Background: In the present study, we assessed the clinical value of circulating tumor cells (CTC) with stem-like phenotypes for diagnosis, prognosis, and surveillance in hepatitis B virus (HBV)–related hepatocellular carcinoma (HCC) by an optimized qPCR-based detection platform.

Methods: Differing subsets of CTCs were investigated, and a multimarker diagnostic CTC panel was constructed in a multicenter patient study with independent validation (total n = 1,006), including healthy individuals and patients with chronic hepatitis B infection (CHB), liver cirrhosis (LC), benign hepatic lesion (BHL), and HBV-related HCC, with area under the receiver operating characteristic curve (AUC-ROC) reflecting diagnostic accuracy. The role of the CTC panel in treatment response surveillance and its prognostic significance were further investigated.

Results: The AUC of the CTC panel was 0.88 in the training set [sensitivity = 72.5%, specificity = 95.0%, positive predictive value (PPV) = 92.4, negative predictive value (NPV) = 77.8] and 0.93 in the validation set (sensitivity = 82.1%, specificity = 94.2%, PPV = 89.9, NPV = 89.3). This panel performed equally well in detecting early-stage and α-fetoprotein–negative HCC, as well as differentiating HCC from CHB, LC, and BHL. The CTC load was decreased significantly after tumor resection, and patients with persistently high CTC load showed a propensity of tumor recurrence after surgery. The prognostic significance of the CTC panel in predicting tumor recurrence was further confirmed [training: HR = 2.692; 95% confidence interval (CI), 1.617–4.483; P < 0.001; and validation: HR = 3.127; 95% CI, 1.360–7.190; P = 0.007].

Conclusions: Our CTC panel showed high sensitivity and specificity in HCC diagnosis and could be a real-time parameter for risk prediction and treatment monitoring, enabling early decision-making to tailor effective antitumor strategies. Clin Cancer Res; 24(9); 2203–13. ©2018 AACR.

Translational Relevance

Whether detection of circulating tumor cells (CTC) with stem-like phenotypes could be a reliable and effective method that integrates in hepatocellular carcinoma (HCC) early diagnosis, outcome prediction as well as treatment response evaluation was unknown. In this study, we constructed a CTC panel including four putative stem cell biomarkers (epithelial cell adhesion molecule, CD133, CD90, and cytokeratin 19). In our clinical trial including 1,006 individuals, the area under the curve of this panel is 0.88 [sensitivity = 72.5%, specificity = 95.0%, positive predictive value (PPV) = 92.4, negative predictive value (NPV) = 77.8] in the training set and 0.93 (sensitivity = 82.1%, specificity = 94.2%, PPV = 89.9, NPV = 89.3) in the validation set. HCC patients with high CTC load show higher recurrence rate than patients with low load (49.1% vs. 24.7%). Our CTC panel showed high sensitivity and specificity in the diagnosis of HCC and could be a real-time parameter for risk prediction and treatment monitoring in treatment response surveillance.

Hepatocellular carcinoma (HCC) is the most prevalent malignancy in the world and ranks the second as the cause of cancer deaths (1). For the absence of an effective method for timely diagnosis, only 30% to 40% of HCC patients qualify for potentially curative treatments at the time of diagnosis (2). Meanwhile, even after curative treatment, approximate 60% to 70% of patients experience recurrence or distant metastasis within 5 years (3). Although serologic tumor markers, clinicopathologic parameters, and radiologic modalities are commonly used in routine clinical practice for management of HCC patients (4), none of these approaches can provide comprehensive and precise information covering diagnosis, outcome prediction, and the evaluation of therapeutic response in HCC (5). The lack of a reliable and versatile method that integrates early diagnosis, precise prediction, and real-time surveillance becomes the main obstacle for further improving the clinical outcome of HCC patients.

Ready and noninvasive access to circulating tumor cells (CTC) is advantageous, constituting a potential surrogate for tumor biopsy with new diagnostic, prognostic, and therapeutic import (6–8). However, evidence indicates that among thousands of cells freeding into circulation from primary tumors, only small populations with stem cell-like properties have the generative capacity for driving tumor progression, metastasis, and resistance to traditional therapies (9, 10). Thus, identifying the subpopulations of CTC with stem cell phenotypes might be more clinically relevant than the analyses on total CTC counts. Recently, we identified the stem cell features of epithelial cell adhesion molecule (EpCAM) CTCs and their clinical significance in HCC (11). Others have also characterized other subpopulations of CTC with stem-like cell features in HCC, based on various surface molecules, such as CD44 (12), CD90 (12), and intercellular adhesion molecule 1 (ICAM1; ref. 13). These observations imply that CTCs with stem-like cell features in HCC are phenotypically diverse, which is in agreement with our previous study that acknowledged not only the heterogeneity of cancer stem cell (CSC) markers expression in HCC, but also the array of CSC subsets that exist and that may vary from patient to patient (14). Therefore, targeting of CTCs with stem-like phenotypes through multimarker strategies might be optional choice for CTC detection in HCC patients.

Our previous study constructed a novel optimized platform for CTC detection in HCC patients, based on quantitative real-time PCR (qRT-PCR) analysis and negative enrichment strategy (15). Based on this CTC detection platform, we further systematically screened the expression patterns of nine putative CSC biomarkers [EpCAM (11), CD90 (12, 16), CD24 (17), ATP-binding cassette subfamily G member 2 (ABCG2; ref. 14), CD44 (12), Nestin (14), CD133 (18), cytokeratin19 (CK19; ref. 19), and ICAM1 (13)] in CTCs of HCC patients, and a multimarker diagnostic panel, targeting CTC subpopulation with stem-like phenotypes, was thereby constructed through a multicenter patient study with independent validation, via a large group of 1,006 subjects including healthy individuals and patients with chronic hepatitis B infection (CHB), liver cirrhosis (LC), benign hepatic lesion (BHL), and hepatitis B virus (HBV)–related HCC. Moreover, the treatment response surveillance and the prognostic significance of our CTC panel were further investigated in the same HCC group. We found that the CTC detection panel showed considerable clinical benefit in HCC early diagnosis, outcome prediction as well as treatment response evaluation.

Study design and patient selection

We recruited consecutive patients with HCC, CHB, LC, BHL, and healthy donor (HD) from four clinical institutions in Shanghai, China (Zhongshan Hospital, Fudan University; Shanghai Public Health Clinical Center; Shanghai Cancer Center, Fudan University; and Longhua Hospital, Shanghai University of Traditional Chinese Medicine), between December 2012 and June 2015. Blood samples were analyzed in three chronologic phases (Fig. 1). First, 100 samples (HCC group, n = 50; HD group, n = 50) were screened for 9 putative CSC biomarkers (EpCAM, CD133, CD90, CK19, ABCG2, CD44, ICAM1, CD24, and Nestin) by qRT-PCR. Next, a multimarker diagnostic panel designed to differentiate HCC and other control groups of training group (n = 401) was subjected to logistic regression, and the diagnostic panel performance was validated in another independent group (n = 505). For the validation group, additional BHL group (62 hepatic hemangioma, 26 hepatic cyst, 10 focal nodular hyperplasia, and 2 hepatocellular adenoma) was included for further exporting the differential potential of the CTC panel. Finally, the clinical significance of the CTC panel for treatment response surveillance was further explored in 60 resectable HCC patients with CTC value after resection, and the prognostic significance of the CTC panel was investigated in 195 resectable HCC patients of the training set and then validated in 130 patients of the validation set.

Figure 1.

Flowchart of study design.

Figure 1.

Flowchart of study design.

Close modal

For those patients who received transcatheter arterial chemoembolization, the HCC was defined according to American Association for the Study of Liver Diseases guidelines (20). And serum α-fetoprotein (AFP), a serum marker routinely used for HCC surveillance in China, was assessed for adjunctive diagnosis (21). For resectable patients, the diagnosis of HCC was confirmed by the histologic examination of the resected primary tumor based on World Health Organization criteria (22). For staging, Barcelona Clinic Liver Cancer (BCLC) criteria were applied (23, 24). Diagnosis of CHB required >6 months of hepatitis B surface antigen (HBsAg), HBsAg positivity, HBV DNA level >103 copies/mL, and elevated serum concentrations of alanine aminotransferase, according to the guidelines for prevention and treatment of chronic HBV infection (25). Diagnosis of cirrhosis stemmed from liver biopsy and clinical imaging evidence. Diagnosis of benign hepatic lesion was based on the standard clinical, biochemical, and clinical imaging evidences, as well as the pathologic data. HD subjects were eligible blood donors with normal biochemistries such as HBsAg and no liver disease (historically), viral hepatitis, or malignancies. Patients with histories of other solid tumors were excluded. The study was approved by the ethics review committee of each institution. Informed consent was obtained from participants in accordance with respective committee regulations.

Follow-up for tumor recurrence

Patients treated with resectable operation were followed up every 2 months during the first postoperative year, and every 6 months afterward. Patients were monitored by serum AFP, abdomen ultrasonography, and chest X-ray every 6 months, according to the postoperative time. For patients with test suspected with recurrence, CT and/or MRI were used to verify whether intrahepatic recurrence and/or distal metastasis had occurred. The diagnosis of recurrence was based on typical imaging appearance in CT and/or MRI scan and an elevated AFP level. Follow-up began form January 2013 and finished on June 2015. The median follow-up time was 25.2 months in the training set and 21.2 months in the validation set. All 325 resectable HCC patients (n = 195 in the training cohort and n = 130 in the validation cohort) have complete follow-up information.

Blood sampling and quantitative real-time PCR

Peripheral blood samples (5 mL each) were collected separately in different centers before initial diagnosis or 1 week after tumor resection and immediately transported to laboratory areas. The diagnosis and clinical information, including disease status, of each sample kept masked in the whole process of measurement until the results enter into statistical process. A qRT-PCR test platform was utilized, with negative enrichment to optimize CTC detection (15). In brief, samples first were negatively enriched, using RosetteSep Human CD45 Depletion Cocktail (STEMCELL Technologies) to minimize background expression by removing leukocyte contamination. Thereafter, RNA extraction and cDNA synthesis were performed according to the manufacturer's instructions (RNeasy Mini Kit & QuantiTect Reverse Transcription Kit, Qiagen), and mRNA expression levels of 10 target genes (β-actin used as an internal control) were assayed by TaqMan-based real-time PCR, using a LightCycler 480 instrument (Roche Diagnostics). All qRT-PCR assay results entailed relative quantification via the 2−ΔΔCq algorithm, where fold expression relative to a calibrator (i.e., average ΔCq of HD subjects in each group) was given (15, 26). All testing was done in triplicate. Primer and probe utilization are itemized in Supplementary Table S1. To calculate relative expression levels of each marker, a 2−ΔΔCq algorithm was applied as follows (15, 26), using average ΔCq of HD subjects as a calibrator, ΔCq = Cq(target) – Cq(β-actin), with ΔΔCq = ΔCq(target) – ΔCq(calibrator).

Statistical analysis

Statistical analysis relied on standard softwares (MedCalc v10.4.7.0, MedCalc Software; SPSS v19.0, SPSS Inc.; and PASS v11.0.7, NCSS Inc.). The Mann–Whitney U test (continuous variables and nonparametric analysis) was used to test differences between groups. Backward logistic regression model was used to construct the diagnostic CTC panel with four CSC biomarkers based on the training dataset, and variable with a P value less than 0.05 was used for further model construction. Afterward, predicted probabilities of individual study enrollees were calculated as continuous variances and functioned as surrogate markers for constructing receiver operating characteristic (ROC) curves, where area under the ROC curve (AUC) reflects diagnostic performance. The cutoff value of our CTC panel used in prognosis was estimated using X-tile 3.6.1 software (Yale University, New Haven, CT; ref. 27). Time to recurrence (TTR) was defined as interval between the surgery and recurrence. The cumulative recurrence and survival rates were calculated using the Kaplan–Meier method and analyzed by the log-rank test. Univariate and multivariate analyses were calculated by the Cox proportional hazards regression model. All P values were two-sided, with P < 0.05 considered statistically significant.

Based on the results obtained from the first 100 subjects for 9 CSC markers' screening, a required minimum sample size of 127 patients and 77 controls was calculated by PASS with sensitivity of 0.70, specificity of 0.85, and permissible error of 0.08.

Patient characteristics

The clinical characteristics of patients recruited in this study were summarized in Table 1. Groups were well-matched for serum AFP level, age, and gender (all P > 0.05, Table 1). The validation group had a significantly larger percentage of patients with multiple tumors (17.95% vs. 10.50%, P = 0.03), larger sized tumors (52.82% vs. 40.00%, P = 0.02), and advanced BCLC staging (51.79% vs. 34.50%, P < 0.01, Table 1). All HBV-related characteristics were well-balanced (all P > 0.05, Table 1).

Table 1.

Clinicopathologic characteristics of patients with HCC and CHB/LC groups in training and validation sets

Training setValidation set
N (%)N (%)P
HCC group 
AFP (ng/mL) ≤20 85 (42.50) 86 (44.10) 0.75 
 >20 115 (57.50) 109 (55.90)  
Age (y) ≤50 83 (41.50) 74 (37.95) 0.47 
 >50 117 (58.50) 121 (62.05)  
Gender Female 47 (23.50) 34 (17.44) 0.14 
 Male 153 (76.50) 161 (82.56)  
HBsAg Negative 31 (15.50) 32 (16.41) 0.81 
 Positive 169 (84.50) 163 (83.59)  
Cirrhosis No 44 (22.00) 40 (20.51) 0.72 
 Yes 156 (78.00) 155 (79.49)  
ALT (U/L) ≤40 138 (69.00) 120 (61.54) 0.14 
 >40 62 (31.00) 75 (38.46)  
Tumor number Single 179 (89.50) 160 (82.05) 0.03 
 Multiple 21 (10.50) 35 (17.95)  
Tumor size (cm) ≤5 120 (60.00) 92 (47.18) 0.02 
 >5 80 (40.00) 103 (52.82)  
BCLC stage 0+A 131 (65.50) 94 (48.21) 0.007 
 B+C 69 (34.50) 101 (51.79)  
CHB/LC group 
AFP (ng/mL) ≤20 81 (80.20) 77 (77.00) 0.58 
 >20 20 (19.80) 23 (23.00)  
HBV-DNA (IU/mL) ≤1,000 41 (40.59) 32 (32.00) 0.21 
 >1,000 60 (59.41) 68 (68.00)  
Age (y) ≤50 73 (72.28) 75 (75.00) 0.66 
 >50 28 (27.72) 25 (25.00)  
Gender Female 28 (27.72) 23 (23.00) 0.44 
 Male 73 (72.28) 77 (77.00)  
HBsAg Negative 0 (0.00) 0 (0.00) N.A. 
 Positive 101 (100.00) 100 (100.00)  
Training setValidation set
N (%)N (%)P
HCC group 
AFP (ng/mL) ≤20 85 (42.50) 86 (44.10) 0.75 
 >20 115 (57.50) 109 (55.90)  
Age (y) ≤50 83 (41.50) 74 (37.95) 0.47 
 >50 117 (58.50) 121 (62.05)  
Gender Female 47 (23.50) 34 (17.44) 0.14 
 Male 153 (76.50) 161 (82.56)  
HBsAg Negative 31 (15.50) 32 (16.41) 0.81 
 Positive 169 (84.50) 163 (83.59)  
Cirrhosis No 44 (22.00) 40 (20.51) 0.72 
 Yes 156 (78.00) 155 (79.49)  
ALT (U/L) ≤40 138 (69.00) 120 (61.54) 0.14 
 >40 62 (31.00) 75 (38.46)  
Tumor number Single 179 (89.50) 160 (82.05) 0.03 
 Multiple 21 (10.50) 35 (17.95)  
Tumor size (cm) ≤5 120 (60.00) 92 (47.18) 0.02 
 >5 80 (40.00) 103 (52.82)  
BCLC stage 0+A 131 (65.50) 94 (48.21) 0.007 
 B+C 69 (34.50) 101 (51.79)  
CHB/LC group 
AFP (ng/mL) ≤20 81 (80.20) 77 (77.00) 0.58 
 >20 20 (19.80) 23 (23.00)  
HBV-DNA (IU/mL) ≤1,000 41 (40.59) 32 (32.00) 0.21 
 >1,000 60 (59.41) 68 (68.00)  
Age (y) ≤50 73 (72.28) 75 (75.00) 0.66 
 >50 28 (27.72) 25 (25.00)  
Gender Female 28 (27.72) 23 (23.00) 0.44 
 Male 73 (72.28) 77 (77.00)  
HBsAg Negative 0 (0.00) 0 (0.00) N.A. 
 Positive 101 (100.00) 100 (100.00)  

Abbreviations: ALT, alanine aminotransferase; N.A., not applicable.

Differential expression of CSC biomarkers in CTC for HCC patients

Nine candidate genes as putative CSC biomarkers were assessed by qRT-PCR in an independent and divided screening group of 100 subjects (HCC, 50; HD, 50). Levels of EpCAM, CD133, CD90, and CK19 expression were significantly higher in the HCC (vs. HD) group (Supplementary Fig. S1A–S1D), whereas ABCG2 (P = 0.84), CD24 (P = 0.19), CD44 (P = 0.27), and ICAM1 (P = 0.11) expression levels did not differ significantly, and Nestin was uniformly undetectable (Supplementary Fig. S1E–S1I). Thus, EpCAM, CD90, CD133, and CK19 were further evaluated.

To confirm the detection accuracy of the qRT-PCR method, the existence of CTC subpopulations with stem-like phenotypes was further validated by the immunofluorescent method in 10 HCC patients, which were positive for 4 CSC biomarkers. We found that the corresponding subpopulations of CTC were identified and well matched with the qRT-PCR detection results in HCC patients (Supplementary Fig. S2A). Furthermore, the single-cell transcriptional analysis also confirmed their tumor origin, such as high expression level of tumor-specific gene (AFP), high expression stem cell markers (NANOG, ABCG2, and SOX2), and low expression of hematopoietic markers (CD45, CD16, and CD34) distinguished with leukocytes (Supplementary Fig. S2B).

Formulating CTC detection panel with CSC multimarkers in the training set

Expression levels of EpCAM, CD90, CD133, and CK19 were determined in a training group of 401 subjects (HCC, 200; CHB/LC, 101; HD, 100) via qRT-PCR. Relative expression levels for all four markers were significantly higher in the HCC group, compared with CHB/LC and HD groups (all P < 0.05, Supplementary Fig. S3). Using 2−ΔΔCq = 2.0 as a cutpoint (15, 26), positivity rates in the HCC group for EpCAM (43.50%), CD133 (33.50%), CD90 (34.00%), and CK19 (29.00%) were significantly higher than those in CHB/LC and HD groups (all P < 0.05, Supplementary Fig. S3). Diagnostic accuracies in using these four biomarkers were as follows: EpCAM, AUC = 0.70 [95% confidence interval (CI), 0.65–0.74]; CD133, AUC = 0.65 (95% CI, 0.60–0.70); CD90, AUC = 0.64 (95% CI, 0.59–0.69); and CK19, AUC = 0.64 (95% CI, 0.59–0.69). Multivariate P values for all of four biomarkers were <0.05 by logistic regression (Supplementary Table S2).

Consequently, a backward logistic regression model was applied to estimate the risk of malignant (HCC) diagnosis for the training dataset, issuing the following new variable of predicted probability (P) for HCC on the basis of an equation derived from logistic regression (all HCC vs. control groups in the training set): logit (P = HCC) = −2.15 + 0.74 × EpCAM + 0.65 × CD133 + 0.34 × CD90 + 0.99 × CK19.

Performance of the CTC detection panel in the training set

In accordance with our logistic regression equation, a predicted probability was registered for each member of the training set. These values were significantly higher in patients with HCC (n = 200) versus CHB/LC (n = 101) or HD groups (n = 100; both P < 0.001, Fig. 2A). Further ROC analysis in patients with HCC versus CHB/LC and HD showed optimal diagnostic cutoff value of 0.57 (criteria for optimal sensitivity plus specificity, optimal Youden index) for the CTC panel and 0.88 for AUC, with 72.5% sensitivity and 95.0% specificity, compared with an AUC of 0.77 for AFP, with 57.0% sensitivity and 90.0% specificity (Fig. 3A; Table 2, cutoff value of AFP = 20 ng/mL). A greater proportion of patients with HCC were positive on the basis of the CTC panel, as opposed to AFP (73.0% vs. 57.5%, cutoff value for CTC load/5 mL = 0.57 and AFP = 20 ng/mL; Fig. 2B), and similar CTC panel positivity rates were observed in AFP-negative and AFP-positive patients with HCC (77.9% vs. 71.1%, Fig. 2B). Furthermore, patients with early-stage HCC showed a higher proportion of positive CTC panel results than for AFP (71.8% vs. 53.4%, Fig. 2C).

Figure 2.

Distributions of predicted probabilities and positivity rates for the CTC diagnostic panel in training and validation sets. A, Distributions for predicted probabilities of the CTC panel in training (left) and validation (right) groups. B, Positivity rates for the CTC panel, serum AFP, or both and for the CTC panel stratified by AFP status in all HCC patients of training (left) and validation (right) groups. C, Positivity rates for the CTC panel, serum AFP, or both and for the CTC panel stratified by AFP status in patients with early-stage HCC of training (left) and validation (right) groups. D, Positivity rates for the CTC panel and serum AFP and for the CTC panel with AFP positivity in patients with chronic HBV infection and/or cirrhosis of training (left) and validation (right) groups.

Figure 2.

Distributions of predicted probabilities and positivity rates for the CTC diagnostic panel in training and validation sets. A, Distributions for predicted probabilities of the CTC panel in training (left) and validation (right) groups. B, Positivity rates for the CTC panel, serum AFP, or both and for the CTC panel stratified by AFP status in all HCC patients of training (left) and validation (right) groups. C, Positivity rates for the CTC panel, serum AFP, or both and for the CTC panel stratified by AFP status in patients with early-stage HCC of training (left) and validation (right) groups. D, Positivity rates for the CTC panel and serum AFP and for the CTC panel with AFP positivity in patients with chronic HBV infection and/or cirrhosis of training (left) and validation (right) groups.

Close modal
Figure 3.

ROC analysis of the CTC panel performance in diagnosing HCC for training and validation sets. ROC curves for (A) the CTC diagnostic panel and serum AFP in all patients with HCC versus all controls of training (left) and validation (right) groups; (B) the CTC panel and serum AFP in patients with early-stage HCC versus all controls of training (left) and validation (right) groups; (C) the CTC panel and serum AFP in all patients with HCC versus patients at risk of HCC of training (left) and validation (right) groups; and (D) the CTC panel and serum AFP in patients with early-stage HCC versus patients at risk of HCC of training (left) and validation (right) groups.

Figure 3.

ROC analysis of the CTC panel performance in diagnosing HCC for training and validation sets. ROC curves for (A) the CTC diagnostic panel and serum AFP in all patients with HCC versus all controls of training (left) and validation (right) groups; (B) the CTC panel and serum AFP in patients with early-stage HCC versus all controls of training (left) and validation (right) groups; (C) the CTC panel and serum AFP in all patients with HCC versus patients at risk of HCC of training (left) and validation (right) groups; and (D) the CTC panel and serum AFP in patients with early-stage HCC versus patients at risk of HCC of training (left) and validation (right) groups.

Close modal
Table 2.

Performance of the CTC panel, serum AFP, or both in diagnosing HCCa

Training (n = 401)Validation (n = 505)
AUC (95% CI)Sen (%)Sp (%)PPV (%)NPV (%)PLRNLRAUC (95% CI)Sen (%)Sp (%)PPV (%)NPV (%)PLRNLR
HCC vs. CHB, LC, and HD HCC vs. CHB, LC, BHL, and HD 
CTC 0.88 (0.84–0.91) 72.5 95.0 92.4 77.8 14.6 0.3 0.93 (0.90–0.95) 82.1 94.2 89.9 89.3 14.1 0.2 
AFP 0.77 (0.73–0.81) 57.0 90.0 85.2 68 5.8 0.5 0.80 (0.77–0.84) 55.9 91.6 81.3 76.8 6.9 0.5 
CTC+AFP 0.89 (0.86–0.92) 76.0 95.0 92.7 79.8 14.0 0.3 0.94 (0.91–0.96) 83.1 94.0 90.1 90.0 14.1 0.2 
HCC vs. CHB and LC HCC vs. CHB, LC, and BHL 
CTC 0.87 (0.82–0.90) 72.5 92.0 93.6 62.8 9.2 0.3 0.93 (0.90–0.95) 82.1 93.0 92.0 84.2 11.7 0.2 
AFP 0.72 (0.67–0.77) 57.0 80.0 85.2 48.8 2.9 0.5 0.79 (0.75–0.83) 55.9 87.5 81.3 67.0 4.5 0.5 
CTC+AFP 0.88 (0.84–0.91) 74.5 93.0 93.8 65.5 10.8 0.3 0.93 (0.90–0.96) 83.1 93.0 92.0 85.0 11.1 0.2 
Early-stage HCC vs. CHB, LC, and HD Early-stage HCC vs. CHB, LC, BHL, and HD 
CTC 0.87 (0.83–0.90) 71.8 95.0 88.7 83.6 13.2 0.3 0.93 (0.90–0.96) 85.1 94.2 81.6 95.4 14.7 0.2 
AFP 0.74 (0.69–0.78) 53.4 90.0 77.8 74.8 5.4 0.5 0.75 (0.71–0.79) 46.8 91.6 62.9 85.0 5.6 0.6 
CTC+AFP 0.88 (0.85–0.92) 80.9 87.0 88.8 84.0 6.3 0.2 0.93 (0.90–0.96) 85.1 94.2 81.6 95.4 14.7 0.2 
Early-stage HCC vs. CHB and LC Early-stage HCC vs. CHB, LC, and BHL 
CTC 0.85 (0.80–0.90) 71.8 91.0 90.4 71.1 8.1 0.3 0.93 (0.90–0.96) 85.1 93.0 85.1 93.0 12.2 0.2 
AFP 0.68 (0.62–0.74) 53.4 80.0 77.8 57.0 2.7 0.6 0.74 (0.61–0.75) 46.8 87.0 62.9 77.7 3.6 0.6 
CTC+AFP 0.87 (0.82–0.91) 71.0 93.0 90.5 71.7 10.2 0.3 0.93 (0.90–0.96) 85.1 93.0 85.1 93.0 10.6 0.2 
Training (n = 401)Validation (n = 505)
AUC (95% CI)Sen (%)Sp (%)PPV (%)NPV (%)PLRNLRAUC (95% CI)Sen (%)Sp (%)PPV (%)NPV (%)PLRNLR
HCC vs. CHB, LC, and HD HCC vs. CHB, LC, BHL, and HD 
CTC 0.88 (0.84–0.91) 72.5 95.0 92.4 77.8 14.6 0.3 0.93 (0.90–0.95) 82.1 94.2 89.9 89.3 14.1 0.2 
AFP 0.77 (0.73–0.81) 57.0 90.0 85.2 68 5.8 0.5 0.80 (0.77–0.84) 55.9 91.6 81.3 76.8 6.9 0.5 
CTC+AFP 0.89 (0.86–0.92) 76.0 95.0 92.7 79.8 14.0 0.3 0.94 (0.91–0.96) 83.1 94.0 90.1 90.0 14.1 0.2 
HCC vs. CHB and LC HCC vs. CHB, LC, and BHL 
CTC 0.87 (0.82–0.90) 72.5 92.0 93.6 62.8 9.2 0.3 0.93 (0.90–0.95) 82.1 93.0 92.0 84.2 11.7 0.2 
AFP 0.72 (0.67–0.77) 57.0 80.0 85.2 48.8 2.9 0.5 0.79 (0.75–0.83) 55.9 87.5 81.3 67.0 4.5 0.5 
CTC+AFP 0.88 (0.84–0.91) 74.5 93.0 93.8 65.5 10.8 0.3 0.93 (0.90–0.96) 83.1 93.0 92.0 85.0 11.1 0.2 
Early-stage HCC vs. CHB, LC, and HD Early-stage HCC vs. CHB, LC, BHL, and HD 
CTC 0.87 (0.83–0.90) 71.8 95.0 88.7 83.6 13.2 0.3 0.93 (0.90–0.96) 85.1 94.2 81.6 95.4 14.7 0.2 
AFP 0.74 (0.69–0.78) 53.4 90.0 77.8 74.8 5.4 0.5 0.75 (0.71–0.79) 46.8 91.6 62.9 85.0 5.6 0.6 
CTC+AFP 0.88 (0.85–0.92) 80.9 87.0 88.8 84.0 6.3 0.2 0.93 (0.90–0.96) 85.1 94.2 81.6 95.4 14.7 0.2 
Early-stage HCC vs. CHB and LC Early-stage HCC vs. CHB, LC, and BHL 
CTC 0.85 (0.80–0.90) 71.8 91.0 90.4 71.1 8.1 0.3 0.93 (0.90–0.96) 85.1 93.0 85.1 93.0 12.2 0.2 
AFP 0.68 (0.62–0.74) 53.4 80.0 77.8 57.0 2.7 0.6 0.74 (0.61–0.75) 46.8 87.0 62.9 77.7 3.6 0.6 
CTC+AFP 0.87 (0.82–0.91) 71.0 93.0 90.5 71.7 10.2 0.3 0.93 (0.90–0.96) 85.1 93.0 85.1 93.0 10.6 0.2 

Abbreviations: NLR, negative likelihood ratio; NPV, negative predictive value; PLR, positive likelihood ratio; PPV, positive predictive value.

aDiagnostic cutpoints of CTC panel and serum AFP were 0.57 and 20 ng/mL, respectively.

The CTC panel outperformed AFP as a biomarker in terms of differential diagnostic capability, yielding higher AUC, sensitivity, and specificity in HCC (n = 200) and CHB/LC group (n = 101) comparisons (Fig. 3C; Table 2). In the CHB/LC group (n = 101), 19.8% had above-threshold serum AFP levels, compared with only 9.9% on CTC panel screening, and only 3 of 20 AFP-positive (15.0%) CHB/LC subjects were positive by CTC panel screening (Fig. 2D).

Compared with AFP, the CTC panel also showed higher potential in diagnosing early-stage HCC (BCLC 0+A; n = 131; Fig. 3B and D). At differing BCLC stages, consistently accurate panel results were seen: stage 0, AUC = 0.92 (95% CI, 0.88–0.95); stage A, AUC = 0.86 (95% CI, 0.82–0.90); stage B, AUC = 0.91 (95% CI, 0.87–0.94); and stage C, AUC = 0.86 (95% CI, 0.82–0.91; Supplementary Table S3).

In addition, we found that cancerous and noncancerous states in AFP-negative patients were distinguished satisfactorily via the CTC panel (AUC = 0.89, 95% CI, 0.85–0.92; sensitivity, 77.7%; specificity, 95.0%; Supplementary Table S4). More importantly, this level of performance was maintained in the early-stage HCC subgroup of AFP-negative patients (AUC = 0.89, 95% CI, 0.84–0.92; sensitivity, 75.4%; specificity, 95.0%; Supplementary Table S4).

Validating the diagnostic potential of the CTC detection panel

We found that relative expression levels for all four markers were still significantly lower in HD (n = 110), CHB/LC (n = 100), and BHL groups (n = 100) compared with HCC group (n = 195) in the validation set (P < 0.05, Supplementary Fig. S4). When comparing HCC (including early-stage subset) with other three control groups (HD, CHB/LC, and BHL), the similar differences in patterns of predicted probabilities were observed (Fig. 2A). Relative to three control groups, AUC of the CTC panel was 0.93 (95% CI, 0.90–0.95), with 82.1% sensitivity and 94.2% specificity for all HCC, and 0.93 (95% CI, 0.90–0.96) with 85.1% sensitivity and 94.2% specificity for early-stage HCC (n = 94), all surpassing the performance of serum AFP (Fig. 3A and B; Table 2). Using the 0.57 cutoff point for diagnosis, our CTC panel positivity rates remained high in the validation set (82.0%), including early-stage patients (85.1%), but were low in the CHB/LC group (8.0%) and BHL group (13.0%) regardless of AFP status (Fig. 2B and D; Supplementary Fig. S5). The panel diagnostic performance was outstanding in both negative and positive AFP subgroups (Supplementary Table S4). The capacity of the CTC panel to distinguish patients with HCC (even at early stage) from those with CHB/LC and BHL in the validation set was also confirmed (Fig. 3C and D). Meanwhile, the similar satisfactory diagnostic performances of the CTC panel were observed in different BCLC stages (Supplementary Table S3).

The clinical significance of our CTC panel in treatment response surveillance

The potential utility of the CTC panel in therapeutic response evaluation was further investigated in 60 resectable HCC patients. We found that the CTC load was decreased significantly after resection in 76.7% (46/60) of HCC patients with blood sample collection 1 month after surgery, and the positive rate decreased from 70.0% (42/60) to 31.7% (19/60; Fig. 4A). With a median follow-up time of 25.8 months, 46.7% (28/60) of the patients suffered intrahepatic recurrence. On the basis of changes between preoperative and postoperative CTC levels, we divided the 60 patients into three groups: I, persistent negative (n = 18) at both points; II, preoperatively positive and then postoperatively negative (n = 23); and III, persistent positive (n = 19). The recurrence rates for groups I to III were 11.1% (2/18), 47.8% (11/23), and 78.9% (15/19), respectively. Patients in group III showed a significantly higher recurrence rate than those in groups II and I (both P < 0.050, Fig. 4B).

Figure 4.

The prognostic significance of the CTC panel in HCC patients after operation. A, The CTC positivity rates before and after curative resection. B, The prognostic significance of CTC loads with respect to TTR in patients with persistent positive CTC, conversion of CTC from positive to negative, and persistent negative CTC. C, The Kaplan–Meier analysis of TTR for the CTC panel in training (left) and validation (right) groups. D, The Kaplan–Meier analysis of TTR for the CTC panel in BCLC 0+A subgroups in training (left) and validation (right) groups. E, The Kaplan–Meier analysis of TTR for the CTC panel in AFP ≤ 20 subgroups in training (left) and validation (right) groups (CTC high represents CTC load/5 mL > 0.80; CTC low represents CTC load/5 mL ≤ 0.80).

Figure 4.

The prognostic significance of the CTC panel in HCC patients after operation. A, The CTC positivity rates before and after curative resection. B, The prognostic significance of CTC loads with respect to TTR in patients with persistent positive CTC, conversion of CTC from positive to negative, and persistent negative CTC. C, The Kaplan–Meier analysis of TTR for the CTC panel in training (left) and validation (right) groups. D, The Kaplan–Meier analysis of TTR for the CTC panel in BCLC 0+A subgroups in training (left) and validation (right) groups. E, The Kaplan–Meier analysis of TTR for the CTC panel in AFP ≤ 20 subgroups in training (left) and validation (right) groups (CTC high represents CTC load/5 mL > 0.80; CTC low represents CTC load/5 mL ≤ 0.80).

Close modal

The prognostic significance of the CTC panel in HCC patients undergoing resection

Based on X-Tile analysis, an optimal cutoff point of 0.80 (CTC load/5 mL) for the CTC panel showed the most significant power to predict patients outcome in the training set (Supplementary Fig. S6), and the HCC patients were stratified into CTC load/5 m: ≤ 0.80 (CTC low) or CTC load/5 mL > 0.80 (CTC high). In the training set, with a median follow-up time of 25.2 months, 62.5% (75/195) of these patients suffered tumor recurrence. Patients with CTC load/5 mL > 0.80 had significantly shorter TTR (median, 27.7 months vs. not reached) and higher recurrence rates (49.1% vs. 24.7%) than those with CTC load ≤ 0.80 (P < 0.001, Fig. 4C). On multivariate analysis, the CTC panel was an independent prognostic factor for TTR (HR = 2.692, 95% CI, 1.617–4.483; P < 0.001, Supplementary Table S5), and the similar results were also confirmed in the validation set (HR = 3.127, 95% CI, 1.360–7.190; P = 0.007, Fig. 4C; Supplementary Table S5). Furthermore, in AFP-negative and early-stage subgroups, patients with preoperative CTC load/5 mL > 0.80 also showed a relatively higher risk of developing postoperative tumor recurrence than those with CTC load/5 mL ≤ 0.8 (all P < 0.05, Fig. 4D and E).

Nucleic acid–based CTC detection methods have the advantages of high sensitivity and small sample volume required (28, 29). Using an optimal multimarker qRT-PCR detection platform with negative enrichment strategy (15), the aim of this study was to evaluate expression patterns of putative CSC biomarkers in CTCs and investigate the diagnostic and prognostic values as well as treatment response evaluation of CTCs with stem-like features in HCC. Searching for appropriate mRNA marker expressed specifically by tumor cells is critical for the specificity and reliability of CTC detection. After screening the expression patterns of nine putative CSC biomarkers systematically, a CTC detection panel, including EpCAM, CD90, CD133, and CK19, was then constructed through a multivariate logistic regression model. We found the panel exhibited highly accurate in diagnosing HCC, especially in early-stage and AFP-negative diseases. Meanwhile, our CTC detection panel showed great significance in predicting early recurrence of HCC after resection as well as its potential in monitoring therapeutic response. We also found that the multimarker CTC panel shows better AUC than single-marker EpCAM in differentiated HCC and control groups (Supplementary Fig. S7A), and the prognostic significance of CTC panel still retained in EpCAM subgroup (Supplementary Fig. S7B). These results indicate that our multimarker CTC outperforms EpCAM alone in HCC diagnostic and prognostic evaluation. Thus, our study implied that this panel could serve as a useful complement tool for early diagnosis, outcome prediction, and treatment evaluation, which is urgently needed in HCC management.

CSCs are thought to be responsible for tumor recurrence and metastasis (30), and these cells seem more capable of vascular invasion, tending to circulate in even early-stage tumors (31). Although both CSCs and mature cancer cells can migrate into the blood stream, CSCs are more prone to survive in the circulation and deposit in distant organs or recirculate back to the liver remnant (12). For these reasons, we focused on the CTC subpopulations with stem-like phenotypes instead of whole CTC population detection for HCC diagnosis, prognosis, and therapeutic evaluation. Given the documented heterogeneity of CTCs in HCC and the results of our previous study using a single marker (15), a multimarker qRT-PCR assay for CTC detection was conducted, and nine putative CSC markers were chosen for the biomarker screening of CTCs with stem-like cell features (see references of ours and others; refs. 11–19), and only four were finally included for constructing our CTC detection panel.

Serum AFP has been widely used as a diagnostic and screening marker of HCC. However, approximately 30% to 40% of patients with HCC are AFP negative (32, 33), and elevated AFP concentrations have been reported in 11% to 58% of patients with chronic hepatitis or cirrhosis, without HCC (34, 35). Our data indicated that the differential diagnostic capability of our CTC panel performed much better than serum AFP in both the training and validation sets (Figs. 2 and 3; Table 2). Furthermore, the excellent diagnostic performances of our panel were sustained in AFP-negative subsets of HCC (Supplementary Table S4). Currently, imaging remains an important tool for diagnosis and staging of HCC (20). Nevertheless, approximately 30% of patients with HCC lack typical imaging manifestations, possibly because the diagnostic accuracy of imaging is largely affected by tumor size (20). Our CTC panel had significant value in diagnosing patients with early-stage HCC, including those with single tumor of ≤2 cm. Indeed, small-sized lesions of HCC could be detected in subjects at risk, supporting use of our CTC screening panel in this setting. More importantly, when patients at risk without visible hepatic lesion are positive for CTC panel detection, we should make a close follow-up to check tumor existence for timely intervention. Thus, our findings underscore the promise of our CTC panel as a novel biomarker for early diagnosing of HCC and complementing current diagnostic protocols.

According to the hypothesis that CTCs might be the “seeds” of tumor metastasis and recurrence (9, 11), the prognostic significance of the CTC panel was further confirmed in HCC patients undergoing surgery, and high recurrence rate in patients with high CTC load was observed. Our CTC detection panel might serve as a novel indicator reflecting the micrometastatic status which could not be detected by routine image tools, and estimating the recurrence risk of HCC patients in a real-time manner. Meanwhile, the decrease of CTC load was observed soon after resection, and patients with persistently high CTC load postoperatively showed a propensity of increased tumor recurrence, and this suggested that our CTC panel might be a surrogate indicator for surveillance of the treatment response during HCC management. In addition, further studies on these CTC subpopulations (such as EpCAM+, CD90+, CD133+, and CK19+) will present more information for metastatic mechanism as well as related therapeutic targets.

Epithelial–mesenchymal transition (EMT) is essential for circulatory invasion by tumor cells, which become CTCs, and EMT markers are also candidate indices of CTCs (36). We evaluated expression patterns of two major EMT markers (E-cadherin and Vimentin) in 89 patients with HCC patients and in 70 HD subjects, and there was no significant difference in two groups (Supplementary Fig. S8).

To our knowledge, this is the first large-scale, multicenter study to report the utility of CTC detection in HCC diagnosis, prognosis as well as therapeutic evaluation, using both the training and validation sets. However, HCC stemmed from cirrhosis or HBV infection in all patients we recruited, unlike conditions in the United States, Europe, and Japan (37). Thus, the clinical relevance of our CTC panel should be validated in patients from other geographic areas. In addition, the clinical significances of this CTC panel were only evaluated in HCC patients, and the application in other solid tumors needed to be further confirmed.

In conclusion, we have generated a multimarker CTC detection panel with high sensitivity and specificity, capable of differentiating patients with HCC from healthy control and patients with CHB, LC, and BHL. More importantly, this CTC panel may have merit as a tool for providing information about the patient's current disease state and assessing therapeutic response in a real-time manner, which might play an important role in personalized therapy for patients with HCC in the future.

No potential conflicts of interest were disclosed.

Conception and design: Y.-F. Sun, W.-Q. Chen, Y. Cao, X.-R. Yang, J. Fan

Development of methodology: Y.-F. Sun, M.-N. Shen, X.-L. Ma, W.-Q. Chen, X.-R. Yang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): X.-L. Ma, Y. Zhou, B. Hu, M. Zhang, G. Wang, W.-Q. Chen, X. Zhang, Y.-H. Shi, S.-j. Qiu, J. Zhou, X.-R. Yang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y.-F. Sun, X.-L. Ma, J. Wu, C.-Y. Zhang, W.-Q. Chen, L. Guo, R.-Q. Lu, X.-R. Yang

Writing, review, and/or revision of the manuscript: Y.-F. Sun, X.-L. Ma, J. Wu, W.-Q. Chen, B.-s. Pan, X.-R. Yang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): W. Guo, C.-Y. Zhang, W.-Q. Chen, C.-H. Zhou, B.-s. Pan, J. Zhou

Study supervision: Y. Xu, W.-Q. Chen, X.-R. Yang

The authors thank the participating patients for the source of clinical blood samples.

W. Guo was supported by the National Natural Science Foundation of China (81572064, 81772263), the Key Developing Disciplines of Shanghai Municipal Commission of Health and Family Planning (2015ZB0201), the Shanghai Municipal Commission of Health and Family Planning (201440389), and the Projects from the Shanghai Science and Technology Commission (16411952100). Y.-F. Sun was supported by the National Natural Science of China (81602543) and the Sailing Program from the Shanghai Science and Technology Commission (16YF1401400). C.-Y. Zhang was supported by the Shanghai Municipal Commission of Health and Family Planning (201540052). Y. Xu was supported by the National Natural Science of China (81372317). J. Zhou was supported by the National Natural Science of China (81572823, 81772578). X.-R. Yang was supported by the National Natural Science of China (81472676, 81672839), the Projects from the Shanghai Science and Technology Commission (14411970200, 14140902301), and the Strategic Priority Research Program of the Chinese Academy of Sciences (XDA12020103). J. Fan was supported by the National High Technology Research and Development Program (863 Program) of China (2015AA020401), the State Key Program of National Natural Science of China (81530077), the National Natural Science of China (81772551), the Specialized Research Fund for the Doctoral Program of Higher Education and Research Grants Council Earmarked Research Grants Joint Research Scheme (20130071140008), the Projects from the Shanghai Science and Technology Commission (14DZ1940300), and the Strategic Priority Research Program of the Chinese Academy of Sciences (XDA12020105).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Ferlay
J
,
Soerjomataram
I
,
Dikshit
R
,
Eser
S
,
Mathers
C
,
Rebelo
M
, et al
Cancer incidence and mortality worldwide: sources, methods and major patterns in GLOBOCAN 2012
.
Int J Cancer
2015
;
136
:
E359
86
.
2.
Ji
J
,
Shi
J
,
Budhu
A
,
Yu
Z
,
Forgues
M
,
Roessler
S
, et al
MicroRNA expression, survival, and response to interferon in liver cancer
.
N Engl J Med
2009
;
361
:
1437
47
.
3.
Bruix
J
,
Han
KH
,
Gores
G
,
Llovet
JM
,
Mazzaferro
V
. 
Liver cancer: approaching a personalized care
.
J Hepatol
2015
;
62
:
S144
56
.
4.
Bruix
J
,
Reig
M
,
Sherman
M
. 
Evidence-based diagnosis, staging, and treatment of patients with hepatocellular carcinoma
.
Gastroenterology
2016
;
150
:
835
53
.
5.
Bruix
J
,
Gores
GJ
,
Mazzaferro
V
. 
Hepatocellular carcinoma: clinical frontiers and perspectives
.
Gut
2014
;
63
:
844
55
.
6.
Plaks
V
,
Koopman
CD
,
Werb
Z
. 
Cancer. Circulating tumor cell
.
Science
2013
;
341
:
1186
8
.
7.
Miyamoto
DT
,
Zheng
Y
,
Wittner
BS
,
Lee
RJ
,
Zhu
H
,
Broderick
KT
, et al
RNA-Seq of single prostate CTCs implicates noncanonical Wnt signaling in antiandrogen resistance
.
Science
2015
;
349
:
1351
6
.
8.
Maheswaran
S
,
Haber
DA
. 
Ex vivo culture of CTCs: an emerging resource to guide cancer therapy
.
Cancer Res
2015
;
75
:
2411
5
.
9.
Kim
MY
,
Oskarsson
T
,
Acharyya
S
,
Nguyen
DX
,
Zhang
XH
,
Norton
L
, et al
Tumor self-seeding by circulating cancer cells
.
Cell
2009
;
139
:
1315
26
.
10.
Ruscetti
M
,
Quach
B
,
Dadashian
EL
,
Mulholland
DJ
,
Wu
H
. 
Tracking and functional characterization of epithelial-mesenchymal transition and mesenchymal tumor cells during prostate cancer metastasis
.
Cancer Res
2015
;
75
:
2749
59
.
11.
Sun
YF
,
Xu
Y
,
Yang
XR
,
Guo
W
,
Zhang
X
,
Qiu
SJ
, et al
Circulating stem cell-like epithelial cell adhesion molecule-positive tumor cells indicate poor prognosis of hepatocellular carcinoma after curative resection
.
Hepatology
2013
;
57
:
1458
68
.
12.
Yang
ZF
,
Ho
DW
,
Ng
MN
,
Lau
CK
,
Yu
WC
,
Ngai
P
, et al
Significance of CD90+ cancer stem cells in human liver cancer
.
Cancer Cell
2008
;
13
:
153
66
.
13.
Liu
S
,
Li
N
,
Yu
X
,
Xiao
X
,
Cheng
K
,
Hu
J
, et al
Expression of intercellular adhesion molecule 1 by hepatocellular carcinoma stem cell and circulating tumor cells
.
Gastroenterology
2013
;
144
:
1031
41
.
14.
Yang
XR
,
Xu
Y
,
Yu
B
,
Zhou
J
,
Qiu
SJ
,
Shi
GM
, et al
High expression levels of putative hepatic stem/progenitor cell biomarkers related to tumor angiogenesis and poor prognosis of hepatocellular carcinoma
.
Gut
2010
;
59
:
953
62
.
15.
Guo
W
,
Yang
XR
,
Sun
YF
,
Shen
MN
,
Ma
XL
,
Wu
J
, et al
Clincial significance of EpCAM mRNA-positive circulating tumor cells in hepatocellular carcinoma by an optimized negative enrichment and qRT-PCR based platform
.
Clin Cancer Res
2014
;
20
:
4794
805
.
16.
Yamashita
T
,
Honda
M
,
Nakamoto
Y
,
Baba
M
,
Nio
K
,
Hara
Y
, et al
Discrete nature of EpCAM+ and CD90+ cancer stem cells in human hepatocellular
.
Hepatology
2013
;
57
:
1484
97
.
17.
Lee
TK
,
Castilho
A
,
Cheung
VC
,
Tang
KH
,
Ma
S
,
Ng
IO
. 
CD24(+) liver tumor-initiating cells drive self-renewal and tumor initiation through STAT-3-mediated NANOG regulation
.
Cell Stem Cell
2011
;
9
:
50
63
.
18.
Ma
S
,
Tang
KH
,
Chan
YP
,
Lee
TK
,
Kwan
PS
,
Castilho
A
, et al
miR-130b promotes CD133(+) liver tumor-initiating cell growth and self-renewal via tumorprotein 53-induced unclear protein 1
.
Cell Stem Cell
2010
;
7
:
694
707
.
19.
Govaere
O
,
Komuta
M
,
Berkers
J
,
Spee
B
,
Janssen
C
,
de Luca
F
, et al
Keratin 19: a key role player in the invasion of human hepatocellular carcinomas
.
Gut
2014
;
63
:
674
85
.
20.
El-Serag
HB
,
Marrero
JA
,
Rudolph
L
,
Reddy
KR
. 
Diagnosis and treatment of hepatocellular carcinoma
.
Gastroenterology
2008
;
134
:
1752
63
.
21.
Sato
Y
,
Nakata
K
,
Kato
Y
,
Shima
M
,
Ishii
N
,
Koji
T
, et al
Early recognition of hepatocellular carcinoma based on altered profiles of alpha-fetoprotein
.
N Engl J Med
1993
;
328
:
1802
6
.
22.
Ishak
KG
,
Anthony
PP
,
Sobin
LH
. 
Nonepithelial tumors
.
In:
Ishak
KG
,
Anthony
PP
,
Sobin
LH
,
editors
.
Histological typing of tumors of the liver. World Health Organization international classification of tumors
. 2nd ed.
Berlin
:
Springer
; 
1994
.
p.
22
7
.
23.
Forner
A
,
Gilabert
M
,
Bruix
J
,
Raoul
JL
. 
Treatment of intermediate-stage hepatocellular carcinoma
.
Nat Rev Clin Oncol
2014
;
11
:
525
35
.
24.
Shen
Q
,
Fan
J
,
Yang
XR
,
Tan
Y
,
Zhao
W
,
Xu
Y
, et al
Serum DKK1 as a protein biomarker for the diagnosis of hepatocellular carcinoma: a large-scale, multicentre study
.
Lancet Oncol
2012
;
13
:
817
26
.
25.
Lok
AS
,
McMahon
BJ
. 
Chronic hepatitis B: update 2009
.
Hepatology
2009
;
50
:
661
2
.
26.
Reinholz
MM
,
Kitzmann
KA
,
Tenner
K
,
Hillman
D
,
Dueck
AC
,
Hobday
TJ
, et al
Cytokeratin-19 and mammaglobin gene expression in circulating tumor cells from metastatic breast cancer patients enrolled in North Central Cancer Treatment Group trials, N0234/336/436/437
.
Clin Cancer Res
2011
;
17
:
7183
93
.
27.
Camp
RL
,
Dolled-Filhart
M
,
Rimm
DL
. 
X-tile: a new bio-informatics tool for biomarker assessment and outcome-based cut-point optimization
.
Clin Cancer Res
2004
;
10
:
7252
9
.
28.
Kalinich
M
,
Bhan
I
,
Kwan
TT
,
Miyamoto
DT
,
Javaid
S
,
LiCausi
JA
, et al
An RNA-based signature enables high specificity detection of circulating tumor cells in hepatocellular carcinoma
.
Proc Natl Acad Sci U S A
2017
;
114
:
1123
8
.
29.
Sun
YF
,
Yang
XR
,
Zhou
J
,
Qiu
SJ
,
Fan
J
,
Xu
Y
. 
Circulating tumor cells: advances in detection methods, biological issues, and clinical relevance
.
J Cancer Res Clin Oncol
2011
;
137
:
1151
73
.
30.
Lawson
DA
,
Bhakta
NR
,
Kessenbrock
K
,
Prummel
KD
,
Yu
Y
,
Takai
K
, et al
Single-cell analysis reveals a stem-cell program in human metastatic breast cancer cells
.
Nature
2015
;
526
:
131
5
.
31.
Rhim
AD
,
Mirek
ET
,
Aiello
NM
,
Maitra
A
,
Bailey
JM
,
McAllister
F
, et al
EMT and dissemination precede pancreatic tumor formation
.
Cell
2012
;
148
:
349
61
.
32.
Farinati
F
,
Marino
D
,
De Giorgio
M
,
Baldan
A
,
Cantarini
M
,
Cursaro
C
, et al
Diagnostic and prognostic role of alpha-fetoprotein in hepatocellular carcinoma: both or neither?
Am J Gastroenterol
2006
;
101
:
524
32
.
33.
Forner
A
,
Llovet
JM
,
Bruix
J
. 
Hepatocellular carcinoma
.
Lancet
2012
;
379
:
1245
55
.
34.
Taketa
K
. 
Alpha-fetoprotein: reevaluation in hepatology
.
Hepatology
1990
;
12
:
1420
32
.
35.
Johnson
PJ
. 
The role of serum alpha-fetoprotein estimation in the diagnosis and management of hepatocellular carcinoma
.
Clin Liver Dis
2001
;
5
:
145
59
,
36.
Yu
M
,
Bardia
A
,
Wittner
BS
,
Stott
SL
,
Smas
ME
,
Ting
DT
, et al
Circulating breast cells exhibit dynamic changes in epithelial and mesenchymal composition
.
Science
2013
;
339
:
580
4
.
37.
Poon
D
,
Anderson
BO
,
Chen
LT
,
Tanaka
K
,
Lau
WY
,
Van Custem
E
, et al
Management of hepatocellular carcinoma in Asia: consensus statement from the Asian Oncology Summit 2009
.
Lancet Oncol
2009
;
10
:
1111
.