Background
Circulating tumor cells (CTCs) are tumor cells derived from the tumor origin or metastasis and released into the bloodstream. The dissemination and tumorigenicity of CTCs [
1,
2] indicate that they are important seeds for tumor metastasis, which is the leading cause of poor prognosis in cancer patients. Compared with traditional tissue biopsy, CTCs detection helps make possible non-invasive screening and monitoring of cancer. Recent studies have demonstrated the expanding roles of CTCs counts in the prognosis of breast, colorectal, lung, and prostate cancers [
3‐
5]. However, merely detecting the number of CTCs has limited benefit for clinical interpretation when inter- and intra-tumor heterogeneities are considered [
6,
7]. Further exploration of the genetic and phenotypic features of CTCs could provide information on the nature of individual tumor cells, which would be useful for disease evaluation and therapeutic decision-making.
The epithelial-mesenchymal transition (EMT) classification is the most investigated phenotypic characteristic of CTCs in recent years. EMT plays a vital role in tumor metastasis. The epithelial cells gain mesenchymal properties for cell movement via EMT, whereas the disseminated cells might recover epithelial properties for rapid colonization through mesenchymal-epithelial transition (MET) [
8]. Hence, CTCs phenotypes might include E-CTCs (epithelial), M-CTCs (mesenchymal), and H-CTCs (hybrid). Increased H-CTCs and M-CTCs have been reported as more relevant to metastatic potential and aggressive progression [
9,
10]. Nevertheless, static morphological analysis may be unsatisfactory for determining the change and outcome of CTCs because of the dynamic variation of EMT and MET [
8]. It remains unclear whether all CTCs that present the same plastic EMT phenotype are capable of going through the metastatic cascade to form metastases. Thus, a great need exists for developing biofunctional activity markers to evaluate the actual roles of CTCs in cancer metastasis.
The deregulation of cellular energetics is a hallmark of tumor cells [
11] and plays a crucial role in cancer progression by promoting EMT, anoikis resistance, angiogenesis, and cell stemness [
12]. Glucose metabolic reprogramming is characterized by active aerobic glycolysis together with enhanced pentose phosphate pathway and glutaminolysis. These processes promote the accumulation of precursor molecules and activation of signaling pathways to modulate cell proliferation and migration [
13,
14]. Our previous studies have suggested that tumor cell malignancy is closely associated with metabolic conversion toward a glycolytic pattern [
15]. Many research works have also highlighted the decisive effect of metabolic transition on tumor cell behavior [
16,
17]. For instance, the pro-metastatic protein S100A4 was reported to stimulate invasiveness in poorly motile melanoma cells by suppressing mitochondrial activity and potentiating glycolysis [
16]. PCK2 is a crucial factor for the metabolic switch from oxidative phosphorylation to aerobic glycolysis. Zhao et al. [
17] showed that PCK2 knockdown reduced the tumor-initiating ability of prostate cancer (PCa) cells in the in vivo xenograft models and that the increased PCK2 level was associated with more aggressive tumors and lower survival rates for PCa patients.
Accordingly, it is widely agreed that metabolic reprogramming is a crucial feature of the highly aggressive tumor cells. In the present work, we aim to utilize the metabolic characteristics to establish a functional activity indicator for CTCs, and explore the feasibility and significance of the metabolic markers in CTCs detection.
Methods
Cell lines and cell culture
Paired isogenic PCa cell lines PC-3 M 2B4 and PC-3 M 1E8 which differ in metastatic ability were purchased from the Institute of Basic Medical Sciences (Chinese Academy of Medical Sciences, Beijing, China). LNCAP, PC-3, and DU145 cell lines were bought from the Cell Bank of Type Culture Collection (Chinese Academy of Sciences, Shanghai, China). LNCAP and DU145 cells were maintained in Dulbecco’s Modified Eagle Medium (Gibco, Gaithersburg, MD, USA). PC-3 M 2B4, PC-3 M 1E8, and PC-3 cells were maintained in Roswell Park Memorial Institute 1640 Medium (Gibco). Cells were supplemented with 10% fetal bovine serum (Gibco) and cultured in a humidified incubator (37 °C, 5% CO2).
Wound healing assay
Cell migration rates were assessed by wound healing assay. Cells were routinely cultured until they reached monolayer confluence. Wounds were created with a 200 μL tip (Corning, NY, USA). After washing with phosphate-buffered saline (Gibco), the cells were incubated with fresh serum-free medium for 24 h. The wound widths were measured using an inverted microscope (Olympus CKX41, Tokyo, Japan). The migration rate was calculated as the average wound closure percent of five random fields in three repetitions.
Cell migration and invasion assay
Cell migration and invasion assays were performed using Transwell chambers (Corning). For the migration assays, 5 × 104 cells suspended in 200 μL of serum-free medium were added to the upper chamber and 500 μL of medium containing 10% FBS was added to the lower chamber. After routine incubation for 24 h, the upper chambers were stained with a crystal violet kit (Beyotime, Shanghai, China). Cells outside the upper chamber were imaged and counted in five random fields using a microscope (Olympus Bx51, Tokyo, Japan). Similar protocols were performed for the invasion assays, except that 50 μL of Matrigel (BD Biosciences, NJ, USA) was added to the upper chamber overnight before the cells were seeded. Each assay was repeated three times.
Microarray analysis
The gene profiles of PC-3 M 2B4 and PC-3 M 1E8 cells were compared using the Human Glucose Metabolism RT2 Profiler™ PCR Array (PAHS-006Z, SuperArray, Frederick MD, USA). RNA isolation and purification were conducted according to the manufacturer’s protocol (Qiagen, Hilden, Germany). Qualified RNA was converted to cDNA using the RT2 First Strand Kit (Invitrogen, Carlsbad, CA, USA). The cDNA template was then added to an instrument-specific ready-to-use RT2 SYBR Green qPCR Master Mix (Invitrogen), and the array detection was performed using an ABI PRISM7900 instrument (Applied Biosystems, Foster City, CA, USA). All Ct values were corrected by the average Ct of a housekeeping gene (ACTB). KangChen Bio-tech Company (Shanghai, China) assisted the analysis.
Quantitative real-time polymerase chain reaction (qRT-PCR) analysis
Expression of the array-identified genes was verified in all five PCa cells by qRT-PCR. RNA was isolated using RNAiso Reagent (Takara, Dalian, China) according to the manufacturer’s instructions. RNA sample (1 μg) was used to synthesize the cDNA with a PrimeScript RT reagent kit (Takara). The qRT-PCR reaction mixture was prepared using an SYBR Premix Ex Taq kit (Takara) and detected on an ABI 7500 Fast Real-Time PCR system (Applied Biosystems). The ACTB gene was used as the internal reference. Relative quantification was calculated using the 2
−ΔΔCt method [
18]. Gene expression results were determined as the mean of three independent tests. The information on primers is shown in Additional file
1: Table S1.
Western blot analysis
Total proteins were extracted from the cultured cells using a Whole Cell Lysis Assay Kit (Keygen, Jiangsu, China) according to the operating instructions. Protein sample (30 μg) was separated by 15% sodium dodecyl sulphate-polyacrylamide gel (Bio-Rad, Gladesville, NSW, Australia) and transferred to polyvinylidene difluoride membranes (Millipore, Billerica, MA, USA). The membranes were then blocked with 5% non-fat milk for 2 h and incubated with primary antibodies overnight at 4 °C. After incubation with HRP-labeled secondary antibodies for 1 h at room temperature, the protein bands were detected using a chemiluminescence HRP kit (Millipore). The ACTB protein was measured as the internal reference for normalization. Experiments were repeated three times. Detailed information on the antibodies is presented in Additional file
2: Table S2.
Patients and samples
A total of 118 cancer patients were enrolled from Nanfang Hospital, Southern Medical University (Guangzhou, China) in two cohorts from December 2016 to August 2017. Cohort 1 included 64 patients pathologically diagnosed with common cancers (12 hepatocellular carcinoma, 14 lung cancer, 10 nasopharyngeal carcinoma, 9 esophageal cancer, 6 PCa, and 13 cervical cancer). Cohort 2 included 54 patients pathologically diagnosed with PCa, of which 29 had metastatic disease and 25 had no metastatic lesions. Patients were eligible if they were over 18 years of age and had no other concurrent tumors; otherwise they were excluded. Blood samples were collected for CTCs analysis from both cohorts before any treatment was provided. Other pathological characteristics were simultaneously determined, including tumor features (Gleason score, stage, and metastasis) and disease-related serum biomarkers such as total prostate-specific antigen (tPSA), alkaline phosphatase (ALP), and Hemoglobin B (Hb).
Ethics approval
The study methodologies conformed to the standards set by the Declaration of Helsinki. The research protocol was approved by the Ethics Committee of Nanfang Hospital (Approval No. 2016–172), and all patients provided informed consent.
CTCs enrichment and identification
CTCs isolation and identification were performed on the CanPatrol platform (SurExam, Guangzhou, China) using size-based microfiltration and fluorescence staining, as previously described [
19]. Briefly, 5 mL of blood samples were collected and pre-treated with ammonium chloride-based lysis buffer to remove erythrocytes. The karyocytes were then filtrated with a membrane (Millipore) with calibrated pores (diameter 8 μm). Cell nuclei of the retained cells were stained with 4′,6-diamidino-2-phenylindole (DAPI; Sigma, St. Louis, USA) for microscopic scanning (Zeiss, Germany). The leukocytes were excluded using the Alexa Fluor 740-labeled probes targeting CD45. Large cells with an oval or heteromorphic nucleus and specific chromatin but that had no expression of CD45 were identified as CTCs. The cell-counting results were judged as detectable CTCs (≥ 1/5 mL) and positive CTCs (≥ 3/5 mL). The method for determining the positive standard of cell counting is described in Additional file
3: Determination of the positive standard for CTCs counting.
The multiple RNA in situ hybridization (multi-RNA-ISH) technique was used to detect the molecular characteristics of the enriched CTCs. Five single channels of the automated imaging microscope were used to detect the fluorescent signals of markers that reflect different characteristics of CTCs. Except for the above-mentioned channels for DAPI (F
1) and CD45 (F
2), the F
3 channel was used to detect single or combined glucose metabolic (GM) genes. The capture probes for the previously identified genes were labeled by Alexa Fluor 647 and the probe sequences are shown in Additional file
4: Table S3. The metabolic CTCs phenotype was determined as GM
+CTCs or GM
−CTCs according to the selected GM markers. The Cy3-labeled epithelial markers (EpCAM and CK8/18/19) and Alexa Fluor 488-labeled mesenchymal markers (Vimentin and Twist) were detected in the F
4 and F
5 channel. Sequences of the capture probes for these EMT markers were described previously [
19]. The EMT phenotype of CTCs was determined as epithelial- (E-), hybrid- (H-) or mesenchymal- (M-) type.
Statistical analysis
Statistical analyses were performed using SPSS 13.0 (SPSS Inc., Chicago, IL, USA). Data were presented as mean ± SD for continuous variables (cell line assays). Differences between groups were compared using Student’s t-test or one-way ANOVA. Discontinuous variables (CTCs parameters) were presented as median with the interquartile range (IQR). Mann-Whitney U or Kruskal-Wallis H tests were used to compare differences. Categorical variables (clinical data) were expressed as numbers and percentages. The clinical relevance of the CTCs parameters was evaluated using Chi-square and Spearman’s rank correlation test. All tests were two-tailed, and a P value < 0.05 was considered statistically significant.
Discussion
In the present study, we identified the specific metastasis-related metabolic genes and determined PGK1 and G6PD as combined GM markers for the metabolic profiling of CTCs. The GM markers that reflected the metabolic reprogramming could serve as the CTCs biofunctional activity indicators. We also evaluated the clinical significance of the metabolic characterization of CTCs and conducted comparison analysis between the metabolic CTCs phenotypes and the typical CTCs classification using EMT markers. Although the GM+CTCs had a correlation with the EMT phenotypes of CTCs, they presented higher AUCs than EMT-CTCs in the discrimination of metastatic patients, demonstrating their promise as indicators of PCa metastasis.
Mitochondrial oxidative phosphorylation is the main energy source for normal cells. Tumor cells, however, depend strongly on the enhanced glycolysis (even in the presence of sufficient oxygen), the pentose phosphate pathway, and glutaminolysis, resulting from the damaged mitochondrial function [
11]. This reprogramming of the glucose metabolism plays a vital role in cancer metastasis because it allows tumor cells to escape normal growth processes. The Human Glucose Metabolism Array used in this study is a functional microarray specifically targeting the enzymes and regulators of the glucose and glycogen metabolism. It profiles 84 genes involved in the crucial processes of glycolysis, gluconeogenesis, the tricarboxylic acid cycle, the pentose phosphate pathway, and glycogen synthesis and degradation. The array identified eight metabolic genes differentially expressed among metastatic PCa cell lines: HK2, PDP2, G6PD, PGK1, PHKA1, PYGL, PDK1, and PKM2. The mRNA expressions of these metabolic genes increased significantly in the high metastatic PCa cells (1E8 and DU145) in comparison with the low metastatic cells (2B4, LNCAP, and PC-3). These genes play vital roles in the process of glucose and glycogen metabolism, and studies have demonstrated a remarkable association between their functions and the metastatic capacity of tumor cells [
28‐
33], which was in line with our data. Previous research has found that overexpression of HK2 and PKM2 governs the glucose influx and sustain high levels of glycolysis to promote cancer cell migration and induce stem cell differentiation [
28,
29]. PDP2 is the activator of pyruvate dehydrogenase (PDH), and the function of PDH in the cancer-associated fibroblasts was demonstrated essential for the migration ability of the co-cultured cancer cells [
30]. PHKA1 and PYGL are regulators of glycogen metabolism which balances the glucose supply. Prakash and his colleges [
31] demonstrated that the increased expression of PHKA1 was associated with younger ages of gastrointestinal stromal tumor patients. Hypoxia-induced activation of PYGL resulted in the enhancement of glycogen degradation and further contributed to the in vivo xenograft growth of U87 and MCF-7 cells by optimizing glucose utilization [
32]. Kamarajugadda et al. [
33] reported that the up-regulated PDKs and LDH in glycolysis could modulate the estrogen-related receptor gamma (ERRγ) to prolong the survival of mammary epithelial cells that break away from the matrix. This process strengthened the anoikis resistance to promote cell invasion and migration. These studies suggested the array-selected genes could be potential metabolic markers related to cancer metastasis.
Notably, in the verification assays, we observed inconsistent results between the protein and mRNA levels of HK2, PDP2, G6PD, PYGL, and PKM2 in the comparison of the 2B4 cells and 1E8 cells. Earlier studies reported the inconsistency between the protein and mRNA levels of gene expression in human tissue and cells [
34,
35], which did not conform to the Central Dogma of biological genetics. The reasons for this phenomenon might be biological and technological. Biologically, the complicated post-transcriptional and translational regulations, such as alternative splicing [
36], non-coding RNAs (miRNA and siRNA) interference [
37,
38], and translational recoding [
39] could change the protein level of the transcribed mRNA. Technologically, the characteristics of current methods might decrease the accuracy of mRNA or protein quantification. For instance, if an important splice variation or translational modification occurs in gene expression, the use of monoclonal antibodies directed to the wild-type could result in different protein information from mRNA expression [
40,
41]. Further systematic experiments are needed to explore the concrete mechanism of the specific genes identified in the present study.
The AHP is one of the most widely used methods of multi-criteria decision analysis [
24,
25]. It structures complex decision problems into a hierarchical system to conduct a rational and practical decision-making [
42]. We found that the mRNA expression of the metabolic genes varied among individual CTCs of blood samples, which revealed that the changes in tumor cell lines might not be exactly the same as those in the peripheral CTCs because of the difference in sample types and detecting techniques. Thus, we used the AHP-based multi-criteria weighted model by which PGK1 and G6PD were determined as the optimal markers for CTCs metabolic analysis. They function as key enzymes in glycolysis and the pentose phosphate pathway, respectively. PGK1 catalyzes 1,3-diphosphoglyceric acid to generate 3-phosphoglyceric acid and ATP. G6PD catalyzes the dehydrogenation of glucose 6-phosphate to generate ribulose-5-phosphate and further produce ribose 5-phosphate and NADPH. Experimental and clinical studies have demonstrated the pivotal role of PGK1 and G6PD in cell malignance and cancer metastasis [
43‐
50]. A mouse model study of metastatic gastric cancer demonstrated that the overexpression of PGK1 significantly increased the invasive and metastatic behavior of the implanted gastric tumors [
43]. Apart from the PGK1-induced enhancement of energy supply, the mechanism might involve the stimulation of PGK1-activated oncogenic AKT/mTOR pathway in tumor cells [
44]. Studies by Ahmad et al. [
45] and Xie et al. [
46] revealed that the PGK1 expression was up-regulated in the tissue of metastatic colon cancer and was closely correlated to the poor prognosis of hepatocellular carcinoma patients. Kowalik et al. [
47] used a rat model to identify the most aggressive lesions at early phases of hepatic carcinogenesis and found that the increased G6PD expression induced aggressive preneoplastic hepatocytes. G6PD could activate the STAT3 pathway to induce the EMT process and further promote the migration and invasion of hepatocellular carcinoma cells [
48]. The clinical analysis also revealed that the G6PD level is associated with high risk of recurrence and poor survival in patients with gastric [
49] and renal [
50] cancers. Therefore, increased activity of glycolysis and the pentose phosphate pathway (indicated by the up-regulation of PGK1 and G6PD) might reflect the active status of glucose metabolism in CTCs, which further reveals CTCs aggressiveness.
This theory was confirmed by the metabolic analysis of peripheral CTCs in PCa patients. The results demonstrated that the GM
+CTCs marked by PGK1 and G6PD were closely associated with the patients’ clinical stage, cancer metastasis and serum tPSA level, although they made up only a small part of the total CTCs. A high baseline of total CTCs has proved relevant to the metastatic tendency and decreased survival of PCa [
51,
52], whereas the reports on the associations between total CTCs and the clinical stage or serum tPSA are not exactly the same. For example, Resel et al. [
53] demonstrated a positive correlation between the CTCs count and disease stage or tPSA level. On the contrary, research by Oscar et al. [
51] and Tsumura et al. [
54] suggested that the clinical stage and PSA level at diagnosis had no correlation with the baseline number of CTCs. The different detection methods used in these studies and the heterogeneity of CTCs population might explain the discrepancy. Here, we found that the positive rate and number of GM
+CTCs were remarkably correlated with tumor stage and tPSA level. GM
+CTCs ≥3/5 mL was indicative of the advanced tumor stage and increased tPSA concentration, which were both significant prognostic biomarkers for PCa patients. These data suggest that the GM
+CTCs, as a highly aggressive subpopulation of the total CTCs, might be a more precise marker for tumor malignance and progression.
The EMT phenotype of CTCs is currently the most common indicator of cancer metastasis and prognosis [
55,
56]. Previous reports proved the vital role of metabolic genes in EMT by in vivo and in vitro experiments. Wu et al. [
57] investigated the effect of G6PD knockdown on A549, MDCK cells, and zebrafish embryos. They observed morphological changes and suppression of epithelial markers such as E-cadherin. A subcellular localization study found that the nuclear translocation of PKM2 regulated the EGF-induced EMT by promoting the transactivation of β-catenin [
58]. In return, EMT can reprogram the cancer metabolic profile by activating transcription factors. Twist is a well-known transcription factor of EMT. A Twist-overexpressing assay in breast cancer cells demonstrated the up-regulation of metabolic enzymes such as LDHA, PKM2, HK2, and G6PD, which was mediated by the PI3K/AKT and p53 signaling pathways [
59]. Therefore, the metabolic reprogramming and EMT, which are both crucial for the biological behavior of tumor cells, can mutually regulate each other and synergistically promote cancer metastasis. Here, our work found that the H-CTCs and M-CTCs presented higher GM
+CTCs proportions and GM
+CTCs correlation coefficients than those in the E-CTCs. This result also verified the significant interrelation between the metabolic and EMT phenotypes of CTCs.
Recent studies on gastric [
10], colorectal [
26] and breast [
27] cancers demonstrated that M-CTCs rather than E-CTCs were more relevant to tumor progression and metastasis. Nonetheless, inconsistent with these reports, our results revealed that the H-CTCs but not E-CTCs or M-CTCs were significantly correlated with the metastasis of PCa. This case has two possible explanations. On the one hand, EMT is dynamic, and the reversible EMT-MET occurred during the dissemination and metastasis of tumor cells [
60]. This phenomenon might result in the diversity of cell biological activities and functions even though they present the same morphological EMT phenotype. On the other hand, the importance of epithelial plasticity to metastasis has recently attracted considerable attention. A study by Ruscetti et al. [
61] found that although mesenchymal and hybrid CTCs increased tumorigenesis in transgenic mouse models, only hybrid and epithelial CTCs formed macro-metastases, whereas mesenchymal CTCs persisted as micro-metastatic foci. In fact, hybrid CTCs may represent the most plastic tumor cells because they have both the epithelial and mesenchymal plasticity [
62]. Therefore, the hybrid CTCs might have the greatest potential to contribute to cancer metastasis. Our next investigations showed that the GM
+CTCs number was closely correlated with the number of H-CTCs (
r = 0.807), whereas correlations between GM
+CTCs and E-CTCs (
r = 0.369) or M-CTCs (
r = 0.553) were moderate. These data also supported the highlighted significance of hybrid CTCs in cancer metastasis. Besides, the metabolic analysis of CTCs provides a way to help settle the controversial outcomes of EMT subtypes. The aggressiveness of malignant cells might depend more on the metabolic activity as a functional indicator, compared with the morphological EMT features. We used the simulated ROC to assess the performance of EMT-CTCs and GM
+CTCs in the discrimination of metastatic PCa patients. The AUCs of GM
+CTCs were higher than that of H-CTCs, no matter as a single marker or as combined markers with tPSA and the Gleason score. These results demonstrate that the hypermetabolic CTCs could be a more accurate marker than EMT-CTCs for revealing the metastasis and disease progression of cancer patients.
In addition, the multi-RNA-ISH technique used in this method not only offers sensitive and visualized detection of the metabolic markers, but could also facilitate the integrated CTCs characterization with the combination of EMT phenotypes. Simultaneously profiling the metabolic and EMT subtypes would improve the molecular analysis of CTCs from the functional and morphological aspects, and further contribute to the understanding of tumor transfer mechanism and better application of CTCs tests. Nevertheless, some additional experiments are necessary to perfect this study. In the aspect of the fundamental theory, the biological functions of the PGK1/G6PD-positive CTCs and the related mechanisms remain to be illustrated using in vitro culture assays. In terms of research methodology, new techniques for the protein detection of the metabolic markers in CTCs need to be developed and tested for clinical applications. Moreover, an expanded sample size of PCa and other carcinomas would also be helpful to confirm the clinical significance of this method. Follow-up investigations on the prognostic and monitoring roles of specific metabolic CTCs phenotype will be conducted in the future.