Associations of PGK1 promoter hypomethylation and PGK1-mediated PDHK1 phosphorylation with cancer stage and prognosis: a TCGA pan-cancer analysis

Background Cancer cells reprogram metabolism for proliferation. Phosphoglycerate kinase 1 (PGK1), as a glycolytic enzyme and newly identified protein kinase, coordinates glycolysis and mitochondrial metabolism. However, the clinical significance of PGK1 expression and function in cancer progression is unclear. Here, we investigated the relationship between the progression and prognosis of multiple cancer types and PGK1 expression and its function in the mitochondrial metabolism regulation. Methods We performed pan-cancer analyses of PGK1 mRNA level and DNA methylation in 11,908 tumor tissues and 1582 paired normal tissues across 34 cancer types in The Cancer Genome Atlas datasets. Using specific antibodies against PGK1 S203 and PDHK1 T338 phosphorylation, we performed immunohistochemistry with tissue microarray assay in additional 818 cancer cases with 619 paired normal tissues from five cancer types. Results The PGK1 mRNA level was significantly elevated with hypomethylation in promotor regions and associated with advanced TNM stage in 15 and four cancer types, respectively. In breast carcinoma, elevated PGK1 mRNA level and promoter hypomethylation were associated with poor prognosis. Positively correlated PGK1 S203 and PDHK1 T338 phosphorylation levels were significantly associated with short overall survival (OS) in cancers of the breast, liver, lung, stomach, and esophagus and with advanced TNM stage in breast and esophageal cancers. PGK1 pS203 and PDHK1 pT338 were also independent predictors of short OS in liver, lung, and stomach cancer. Conclusions The elevated expression, promoter hypomethylation, and phosphorylation of PGK1 and PDHK1 were related with disease progression and short OS in diverse types of cancer. PGK1 and PDHK1 phosphorylation may be potential prognostic biomarkers.


Background
Most cancer cells, even in the presence of ample oxygen, predominantly generate adenosine triphosphate (ATP) by a high rate of glycolysis followed by lactate fermentation in the cytosol rather than by oxidation of pyruvate in the mitochondria, as in most normal cells. This phenomenon, known as aerobic glycolysis or the Warburg effect, facilitates tumor cell growth [1][2][3][4][5]. In the glycolytic pathway, phosphoglycerate kinase 1 (PGK1), the first enzyme to generate ATP, catalyzes the transfer of the high-energy phosphate from 1,3-diphosphoglycerate to adenosine diphosphate (ADP), leading to the generation of 3-phosphoglycerate and ATP. In Homo sapiens, PGK has two isozymes, ubiquitously expressed PGK1 and testis-expressed PGK2, both with 87%-88% identical amino acid sequence identity [6].
The reprogramming of metabolism is an emerging hallmark of cancer biology [4,7,8]. Recent studies have shown that the protein level of PGK1 was elevated in breast cancer [9], astrocytoma [10], metastatic colon cancer [11], and pancreatic ductal adenocarcinoma [12]; its mRNA levels were increased in gastric cancer [13].
Our previous studies [14] revealed that in tumor cells, PGK1 possesses protein kinase activity in addition to performing its well-established glycolytic function. In response to receptor tyrosine kinase activation, the expression of K-Ras G12V and B-Raf V600E, hypoxia, pyruvate metabolism in mitochondria is suppressed [14,15]. This is primarily regulated by the mitochondrial translocation of PGK1, which is phosphorylated at S203 by extracellular signal-regulated kinase 1/2 (ERK1/2) and cis-trans isomerized by peptidyl-prolyl cis-trans isomerase NIMA-interacting 1 (PIN1), leading to exposure of the pre-sequence of PGK1 for binding to the translocase of the outer membrane (TOM) complex of mitochondria. In the mitochondria, PGK1 functions as a protein kinase to phosphorylate pyruvate dehydrogenase kinase 1 (PDHK1, also known as PDK1) at T338, which activates PDHK1 to phosphorylate and inhibit the pyruvate dehydrogenase (PDH) complex [14,15]. Suppression of PDH activity reduces mitochondrial pyruvate utilization and reactive oxygen species production and increases lactate production, thereby promoting tumorigenesis. In addition, PGK1 S203 and PDHK1 T338 phosphorylation levels were found to be positively correlated with each other, and both were correlated with PDH S293 inactivating phosphorylation levels and poor prognosis in patients with glioblastoma (GBM) [14]. However, whether the newly identified protein kinase function of PGK1 applies to other cancer types and the relationship between PGK1 kinase activity and tumor progression remain unknown.
Here, we performed a pan-cancer analysis of clinical relevance of PGK1 using data from 11,908 cases (including 1582 with paired normal tissues) across 34 cancer types from The Cancer Genome Atlas (TCGA) datasets. We also analyzed the clinical relevance of PGK1 S203 and PDHK1 T338 phosphorylation levels by conducting immunohistochemical experiments in an additional 818 independent cancer cases (including 619 with paired normal tissues). We aimed to evaluate the pathological progression value and prognostic values of PGK1 mRNA high expression, PGK1 promoter methylation, and PGK1 mediated-PDHK1 activating phosphorylation in multiple human cancers.

Data resource
We downloaded clinical records, RNAseqV2 level 3 gene level data, and DNA methylation level 3 data for 11,908 cases across 34 cancer types from TCGA (http://xena. ucsc.edu/welco me-to-ucsc-xena/). Profiling data of the TCGA-retrieved cases were generated using the Illumina HiSeq 2000 RNA Sequencing and Illumina Infinium Human Methylation 450 platforms, as described by the TCGA network [16,17]. Gene transcription estimates for each gene were presented as in RNA-Seq using the Expectation Maximization (RSEM) software. DNA methylation values are presented as beta values for each CpG probe transformed into M values. The detailed information about data processing is provided in Additional file 1: Methods. A summary of the sample sizes for the PGK1 RNA-Seq and DNA methylation analyses for each cancer type is shown in Additional file 1: Table S1. There are 16 methylation probes that cover the PGK1 gene (Chromosome X; UCSC Gene Accession: NM_000291) (Additional file 1: Table S2).
The histopathologic diagnoses of the TCGA cases are available in the Genomic Data Commons (GDC, https :// porta l.gdc.cance r.gov/). PGK gene level data in 16 tissue types were downloaded from the Illumina Body Map Project (https :// www.ebi.ac.uk/gxa/home), and the results are presented as transcripts per million (TPM) values.

Patients and tissue samples
We retrospectively collected surgically resected, formalin-fixed, paraffin-embedded tissue samples from the biobank of National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital in Chinese Academy of Medical Sciences and Peking Union Medical College (Beijing, China). Tissue samples of 818 treatment-naïve patients who underwent surgery for pathologically diagnosed cancer between 2006 and 2015 were selected as an additional independent cohort, including 145 cases of breast carcinoma (BRCA) (with 69 paired normal specimens), 185 cases of liver hepatocellular carcinoma (LIHC) (with 174 paired normal specimens), 179 cases of lung adenocarcinoma (LUAD) (with 175 paired normal specimens), 95 cases of stomach adenocarcinoma (STAD) (with 55 paired normal specimens), and 214 cases of esophageal carcinoma (ESCA) (with 146 paired normal specimens).

Tissue microarray construction
Rabbit polyclonal antibodies recognizing phospho-PGK1 S203 and phospho-PDHK1 T338 were obtained from Signalway Antibody (College Park, MD, USA). The specificities of these antibodies were previously validated [14]. A rabbit monoclonal antibody recognizing IgG was purchased from Cell Signaling Technology (Danvers, MA, USA). Formalin-fixed, paraffin-embedded tissues were obtained by surgical resection, archived after clinical use for pathological diagnosis, and stained with Mayer's haematoxylin and eosin (H&E; Biogenex Laboratories, San Ramon, CA, USA).
Tumor samples from the 818 cancer cases with 550 paired normal tissues were subjected to tissue microarray (TMA). Employing an automated tissue array instrument (Minicore ® 3, Alphelys, Plaisir, France), cancer tissue (diameter at 2 mm, selected by a pathologist) from each specimen was extracted and fixed into a paraffin block. After quality control, the TMA blocks were sectioned into 3-μm-thick slides for immunohistochemistry analysis.

Statistical analysis
SPSS version 20.0 software (SPSS Inc., Chicago, IL, USA) was used for data analysis. PGK1 and PGK2 mRNA levels in tumor and normal tissues were compared using the independent variable t test. The associations between PGK1 mRNA levels, PGK1 pS203 and PDHK1 pT338 levels and clinicopathologic characteristics of patients were analyzed using one-way analysis of variance (ANOVA) with the post hoc Bonferroni test for multiple comparisons and least significant difference test. The methylation value (M value) method is regarded as more statistically valid than the beta value method [19]. The correlation between the M values of the 11 probes covering the PGK1 gene and PGK1 mRNA levels was analyzed using Spearman's correlation coefficient. The correlation between PGK1 pS203 and PDHK1 pT338 levels was analyzed using the Pearson correlation coefficient. Overall survival (OS) was defined as the duration from the date of diagnosis to death or the last known date of follow-up. The survival analyses were performed using the K-means cluster analysis to stratify the expression levels of related markers, the Kaplan-Meier method to plot survival curves, the log-rank test to compare survival rate, and a Cox regression model with two-sided Wald tests to calculate hazard ratios (HR) and 95% confidence intervals (CIs). Censored data were used for patients who were alive at last follow-up or lost to follow-up. Variables in univariate analysis with P values less than 0.05 were included in multivariate analysis. P < 0.05 was considered statistically significant. All statistical tests were two-sided.

PGK1 and PGK2 expression in human cancers
To compare PGK1 and PGK2 mRNA levels between tumor and normal tissues, we analyzed the RNA-Seq data of 11,908 tumor cases with 1582 paired normal tissues across 34 cancer types from TCGA datasets. In all types of cancer, the PGK1 mRNA levels in tumor and matched normal tissues were approximately 2 12 to 2 14 times higher than those of PGK2, which could hardly be detected (Fig. 1). In addition, the data from the Illumina Body Map Project further confirmed that PGK1, which had the highest mRNA level in leukocytes, was the major isozyme in all tissues except in the testes (Additional file 1: Fig. S1).

Fig. 1 PGK1
and PGK2 mRNA levels in human cancer tissues and matched normal tissues. The data of PGK1 (a) and PGK2 (b) mRNA levels in human tumors and paired normal tissues were obtained from TCGA RNA-Seq datasets as log RSEM values (see "Materials and methods" section) and plotted. The fold changes of median PGK1 mRNA level in tumors compared to that in normal controls in four cancer types are provided. Significant changes in median PGK1 mRNA level between tumors and normal controls are marked with asterisks (*P < 0.05, **P < 0.01, ***P < 0.001, independent variable t test). PGK1/2 phosphoglycerate kinase 1/2, TCGA The Cancer Genome Atlas, ACC adrenocortical cancer, BLCA bladder urothelial carcinoma, BRCA breast carcinoma, CESC cervical and endocervical cancer, CHOL cholangiocarcinoma, COAD colon adenocarcinoma, COADREAD colon and rectum adenocarcinoma, DLBC diffuse large B-cell lymphoma, ESCA esophageal carcinoma, GBM glioblastoma multiforme, HNSC head and neck squamous cell carcinoma, KICH kidney chromophobe, KIRC kidney clear cell carcinoma, KIRP kidney papillary cell carcinoma, LAML lymphoblastic acute myeloid leukaemia, LGG brain lower grade glioma, LIHC liver hepatocellular carcinoma, LUAD lung adenocarcinoma, LUSC lung squamous cell carcinoma, MESO mesothelioma, OV ovarian serous cystadenocarcinoma, PAAD pancreatic adenocarcinoma, PCPG pheochromocytoma and paraganglioma, PRAD prostate adenocarcinoma, READ rectum adenocarcinoma, SARC sarcoma, SKCM skin cutaneous melanoma, STAD stomach adenocarcinoma, TGCT testicular germ cell tumor, THCA thyroid carcinoma, THYM thymoma, UCEC uterine corpus endometrioid carcinoma, UCS uterine carcinosarcoma, UVM uveal melanoma Since nine cancer types did not have mRNA level data available for matched normal tissues, we analyzed 25 cancer types and found that 15 had significantly higher PGK1 mRNA level in tumor tissues than in normal tissues and that PGK1 mRNA levels were increased by more than twofolds in 4 of the 15 cancer types, namely, esophageal carcinoma (ESCA), uterine corpus endometrioid carcinoma (UCEC), cervical and endocervical cancer (CESC), and lung squamous cell carcinoma (LUSC) (Fig. 1a). In addition, the 15 cancer types account for 55.0% of cancer incidence and 63.1% of cancer mortality each year worldwide (data from GLOBOCAN 2018) [20], indicating that PGK1 overexpression is prevalent among the most deadly human cancers.

Association between PGK1 mRNA levels and human cancer progression
To access whether the PGK1 overexpression in tumor tissues was related to cancer progression, we analyzed the TNM staging data of the 15 cancer types with significantly increased PGK1 mRNA levels. We found that PGK1 mRNA level was significantly associated with progressive pathologic TNM stage in breast carcinoma (BRCA), CESC, liver hepatocellular carcinoma (LIHC), and lung adenocarcinoma (LUAD) (Fig. 2a). Further analyses showed that PGK1 mRNA level was significantly associated with T stage in BRCA, CESC, LIHC, and LUAD (Fig. 2b); N stage in LUAD (Fig. 2c); and M stage in BRCA (Fig. 2d).
Association between PGK1 promoter hypomethylation and PGK1 mRNA level elevation DNA methylation regulates gene expression and is implicated in tumor progression and therapeutic response [26,27]. We next determined the methylation status of the PGK1 gene in 14 cancer types with significantly elevated PGK1 mRNA levels in the TCGA data. GBM was not included in further comparison and association analysis because there were only two matched normal tissues with DNA methylation data available.
Median differential M values between tumor and normal tissues were plotted in the 14 cancer types (Additional file 1: Fig. S2). We found that 11 methylation probes had unanimous values in five types of cancer [stomach adenocarcinoma (STAD), bladder urothelial carcinoma (BLCA), ESCA, LIHC, and BRCA], and these values were lower in tumor tissues than in normal tissues (Fig. 3a, Additional file 1: Fig. S2). Intriguingly, these 11 probes were all located in the PGK1 promoter regions, ranging from 500 nt upstream of the transcription start site (TSS500) to the 5′-untranslated region (5′-UTR) (Additional file 1: Table S2; Additional file 1: Fig. S3), indicating that the PGK1 promoter regions were hypomethylated in these five types of cancer.
The methylation data for these probes did not follow a normal distribution (1-sample Kolmogorov-Smirnov test, asymptotic P < 0.001, two-tailed; Additional file 1: Table S3). We identified a significant inverse correlation between the methylation levels and mRNA levels of PGK1 in STAD, BLCA, ESCA, LIHC, and BRCA ( Fig. 3b; Additional file 1: Table S4), suggesting promoter hypomethylation as a mechanism promoting PGK1 expression. Among the five cancer types, BRCA showed the strongest correlation (Fig. 3b).

Associations between PGK1 promoter hypomethylation and PGK1 mRNA level elevation and poor prognosis in BRCA patients
We next analyzed the association between PGK1 promoter hypomethylation and the survival of patients with STAD, BLCA, ESCA, LIHC, and BRCA, and found that only in BRCA, hypomethylation of cg13203541 was associated with short OS (HR = 0.551, 95% CI 0.361-0.841, P = 0.005; Additional file 1: Table S5; Fig. 4a). A multivariate Cox regression model showed that cg13203541 methylation was an independent predictor of prolonged OS in BRCA (HR = 0.599, 95% CI 0.382-0.939, P = 0.026; Additional file 1: Table S5). In line with these results, an inverse correlation between PGK1 mRNA level and the OS of BRCA patients was also identified (HR = 1.966, 95% CI 1.535-2.519, P < 0.001; Fig. 4b). These results suggest that PGK1 promoter methylation and mRNA level may be prognostic markers for BRCA patients.

Correlation between elevated PGK1 pS203 and PDHK1 pT338 levels and their associations with cancer prognosis
ERK-phosphorylated PGK1 S203 (PGK1 pS203) and PGK1-phosphorylated PDHK1 T338 (PDHK1 pT338) levels were found to be strongly correlated with each other and were both associated with GBM prognosis [14]. Here, we examined the phosphorylation levels of these two proteins in additional 818 independent cancer cases. We found that in all five cancer types, PGK1 pS203   e Esophageal carcinoma (ESCA). *P < 0.05, **P < 0.01, ***P < 0.001 (2-tailed). PGK1 pS203 phosphorylated phosphoglycerate kinase 1 (PGK1) S203, PDHK1 pT338 phosphorylated phosphorylate pyruvate dehydrogenase kinase 1 (PDHK1) T338, HE hematoxylin-eosin staining and PDHK1 pT338 levels were higher in most tumor tissues than in their matched normal tissues (Additional file 1: Fig. S4), increasing from normal tissues to earlystage cancer tissues and to advanced carcinoma tissues (Fig. 5a-e), and positively correlating with each other in tumor tissues (Fig. 6). Kaplan-Meier analysis showed that higher levels of both PGK1 pS203 and PDHK1 pT338 were associated with shorter OS in patients with these five cancer types (all P < 0.05) (Fig. 7).

Prognostic values of PGK1 pS203 and PDHK1 pT338 in cancer
An independent variable t test showed that both PGK1 pS203 and PDHK1 pT338 were associated with advanced TNM stage in patients with BRCA and ESCA (all P < 0.05) ( Table 1). Univariate and multivariate Cox regression analyses showed that PGK1 pS203 was an independent predictor of short OS for LIHC (HR = 1.574, 95% CI 1.064-2.327, P = 0.023), LUAD

Discussion
In the present study, we identified relationships of high PGK1 mRNA level and promoter hypomethylation with advanced TNM stage and short OS in multiple cancer types in a pan-cancer analysis of TCGA data involving 11,908 cases covering 34 cancer types. Additional analyses of a cohort of 818 cases revealed that the phosphorylation levels of PGK1 S203 and PDHK1 T338 were independent prognostic biomarkers for LIHC, LUAD, and STAD. All these findings suggest that PGK1 gene modification and PGK1-mitochondrial function were significantly associated with clinical behaviors of cancer patients. Metabolic reprogramming plays an important role in tumorigenesis [4,[28][29][30][31][32]. It is emerging that the nonmetabolic functions of metabolic enzymes are fundamental to tumorigenesis [33]. We reported that the protein kinase activity of metabolic enzymes, such as PGK1 [14,15,34], pyruvate kinase M2 (PKM2) [35,36], and ketohexokinase isoform A (KHK-A) [37], regulates the Warburg effect, gene expression, cell proliferation, and autophagy [28,29].We previously found that mitochondrial PGK1 functions as a protein kinase to promote tumor cell proliferation and brain tumorigenesis [14]. In the present study, we investigated the clinical relevance of PGK1 in human cancers and found that    elevated PGK1 mRNA level and PGK1 protein kinase activity were associated with advanced TNM stages and poor prognosis in multiple human cancers. Whether other metabolic enzymes with protein kinase activity can act as potential biomarkers for the prediction of progression and prognosis of human cancers should be further analyzed.
The up-regulation of PGK1 involved in the Warburg effect has been detected in several types of human cancer [9][10][11][12][13]. Several other studies also reported the relationship between PGK1 acetylation and its innate enzymatic activity [38,39]. However, all these studies focused on the metabolic function of PGK1 without elucidating the relationship between cancer progression and gene modification and the mitochondrial function of PGK1. Importantly, recent studies showed that metabolic changes in cancer alter the epigenetic landscape, especially DNA modifications, leading to malignant transformation, adaptation to inadequate nutrition, and tumor development [40]. Therefore, in the present study, we analyzed the DNA methylation data for 14 cancer types from TCGA datasets and identified hypomethylation of the PGK1 promoter (cg13203541) as an independent prognostic biomarker in BRCA patients (Additional file 1: Table S5). We also detected mitochondrial PGK1-dependent PDHK1 T338 phosphorylation in additional cases of five cancer types and demonstrated that mitochondrial function of PGK1 significantly affected the clinical behaviors of patients with these cancers.
Reprogrammed energy metabolism is an emerging hallmark of cancer biology and is an important way to treat cancer [4,[29][30][31]41]. One important example is isocitrate dehydrogenase 1 (IDH1) mutation, which has important clinical significance and was found in GBM [42] and myeloid malignancies, such as acute myelocytic leukaemia (AML) [43] and myelodysplastic syndromes (MDS) [44]. A clinical study suggested that IDH1 mutation was an independent, favorable prognostic marker in grade 2-4 glioma [45]. Related clinical trials are ongoing in AML [43,44]. In the present study, we found another metabolism reprogramming mediated by PGK1 protein kinase activity-dependent phosphorylation, which was associated with clinical behaviors of cancer patients and was an independent prognostic biomarker in multiple types of cancer. In addition, we revealed a high percentage of patients exhibiting elevated protein kinase activity of PGK1 compared to a relatively low IDH1 mutations in cancer patients [42,43,45]. Thus, we underscore that PGK1 protein kinase activity is a potential target for cancer treatment.
Our research has several limitations. First, the number of cases of some cancer types were rather limited in TCGA datasets. For example, only 36 tumor samples of cholangiocarcinoma and 66 tumor samples of kidney chromophobe were identified. Second, data of normal samples were not available in several cancer types, therefore, some analyses could not be performed for these cancer types. Third, validation of the association of PGK1 phosphorylation with clinicopathological characteristics could only be made in independent Chinese cohorts covering five cancer types as there were few publicly available datasets regarding the phosphorylation levels of proteins, and we could not validate those associations in Caucasians tumor samples.

Conclusions
We demonstrated a relationship between PGK1 promoter methylation and PGK1 mRNA level and demonstrated the significance of PGK1 mRNA level, PGK1 promoter methylation, and PGK1 pS203 and PDHK1 pT338 levels in tumor progression and cancer patient survival. These findings highlight the potential use of PGK1 mRNA level, PGK1 promoter hypomethylation, and PGK1 pS203 and PDHK1 pT338 levels as biomarkers for cancer progression and prognosis, and the promising significance of PGK1 as a target in cancer treatment.