High Expression of CPT1A Predicts Adverse Outcomes: A Potential Therapeutic Target for Acute Myeloid Leukemia

Abstract

Carnitine palmitoyl transferase 1A (CPT1A) protein catalyzes the rate-limiting step of Fatty-acid oxidation (FAO) pathway, which can promote cell proliferation and suppress apoptosis. Targeting CPT1A has shown remarkable anti-leukemia activity. But, its prognostic value remains unclear in Acute Myeloid Leukemia (AML). In two independent cohorts of cytogenetically normal AML (CN-AML) patients, compared to low expression of CPT1A (CPT1A^low), high expression of CPT1A (CPT1A^high) was significantly associated with adverse outcomes, which was also shown in European Leukemia Network (ELN) Intermediate-I category. Multivariable analyses adjusting for known factors confirmed CPT1A^high as a high risk factor. Significant associations between CPT1A^high and adverse outcomes were further validated whether for all AML patients (OS: P = 0.008; EFS: P = 0.002, n = 334, no M3) or for National Comprehensive Cancer Network (NCCN) Intermediate-Risk subgroup (OS: P = 0.021, EFS: P = 0.024, n = 173). Multiple omics analysis revealed aberrant alterations of genomics and epigenetics were significantly associated with CPT1A expression, including up- and down-regulation of oncogenes and tumor suppressor, activation and inhibition of leukemic (AML, CML) and immune activation pathways, hypermethylation enrichments on CpG island and gene promoter regions. Combined with the previously reported anti-leukemia activity of CPT1A's inhibitor, our results proved CPT1A as a potential prognosticator and therapeutic target for AML.

Highlights

• •High expression of CPT1A is an adverse prognostic biomarker in AML.
• •Aberrant alterations of genomic and epigenomic patterns are significantly associated with CPT1A expression.

Identification of prognostic biomarkers is essential for therapeutic choice of AML. This study represents direct evidences that high expression of CPT1A is significantly associated with poor outcomes and abnormal genomic and epigenomic patterns in AML patients. CPT1A is an important catalyzer for fatty-acid oxidation pathway, which may provide alternative carbon source for leukemia proliferation. Findings of this study may indicate the significance of fat metabolism in leukemogenesis.

1. Introduction

Acute myeloid leukemia (AML) represents a group of myeloid malignancies with remarkably heterogeneous outcomes. Finding effective prognostic biomarkers has been being one of the most urgent clinical needs and research hotspots. Cytogenetically normal acute myeloid leukemia (CN-AML) accounts for about one-half of total AML and constitutes the main body of intermediate-risk AML (Dohner et al., 2010). Since there are no microscopically detectable chromosome abnormalities in leukemic blasts, CN-AML shows a particular uniformity of cytogenetics and provides a perfect platform of biomarker recognition for AML. Many kinds of molecular signals (such as DNA mutations, aberrant expression of mRNA and microRNAs) have been identified independently as favorable or adverse prognostic biomarkers. Recent studies showed that metabolic signatures were involved in leukemogenesis and could inspire novel therapeutic regimens for AML (Cheong et al., 2012).

Aggressive metabolic changes are key hallmarks of cancer (Hanahan and Weinberg, 2000), Warburg effect of aerobic glycolysis has been regarded as an important bioenergetic source for rapid cell proliferation (Hsu and Sabatini, 2008). Beyond the Warburg effect, other effects, especially fatty acid oxidation (FAO) (Carracedo et al., 2013), also show important roles in the cancer pathogenesis. Series reports of therapeutic applications for metabolic signatures have appeared for AML, including a distinct glucose metabolism signature proved as a prognostic biomarker (Chen et al., 2014), glutamine uptake (Willems et al., 2013) and arginine dependence (Mussai et al., 2015) reported as novel therapeutic targets. Particular concerns were paid on FAO, an original article and editorial discussed function blocking of FAO with CPT1A's inhibitor (ST1326) and its inherent mechanisms for anti-leukemia treatment (Ricciardi et al., 2015, Samudio and Konopleva, 2015). But, up to now, the prognostic value of CPT1A expression remains unclear for AML patients.

Here, we presented CPT1A^high as an adverse prognostic biomarker for AML with concrete data, and also explored the distinctive genomic and epigenomic patterns associated with CPT1A expression. Compared to the existing basic experiments of CPT1A, our work offered more direct evidences for using CPT1A expression as a prognostic biomarker in risk stratification or as a potential therapeutic target for AML patients.

2. Materials & Methods

2.1. Patients and Treatment

The first cohort was derived from a whole AML cohort (n = 334, no M3), which included 156 de novo CN-AML patients (median age, 50 years, range: 16–77 years), which were all diagnosed and collected at Erasmus University Medical Center (Rotterdam) between 1990 and 2008. 83% of patients (n = 129) were aged under 60 years (younger patients) and 17% patients were ≥ 60 years (older patients). The study was approved by the institutional review boards at Weill Cornell Medical College and Erasmus University Center, and all subjects provided written informed consent in accordance with the Declaration of Helsinki. The methods were carried out in accordance with the approved guidelines. All patient were uniformly treated under the study protocols of Dutch-Belgian Cooperative Trial Group for Hematology Oncology (HOVON, details of therapeutic protocol available at http://www.hovon.nl). All clinical, cytogenetic and molecular information as well as microarray data of these patients can be publicly downloaded at the Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo, GSE6891 (Verhaak et al., 2009)). All samples contained 80–100% blast cells after thawing. Conventional cytogenetic examination of at least 20 metaphases from bone marrow (BM) was used to determine normal karyotype. RT-PCR assays were used to assess the mutations of NPM1, CEBPA, N-RAS, K-RAS, IDH1 and IDH2, the presence of FLT3-ITD and FLT3-TKD [D835].

Another independent cohort included 162 previously untreated CN-AML patients (median age: 57.5, range: 17–83 years), which also received uniform intensive double-induction and consolidation chemotherapy provided by the multicenter AMLCG-1999 trial of German AML Cooperative Group between 1999 and 2003. The AMLCG-1999 clinical trial was approved by the local institutional review boards, and written informed consent for each patient were obtained in accordance with the Declaration of Helsinki. The methods were carried out in accordance with the approved guidelines. Microarray data as well as patient information were also publicly available (GEO accession number: GSE12417 (Metzeler et al., 2008)).

2.2. Microarray and Sequencing Data Analyses

Several previously published dataset were used for gene expression profiles, including GSE6891 (Verhaak et al., 2009), GSE12417 (Metzeler et al., 2008), GSE1159 (Valk et al., 2004), GSE9476 (Stirewalt et al., 2008) and GSE30029 (de Jonge et al., 2011), all of which can be obtained from GEO. Microarray expression profiles were obtained by Affymetrix Human Genome 133 plus 2.0 and U133A Gene Chips. All experiments' design, quality control and data normalization were in line with the standard Affymetrix protocols. Expression profiles of mRNA and microRNA were obtained by high throughput sequencing (RNA-Seq), and genome-wide methylation data were obtained by Illumina infinium 450K beadchips, derived from the Cancer Genome Atlas (TCGA), which provided 73 CN-AML patients with all mRNA and microRNA expression and methylation profiles (Cancer Genome Atlas Research, 2013). For microarray data, expression levels of a gene were computed as the mean value of all probe sets annotated to it, while for RNA-Seq data, expression levels of mRNA and microRNA were computed as RPKM and RPM (Reads Per Kilo-base per Million reads). To choose the appropriate cut-off value for subdivision, we accessed the distribution of CPT1A expression level and compared survivals for the 156 CN-AML patients dividing into 4 quartiles, results represented a normal distribution (Fig. S1A) and evident distinction along the median value (Fig. S1B and C). Thus, median value of CPT1A expression was used to divide patients into CPT1A^high and CPT1A^low groups. Other dividing of patients according to a gene's expression (such as ERG, WT1 and DMNT3A etc.) were dealt using the same strategy.

2.3. Statistical Analyses

The definition of overall survival (OS) was the time from date of diagnosis to death due to any causes. Event free survival (EFS) was defined as the time from date of diagnosis to removal from the study due to the absence of complete remission (CR), relapse or death. The Kaplan-Meier method and log-rank test were used to estimate the association between CPT1A expression and OS, EFS. The Fisher exact and Wilcoxon rank-sum tests were used, respectively for categorical and continuous variables, to assess the association between expression levels and clinical, molecular characteristics. Multivariable hazards models were used to evaluate the impacts of CPT1A expression to OS and EFS in the presence of other known risk factors. Student's t-test and multiple hypothesis correction (False Discovery Rate, FDR) were used to identify differences in genome-wide gene, microRNA and methylation profiles between CPT1A^high and CPT1A^low groups. All analyses were performed using the R 3.2.3 software packages.

3. Results

3.1. Overexpression of CPT1A in AML Patients

Three public microarray datasets were used to compare CPT1A expression between AML patients and heathy donors using bone marrow (BM), peripheral blood (PB) and CD34 + cells. For the comparison of BM samples, CPT1A was highly expressed in AML (Fig. 1A, P = 0.049, 7 AML vs 10 NBM, GSE9476), which was validated in PB samples (Fig. 1B, P < 0.001, 19 AML vs 10 normal, GSE9476). Additionally, CD34 + cells derived from AML patients and healthy donors were used to further verify CPT1A's overexpression in AML patients (Fig. 1C, P < 0.001, 46 AML CD34 + vs 31 normal CD34 +, GSE30029). Besides, higher expression levels of CPT1A was shown in various different AML-subtypes than normal BM, including 19 CBFB-MYH11, 115 CN-AML, 10 Complex, 17 MLL-translocation, 86 Others, 22 RUXN1-RUNX1T1 and 5 normal BM (Fig. S2, GSE1159). All these results showed significant overexpression of CPT1A in AML patients.

Fig. 1.

Differential expression of CPT1A. (A) AML-BM (n = 7) vs NBM (n = 10). (B) AML-PB (n = 19) vs NPM (n = 10). (C) AML CD34 + cells (n = 46) vs NBM CD34 + cells (n = 31).

3.2. Different Molecular Characteristics Between CPT1A^high and CPT1A^low Groups

In the cohort of 156 CN-AML derived from GSE6891 (334 AML, no M3), patients in CPT1A^high group tended to be younger (P = 0.03) and belong more to FAB M1 (P = 0.03) than that in CPT1A^low. CPT1A^high group carried more FLT3-ITD than CPT1A^low (P = 0.05), while no additional significant associations between CPT1A expression and other mutations were found. In addition, CPT1A^high was more likely to accompany with higher expression of many known adverse prognostic biomarkers, such as ERG (P < 0.001), BAALC (P < 0.04), WT1 (P < 0.001), DNMT3A (P < 0.001), DNMT3B (P = 0.006), MAPKBP1 (P = 0.04), ITPR2 (P < 0.001), ATP1B1 (P < 0.001), RUNX1 (P < 0.001), TCF4 (P = 0.002) and CXXC5 (P < 0.001). (See Table 1 and Fig. S3A)

Table 1.

Patients' characteristics in the primary cohort of 156 CN-AML patients according to CPT1A expression.

Variable	CPT1Ahigh, n = 78	CPT1Alow, n = 78	P
Median age. y (range)	47.5 (16–77)	53 (18–77)	0.03
FAB subtype, no.
M0	0	3	0.25
M1	29	16	0.03
M2	20	12	0.16
M4	10	14	0.51
M5	16	23	0.62
M6	1	0	1
Other	2	10	0.03
Double CEBPA (No FLT3-ITD), no	6	8	0.78
NPM1 (No FLT3-ITD), no	17	18	1
FLT3-ITD, no.	39	26	0.05
High ERG, no.	52	26	< 0.001
High BAALC, no.	46	32	0.04
High WT1, no.	51	27	< 0.001
High DNMT3A, no	50	28	< 0.001
High DNMT3B, no	48	30	0.006
High MAPKBP1, no	46	32	0.04
High ITPR2, no	51	27	< 0.001
High ATP1B1, no	51	27	< 0.001
High RUNX1, no	55	23	< 0.001
High TCF4, no	49	29	0.002

FAB, French-American-British classification; ITD, internal tandem duplication; TKD, tyrosine kinase domain; ELN, European Leukemia Net.

High ERG, BAALC, WT1, DNMT3A, DNMT3B, MAPKBP1, ITPR2, ATP1B1, RUNX1 and TCF4 expression were defined as an expression level above the median of all samples, respectively.

3.3. CPT1A^high Is Associated With Adverse Outcomes in AML

Survival analyses were carried out in the whole cohort of 156 CN-AML patients and European Leukemia Net (ELN) Intermediate-I category (n = 121). Compared to CPT1A^low groups, the former results showed that CPT1A^high group was significantly associated with shorter OS and EFS (P = 0.042 and P = 0.037, respectively, Fig. 2A), and the latter results also showed significant associations between CPT1A^high and shorter OS and EFS (P = 0.013 and P = 0.008, respectively, Fig. 2B). Further, in order to confirm the prognostic significance of CPT1A expression, multivariable analyses were performed in the whole CN-AML and ELN Intermediate-I category. After adjusting for the impacts of known risk factors, multivariable models of OS and EFS were built respectively. For the whole CN-AML, CPT1A^high had 1.68 times higher risks on OS (P = 0.014) and 1.64 times higher risks on EFS (P = 0.014). Other negative factors included older age (P = 0.031 for OS), wild type of NPM1 (OS: P = 0.007, EFS: P = 0.003) and the presence of FLT3-ITD (OS: P = 0.002, EFS: P = 0.003). For the ELN Intermediate-I category, CPT1A^high still showed adverse impacts on OS and EFS (P = 0.006 and P = 0.004, respectively) (See Table 2). To further investigate the prognostic significance of CPT1A expression, same analyses were executed for the whole cohort of 334 AML (No M3) and National Comprehensive Cancer Network (NCCN) Intermediate-Risk category (n = 173). Compared to CPT1A^low groups, CPT1A^high groups still had remarkable shorter OS and EFS in the total AML (P = 0.008 and P = 0.002, respectively, Fig. 2C) and in the NCCN Intermediate-Risk category (P = 0.021 and P = 0.024, respectively, Fig. 2D).

Fig. 2.

The prognostic value of CPT1A expression in CN-AML and AML patients. (A) OS and (B) EFS of the entire 156 CN-AML and 121 patients of ELN Intermediate-I category. (C) OS and (D) EFS of the entire 334 AML and 173 patients of NCCN Intermediate-Risk.

Table 2.

Multivariable analysis with OS and EFS in the primary cohort of 156 CN-AML patients.

Variables in final model by end point	HR	95% CI	P-value
OS (All CN-AML, n = 156)
CPT1A expression, high vs low	1.68	1.11–2.55	0.014
Age, per 10-y increase	1.19	1.02–1.40	0.031
Single CEBPA mutation vs wild type	0.95	0.40–2.26	0.915
Double CEBPA mutation vs wild type	0.49	0.21–1.13	0.095
NPM1, mutated vs wild type	0.54	0.34–0.84	0.007
FLT3-ITD, presented vs others	2.03	1.29–3.17	0.002
EFS (All CN-AML, n = 156)
CPT1A expression, high vs low	1.64	1.11–2.42	0.014
Age, per 10-y increase	1.08	0.93–1.26	0.305
Single CEBPA mutation vs wild type	1.03	0.43–2.45	0.952
Double CEBPA mutation vs wild type	0.49	0.23–1.04	0.064
NPM1, mutated vs wild type	0.52	0.34–0.80	0.003
FLT3-ITD, presented vs others	1.92	1.25–2.95	0.003
OS (ELN Intermediate-I, n = 121)
CPT1A expression, high vs low	1.90	1.12–3.01	0.006
Age, per 10-y increase	1.18	1.00–1.40	0.050
Single CEBPA mutation vs wild type	0.94	0.30–2.24	0.891
Double CEBPA mutation vs wild type	0.47	0.20–1.11	0.085
NPM1, mutated vs wild type	0.57	0.30–1.11	0.098
FLT3-ITD, presented vs others	1.86	0.96–3.61	0.065
EFS (ELN Intermediate-I, n = 121)
CPT1A expression, high vs low	1.90	1.23–2.95	0.004
Age, per 10-y increase	1.11	0.94–1.30	0.213
Single CEBPA mutation vs wild type	1.0	0.41–2.41	0.999
Double CEBPA mutation vs wild type	0.49	0.23–1.05	0.068
NPM1, mutated vs wild type	0.50	0.27–0.94	0.032
FLT3-ITD, presented vs others	2.01	1.07–3.76	0.030

3.4. CPT1A^high Is Associated With Adverse Outcomes in Another Independent Cohort of 162 CN-AML Patients

Another independent patient cohort of 162 de novo CN-AML was further studied (GSE12417), results showed that CPT1A^high was significantly associated with shorter OS (P = 0.01, Fig. S4). In addition, CPT1A^high patients were more likely to have higher expression of ERG, WT1, DNMT3B, TCF4, ITPR2, MAPKBP1, ATP1B1 and RUNX1 (All P < 0.001, see Table S1 and Fig. S3B.

3.5. Associations Between Genome-wide Expression Profiles and CPT1A Expression

To further investigate the biological role of CPT1A in leukemogenesis, gene expression profiles associated with CPT1A were derived based on genome-wide microarray analysis. 131 up-regulated and 271 down-regulated genes were identified as to be significantly associated with CPT1A expression (fdr-adjusted P < 0.05 and FC > 1.5 or FC < 1/1.5, Fig. 3A). Further, these aberrant genes were presented as an expression heatmap (Fig. 3B). The up-regulated genes included: 1) genes involving in leukemogenesis (such as HHEX (Shields et al., 2016), NET1 (Ahmad et al., 2014)), tumorigenesis promoters (such as CDK6, HOX family genes (Shah and Sukumar, 2010)), tyrosine kinase genes (c-KIT, GRB10 (Kazi and Ronnstrand, 2013)); 2) Independent adverse prognostic factors in AML including WT1, CXXC5 (Kuhnl et al., 2015), MSI2 (Byers et al., 2011), etc.; 3) CD34, which acts as a marker for hematopoietic progenitor cells; 4) genes correlating with drug-resistant in AML (such as IGFBP2 (Kuhnl et al., 2011) and ABCC1 (Xu et al., 2013a)). The down-regulated genes included: 1) immune system activators such as ICAM1 and CD86; 2) Hematopoietic tumor suppressor ID2 (Ishiguro et al., 1996) and KLF4 (Seipel et al., 2016); 3) CEBPB, a gene negatively regulated by BCR/ABL, inhibits proliferation and promotes differentiation of BCR/ABL-expressing cells (Guerzoni et al., 2006).

Fig. 3.

Genome-wide genes and cell signaling pathways associated with CPT1A expression. (A) Volcano plot of differential gene profiles between CPT1A^high and CPT1A^low. (B) Expression heatmap of CPT1A-associated genes. The top curve shows CPT1A's expression distribution of 156 CN-AML samples. (C) Boxplot of direct-related signaling pathways.

Cell signal pathways, acting as functional units of gene groups, play important biological roles in major cellular processes, including cancer development and progression. Thus, we assessed the associations between CPT1A expression and cell signal pathways based on MSigDB database (Subramanian et al., 2005). Using the mean expression of all genes in a pathway to quantize its expression level, 14 down-regulated and 38 up-regulated pathways were found to be significantly associated with CPT1A^high (P < 0.05). The up-regulated included: 1) Leukemia-related pathways, such as AML and CML; 2) Metabolic pathways of nucleic acid, amino acid and protein, such as pyrimidine metabolism, basal transcription factors and arginine and proline metabolism; 3) Pathways of lipid metabolism, such as biosynthesis of unsaturated fatty acids, fatty acid metabolisms; 4) Other molecule-related pathways, such as RNA polymerase, DNA replication and Cell-cycle. However, the down-regulated were mainly immune-activation pathways, such as Natural Killer Cell-Mediated-Cytotoxicity, Toll-like Receptor, Nod-like Receptor and Antigen Processing-Presentation. These dysregulated genes and pathways were consistent with their known understanding about leukemogenesis, which might explain the involvement between CPT1A and the prognosis of CN-AML. (See Figs. 3C and S5)

3.6. Associations Between Genome-wide MicroRNA Profiles and CPT1A Expression

A genome-wide analysis of microRNA-sequencing data were carried out to identify microRNA profiles showing significant correlation to CPT1A expression. 109 microRNAs were found, including 89 positive and 20 negative (P < 0.05, Fig. 4A). Via filtering profiles with missing values, expression profiles of 76 microRNAs were presented as a heatmap (Fig. 4B). Firstly, positively correlated microRNAs included miR-222, miR-221, miR-20a, miR-17, miR-155, miR-26a, miR-335 and so on. All of those microRNAs have been found to have important tumor-promoting values in previous studies. miR-222/221 can enhance proliferation and differentiation blockade of melanoma cells (Felicetti et al., 2008). miR-20a and miR-17 can participate in the acquisition of apoptotic resistance in HL50, K562 and K562/ADR cells, inhibition of which helps to decrease the chemoresistance in leukemia therapy (Weng et al., 2014). Up-regulation of miR-155 is significantly associated with lower complete remission (CR), shorter disease-free survival (DFS) and OS, and can independently identify high-risk patients of CN-AML (Marcucci et al., 2013). miR-26a can target E2F7, and significantly sustains cell cycle progression and inhibits monocytic differentiation of AML cells (Salvatori et al., 2012). High bone marrow miR-335 level was significantly associated with a poor treatment response and shorter relapse-free survival and OS in AML patients (Yingchun et al., 2015). Secondly, negatively correlated microRNAs included miR-193a, miR-708, miR-152, miR-148a, miR-326, miR-22 and miR-27a. Previous study reported that miR-193a could repress c-Kit expression, and silencing of miR-193a contributed to leukemogenesis in AML (Li et al., 2013). Its down-regulation in CPT1A^high hints higher expression of c-Kit, which further aggravates worse outcome, which is consistent to CPT1A's prognostic role. miR-152 is crucial for anti-tumor effect of nature killer-cells by upregulating HLA-G (Bian et al., 2015). Down-regulation of miR-326 promotes cell proliferation in CML CD34 + cells (Babashah et al., 2013). Silencing of miR-708 could promote NF-kB signaling in CLL (Baer et al., 2015). miR-27a functions as a tumor suppressor in acute leukemia cells via regulation of apoptosis (Scheibner et al., 2012). All these aberrant regulations of microRNAs would contribute to the adverse outcome of CPT1A^high group.

Fig. 4.

Genome-wide microRNAs and microRNA-mRNA regulations associated with CPT1A expression. (A) Volcano plot of differential microRNAs. (B) Expression heatmap of associated microRNAs. The top curve shows CPT1A's expression distribution in 73 CN-AML samples. (C) Networks of microRNA-mRNA regulation.

To further clarify the biological functions of microRNAs, interaction-network was constructed based on the overlapping of results derived from microRNA-target predicting algorithm (Garcia et al., 2011) and correlation analysis (Fig. 4C). Up- and down-regulation of microRNAs and mRNAs were represented separately as red and green rectangles and ellipses, separately. Results showed some tumor suppressors were repressed by the up-regulated microRNAs, thus aggravated worse outcome. For example, CEBPB, targeted by miR-155, could inhibit proliferation and promote differentiation of BCR/ABL-expressing cells (Guerzoni et al., 2006). Meanwhile, targets of the down-regulated microRNAs included many oncogenes, such as ABI2, targeted by miR-193a, participated in the MLL translocation in AML (Coenen et al., 2012). RICTOR was targeted by miR-152. HHEX (Shields et al., 2016), PBXIP1 (Xu et al., 2013b) and EZH1 (Xu et al., 2015) were all targets of miR-148a, NET1 (Ahmad et al., 2014) and HOXA4 (Zangenberg et al., 2009) were targeted by miR-22; MAP3K4, ERG, ABL2, CDK6 and CDK8 were targeted by miR-27a; JAK1, MAP3K3, SRSF6 (Cohen-Eliav et al., 2013) and SRSF1 (Zou et al., 2012) were targeted by miR-708. These discoveries perhaps helped to understand why CPT1A^high could act as an adverse prognosticator.

In addition, the dynamic microRNA-mRNA pairs (MMR) were also investigated between CPT1A^high and CPT1A^low groups. Distinct MMR pairs were screened out for CPT1A^high and CPT1A^low separately, with significant negative correlation (P < 0.01) in CPT1A^high or CPT1A^low and without significant negative correlation (P > 0.1) in CPT1A^low or CPT1A^high groups, additionally with significant changes in absolute correlation coefficients between these two groups (Fig. S6A for CPT1A^high, Fig. S6B for CPT1A^low). 14 most significant NMR pairs were filtered out in CPT1A^high, including miR-196a and PIP4K2C (a tyrosine kinase), miR-32 and AKIRIN2, which promotes tumorigenicity and metastasis of Lewis lung carcinoma cells (Komiya et al., 2014), miR-7 and TRIM11, which is overexpressed in high-grade gliomas (Di et al., 2013). Our discoveries were consistent to all those known studies. 13 MMR pairs were found to be significantly correlated in CPT1A^low, including miR-374b and ZMAT3, the gene can promote cell cycle rather than apoptosis via directing the p53 stress response (Bersani et al., 2014). Taking together, those observations possibly provided rational interpretations for the prognostic role of CPT1A expression.

3.7. Abnormal Genome-wide Methylation Alterations Associated With CPT1A^high

DNA methylation is an important epigenetic mechanism, which can regulate gene expression through 3 DNA methyltransferases (DNMT1, DNMT3A and DNMT3B) and affect the behaviors of cancer cells. Considering the fact that CPT1A^high remarkably accompanied with higher expression of DNMT3A and DNMT3B, differential methylated regions (DMR) were derived to see the different methylation patterns between CPT1A^high and CPT1A^low groups of CN-AML. Firstly, compared to CPT1A^low, DNMT3A and DNMT3B showed significantly higher expression in CPT1A^high group (P = 0.001 and P < 0.001, respectively), and DNMT1 also showed a trend of higher expression (P = 0.182) (Fig. 5A). Secondly, 1201 hypermethylation and 455 hypomethylation DMRs were derived from the comparison between CPT1A^high and CPT1A^low (P < 0.05, FC > 2 or FC < 1/2, Fig. 5B). Thirdly, position distribution around CpG islands for these aberrant DMRs were compared. More hypermethylated DMRs were around the CpG islands (Island: P < 0.001, S_Shore: P = 0.03 and N_Shore: P < 0.001), while more hypomethylated DMRs lied in open sea regions (P = 0.02). (See upper subfigure of Fig. 5C). Finally, relative position distributions of different structural fragments for genes were presented. More hypomethylated DMRs lied on UTR regions (5′ UTR: P < 0.001 and 3′ UTR: P = 0.01), while more hypermethylation DRMs fell onto other regions (TSS200, TSS1500, Body and Others, all P < 0.001, lower subfigure of Fig. 5C). Further, CPT1A-associated DMRs were presented as heatmaps for those enriched on island and gene promoter regions (Fig. 5D and E).

Fig. 5.

Genome-wide methylation patterns associated with CPT1A expression. (A) Differential expression of three DNA methyltransferases (DNMT1, DNMT3A and DNMT3B). (B) Volcano plot of differential methylated regions. (C) Distribution of DMRs around the islands and on gene's different structural regions. (D) Methylation heatmap of DMRs on genome-wide CpG island regions. (E) Methylation heatmap of DMRs on genome-wide promoter regions.

4. Discussion

CPT1A protein can catalyze the rate-limiting step of FAO pathway, which may represent an alternative carbon source for anabolic processes of dNTP (Schoors et al., 2015). Targeting CPT1A has shown positive results for impairing cancer cell survival and inhibiting tumor cell proliferation in in vitro and in vivo models of Burkitt lymphoma (Pacilli et al., 2013). Recent study showed that CPT1A was widely expressed in all leukemia cell lines, and targeting CPT1A had a strong anti-leukemia activity in in vitro on human leukemia cell lines and primary cells of different hematologic diseases (Ricciardi et al., 2015). Furthermore, an editorial illustrated molecular mechanisms of FAO's support to mitochondrial function, and confirmed the therapeutic effectiveness of targeting CPT1A for hematological malignancies, alone or in combination with cytarabine (AraC) (Samudio and Konopleva, 2015).

Our study provided direct evidences that CPT1A^high predicted adverse outcomes for AML. Firstly, overexpression of CPT1A was shown in AML than normal BM and PB, and also in AML CD34 + and normal BM CD34 + cells (Fig. 1A, B and C), which facilitated the clinical use by qPCR detection. Secondly, based on two relatively larger, independent CN-AML cohorts, CPT1A^high was proved to have significant associations with adverse outcomes, which was also shown in ELN Intermediate-I subgroup and multivariable analyses (Fig. 2A and B, Table 2). Further, the prognostic value of CPT1A^high was confirmed in the whole AML patients and NCCN Intermediate Risk group (Fig. 2C and D), both of which contained a variety of karyotypes of AML samples. Combined the anti-leukemia activity of CPT1A's inhibitors (Ricciardi et al., 2015, Pacilli et al., 2013), all these results proved CPT1A as a potential prognosticator and therapeutic target of AML, which could promote further fine stratification of ELN Intermediate-I category and NCCN Intermediate-Risk group.

The pathogenesis of AML remains unclear, but previous studies have reported many aberrant molecular signatures, which could be used independently as favorable or adverse prognosticators and promote the understanding of pathogenesis for AML. The former includes NPM1 and double CEBPA mutation. The latter includes the presence of FLT3-ITD, higher expression of ERG and BAALC, WT1, DNMT3B, ITPR2 (Shi et al., 2015), TCF4 (In 'T Hout et al., 2014), ATP1B1 (Shi et al., 2016), MAPKBP1 (Fu et al., 2015) and RUNX1 (Fu et al., 2016). Associations between CPT1A expression and these known prognosticators were assessed, including some pretreatment molecular characteristics. Compared to the CPT1A^low group in the 156 CN-AML cohort, CPT1A^high group contained significantly more patients with M1 (P = 0.03) FAB subtype, suggesting that CPT1A^high expressers carried more immature cells, which might indicate adverse malignancy. CPT1A^high expressers presented significantly more FLT3-ITD and only slight trends of fewer NPM1 and double CEPBA mutations. In addition, CPT1A^high remarkably companied with high expression of ERG, BAALC, WT1, DNMT3A, DNMT3B, MAPKBP1, ITPR2, ATP1B1, RUNX1, TCF4 and CXXC5 in both cohorts of CN-AML (Fig. S3A and B), suggesting that CPT1A might be another independent biomarker of adverse outcomes for CN-AML.

Recent studies shows both genetics and epigenetics mechanisms play important roles in tumorgenesis, the former focuses primarily on aberrant gene expression, the latter mainly included microRNA's post-transcriptional regulation and DNA methylation. Thus, we further investigated possible mechanisms by which CPT1A^high affects outcomes from these three aspects. Firstly, we derived CPT1A-associated gene and microRNA expression profiles, including related cell signaling pathways. Many genes involved in tumorigenesis were up-regulated such as c-Kit, CDK6 and HOX family genes, including some known independent adverse prognosticators (WT1, ATP1B1, MSI2, etc.), down-regulated genes included many tumor suppressors and immune factors, such as ID2 (Ishiguro et al., 1996) and CD86 (Fig. 3B). The consistency to CPT1A's role of adverse prognosticator could also be seen in aberrant expression of cell signaling pathways, including up-regulation of AML, CML and cell cycle pathways, and down-regulation of immune-activation pathways such as Natural Killer pathway and Toll-like or Nod-like receptor pathways (Figs. 3C and S5).

More interesting results were the aberrant changes related to metabolisms. Many pathways of glycometabolism and lipid metabolisms were significantly associated with CPT1A expression, including many metabolic pathways of nucleic acid, amino acid and protein (Fig. S5). Especially, fatty acid pathway also showed a trend of up-regulation, which was reported previously as involved by CPT1A and promote tumor cell proliferation.

Secondly, genome-wide analysis also revealed some microRNA-involved mechanisms consist with CPT1A's prognostic role (Fig. 4B and C). Except those CPT1A-positive or -negative correlated microRNA individuals (such as miR-155 and miR-193a), which were known involved in tumor-promotion or progression or tumor suppression, many pairs of microRNA-mRNA were also shown aberrantly synchronized changes along with CPT1A expression, such as positive correlation of miR-155 and down-regulation of CEBPB, negative correlation of miR-193a and up-regulation of ABI2, etc. (Fig. 4C). Those synchronized changes of microRNAs and target genes may aggravate the malignancy of AML. Further, aberrant regulations of dynamic MMR pairs were also explored, which also showed some consistency to CPT1A's prognostic value, such as miR-196a and PIP4K2C, the significantly negative regulation of miR-196a to PIP4K2C could contribute to adverse outcome for CPT1A^high patients (Fig. S6A and B). Those aberrant genes, pathways, microRNAs and MMR pairs may contribute to the unfavorable outcomes for CN-AML.

Many microRNAs associated with CPT1A also participated in metabolic regulations, especially fatty acid metabolisms. miR-22 could repress fatty acid synthesis and was an important regulator of lipid and folate metabolism in breast cancer cells (Koufaris et al., 2016). miR-27a was reported to accelerate adipolysis releasing significantly more glycerol and free fatty acids (Wang et al., 2011). miR-155 could driver the metabolic reprogramming in ER + breast cancer cells (Bacci et al., 2016), and its overexpression could promote the aberrant expression of hepatic genes associated with lipid metabolism in a mice model of liver (Lin et al., 2015). All those results might illustrate the metabolic alterations derived from expression changes of CPT1A. Metabolic alterations have been proposed to support extra bioenergy and special microenvironment needed by tumor cell proliferations (Cairns et al., 2011). Our work and previous studies about FAO further showed catabolic signatures could participate in controlling cell growth, survival and chemo-resistance and might be used as a potential target for leukemia treatment (Ricciardi et al., 2015, Samudio and Konopleva, 2015).

Finally, inspired by the aberrant expression of those three DNA methyltransferases (DNMT1, DNMT3A and DMNT3B, Fig. 5A), associations between CPT1A expression and genome-wide methylation were exploited. Result showed that the nearer to the CpG islands lied more hypermethylation DMRs (Fig. 5C). Also, more DMRs related to genes' different functional fragments were in hypermethylation, especially the gene promoter regions, while hypomethylation DMRs fell more on the 3′ UTR or 5′ UTR regions (Fig. 5C). Hypermethylation on gene promoters and CpG islands were consistent to known understanding of the methylation influences to gene expression, which might contribute to the down-regulation of tumor suppressors in CPT1A^high expressers. Possibly, those abnormal methylation might be accompanying phenomenon to the adverse prognosis, and indicated extra understanding to leukemogenesis.

In summary, the involvement of metabolic alterations in tumorigenesis has obtained more and more acceptance, and attracted great attentions. In recent publications, CPT1A has been proved to have clear anti-leukemia efficacy in basic experiments. Combined with existing results of CPT1A in AML, our results in clinical further confirmed that high expression of CPT1A was an adverse biomarker for AML, and CPT1A^high patients carried various kinds of molecular adverse events. Taken together, CPT1A plays an important role in leukemogenesis and will be potentially a therapeutic target for AML.

The following are the supplementary data related to this article.

Table S1

Patients' characteristics in another cohort of 162 CN-AML patients according to CPT1A expression.

Fig. S1

Median value of CPT1A expression as the cut-off. (A) CPT1A expression is normally distributed. (B). OS and (C) EFS CN-AML patients were subdivided into four quartiles based on the quartile of CPT1A expression.

Fig. S2

Expression levels of CPT1A in various different subgroups of AML patients.

Fig. S3

Associations between CPT1A expression and known prognostic biomarkers (A) in the test cohort of 156 CN-AML; (B) in the validating cohort of 162 CN-AML.

Fig. S4

The prognostic value of CPT1A expression in another independent cohort of 162 CN-AML patients.

Fig. S5

Expression heatmap of CPT1A-associated cell signaling pathways.

Fig. S6

Significant MMR pairs in CPT1A^high and CPT1A^low.

Funding Sources

This work was supported by grants from the National Natural Science Foundation of China (61501519, 81500118), Chinese PLA General Hospital Nursery Foundation (14KMM38) and the China Postdoctoral Science Foundation funded project (project no. 2016M600443).

Conflict of Interest Disclosures

The authors declare no competing financial interests.

Authorship Contributions

J.L. Shi, L. Fu and H.P. Fu designed and performed the study and wrote the manuscript; Z.L. Jia participated in analyzing data and revising manuscript; L. Fu, W.D. Wang and K.L. He coordinated the study over the entire time. All authors reviewed the final manuscript.