PerspectiveCancer Biomarkers

Going with the Flow: From Circulating Tumor Cells to DNA

See allHide authors and affiliations

Science Translational Medicine  16 Oct 2013:
Vol. 5, Issue 207, pp. 207ps14
DOI: 10.1126/scitranslmed.3006305


Molecular analyses of circulating tumor DNA (ctDNA) in plasma from cancer patients have the potential to deliver minimally invasive diagnostic and disease-monitoring biomarkers. Drawing from experience gained through the translation of circulating tumor cell detection to clinical tests, we discuss ctDNA as a source of tumor material for biomarker development.

High-throughput molecular technologies developed over the past decade have contributed to cancer treatment by integrating clinicopathological parameters with the genomic profile of a tumor to delineate diagnostic, prognostic, and predictive strategies tailored to an individual patient—a therapeutic approach called precision medicine. Massively parallel DNA sequencing [also called next-generation sequencing (NGS)] has revealed that cancers harbor heterogeneous repertoires of somatic mutations (1). In addition, it has been demonstrated that cancers display intratumor genetic heterogeneity (25 ) and that there are differences in the mutational make-up of primary tumors and those that develop at metastatic sites (6, 7). This spatial and temporal intratumor genetic heterogeneity may constitute a challenge for the realization of precision medicine (8, 9 ), and it is unclear as to whether the optimal assessment of the constellation of molecular alterations in cancer cells should be performed on primary tumor tissue, metastatic deposits, circulating biomarkers, or all three.

Given their minimally invasive nature, blood-borne biomarkers hold the promise of becoming noninvasive surrogates for tissue-based biomarkers. Circulating tumor cells (CTCs) and, more recently, circulating tumor DNA (ctDNA) are being investigated as potential sources of tumor material for dissection of the molecular and clinical heterogeneity of cancers as well as for the development of biomarkers for prognostication, prediction, and monitoring of response to therapy. Unlike tumor biopsies, CTCs and ctDNA in the blood of cancer patients may overcome problems such as sampling bias because these materials are thought to be shed in the bloodstream by both primary tumors and the whole set of metastases (if present). Therefore, CTCs and ctDNA should be a source of tumor-derived nucleic acids from all disease sites.

Although CTC enumeration is a prognostic biomarker in patients with metastatic breast cancer (10), its use as a source of biological material for biomarker assessment has not experienced the same level of success; hence, there is growing interest in whether ctDNA can be used for this purpose. There is now evidence that ctDNA may be used not only as a quantitative marker for longitudinal follow-up and disease monitoring (11) but also for characterization of the genomic landscape of a given cancer throughout the various stages of the disease (1215) and as a means to address the challenge of intratumor genetic heterogeneity (8, 16). Here, we provide an overview of the current clinical applications of CTCs and ctDNA and discuss the challenges associated with the translation of ctDNA into clinically useful tools, taking into account experience gained through the development and validation of CTCs as biomarkers.


In addition to the U.S. Food and Drug Administration (FDA)–approved CellSearch system, a variety of platforms are currently available for the detection of CTCs (1720). However, not all of the cells identified are CTCs, nor are CTCs of all phenotypes captured by these approaches (21). This poses challenges for the use of CTCs for tumor biomarker assessment. In contrast, DNA extraction from plasma samples is relatively straightforward. The detection of ctDNA within total circulating cell-free DNA in plasma requires the use of molecular methods and is based on the genetic or epigenetic differences between normal and tumor-derived DNA (Table 1). Given the small fragment size of ctDNA and the varying proportions of tumor-derived DNA in the pool of circulating cell-free plasma DNA, ctDNA detection initially proved to be technically challenging (22, 23). Standard polymerase chain reaction (PCR)–based assays have a relatively limited sensitivity and cannot detect mutations that represent <5 to 10% of the total pool of alleles. Identification of somatic genetic and epigenetic aberrations now has been facilitated by the advent of highly sensitive techniques (2428). Tailored PCR focusing on specific structural genomic variants—translocations (29), insertions, or deletions (30)—known to be present in the primary tumor allows for high sensitivity (0.001%; that is, the true detection of one mutant allele among 100,000 wild-type alleles). In addition, new PCR-based approaches, including digital PCR and BEAMing, can detect somatic point mutations with sensitivities ranging from 1 to 0.001% (Table 2). These PCR-based methods with a high sensitivity are limited by the fact that the exact genomic aberrations to be investigated must be known a priori or because the methods are labor intensive or challenging to design.

Table 1

Potential and limitations of ctDNA-based biomarkers.

View this table:
Table 2

Sensitivity of techniques for detecting point mutations in ctDNA.

View this table:
Unearthing evidence.

Thorough detectives use modern techniques to profile and track down suspected killers. Deadly tumors scatter evidence of their existence and molecular character in the form of CTCs, which are shed from tumors into the bloodstream, and cell-free DNA, which is thought to be released by dying tumor cells. Because both primary tumors and metastases release CTCs and ctDNA, these biological clues may capture the entire genetic heterogeneity of a given cancer, making it more likely that a deadly cancer can be arrested.


Owing to the quantitative nature of the sequencing information provided, NGS technologies have allowed for the development of new approaches for ctDNA analysis. NGS analysis of cell-free plasma DNA can be used to (i) determine the presence of a given mutation and estimate its allelic frequency within a sample (31) and (ii) perform whole-exome or whole-genome characterization of the entire repertoire of mutations in a cancer (12, 15). If sets of selected sequences of known mutations are tested, high sequencing depth (>10,000×) can now be readily achieved. Although high-depth targeted sequencing allows for the identification and quantification of rare sequence variants within a sample, artifacts generated in the amplification step, library construction, and base calling may lead to false-negative and false-positive results (31). Because of these intrinsic error rates, the detection of mutations present in <1% of total DNA has proven challenging with standard NGS techniques (14, 32). Whole-exome or whole-genome (14) sequencing approaches face similar challenges and, if performed at standard depths (for example, 100×), preferentially identify mutations with high allelic frequency (such as those present in the modal population of cancer cells) (15). Techniques to overcome these issues are being developed (3335).


CTCs are cells shed from tumors into the bloodstream, whereas the main source of free circulating DNA in serum and plasma is thought to be cells undergoing apoptosis, necrosis, or necroptosis, processes during which DNA is fragmented [for reviews, see (28, 36, 37)]. Because distinct biological processes give rise to CTCs and ctDNA, it is plausible that the qualitative and quantitative information and potential clinical applications of these modalities of blood-borne biomarkers will not overlap entirely.

A decade ago, the proposed clinical applications of CTC detection were defined as prognostication, early assessment of tumor response to treatment in metastatic cancer patients, and biological characterization of isolated cells (38). In breast cancer, the levels of evidence (LOE) reached for these different applications of CTCs as biomarkers are described in table S1. Briefly, the use of CTC enumeration as a prognostic biomarker for metastatic breast cancer patients is currently supported by level-I evidence (evidence from at least one prospective randomized trial) (10), whereas the molecular characterization of CTCs has not yet delivered on the promise of being a replacement or even an ancillary source of information for tissue-based biomarkers. Several studies focusing on the commonly used predictive factors in breast cancer—estrogen receptor, progesterone receptor, and HER2—report significant but heterogeneous discrepancy rates between the expression of these predictive factors in CTCs versus the matched primary tumor, together with a significant heterogeneity within the CTC population from a given patient (3941). Such discrepancies can result from either the intrinsic biology of CTCs (and, therefore, possibly relevant for the treatment of patients) or technical limitations, mostly because of inappropriate technique, inadequate cut-offs to determine the expression or amplification of a given biomarker, and limited numbers of CTC assessed. Despite the controversies, the clinical relevance of these CTC-based assessments of predictive factors is being investigated in several interventional studies addressing the efficacy of diverse anti-HER2 treatments (for example, chemotherapy-lapatinib NCT01619111, or trastuzumab-pertuzumab without chemotherapy NCT01048099) in patients with HER2-positive CTCs but HER2-negative primary breast cancer.

With regards to DNA sequencing from CTCs, a recent study of colorectal cancer patients with high CTC counts showed that most known driver mutations were shared by the pool of CTCs and matched tumor biopsies, whereas the driver mutations initially found only in CTCs were, in fact, detectable at subclonal levels upon further deep sequencing of the tumor tissue (42). In contrast to pooled analysis of CTCs, analyses of individual CTCs from patients in the same setting have been shown to display considerable intra- and interpatient heterogeneity with regards to mutated driver genes (43).

As opposed to CTCs, which are by definition tumor-derived, the release of cell-free DNA into the blood by dying cells is not restricted to cancer patients; in fact, cell-free DNA can be detected in the plasma of healthy subjects, and fetal DNA can be readily found in the plasma of pregnant women. In cancer patients, the proportion of ctDNA detectable within the pool of cell-free plasma DNA is related to tumor burden (11, 27). It is plausible, however, that the yields of ctDNA may also be related to cancer cell death–inducing factors such as hypoxia (for example, antiangiogenic drugs), proliferation, response to treatments, and tumor handling (surgery or biopsy). These variables will need to be assessed, as they may have an impact on the prognostic and predictive value of ctDNA-derived biomarkers. Importantly, however, ctDNA may provide a relatively unbiased sample of the whole tumor burden and, therefore, a means to circumvent the problem of spatial intratumor genetic heterogeneity, which is one of the main causes of the potential nonrepresentative nature of single-tumor biopsies (16, 44).

ctDNA detection is not yet used in the clinical setting, and we discuss its potential future applications below. Some applications, however, can have an immediate clinical impact. For instance, in ~20% of cases, differences in the HER2 status of primary tumors and their metastatic deposits have been reported (45). There is evidence to suggest that ctDNA analysis coupled with digital PCR may constitute a means to detect HER2 gene amplification in plasma samples (46). However, additional validation of these results is required to determine whether the discrepancies between the HER2 status in primary tumor and metastasis were caused by intratumor genetic heterogeneity. Also, the positive (70%) and negative (92%) predictive values observed (46) may not be sufficient for introduction of this approach into clinical practice.

Direct comparisons of CTC counts and ctDNA yields in the same patients (table S2) have revealed that these blood-borne biomarkers are largely correlated because both are associated with tumor burden. In a recent proof-of-concept study (11), PIK3CA and TP53 somatic mutations or private structural DNA variants (that is, a variant found in only one or few cases) were investigated in circulating plasma DNA of patients with metastatic breast cancer. The genetic aberrations retrieved in matched archived tumor tissues were detected in ctDNA from 82% of plasma samples, and the levels of ctDNA were associated with overall survival (P < 0.001). Quantitative changes in ctDNA during treatment were reported to show a higher dynamic range and a better correlation with tumor burden changes than CTC enumeration changes (11). The field eagerly awaits further studies validating these findings.


The implementation of biomarkers is fraught with difficulties, and awareness of the impact of preanalytical and analytical variables has proven crucial for biomarker development and validation. In the case of blood-based biomarkers, variability in blood sampling and handling can have substantial effects on quantitative measurements (19), and standardized methods for CTC and ctDNA analysis are fundamental. A major challenge for the use of CTC and ctDNA detection as biomarkers in the clinical setting is their low concentration in blood and the fact that they correlate with disease extent. Therefore, with the currently available technologies, CTCs and ctDNA are not yet useful for cancer early-detection screening or as diagnostic biomarkers in early-stage disease. With the FDA-approved CellSearch technique, a median of five CTCs per 7.5 ml of blood is found in metastatic breast cancer patients, and in ~20% of these patients, no CTCs are detected. In early breast cancer, only ~20% of patients are found to have ≥1 CTC. This limited sensitivity may arise from incomplete characterization of the phenotype of CTCs: Tumor cells in circulation may not necessarily express epithelial markers, which form the basis of the validated CTC detection test. Instead, these cells may undergo epithelial-to-mesenchymal transition and express mesenchymal markers (47); in this context, bona fide CTCs would be missed using most of the current methods.

Because the reliability of target assessment using CTCs is strongly influenced by the number of CTCs assessed in a given patient (39), any predictive-marker assessment that uses CTCs (for example, HER2 status) may be confounded by the intrinsic prognostic value conferred by their enumeration (that is, the more CTCs analyzed, the more reliable the target assessment and the worse the prognosis). Given the low sensitivity and selective nature of the cells retrieved with current CTC detection methods, it is arguable whether CTCs are a representative source of tumor material for the assessment of predictive markers or molecular targets in cancer patients. Other obstacles to the use of CTCs for the molecular analysis of cancers, in particular with sequencing-based methods, are contamination of CTCs by nonneoplastic cells (such as leukocytes), the difficulty of sorting single CTCs from a CTC-leukocyte mix, and the technical challenges posed by downstream single-cell sequencing analysis and imaging methods.

From a biomarker perspective, the sensitivity and dynamic range of ctDNA detection reported to date (1115, 29, 46) suggest that ctDNA has potential as a prognostic marker for survival, as a predictive marker of response to specific therapies, and for disease monitoring in the metastatic stage. However, the ctDNA detection rate in patients with early-stage cancers remains unknown because studies have largely focused on advanced disease settings. From a therapeutic target perspective, one challenge associated with target discovery from molecular analyses of ctDNA stems from the correlation between ctDNA levels and tumor burden. Given that mutant allelic frequencies in plasma are directly related to the ratio of ctDNA to normal DNA and that this ratio is associated with tumor burden, it is plausible that somatic mutations may only be reliably identified in patients with high disease burden. We anticipate, however, that with the development of more sensitive detection techniques it will be possible to detect mutations in the modal clones of early-stage tumors and to identify mutations heterogeneously distributed in a given cancer in patients with advanced disease.

In a way akin to CTCs, however, the use of ctDNA as a source of biological material for biomarker analyses may also have important limitations. The yield of ctDNA is limited in some clinical settings—for example, after surgical removal of a small primary tumor for which the ctDNA released by any minimal residual disease may be beyond the resolution of the molecular techniques currently available when applied to a “clinically reasonable” volume of blood (5 to 10 ml). This would make the use of ctDNA as a cancer-screening tool challenging. Furthermore, given the inter- and intratumor genetic heterogeneity and the limited number of highly recurrent mutations in some tumor types, such as breast cancer, tracking of a single-point mutation in ctDNA is unlikely to account for this heterogeneity; thus, it would be necessary either to have a priori knowledge of the tumor’s “truncal” driver genetic aberration (which is present in all cancer cells) or to test a large panel of the most recurrent cancer-specific alterations or private chromosomal rearrangements. This is exemplified by a recent study in which targeted deep sequencing of PIK3CA and TP53, the two most frequently mutated genes in breast cancer, in primary tumor samples for subsequent ctDNA detection were identified in only 25 of 52 patients (11).


CTCs and ctDNA have been heralded as blood-borne biological surrogates of tissues for molecular analysis of cancers (so-called liquid biopsies). The genetic approaches used to date rely largely on the detection of known mutations that were previously characterized in archived tumor tissues. Repeated sampling of tumor tissue is, generally, clinically and ethically challenging, whereas CTCs and ctDNA can be assessed several times before, during, or after a given therapeutic intervention. This longitudinal follow-up, combined with the multiplicity of the molecular targets that can be assessed, may provide a wealth of data that can ultimately change the way tumor response to specific therapies is perceived and determined. In a proof-of-concept study, CTCs were detected before and during anti–epidermal growth factor receptor (EGFR) treatment in non–small cell lung cancer patients by using a microfluidic platform (48), which allowed for monitoring of the emergence of the resistance-associated gatekeeper EGFR T790M mutation. Similarly, clinical resistance to anti-EGFR antibodies has been shown to be preceded by the emergence and the progressive increase of detectable KRAS mutations in ctDNA from a subset of colorectal cancer patients (49). With the advent of NGS, which allows parallel assessment of multiple genes associated with treatment response (13), ctDNA and CTCs may help not only to monitor known mutations but also to discover new mechanisms of resistance to treatment. In fact, whole-exome sequencing of ctDNA revealed the emergence of mutations in ctDNA during the course of treatment in six patients with various metastatic tumor types; some of the mutations identified likely caused resistance to the therapies administered to patients analyzed in this study (15). On the other hand, as discussed above, CTCs may not be as useful in this context, given the limitations of the platforms for their identification and retrieval for subsequent molecular analyses.

One of the best documented mechanisms of clinically acquired therapeutic resistance (that is, resistance that occurs after a first phase of tumor sensitivity to treatment) is the outgrowth of drug-resistant clones present at low frequencies in the primary cancer. These clones likely harbor genetic aberrations in addition to the founder genetic events and are thought to expand under the selective pressure of drug treatment (16, 50). Despite the limited data to support this contention, it is plausible that ctDNA characterization and quantification may provide a clinical window to monitor in real time relative variations of the genetically distinct subclones in patients. This would enable medical oncologists to forecast both primary resistance (that is, a mutation known to be a predictor of drug resistance found at high percentages in ctDNA) and secondary resistance to targeted therapies or chemotherapy. With this information, it might also be possible to delay resistance development by devising combinatorial therapeutic regimens based on the percentage of drug-resistant and drug-sensitive cells and the growth kinetics of drug-resistant subclones in a cancer.

Although only a limited number of primary breast tumors and matched metastatic deposits have been analyzed at the base-pair level thus far (1, 44), these analyses supported the clonal origin of metastases—that is, that metastases may arise from one or a few subclones of the primary tumor, each of which may have acquired a metastatic phenotype. Few insights into the biology of metastasis have emerged thus far from the molecular analysis of CTCs because of their biological characteristics and technical issues, as discussed above.

Molecular analyses of ctDNA, on the other hand, may provide insights about the heterogeneity between different sites of the disease and tumor evolution. There is evidence to suggest that in advanced cancer patients, ctDNA is released by and representative of all tumor sites and a large contingent of subclones within these sites (15). This would allow for capturing the entire genetic heterogeneity of a given cancer, whereas a tumor biopsy assesses, by definition, a small portion of the tumor.

ctDNA analysis also provides additional opportunities to capture distinct aspects of the biology of a cancer. For instance, the repertoire of somatic genetic aberrations identified in ctDNA may provide an estimate of the level of genetic instability and intratumor genetic heterogeneity of a given primary tumor at the time of diagnosis. One may speculate that the level of genetic heterogeneity may be associated with the likelihood of a tumor to harbor subclones with metastatic capabilities or resistance-associated mutations (16, 51), even if such subclones are not detected directly by ctDNA analysis. An intratumor heterogeneity score would potentially add information to the current clinicopathological and molecular tools used to estimate the course of disease in early-stage cancer patients.

The development of CTC enumeration as a biomarker in oncology has highlighted many of the challenges associated with the translation of blood-borne biomarkers into clinically useful tools. Meeting the challenges posed by disease monitoring and intratumor genetic heterogeneity will be essential for the realization of precision medicine. We predict that lessons learned from the study of CTCs coupled with technological developments for genetic analyses will provide fertile ground for the implementation of noninvasive cancer biomarkers.



Table S1. Clinical validity and utility of CTC detection as a biomarker in breast cancer.

Table S2. Studies on synchronous detection of CTCs and ctDNA.


  1. Funding: F.-C.B. is supported by a fellowship from the Nuovo-Soldati Foundation for Cancer Research. Competing interests: The authors declare that they have no competing interests.
View Abstract

Navigate This Article