Comprehensive analysis of the ability of the Cuproptosis- related gene signature to predict the prognosis of patients with pediatric acute myeloid leukemia

doi:10.21203/rs.3.rs-4280593/v1

Download PDF

Research Article

Comprehensive analysis of the ability of the Cuproptosis- related gene signature to predict the prognosis of patients with pediatric acute myeloid leukemia

https://doi.org/10.21203/rs.3.rs-4280593/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background: Cuproptosis is a new type of Regulated Cell Death (RCD) that is distinct from other known types of cell death, and novel prognostic models for leukemia have been proposed based on this process. However, most studies have been performed predominantly in adults. The potential roles of cuproptosis in pediatric leukemia remain unclear.

Methods: We comprehensively analyzed the expression and mutation sites of CRGs in both pediatric and adult AML patients. Least absolute shrinkage and selection operator (LASSO) was used to construct a cuproptosis-related signature, and univariate Cox regression and differential gene analysis were also performed. According to the risk model, AML patients were divided into two groups. Several statistical approaches, such as Kaplan‒Meier, receiver operating characteristic (ROC), Cox regression, and calibration curve analyses, were used to evaluate the prognostic value of the model. Furthermore, the associations between risk group and immune cell infiltration, HLA expression, immune checkpoint expression, and therapeutic drug sensitivity were also analyzed. Finally, the expression of the mRNAs in the signatures was investigated in AML clinical samples by quantitative polymerase chain reaction (qPCR) to demonstrate the reliability of our analysis.

Results: Our results revealed differences in the expression levels and prognostic value of CRGs between adult and pediatric AML patients. Subsequently, four CRG signatures (FAU, RPL19, TRAIP, and TEFM) were generated. Cox regression analyses demonstrated that the signature was an independent predictor of prognosis. GO and KEGG analyses also revealed that the signature was significantly associated with memory T-cell activation and negatively correlated with natural killer (NK) and regulatory T-cell activation. In addition, we identified the distinctive agents neopeltolide, gemcitabine, oligomycin A, ect. weresensitive to high-risk pediatric AML patients. Finally, we used clinical samples to validate the expression and prognostic value of the four CRG signatures to verify the results.

Conclusions: Our observations revealed disparate functions of CRGs in pediatric and adult AML patients. A cuproptosis-related signature was established to predict patient prognosis and inform potential therapeutic strategies for pediatric AML patients. As the first attempt to determine pediatric AML prognosis using RCD biomarkers, we provide new insight into the prognosis assessment field.

Acute myeloid leukemia (AML)

Cuproptosis

Pediatric oncology

Tumor microenvironment

Survival prognosis

Drug sensitivity analysis.

Despite the broad overlap in the diagnostic and treatment recommendations between pediatric and adult AML patients, there are important differences in terms of the molecular landscape and clinical outcomes ^1-3. The molecular basis may unveil novel diagnostic and prognostic biomarkers of AML, holding immense significance not only for clinical diagnostics but also for future exploratory research ^4-6. However, most studies have been performed predominantly in adults ⁷. It is vital to find reliable biomarkers that will contribute to early diagnosis and more specialized therapy for pediatric AML patients.

Recently, various types of regulated cell death (RCD), such as necroptosis, ferroptosis, and pyroptosis, which are crucial for tissue homeostasis and growth, have been identified ⁸. Cuproptosis, a new type of RCD characterized by a dependence on mitochondrial respiration and protein lipoylation, is distinct from known death mechanisms ^{9, 10}. Additionally, accumulating evidence indicates crosstalk between cuproptosis and AML. Elesclomol, a molecule that binds copper in the environment and transports it into the cell to induce cuproptosis, has been proven to have anti-AML effects¹¹. Disulfiram combined with copper selectively activated the ROS-JNK while inhibiting the NF-kB and Nrf2 pathways, eradicating AML stem cells ¹². One clinical trial indicated the impact of copper on malignancies, including AML ¹³. Using the TCGA database, a signature of cuproptosis-related genes (CRGs) specific to adult AML was constructed, which demonstrated the prognostic value of these genes ¹⁴. However, the application of cuproptosis in pediatric AML has yet to be explored.

Here, our analysis revealed that the number of CRGs differed between pediatric and adult AML patients. Then, we established a prognostic signature model for the CRGs to serve as prognostic markers and elucidate the underlying mechanism of the immune microenvironment. Finally, we explored and summarized the relationships between drug sensitivity and CRGs in patients. An overview of the research design is presented in Supplemental Figure 1. In conclusion, these results highlight the role of cuproptosis in AML and contribute to further investigations of molecular mechanisms and medical interventions. We made the first attempt to explore the unknown field with broad prospects, offering inspiration for more in-depth research related to RCDs in pediatric AML in the future.

Acquisition of data

The RNA sequencing (RNA-seq) data and clinical information of 151 adult AML patient bone marrow (BM) samples, 187 pediatric AML patient bone marrow (BM) samples, and 606 normal peripheral blood samples were acquired from The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/), the Therapeutically Applicable Research To Generate Effective Treatments (TARGET) database (https://ocg.cancer.gov/programs/target), and the Genotype-Tissue Expression (GTEx) database (https://www.gtexportal.org/), respectively. The clinicopathological characteristics of the patients are shown in Supplemental Table 4. The RNA-seq data of raw counts normalized from the aforementioned databases were utilized for differential expression analysis, and the “sva” R package was used to merge every two datasets and remove batch effects (Supplemental Figure 2D-F). Furthermore, the gene expression profile and relevant clinical characteristic data of 52 AML patients were downloaded from previous research (GSE192638) as a validation cohort ¹⁵. All patients’ clinical features are detailed in Supplemental Table 3.

Derived from the genome-wide CRISPR-Cas9 loss-of-function test reported in previous literature (11), 347 potential cuproptosis-related genes (FDR < 0.05) were identified (Supplemental Table 1). In addition, the annotation of lncRNAs was obtained from GENCODE (https://www.gencodegenes.org/) ¹⁶.

Differential expression analysis

To assess the expression levels of Cuproptosis-Related Genes (CRGs) in adult (TCGA cohort) and pediatric (TARGET cohort) AML patients, after removing the batch effect, we first summed the normalized expression of all CRGs to obtain the total number of CRGs in each cohort and plotted a boxplot using the “ggplot2” package. Differential expression analysis (pAML vs. aAML) was performed using the R package “limma”. The significance thresholds were set to |log2-fold change (FC)| ≥ 1 and p < 0.05. A heatmap showing the distribution of the identified DEGs was generated with the R package “pheatmap”.

The distribution of somatic mutations and copy number variations

The somatic mutation and copy number variation (CNV) data were obtained from the TARGET and TCGA databases. The somatic mutation landscape was analyzed via the “maftools” R package. The frequencies of genetic amplification and deletion were subsequently summarized. Additionally, genes in CNV regions were annotated using the Genome Research Consortium Human Build 38 (GRCh38) as the reference genome.

Consensus clustering

Unsupervised clustering analysis using the k-means algorithm was performed on adult and pediatric AML patients ¹⁷, determined by the consensus clustering algorithms of the “ConsensuClusterPlus” R package. We conducted 1,000 repetitions to ensure classification stability, and the consensus index, relative change in area under the CDF curve, and sample cluster heatmap were generated. Principal component analysis (PCA) and t-distributed stochastic neighbor embedding (tSNE) analysis were subsequently applied to visualize sample clustering. Kaplan–Meier (KM) overall survival curves between different clusters were generated by the “survival” R package.

Construction of a prognostic cuproptosis-related signature

Differential expression of CRGs between AML and normal samples after removing batch effects was analyzed using the “limma” R package, and genes with a p value less than 0.05 were considered to be significantly different. Univariate Cox regression analysis of OS was subsequently performed to screen for CRGs associated with survival (FDR < 0.05). The significantly prognosis-related DECRGs were identified using the “VennDiagram” R package. Subsequently, least absolute shrinkage and selection operator (LASSO) Cox regression analysis using the “glmnet” R package was performed to remove redundant prognostic genes to establish the CRG signature prognostic model ¹⁸. The model penalty parameter λ was determined by 10-fold cross-validation based on the minimum criteria. According to the coefficient calculated by LASSO regression, the individual risk score was calculated as follows:

TARGET-AML patients were categorized into high- and low-risk groups according to the median risk score. Moreover, the difference in survival between the two categories was assessed by the “survival” and “survminer” R packages. Principal component analysis (PCA) and t-distributed stochastic neighbor–embedding (t-SNE) analysis were conducted to visualize the distribution of various groups in terms of gene expression levels in the constructed model using the “Rtsne” and “ggplot2” R packages. Then, the “survival” and “time-ROC” R packages were utilized for 1-, 3-, and 5-year receiver operating characteristic (ROC) curve analyses. Moreover, the same formula and statistical methods were used to further validate the prognostic capacity of the model in the GEO (GSE192638) and TCGA databases.

Independent Prognostic Analysis of the Prognostic Risk-Score Model

The prognostic significance of the prognostic risk scoring model and other clinical characteristics were further investigated using univariate and multivariate Cox regression analyses. Ultimately, variables with p < 0.05 according to multivariate Cox regression analysis were considered independent prognostic factors for AML.

To further explore the relationship between clinicopathologic characteristics and the prognosis of pediatric AML patients, we applied the chi-squared test to explore the associations between the gene signature and clinicopathological characteristics and visualized the results with a heatmap using the “pheatmap” package. Then, the Wilcoxon signed-rank test and Kruskal–Wallis H test were utilized to compare the risk scores among various categories of these clinicopathological characteristics, and the boxplots present the results. For clinical characteristics analysis, patients were categorized into different groups based on clinical consensus: those with an age > 12 years were classified as 'juvenile', while the rest were labeled 'children'; those with a white blood cell (WBC) value > 10 * 10⁹ were assigned to the high WBC group; those with a bone marrow (BM) value > 70 were placed in the high BM group; those with a peripheral blood (PB) value > 30 were classified as the high PB group; and those with an FLT3.ITD. Patients with an AR > 0.5 were categorized into the high AR group.

Establishment of the predictive nomogram

The independent clinical features (FLT3. ITD mutation status, age, WBC count, and risk score) were used to construct a nomogram for 1-, 3-, and 5-year OS and EFS prediction. The nomogram was built using the “rms” R package. Time-dependent calibration curves and area under the curve (AUC) curves for 1, 3, and 5 years were plotted to assess the accuracy of the nomogram with 60 samples per iteration and 1000 samples. Furthermore, an alluvial diagram was drawn to show the changes in cuproptosis-related clusters, risk scores, age, and FLT3. ITD mutation using the “ggalluvial” R package. To better illustrate the role of our risk score in AML development, we compared the AUC for 2 years between our risk score and the previous risk score for adults in the TARGET pediatric cohort.

Functional enrichment analysis

To identify genes associated with pediatric AML patients, the TARGET cohort was divided into two groups according to the median risk score. Afterward, the differentially expressed genes (DEGs) (|logFC| > 1, p value < 0.05) between the two groups were determined using the limma package in R, and a volcano plot of the DEGs was drawn using the ggplot package. Then, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses of the DEGs were performed using the “clusterProfiler” R package ¹⁹. The GO terms included the following three categories: biological process (BP), cellular component (CC), and molecular function (MF).

Assessment of the tumor immune environment

To investigate the immune-related characteristics of the two groups of pediatric AML patients, immune cell infiltration and immune-related functions were quantified by single sample gene set enrichment analysis (ssGSEA) using the “GSVA” R package and the CIBERSORTx algorithm to evaluate the expression of different types of immune cells ^{20, 21}. For ssGSEA (single-sample GSEA), the gene expression of AML samples was used as an input to generate scores of 29 types of immune cells using the “gsva” R package. For CIBERSORTx, normalized gene expression data were uploaded to the web portal with the LM22 signature and 1,000 permutations, and the infiltration scores of 22 kinds of human immune cells in each AML sample were computed. Subsequently, the multigene correlation map was displayed by the R package “pheatmap.” In addition, we applied the Wilcoxon signed-rank test to estimate the differences in the expression of human lymphocyte antigen (HLA) signatures and immune checkpoint genes between the two groups.

Drug Sensitivity Prediction

To investigate the differences in the efficacy of routine clinical treatment between patients in the high- and low-risk categories, we downloaded treatment data from the TARGET datasets and performed K‒M analysis to assess the outcomes after gemtuzumab (GO) chemotherapy or stem cell transplantation (SCT) treatment in different groups. On the other hand, we utilized the risk score distribution of patients in different drug response groups to select efficient treatment drugs for a prognostic model. Based on the half-maximal inhibitory concentration (IC50) obtained from the Genomics of Drug Sensitivity in Cancer (GDSC) (https://www.cancerrxgene.org/) and clinical gene expression data, we used the “pRRophetic” R package to predict sensitivity to potential chemotherapeutic drugs. ^{22, 23}. A total of 2281 IRGs were obtained from the InnateDB database.

Primary AML sample collection and quantitative real-time PCR analysis

Bone marrow samples from 20 AML patients were obtained from the sample database of the First Affiliated Hospital of Sun Yat-sen University. Peripheral blood samples were donated by 10 healthy volunteers. Total RNA was extracted via TRIzol (Thermo Fisher, USA, 15596018) and then reverse transcribed into cDNA by HiScript III RT SuperMix (Vazyme, China, R323). Quantitative real-time polymerase chain reaction (qRT‒PCR) was performed to detect RNA expression using 2 × SYBR Green qPCR Master Mix (Toyobo, Japan, QPK-201) on a Quantstudio 3 Real-Time PCR instrument (Thermo Fisher, USA, A28567). qRT‒PCR was conducted in triplicate for each sample. All gene expression levels were normalized to that of β-actin using the 2−ΔΔCt method. The primers used in this study are listed in Supplemental Table 10.

Statistical analysis

R software (v4.2.2) and GraphPad Prism 9 were used to complete the statistical analyses and plot the data. Log-rank tests were used for identifying the significance of differences in the K‒M analysis. The area under the time-dependent ROC curve (AUC) served as an indicator of prognostic accuracy. For comparisons of immune cell infiltration and immune scores between the two risk groups, the Mann‒Whitney test was used. Unless otherwise indicated, Student’s t test, the Mann‒Whitney test, and the Wilcoxon signed-rank test were used to compare two groups. One-way ANOVA and the Kruskal–Wallis test were used for comparisons among three or more groups. The chi-square test was used for categorical variables. p ≤ 0.05 was considered to indicate statistical significance. Asterisks indicate that *p < 0.05, **p < 0.01, or ***p < 0.001.

Differences in the landscape of CRGs in pediatric and adult AML patients.

In a previous study, Li Pang et al. established a signature of four ferroptosis-related lncRNAs to predict the prognosis of AML patients in the TCGA cohort ¹⁴, which included only adult AML patients. To validate the prognostic value of these signatures in pediatric AML patients, we used univariate Cox analysis of four Cuproptosis-lncRNAs in pediatric datasets (TARGET), namely, LINC01679, AC133961.1, and AC093278.2. The three lncRNAs had no significant effect on the prognosis of patients with pediatric AML (Figure 1A). Then, we validated these results in the GEO database (Supplemental Figure 2A).

We analyzed the expression of Cuproptosis-related genes in adult and pediatric patients. We found that the expression of Cuproptosis-related genes in the pediatric cohort was higher than that in the adult cohort (Supplemental Figure 2B). Moreover, Kaplan‒Meier (K–M) analysis revealed that the prognostic significance of CRGs in the pediatric group (p = 0.016) (Figure 1B) was more pronounced than that in the adult group (p = 0.061) (Figure 1C). After that, we used the “Limma” package to analyze the genes related to differentially expressed CRGs between the pediatric (TARGET) and adult (TCGA) datasets. Subsequently, we compared the expression of CRGs between the adult and pediatric samples. A heatmap revealed differences in the gene expression levels of the CRGs (Supplemental Figure 2C). These results showed that CRGs have different values in pediatrics and adults.

After that, we summarized the incidence of somatic mutations and CNVs in pediatric and adult AML patients. The landscape of somatic variants in pediatric AML patients was different from that reported in adults, consistent with previous studies²; alterations in RAS genes (26%) were more common in children (Supplemental Figures 3A, 3B). Focusing on the CRGs, we found that the most frequently mutated genes in the pediatric samples were DHX15 (22%) and GSS (22%) (Figure 1D), whereas IDH2 (29%) was mutated in the adult samples (Figure 1E). Interestingly, we noticed that missense mutations were the most frequent. Single nucleotide polymorphisms (SNPs) were the most prevalent variant type, with C > T ranking at the top among the single nucleotide variant (SNV) classes for both the pediatric and adult groups (Supplemental Figures 3C, 3D). The exploration of CNVs revealed widespread CNV alterations in all these genes. Notably, in the pediatric group, the CNV of CDC16 was significantly increased, and the GCSH was significantly decreased, while in adults, UTP25 was significantly increased, and ATPAF2 was significantly decreased (Figure 1F, Supplemental Table 4). Overall, these results indicated that CRGs play different roles in adult and pediatric AML patients and that CRGs have stronger prognostic value in pediatric patients than in adults.

Identification of Cuproptosis Gene Clusters in Pediatric and Adult AML Patients

To further explore the expression features of CRGs in AML, we used a consensus clustering algorithm to identify pediatric and adult patients with AML based on the expression profiles of the CRGs. We found that k = 4 seemed to be the best alternative for categorizing both the pediatric and adult cohorts (Figure 2A, 2E), and this division was supported by the consensus CDF curve and optimal number in Nbclust (Figure 2B, 2F; Supplemental Figure 4A, 4B, 4D, 4E). Based on this, satisfactory separation across the four clusters was achieved according to the PCA and t-SNE plots (Figure 2C, 2G; Supplemental Figure 4C, 4F). As determined by the KM curves, the overall survival of the four subclusters in the pediatric cohort demonstrated that the CRGs could serve as a bioinformatics tool with prognostic value in clinical practice (p = 0.022) (Figure 2D). However, no significant difference was detected in the adult cohort (p = 0.067) (Figure 2H), and similar results were detected in the adult samples when the data were categorized under the k=2, 3, 5, and 6 conditions (Supplemental Figures 4G-4J). Together, the results confirmed that the CRGs had significant prognostic value in children with AML.

Construction of the Prognostic Signature of Cuproptosis-Related Genes in Pediatric AML Patients

To better investigate whether the cuproptosis-related gene set can be used as a predictive biomarker for pediatric AML patients. We further conducted a differentially

DEG analysis between the TARGET and GTEx datasets was performed using the "limma" R package (the cutoff was |log2FoldChange| > 1 and FDR < 0.05) (Figure 3A). Eight differentially expressed genes (DEGs) were identified. Moreover, we performed univariate Cox regression analysis and revealed that 120 CRGs were significantly associated with patient survival (p < 0.05) (Supplemental Table 5), among which five genes overlapped between the DECRGs and prognostic genes and were identified as prognostic DECRGs (Figure 3B). The specific prognostic value of the 8 CRGs was further assessed for three risk factors and two favorable factors. (Figure 3C). Overall, five prognostic DEG candidates were identified according to specific criteria.

Subsequently, based on the LASSO Cox regression analysis, these genes were analyzed using the minimum λ value (0.009) to construct a prognostic model (Figure 3D, 3E; Supplemental Table 6), leading to the identification of four risk genes (TRAIP, FAU, TEFM, and RPL19). TEFM and TRAIP exhibited oncogenic features, the overexpression of which was associated with worse survival in pediatric AML patients, while higher expression of FAU and RPL19 predicted better survival. The cuproptosis model was constructed as follows: Cuproptosis risk score = Exp (TRAIP) × (0.102) – Exp (FAU) × (0.132) + Exp (TEFM) × (0.465) – Exp (RPL19) × (0.529).

Then, the samples in the TARGET cohort (training group) were evenly divided into low- and high-risk groups (Figure 4A; Supplemental Table 7). The distribution of the risk score indicated that patients in the low-risk group experienced less cell death and longer survival than did those in the high-risk group (Figure 4B). In addition, the K‒M survival curves showed significant differences (p = 0.0011), with the low-risk group demonstrating a notably longer overall survival than the high-risk group (Figure 4C). Further PCA and t-SNE analyses revealed identifiable differences between the high-risk and low-risk groups (Figure 4D; Supplemental Figure 5A). Subsequently, time-dependent ROC analysis also revealed a favorable prognostic performance of this gene signature, with areas under the ROC curve (AUCs) of 0.61, 0.68, and 0.69 at 1-, 2-, and 3-year intervals, respectively (Figure 4E). In addition, the expression of risk genes was analyzed using the cuproptosis risk score, which revealed that TEFM and TRAIP were high-risk genes, while FAU and RPL19 were low-risk genes (Supplemental Figure 5B).

In the external validation cohort from the GEO, we calculated the risk scores of patients with the same formula and then categorized them into high-risk and low-risk groups (Supplemental Table 9). The findings from the risk score distribution plot and scatter plot analyses aligned with those from the training set (Figure 4F, G). It is worth noting that although the KM plot of the verification set showed a noticeable trend toward a prognostic advantage in the low-risk group, the difference was not significant (p = 0.25) (Figure 4H), possibly due to the limited amount of data. Similar to the results of the TARGET cohort, patients in the two groups of the GEO cohort were distributed in various directions based on the PCA and t-SNE analyses (Figure 4I; Supplemental Figure 5C). In addition, the area under the ROC curve (AUC) values were 0.54, 0.6, and 0.56, respectively (Figure 4J). The heatmap showed the same results as those in the training set (Supplemental Figure 5D).

Finally, we used the same algorithm to compute the risk scores in the TCGA datasets. Consistent with our hypothesis, the risk score distribution plot (Figure 4K, 4 L) and K‒M plot (Figure 4M) did not significantly differ among the adult patients. Although the tSNE plot showed a distinct distribution (Figure 4F), the PCA (Figure S5E) and heatmap (Supplemental Figure 5F) showed different results than those for children, and the ROC curves indicated worse predictive ability (Figure 4O). Our constructed individual-level cuproptosis-related risk signature showed a significant association with the survival of patients with pediatric AML, independent of age.

Independent prognostic value of the 4-gene signature and evaluation of clinical characteristics

To further enhance the applicability of the risk score in clinical settings, we performed univariate and multivariate Cox analyses of OS and EFS to assess the possibility of the risk score functioning as an independent prognostic factor (Supplemental Table 8). The results of univariate Cox regression analysis demonstrated that the risk score was a significant independent predictor of poor survival in the pediatric cohort (p < 0.0001, HR = 2.21, 95% CI = 1.53 - 3.18; Figure 5A). Multivariate analysis also revealed that the risk score was a critical prognostic factor for AML patients after accounting for other confounders (p = 0.008, HR = 1.84, 95% CI = 1.17-2.88; Figure 5B). In addition, we used the same algorithm to assess the functioning risk score using univariate and multivariate Cox analyses for EFS. The results were similar to those for OS, and univariate Cox regression analysis revealed that the risk scores were significantly different (p < 0.0001, hazard ratio (HR) = 2, 95% CI = 1.41). Figure 5C, 5D). After that, we compared the clinical characteristics of patients in different risk groups. Our analysis revealed that the risk score was significantly altered among the peripheral blood (PB group; p = 0.00083; Supplemental Figure 6D), event (p = 0.022; Supplemental Figure S6C), FAB (p = 0.00075; Supplemental Figure 6B) and risk (p = 0.016; Supplemental Figure 6A) groups between the high- and low-risk groups (Figure 5E). Moreover, the risk score had a statistically superior ability to predict overall survival (OS) compared to previously reported prognostic signatures in adult patients ¹⁴ (Supplemental Figure 6F). The AUC of Model 1 (previously performed for an adult patient) was 0.62, while that of Model 2 (we performed) was 0.69. These results confirmed the accuracy of this predictive model in pediatric patients.

Establishment of a Predictive Nomogram for AML Patients

Recognizing that the risk score alone was insufficient to predict the prognosis of AML patients, a nomogram incorporating the risk score was constructed to predict the 1-, 3-, and 5-year OS of AML patients based on the results of multivariate Cox regression analysis (Figure 6A). The subsequent calibration plot showed that the performance of the proposed nomogram was consistent with that of an ideal model (Figure 6B), with a calculated C index of 0.649. Furthermore, ROC curves indicated that this nomogram exhibited better prognostic performance, with AUCs of 0.62, 0.71, and 0.69 at 1, 3, and 5 years, respectively (Figure 6C). Moreover, an alluvial diagram was constructed to visualize variations in the aforementioned characteristics of AML patients (Supplemental Figure 6E). Then, we developed a nomogram containing our prognostic risk score model and multiple clinical factors related to EFS according to the same algorithm (Figure 6D). The calculated C index was 0.633, and the calibration plots of the 1-, 3- and 5-year EFSs showed no deviations from the Platt calibration curves (Figure 6E). ROC curves revealed AUCs of 0.69, 0.69, and 0.72 for 1-, 3-, and 5-year survival, respectively (Figure 6F).

Functional annotation analysis of the 4-gene signature

To further investigate the potential biological functions and pathways of the 4-gene signature, the DEGs across the high-risk and low-risk categories were subjected to GO and KEGG analyses (Supplemental Figure 7A). Furthermore, these DEGs were significantly enriched in biological processes and molecular functions related to immunity, such as the regulation of T-cell proliferation and activation involved in the immune response, the regulation of leukocyte proliferation, mononuclear cell proliferation and lymphocyte proliferation (Figure 7A). In addition, KEGG analysis revealed several cancer-related pathways, such as the hematopoietic cell lineage, ECM-receptor interaction, Hippo, TGF-beta, MPK, and P13K-Akt signaling pathways, in the cohorts (Figure 7B). These results revealed that the cuproptosis-related 4-gene signature was significantly associated with cancer progression and strongly influenced the immunoregulation of the TME.

Tumor immune microenvironment analysis of the 4-gene signature

Next, we further explored the association between the cuproptosis-related gene signature and the tumor immune microenvironment. We observed a significant correlation among the low-risk patients within the TARGET cohort. Comparisons of 29 immune signatures identified by the ssGSEA algorithm revealed that high-risk patients exhibited increased infiltration of activated CD4 T cells, effector memory CD4 T cells, memory B cells, and type 2 T helper cells but decreased infiltration of CD56 bright natural killer cells, CD56 dim natural killer cells, monocytes, T follicular helper cells, and type 1 T helper cells (Figure 8C). Furthermore, based on the CIBERSORT algorithm, we compared the distributions of 22 kinds of immune cells in diverse risk subgroups. A significant difference in the distribution of immune cells was observed in high-risk patients with superior infiltration of resting NK cells and neutrophils (Figure 8D). Concurrently, we analyzed the association between the 4-gene signature and the infiltration score of immune cells, which further suggested that the 4-gene signature was significantly associated with the six kinds of immune cells (Supplemental Figure 7B).

Considering the critical role of HLA-related genes in regulating the immune response, we compared the expression of HLA-related genes across different risk groups (Supplemental Figure 7C). Furthermore, we investigated the correlation between 33 immune checkpoint molecules and the risk score (Supplemental Figure 7D). The results demonstrated that high-risk patients exhibited significantly greater expression of VTCN1, CD160, CD28, CTLA4, PDCD1, TIGIT, CD80, IDO1, IDO2, and CD274 and lower expression of CD44 and LGALS9 than low-risk patients (Supplemental Figure 7D).

Two Risk Groups Showed Inconsistent Sensitivities to Different Drugs

The sensitivity of AML patients within different risk groups to gemtuzumab (GO) chemotherapy and stem cell transplantation (SCT) treatment was analyzed in the TARGET datasets. However, the KM plot revealed that GO and SCT treatment did not significantly improve the prognosis of AML patients in either the high-risk or the low-risk group. Although treatment data for survival analysis were limited, we also found that common routine clinical treatments were not specific to high-risk patients.

Moreover, to predict the precise treatment of high-risk patients, we further selected the drugs currently used in AML treatment to assess their sensitivity in both high-risk and low-risk subgroups. Interestingly, we observed that patients in the high-risk group presented significantly lower IC50 values for RSL3, KPT185, neopeltolide, gemcitabine, oligomycin A, and cucurbitacin.1 than did those in the low-risk group (Figure 8C), while parbendazole, SNS.032, SNX.2112, ouabain, AZD8055, and alvacidib presented significantly greater IC50 values. Therefore, patients with a low CRG risk score might exhibit better treatment benefits from these drugs.

qPCR analysis of the expression levels of four cuproptosis-related genes in AML samples

The expression levels of four cuproptosis-related genes were detected in clinical AML patients and healthy individuals through RT‒qPCR. The patients’ clinical characteristics and risk gene expression data are presented in Supplemental Table 9. The results revealed no significant difference in the expression of FAU between the two groups (Figure 9A, P=0.1168), whereas the TEFM, RPL19, and TRAIP expression levels were significantly greater in the AML samples than in the normal samples (Figure 9B, P<0.0001; Figure 9C, P=0.0033; Figure 9D, P<0.0001).

Pediatric AML patients exhibit distinctive and critically important characteristics that are different from those of adults ^{24, 25}. Approximately 30% of all pediatric AML patients have a poor prognosis ²⁶. This is partly because current prognostic schemas classify many children who ultimately succumb to their disease as low or intermediate risk. Several novel prognostic models for AML, such as those involving cuproptosis-related genes, have been proposed in recent years based on regulated cell death (RCD) ¹⁴, but these models are limited to pediatric patients. Therefore, our study aimed to assess the prognostic potential of the CRG signature specifically in pediatric patients. We further constructed a novel prognostic model tailored to the unique characteristics of pediatric AML patients. Finally, we evaluated the accuracy of this predictive model through both bioinformatics and experimental validation.

The recent TARGET AML dataset showed different features of age-specific AML disease, necessitating a deeper understanding of this disease to drive pediatric-specific therapy ². We proved that the CRG signature, proposed based on the adult cohort, lost its prognostic ability in pediatric patients (Figure 1A; Supplemental Figure 2A). We observed that the pediatric patients exhibited lower expression levels of CRGs than did the adults, most of which were associated with the family encoding mitogenomes (COX411, RPL19, RPL10A, etc.), mitochondrial respiratory chain-related genes (UQCRC2), and tricarboxylic acid cycle (TCA)-related genes (CSs). Further analysis of the mutation status and CNV revealed that CNV was lost in the mitogenome (MFF, MRPL41m MRPL44) and that there was a mutation in glutathione synthetase (GSS) within the pediatric cohort. All of these genes are vital components of Cuproptosis, an unconventional mechanism of cell death involving protein lipoylation in the TCA cycle. These different expression levels of CRGs resulted in a more significant difference in survival in pediatric patients (Figure 1B, 1C). In addition, these findings confirm that the present CRG signature can be used as a potentially reliable and distinctive prognostic biomarker for pediatric AML patients.

Cuproptosis, a novel type of regulated cell death triggered by excess Cu²⁺, is poorly understood ²⁷. In our study, we identified a novel signature composed of four CRGs (FAU, RPL19, TEFM, and TRAIP) that accurately and independently predicted OS in pediatric AML patients. Then, we constructed and validated the prognostic scoring model with internal and external databases, showing its outstanding prognostic performance. FAU, known for its role in ribosome biosynthesis ²⁸, potentially functions as a tumor suppressor gene ²⁹. RPLP2 is involved in carcinogenesis by affecting the translation of specific cellular mRNAs ³⁰, and RPL19 has been proven to be a prognostic marker for pediatric AML ³¹. Moreover, TEFM is a key molecule for the mitochondrial DNA (mtDNA) replication-transcription switch ³², and TRAIP is an E3 ubiquitin-protein ligase ³³. Both genes have been proven to be involved in the prognosis of many cancers ^{34, 35}. Our findings further revealed that these genes serve as favorable prognostic indicators for survival outcomes in pediatric AML patients, consistent with previous studies.

Patients divided into high- and low-risk groups according to the cuproptosis risk model had significantly different prognoses. We further explored the reasons for the difference in prognosis. Interestingly, the GO analysis showed obvious enrichment of immune mechanisms (Figure 9A), consistent with the known importance of copper in cellular and humoral immunity ³⁶. Then, we assessed the association between the cuproptosis risk score and immune cell abundance. The results showed that the risk score was positively correlated with activated and effector memory CD4 T cells, memory B cells, and follicular helper T cells but negatively correlated with activated NK and resting CD4⁺ memory T cells, both of which are known for their pivotal roles in the antitumor immune response ³⁷. Studies have reported that patients with high memory T-cell activation have shorter OS, whereas those with high memory T-cell quiescence have longer OS ³⁸. NK cells play a central role in innate immunity by eliminating cells lacking MHC-I molecules and evading T-cell recognition ²¹, whereas tumorigenesis-, angiogenesis- and immune suppression-related cell types play a central role ^{39, 40}. Our results revealed a decrease in the number of NK cells, an increase in the number of NK cells and a poor prognosis in the high-risk group. In addition, immune escape genes such as PD-L1 (CD274) and IDO1 were upregulated in patients with high CRG scores. We also found that most HLA-related genes were expressed at higher levels in the low-risk group, consistent with the notion that HLA presentation promotes tumor-associated antigen recognition ⁴¹.

Current therapies for pediatric AML have largely been adapted from adult AML treatment, but the cure rates with standard intensive chemotherapy vary between pediatric and adult AML patients. To gain deeper insights into the role of the prognostic risk score model in pediatric AML, we performed an anti-neoplastic drug sensitivity analysis of the different groups. The results showed that the high-risk group responded better to several drugs, including gemcitabine, oligomycin A, cucurbitacin I (a JAK2/STAT3 inhibitor), and ferroptosis inducers (X1S.3R.RSL3). Furthermore, KPT185 (an antileukemia agent) may contribute to the prognosis of patients in the high-risk group, who are less sensitive to most drugs, including SNX.2112, alvocidib, and albendazole. These findings suggested the use of different treatment approaches for different risk groups. Therefore, this risk score can provide a therapeutic indication and support its translational value.

The following points remained inadequate in our study. First, our prognostic model was constructed from retrospective data from public databases. Although it has been validated using a qPCR expression test, substantial prospective studies and additional in vivo and in vitro experiments are still required to validate its predictive power. Second, the CRGs used in this study were identified through a genome-wide CRISPR-Cas9 loss-of-function test reported in the previous literature. The exact roles of these genes in the process of cuproptosis remain unclear and need further investigation. In addition, imperfections in the available information have contributed to unsatisfactory results. Public databases do not provide sufficient clinical information or information on the efficacy of drug therapy. As a result, the risk model could not accurately reflect the patient's current state and treatment effect, and more comprehensive real-world data are needed to validate its clinical utility. In the future, we should continue to investigate the role of these four genes in AML pathology through experimental and clinical studies. We believe that our prognostic model, based on CRGs, holds significant value in predicting the survival of pediatric AML patients and will offer novel insight for AML research.

We performed a comprehensive analysis of the expression profiles and mutation landscapes of cuproptosis-related genes in both adult and pediatric AML patients. Our study demonstrated that these cuproptosis-related genes may play distinct roles in pediatric patients compared to adults. Then, we established a specific cuproptosis risk model for pediatric patients, which revealed worse survival outcomes and greater immune infiltration in children with high CGRP scores. The risk model has also been proven to effectively guide clinical practice and allows for more personalized treatment strategies for predicting chemotherapeutic effectiveness. Our findings can be used to design individualized treatment strategies for children with different CRG subtypes and provide novel insights for developing specific pediatric AML research.

AML: acute myeloid leukemia

RCD: regulated cell death

CRGs: Cuproptosis related genes

DEGs: differentially expressed genes

BM: Bone Marrow

WBC: white blood cell

PB: peripheral blood

HLA: human lymphocyte antigen

pAML: pediatric acute myeloid leukemia

aAML: adult acute myeloid leukemia

CNV: copy number variation

LASSO: least absolute shrinkage and selection operator

PCA: principal component analysis

KM: Kaplan–Meier

ROC: receiver operator characteristic

GO: Gemtuzumab

SCT: Stem Cell Transplant

Ethics approval and consent to participate

This study received approval from the Institutional Ethics Committee (IEC) for Clinical Research and Animal Trials of the First Affiliated Hospital of Sun Yat-sen University (reference number [2023]070). All participants provided written consent before donating bone marrow or peripheral blood specimens. All participants were anonymous.

Consent for publication

Not applicable.

Availability of data and materials

The data used for this study will be made available upon reasonable request from the corresponding author.

Competing interests

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as potential conflicts of interest.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The study was supported by the Guangdong Basic and Applied Basic Research Foundation [Grant No. 2023A1515030060], the Youth Talent Support Programme of Guangdong Provincial Association for Science and Technology [Grant No. SKXRC202316], the Science and Technology Planning Project of Guangdong Province, China [Grant No. 2021A0505060004], and the Shenzhen Fundamental Research Program [Grant No. JCYJ20190809142619278]. Dongguan Social Development Science and Technology Key Project (Project Number: 20211800904872). Shenzhen Science and Technology Innovation Commission, grant number:JCYJ20190809160609727.

Author contributions

SYS and YLT designed the study. SYS, YJC and LBH analyzed the data. XQL, LBH and YLT supervised the study. ZHZ and CL performed data preprocessing. CL, HMX implemented statistical analysis. MYX and XLZ developed the model. YL and JSL conducted the data analysis. ZF and LDZ assisted editing and sorted data. SYS and YLT drafted the manuscript. XQL, LBH, YLT and SYS revised the manuscript.

Acknowledgments

We extend our gratitude to the members of the Pediatric Department at The First Affiliated Hospital of Sun Yat-sen University, including Huang Xiuya, Cui Xiaoyu, Yue Tianfang, Li Zhen, Chen Zhiyan, Chen Runyu, Lv Xiaoyang, Yan Juan, Yang Cuiyun, Liu Yingli, and Wang Lina, for their invaluable support in this work.

Tarlock K, Zhong S, He Y, et al. Distinct age-associated molecular profiles in acute myeloid leukemia defined by comprehensive clinical genomic profiling. Oncotarget 2018; 9: 26417-26430. 2018/06/15. DOI: 10.18632/oncotarget.25443.
Bolouri H, Farrar JE, Triche T, Jr., et al. The molecular landscape of pediatric acute myeloid leukemia reveals recurrent structural alterations and age-specific mutational interactions. Nat Med 2018; 24: 103-112. 2017/12/12. DOI: 10.1038/nm.4439.
Stratmann S, Yones SA, Mayrhofer M, et al. Genomic characterization of relapsed acute myeloid leukemia reveals novel putative therapeutic targets. Blood Adv 2021; 5: 900-912. 2021/02/10. DOI: 10.1182/bloodadvances.2020003709.
Mrozek K, Marcucci G, Paschka P, et al. Clinical relevance of mutations and gene-expression changes in adult acute myeloid leukemia with normal cytogenetics: are we ready for a prognostically prioritized molecular classification? Blood 2007; 109: 431-448. 2006/09/09. DOI: 10.1182/blood-2006-06-001149.
Cheng C-K, Yung Y-L, Chan H-Y, et al. Deep genomic characterization highlights complexities and prognostic markers of pediatric acute myeloid leukemia. Commun Biol 2023; 6: 356. DOI: 10.1038/s42003-023-04732-2.
Wang H, Chan KYY, Cheng CK, et al. Pharmacogenomic Profiling of Pediatric Acute Myeloid Leukemia to Identify Therapeutic Vulnerabilities and Inform Functional Precision Medicine. Blood Cancer Discov 2022; 3: 516-535. DOI: 10.1158/2643-3230.BCD-22-0011.
Li S, Garrett-Bakelman FE, Chung SS, et al. Distinct evolution and dynamics of epigenetic and genetic heterogeneity in acute myeloid leukemia. Nat Med 2016; 22: 792-799. 2016/06/21. DOI: 10.1038/nm.4125.
Vanden Berghe T, Linkermann A, Jouan-Lanhouet S, et al. Regulated necrosis: the expanding network of non-apoptotic cell death pathways. Nat Rev Mol Cell Biol 2014; 15: 135-147. 2014/01/24. DOI: 10.1038/nrm3737.
Tang D, Chen X and Kroemer G. Cuproptosis: a copper-triggered modality of mitochondrial cell death. Cell Res 2022; 32: 417-418. 2022/04/01. DOI: 10.1038/s41422-022-00653-7.
Tsvetkov P, Coy S, Petrova B, et al. Copper induces cell death by targeting lipoylated TCA cycle proteins. Science 2022; 375: 1254-1261. 2022/03/18. DOI: 10.1126/science.abf0529.
Hedley D, Shamas-Din A, Chow S, et al. A phase I study of elesclomol sodium in patients with acute myeloid leukemia. Leuk Lymphoma 2016; 57: 2437-2440. 2016/01/07. DOI: 10.3109/10428194.2016.1138293.
Xu B, Wang S, Li R, et al. Disulfiram/copper selectively eradicates AML leukemia stem cells in vitro and in vivo by simultaneous induction of ROS-JNK and inhibition of NF-kappaB and Nrf2. Cell Death Dis 2017; 8: e2797. 2017/05/19. DOI: 10.1038/cddis.2017.176.
Sanaat Z, Nouri M, Hajipour B, et al. Evaluation of Copper, Zinc, Cu/Zn, and VEGF in Patients with AML in Iran. Iran J Cancer Prev 2011; 4: 151-153. 2011/10/01.
Li P, Li J, Wen F, et al. A novel cuproptosis-related LncRNA signature: Prognostic and therapeutic value for acute myeloid leukemia. Front Oncol 2022; 12: 966920. 2022/10/25. DOI: 10.3389/fonc.2022.966920.
Wang H, Chan KYY, Cheng CK, et al. Pharmacogenomic Profiling of Pediatric Acute Myeloid Leukemia to Identify Therapeutic Vulnerabilities and Inform Functional Precision Medicine. Blood Cancer Discov 2022; 3: 516-535. 2022/08/13. DOI: 10.1158/2643-3230.BCD-22-0011.
Frankish A, Diekhans M, Ferreira AM, et al. GENCODE reference annotation for the human and mouse genomes. Nucleic Acids Res 2019; 47: D766-D773. 2018/10/26. DOI: 10.1093/nar/gky955.
Wilkerson MD and Hayes DN. ConsensusClusterPlus: a class discovery tool with confidence assessments and item tracking. Bioinformatics 2010; 26: 1572-1573. 2010/04/30. DOI: 10.1093/bioinformatics/btq170.
Tibshirani R. The lasso method for variable selection in the Cox model. Stat Med 1997; 16: 385-395. 1997/02/28. DOI: 10.1002/(sici)1097-0258(19970228)16:4<385::aid-sim380>3.0.co;2-3.
Papaemmanuil E, Gerstung M, Bullinger L, et al. Genomic Classification and Prognosis in Acute Myeloid Leukemia. N Engl J Med 2016; 374: 2209-2221. 2016/06/09. DOI: 10.1056/NEJMoa1516192.
Wang Y, Cai YY, Herold T, et al. An Immune Risk Score Predicts Survival of Patients with Acute Myeloid Leukemia Receiving Chemotherapy. Clin Cancer Res 2021; 27: 255-266. 2020/12/03. DOI: 10.1158/1078-0432.CCR-20-3417.
Liang Y, Hu F, Dai Y-J, et al. An Immune Risk Score Predicts Survival of Patients with Myelodysplastic Syndrome. Blood 2020; 136: 33-34. DOI: 10.1182/blood-2020-137349.
Paul Geeleher NJCaRSH. Clinical drug response can be predicted using baseline gene expression levels and in vitro drug sensitivity in cell lines. Genome Biol 2014; 2014 Mar 3;15(3):R47. DOI: 10.1186/gb-2014-15-3-r47.
Yang W, Soares J, Greninger P, et al. Genomics of Drug Sensitivity in Cancer (GDSC): a resource for therapeutic biomarker discovery in cancer cells. Nucleic Acids Res 2013; 41: D955-961. 2012/11/28. DOI: 10.1093/nar/gks1111.
Gruber TA, Larson Gedman A, Zhang J, et al. An Inv(16)(p13.3q24.3)-encoded CBFA2T3-GLIS2 fusion protein defines an aggressive subtype of pediatric acute megakaryoblastic leukemia. Cancer Cell 2012; 22: 683-697. 2012/11/17. DOI: 10.1016/j.ccr.2012.10.007.
Ostronoff F, Othus M, Gerbing RB, et al. NUP98/NSD1 and FLT3/ITD coexpression is more prevalent in younger AML patients and leads to induction failure: a COG and SWOG report. Blood 2014; 124: 2400-2407. 2014/08/26. DOI: 10.1182/blood-2014-04-570929.
Perry H, Morrow M, Davis T, et al. Care Groups II: A Summary of the Child Survival Outcomes Achieved Using Volunteer Community Health Workers in Resource-Constrained Settings. Glob Health Sci Pract 2015; 3: 370-381. 2015/09/17. DOI: 10.9745/GHSP-D-15-00052.
Chen L, Min J and Wang F. Copper homeostasis and cuproptosis in health and disease. Signal Transduct Target Ther 2022; 7: 378. 2022/11/23. DOI: 10.1038/s41392-022-01229-y.
Hochstrasser M. Biochemistry. All in the ubiquitin family. Science 2000; 289: 563-564.
L Michiels EVdR, F Van Hasselt, K Kas, J Merregaert. fau cDNA encodes a ubiquitin-like-S30 fusion protein and is expressed as an antisense sequence in the Finkel-Biskis-Reilly murine sarcoma virus. Oncogene 1993; 1993 Sep;8(9):2537-46.
Campos RK, Wong B, Xie X, et al. RPLP1 and RPLP2 Are Essential Flavivirus Host Factors That Promote Early Viral Protein Accumulation. J Virol 2017; 91 2016/12/16. DOI: 10.1128/JVI.01706-16.
Cai Z, Wu Y, Zhang F, et al. A three-gene signature and clinical outcome in pediatric acute myeloid leukemia. Clin Transl Oncol 2021; 23: 866-873. 2020/08/31. DOI: 10.1007/s12094-020-02480-x.
Soriano ME and Scorrano L. Traveling Bax and forth from mitochondria to control apoptosis. Cell 2011; 145: 15-17. 2011/04/05. DOI: 10.1016/j.cell.2011.03.025.
Christophe Chapard DHaMH. The TRAF-interacting protein (TRAIP) is a novel E2F target with peak expression in mitosis. Oncotarget 2015; Aug 28;6(25):20933-45. DOI: 10.18632/oncotarget.3055.
Guo Z, Zeng Y, Chen Y, et al. TRAIP promotes malignant behaviors and correlates with poor prognosis in liver cancer. Biomed Pharmacother 2020; 124: 109857. 2020/01/24. DOI: 10.1016/j.biopha.2020.109857.
Li S, Wang W, Zi J, et al. Elevated expression of mitochondrial transcription elongation factor (TEFM) predicts poor prognosis in low grade glioma-an analysis of the Cancer Genome Atlas (TCGA) dataset. Transl Cancer Res 2020; 9: 3610-3622. 2020/05/01. DOI: 10.21037/tcr.2020.04.16.
de Rooij JD, Zwaan CM and van den Heuvel-Eibrink M. Pediatric AML: From Biology to Clinical Management. J Clin Med 2015; 4: 127-149. 2015/08/04. DOI: 10.3390/jcm4010127.
Watanabe T. The Tumor Microenvironment in Follicular Lymphoma: Its Pro-Malignancy Role with Therapeutic Potential. Int J Mol Sci 2021; 22 2021/06/03. DOI: 10.3390/ijms22105352.
Cao J, Yang X, Li J, et al. Screening and Identifying Immune-Related Cells and Genes in the Tumor Microenvironment of Bladder Urothelial Carcinoma: Based on TCGA Database and Bioinformatics. Front Oncol 2019; 9: 1533. 2020/02/06. DOI: 10.3389/fonc.2019.01533.
Nahas MR, Stroopinsky D, Rosenblatt J, et al. Hypomethylating agent alters the immune microenvironment in acute myeloid leukaemia (AML) and enhances the immunogenicity of a dendritic cell/AML vaccine. Br J Haematol 2019; 185: 679-690. 2019/03/05. DOI: 10.1111/bjh.15818.
Fu Y, Liu S, Zeng S, et al. From bench to bed: the tumor immune microenvironment and current immunotherapeutic strategies for hepatocellular carcinoma. J Exp Clin Cancer Res 2019; 38: 396. 2019/09/11. DOI: 10.1186/s13046-019-1396-4.
Rizvi NA, Hellmann MD, Snyder A, et al. Cancer immunology. Mutational landscape determines sensitivity to PD-1 blockade in non-small cell lung cancer. Science 2015; 348: 124-128. 2015/03/15. DOI: 10.1126/science.aaa1348.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Comprehensive analysis of the ability of the Cuproptosis- related gene signature to predict the prognosis of patients with pediatric acute myeloid leukemia

Status:

Version 1

Abstract

Figures

Background

Methods

Results

Discussion

Conclusion

Abbreviations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1