Annals of Surgical Oncology

, Volume 19, Issue 2, pp 402–408

Prognostic Value of Putative Circulating Cancer Stem Cells in Patients Undergoing Hepatic Resection for Colorectal Liver Metastasis


  • Pierluigi Pilati
    • Surgery Branch, Department of Oncological and Surgical SciencesUniversity of Padova
    • Surgery Branch, Department of Oncological and Surgical SciencesUniversity of Padova
  • Loris Bertazza
    • Surgery Branch, Department of Oncological and Surgical SciencesUniversity of Padova
    • Istituto Oncologico Veneto (IOV-IRCCS)
  • Francesca Galdi
    • Surgery Branch, Department of Oncological and Surgical SciencesUniversity of Padova
  • Marta Briarava
    • Surgery Branch, Department of Oncological and Surgical SciencesUniversity of Padova
  • Enzo Mammano
    • Surgery Branch, Department of Oncological and Surgical SciencesUniversity of Padova
  • Emanuela Tessari
    • Surgery Branch, Department of Oncological and Surgical SciencesUniversity of Padova
  • Giorgio Zavagno
    • Surgery Branch, Department of Oncological and Surgical SciencesUniversity of Padova
  • Donato Nitti
    • Surgery Branch, Department of Oncological and Surgical SciencesUniversity of Padova
Colorectal Cancer

DOI: 10.1245/s10434-011-2132-2

Cite this article as:
Pilati, P., Mocellin, S., Bertazza, L. et al. Ann Surg Oncol (2012) 19: 402. doi:10.1245/s10434-011-2132-2



Although surgery is the gold standard treatment of hepatic metastasis from colorectal cancer (CRC), many patients ultimately die of their disease. We tested the hypothesis that the detection of circulating tumor cells (CTC) might identify patients at high risk of dying of disease recurrence after apparently radical liver surgery.


We considered 50 patients undergoing radical surgery for liver-confined hepatic metastasis from CRC. The expression of a panel of cancer-related genes, as assessed by quantitative real-time PCR, was used to detect CTC in the peripheral blood of these patients immediately before surgery. Survival analysis was performed by the Cox regression model.


Univariate analysis of the expression levels of CD133 (a marker of colon cancer stem cells) and survivin (an antiapoptotic factor) resulted in statistically significant association with patient survival [hazard ratio (HR) 2.7, 95% confidence interval (CI) 1.9–3.7, P < 0.0001; and hazard ratio 2.1, 95% CI 1.4–3.2, P < 0.0001, respectively]. Remarkably, multivariate analysis found that only the transcriptional amount of CD133 resulted in statistical significance (HR 2.6, 95% CI 1.9–3.6, P < 0.0001), indicating that this biomarker can independently predict the survival of these patients.


CD133-positive CTC may represent a suitable prognostic marker to stratify the risk of patients who undergo liver resection for CRC metastasis, which opens the avenue to identifying and potentially monitoring the patients who are most likely to benefit from adjuvant treatments.

Despite the improvements of systemic chemotherapy in terms of both tumor response rates and overall survival benefit, surgical resection represents the only potentially curative therapeutic option for patients with liver metastasis from colorectal cancer (CRC).1,2 Unfortunately, only 20% of patients with colorectal liver metastases present with liver-confined resectable disease and/or are candidate for major surgical operation (depending on comorbidities). Importantly, liver resection with curative intent (so called R0 resection, i.e., all macroscopic disease has been removed with histologically negative resection margins) results in 5-year overall survival rates of 20% to 40%.3,4 This consideration underscores the fact that most of these patients still harbor minimal residual disease after apparently radical surgery and thus should be offered postsurgical systemic treatment to improve their life expectancy. However, randomized clinical trials of adjuvant chemotherapy have failed to demonstrate any overall survival advantage in this patient population.57

Besides the need for more effective anticancer drugs, one way of interpreting these unfavorable results is that the trials so far carried out have included patients who do not harbor minimal residual disease and thus cannot benefit from adjuvant treatment (between 20% and 40% of all patients undergoing surgical resection with curative intent, according to the above-reported 5-year survival rates). Ultimately, this might have diluted the potential benefit of adjuvant therapy. In other words, the identification of patients at high-risk of disease recurrence after surgery for resectable liver-confined metastatic disease might lead to better selection of patients who would most benefit from adjuvant chemotherapy, according to the principles of personalized medicine.8,9 Thus, detection of circulating tumor cells (CTC) in the peripheral blood of patients may be a promising approach for the following reasons. First, evidence already exists that CTC have a prognostic role both in CRC and in other tumor types.1016 Second, unlike prognostic biomarkers found in the primary tumor, in metastatic lesions, or in the serum of patients (which are surrogate parameters of cancer aggressiveness or tumor bulkiness, respectively), CTC are believed to be directly involved in the biology of the metastatic process.13,16 In analogy with serum biomarkers [e.g., carcinoembryonic antigen (CEA) and CA19-9], CTC can be easily obtained from the peripheral blood of patients.

In light of these considerations, we assessed the expression levels of a panel of potential CTC markers in the peripheral blood of patients undergoing liver resection for CRC metastasis. We then used this information to test the hypothesis that CTC might help identify patients at high risk of dying from disease. With this aim, we selected a panel of molecular markers associated not only with the presence of CTC (as already reported by others), but also with the metastatic potential of CTC. To our knowledge, this approach is novel.

Patients and Methods

Study Design

The study was designed to retrospectively assess the prognostic role of CTC as detected by the peripheral blood transcriptional levels in patients with resectable liver metastasis from CRC of seven genes: cytokeratin-19 (CK19), CK20, CEA, prominin-1 (CD133), vascular endothelium growth factor (VEGF), epithelial growth factor receptor (EGFR), and Survivin.

Disease-specific survival (DSS) was considered the clinical outcome of interest.

The study was approved by the local ethical committee of the University of Padova, Padova, Italy. Written informed consent regarding the use of biological specimens for investigational purposes was obtained from all patients.

Patients were enrolled onto this study according to the following inclusion criteria: (1) metastatic disease confined to the liver as assessed by preoperative radiological imaging assessment and confirmed by intraoperative evaluation; (2) radical surgery of liver metastatic disease (R0 resection, that is, histologically negative resection margins); and (3) availability of a preoperative peripheral blood sample.

Patient follow-up consisted of physical examination, assessment of serum CEA levels, and thoracoabdominal computed tomographic scan every 6 months for the first 5 years, then annually thereafter.

As a validated multivariate prognostic index, we adopted the clinical risk score proposed by Fong et al.17 Briefly, one point was assigned for each of the following criteria: (1) positive lymph node status of the primary colorectal lesion; (2) disease-free interval from the diagnosis of the primary lesion to discovery of liver metastases less than 12 months; (3) number of hepatic metastases >1 on preoperative imaging; (4) size of the largest hepatic metastases >5 cm on preoperative imaging; and (5) preoperative CEA serum level >200 ng/ml.

The study sample size was calculated using the following parameters: (1) alpha and beta level of significance of 5% and 20%, respectively; (2) minimum detectable hazard ratio (HR) of 3; (3) 3-year DSS rate for patients of the worse prognosis group of 30%; and (4) ratio between number of better- and worse-survival patients of 1. With these settings, the minimum number of patients to be enrolled onto the study was 50 (with 26 events).

The expression levels of seven genes, as assessed by quantitative real-time PCR (qrtPCR), was utilized for the detection of CTC in the peripheral blood of enrolled patients, as we have already reported.18,19 These biomarkers were selected because they can identify epithelial cells (the cytokeratins CK19 and CK20); because they are expressed preferentially by malignant cells (CEA and survivin); or because they play a key role in cancer biology (CD133, VEGF, EGFR, and survivin).

Biopsy samples of liver metastasis obtained from enrolled patients and commercially available cell lines were used to investigate the most likely cell source of the biomarkers associated with patient survival.

Specimens and Cell Lines

A 6-ml aliquot of venous peripheral blood was obtained from each patient immediately before surgery. Samples were immediately processed by centrifuging them at 2000×g for 10 min. The nucleated cell fraction was then collected and stored in liquid nitrogen until use.

Biomarker expression levels were also measured in 10 biopsy samples obtained from liver metastasis of 10 enrolled patients, 10 peripheral blood samples from healthy donors, and in the following human cell lines: primary colon adenocarcinoma cells (HCT-15; ATCC, Manassas, VA), primary fibroblasts (IMR-90; ATCC), and primary endothelial cells (HUVEC; Genlantis, San Diego, CA).

RNA Extraction and cDNA Synthesis

Frozen samples were thawed and total RNA was extracted with the guanidinium thiocyanate–phenol–chloroform method (Trizol; Invitrogen, Carlsbad, CA). The quality of the isolated RNA was established spectrophotometrically and by qrtPCR analysis of the endogenous reference gene beta-glucuronidase (GUS) as described below.

Total RNA (7 μg per 100-μl final reaction volume) was reverse transcribed using random primers and Multi Scribe reverse transcriptase (high-capacity cDNA reverse transcription kit; Applied Biosystem, Foster City, CA). The reaction mixture was incubated for 10 min at 25°C, then at 37°C for 120 min. cDNA was stored at −80°C until use.

Real-Time Quantitative PCR

The transcriptional levels of the seven genes (CK19, CK20, CEA, CD133, VEGF, EGFR, and Survivin) were measured by means of qrtPCR by the relative quantification method, as we have already reported.18,20 As opposed to absolute qrtPCR, this method allows quantification of gene levels only relative to a reference sample, called a calibrator. Therefore, we did not classify patients as positive and negative. Instead, we assigned them a continuous value expressing the amount of the target mRNA relative to the calibrator. Thus, each patient presents a value higher, lower, or equal to that of the patient selected as the calibrator.

Beta-glucuronidase was used as the housekeeping gene (to normalize gene levels according to the sample cellularity), and one of the samples was used as the calibrator (to make relative comparison between samples feasible).

The qrtPCR assay was performed with the ABI PRISM 7500 Sequence Detection system. The PCR reaction proceeded in a mixture (20 μl) containing 10 μl of 2× TaqMan Universal PCR Master mix, 1 μl of 20× TaqMan Gene Expression assay (all reagents purchased from Applied Biosystems, Carlsbad, CA), 7 μl of water, and 2 μl of cDNA template. Fifty cycles of amplification were performed at 95°C (15 s) and 60°C (1 min), and mRNA expression levels were normalized by the housekeeping gene transcriptional levels.

Statistical Analysis

DSS, as defined by the interval between the date of liver resection to death by disease, was the outcome of interest

Univariate and multivariate survival analysis was performed by the Cox regression model. The proportional hazard assumption was checked by the Grambsch-Therneau method, which is based on Schoenfeld residuals.21 Robust standard errors and the Efron’s method for handling ties were used while running the Cox model.

DSS curves for high- and low-risk patients (as determined by the values of independent predictors) were estimated with the Kaplan–Meier method, and the survival functions were compared by the log rank test. The median follow-up was calculated according to the reverse Kaplan–Meier method.22 Completeness of follow-up was calculated as per Clark et al.23

The following prognostic factors were considered at univariate analysis: age (years), sex, clinical risk score, preoperative (neoadjuvant) chemotherapy (yes/no), and the seven biomarkers. Covariates significantly associated with DSS at univariate analysis were entered into the multivariate model.

Multiple imputation was utilized for two variables with missing values (i.e., serum CEA [6 cases] and VEGF gene [4 cases]).24 All tests were two-sided, and the alpha level of significance was set at 5%. All statistical analyses were performed by Stata 11 SE software (StataCorp, College Station, TX).


We searched our prospectively maintained database of patients undergoing liver resection for CRC metastasis at our institution and retrieved the records of 50 patients who met the eligibility criteria and who had undergone radical liver resection between 2002 and 2008. The main characteristics of these patients are listed in Table 1.
Table 1

Main characteristics of 50 patients who underwent radical hepatic resection of metastatic disease from CRC



Age, years, median (IQR)

63 (55–68)

Sex, M/F


Liver metastasis, n






Time to liver progression, mo, median (IQR)

10 (4–37)

Preoperative computed tomography, yes/no (%)

22/28 (44)

Clinical risk score, median (IQR)

2 (1–3)

Type of surgery, n








Length of follow-up, mo, median (95% CI)

36 (23–46)

IQR interquartile range, R0 histologically negative resection margins, R1 histologically positive resection margins, R2 macroscopically incomplete resection

In all cases, liver metastatic disease was metachronous with respect to the primary tumor, with the median time to disease progression to the liver being 10 months (interquartile range 4–37 months).

No patient underwent postoperative (adjuvant) systemic or locoregional chemotherapy; 22 of 50 received neoadjuvant systemic chemotherapy.

After a median follow-up of 36 months (interquartile range 23–46 months; follow-up completeness 96%), 22 are alive (10 with no evidence of disease, 12 alive with disease), and 28 have died of disease.

The findings of univariate and multivariate survival analysis are reported in Table 2.
Table 2

Survival analysis of 50 patients who underwent radical hepatic resection of metastatic disease from CRC


Univariate analysis

Multivariate analysis

HR (95% CI)

Beta (SE)


HR (95% CI)

Beta (SE)



0.986 (0.960–1.012)

−0.014 (0.0134)



0.726 (0.351–1.501)

−0.320 (0.371)


Clinical risk score

1.223 (1.026–1.457)

0.201 (0.089)


1.124 (0.981–1.289)

0.117 (0.070)


Neoadjuvant chemotherapy

1.501 (0.479–4.705)

0.406 (0.583)


Gene levelsa


2.718 (1.961–3.766)

0.999 (0.166)


2.611 (1.941–3.600)

0.960 (0.158)



2.190 (1.494–3.211)

0.784 (0.195)


0.964 (0.913–1.018)

–0.037 (0.028)



1.134 (0.429–2.994)

0.126 (0.496)



1.160 (0.838–1.606)

0.148 (0.166)



0.825 (0.418–1.626)

−0.192 (0.347)



1.118 (0.972–1.287)

0.112 (0.072)



0.691 (0.303–1.578)

−0.370 (0.421)


aMeasured in the peripheral blood by means of qrtPCR

Remarkably, at univariate analysis, only the expression levels of CD133 [HR 2.718, 95% confidence interval (CI) 1.961–3.766] and survivin (HR 2.19, 95% CI 1.494–3.211) resulted significantly associated with patient survival, while the other potential prognostic factors (age, sex, clinical risk score, preoperative chemotherapy, and the other five biomarkers) did not.

At multivariate analysis, only the transcriptional amount of CD133 resulted significantly associated with DSS (HR 2.611, 95% CI 1.941–3.600), indicating that this marker can independently predict survival in this patient cohort. The effect of CD133 expression on DSS did not significantly change over time (proportional hazard test rho = 0.051, Chi-square 0.03, P = 0.856).

By using the median value of CD133, we could identify two groups of patients with different prognosis, as shown in Fig. 1 (P < 0.0001, log rank test).
Fig. 1

Kaplan–Meier survival curves of patients with hepatic liver metastasis from CRC undergoing liver resection with curative intent, grouped by CD133 transcriptional levels in peripheral blood

Interestingly, the expression of CD133 was higher in the 10 surgical specimens of liver metastasis from CRC compared to both primary colon carcinoma cell line and normal human cells (i.e., fibroblasts, endothelial cells, and peripheral blood mononucleated cells from healthy donors) (Fig. 2). This finding indirectly strengthens the hypothesis that CD133 might represent a suitable candidate as a biomarker for CTC originating from metastatic CRC.
Fig. 2

CD133 mRNA levels in normal cells [fibroblasts (IMR90), peripheral blood mononuclear cells (PBMC), and endothelial cells (HUVEC)] and CRC cells [primary tumor-derived cell line (HCT15) and liver metastasis biopsy samples]


Surgical resection is the gold standard treatment for resectable liver metastasis from CRC. However, many patients undergoing surgery ultimately die of their disease, which means that surgery is not sufficient in a many cases. Unfortunately, no adjuvant chemotherapy regimen has been definitively demonstrated to significantly impact on the survival of these patients, and thus no medical treatment is currently considered as a standard in this setting.3,25,26

Until more effective therapies can be developed, one strategy for improving the outcome of these patients is to identify those who would most benefit from postoperative treatments—that is, those with the highest likelihood of disease progression. This would allow us to spare low-risk patients the toxicity of chemotherapy; further, it would limit treatment to patients with the clinically latent residual disease responsible for disease recurrence and ultimately death.

According to the principles of personalized medicine, many biomarkers have been proposed to identify CRC patients at high risk of disease relapse, although no general consensus has been reached, especially in this setting (i.e., after resection of liver metastasis with curative intent).9,27

CTC are a particularly appealing biomarker because it is believed they are directly involved in the metastatic process.13,15 CTC were first described in patients with cutaneous melanoma; since then, their detection has been reported in many other solid tumors, such as breast cancer, prostate cancer, and CRC.13,15 For the prognostic value of CTC, the published findings are heterogeneous and are for other tumor types, which is why CTC are not currently recommended for patient risk stratification in the routine clinical setting.1016

One reason for the failure of CTC to consistently and reproducibly identify high-risk patients is that not all of them are biologically meaningful (that is, capable of generating metastatic deposits) and thus prognostically informative. In fact, it is estimated that only one out of every 10,000 CTC is able to form a metastatic deposit. This phenomenon, known as metastatic inefficiency, would explain why the presence of CTC in the peripheral bloodstream is a necessary but insufficient condition for the occurrence of the metastatic process.28 To identify the relevant CTC subset, molecular characterization is the logical step; some promising results are already available.29,30

For CRC, CTC detection with both PCR-based and cytometric methods has yielded mixed results; it is of note, as for other carcinomas, that most researchers have focused on epithelial cell biomarkers (e.g., cytokeratins) associated with the presence or absence of CTC, regardless of their characterization. An important caveat in the study of the prognostic role of CTC is that the simple detection of CTC might not be informative; in fact, it is universally accepted that most CTC, which can be found in many patients with advanced solid tumors, ultimately die in the bloodstream before forming a metastatic deposit in a target organ (e.g., lung, liver, brain, bone).13,16 Investigators believe that only a minority of CTC possess the molecular setting that makes them capable of giving rise to a metastatic nodule; clearly the molecular characterization of this CTC subset would allow not only to develop new therapeutic targets but also to identify a valuable prognostic biomarker.29,31 This is why we chose a set of biomarkers capable not only of revealing the presence of CTC (as assessed by the expression of CK19, CK20, and CEA), but also to identify those that are biologically more aggressive (as assessed by the expression of CD133, survivin, VEGF, and EGFR) and thus prognostically more informative.

Here we showed that in a homogeneous series of patients such as those surgically resected for liver-only metastatic disease, CTC detection with epithelial cell biomarkers (e.g., cytokeratins) are of not of prognostic value. In contrast, biomarkers suggestive of tumor aggressiveness such as survivin and CD133 do correlate with patient survival by univariate analysis, and one of them (i.e., CD133) independently predicts patient survival by multivariate analysis (Table 2; Fig. 1).

Interestingly, CD133 has been recently proposed as a marker of colon cancer stem cells, although the expression of this single molecule does not appear to reliably identify the entire population of tumor-initiating cells in human metastatic colon cancer.3236 Furthermore, some investigators have reported that the expression of CD133 in different cancers, normal tissues, or compartments correlates with CRC patient survival.3739 However, some conflicting results have also been published.40,41

Overall, these observations support the hypothesis that expression of CD133 in the peripheral blood of patients affected with CRC might identify high-risk patients by detecting putative circulating cancer stem cells that might be responsible for disease progression after apparently radical surgery.

Although our results appear appealing because it may permit personalized treatment of patients with resectable liver metastasis, the PCR-based method we used does not allow identification of the cell source of marker expression. For instance, CD133 is also known to be expressed by endothelial cells. In this regard, some investigators have reported that CD133 mRNA levels in the peripheral blood correlate with survival of patients affected with tumor, node, metastasis system stage I to IV CRC, and they attribute this expression to bone marrow–derived circulating endothelial cells.42

As expected, the PCR-based method did not allow us to directly address the issue of the cell source (as opposed to the cytometric method); however, our in vitro experiments showed far higher CD133 transcriptional levels from surgical specimens of CRC liver metastasis than from normal cells (fibroblasts, peripheral blood mononuclear cells, and endothelial cells) as well as primary CRC cells (Fig. 2). These findings support the hypothesis that the detection of CD133 in the peripheral blood is likely to derive from metastatic CRC cells.

In conclusion, although our results need to be confirmed in larger series, analysis of our data suggests that CD133 transcriptional levels in the peripheral blood of patients who have undergone radical liver resection for CRC metastatic disease might be a useful prognostic marker for risk stratification and selection of patients most likely to benefit from adjuvant chemotherapy. Moreover, our findings indicate that further research should be performed on the use of CD133 as a marker for responsiveness to adjuvant therapies as well as a marker for early detection of disease recurrence during follow-up.


Supported in part by a grant from AIRC (Associazione Italiana per la Ricerca contro il Cancro), an Italian association for cancer research.

Copyright information

© Society of Surgical Oncology 2011