Circulating tumour cells: their utility in cancer management and predicting outcomes

Abstract

Recent advances in technology now permit robust and reproducible detection of circulating tumour cells (CTCs) from a simple blood test. Standardization in methodology has been instrumental in facilitating multicentre trials with the purpose of evaluating the clinical utility of CTCs. We review the current body of evidence supporting the prognostic value of CTC enumeration in breast, prostate and colorectal cancer, using standardized approaches, and studies evaluating the correlation of CTC number with radiological outcome. The exploitation of CTCs in cancer management, however, is now extending beyond prognostication. As technologies emerge to characterize CTCs at the molecular level, biological information can be obtained in real time, with the promise of serving as a ‘surrogate tumour biopsy’. Current studies illuminate the potential of CTCs as pharmacodynamic and predictive biomarkers and potentially their use in revealing drug resistance in real time. Approaches for CTC characterization are summarized and the potential of CTCs in cancer patient management exemplified via the detection of epidermal growth factor receptor mutations from CTCs in patients with non-small cell lung cancer. The opportunity to learn more about the biology of metastasis through CTC analysis is now being realized with the horizon of CTC-guided development of novel anticancer therapies.

Introduction

Metastatic disease is responsible for over 90% of cancer deaths [Wittekind and Neid, 2005; Weiss, 2000]. The hypothesis that circulating tumour cells (CTCs) are a fundamental prerequisite to metastasis was first proposed in the mid 19th Century by Thomas Ashworth, an Australian pathologist [Ashworth, 1869]. Today, the identification and molecular characterization of CTCs in cancer patients remains key to unprecedented insights into the metastatic process and is anticipated to unmask novel therapeutic targets for the treatment of cancer. Progress, however, has been hindered until quite recently by the technical challenges posed by CTC detection, analogous to looking for the proverbial ‘needle in a haystack’. However, there have been major technological advances, in recent years, leading to the genuine prospect of comprehensive interrogation of CTCs. Tumour cells can now be reliably identified in the peripheral bloodstream of cancer patients with metastatic disease and their biological significance is being revealed. This review focuses on some of the more developed methods for CTC detection and the current body of evidence supporting the clinical utility of detecting and characterizing CTCs.

Potential therapeutic uses of circulating tumour cells

The potential applications of CTC enumeration and characterization are far-reaching and could facilitate several key areas of cancer therapeutics (Table 1). Most clinical studies, so far, have focused on CTC enumeration in guiding prognosis in metastatic cancer patients and current research is exploring the pharmacodynamic and predictive biomarker utility of CTCs. A much-anticipated application relates to the detection of CTCs in patients with early stage, resectable cancers with the potential to discriminate selectively for patients most likely to benefit from adjuvant treatment and indeed to monitor the efficacy of therapy during the adjuvant course. However, most methods in current use are not sensitive enough for reliable enumeration of CTCs in early stage cancer patients. In addition, CTC number surveillance, upon completion of therapy, may detect relapse at a much earlier stage than traditional clinical or radiological parameters and this is under investigation.

Table 1.

Potential applications of circulating tumour cell (CTC) analysis.

Enumeration of CTCs	Molecular characterisation of CTCs
• Guide prognosis • Assist in measuring response to anticancer therapy – predictive and/or pharmacodynamic biomarker	• Surrogate for biological activity of underlying tumour – ‘real-time biopsy’ • Elucidate prognostic and predictive molecular features
• May lead to more accurate prognosis when added to existing staging classifications	• Detection of treatment-resistant profiles –ease of serial sampling
• Select patients for adjuvant chemotherapy • Detect recurrent disease • Aid diagnostic process	• Improve understanding of mechanisms of biological processes • Discover and identify new targets for therapeutic manipulation

The molecular characterization of CTCs offers a unique ability to assess genotypic and phenotypic features of a cancer without the need for invasive biopsy. As our understanding of tumour biology, the signalling pathways on which tumour formation and maintenance depends and treatment resistance mechanisms improves, it becomes increasingly clear that mechanism-based anticancer therapeutics must be tailored according to individuals’ tumour characteristics: the concept of personalized medicine. Traditionally, treatment decisions have been empiric based on the histology of tumour biopsies, normally taken for diagnosis, that in some cases were procured several months or even years prior to the patient treatment decisions at hand. The characterization of CTCs, derived from a ‘simple’ blood test, have the potential to serve as an invaluable ‘real-time tumour biopsy’ permitting an up-to-date snapshot of tumour biology, ideally without the need for invasive tests. Furthermore, CTCs can be analysed on a serial basis, permitting the identification of emergent treatment resistance profiles in real time. If successful, this application could herald a reduction in the incidence of unnecessary toxicity experienced by a patient taking futile treatment regimens. However, well-designed, prospective, randomized, multicentre clinical trials coupled with robust CTC methodologies will be needed to confirm that changes in therapy based on CTC evaluation, imposed earlier in a treatment course than possible with traditional radiological methods, make a significant difference to patient outcomes. Early data to support these potential clinical utilities of CTCs are discussed herein, with specific focus on prognostication, correlation of CTC number with radiological outcomes and a survey of approaches used to characterize CTCs at the molecular level.

The metastatic process and difficulties presented for CTC detection

In order to appreciate the challenges of CTC detection, it is important to consider the concepts and theories relating to the metastatic process itself. Paget first described the ‘seed and soil’ theory of tumour invasion and dissemination in 1889 to explain the nonrandom formation of metastasis [Paget, 1889]. The intrinsic properties of the tumour cells (seeds) and host microenvironment (soil) are vital determinants of sites of formation. The anatomical structure of the vasculature is also thought to play a (more mechanical) role in the determination of the destination of tumour cells that enter the circulation [Fidler and Poste, 2008; Coman et al. 1951; Ewing, 1928]. The entire process of metastasis appears to be inefficient in that only the ‘decathlon champions’ of CTCs (as coined by Isaiah Fidler [Fidler, 2003]) are successful in establishing distant metastatic disease. In support of this hypothesis, preclinical models demonstrate that within 24 hours of intravenous administration of tumour cells, less than 0.1% cells remain viable and that less than 0.01% of these surviving CTCs can produce metastasis [Fidler, 1970]. This low ‘success rate’ could be explained by a stochastic/random survival and growth of a few neoplastic cells or, more likely, by the selection of a unique subpopulation of cells with greater metastatic potential. Indeed, the hypothesis that some CTCs express ‘tumour-initiating’ properties has been purported and perhaps it is these cells that are CTC ‘decathlon champions’ capable of seeding distant metastatic disease [Theodoropoulos et al. 2010; Aktas et al. 2009]. The identification of clusters of contiguous tumour cells in the circulation of cancer patients, termed tumour microemboli [Friedl and Gilmour, 2009; Ilina and Friedl, 2009; Liotta et al. 1976], has given rise to an as yet untested hypothesis that these cells, via cell–cell contact survival signaling, have a survival advantage over single CTCs and thus might be more likely to contain the minority of culprit tumour cells that form metastasis.

Contemporary reviews of the metastatic process include the suggestion that a reversible epithelial-to-mesenchymal transition (EMT), describing a major phenotypic change in a subset of cells within the primary tumour, is essential for metastasis to proceed [Thiery and Sleeman, 2006; Yang et al. 2006; Thiery, 2003]. During EMT, epithelial tumour cells lose cell-to-cell to contacts and develop a more motile and invasive mesenchymal phenotype, facilitating their entry into the bloodstream, and revert back to an epithelial phenotype upon extravasation in host tissue (the so-called mesenchymal-to-epithelial transition (MET)) [Christiansen and Rajasekaran, 2006; Thiery and Sleeman, 2006]. Many CTC detection techniques depend on capture of CTCs based on defined epithelial protein expression (e.g. EpCam and cytokeratins); thus, the very process of CTC detection may be inherently flawed if EMT has occurred. However, it is recognized that EMT is not a homogenous ‘black and white’ cellular scenario and it seems likely that CTCs can express both epithelial and mesenchymal properties, to varying degrees, giving rise to heterogeneous CTC populations [Christiansen and Rajasekaran, 2006]. A recent in vivo study postulated a model of cell cooperativity where mesenchymal expressing tumour cells were responsible for breakdown of the extracellular matrix and vascular wall invasion, which facilitated entry of both mesenchymal and ‘passenger’ epithelial tumour cells into the circulation [Tsuji et al. 2009]. Only the epithelial expressing tumour cells had the capabilities to subsequently form distant metastases in this model. The relationships between EMT and collective cell migration are not yet clear.

With this framework of metastatic tumour cell behaviour in mind, the challenge for CTC research is thus not only to discriminate tumour cells circulating in the bloodstream from the vast majority of red and white blood cells (in 10 ml of blood there are an estimated 100 million leucocytes and 50 billion erythrocytes) but additionally to identify that subpopulation of CTCs with the elite, lethal metastatic potential that are ultimately responsible for mortality.

Methods of CTC detection

A battery of novel CTC detection technologies has emerged over the last few years, elegantly summarized in recent reviews [Alunni-Fabbroni and Sandri, 2010; Pantel et al. 2009; Paterlini-Brechot and Benali, 2007; Stebbing and Jiao, 2009]. In principle, methods can be divided into nucleic-acid-based and cytometric approaches.

Nucleic-acid-based methods were predominantly adopted throughout the 1990s following the development of the polymerase chain reaction (PCR) and indeed can be very sensitive techniques, relying on the detection of specific DNA or RNA sequences differentially expressed by tumour cells [Alunni-Fabbroni and Sandri, 2010; Paterlini-Brechot and Benali, 2007]. However, in recent years there has been a preferential shift toward cytometric assays where cells remain intact, hence morphology can be visualized, cells can be enumerated and further analysis by techniques such as fluorescent in situ hybridization (FISH) or even DNA/RNA extraction are practically or theoretically possible, contrasting with the more limited capabilities using nucleic-acid-based methods (Table 2).

Table 2.

Main advantages and disadvantages of techniques used for circulating tumour cell (CTC) enrichment and analysis [Alunni-Fabbroni and Sandri, 2010; Paterlini-Brechot and Benali, 2007].

	Advantages	Disadvantages
Enrichment
Immunomagnetic  separation (manual)	Cells selected on basis of antigen expression  using antibody coated magnetic beads.  Versatility for positive or negative  cell selection.	Dependent on epithelial marker expression  which tumour cells may or may not express;  cells may be lost in sample preparation
Immunomagnetic  separation  (CellSearch)	Semi-automated magnetic separation – easy  to use. Blood samples stable for up to  96 hours. Automation permits reproducible  results between users, laboratories and  samples thus robust for multisite trials.	Dependent on epithelial marker (EpCam)  expression. Challenging to extract pure CTC  for further characterisation but enrichment  good compared with other techniques.  Expensive.
Centrifugation	Simple method. Isolates mononuclear cells  on the basis of differences in density gradient.  Nonexpensive.	Poor enrichment for CTCs - high contamination  with WBCs. Large number CTCs potentially  lost in processing
Filtration (ISET)	Isolates cells on basis of size differences (CTC  large; WBC small). Effective method. Tumour  cells amenable for molecular characterization.	Small tumour cells may not be detected by this  method. Lower sensitivity compared with  other techniques.
Analysis
Cytometric	Cells directly visualized by IHC or IF - permits  evaluation of morphology and CTC  enumeration. Cells can be extracted for  molecular characterization, e.g. FISH, RT-PCR.  In combination this allows greater specificity  than nucleic acid methods.	Detection depends on epithelial or tumour  marker expression. No pure CTC marker  exists. False-positive cells may be enriched.  Current techniques probably less sensitive  than nucleic acid methods.
Nucleic Acid Based  (RT-PCR)	Highly sensitive in particular with multimarker  assay. Can be performed without enrichment  step but at expense of sensitivity. Quantitative  RT-PCR permits relative quantification.	Cells cannot be visualized directly for  morphology or enumeration. Clinical  correlations reported but difficult to  standardize techniques across laboratories.  No transcripts purely specific for tumour cells  thus high false-positive rate.

FISH, fluorescent in situ hybridization; IF, immunofluorescence; IHC, immunohistochemistry; ISET, isolation by size of epithelial tumour cells; RT-PCR, reverse transcription polymerase chain reaction; WBC, white blood cell.

Cytometric approaches use immunostaining profiles to identify and characterize CTCs. These assays need to be highly sensitive, highly specific and highly reproducible if they are to be useful in the clinical setting and used to make patient treatment decisions. Most methods employ an initial enrichment step to optimize the probability of rare cell detection, achievable through immunomagnetic separation, centrifugation or filtration (Table 2). Cytometric-based techniques subsequently interrogate cells by fluorescence microscopy or immunohistochemistry.

CellSearch^™ system

The most widely used cytometric CTC technology currently in clinical testing is the CellSearch^™ platform (Veridex LLC, Huntingdon Valley, PA, USA) and is the only technology to have received FDA approval for the enumeration of CTC in whole blood in specific cohorts of cancer patients [Miller et al. 2010]. The major advantage of this system is its semi-automation and proven reproducibility, reliability, sensitivity, linearity and accuracy [Riethdorf et al. 2007; Allard et al. 2004]. These are features crucial to any biomarker technology to ensure validity of results in clinical testing across multiple sites and have so far been lacking with previous techniques. CellSearch employs immunomagnetic bead-based separation to enrich for CTCs, and the platform characteristics have been described in detail previously [Miller et al. 2010; Allard et al. 2004].

Application of CTC technologies to clinical testing

Prior to the introduction of the CellSearch platform, a variety of combinations of enrichment techniques and detection assays had been applied in an attempt to draw clinically relevant conclusions regarding the prevalence of CTCs in cancer patients. Unfortunately, many of these studies accrued small numbers of patients with insufficient statistical power for translation to routine clinical practice. The inclusion within a single study of patients with different disease stage and treatments also hindered the ability to draw robust conclusions. Other variables between reported studies included: the timing of blood samples in relation to treatment interventions; whether or not the first few millilitres of blood drawn should be discarded on the basis that this will avoid contamination with skin epithelial cells; and, importantly, few studies reported conditions for sample handling, sample transport and the conditions and duration of sample storage prior to analysis. All of these factors, if not standardized, are likely sources of variability. Other biological factors have yet to be examined in sufficient depth, such as the relationship between the rate of cancer cell entry into the circulation and circadian rhythms although studies in 8 prostate and 51 breast cancer patients suggested that multiple sampling over a 24-hour period, within a given individual, did not fluctuate [Martin et al. 2009; Moreno et al. 2001]. Once the patients’ blood samples are presented for evaluation, the choice of antigen for CTC capture in cytometric assays is critical and even if the same capture antigen is exploited across different studies and sites, if different antibodies are used they may have different sensitivity and specificities. Similarly, different reverse transcription (RT)-PCR primers for target sequences in nucleic-acid-based approaches, coupled with subtle differences in assays, might amplify cDNA with different efficiency with wide ranging variation in PCR product. Criteria for the definition of CTCs have also varied. The requirements for exploratory CTC research and the application of CTCs as biomarkers in clinical trials have different requirements, where standardization is essential for the latter application.

CellSearch technology has addressed many of these variables with its semi-automation, standardized kits and user-proficiency assessments and is consequently the predominate technology for clinical testing. Its robust validation permits for reliable comparisons of studies and, as such, will remain a focus for this review. Many of the next-generation CTC technologies hold great promise for improving sensitivity and specificity for CTC detection but validation studies are outstanding prior to their integration into large-scale clinical studies. In this regard, the ‘CTC-chip’ is a very promising approach based on microfluidics and immunomagnetic cell capture and its capability in lung cancer will be discussed [Nagrath et al. 2007].

Clinical studies using CellSearch technology

The first reported study using CellSearch assessed CTC numbers in 145 healthy women volunteers, 199 women with nonmalignant disease and 964 patients with a range of metastatic epithelial cancer types [Allard et al. 2004]. Of all healthy women donors and women with nonmalignant disease, only one sample contained ≥2 CTCs (CTC number (#) is expressed per standard 7.5 ml blood). In contrast, out of 2183 samples taken from 964 metastatic carcinoma patients CTC# ranged from 0 to 23,618 and 36% of specimens had ≥2 CTCs. The presence varied widely in samples from different carcinomas with frequency being highest in metastatic prostate, unknown primary, ovary and breast carcinomas [Allard et al. 2004]. Results were highly reproducible between duplicate measurements and between measurements performed in different laboratories, paving the way for further studies aimed at determining the prognostic, pharmacodynamic and predictive value of CTC enumeration in patients with advanced cancers. Three seminal studies in breast, prostate and colorectal cancer [Cohen et al. 2008; de Bono et al. 2008; Cristofanilli et al. 2004] have subsequently shown that CTC#, over a specified cut-off value, predicts for worse prognosis, and a change in CTC# following initiation of therapy demonstrates predictive value for survival outcome (Tables 3 and 4). These studies have led to FDA approval of the CellSearch system as an aid to prognosis and aid to monitoring patients on therapy in these three clinical situations. Several smaller studies have supported these findings and our own group has begun to investigate the role of CTCs in the management of small cell lung cancer (SCLC) and non-small cell lung cancer (NSCLC) [Krebs et al. 2010; Hou et al. 2009].

Table 3.

Summary of selected prognostic studies published to date using CellSearch technology.

Study and cancer type	Breast [Hayes et al. 2006; Cristofanilli et al. 2005, 2004]	Breast [Nole et al. 2008]	Castrate resistant prostate cancer [de Bono et al. 2008]	Castrate resistant prostate cancer [Scher et al. 2009]	Castrate resistant prostate cancer [Olmos et al. 2009]	Castrate resistant prostate cancer [Danila et al. 2007]	Colorectal [Cohen et al. 2009, 2008]	Colorectal [Tol et al. 2010]	Small cell lung cancer [Hou et al. 2009]
No of patients	177	80	219	156	119	112	413	451	50
% Positive for CTCs at BL (above cut-off)	49% (≥5)	61% (≥5)	58% (≥5)	55% (≥5)	50% (≥5)	62% (≥5)	26% (≥3)	29% (≥3)	86% (≥1)
Line of therapy	Any	1st, 2nd or 3rd	Any	1st	Any	Any	1st, 2nd or 3rd	1st	1st
CTC prognostic cut-off	5	5	5	no cut-off	5	no cut-off	3	3	no cut-off
Median OS according to BL CTC count:	–	–	–	–	–	–	–	–	–
<cut-off	18 m	–	21.7 m	–	>30 m	–	20.6 m	22.0 m	–
≥cut-off	10.6 m	–	11.5 m	–	19.5 m	–	9.4 m	13.7 m	–
P value	<0.001	–	<0.0001	–	<0.0001	–	<0.0001	<0.0001	–
Median OS according to BL and FU CTC counts:	–	–	–	–	–	–	–	–	–
<cut-off at BL <cut-off at FU	22.6 m	–	>26 m	–	>30 m	–	17.7 m	21.9 m	–
≥ cut-off at BL <cut-off at FU	19.8 m	–	21.3 m	–	21.4 m	–	11.0 m	14.5 m	–
<cut-off at BL≥cut-off at FU	10.6 m	–	9.3 m	–	11.2 m	–	10.9 m	6.3 m	–
≥cut-off at BL ≥cut-off at FU	4.1 m	–	6.8 m	–	9.2 m	–	3.7 m	6.3 m	–
Timing of FU blood sampling:	Up to 15-20 weeks	–	Up to 13-20 weeks	–	After 1-2 cycles of novel agent in phase 1 setting	–	3–5 weeks	1–2 weeks	–
Additional comments	CTC values assessed for correlation with radiological response (see Table 4)	Patients with ≥5 CTC at baseline or follow up compared to <5 CTC predicted worse PFS; S not reached	CTC counts predicted OS better than PSA decrement algorithms at all time points	Higher baseline CTCs (as a continuous variable) predicted for worse OS (HR 1.58, p < 0.0001)	Patients with >50 CTC at baseline had poorer OS compared with patients with 5–50 or <5 CTCs (6.3 vs 21.1 vs 30 months, p < 0.001).	Higher baseline CTCs (as a continuous variable) predicted for worse OS. Predictive model improved with inclusion of PSA and albumin levels	CTC values assessed for correlation with radiological response (see Table 4)	CTC values assessed for correlation with radiological response (see Table 4)	Higher baseline CTCs (as a continuous variable) predicted for worse OS (HR 1.1, p = 0.015)

CTC, circulating tumour cell; HR, hazard ratio; m, months; OS, overall survival; PFS, progression free survival; FU, follow up; BL, baseline.

Table 4.

Summary of selected CellSearch-based studies comparing circulating tumour cell numbers with radiological outcomes and overall survival.

Study and cancer Type	Breast [Budd et al. 2006]	Breast [De Giorgi et al. 2009]	CRC [Cohen et al. 2008]	CRC [Tol et al. 2010]	Breast [Nakamura et al. 2009]	Breast [Liu et al. 2009]
Number of patients	138	102	364	307	107	68
Imaging technique	CT	PET-CT	CT	CT	CT	CT
Timing of imaging from  baseline	10 weeks	9–12 weeks	6–12 weeks	9 weeks	12 weeks	9–12 weeks
Timing of CTC from baseline	4 weeks	9–12 weeks	Within 1 month  imaging/death	1–2 weeks	3–4 weeks	3–12 weeks
CTC cut off	5	5	3	3	5	5
Concordance (% patients with  <cut off CTC and NPD by  imaging and patients  with ≥ cut off CTC and PD  by imaging)	76%	75%	78%	90%	Not reported	70%
Median OS according to  Imaging and CTC  parameters:
1. NPD by Imaging and CTC  <cut off	26.9 m	Not reached	18.8 m	21.6 m	Not reported	Not reported
2. PD by Imaging and CTC  <cut off	19.9 m	14.7 m	8.3 m	7.1 m
3. NPD by Imaging and CTC  ≥cut off	15.3 m	4.9 m	7.1 m	9.4 m
4. PD by Imaging and CTC  ≥cut off	6.4 m	9.8 m	3.1 m	3.4 m
Statistical significance  between groups:
1v2	0.0785	0.0018	<0.0001	<0.0001	Not reported	Not reported
1v3	0.0389	<0.0001	<0.0001	<0.0001
1v4	<0.0001	<0.0001	<0.0001	<0.0001
2v3	0.7549	0.1073	0.4662	Not reported
2v4	0.0039	0.2244	0.0001	Not reported
3v4	0.0777	0.6441	0.0330	Not reported
Additional comments	CTCs strongest predictor for OS in multivariate analysis. However, both CTC values and imaging were useful in combination in predicting survival outcome.	CTCs strongest predictor for OS in multivariate analysis. In this series imaging helpful in determining prognosis in patients with <5 CTCs. In patients with ≥ 5 CTCs, difference in survival was not significant between those with PD or NPD by imaging	CTCs strong predictor for OS in multivariate analysis. However, both CTC values and imaging were useful in combination in predicting survival outcome	CTCs at baseline and at 1–2 weeks, strongest predictor for PFS and OS in multivariate analysis. However, both CTC values and imaging were useful in combination in predicting survival outcome	In cases where CTCs increased at 3–4 weeks, 63.6% patients showed PD at time of CT scan. In cases where CTCs decreased by >90% at 3–4 weeks, 85.7% patients showed complete or partial response at imaging	If ≥5 CTCs detected at time of imaging, 3–5 weeks prior to imaging or 7–9 weeks prior to imaging, Hazard Ratio for PD at scan = 6.3, 3.1, 4.9 respectively (p < 0.001). Study included patients on hormonal therapy as well as chemotherapy

FU, follow up; OS, overall survival; PD, progressive disease; NPD, nonprogressive disease (defined as combination of complete response, partial response and stable disease); CTC, circulating tumour cell; CRC, colorectal cancer.

CTC number as a prognostic biomarker

Breast cancer

The first prospective, multicentre prognostic study using CellSearch was performed in patients with breast cancer. Out of 177 metastatic breast cancer patients starting a new line of therapy, a high CTC# at baseline defined as ≥5 CTCs compared with patients with <5 CTCs, was an independent predictor for poor outcome in terms of progression-free survival (PFS; 2.7 versus 7.0 months p < 0.0001) and overall survival (OS; 10.9 versus 21.9 months p < 0.0001) [Hayes et al. 2006; Cristofanilli et al. 2005, 2004]. Assessment of CTC levels at any subsequent time point (3 weeks onwards) reproducibly predicted clinical outcome with the worst PFS and OS occurring in those patients with CTC counts ≥5 at any time point tested. Further analysis determined that a dynamic change in CTC# from baseline, up to 15–20 weeks after initiation of therapy, predicted more discriminately for survival outcome. For example, patients with persistently elevated CTC counts (≥5 at baseline and follow up) exhibited significantly worse OS compared with patients with persistently low CTC counts (<5 at baseline and follow up). Furthermore, patients who converted from an elevated CTC count (≥5) at baseline to a low CTC count (<5) at follow up survived significantly longer than those patients who converted from a low baseline CTC count to elevated CTC count at follow up (Tables 3 and 4). These were the first data to support CTCs as a potentially useful marker for determining response to therapy.

Metastatic castrate-resistant prostate cancer

Studies in metastatic castrate-resistant prostate cancer (mCRPC) have shown CTC# determined by CellSearch to be prognostic. In 219 patients with mCRPC commencing a new line of chemotherapy, baseline CTC# ≥5 predicted for significantly worse OS compared with patients with <5 CTCs (median OS 11.5 months versus 21.7 months respectively, p < 0.0001) [de Bono et al. 2008]. Blood samples were taken at monthly intervals following commencement of chemotherapy and CTCs ≥5 at any subsequent time point consistently predicted for worse OS, in a pattern similar to that seen in the breast cancer study (Table 3).

Not only were CTCs a strong predictor for OS in multivariate analysis in this study but their predictive value for survival was superior to that of monitoring a reduction in prostate-specific antigen (PSA) of >30% or >50%, a routinely tested tumour maker with potential predictive value for survival in mCRPC [de Bono et al. 2008].

The use of a low cut-off value (≥5 CTCs) for separating patients into ‘favourable’ and ‘unfavourable’ prognostic groups has been questioned by statistical considerations that challenge the ability of CTC assays to accurately enumerate small numbers of cells, with a risk of stratifying patients into the wrong prognostic group [Allan and Keeney, 2010; Tibbe et al. 2007]. An alternative approach is to consider CTC# as a continuous variable where one would hypothesize the higher the number of CTCs, the worse the prognosis. To this end, data from the above-detailed prostate cancer study were reanalysed considering baseline and post-treatment CTC values as continuous variables [Scher et al. 2009]. Results confirmed that the higher the CTC#, the worse the OS for patients, irrespective of cut off (Table 3). Two smaller prostate cancer studies exhibited similar results. [Olmos et al. 2009; Danila et al. 2007] (Table 3).

Colorectal cancer

Studies in colorectal cancer (CRC) have also shown CTC# determined by CellSearch to be prognostic. In a study of 413 patients with metastatic CRC commencing first-, second- or third-line therapy, patients with baseline CTC# ≥3 had significantly worse median PFS (4.4 versus 7.8 months, p = 0.004) and OS (9.4 versus 20.6 months, p < 0.0001) compared with patients with <3 CTCs [Cohen et al. 2009, 2008]. CTC# at any subsequent blood draw, up to 13–20 weeks after initiation of treatment, was consistently able to predict for worse PFS and OS on the basis of ≥3 CTCs. A dynamic change in CTC number from baseline to follow up predicted more discriminately for survival outcome as seen for patients with breast and prostate cancer, e.g. patients with persistently elevated CTC counts (≥3 at baseline and follow up) exhibited significantly worse OS compared with patients with persistently low CTC counts (<3 at baseline and follow up) (Table 3).

A second study in metastatic CRC assessed the prognostic value of CTCs in patients starting first-line treatment with chemotherapy and targeted agents [Tol et al. 2010]. The same patterns of results were identified (Table 3).

Lung cancer

Our group has shown that CTCs are detectable in much higher numbers in patients with SCLC compared with other disease types. In a cohort of 50 patients, the median number of CTCs detected was 28, ranging from 0 to 44,896 per 7.5 ml blood. Higher numbers of CTCs were associated with worse survival in univariate analysis with >300 CTCs associated with significantly worse survival compared with patients with <2 CTCs (4.5 months versus 14.8 months respectively, p < 0.005) [Hou et al. 2009]. A persistently elevated CTC count after one cycle of chemotherapy was a stronger predictor for OS. Studies are ongoing regarding the prognostic significance of CTC number in both SCLC and NSCLC where in the latter the range of CTC# is smaller than the former.

Summary of prognostic studies

Common to all of these studies is that CTC# evaluated by CellSearch, prior to commencing a new line of chemotherapy, has prognostic value in breast, prostate and colorectal cancer, leading to FDA approval in these disease types. Furthermore, changes in CTC number during treatment predict survival outcome, highlighting the potential to use CTCs as a surrogate endpoint biomarker. In the clinical setting this may influence changes in treatment strategy, within just a few weeks of starting therapy, in cases where CTCs remain above prognostic cut off or demonstrate an increasing trend. Indeed, at least one randomized study is currently underway to test this hypothesis. The SWOG 0500 study randomizes women with metastatic breast cancer to either continuation of the same chemotherapy or change in therapy on the basis of a persistently elevated CTC count at 3 weeks (≥5 CTCs), in those with an initially elevated count (≥5 CTCs) at baseline. Target accrual is 500 patients and primary outcome measures are PFS and OS. Results are anticipated early 2012.

CTCs as a surrogate response biomarker

Hand-in-hand with changes in CTC number predicting survival outcome is the question of whether changes in CTC number correlate with traditional radiological outcomes, the ‘gold standard’ for monitoring disease status. Several studies have investigated this correlation, some of which were reported in parallel to the aforementioned prognostic studies (Table 4).

The approach has been to take CTC samples at a given time point after commencing a new line of therapy and to compare CTC values with response to treatment according to standard-of-care imaging techniques. Studies correlated CTC number (above and below a specified cut off) with response to imaging according to Response Evaluation Criteria in Solid Tumours (RECIST) and OS was calculated according to both parameters independently or as combined modalities (Table 4).

Although studies were slightly heterogeneous in their methodological approaches with regard to timing of CTC samples and imaging after commencing treatment, the results favoured CTCs over conventional imaging to be the most significant factor in determining prognosis in multivariate analysis. Concordance between CTCs and imaging was reported in the range 70–90% although this should be interpreted with caution as patients with stable disease were grouped together with patients with partial response (PR) and complete response (CR), collectively termed nonprogressive disease (NPD). This means that in cases where CTC# was reduced but stable disease was seen by imaging, a positive correlation was concluded, as opposed to drawing positive correlation only when reduction in CTC# was seen with a true radiological response (i.e. PR or CR). Analysing CTCs and imaging, however, as combined modalities enhanced the ability to predict survival. For example, in a study of breast cancer, both CTC values at 4 weeks and imaging results at 10 weeks were independently predictive for survival [Budd et al. 2006]. In combination, however, greater discrimination for survival was identified, e.g. patients who exhibited PD by CT scan at 10 weeks, could be categorized into those with a very poor prognosis (≥5 CTCs; median OS 6.4 months) and those with a better prognosis (<5 CTCs; median OS 19.9 months, p = 0.0039). Further details are summarized in Table 4.

Studies in CRC showed similar results. In multivariate analysis, CTCs were the most significant factor for determining prognosis but the combination of CTCs and imaging provided additive data (Table 4) [Tol et al. 2010; Cohen et al. 2008]. Such studies have not been reported in prostate cancer as it is notoriously difficult to assess tumour response by standard imaging in this disease.

For future clinical use, the biological significance of CTC# in predicting patient survival soon after commencing chemotherapy is encouraging and may, in certain situations, be more informative than response by imaging, e.g. with novel targeted therapies which are often cytostatic in their effect. There is also a body of evidence to support response by standard CT imaging as an unreliable surrogate for survival [Mandrekar et al. 2010; Desar et al. 2009]. On the basis of data reported so far, a particularly pertinent clinical setting to which CTCs could be applied is that of phase 1 trials. CTC# may be used to predict the activity of a drug early in its development, potentially facilitating go/no-go decision-making and thus expediting the drug development process and reducing expenditure by avoiding the all-too-frequently seen ‘drug fails late and expensive’ scenario. The ability to ascertain drug–target expression and downstream measures of drug–target hitting in CTCs are other exciting prospects beginning to be realized (as discussed in the following).

CTCs as a real-time biopsy

The characterization of CTCs at the molecular level will facilitate personalized medicine via analysis of pharmacodynamic and/or predictive biomarkers, and biomarkers of drug sensitivity or inherent/acquired resistance. This is already being realized, exemplified by studies of human epidermal growth factor receptor-2 (HER2) status in patients with breast cancer. Expression of HER2 has been detected in CTCs in metastatic breast cancer patients in cases where primary tumours were negative at original diagnosis [Pestrin et al. 2009; Fehm et al. 2007; Hayes et al. 2002]. This suggests change in biological activity over time and may justify treatment with HER2-receptor antagonists, such as trastuzumab, in patients who previously would not have been eligible. Indeed one study showed 9 of 24 breast cancer patients whose primary tumour was HER2-negative each acquired HER2 gene amplification in their CTCs during cancer progression [Meng et al. 2004]. Four of the nine patients were treated with trastuzumab-containing therapy, three of whom showed evidence of complete or partial response.

An exemplary model for personalized medicine lies in the detection of epidermal growth factor receptor (EGFR) mutations in patients with NSCLC, predicting for response to EGFR tyrosine kinase inhibitors such as gefitinib and erlotinib [Sequist et al. 2007]. In a large phase 3 study of East Asian patients with NSCLC, randomized to gefitinib (250 mg/day) or carboplatin (area under the curve 5 or 6 mg/ml/minute) with paclitaxel (200 mg/m²) (CP), patients with EGFR mutations treated with gefitinib exhibited significantly longer PFS compared with patients treated with CP (hazard ratio [HR] 0.48, 95% confidence interval [CI], 0.36–0.64, p < 0.001). Conversely, patients negative for the EGFR mutation exhibited significantly longer PFS if they received CP chemotherapy versus gefitinib (HR with gefitinib 2.85, 95% CI 2.05–3.98, p < 0.001) [Mok et al. 2009]. Mutation analyses were performed using tumour specimens; however, EGFR-mutation analysis is now possible in CTCs.

CTC chip technology and EGFR mutation analysis

Pre-eminent studies using a novel CTC technology have demonstrated that EGFR mutations can be detected from CTCs [Sequist et al. 2009; Maheswaran et al. 2008; Nagrath et al. 2007]. The ‘CTC chip’ isolates CTCs by passing whole blood through a microchip containing 78,000 geometrically arranged microposts coated in EpCam. Sensitivity and specificity is reported as 99.1% and 100%, respectively, and CTCs were detected in 115/116 samples taken from 68 patients with a range of tumours including NSCLC (n = 55), prostate (n = 26), pancreatic (n = 15), breast (n = 10) and colon (n = 10) [Nagrath et al. 2007].

Using this technology, DNA was extracted from CTCs isolated from patients with NSCLC and EGFR-mutation analysis performed using the scorpion amplification refractory mutation system (SARMS) [Maheswaran et al. 2008]. CTCs were detected in 27/27 patients with a median number of 74 cells/ml. Twenty patients with known EGFR-mutant primary tumours had CTCs available for molecular analysis, from whom EGFR mutations were detected in CTCs from 19 patients (95%). In addition to the primary activating mutation, the resistance mutation T790M, predictive for EGFR tyrosine kinase inhibitor resistance [Kobayashi et al. 2005], was detected in 11/20 patients. Serial molecular analysis of CTCs in four patients demonstrated evolving genotypes with the emergence of T790M drug-resistance mutation in all four patients, consistent with development of clinical drug resistance.

The potential to use this test in the clinical setting is extremely promising for predicting both response to therapy and emergence of resistance genotypes. It exemplifies perfectly the crucial role CTCs may play in tandem with development of novel molecular agents and routine clinical management.

CTC characterization after enrichment using the CellSearch platform

Several other methods exist to characterize CTCs with the common goal of facilitating personalized medicine and to learn more about the biology of metastasis. Our own group has characterized the malignant origin of CTCs isolated by CellSearch in patients with SCLC. The CellSearch system has a channel, not required for CTC identification, that can be used to detect an appropriately fluorochrome labelled antibody for further CTC characterization. This was exploited in our studies to examine a neuroendocrine marker, CD56 expression in SCLC CTCs. The heterogeneity in CTC expressions of CD56 broadly mirrored that found in matched tumour biopsy [Hou et al. 2009].

In a series of phase I studies, the CellSearch spare channel was used for the detection of insulin-like growth factor-1 receptor (IGF-1R) in patients receiving a monoclonal antibody against IGF-1R either alone or in combination with chemotherapy [de Bono et al. 2007]. Out of 26 patients with detectable CTCs, 23 had evidence of IGF-1R-positive CTCs. In patients with prostate cancer receiving anti-IGF-1R treatment and docetaxel, there was some evidence to support a greater and more rapid decline in PSA in those patients with positive IGF-1R CTCs than in those patients with no CTCs, suggesting a potential predictive marker.

Revealing the genotype/phenotype of CTCs will further our understanding of the heterogeneous populations of CTCs that are beginning to be observed and potentially reveal the CTC ‘decathlon champions’ with the highest malignant potential. To the best of the authors’ knowledge, only one study, so far, has reported global gene expression profiling from CTCs isolated by CellSearch [Smirnov et al. 2005]. This was performed using samples from three patients, one with breast, one with prostate and one with colorectal cancer, each having >100 CTCs/7.5 ml sample. Differences in genetic profiles were seen between each of the patients and candidate tumour-specific genes were selected and prospectively validated by qPCR in a cohort of 74 metastatic cancer patients and 50 healthy donors. The CTC gene signatures were successfully able to discriminate cancer patients from healthy donors, and discriminate between the three different cancer types. This is promising initial data but methods will need to advance further to reliably isolate RNA from smaller numbers of CTCs and to improve purification from contaminating white blood cells which overwhelm and dilute the tumour gene signatures. The move toward single cell gene expression profiling from purified CTCs is greatly anticipated.

FISH has also been performed in CTCs following enrichment by CellSearch to confirm malignant origin or identify predictive aberrations. Three studies in prostate cancer have assessed for FISH abnormalities in CTCs including amplification of androgen receptor (AR) and MYC [Leversha et al. 2009]; gene copy number for chromosomes 1, 7, 8 and 17 [Swennenhuis et al. 2009]; and fusions of TMPRSS2-ERG together with AR and phosphatase and tensin homologue (PTEN) amplifications [Attard et al. 2009]. All have shown genetic aberrations of CTCs confirming their malignant origin.

Progress has been made in terms of ability to characterize CTCs but there is much still to be learned of their biology. The genetic and epigenetic determinants of tumour cell invasion, cell survival within the circulation and ability to colonize distant organs is an area of intense research [Nguyen et al. 2009]. CTC characterization of single cells, once achievable, will offer unprecedented opportunity to explore this further.

Conclusion

Since their first description in 1869, a multitude of studies have provided evidence for CTCs in the blood of cancer patients. Until recently there has been substantial variability in CTC data using several different methods of detection and analysis. CellSearch technology provides robust, reproducible CTC enumeration and by unifying methodology across the scientific community meaningful interpretations are being drawn. The limitations of this approach, however, are founded on the bias towards high enough levels of EpCam expression on CTCs, and evidence is emerging to suggest that CTCs are heterogeneous in this regard. Nevertheless, the prognostic utility of CTCs, enumerated using CellSearch, has been demonstrated for breast, prostate and colorectal cancer and our own studies have shown prognostic significance in SCLC. The potential for serial CTC enumeration to predict response to therapy has also been demonstrated in breast, prostate and colorectal cancer and a prospective evaluation of changing therapy based on CTC enumeration alone, early in the treatment course, is underway in breast cancer.

Whilst data are emerging on several cancer types (breast, prostate, CRC and lung), there remains little information regarding CTCs in many others (e.g. melanoma, pancreatic cancer, head and neck cancer, biliary cancer) and much more research across the cancer spectrum is required. The promise of CTC research in early stages of cancer is largely unmet, requiring more sensitive technologies. However, the ‘real-time biopsy’ potential based on CTC molecular characterization is intuitive and promising and remains a key area for further intensive research. The envisioned future is of a model synonymous to antimicrobial therapy whereby a simple blood test will permit molecular tumour characterization, identification of treatment targets and aid in selection of the most appropriate targeted therapy from an armamentarium of targeted agents.

Funding

Supported by Cancer Research UK grant C147/A12328. M.K. was supported by a Cancer Research UK/AstraZeneca Clinical Pharmacology Fellowship.

Conflict of interest statement

The authors declare no conflict of interest in the preparation of this manuscript.