Circulating Tumor Cells and Circulating Tumor DNA: Challenges and Opportunities on the Path to Clinical Utility

M. Ignatiadis is the principal investigator for the Treat CTC trial, which is supported by grants from Janssen Diagnostics and Roche. M. Lee was an employee of Boreal Genomics and Genomic Health. S.S. Jeffrey is an inventor of intellectual property related to the MagSweeper device for rare cell capture that is owned by Stanford University and licensed to Illumina. The Jeffrey Laboratory has a research collaboration with Vortex BioSciences to help optimize and validate specific applications for Vortex technology that is administered by Stanford University. No other potential conflicts of interest were disclosed.

The following editor(s) reported relevant financial relationships: J.R. Grandis—None.

The members of the planning committee have no real or apparent conflicts of interest to disclose.

Upon completion of this activity, the participant should have a better understanding of the technologies used to detect and characterize circulating tumor cells (CTC) and circulating tumor DNA (ctDNA), different strategies of testing their clinical utility, and the potential future applications of CTCs and ctDNA in the field of precision medicine.

This activity does not receive commercial support.

Next-generation sequencing (NGS) studies performed in bulk primary tumor specimens have demonstrated extensive interpatient (1) and, more importantly, intrapatient (2) heterogeneity. Recently, single-cell analyses of primary breast tumors have provided higher-resolution evidence of intratumor heterogeneity (3) with the finding of substantial clonal diversity and subclonal heterogeneity, such that no two individual tumor cells are genetically identical. Beyond spatial heterogeneity, solid tumors also exhibit temporal heterogeneity, evolving over time under selection pressure from treatment (4, 5). Thus, there is an increased appreciation that the management of metastatic disease should rely on analysis of contemporary tumor tissue rather than on the primary tumor diagnosed years ago (6). However, obtaining serial samples of metastatic tissue is impractical and complicated by spatial heterogeneity and sampling bias. Analysis of circulating tumor cells (CTC) and circulating tumor DNA (ctDNA) thus holds appeal and promise for noninvasive real-time assessment of tumor molecular profiles during the course of disease. Evaluation of CTCs and ctDNA may enable more sensitive monitoring of treatment efficacy and thereby guide drug selection, even potentially in the adjuvant setting where no such tools exist today.

CTCs can be found in the bloodstream of patients with cancer as single cells or, less commonly, as cell clusters, and CTC levels have been shown to have clinical associations with survival and response to therapy (7). CTCs are presumptively shed into the vasculature from primary tumor or distant metastatic foci and are postulated to contain subpopulations of “culprit cells,” which are responsible for seeding and reseeding metastases, eventually leading to patient demise (8). It is thus appealing to not only enumerate CTCs for measuring disease burden and detection of minimal residual disease but also to characterize CTCs as a means to target therapy to these putative culprit cells.

Several reviews have discussed the various CTC enrichment and detection technologies (7, 9–12). Table 1 presents an updated list of CTC assays that have been used to test patient samples within the past 5 years, along with their commercialization status. CTC detection or capture methods can be broadly categorized as either label dependent, using positive enrichment with cell surface markers such as epithelial cell adhesion molecule (EPCAM), also used in in vivo capture techniques, or label independent, enriching for CTCs based on negative selection, size, or other biophysical properties; other strategies include direct imaging of CTCs and functional assays. A significant issue in detecting and capturing CTCs by label-dependent methods is the lack of reliable immunocytochemically identifiable markers that distinguish them from normal epithelial cells. As such, because epithelial cells are rarely present in blood samples from healthy individuals and because circulating epithelial cells (CEC) in patients with cancer often carry the same genetic aberrations seen in the primary tumor (13), the common definition of CTCs has been equivalent to that of CECs: nucleated cells in the bloodstream that express epithelial cytokeratins and do not express the white blood cell surface antigen CD45. More recently, cytokeratin-negative CTCs have been identified, potentially representing tumor cells undergoing epithelial–mesenchymal transition (EMT), or alternatively, cancer stem cells that have not yet shown epithelial differentiation (14–17).

Some technologies listed in Table 1 may detect distinct CTC subpopulations and therefore could be used in different clinical scenarios in the future. However, for any technology to be used in the clinic, demonstration of analytic validity (the accuracy of the test to measure the target of interest), clinical validity (the value of the test to predict the clinical outcome), and ultimately clinical utility (ability of the test to lead to improved clinical outcome when treatment choice is informed by test results) is required (9, 18). The only system currently approved by the FDA as an aid in monitoring patients with metastatic breast, colorectal, or prostate cancer is CELLSEARCH (Janssen Diagnostics; refs. 19–21). Recent data suggest that this technology can also be used for clinical trials across multiple laboratories in the nonmetastatic setting provided that continuous training and central image review is performed (22).

In April 2015, a search in the “ClinicalTrials.gov” website using the keywords “circulating tumor cell” revealed 296 studies involving CTCs. Table 2 refers to studies in multiple tumor types using CTC enumeration or characterization as an inclusion criterion. Table 3 refers to studies for the development and/or validation of CTC assays for a particular indication.

A recent pooled analysis provided level-one evidence for the clinical validity of elevated CTC levels as a marker of poor prognosis in metastatic breast cancer (23). However, the value of CTC enumeration for treatment decision making in metastatic breast cancer was prospectively tested in the Southwest Oncology Group (SWOG) S0500 clinical trial (24). The SWOG trial evaluated the benefit of an early change in chemotherapy for patients with persistently increased CTCs at first follow-up after starting first-line chemotherapy. Of 595 evaluable patients, 123 patients with persistently elevated CTCs on day 21 of therapy were randomized to either continue the same treatment or to switch to an alternative chemotherapy of physician's choice. In this trial, an early switch to an alternative chemotherapy did not increase overall survival (OS). Although CTCs were strongly prognostic, the absence of a survival benefit from changing treatment based on elevated CTC counts suggests that earlier detection of relapse can only be important when a more effective treatment is available: Switching from one ineffective therapy to another ineffective therapy does not change outcome. Instead, changing treatment based on CTC molecular characterization might be a more promising approach to test.

In nonmetastatic breast cancer, detection of elevated CTC levels using CELLSEARCH and other platforms (25–29) is also associated with adverse prognosis. The ongoing Treat CTC trial (NCT01548677; Table 2) is assessing CTC dynamics as an early signal of drug activity. This trial enrolls women with primary high-risk HER2-nonamplified breast cancer who have detectable CTCs after completing surgery and (neo)adjuvant chemotherapy. It evaluates whether six cycles of trastuzumab (a humanized monoclonal antibody targeting the HER2 growth factor receptor) compared with observation alone will eliminate the persistent CTCs. This trial was based on several lines of evidence: (i) in preclinical models, trastuzumab appears to target the cancer stem cell population in a process that does not require HER2 gene amplification (30); (ii) subset analyses of prospective trials demonstrate similar trastuzumab benefit for women with HER2-positive tumors by local testing but deemed HER2-negative by central pathology review (31, 32); and (iii) a single-center randomized phase II study of 75 patients with HER2-negative early breast cancer found that short-course trastuzumab can eliminate chemotherapy-resistant CK19 mRNA-positive CTCs and improve patient outcome compared with observation (33). Additional studies using CTCs in breast cancer are listed in Tables 2 and 3.

CTC enumeration has also been shown to provide prognostic information in metastatic castration-resistant prostate cancer (34, 35). In the COU-AA-301 registration trial that compared abiraterone plus prednisone with prednisone alone, the combination of CTC enumeration and lactate dehydrogenase (LDH) levels at 12 weeks posttreatment was shown to be a surrogate for OS at the individual patient level (36). Efforts are ongoing to validate this biomarker panel.

Clinical studies using CTCs in tumor types other than breast and prostate cancer are described in Tables 2 and 3.

Beyond CTC enumeration, characterization of protein expression on CTCs has also been used to guide treatment selection in clinical trials (Table 2). In breast cancer, HER2 protein expression on CTCs has been assessed using the CELLSEARCH technology, with demonstration that some women with HER2-negative breast cancer may have detectable HER2-positive CTCs (37, 38). However, a phase II study of single-agent lapatinib (an anti-HER2 tyrosine kinase inhibitor) did not find objective responses in patients with metastatic breast cancer with HER2-negative primary tumors and HER2-positive CTCs at study entry (39). Of 139 screened HER2-negative patients, only 96 (69%) had ≥2 CTCs, and only 7 (5%) had ≥50% HER2-positive CTCs and received treatment. One patient (1 of 139 screened patients) had durable disease stabilization, having received lapatinib as a third line of therapy, although efficacy analysis could not be done due to the low number of treated patients. Of note, of the 7 treated patients, the 6 who progressed received lapatinib as ≥fourth line therapy for advanced disease, raising the question of whether HER2 status of CTCs represents HER2 status of the bulk of metastases in very late-stage disease. An ongoing phase III trial (DETECT III, NCT01619111) is evaluating the role of adding lapatinib to chemotherapy in this same patient population (Table 2). Other investigators are using CELLSEARCH to monitor endocrine resistance in ER-positive HER2-negative metastatic breast cancer. To that end, a score based on CTC enumeration and characterization for estrogen receptor (ER), Bcl-2, HER2, and Ki67, the CTC-Endocrine Therapy Index (CTC-ETI; ref. 40), is currently being tested in the COMETI Phase II study (NCT01701050; Table 3).

CTC protein expression has also been characterized in other tumor types. In patients with metastatic colorectal cancer, thymidylate synthase expression in CTCs has been studied as a potential marker of resistance to 5-fluorouracil (41). Immunofluorescent markers have been used to study androgen receptor (AR) signaling in CTCs from patients with metastatic prostate cancer to tailor hormonal treatment approaches (42).

An emerging area of investigation is transcriptional profiling of CTCs to help guide real-time drug selection. For example, a postulated reason that patients with castration-resistant prostate cancer (CRPC) may not respond to drugs that inhibit or impair AR signaling may be the presence of AR splice variants within their tumor cells. This was demonstrated by examining mRNA from CTCs collected prospectively from patients with metastatic CRPC who were enrolled in a clinical trial of abiraterone or enzalutamide treatment. When CTC mRNA was assayed for the splice variant AR-V7, a constitutively active isoform of the AR that lacks the ligand-binding domain, there was a significant association with therapeutic resistance to abiraterone and enzalutamide, drugs that indirectly (abiraterone) or directly (enzulatamide) target the AR, where this ligand-binding domain is present (43).

However, an important challenge with RNA expression analysis of CTC-enriched cell fractions is the potentially confounding signal from contaminating leukocytes. To address this challenge, multiplex PCR with CTC-specific mRNAs (44–46) and single-cell approaches are being explored. High-dimensional single-cell transcriptional profiling of CTCs purified using the MagSweeper, an immunomagnetic enrichment technology (47), revealed significant CTC heterogeneity, even within the same blood draw, suggesting a need for multidrug therapy to approach tumor molecular diversity (8). Importantly, CTCs also showed markedly different gene expression profiles compared with those in single cells from breast cancer cell lines, suggesting that CTC analysis might complement cell-line analysis in drug development. Similarly, prostate cancer also appears to be characterized by CTC heterogeneity, with distinct differences in single-cell expression of EMT-related genes between CTCs from castration-sensitive and castrate-resistant cancers (48).

Several proof-of-concept studies have demonstrated the feasibility of detecting specific somatic mutations or other genetic alterations in pooled or single CTCs from patients with various tumor types (49–54). Moreover, nontargeted approaches are now being used to analyze whole-genome copy number aberrations in single CTCs by array comparative genomic hybridization (aCGH) and NGS techniques (55–57), including whole-exome sequencing of single CTCs (58).

A challenge when analyzing single CTCs using Sanger or NGS methods is to exclude false-positive and false-negative findings due to biases introduced by whole-genome amplification. Moreover, it remains unclear how many CTCs need to be analyzed to capture tumor heterogeneity sufficiently to predict treatment efficacy.

CTCs from patients have been propagated in vitro by multiple groups. These include short-term cultures (28 days or less) of CTCs from patients with breast, colorectal, pleural mesothelioma, prostate, urothelial, bladder, esophageal, pancreatic, gastric, and lung cancers (59–70) and long-term cultures (6–24 months) of CTCs from patients with breast, prostate, and colorectal cancer (71–74). The purpose of such model systems would be to study drug response or, in one study, to identify multilayer clusters, which if they appear in culture by day 14, may be an early predictor of therapy resistance (70).

Several recent studies have reported the development of mouse xenografts generated directly from CTCs or from CTC cultures from patients with advanced breast, colorectal, prostate, hepatocellular, small cell lung, and gastric cancers (68, 72, 74–79). Some of these assays explore metastatic subpopulations of CTCs and others generate patient-specific models for guiding treatment.

In a xenograft assay of luminal breast cancer CTCs, it was demonstrated that, in contrast to bulk EPCAM⁺ CTCs, an EPCAM⁺CD44⁺CD47⁺MET⁺ CTC subpopulation is highly enriched for metastasis-initiating cells when injected into the bone marrow of immunocompromised mice (75). In another study, a subset of EpCAM-negative CTCs taken from the peripheral blood of patients with breast cancer (EpCAM⁻/ALDH1⁺/CD45⁻) was grown in vitro. While all EpCAM-negative CTCs grown in culture caused lung metastases after tail vein or intracardiac injection, only cells enriched for a “brain metastasis selected marker (BMSM) signature,” HER2⁺/EGFR⁺/HPSE⁺/Notch1⁺, showed increased potential for both lung and brain metastasis (71).

CTCs enriched from the blood of patients with small cell lung cancer have been implanted subcutaneously into immunocompromised mice as CTC-derived explants (CDX). The CTCs were tumorigenic when there were greater than 400 CTCs in 7.5 mL of blood; generated CDX not only showed similar genomic profiles to those from the patients' CTCs but accurately reflected the donor patient's response to cisplatin/etoposide chemotherapy (78). Xenografts derived from CTC cell lines can also be used to test the efficacy of different drug combinations (68, 72). Such studies open exciting possibilities for the use of CTC genotyping and functional testing to identify rational drug combinations for clinical evaluation.

Potential limitations, however, of in vitro and in vivo models are that they are usually generated from highly aggressive tumor subclones that may not accurately reflect the spectrum of tumor cell heterogeneity, and, perhaps more importantly in the current era of burgeoning cancer immunotherapy, in vivo xenograft models do not recapitulate tumor–host interactions that may play a role in drug resistance. Additional work is required to optimize experimental conditions for efficient generation of these models from the majority of patients with metastatic cancer and to demonstrate that results from these models accurately reflect outcomes in the clinical setting.

A variety of mesenchymal markers have been identified in CTCs, suggesting an EMT phenotype (8, 80–82) that may contribute to metastatic progression. Plastin3 (encoded by PLS3 gene), an actin-binding/bundling protein, has been identified in tumors and CTCs of patients with primary and metastatic colorectal cancer, with higher expression in advanced and metastatic tumors; CTCs that express plastin 3 show an EMT phenotype and are associated with poor prognosis (83).

Although the majority of CTCs are single, CTC clusters have also been identified in blood samples from patients with metastatic cancer. Also known as circulating tumor microemboli, clusters immunomagnetically captured usng the CELLSEARCH platform have been identified in more than 30% of patients with small cell lung cancer and appear to lack apoptotic or proliferating cells, perhaps offering a survival advantage (84). The use of a new device, the Cluster-Chip, offers label-free isolation of CTC clusters with the use of triangular pillars acting as microfluidic “cluster traps” (85). This method showed that clusters were identified in 30% to 40% of patients with metastatic melanoma, breast cancer, and prostate cancer, with the majority of clusters containing 2 to 10 cells, although sometimes up to 19 cells. However, in contrast to CTC clusters isolated by CELLSEARCH in metastatic small cell lung cancer, about half the CTCs within clusters isolated by the Cluster-Chip in metastatic breast cancer were proliferating.

Similar to single migratory mesenchymal-like CTCs, CTC clusters appear to be enriched for mesenchymal markers (82, 86). It is postulated that CTC clusters may arise from oligoclonal tumor cell groups with high metastatic potential and that EMT in the CTC clusters may be mediated through TGFβ signaling by platelets attached to these clusters (refs. 82, 87, 88; Fig. 1).

Beyond CTCs, an increasing interest has been expressed in the role of circulating stromal cells and macrophages in metastatic progression. In mouse models, tumor cells entering the circulation together with primary tumor-derived stromal cells have a survival advantage compared with single CTCs and are more efficient in forming lung metastases (89). Similarly, circulating cancer-associated macrophage-like cells from patients with metastatic breast, prostate, and pancreatic cancers can disseminate into the circulation and interact with CTCs (90), with some observed to migrate bound to CTCs, potentially facilitating distant colonization and neovascularization. CTC clusters may express markers associated with platelet transcripts and/or tissue-derived macrophages, but not T/B/natural killer (NK) cells (85). Refinements in our ability to interrogate CTCs and associated cells may enable more rapid clinical development of targeted agents that can affect the metastatic cascade.

First recognized more than 20 years ago (91), plasma ctDNA species are identifiable by the presence of pathognomonic or previously characterized molecular alterations in corresponding tumor tissue (i.e., single nucleotide, copy number, structural, and methylation variants) and thus afford tremendous specificity (92–94). Recent advances in our understanding of the biologic properties and clinical associations of ctDNA, as well as the analytic platforms for its detection, have provided evidence that this class of biomarker may also enable a level of sensitivity suitable for noninvasive tumor monitoring.

As with CTCs, proposed clinical applications of ctDNA segregate broadly into two categories: profiling, noninvasive characterization of tumor molecular features, and quantitation, where ctDNA levels serve as a surrogate of tumor burden (92). For both categories, clinical utility will depend on reliable detection of ctDNA when it is present (analytic sensitivity), as well as the proportion of patients for whom ctDNA should be detectable (clinical sensitivity). Although ctDNA can be detected across several tumor types and generally correlates with tumor stage, absolute ctDNA levels vary widely within each subpopulation (95). Detection of ctDNA is further challenged by the high background levels of circulating wild-type DNA observed in individuals with and without cancer (96). In early-stage disease (but also in some metastatic cases), ctDNA may represent an exceedingly rare subpopulation within total cell-free DNA, at levels corresponding to one genome equivalent in 5 mL of plasma (∼0.01% allele fraction), and may be undetectable in plasma volumes typically sampled (95, 97, 98). Although incompletely understood, ctDNA levels may vary according to tumor burden, anatomic proximity to vasculature, and biologic features, including apoptotic rate and metastatic potential.

Given a low signal-to-noise ratio, ctDNA detection methods must account for multiple sources of variability to have the robustness desired for clinical use. Analytic variability can arise from inefficient ctDNA recovery during sample preparation, intrinsic error rates for PCR and sequencing which exceed the lower range of ctDNA abundance, and biases in enrichment of genomic regions for analysis (97). Preanalytic variability can affect levels of background wild-type DNA due to lysis of white blood cells during plasma preparation (99), which has prompted development of standardized protocols incorporating use of specialized preservative-containing tubes (e.g., Streck Cell-Free DNA BCT; ref. 100). Detection methods should also accommodate the presence of ctDNA predominantly as 160- to 180-bp fragments, consistent with the nucleosomal pattern of DNA fragmentation arising from tumor cell apoptosis, the rate of which is likely to be the key driver of ctDNA levels; however, it has been shown that a high portion of ctDNA fragments are <100 bp and that optimal detection would then require the use of primers that target amplicons <100 bp (60 bp may be best; refs. 101–103).

The most challenging source of variability, however, comes from tumor heterogeneity. As previously discussed, tumors are characterized by marked spatial heterogeneity resulting from clonal evolution of cells harboring tumor-initiating molecular alterations (the “trunk”) to subclones with additional mutations (the “branches”; ref. 4). If a “branch” mutation is selected to detect ctDNA, an absent or low-level signal may not accurately represent the overall level of ctDNA, with potential implications for clinical utility. As an example, a low level of circulating KRAS-mutant DNA in a patient with metastatic colorectal cancer could have a concordant result in tissue but could alternatively represent a rare subclone that would not have been detected by conventional tissue KRAS testing. Withholding anti-EGFR therapy for this patient might be appropriate in the former case but would be controversial in the latter. Moreover, for applications in which ctDNA is assessed longitudinally, an added challenge will be temporal heterogeneity, whereby tumor molecular profiles evolve with emergence and disappearance of dominant subclones due to the selective pressure of treatment (104–106). For broad applicability, ctDNA detection platforms should not only have high analytic sensitivity but also sufficient genomic coverage to identify a tumor with multiple molecular markers (for redundancy and inclusion of “trunk” mutations) and to anticipate molecular alterations expected with tumor evolution.

Several methods have been developed to detect ctDNA, with the predominant platforms at present based on digital PCR and NGS. Comparisons of clinical sensitivity across studies are challenging due to variability in methods, the number and type of targeted molecular alterations, tumor type, tumor stage, and preselection of patients (Table 4). With the notable exception of the studies by Bettegowda and colleagues (95) and Douillard and colleagues (107), published studies have been limited by small sample sizes. Nevertheless, a number of themes emerge. First, PCR-based approaches have very high sensitivity for ctDNA but are limited in the number of foci that can be assessed and, consequently, the addressable proportion of each population (compare the tested populations and the populations evaluable for sensitivity; Table 4). This limitation can be addressed by first identifying patient-specific molecular alterations in tumor tissue and then developing customized ctDNA assays [e.g., PCR assays for single-nucleotide variants (SNV; refs. 95, 108) or structural variants (personalized analysis of rearranged ends, PARE (ref. 109)]. However, this approach may be limited by practicality and cost considerations. Second, newer sequencing-based platforms that account for sequencing error rate and PCR errors during library preparation [e.g. SafeSeq (ref. 110), TAm-Seq (ref. 111), CAPP-Seq (ref. 97), Ampli-Seq (ref. 112, the first published study to demonstrate the feasibility of performing deep-coverage NGS in breast cancer for the detection of mutations in hot spot regions of 50 genes in an ISO-certified laboratory), and others (113)] are achieving analytic sensitivities on par with PCR while maintaining broader genomic coverage. Finally, ctDNA detection with state-of-the-art techniques remains consistently lower for early-stage disease than for metastatic disease.

The first area in which ctDNA will be proven to have clinical utility is in noninvasive profiling for the presence of actionable mutations. Several studies have now demonstrated high concordance for selected actionable mutations between paired tumor and plasma specimens, particularly for metastatic disease in breast (95, 105, 112, 114–116), colorectal (95, 108, 117–119), and non–small cell lung cancer (refs. 97, 113, 120; Table 4). Larger, prospective studies with standardized analytic methods should be conducted to validate concordance in each tumor type, enabling a more precise understanding of false-negative and false-positive rates. For metastatic disease patients who have tumors that are difficult to biopsy, who have contraindications to biopsy, or who have tumors that are traditionally challenging to diagnose by conventional means (e.g., cholangiocarcinoma), a validated ctDNA assay could have clinical utility in the near term as an “alternative to tissue biopsy.”

Discordances in the molecular profiles between ctDNA and tumor tissue specimens may reflect underappreciated tumor heterogeneity within and between tumor foci. Serial ctDNA assessment may also detect the process of tumor evolution (105, 121), where the appearance of new molecular alterations on treatment may herald the emergence of resistance and potentially also predictive markers for different therapies. This phenomenon has been best described for colorectal cancer, in which ctDNA obtained after treatment of KRAS wild-type tumors with anti-EGFR therapies has demonstrated new molecular alterations that plausibly confer resistance, including KRAS, NRAS, BRAF, and EGFR mutations (96, 104) as well as MET amplification (122). Ultimately, however, prospective studies will be needed to demonstrate that treatment strategies guided by unique information provided by ctDNA yield superior clinical outcomes when compared with tissue-based approaches.

Reliable and sensitive methods to detect and quantitate ctDNA may enable noninvasive disease monitoring in a manner analogous to BCR-ABL testing in chronic myeloid leukemia or HIV viral testing. Case studies in breast, colorectal, and non–small cell lung cancer have suggested that ctDNA dynamics can provide an early indicator of tumor response, which could help optimize neoadjuvant therapy or treatment of metastatic disease. Ineffective therapies could be halted with the appearance of resistance, avoiding unnecessary toxicity. Posttreatment ctDNA levels may also be useful in detecting previously unrecognized residual disease following definitive therapy. Anecdotal evidence has been reported to support this concept (7, 17, 123, 124), but large, prospective studies will be needed to demonstrate the prognostic value of residual disease detected by ctDNA. Tie and colleagues (125) have reported preliminary results of a prospective trial in stage II colon cancer evaluating the relationship of postoperative ctDNA levels with tumor recurrence. At a median follow-up of 507 days, recurrence rates were >10-fold higher in patients with detectable postoperative ctDNA (5 of 6, 88% with detectable ctDNA vs. 5 of 72, 7% without detectable ctDNA). Finally, early detection of cancer is a tantalizing application for ctDNA. At present, the relatively poor clinical sensitivity of ctDNA for early-stage disease (Table 4 and ref. 95) would result in high proportions of false negatives and, more significantly, would limit the degree of stage migration, which is critically important for screening programs to affect patient outcome. Further development of ctDNA platforms will be required before the challenges of cancer screening should be considered.

A major reason for treatment failures is our inability to monitor tumor evolution and adapt treatment accordingly. Identifying tumor recurrence at an earlier time point does not improve clinical outcome if an effective therapy is not selected or available. Liquid biopsy technologies are potentially important advances in this regard, with CTCs and ctDNA expected to play complementary roles, on the basis of their relative strengths and limitations (Table 5). Plasma ctDNA assays (disease-specific, treatment-specific, or personalized) may prove more useful for monitoring disease burden and limited molecular profiling. Once increased disease burden is recognized, then CTC analysis for comprehensive characterization of tumor DNA, RNA, and/or protein levels, including their co-localization, in known residual cancer cells may help to optimize therapy selection (92). It is also quite likely that CTCs may be particularly useful ex vivo, incorporated into functional studies using CTC cultures, mouse xenografts, or real-time in vitro assays for drug sensitivity evaluation.

Significant challenges remain, particularly with respect to analytic and clinical sensitivity. Adoption of these tools into routine clinical practice will necessitate rigorous demonstration of analytic validity, clinical validity, and, most importantly, clinical utility. One consideration in population screening is that 11% to 19% of patients with benign inflammatory conditions (e.g., Crohn disease) have small numbers of morphologically benign circulating epithelial cells detectable, which could potentially give a false-positive CTC result (126). Another risk is the detection of clinically irrelevant molecular changes due to the high sensitivity of the methods. Therefore, large annotated datasets and bioinformatic tools will be needed to distinguish potentially important genomic aberrations from noise. Moreover, only clinical studies will provide evidence about whether a genomic aberration detected in blood can predict benefit from a specific targeted agent. Although most efforts are currently focused on testing liquid biopsy in the metastatic setting, we expect that future studies will evaluate its role in the early disease setting or even as a potential tool to assist early cancer diagnosis.

M. Ignatiadis was supported by Fonds de la Recherche Scientifique (FNRS), the Breast Cancer Research Foundation (BCRP), the MEDIC foundation, and Les Amis de l'Institut Bordet.