Effects of PKM2 on global metabolic changes and prognosis in hepatocellular carcinoma: from gene expression to drug discovery
- 571 Downloads
Hepatocellular carcinoma (HCC) is a malignant tumor that threatens global human health. High PKM2 expression is widely reported in multiple cancers, especially in HCC. This study aimed to explore the effects of PKM2 on global gene expression, metabolic damages, patient prognosis, and multiple transcriptional regulation relationships, as well as to identify several key metabolic genes and screen some small-molecule drugs.
Transcriptome and clinical HCC data were downloaded from the NIH-GDC repository. Information regarding the metabolic genes and subsystems was collected from the Recon 2 human metabolic model. Drug-protein interaction data were obtained from the DrugBank and UniProt databases. We defined patients with PKM2 expression levels ≥11.25 as the high-PKM2 group, and those with low PKM2 expression (< 11.25) were defined as the low-PKM2 group.
The results showed that the global metabolic gene expression levels were obviously divided into the high- or low-PKM2 groups. In addition, a greater number of affected metabolic subsystems were observed in the high-PKM2 group. Furthermore, we identified 98 PKM2-correlated deregulated metabolic genes that were associated with poor overall patient survival. Together, these findings suggest more comprehensive influences of PKM2 on HCC. In addition, we screened several small-molecule drugs that target these metabolic enzymes, some of which have been used in antitumor clinical studies.
HCC patients with high PKM2 expression showed more severe metabolic damage, transcriptional regulation imbalance and poor prognosis than low-PKM2 individuals. We believe that our study provides valuable information for pathology research and drug development for HCC.
KeywordsHepatocellular carcinoma PKM2 Metabolic Overall survival Drugs
False discovery rate
Gene set enrichment analysis
Normalized enrichment score
RNA-seq by expectation-maximization
The Cancer Genome Atlas
Transcriptional regulatory network
Hepatocellular carcinoma (HCC) is one of the two major forms of primary liver cancer and accounts for 85–90% of all primary liver cancers . HCC is typically caused by viral hepatitis infection or fatty liver disease. Cryptogenic cirrhosis, which is frequently linked with nonalcoholic steatohepatitis (NASH) / nonalcoholic fatty liver disease (NAFLD) in patients with metabolic syndrome, diabetes and obesity, is an increasingly significant cause of HCC . Furthermore, metabolic syndrome itself is also a strong risk factor for HCC . Increasing evidence suggests that diabetes is an independent risk factor for HCC , particularly diabetes controlled by diet, insulin or sulfonylureas . Additionally, a series of metabolic alterations in HCC, including elevated glycolysis, gluconeogenesis, and b-oxidation levels and reduced tricarboxylic acid cycle and D-12 desaturase levels, were observed .
Pyruvate kinase (PK) catalyzes the last and physiologically irreversible step in glycolysis, the conversion of phosphoenol pyruvate (PEP) to pyruvate via the transfer of a phosphate group to ADP . In mammals, four PK isoforms exist that are encoded by two genes. The PKLR gene encodes PKL and PKR . The former is exclusive to erythrocytes, and the latter is expressed primarily in the liver, with low expression in the kidney. The PKM1 and PKM2 isoforms are encoded by PKM through alternative splicing of the mutually exclusive exons 9 and 10, which generate 56-amino acid regions that differ at 22 residues [7, 8]. PKM1 is mainly expressed in brain, heart, and muscle tissue, whereas PKM2 is expressed in most tumor cells, embryonic tissue and many adult tissues, including the kidney, spleen, lung and intestine [7, 9]. PKM2 plays a central role in maintaining the metabolism program of cancer cells and other proliferating cell types and is over-expressed in a broad range of human cancers .
High PKM2 expression is correlated with poor prognosis for patients with multiple types of solid tumors compared with the prognosis of patients with low PKM2 expression levels [10, 11]. Our previous study also found high PKM2 expression is independently associated with poor overall survival in HCC patients . The knockdown of PKM2 in HCC cells inhibited cell proliferation and induced apoptosis in vitro and in vivo . Furthermore, PKM2 plays critical in Warburg effect, gene expression, cell cycle progression and many other fundamental cellular functions and is highlighted as an important integrator of diverse cellular stimuli to modulate metabolic flux and cancer cell proliferation . Recent studies found that PKM2 is the dominant form highly expressed in HCC and is a direct target of miR-122, which serves as a prognostic biomarker and induces apoptosis and growth arrest by downregulating PKM2 in HCC [15, 16].
Establishment of a human metabolic model provided considerable data support and systematic analytical methods for human metabolic disease research . In this study, we obtained information regarding all human metabolic genes and metabolic subsystems from the Recon2 human metabolic model . Considering that HCC is a metabolic disease, we aimed to explore the associations between PKM2 expression and overall metabolic changes in HCC, identify several risk metabolic genes and discover some drugs.
HCC RNA-seq and clinical data were downloaded from the NIH-GDC repository (https://portal.gdc.cancer.gov/) and cBioPortal (http://www.cbioportal.org/) databases . In total, we downloaded TCGA level 3 data that included 371 primary hepatocellular carcinoma patients and 50 controls. All samples were analyzed by Illumina HiSeq 2000 RNA Sequencing Version 2. RNA-seq by expectation-maximization (RSEM) expression values were used for statistical analysis. We divided all HCC samples into two groups according to their log2-converted median expression of PKM2 (11.25). Case samples with PKM2 expression ≥11.25 comprised the high-PKM2 group, and those with low PKM2 expression (< 11.25) comprised the low-PKM2 group. PKM2 copy number data were downloaded from the cBioPortal database and PKM2 promoter methylation data were obtained from MethHC (http://methhc.mbc.nctu.edu.tw/php/search.php?opt=gene). The methylation data contains 204 HCC patients and 49 controls, and the copy number data contains 364 HCC patients but no controls. We mapped these samples to the transcriptome data and compared the difference among controls, low-PKM2 patients and high-PKM2 patients. The validation datasets of GSE14520 (contributed by Roessler S, including 225 patients) and GSE25097 (contributed by Tung EK, including 268 patients) were downloaded from NCBI-GEO (https://www.ncbi.nlm.nih.gov/geo/). We used these two datasets to verify the effect of PKM2 on global metabolic gene expression.
Preprocessing and differential expression analysis
R statistical software v3.3.3 was used to perform data preprocessing. We removed genes with median expression values = 0 in all patients. In total, we filtered 16706 genes in 371 HCC patients and 50 controls. All expression values were log2-transformed. Human metabolic genes were extracted from the Recon 2 human metabolism model (https://www.vmh.life/). This model contains 1789 unique genes that belong to 99 metabolic subsystems. We mapped 1492 metabolic genes and 95 subsystems in our HCC data. Differentially expressed gene analysis between HCC patients and controls in the high- and low-PKM2 groups was performed using the empirical Bayes algorithm (function “eBayes”) in the “limma” package . Differences (up- or down-regulated) were considered statistically significant for absolute value log2-transformed fold-changes ≥1 and false discovery rate (FDR) adjusted P values ≤0.05.
Metabolic subsystem enrichment and correlation analysis
We used javaGSEA desktop application v3.0 (http://software.broadinstitute.org/gsea/index.jsp)  to perform gene set enrichment analysis (GSEA) of the mapped metabolic subsystems in high-PKM2 vs. controls, low-PKM2 vs. controls and high-PKM2 vs. low PKM2. Gene sets with less than 10 genes or more than 500 genes were excluded. The t-statistic mean of the genes was computed in each metabolic subsystem using a permutation test with 1000 replications. Subsystems with normalized enrichment scores (NESs) > 0 were considered up-regulated, and subsystems with NESs < 0 were considered down-regulated. Statistical significance was identified as P values ≤0.05. A univariate correlation model was used to analyze correlation between PKM2 and other metabolic genes in high- and low-PKM2 patients. Genes with absolute correlation coefficient values ≥0.5 and P values ≤0.05 were considered significant.
Single gene and gene interaction survival analyses
We used Venn diagrams to show the overlapping genes among the up-regulated and PKM2-positively correlated genes as well as the down-regulated and PKM2-negatively correlated genes in the high- and low-PKM2 groups. All survival analyses were conducted using “survival” package in R. A Cox proportional hazards model was used to estimate the independent effects of these overlapping genes on the total overall survival of the HCC patients. Each gene was divided into two groups according to its median expression. Patients with gene expression levels less than the median comprised the low-expression group, and those with gene expression levels higher than the median comprised the high-expression group. The reference group of up-regulated genes was defined as patients in the low-expression group, and the reference group of down-regulated genes was defined as patients in the high-expression group. According to our grouping, genes with hazard ratios (HRs) > 1 were considered risk factors. Conversely, genes with HRs < 1 were considered protective factors. To explore the effect of interactions between PKM2 and metabolic genes on patients’ overall survival, we used Cox proportional hazards model to perform interaction survival analysis of PKM2 and metabolic genes that uncorrelated with PKM2. We screened these uncorrelated genes that were not affect patients’ prognosis alone and no expression difference in high- or low-PKM2 patients compared to controls. The “simPH” package  in R was used to simulate and plot quantities of interactions from Cox proportional hazard models. The Kaplan-Meier curves were used to show the difference of these interactions on patients’ survival. A P value ≤0.05 was considered as significant.
Transcriptional regulatory network analysis
Using the analysis described above, we defined overlapping genes with survival differences as risk metabolic genes. We used the TRRUST v2.0 web server  to find the transcription factors (TFs) targeting these genes. TRRUST v2.0 is a manually curated database of human and mouse transcriptional regulatory networks. It contains 9,996 TF-target regulatory relationships for 824 human TFs and provides information regarding the regulation type (such as activation or repression) between the queried TFs and target genes (http://www.grnpedia.org/trrust/). All these relationships were derived from PubMed articles. We used these TFs and risk metabolic genes to construct transcriptional regulatory networks (TRNs) in high- and low-PKM2 groups. Because the transcriptional regulation information of multiple risk metabolic genes is not recorded in the database, the co-expression method was used to construct TRNs. The correlation coefficients of these TFs and risk metabolic genes were calculated in each group, and the TF-target pairs with absolute value of correlation coefficients ≥0.5 and FDR P-values ≤0.05 were selected to construct the TRN.
Construction of the risk metabolic genes and drug network
We used these risk metabolic genes and drugs that target these proteins to construct the network. Information regarding the drugs and targeting proteins was downloaded from the DrugBank database (https://www.drugbank.ca/) . Because the records in the DrugBank database are protein IDs and drugs interactions, we used UniProt (http://www.uniprot.org/) to find the protein IDs corresponding to these risk metabolic genes. By combining the DrugBank and UniProt databases, we obtained 14834 gene-protein-drug interactions in total, including 2127 unique genes and 6255 unique drugs. We extracted the interactions of risk metabolic genes and drugs and used Cytoscape v3.4.0 to visualize the network.
Global metabolic gene expression levels were different in the high- and low-PKM2 groups
Patients in the high-PKM2 group showed more severe metabolic abnormalities
PKM2-correlated deregulated genes associated with poor overall survival
Transcriptional regulatory networks of risk metabolic genes were related to PKM2 expression
Network of risk metabolic genes and drugs that target these genes
This study found that patients expressing PKM2 at high levels exhibited abnormal expression of whole metabolic genes, detrimental metabolic subsystems changes, unfavorable prognoses, and transcriptional regulation imbalances compared with low-PKM2 patients. In addition, we identified 98 risk metabolic genes that were significantly correlated with PKM2. Utilizing the DrugBank and UniProt databases, we screened 204 drugs that target these genes.
HCC is metabolically different from normal liver tissue in many ways, including glycolysis, lipid metabolism, the TCA cycle, amino acid metabolism, and several PPAR signaling pathways . In addition, close relationships have been observed among metabolic syndrome, nonalcoholic fatty liver disease, and liver cancer . In the American population, metabolic syndrome occurrences were significantly higher in HCC (37.1%) and intrahepatic cholangiocarcinoma (ICC) (29.7%) patients than in the normal group (17.1%, P < 1E-4) . Furthermore, metabolic syndrome is not only a risk factor for HCC in elderly people, but it also promotes cancer in people younger than 65 years . A prospective study of 578,700 adults suggested that high BMI, blood glucose and composite metabolic syndrome z-scores were positively associated with liver cancer risks . Another study in a Swedish population also drew a similar conclusion that metabolic syndrome components were strongly associated with primary liver cancer . In a normal liver, glucose is phosphorylated to glucose-6-phosphate (G6P) by glucokinase (GCK). Normally, G6P undergoes glycolysis and is converted to pyruvate before going through the citric acid cycle in mitochondria to generate ATP through oxidative phosphorylation . However, liver tumor cells produce ATP by aerobic glycolysis, called the Warburg effect, which is accompanied by increased glycolysis and decreased oxidative phosphorylation. In our study, we observed many genes that were over-expressed in glycolysis (including HKDC1, HK2, PFKP, ENO2, ALDOA, GPD2, G6PC3, BPGM, etc.) in high-PKM2 patients. By contrast, more genes were down-regulated in gluconeogenesis, resulting in negative normalized enrichment scores (NESs) in the GSEA results. Pyruvate kinase is activated by its substrate and inhibited by high levels of ATP. In tumors, low enzymatic activity of PKM2 favors the Warburg effect. In addition, high PKM2 mRNA expression was associated with tumor growth and poor outcomes . In our study, patients with high PKM2 expression showed more severe metabolic abnormalities and poorer overall survival than low-PKM2 patients. Furthermore, many down-regulated metabolic genes were significantly negatively correlated with PKM2 in the high-PKM2 group. These results suggest a pleiotropic effect of PKM2 on HCC.
Our study observed that 98 PKM2-correlated genes had a certain relationship with prognosis. The expression statuses of these genes in high-PKM2 patients were linked to poor survival, and some have been widely reported. Solute carrier family 2 member 1 (SLC2A1), also known as GLUT1, encodes a major glucose transporter in the blood-brain barrier. Studies showed that GLUT1 is over-expressed in HCC and suggested GLUT1 as a potential target . In our study, high GLUT1 expression was independently associated with poor overall survival, and its interaction with PKM2 exacerbated this consequence. SLC16A3, belonging to the proton-linked monocarboxylate transporter (MCT) family, was significantly up-regulated in liver cancer in multiple datasets  and in our study. Deregulation of this gene may influence intracellular acidic environments . Our study showed that high levels of SLC16A3 and PKM2 expression interactively led to poor prognosis. Dimethylglycine dehydrogenase (DMGDH) was expressed at low levels, negatively correlated with PKM2 and correlated with low patient survival in this study. Decreased DMGDH levels in liver tumor cells and the effects of DMGDH over-expression on the suppression of migration, invasion and metastasis has already been proven. Thus, DMGDH is suggested to be a potential diagnostic and prognostic marker for HCC . SLC22A1, encoding organic cation transporter-1 (OCT1), was expressed at low levels, associated with worse HCC patient survival and significantly associated with advanced HCC stages . Furthermore, a previous study showed that HCC and cholangiocarcinoma development is accompanied by the appearance of aberrant SLC22A1 variants. Thus, aberrant SLC22A1 variants together with its low expression may dramatically affect the ability of sorafenib to reach active intracellular concentrations in HCC tumors . In this study, we also observed that decreased SLC22A1 expression was associated with poor prognosis, and its interaction effect with PKM2 further reduced patient survival. In addition, this study identified ATP1B3, SLC25A15 and other key metabolic genes that were deregulated in HCC and associated with patient overall survival. However, these genes have not yet been reported. Given that these genes were correlated with PKM2, we believe that a more complex mechanism of PKM2 and metabolic genes may exist in HCC and needs to be further studied.
The transcriptional regulation relationships of these key metabolic genes showed a relatively large difference between the high- and low-PKM2 groups, especially HNF4A and its target genes. Hepatocyte nuclear factor 4 alpha (HNF4A) is a nuclear transcription factor that binds DNA as a homodimer, and it has distinct transcriptional regulatory mechanisms in human liver and brain tissues . Similar to our findings, HNF4A and several HNF4A target genes were down-regulated in liver cancer patients . Suppression of HNF4A in mouse livers showed profound influences on zonated metabolic functions, cell proliferation and oncogenesis . An HNF4A-microRNA-194/192 signaling axis was suggested to influence glucose metabolism, cell adhesion and migration, tumorigenesis and tumor progression, as well as epigenetic regulation in a recent study . In our study, HNF4A transcriptionally regulated LIPC, ACMSD, AKR1C4, CYP27A1 and other metabolic genes and showed differential expression in the high- and low-PKM2 groups. In addition, FOS, MYCN, POU5F1 and other TFs also showed differential expression and regulation in high- and low-PKM2 patients. Combined with previous reports, we speculate that the carcinogenic effects of these TFs may be influenced by PKM2 expression in HCC.
High PKM2 expression in liver tumor cells is accompanied by a decrease in total cellular pyruvate kinase activity. Notably, elevated PKM2 enzyme activity may compromise both its proanabolic and antioxidant functions and shows an anti-cancer effect in HCC . Therefore, the use of small-molecule PKM2 activators may be an appropriate approach to disrupt cancer cell metabolism for therapeutic purposes , and PKM2’s substrate may be a good target choice . Many of the drugs screened in this study are essential human metabolites. Although some of these supplements may promote cancer cell growth, many studies also reported that metabolites supplement have a tumor suppressor effect. A previous study reported that S-adenosylmethionine (SAMe) can selectively induce Bcl-xS that promotes apoptosis in liver cancer cells . Aminooxyacetic acid (AOAA) is a cystathionine-β-synthase (CBS) inhibitor, studies showed it suppresses the proliferation of colon cancer cells in vitro and reduces tumor growth in vivo [41, 42]. Another study suggested that hexachlorophene cause the growth arrest of B lymphoma cells by reducing the expression of cyclin-D1 and c-Myc . Furthermore, studies reported that nontoxic compound disulfiram (DSF) targeted tumor cellular copper inhibited of the proteasomal activity, resulting in tumor cells apoptosis induction . The compounds screened in this study contain the above anti-cancer activity drugs, therefore, we believe that these screened compounds may help to liver cancer treatment.
In summary, this study showed several unfavorable effects of high PKM2 expression on whole gene expression, metabolic functions, prognosis and transcriptional regulation relationships in HCC patients. Therefore, we recommend PKM2 and PKM2-correlated risk genes can be used as biomarkers or therapeutic targets of HCC. In addition, we screened several drugs that target with these risk genes. This study pointed complex regulatory mechanisms of PKM2 and other risk metabolic genes on HCC, and the potential carcinogenic mechanisms require further study.
This work was supported by the National Natural Science Foundation of China (grants no. 81570376 and no. 81573231), the Natural Science Foundation of Shanghai (grant no. 15ZR1424400), the Three-year Action Plan on Public Health, Phase IV, Shanghai, China (grant no. 15GWZK0801), SJTU Interdisciplinary Research Key (grant no. YG2015ZD01) and Shanghai Municipal Planning Commission of Science and Research Fund (grant no. 20144Y0249). The funding bodies allowed to collect, analyze, interpret the data and write the manuscript.
Availability of data and materials
The transcriptome data and clinical data are available in the NIH-GDC repository (https://portal.gdc.cancer.gov/). All human metabolic genes and metabolic subsystems are available in the Recon2 human metabolic model . Copy number data are available in cBioPortal database (http://www.cbioportal.org/). Methylation data are available in MethHC (http://methhc.mbc.nctu.edu.tw/php/search.php?opt=gene). Human transcriptional regulatory relationship data are available in TRRUST (http://www.grnpedia.org/trrust/). Drug and protein information are available in DrugBank (https://www.drugbank.ca/) and UniProt (http://www.uniprot.org/).
WXL and WWL designed the study. WXL, WWL, DL and XCL performed data analysis. WXL, WWL, TNF, LL and BYQ wrote the manuscript. WXL, WWL and BYQ revised the manuscript. All authors read and approved the final manuscript.
Ethics approval and consent to participate
Consent for publication
The authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
- 21.Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A. 2005;102(43):15545–50.CrossRefGoogle Scholar
- 26.Kanda T, Imazeki F, Wu S, Nakamoto S, Yokosuka O. The assessment of serum hepatitis C virus RNA 12 weeks after the end of treatment using TaqMan polymerase chain reaction is less relevant than after 24 weeks for predicting sustained virological response. Hepatology. 2011;54(4):1482 author reply 1482-1483.CrossRefGoogle Scholar
- 29.Nderitu P, Bosco C, Garmo H, Holmberg L, Malmstrom H, Hammar N, Walldius G, Jungner I, Ross P, Van Hemelrijck M. The association between individual metabolic syndrome components, primary liver cancer and cirrhosis: a study in the Swedish AMORIS cohort. Int J Cancer. 2017;141(6):1148–60.CrossRefGoogle Scholar
- 33.Heise M, Lautem A, Knapstein J, Schattenberg JM, Hoppe-Lotichius M, Foltys D, Weiler N, Zimmermann A, Schad A, Grundemann D, et al. Downregulation of organic cation transporters OCT1 (SLC22A1) and OCT3 (SLC22A3) in human hepatocellular carcinoma and their prognostic significance. BMC Cancer. 2012;12:109.CrossRefGoogle Scholar
- 41.Szabo C, Coletta C, Chao C, Modis K, Szczesny B, Papapetropoulos A, Hellmich MR. Tumor-derived hydrogen sulfide, produced by cystathionine-beta-synthase, stimulates bioenergetics, cell proliferation, and angiogenesis in colon cancer. Proc Natl Acad Sci U S A. 2013;110(30):12474–9.CrossRefGoogle Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.