumor immunity and influencing leukemia progression [29]. CXCR3 expression in regulatory T cells affects CD8(+) T-cell immunity and impacts cancer dissemination by guiding T-cell fate. Targeting CXCR3 signaling could enhance immune responses and improve therapeutic outcomes in AML, making it a promising treatment strategy. These enriched Gene Ontology (GO) terms underscore the complex interplay of genetic regulation, immune response, and cellular interactions in pediatric AML, providing insights into potential therapeutic targets and the biological foundations of risk stratification.
Notably, the CIBERSORTx analysis reveals a significant enrichment of monocytes in the high-risk group and an increased presence of resting mast cells in the low-risk group. These findings suggest distinct immune landscapes, with the high-risk group exhibiting a potential activation profile that may contribute to aggressive disease progression. According to the literature, the presence of monocytes is associated with a pro-tumorigenic environment due to their role in suppressing anti-tumor responses and promoting angiogenesis and tumor growth [30]. Conversely, an increased presence of mast cells has been linked to improved outcomes in certain cancers, potentially due to their role in modulating immune responses and inflammation. This pattern underscores the complexity of the immune landscape in pediatric AML and highlights the potential of specific immune profiles as markers of disease severity and prognosis.
Notably, the LBS6 gene signature was revealed as an independent risk factor in multivariate analysis, irrespective of other established risk factors such as poor karyotypes, elevated WBC, advanced age, and gene mutations. An important limitation of our study is that key prognostic factors such as AML M3 and core binding factor AML status and detailed treatment were not consistently available across the public databases utilized. We employed multivariate analysis to control known clinical variables and minimize potential confounding effects, helping to mitigate the impact of treatment heterogeneity and key prognostic factors. Future studies incorporating these important clinical and molecular features will be valuable for a more comprehensive understanding of IED signatures across AML subtypes.
While our study suggests the potential prognostic value of the LBS6 signature, it may serve as a complement to, rather than a replacement for, established risk factors such as cytogenetics and molecular abnormalities. The independent prognostic significance of LBS6 in multivariate analysis indicates that it might provide additional, complementary information that could potentially enhance current risk stratification strategies. Based on LBS6-defined risk groups, different therapeutic approaches could be considered: patients with high LBS6 scores might potentially benefit from more intensive therapeutic approaches, possibly including consideration for allogeneic stem cell transplantation, while standard treatment protocols could be sufficient for those with low LBS6 scores. However, these therapeutic suggestions would require prospective validation through clinical trials before any consideration for implementation in clinical practice.
Our study is significant in that the LBS6 signature was constructed based on the expression levels of genes derived from well-documented signatures that reflect the key biological processes associated with AML across various datasets. Additionally, a literature-oriented approach presents a viable option for constructing a robust gene signature at a reasonable cost, particularly when multiple studies are available.
In addition, our study demonstrated that the six-gene signature performed consistently across various AML cohorts, regardless of differences in age, genetic backgrounds, and therapies, establishing it as an independent prognostic factor for AML patients. By utilizing only six genes in conjunction with cytogenetic and molecular tests, patients with AML can be accurately stratified by risk through practical techniques such as multiplex real-time quantitative polymerase chain reaction.
In conclusion, the LBS6 score, derived from well-validated gene signatures of 300 genes across multiple independent AML datasets, has the potential to redefine early risk categorization and identify low-risk AML cases. By refining the gene panel while preserving its predictive power, the LBS6 score enhances clinical value and may inform the development of new therapeutic strategies.
XML Download