Sage Journals: Discover world-class research

Abstract

BACKGROUND:

Acute myeloid leukemia (AML) has a poor prognosis, and the current 5-year survival rate is less than 30%.

OBJECTIVE:

The present study was designed to identify the significant genes closely related to AML prognosis and predict the prognostic value by constructing a risk model based on their expression.

METHODS:

Using bioinformatics (Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, univariate and multivariate Cox regression analysis, Kaplan-Meier survival analysis, and receiver operating characteristic (ROC) analysis) to identify a prognostic gene signature for AML. Finally, The Cancer Genome Atlas (TCGA) database was used to validate this prognostic signature.

RESULTS:

Based on univariate and multivariate Cox regression analysis, eighteen prognostic genes were identified, and the gene signature and risk score model were constructed. Multivariate Cox analysis showed that the risk score was an independent prognostic factor [hazard ratio (HR) $=$ 1.122, 95% confidence interval (CI) $=$ 1.067–1.180, $P<$ 0.001]. ROC analysis showed a high predictive value of the risk model with an area under the curve (AUC) of 0.705.

CONCLUSIONS:

This study evaluated a potential prognostic signature with eighteen genes and constructed a risk model significantly related to the prognosis of AML patients.

Keywords

Acute myeloid leukemia prognostic value bioinformatics analysis TARGET TCGA

1. Introduction

AML is a hematopoietic system malignant hyperplasia disease [1, 2]. The risk model and prognosis assessment are important in treating AML patients [3, 4, 5]. The treatment effect of patients with AML is still not optimistic, and the 5-year overall survival (OS) rate of patients is less than 30% [6].

AML is the most common acute leukemia in adults. Currently, chemotherapy is still the main treatment for AML patients, but around 70% of patients who have achieved remission eventually relapse and evolve into refractory leukemia, leading to treatment failure. Usually, patients diagnosed with AML often have severe clinical symptoms [7]. As a result, the disease progression of early untreated AML patients is very rapid, and the mortality rate is also very high [8]. Novel therapies for AML, including genetic and epigenetic targeted drugs, immunotherapies, and refinements of conventional cytotoxic chemotherapies, have significantly improved patient outcomes in recent years [9, 10, 11]. In addition, with the continuous exploration of the pathogenesis of AML at the cellular and molecular levels, the pathogenic genes of AML have been continuously discovered, and the development of new drugs for AML has also made new progress [12, 13, 14]. A previous study demonstrated that the co-inheritance of FLT3, CD177, and TTPAL contributed to AML [15]. Some researchers have found that fms-like tyrosine kinase 3 (FLT3) gene internal tandem duplications (FLT3-ITD) are frequently found in adult AML [16, 17, 18, 19]. AML patients with FLT3-ITD mutation have a higher relapse rate and poorer survival than AML patients without FLT3-ITD mutation [20, 21]. Previous studies have shown that FHL1 was a powerful prognostic candidate and potential therapeutic target in acute myeloid leukemia [22, 23]. Despite the continuous improvement of studies in this area, AML has not been completely cured or achieved an ideal survival duration in clinical practice [24], and the current prognostic evaluation system of AML is insufficient. Thus, the exploration of effective prognostic biomarkers is pivotal for improving the clinical outcomes of AML.

The present study was designed to pursue the molecular mechanism of AML using comprehensive bioinformatics analysis and build an applicable model to predict the survival probability of AML patients and provide new research ideas and potential targets for treating AML. To explore the genes related to the prognosis of AML patients, we obtained 187 bone marrow samples with AML and 37 normal bone marrow samples from the GDC TARGET database. We analyzed the RNA expression profiles of these samples and identified 1264 differentially expressed genes (DEGs). GO enrichment and KEGG pathway analyses performed functional annotations on these genes. We identified eighteen prognostic genes from GDC TARGET AML datasets using univariate Cox regression analysis. To predict the prognosis of AML, we constructed an eighteen prognostic genes signature using INPP4A, CFLAR, LNPEP, KLRF1, APAF1, UTRN, TNPO3, G3BP1, ZNF276, LPCAT1, PSD4, ADCY7, HECTD3, VPS52, ANAPC2, SPG11, PHRF1, and FAM111A. We divided AML patients into high-risk and low-risk groups according to the median risk score. OS analysis showed that AML patients in the high-risk group had a worse prognosis than those in the low-risk group. Finally, the TCGA database was used to validate this prognostic signature.

2. Materials and methods

2.1 The workflow of the model construction process

A brief workflow for this study is shown in Fig. 1.

Figure 1.

The workflow of the model construction process.

2.2 Data collection

The normalized GDC TARGET-AML datasets (cohort: GDC TARGET-AML) were downloaded from UCSC Xena, which provides 187 AML patients with RNA expression profiles based on high-throughput sequencing (HTS) and detailed clinical information. The GDC TARGET-Bone Marrow Normal datasets were downloaded from the TCGA database. The TCGA database, which provides independent GDC TCGA AML datasets (cohort: GDC TCGA Acute Myeloid Leukemia), was utilized to validate this prognostic signature. Only samples whose sample identities (IDs) end in “A” were retained.

2.3 Cell culture

Human bone marrow stromal cell line HS-5 was cultured in DMEM supplemented with 10% heat-inactivated fetal bovine serum (FBS), 1% penicillin-streptomycin, and 2% L-Glutamine. Human acute myeloid leukemia cell line HL-60 was cultured in RPMI-1640 supplemented with 10% fetal calf serum (FCS), 100 units/ml penicillin, 100 mg/ml streptomycin, and 2 mM L-Glutamine. Cell cultures were maintained in a humidified incubator at 37.5 ${}^{\circ}$ C in 5% CO ${}_{2}$ . HS-5 and HL-60 cells were obtained from the American Type Culture Collection (ATCC, United States). All media, supplements, and antibiotics were obtained from Sigma Aldrich.

2.4 Functional enrichment analysis

The “tidyverse” and “DESeq2” packages [25] in R (version 4.1.2) were used to identify DEGs, then GO and KEGG analyses were performed using the packages of “tidyverse”, “BiocManager”, “org.Hs.eg.db”, and “clusterProfiler” in R [26]. $|$ Log2 (fold change) $|$ $>$ 1 and adjusted $P$ -value $<$ 0.05 were selected as the significantly differentially expressed genes. GO is an internationally standardized gene function classification system. There are a total of three ontologies, which describe molecular functions (MF), cellular components (CC), and biological processes (BP) [27]. The KEGG is a knowledge base for systematically analyzing gene functions, and linking genomic information with higher-order functional information [28, 29].

2.5 Construction of the eighteen prognostic genes’ signature

Eighteen prognostic genes were identified using univariate Cox regression analysis by the packages of “tidyverse”, “survival” and “forest plot” in R. The predictive ability of these genes was assessed by the Kaplan-Meier survival analysis and log-rank test ( $P<$ 0.05) [30] using the packages of “tidyverse”, “survival”, and “survminer” in R [31]. Then these prognostic genes were used to construct a multivariate Cox regression model using the packages of “tidyverse” and “survival” in R. The multivariate Cox analysis was used to obtain the coefficients of these genes, constructing the eighteen prognostic genes-associated signatures [32]. The risk score model was performed using the packages of “tidyverse”, “survival”, and “ggrisk” in R.

2.6 Quantitative real-time polymerase chain reaction (qRT-PCR)

qRT-PCR was used to verify the results. Five of the eighteen genes were validated between HS-5 and HL-60 cells. Total RNA was isolated from cells using TRIzol reagent (Tiangen Biotech, China). To obtain mRNA, total RNA was reverse transcribed to cDNA, and qRT-PCR was performed using a SYBR Green PCR Kit (TaKaRa Bio, Japan). The expression levels were analyzed using the 2 ${}^{-\Delta\Delta\text{Ct}}$ method. GAPDH was used for normalization. The specific primer sequences are presented below: KLRF1 (F: 5’-TACTGGGAATATCTGGAACCGT-3’; R: 5’-TTGAGCCATTCTGATTGGCAT-3’); INPP4A (F: 5’-TGGCAGGAAATATACAAGACCCA-3’; R: 5’-TCGATCTAGCGATGGAGTATGC-3’); ADCY7 (F: 5’-GATGTACGTCGAGTGTCTCCT-3’; R: 5’-CTTTGTCCATGCGTCGAACA-3’); UTRN (F: 5’-TGAAGGATGTCATGTCGGACC-3’; R: 5’-GTTGAGGACGTTGACTTGGCT-3’); LPCAT1 (F: 5’-CGCCTCACTCGTCCTACTTC-3’; R: 5’-TTCCCCAGATCGGGATGTCTC-3’); GAPDH (F: 5’-GGAGTCCACTGGCGTCTTCA-3’; R: 5’-GTCATGAGTCCTTCCACGATACC-3’).

2.7 Statistical analysis

Our study used R software (version 4.1.2) to perform all statistical analyses. Univariate Cox regression analysis was performed for the DEGs to screen OS-associated genes. Multivariate Cox regression was used to construct a risk model. ROC analysis [33, 34] and the calculation of the AUC were analyzed by the “tidyverse”, “ROCR”, and “rms” packages of R software. A $P$ -value $<$ 0.05 was considered as statistical significance.

3. Results

3.1 Characteristics of AML patients in the dataset from the GDC TARGET database

We obtained 187 patient samples and clinical information from the GDC TARGET database. Among them, 96 (51.3%) patients were female, and 91 (48.7%) were male. The age at initial diagnosis ranged from 0.4 to 22.6 years, with a median age of 8.8 years. The clinical characteristics of the patients are listed in Table 1, including white blood cell (WBC) count at diagnosis ( $\times$ 10 ${}^{9}$ /L), vital status, protocol, central nervous system (CNS) disease, and French-American-British (FAB) category. In vital status, 91 (48.7%) patients were alive, and 96 (51.3%) died. There were 175 (93.6%) patients without CNS disease. In the FAB category, M4 had the highest proportion of all subtypes.

Table 1
Characteristics of patients

Characteristic	Number of cases	Percentages (%)
Patients	187	100
Age at diagnosis (year)
Median	8.8	/
Range	0.4–22.6	/
Sex
Female	96	51.3
Male	91	48.7
WBC count at diagnosis ( $\times$ 10 ${}^{9}$ /L)
$\geqslant$ 50	83	44.4
$<$ 50	104	55.6
Vital status
Alive	91	48.7
Dead	96	51.3
Protocol
AAML0531	117	62.6
CCG-2961	27	14.4
AAML03P1	43	23.0
CNS disease
Yes	12	6.4
No	175	93.6
FAB category
M0	4	2.1
M1	21	11.2
M2	45	24.1
M4	52	27.8
M5	35	18.7
M6	3	1.6
M7	8	4.3
NOS	10	5.3
Unknown	9	4.8

3.2 Identification of DEGs

We compared 187 AML samples and 37 bone marrow normal samples from the GDC TARGET database. 1,264 DEGs were identified, including 693 significantly up-regulated genes and 571 significantly down-regulated genes, as shown in the heat map by the “tidyverse” and “pheatmap” packages of R software (Fig. 2A) and the volcano plot using the packages of “tidyverse”, “ggpubr” and “ggthemes” in R (Fig. 2B).

Figure 2.

Identification of differentially expressed genes. (A) Heat map of differentially expressed gene expression levels in the GDC TARGET dataset. (B) Volcano plot for differentially expressed gene expression levels in GDC TARGET dataset.

3.3 GO enrichment and KEGG pathway analysis

The R package “clusterProfiler” was used for GO enrichment and KEGG pathway analyses. Enriched results with a significant level at $P<$ 0.05 were selected. The up-regulated and down-regulated genes were analyzed by GO and KEGG, respectively. In GO enrichment analysis of the up-regulated genes, the results showed that the most enriched GO terms involved cytoplasmic translation, ribonucleoprotein complex biogenesis, ribosome, ribosomal subunit, and structural constituent ribosome (Fig. 3A). GO enrichment analysis of the down-regulated genes showed that the most enriched GO terms were involved in T cell activation, positive regulation of cytokine production, secretory granule membrane, GTPase regulator activity, and nucleoside-triphosphatase regulator activity (Fig. 3B). The top pathway was ribosome in the KEGG pathway analysis of the up-regulated genes (Fig. 3C). In the KEGG pathway analysis of the down-regulated genes, the top four pathways were the chemokine signaling pathway, T cell receptor signaling pathway, Th1, and Th2 cell differentiation, and Th17 cell differentiation (Fig. 3D). These results are closely associated with AML.

Figure 3.

Go and KEGG pathway enrichment analyses. (A) The GO analysis of 693 up-regulated genes. “BP” represents “biological process”, “CC” represents “cellular component”, and “MF” represents “molecular function”. (B) The GO analysis of 571 down-regulated genes. (C) KEGG analysis of 693 up-regulated genes. (D) KEGG analysis of 571 down-regulated genes.

3.4 Eighteen prognostic genes predict a prognosis in AML

We screened out eighteen prognostic genes associated with AML using univariate Cox analysis (Fig. 5A). The significance level was $P$ -value $<$ 0.05. The genes that have been identified are INPP4A, CFLAR, LNPEP, KLRF1, APAF1, UTRN, TNPO3, G3BP1, ZNF276, LPCAT1, PSD4, ADCY7, HECTD3, VPS52, ANAPC2, SPG11, PHRF1, and FAM111A. We performed survival analysis from the Kaplan-Meier Plotter online database [35] to validate the potential values of the eighteen genes predicting prognosis. Kaplan-Meier survival curves showed high expression of INPP4A, CFLAR, LNPEP, KLRF1, APAF1, UTRN, TNPO3, G3BP1, ZNF276, LPCAT1, PSD4, ADCY7, HECTD3, VPS52, ANAPC2, SPG11, PHRF1, and FAM111A correlated with a poor prognosis. The results of the survival analysis are shown in Fig. 4A–R.

Figure 4.

Kaplan-Meier survival analysis of AML gene signatures. (A) INPP4A. (B) CFLAR. (C) LNPEP. (D) KLRF1. (E) APAF1. (F) UTRN. (G) TNPO3. (H) G3BP1. (I) ZNF276. (J) LPCAT1. (K) PSD4. (L) ADCY7. (M) HECTD3. (N) VPS52. (O) ANAPC2. (P) SPG11. (Q) PHRF1. (R) FAM111A.

Figure 5.

Regression analysis and characteristics of prognostic gene signatures. (A) Forest map of 18 prognostic genes by univariate Cox regression. (B) Distributions of risk scores in all samples. (C) Distribution of follow-up times in the training samples. (D) Heat map of gene expression in the prognostic signature of AML.

3.5 Eighteen prognostic genes risk score model construction

We construct a multivariate Cox regression model to explore the potential progonostic value of the eighteen genes. The coefficients of the eighteen genes are shown in Table 2. We used these genes to construct the risk model. The formula of the AML risk score [36] was constructed, integrating the gene expression value and the coefficients derived from multivariable Cox regression analysis. The formula was exhibited below: Risk score $=$ ( $-$ 0.05657 $\times$ expression value of INPP4A) $+$ (0.143723 $\times$ expression value of CFLAR) $+$ ( $-$ 0.00636 $\times$ expression value of LNPEP) $+$ (0.102319 $\times$ expression value of KLRF1) $+$ (0.094404 $\times$ expression value of APAF1) $+$ (0.086947 $\times$ expression value of UTRN) $+$ ( $-$ 0.09493 $\times$ expression value of TNPO3) $+$ (0.152613 $\times$ expression value of G3BP1) $+$ ( $-$ 0.07936 $\times$ expression value of ZNF276) $+$ (0.054006 $\times$ expression value of LPCAT1) $+$ (0.002892 $\times$ expression value of PSD4) $+$ (0.077592 $\times$ expression value of ADCY7) $+$ (0.051897 $\times$ expression value of HECTD3) $+$ (0.096815 $\times$ expression value of VPS52) $+$ ( $-$ 0.08058 $\times$ expression value of ANAPC2) $+$ ( $-$ 0.24259 $\times$ expression value of SPG11) $+$ ( $-$ 0.02295 $\times$ expression value of PHRF1) $+$ (0.005943 $\times$ expression value of FAM111A).

Table 2
Genes included in prognostic AML genes signature

Gene	Coef	HR	$Z$	$P$ -value	Lower	Upper
INPP4A	$-$ 0.05657	0.944998	$-$ 0.27358	0.784404	0.630113	1.417241
CFLAR	0.143723	1.154565	1.641381	0.100718	0.97249	1.370729
LNPEP	$-$ 0.00636	0.993659	$-$ 0.06409	0.948901	0.817983	1.207064
KLRF1	0.102319	1.107737	1.709324	0.087391	0.985109	1.24563
APAF1	0.094404	1.099004	1.214666	0.224494	0.943721	1.279838
UTRN	0.086947	1.090839	1.328451	0.184029	0.95951	1.240144
TNPO3	$-$ 0.09493	0.90944	$-$ 1.15302	0.2489	0.773921	1.06869
G3BP1	0.152613	1.164874	2.5982	0.009371	1.0382	1.307004
ZNF276	$-$ 0.07936	0.923709	$-$ 1.05781	0.29014	0.797402	1.070023
LPCAT1	0.054006	1.055491	1.116114	0.264373	0.95999	1.160493
PSD4	0.002892	1.002896	0.032727	0.973892	0.843404	1.19255
ADCY7	0.077592	1.080682	1.189095	0.234402	0.950943	1.228121
HECTD3	0.051897	1.053267	0.794775	0.426744	0.926739	1.19707
VPS52	0.096815	1.101656	1.097609	0.272375	0.926757	1.309563
ANAPC2	$-$ 0.08058	0.922581	$-$ 1.05546	0.291214	0.794362	1.071495
SPG11	$-$ 0.24259	0.784595	$-$ 2.1047	0.035318	0.625945	0.983456
PHRF1	$-$ 0.02295	0.977309	$-$ 0.28671	0.774332	0.835394	1.143332
FAM111A	0.005943	1.00596	0.089841	0.928413	0.883645	1.145207

Patients were divided into high-risk and low-risk groups based on the median risk score. As the risk score increased, the survival time of the patients tended to decrease, and the death rate of patients in the high-risk group was higher than that in the low-risk group. The risk score distribution, survival status, and heat map of genes are shown in Fig. 5B–D.

3.6 Cox proportional hazard regression analysis

In the univariate Cox regression analysis, we found that age [HR $=$ 1.034, 95% CI $=$ 0.999–1.070, $P=$ 0.060], gender (HR $=$ 1.354, 95% CI $=$ 0.904–2.028, $P=$ 0.141), WBC at diagnosis (HR $=$ 1.000, 95% CI $=$ 0.998–1.002, $P=$ 0.860), CNS disease (HR $=$ 1.320, 95% CI $=$ 0.611–2.850, $P=$ 0.479), FLT3 point mutation (PM) (HR $=$ 0.317, 95% CI $=$ 0.100–1.002, $P=$ 0.050), and risk score (HR $=$ 2.718, 95% CI $=$ 1.971–3.749, $P<$ 0.001) were significantly correlated with OS. Furthermore, in the multivariate Cox regression analysis, the risk score (HR $=$ 2.645, 95% CI $=$ 1.915–3.654, $P<$ 0.001) was illustrated to be a significant independent prognostic factor for OS (Fig. 6A and B). In Kaplan-Meier survival analysis, the low-risk group had longer OS than those in the high-risk group (Fig. 6C). In Multi-index ROC analysis, the AUCs of the ROC curve of risk score, age, gender, WBC at diagnosis, CNS disease, and FLT3 PM were 0.705, 0.578, 0.550, 0.566, 0.509, and 0.461 respectively (Fig. 6D). The AUC value of the risk score was the highest one than that of other clinical parameters, showing a convincing predictive value of the risk model and revealing that this Cox regression model predicted prognosis well in AML patients.

Figure 6.

Cox proportional hazard regression analysis of AML risk factors. (A) Univariate Cox regression analysis of risk score and other indicators. (B) Multivariate Cox regression analyses of risk score and other indicators. (C) Kaplan-Meier analysis of GDC TARGET AML patients grouped according to median risk. (D) Multi-index ROC curve of risk score and other indicators.

Figure 7.

External verification of the prognostic AML gene signature. (A) Risk scores distribution. (B) Survival status of patients. (C) Heat map of gene expression pattern. (D) Kaplan-Meier plot for OS of patients in different risk groups. (E) ROC curve of risk score. (F) Validation of prognostic genes by qRT-PCR. $P$ -values were conducted by Multiple $t$ -tests, ${}^{*}p<$ 0.05, ${}^{**}p<$ 0.01, ${}^{***}p<$ 0.001.

3.7 External verification of the prognostic gene signature

The GDC TCGA database validated the prognostic signature using the same risk score formula. The gene signature was tested in independent AML datasets.

The risk score distribution, survival status of patients, and gene expression pattern are shown in Fig. 7A–C. In the validation datasets, the Kaplan-Meier survival analysis showed that patients in the high-risk group had a significantly poorer OS than those in the low-risk group (Fig. 7D). In ROC analysis, the AUC of the ROC curve of risk score was 0.696, showing a good accuracy in predicting OS (Fig. 7E). These results showed that risk score was an independent prognostic factor for prognosis in patients with AML.

3.8 Experimental verification of prognostic genes in mRNA levels

To prove the accuracy and repeatability of the data, five genes were selected for qRT-PCR, as shown in Fig. 7F. The prognostic genes were successfully validated and showed a good correspondence with the transcriptome analysis results.

4. Discussion

Our study obtained 187 bone marrow samples with AML and 37 normal bone marrow samples from the GDC TARGET database. The significantly different expression levels of mRNAs in AML were calculated. The GO and KEGG analyses found that AML patients’ dysfunctions were primarily enriched in cytoplasmic translation, ribosome, and T-cell activation. Interestingly, a previous study demonstrated that T-cell function was impaired in AML [37]. Our study discovered that T cell activation is one of the pathways through which down-regulated genes were enriched, consistent with the previous study [38].

Meanwhile, up-regulated translation pathways, including cytoplasmic translation and ribonucleoprotein complex biogenesis [39], maybe a potential therapeutic target in AML, confirming previous studies that translational control can play a critical role in modulating gene expression [40, 41]. With univariate and multivariate Cox regressions, eighteen prognostic genes were identified. Finally, the signature of eighteen prognostic genes and a risk score model were constructed. Kaplan-Meier survival analysis was used to validate the prognostic value of the eighteen genes. The Kaplan-Meier plot for OS of AML patients showed that patients with predicted high risk had significantly shorter OS time than those with low risk. The AUC value of ROC of the risk score was higher than that of other clinical parameters, showing a high predictive value of the risk model.

Our study developed and validated the prognostic signature of eighteen genes using GDC TARGET and GDC TCGA databases. Then, INPP4A, CFLAR, LNPEP, KLRF1, APAF1, UTRN, TNPO3, G3BP1, ZNF276, LPCAT1, PSD4, ADCY7, HECTD3, VPS52, ANAPC2, SPG11, PHRF1, and FAM111A were identified. In the Multi-index ROC analysis of the training set, the AUC of the risk score was 0.705, showing a high predictive value of the risk model.

Among the eighteen AML prognostic genes, CFLAR, KLRF1, APAF1, G3BP1, LPCAT1, and ADCY7 have been associated with AML. The protein encoded by CFLAR (CASP8 and FADD-like apoptosis regulator) is an apoptosis regulator structurally similar to caspase-8. Previous studies demonstrated that patients with higher than median CFLAR mRNA expression had a significantly lower 3-year OS than those with low CFLAR expression. High CFLAR mRNA expression was potentially a biomarker of overall adverse survival in an adult AML patient cohort [42]. KLRF1 (killer cell lectin-like receptor, subfamily F, member1, KLRF1) is expressed on most NK cells’ surfaces and a part of T cells [43, 44, 45], stimulating cytotoxicity and cytokine secretion and was seldom reported involving leukemia. KLRF1 expression was significantly lower in AML patients than in non-malignant controls, which might be predominantly caused by the down-regulation of KLRF1 on NK cells or the decrease in NK cell numbers, companies with the reduced NK cell surveillance of tumors [46]. APAF1 (apoptotic peptidase activating factor 1) encodes a cytoplasmic protein that initiates apoptosis. Low expression of APAF-1XL, one of the APAF-1 transcripts in acute myeloid leukemia, may be associated with the failure of remission induction therapy [47]. G3BP1 (G3BP stress granule assembly factor 1) encodes one of the DNA-unwinding enzymes, which prefers partially unwound 3’-tailed substrates and can also unwind partial RNA/DNA and RNA/RNA duplexes in an ATP-dependent fashion. A previous study demonstrated that genomic profiling of acute myeloid leukemia associated with ataxia-telangiectasia identifies a complex karyotype with wild-type TP53 and mutant KRAS, G3BP1, and IL7R [48]. LPCAT1 (lysophosphatidylcholine acyltransferase, 1) encodes a member of the 1-acyl-sn-glycerol-3-phosphate acyltransferase family of proteins. The encoded enzyme plays a role in phospholipid metabolism, specifically in converting lysophosphatidylcholine to phosphatidylcholine in the presence of acyl-CoA. A previous study investigated the clinical implications of the expression of LPCAT family members in AML and demonstrated that LPCAT1 expression is significantly associated with the survival of patients with AML [49]. ADCY7 (adenylate cyclase 7) encodes a membrane-bound adenylate cyclase that catalyzes cyclic AMP formation from ATP and is inhibited by calcium. ADCY7 is a promising target for AML treatment, and high expression of ADCY7 correlates with poor survival of AML patients [50]. Based on this, we speculate that other genes may also have a guiding role in the prognosis of AML and have the potential to be used as new biomarkers for the diagnosis and prognosis of AML. We anticipate that our findings may help identify critical genes associated with AML and provide new ideas and prospective targets for treating AML. Our findings still need more clinical samples and broader prospective studies to validate further.

In summary, we identified eighteen feature genes tightly correlated with AML prognosis, and high expression levels of these genes were related to bad prognosis in AML patients. Meanwhile, a risk score model was developed and could be used to predict survival and applied to AML treatment and prognosis.

Author contributions

Conception: Jingxin Zhang.

Interpretation or analysis of data: Panpan Chen, Jiaming Cao, Lingling Chen, Guanfei Gao.

Preparation of the manuscript: Jingxin Zhang, Panpan Chen.

Revision for important intellectual content: Yuanlin Xu, Peijun Jia, Yan Li, Yating Li, Shijie Zhang, Jiangfeng Du.

Supervision: Jingxin Zhang.

Funding

This study was supported by the National Natural Science Foundation of China under Grants No. 81900111 and 81870095.

Footnotes

Acknowledgments

We acknowledge GDC TARGET and GDC TCGA databases for providing their platforms and contributors for uploading their meaningful datasets.

References

Cai

S.F.

and Levine

R.L.

, Genetic and epigenetic determinants of AML pathogenesis, Semin Hematol 56 (2019), 84–89.

Newell

L.F.

and Cook

R.J.

, Advances in acute myeloid leukemia, Bmj 375 (2021), n2026.

Döhner

Weisdorf

D.J.

and Bloomfield

C.D.

, Acute myeloid leukemia, N Engl J Med 373 (2015), 1136–1152.

Marcucci

Haferlach

and Döhner

, Molecular genetics of adult acute myeloid leukemia: Prognostic and therapeutic implications, J Clin Oncol 29 (2011), 475–486.

Papaemmanuil

Gerstung

Bullinger

Gaidzik

V.I.

Paschka

Roberts

N.D.

Potter

N.E.

Heuser

Thol

Bolli

Gundem

Van Loo

Martincorena

Ganly

Mudie

McLaren

O’Meara

Raine

Jones

D.R.

Teague

J.W.

Butler

A.P.

Greaves

M.F.

Ganser

Döhner

Schlenk

R.F.

Döhner

and Campbell

P.J.

, Genomic classification and prognosis in acute myeloid leukemia, N Engl J Med 374 (2016), 2209–2221.

Beaton

Peterson

G.J.

and O’Brien

, Acute myeloid leukemia: Advanced practice management from presentation to cure, J Adv Pract Oncol 11 (2020), 836–844.

Medinger

and Passweg

J.R.

, Acute myeloid leukaemia genomics, Br J Haematol 179 (2017), 530–542.

Sung

Ferlay

and Siegel

R.L.

, Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries, CA Cancer J Clin 71 (2021), 209–249.

Yang

and Wei

, Molecular landscape and targeted therapy of acute myeloid leukemia, Biomark Res 6 (2018), 32.

10.

and Zhu

, FLT3 inhibitors in acute myeloid leukemia, J Hematol Oncol 11 (2018), 133.

11.

Yang

and Wang

, Precision therapy for acute myeloid leukemia, J Hematol Oncol 11 (2018), 3.

12.

Abdel-Wahab

and Levine

R.L.

, Mutations in epigenetic modifiers in the pathogenesis and therapy of acute myeloid leukemia, Blood 121 (2013), 3563–3572.

13.

Luo

Qin

Reu

Dai

Huang

Wang

Zhang

Pan

Zhu

Niu

Xiao

Zheng

and Liu

, Microarray-based analysis and clinical validation identify ubiquitin-conjugating enzyme E2E1 (UBE2E1) as a prognostic factor in acute myeloid leukemia, J Hematol Oncol 9 (2016), 125.

14.

Medinger

Lengerke

and Passweg

, Novel prognostic and therapeutic mutations in acute myeloid leukemia, Cancer Genomics Proteomics 13 (2016), 317–329.

15.

Zhang

Guo

Wang

Huang

and Zhou

, Novel gene signature reveals prognostic model in acute myeloid leukemia, Front Genet 11 (2020), 566024.

16.

D.H.

Chen

Liu

X.P.

Yao

and Ruan

X.L.

, Dysregulation of miR-138-5p/RPS6KA1-AP2M1 is associated with poor prognosis in AML, Front Cell Dev Biol 9 (2021), 641629.

17.

Jiang

P.Y.Z.

Sun

Zhang

Jiang

and Song

, Advances in targeted therapy for acute myeloid leukemia, Biomark Res 8 (2020), 17.

18.

Lan

Song

and Pan

, High expression of the SH3TC2-DT/SH3TC2 gene pair associated with FLT3 mutation and poor survival in acute myeloid leukemia: An integrated TCGA analysis, Front Oncol 10 (2020), 829.

19.

Zhao

Song

and Liu

, Gilteritinib: A novel FLT3 inhibitor for acute myeloid leukemia, Biomark Res 7 (2019), 19.

20.

Fröhling

Scholl

Levine

R.L.

Loriaux

Boggon

T.J.

Bernard

O.A.

Berger

Döhner

Ebert

B.L.

Teckie

Golub

T.R.

Jiang

Schittenhelm

M.M.

Lee

B.H.

Griffin

J.D.

Stone

R.M.

Heinrich

M.C.

Deininger

M.W.

Druker

B.J.

and Gilliland

D.G.

, Identification of driver and passenger mutations of FLT3 by high-throughput DNA sequence analysis and functional assessment of candidate alleles, Cancer Cell 12 (2007), 501–513.

21.

Garg

Nagata

Kanojia

Mayakonda

Yoshida

Haridas Keloth

Zang

Z.J.

Okuno

Shiraishi

Chiba

Tanaka

Miyano

Ding

L.W.

Alpermann

Sun

Q.Y.

Lin

D.C.

Chien

Madan

Liu

L.Z.

Tan

K.T.

Sampath

Venkatesan

Inokuchi

Wakita

Yamaguchi

Chng

W.J.

Kham

S.K.

Yeoh

A.E.

Sanada

Schiller

Kreuzer

K.A.

Kornblau

S.M.

Kantarjian

H.M.

Haferlach

Lill

Kuo

M.C.

Shih

L.Y.

Blau

I.W.

Blau

Yang

Ogawa

and Koeffler

H.P.

, Profiling of somatic mutations in acute myeloid leukemia with FLT3-ITD at diagnosis and relapse, Blood 126 (2015), 2491–2501.

22.

Eshibona

Giwa

and Rossouw

S.C.

, Upregulation of FHL1, SPNS3, and MPZL2 predicts poor prognosis in pediatric acute myeloid leukemia patients with FLT3-ITD mutation, Leuk Lymphoma, 2022, 1–10.

23.

Cui

Yang

Zhang

Yin

Huang

Zhou

Wang

and Chen

, Genome-wide identification of FHL1 as a powerful prognostic candidate and potential therapeutic target in acute myeloid leukaemia, EBioMedicine 52 (2020), 102664.

24.

Manola

K.N.

Panitsas

Polychronopoulou

Daraki

Karakosta

Stavropoulou

Avgerinou

Hatzipantelis

Pantelias

Sambani

and Pagoni

, Cytogenetic abnormalities and monosomal karyotypes in children and adolescents with acute myeloid leukemia: Correlations with clinical characteristics and outcome, Cancer Genet 206 (2013), 63–72.

25.

Love

M.I.

Huber

and Anders

, Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2, Genome Biol 15 (2014), 550.

26.

Wang

L.G.

Han

and He

Q.Y.

, clusterProfiler: An R package for comparing biological themes among gene clusters, Omics 16 (2012), 284–287.

27.

Peng

Wang

Hui

Wang

and Shang

, Identifying term relations cross different gene ontology categories, BMC Bioinformatics 18 (2017), 573.

28.

Kanehisa

Furumichi

Tanabe

Sato

and Morishima

, KEGG: New perspectives on genomes, pathways, diseases and drugs, Nucleic Acids Res 45 (2017), D353–d361.

29.

Kanehisa

and Goto

, KEGG: Kyoto encyclopedia of genes and genomes, Nucleic Acids Res 28 (2000), 27–30.

30.

George

Seals

and Aban

, Survival analysis and regression models, J Nucl Cardiol 21 (2014), 686–694.

31.

Menyhárt

and Nagy

Á.

, Determining consistent prognostic biomarkers of overall survival and vascular invasion in hepatocellular carcinoma, R Soc Open Sci 5 (2018), 181006.

32.

Qiu

Sun

Wang

and Chen

, Identification and validation of an individualized autophagy-clinical prognostic index in gastric cancer patients, Cancer Cell Int 20 (2020), 178.

33.

Cao

and López-de-Ullibarri

, ROC curves for the statistical analysis of microarray data, Methods Mol Biol 1986 (2019), 245–253.

34.

Janssens

and Martens

F.K.

, Reflection on modern methods: Revisiting the area under the ROC Curve, Int J Epidemiol 49 (2020), 1397–1403.

35.

Nagy

Á.

Lánczky

Menyhátt

and Györffy

, Validation of miRNA prognostic power in hepatocellular carcinoma using expression data of independent datasets, Sci Rep 8 (2018), 9227.

36.

Wang

Niu

Liu

Wang

Chen

and Qin

, Identification of seven-gene signature for prediction of lung squamous cell carcinoma, Onco Targets Ther 12 (2019), 5979–5988.

37.

Lamble

A.J.

Kosaka

Laderas

and Maffit

, Reversible suppression of T cell function in the bone marrow microenvironment of acute myeloid leukemia, Proc Natl Acad Sci U S A 117 (2020), 14331–14341.

38.

Buggins

A.G.

Milojkovic

Arno

M.J.

Lea

N.C.

Mufti

G.J.

Thomas

N.S.

and Hirst

W.J.

, Microenvironment produced byacute myeloid leukemia cells prevents T cell activation and proliferation by inhibition of NF-kappaB, c-Myc, and pRb pathways, J Immunol 167 (2001), 6021–6030.

39.

Zhang

Han

Song

and Ge

, CD37 high expression as a potential biomarker and association with poor outcome in acute myeloid leukemia, Biosci Rep 40 (2020).

40.

Sonenberg

and Hinnebusch

A.G.

, Regulation of translation initiation in eukaryotes: Mechanisms and biological targets, Cell 136 (2009), 731–745.

41.

Vogel

, Translation’s coming of age, Mol Syst Biol 7 (2011), 498.

42.

McLornan

Hay

McLaughlin

Holohan

Burnett

A.K.

Hills

R.K.

Johnston

P.G.

Mills

K.I.

McMullin

M.F.

Longley

D.B.

and Gilkes

, Prognostic and therapeutic relevance of c-FLIP in acute myeloid leukaemia, Br J Haematol 160 (2013), 188–198.

43.

Huang

Zheng

Tian

and Sun

, Peptide mimicry of AICL inhibits cytolysis of NK cells by blocking NKp80-AICL recognition, Immunol Invest 39 (2010), 587–597.

44.

Kuttruff

Koch

Kelp

Pawelec

Rammensee

H.G.

and Steinle

, NKp80 defines and stimulates a reactive subset of CD8 T cells, Blood 113 (2009), 358–369.

45.

Vitale

Falco

Castriconi

Parolini

Zambello

Semenzato

Biassoni

Bottino

Moretta

and Moretta

, Identification of NKp80, a novel triggering molecule expressed by human NK cells, Eur J Immunol 31 (2001), 233–242.

46.

Tian

Huang

Wang

Yin

Zhou

Zhang

Huang

Zhou

Jiang

Liao

Yang

and Meng

, Identification of novel molecular markers for prognosis estimation of acute myeloid leukemia: over-expression of PDCD7, FIS1 and Ang2 may indicate poor prognosis in pretreatment patients with acute myeloid leukemia, PLoS One 9 (2014), e84150.

47.

Benites

B.D.

Fattori

Hackel

Lorand-Metze

De Souza

C.A.

Schulz

Costa

F.F.

and Saad

S.T.

, Low expression of APAF-1XL in acute myeloid leukemia may be associated with the failure of remission induction therapy, Braz J Med Biol Res 41 (2008), 571–578.

48.

Goldgraben

M.A.

Fewings

Larionov

Scarth

Redman

Telford

Arkwright

P.D.

Bonney

Wilks

Kulkarni

Taylor

A.M.R.

Tischkowitz

M.D.

and Meyer

, Genomic profiling of acute myeloid leukaemia associated with ataxia telangiectasia identifies a complex karyotype with wild-type TP53 and mutant KRAS, G3BP1 and IL7R, Pediatr Blood Cancer 67 (2020), e28354.

49.

Wang

Tang

Cheng

and Lin

, Identification of LPCAT1 expression as a potential prognostic biomarker guiding treatment choice in acute myeloid leukemia, Oncol Lett 21 (2021), 105.

50.

Xie

Chen

Yin

Zhan

Fang

and Zhang

C.C.

, ADCY7 supports development of acute myeloid leukemia, Biochem Biophys Res Commun 465 (2015), 47–52.

Prognostic value of an eighteen-genes panel in acute myeloid leukemia by analyzing TARGET and TCGA databases

Abstract

BACKGROUND:

OBJECTIVE:

METHODS:

RESULTS:

CONCLUSIONS:

Keywords

1. Introduction

2. Materials and methods

2.1 The workflow of the model construction process

2.3 Cell culture

2.4 Functional enrichment analysis

2.5 Construction of the eighteen prognostic genes’ signature

2.6 Quantitative real-time polymerase chain reaction (qRT-PCR)

2.7 Statistical analysis

3. Results

3.1 Characteristics of AML patients in the dataset from the GDC TARGET database

Table 1 Characteristics of patients

Table 2 Genes included in prognostic AML genes signature

3.8 Experimental verification of prognostic genes in mRNA levels

4. Discussion

Author contributions

Funding

Footnotes

Acknowledgments

References

Table 1
Characteristics of patients

Table 2
Genes included in prognostic AML genes signature