Review Article | Published:

Current and future perspectives of liquid biopsies in genomics-driven oncology


Precision oncology seeks to leverage molecular information about cancer to improve patient outcomes. Tissue biopsy samples are widely used to characterize tumours but are limited by constraints on sampling frequency and their incomplete representation of the entire tumour bulk. Now, attention is turning to minimally invasive liquid biopsies, which enable analysis of tumour components (including circulating tumour cells and circulating tumour DNA) in bodily fluids such as blood. The potential of liquid biopsies is highlighted by studies that show they can track the evolutionary dynamics and heterogeneity of tumours and can detect very early emergence of therapy resistance, residual disease and recurrence. However, the analytical validity and clinical utility of liquid biopsies must be rigorously demonstrated before this potential can be realized.


The key objective of precision oncology is to improve diagnosis and treatment of cancer1,2. To this end, a variety of genomic and other molecular analyses can be applied to tumour material to help identify known predictive markers to guide the selection of treatment, derive a molecular subtype classification that might enable estimation of the prognosis, characterize somatic alterations involved in tumour progression, detect disrupted pathways and identify molecular discriminants of metastatic disease3. Although a range of next-generation sequencing (NGS)-based approaches have been used to characterize tumour genomes in detail4, a more accurate classification of tumour types can be achieved through comprehensive, multiparameter analyses. For example, The Cancer Genome Atlas (TCGA) research network generated comprehensive molecular profiles at the DNA, RNA, protein and epigenetic levels for hundreds of tumours5. These multiparameter analyses resulted in an improved understanding of molecular aberrations and their functional roles across tumour types and identified novel tumour subtypes5. Importantly, these efforts led to the identification of new druggable targets, which is a prerequisite for realizing the promise of precision medicine. However, access to tumour material for molecular profiling usually depends on invasive procedures that are not always feasible and do not lend themselves to serial monitoring of tumour genotypes.

Therefore, the focus of precision oncology is increasingly turning to liquid biopsies because they are non-invasive and can be repeated at multiple time points, which facilitates the monitoring of disease courses. Indeed, attempts are now being made to use them for early detection of cancer6,7,8. The term liquid biopsy was first used to describe methods that can derive the same diagnostic information from a blood sample that is typically derived from a tissue biopsy sample9. In oncology, the term is used in a broad sense to refer to the sampling and analysis of analytes from various biological fluids, mostly blood but also other fairly easily accessible fluids such as urine, ascites or pleural effusions10,11. Analytes in the peripheral blood include circulating tumour cells (CTCs); circulating cell-free DNA (cfDNA), which in patients with cancer contains circulating tumour DNA (ctDNA); circulating cell-free RNA (cfRNA), which contains predominantly small RNAs but also mRNAs; circulating extracellular vesicles (EVs), such as exosomes; tumour-educated platelets (TEPs); proteins; and metabolites10,11,12,13,14,15 (Table 1). Together, these analytes have the potential to provide information about features of primary tumours or metastases that are usually obtained by pathologists (Table 2). In addition to the information about genomic mutations and copy number alterations that is usually obtained from CTCs or ctDNA10, liquid biopsies are increasingly being used to generate information about the transcriptome16, the epigenome17, the proteome18 and the metabolome19. Moreover, novel artificial-intelligence-based bioinformatics tools are beginning to move the liquid biopsy field towards truly multiparametric analyses20.

Table 1 Comparison of the analytes found in solid and liquid biopsy samples
Table 2 Histopathological features of solid biopsy samples that can be recapitulated in liquid biopsy samples

In this Review, we first summarize the few examples of liquid biopsy applications that have been approved for clinical care. We then discuss the current evidence for analytical and clinical validity of ctDNA, which is currently one of the most intensively studied analytes found in liquid biopsy samples. This example illustrates not only the great potential of liquid biopsies but also the many challenges that must be overcome before they can be widely implemented for precision medicine. We conclude by reflecting on how these challenges might be met and, in particular, on the likely contributions of multianalyte approaches. Technologies for obtaining and interrogating analytes and the applications of liquid biopsies have been the subjects of recent comprehensive reviews10,11,12,13,14,15 and are not discussed in detail here.

Liquid biopsies in clinical care

The Evaluation of Genomic Applications in Practice and Prevention (EGAPP) initiative established three criteria as requisites for adoption of a tumour biomarker test in clinical care21: analytical validity, which measures the accuracy, reliability and reproducibility of a test; clinical validity, which assesses the ability of a test to divide a population into separate groups with significantly different clinical outcomes; and clinical utility, which evaluates whether outcomes are improved for patients who received the test compared with those who did not21,22. However, regulation of tumour biomarker tests is complex23, and even biomarker tests approved by the US Food and Drug Administration (FDA) do not necessarily need to establish clinical utility23. Furthermore, many commercially available tumour biomarker tests are never submitted to the FDA for review but are instead marketed as laboratory-developed tests (LDTs). Although laboratories performing LDTs must be certified and many LDTs function reliably, these tests are not subject to the same standards that are applied by the FDA to other diagnostic tests and their analytical validity and clinical utility are not always rigorously reviewed23. This discrepancy is exemplified by tests based on plasma ctDNA and CTCs, which have the largest reported evidence base for utility as biomarkers in precision oncology14 (Table 1).

The CellSearch system (Menarini Silicon Biosystems)24,25,26 is an FDA-approved CTC detector for patients with metastatic breast, prostate or colorectal cancer (CRC). It enriches for cells that express epithelial cell adhesion molecules (EpCAMs) but lack the leukocyte common antigen CD45. In patients with metastatic breast or prostate cancer, the presence of ≥5 CTCs per 7.5 ml blood was found to be a strong predictor for reduced progression-free survival (PFS) and <5 CTCs per 7.5 ml was predictive of improved overall survival (OS)27,28. For metastatic CRC, the predictive thresholds are ≥3 CTCs per 7.5 ml for PFS and <3 CTCs per 7.5 ml blood for OS29. The analytical and clinical validity of CTC quantification for prognostication of patients with metastatic breast, prostate or colon cancer, as well as patients with non-metastatic breast cancer treated with neoadjuvant chemotherapy, have been verified in studies and/or pooled meta-analyses involving thousands of patients30,31,32,33,34,35. However, CellSearch CTC enumeration has not become a widely adopted test for any of these tumour entities, largely because it has not been demonstrated to have clinical utility. For example, studies investigating whether patients with rising CTC numbers might benefit from an early switch to another chemotherapy did not result in prolonged survival36. Hence, CTC enumeration represents an established prognostic, but not a predictive, biomarker.

By contrast, clinical utility has been demonstrated for two FDA-approved cfDNA-based tests: the cobas EGFR Mutation Test v2 (Roche Molecular Diagnostics), which detects EGFR mutations37 in plasma cfDNA from patients with lung cancer38, and Epi proColon (Epigenomics AG), which reports on the methylation status of the SEPT9 promoter in plasma cfDNA from patients undergoing screening for CRC39. Hypermethylation has been associated with the occurrence of CRC, and the clinical utility of Epi proColon has been evaluated in several studies that encompass several thousand clinical samples and in recent meta-analyses that described the test as an effective marker for blood-based CRC detection, albeit with a low sensitivity for early stage I CRC40,41. In patients with non-small-cell lung cancer (NSCLC), the cobas EGFR Mutation Test v2 detects activating mutations in EGFR that result in increased sensitivity to tyrosine kinase inhibitors (TKIs); these mutations, which include deletion of exon 19 and L858R substitution in exon 21, are strong predictors for increased PFS in patients who receive EGFR-targeted therapies, such as the first-generation drug erlotinib (Tarceva)42. The test can also be used to detect the T790M EGFR mutation in patients whose cancer has progressed after treatment with EGFR-targeted TKIs; this mutation leads to therapy resistance and its detection indicates that treatment can be switched from erlotinib to the third-generation drug osimertinib. The demonstrated predictive value of this test has led several professional societies to recommend the use of plasma cfDNA to identify EGFR mutations in patients with NSCLC43. Importantly, the clinical utility of the cobas EGFR Mutation Test v2 and Epi proColon emphasized the potential of ctDNA as a biomarker.

Evaluating ctDNA as a biomarker

The potential of ctDNA as a biomarker in precision oncology

Studies show that ctDNA is likely derived from apoptotic cells44,45 and may, therefore, represent a particular subtype of tumour cells. Nonetheless, evidence indicates that ctDNA provides a comprehensive view of the tumour genome as it reflects DNA released from multiple tumour regions46,47,48 or different tumour foci17,49,50. Indeed, ctDNA studies have detected somatic mutations that have been missed in corresponding tissue samples46,47,51,52. Furthermore, deep sequencing of genomic regions spanning thousands of bases have enabled assessment of intratumour heterogeneity53 and detection of subclonal mutations52,54,55,56 and have uncovered specific molecular subtypes with distinct genomic signatures57,58. By contrast, a large number of single CTCs or several solid biopsies would be needed to obtain the same level of information. A particular advantage of ctDNA as a biomarker compared with CTCs is that it might be informative about tumour mass, which reflects the extent of the disease and hence its response to a given therapy. A linear relationship between tumour volume and ctDNA plasma variant allele frequency (VAF) was reported in NSCLC and high-grade serous ovarian cancer (HGSOC)56,59,60, and another study found a high correlation between pretreatment ctDNA concentration and metabolic tumour volume in patients with lung cancer61. Furthermore, ctDNA has been shown to have prognostic value. For example, the presence of ctDNA after curative-intent surgery in patients with localized disease and/or chemotherapy was demonstrated to be a powerful prognostic marker associated with recurrence and poor outcome in patients with colon62, ovarian60 and lung cancer61. Identification of tumour-specific changes in ctDNA from blood has also enabled detection of minimal residual disease and prediction of recurrence with lead times of several months in patients with colon63, breast64 and lung cancer59. Finally, ctDNA analysis enables the detection of resistance markers, such as KRAS mutations in patients with CRC receiving anti-EGFR therapy65,66,67,68; increasing VAFs for genes such as PIK3CA, MED1 or EGFR in patients with breast cancer treated with various therapies69,70; and the EGFR T790M resistance mutation in patients with lung cancer treated with EGFR-targeted TKIs37,51. Together, these features of ctDNA make it a promising analyte for precision oncology applications. However, ctDNA also has some substantial limitations (Box 1), such as its frequently low levels in plasma and confounding factors resulting from ageing or validation issues.

Plasma often contains low levels of ctDNA

The ability to detect mutations in plasma cfDNA correlates with the tumour burden. The Tracking Non-Small-Cell Lung Cancer Evolution Through Therapy (TRACERx) study predicted that primary tumour burdens of 1 cm3, 10 cm3 or 100 cm3 (that is, 9.4 × 107 cells, 3.02 × 108 cells and ~109 cells, respectively)71 would result in mean clonal plasma VAFs of 0.006%, 0.1% and 1.3%, respectively59. However, ctDNA levels may vary greatly between patients, even those with the same tumour type54 (Fig. 1). Indeed, even in metastatic disease, a substantial proportion of patients present with unexpectedly low fractions of ctDNA54,72,73. Clearly, biological factors other than tumour burden affect the release of tumour DNA, as further demonstrated by a study of HGSOC in which patients with relapsed disease had pretreatment TP53 VAFs of 0.04% per cm3 tumour compared with VAFs of 0.0008% per cm3 in newly diagnosed untreated patients60. The number of available template molecules dictates whether such low VAFs are detectable (Fig. 1). On average, 1 ml plasma from a patient with cancer contains approximately 1,500 diploid genome equivalents (GE) (~10 ng DNA)74 (Fig. 2a), with considerably higher amounts often observed in patients with metastatic cancer. A typical 10 ml blood draw yields on average 4 ml plasma containing 6,000 GE (12 × 103 molecules per region or gene), which implies a theoretical sensitivity limit of ~0.01% (that is, the ability to detect 1 fragment in 12,000 copies). If the VAF of ctDNA corresponds to 0.1%, there are on average just six molecules per tube carrying the respective mutation, which may be affected by stochastic sampling (Fig. 2b). Thus, the reliable, accurate and reproducible detection of these mutations poses a considerable challenge and requires specialized technologies.

Fig. 1: Factors influencing the sensitivity of a plasma cell-free DNA test.

Technical and physiological factors limit broad implementation of liquid biopsies in clinical practice. The achievable sensitivity of a circulating cell-free DNA (cfDNA) test on plasma is dependent on several factors, including sampling volume, the number of input molecules, the tumour fraction and the analytical sensitivity of the method being used. GE, genomic equivalents.

Fig. 2: The amount of input cell-free DNA affects the ability to detect rare variants.

a | The graph on the left shows the relationship between the amount of input DNA (in nanograms) and the number of template molecules in terms of diploid genome equivalents (GE). The graph on the right shows the relationship between the number of input molecules (copies) and the probability that 10 mutated copies will be detected in a total of 1,500, 6,000 or 10,000 copies at different variant allele frequencies (VAFs). For a reliable detection of a mutation with a VAF of 0.1% (10 mutated copies), at least 10,000 template molecules (5,000 GE or 33 ng of input DNA) are necessary. b | Simulations of repeated sampling of samples containing 6,000 copies (~20 ng input DNA) with VAFs of 0.01%, 0.1% and 0.5% using a Poisson distribution. At a VAF of 0.01%, the vast majority of samples will not contain any mutated copies when sampling 6,000 copies. By contrast, at a VAF of 0.5%, 99.5% of samples contain 10 or more mutated copies. c | Comparison of single locus assays (left) and gene panel profiling (right). At low tumour fractions, the likelihood of capturing a mutated copy of at least one gene is greatly increased by comprehensive profiling using gene panels.

Approaches for detecting low levels of ctDNA in plasma

The number of detectable molecules might be increased by substantially increasing the sample volume (that is, drawing more blood), but this approach is usually not clinically feasible in patients with advanced disease. Alternatively, the low number of input molecules could be compensated for by simultaneously testing a large number of mutations and establishing the minimum number of detected mutations required to classify a plasma sample as tumour positive56,59,74 (Fig. 2c). For instance, the TRACERx study used multiplex-PCR assay panels to target a median of 18 single-nucleotide variants (SNVs) per patient and implemented a threshold of 2 detected SNVs for a sample to be deemed ctDNA positive59.

Recent developments in ctDNA techniques have focused on increasing the analytical sensitivity of sequencing approaches, which may be affected by errors introduced during DNA preparation or during sequencing itself. For example, using baits (that is, oligonucleotides that hybridize to the desired DNA) to enrich for selected genomic regions during DNA preparation can cause oxidative damage to the DNA, resulting in sequence alterations75Sequencing sensitivity is also affected by the accuracy (that is, the number of errors made per base pair sequenced)76 and the coverage (that is, the number of DNA molecules that are independently sequenced) of the chosen sequencing method. The effects of errors caused by these factors can be mitigated (and sensitivity improved) by using molecular barcodes called unique molecular identifiers (UMIs)56,74,76,77,78,79; these unique tags facilitate bioinformatic alignment of sequences derived from the same DNA fragment and enable errors to be easily identified and excluded from subsequent analyses80. For instance, the cancer personalized profiling by deep sequencing (CAPP-Seq) method assesses plasma ctDNA for alterations in 139 genes that are recurrently mutated in NSCLC and achieved a sensitivity of 85% for disease monitoring and minimal residual disease detection56. When CAPP-Seq was combined with a computational error correction approach called integrated digital error suppression (iDES), the sensitivity improved threefold74. Targeted error correction sequencing (TEC-Seq) is a similar UMI-based approach that assays ctDNA for 58 genes frequently mutated in colorectal, lung, ovarian and breast cancers, among others, and can detect at least 1 mutation in more than 75% of patients79.

Other attempts to increase sensitivity have focused on enriching for ctDNA fragments on the basis of their length, which is reported to be shorter (132–145 bp) than cfDNA from normal cells (~166 bp)81,82,83,84. Indeed, a recent study reported that ctDNA could be enriched by excising appropriately sized DNA bands from plasma DNA electrophoresed on polyacrylamide gels82. However, targeting single-stranded cfDNA, which is usually shorter than double-stranded DNA, did not result in an enrichment of tumour DNA fragments85,86. Furthermore, an increased frequency of dinucleosomal DNA fragments with lengths >320 bp was reported in patients with CRC or breast cancer and high ctDNA concentrations72,73. A recent study suggests that DNA fragment length may vary according to nucleosome accessibility in the tissue of origin; specifically, placental cells were shown to have higher nucleosome accessibility than white blood cells, which exposes different positions in the nucleosome core to nuclease activity and possibly explains why fetal DNA in the plasma of pregnant females is shorter than other cfDNA87. If similar mechanisms are in operation in cancer cells, the measurement of DNA fragment lengths could have the potential to distinguish between the various cell sources that contribute DNA to the circulation. Hence, further work is required to determine whether particular cfDNA subpopulations can be selected on the basis of possible size differences.

Constraints imposed by ageing

Despite these improved methods for detecting low VAF mutations, several limitations are imposed on the utility of ctDNA as a biomarker by the biology of ageing cells. An important confounding factor is benign somatic heterogeneity, which is the accumulation of somatic mutations in non-cancerous lesions during ageing. Mutations in adult stem cells, which can be propagated to daughter cells during self-renewal, may occur at a rate of ~36 mutations per year88 and have a large effect on the mutational load of tissues89. If a mutation occurs in a cancer driver gene, it can confer a fitness advantage on the stem cell and cause it to undergo a clonal expansion90, which can be detected in ctDNA and misattributed to tumorigenesis. Clonal expansions in the haematopoietic system have been extensively documented91,92,93, and genes known to be frequently altered in clonal haematopoiesis have already been reported to cause false-positive genotyping results in plasma94,95,96. Clonal expansions in solid tissues are more difficult to detect but have been reported97,98,99. In particular, TP53 mutations that closely resemble those observed in cancers have been discovered in non-cancerous tissues100, and these have the potential to be identified in plasma cfDNA from individuals without cancer100,101. Hence, detection of a mutation in a cancer driver gene in ctDNA cannot be equated with proof for neoplasia.

Pre-analytical, analytical and clinical validation

The concordance and clinical validity of commercial ctDNA tests are currently under debate. A recent study that evaluated the analytical and clinical validity of the Guardant360 (Guardant Health) ctDNA assay on 10,593 consecutive patients found a high concordance with orthogonal clinical plasma-based and tissue-based genotyping methods and concluded that the assay was clinically applicable102. However, another recent study reported incongruent results for plasma samples sent to two different commercial ctDNA tests, Guardant360 and PlasmaSELECT (Personal Genome Diagnostics)103. Complete concordance between the 2 assays was observed in only 9 of 34 (35%) evaluable samples, of which only 3 had concordant overlapping mutational profiles; no mutations were detected by either assay for the remaining 6 samples103. These data raise questions about the analytical and clinical validity of these tests. In fact, a recent joint review by the American Society of Clinical Oncology and the College of American Pathologists found insufficient evidence of clinical validity and utility for the majority of ctDNA assays in advanced cancer and no evidence of clinical utility and little evidence of clinical validity of ctDNA assays in early-stage cancer, treatment monitoring or residual disease detection104. Indeed, there is currently a lack of multicentre studies with sufficiently large patient numbers to establish both analytical and clinical validity of ctDNA. Hence, at present, ctDNA assays are best used within carefully designed clinical trials104.

Furthermore, standard operating procedures and guidelines are needed. The European Committee for Standardization (CEN) has already published standards and a technical specification for recommended handling, documentation and processing of blood specimens intended for cfDNA analysis (CEN/TS 16835-3:2015). Standardization of generic pre-analytical procedures for in vitro diagnostics for personalized medicine (SPIDIA4P) focuses on establishing pre-analytical workflows to support the processes of European (CEN) and international standardization organizations (the International Organization for Standardization (ISO)). International networks, such as CANCER-ID15 and BloodPAC105, are working on standardization and protocols for liquid biopsy approaches. Comprehensive validation, round robin tests and external quality assessment by established standards (for example, according to ISO 15189 for a medical laboratory) are required before NGS-based liquid biopsy technologies can be routinely used in clinical practice.

Emerging liquid biopsy analytes

Although current liquid biopsy-based tests fail to meet the needs of precision oncology, expanding the range of information extractable from ctDNA and CTCs and the range of analytes examined might help them to realize their full potential. Analytes such as EVs and cfRNAs (and microRNA (miRNA) in particular) are the best candidates to advance the field in the near future106; TEPs107,108, metabolites such as branched-chain amino acids (BCAAs) and proteins19,109 are likely to contribute in the longer term (Table 3). Implementation of novel multiparameter strategies to combine information from multiple sources will play a key part in establishing liquid biopsies in the clinic.

Table 3 Challenges and possible solutions to the use of various liquid biopsy components

Tissue and cancer-specific DNA methylation patterns

Methylation of CpG sites is an important epigenetic regulator of gene expression and tissue differentiation. The International Human Epigenome Consortium (IHEC) and the Blueprint Epigenome consortium have generated large-scale repositories that contain reference methylomes for multiple tissue types, and information about cancer-type-specific methylation changes is increasingly available in the literature110. Methylation patterns can be used to map the tissue of origin of cfDNA or to directly detect cancer. Changes in DNA methylation occur early in carcinogenesis111, and methylation biomarkers therefore represent an alternative to sequence alterations112. In fact, two studies recently suggested that tumour-specific methylation changes may be detectable in the peripheral blood with a lead time to clinical diagnosis of 2 years in breast and ovarian cancer113,114.

Several technical developments, such as methyl-BEAMing115 and single-cell reduced representation bisulfite sequencing116, have improved data resolution and reduced the amount of template DNA required for methylation detection. Resolution can be further improved by using stretches of several CpG sites adjacent to the tissue-specific methylation marker117, and >140,000 blocks of tightly coupled CpG sites that have co-regulated epigenetic statuses were recently defined116.

To date, cfDNA is the liquid biopsy analyte most extensively subjected to methylation analyses. In principle, CTCs are also amenable to methylation studies, but because of the technical challenges of CTC isolation and single-cell methylation analysis118, combined with the low number of CTCs available for analysis, only a few studies have been published, and these have largely focused on the analysis of a single selected region119. However, as single-cell multiomics technologies evolve120, CTCs may become amenable to more comprehensive analyses.

Mapping nucleosome positions

As cfDNA is preferentially released from apoptotic cells, it circulates in the form of nucleosome-protected, mostly mononucleosomal, DNA34,35. Hence, deep sequencing of cfDNA allows for analysis of peak-to-peak spacing of nucleosome positioning and the generation of maps of nucleosome occupancy. Similar to methylation changes, nucleosome positions vary between cell types and can be used to track cfDNA tissues of origin44. Differences in nucleosome patterns between transcribed and silent genes can be detected as differences in the sequencing depth pattern at transcription start sites45, which enables the gene expression profile of cells releasing DNA into the circulation to be directly inferred from nucleosome positioning45. These achievements were made feasible by novel single-stranded DNA library protocols44 and innovative bioinformatics tools, including machine learning for correct classification of nucleosome patterns45.

Defining microRNA signatures

Circulating RNAs have been studied by various approaches, such as RNA sequencing (RNA-seq)121,122, quantitative PCR (qPCR)123 and microarrays124. miRNAs are of particular interest because they are surprisingly stable in plasma or serum125,126, which has been attributed to their containment and preservation in EVs127. They can occur in virtually any body fluid and, in addition to EVs, they have been detected in apoptotic bodies, high-density lipoprotein structures and complexes with Argonaute proteins125,126. There are multiple ongoing efforts to use blood-based miRNA signatures as diagnostic tools. For example, low-dose computed tomography (LDCT) enables early detection of lung cancer in high-risk individuals. Two large-scale studies assessed whether circulating miRNAs can mitigate LDCT’s high false-positive rates. One study, the Continuous Observation of Smoking Subjects (COSMOS) programme, found that a signature of 13 miRNAs had a sensitivity and negative predictive value comparable to LDCT alone and may therefore be suitable for lung cancer screening in high-risk individuals128. The randomized Multicentre Italian Lung Detection (MILD) trial developed a plasma miRNA signature classifier that, when combined with LDCT, reduced the LDCT false-positive rate fivefold129. Hence, the development of miRNA signatures or the detection of long non-coding RNA (lncRNA) in biofluids for the diagnosis of a variety of different cancer types is an active field of research130,131,132.

Transcriptomic and genomic analysis of extracellular vesicles

EVs transport nucleic acids and proteins as a function of their role in intercellular and interorganismal communication133. Depending on their origin, up to 100 proteins and ~10 kb of nucleotides can be packed within an exosome134. A recent analysis of RNA contained in EVs determined it to comprise miRNA (40.4%), piwi-interacting RNA (40%), pseudo-genes (3.7%), lncRNA (2.4%), tRNA (2.1%) and mRNA (2.1%)135. Other studies suggest that EVs may contribute to elucidating the origin of cancer, early detection and establishing the prognosis of patients with different cancers136,137,138. A current focus is the presence of adhesion molecules, such as integrins, on tumour-derived exosomes, which were shown to be informative about organ-specific metastasis139,140. Exosome-derived DNA (exoDNA) may be representative of the entire genome and the mutational status of parental tumour cells and could, therefore, be of relevance for genomic analysis141. Unlike ctDNA, exosomes arise from viable cancer cells, and these two analytes may therefore reflect different aspects of tumour biology. Indeed, a study conducted with patients with localized pancreatic ductal adenocarcinoma found a higher percentage of detectable KRAS mutations in exoDNA than in cfDNA142. However, mutant KRAS exoDNA was also detected in 20% of healthy controls142. Cells release hundreds of thousands of exosomes per day. Thus, determination of how exactly proteins and miRNAs are selected and loaded to exosomes and how trafficking is regulated will be crucial for clinical applications143. Novel chip-based enrichment platforms could facilitate high-throughput isolation and analysis of EVs144, including generation of validated markers for EV classification and detailed characterization of EV subfamilies for diagnostic purposes143,145.

Tumour-educated platelets

Tumour cells may transfer tumour-associated biomolecules to platelets, a process referred to as ‘education’107,108. A study of localized and advanced metastatic cancer, comprising six tumour types (NSCLC, CRC, glioblastoma, pancreatic cancer, hepatobiliary cancer and breast cancer), showed that TEP RNA-seq profiles from all 228 cancer samples differed from those from healthy individuals. In fact, the TEP analyses enabled discrimination of patients with localized and metastasized cancer from healthy individuals with 84–96% accuracy, established the organ of origin of the primary tumour with 71% accuracy108 and showed that different molecular tumour subtypes were associated with different platelet profiles. Particle-swarm optimization-enhanced algorithms for the efficient selection of TEP RNA biomarker panels enabled the diagnosis of late-stage NSCLC with an accuracy of 89% (518 cases) and with an accuracy of 81% in locally advanced stage I–III NSCLC (106 cases)107. These initial studies have focused on the utility of TEPs as a non-invasive biomarker for cancer detection and classification. However, their true potential as a biomarker will become clear only with a fuller understanding of the process of platelet education and the extent to which the relative contribution of platelet subpopulations changes during the clinical course of patients with cancer108.


Proteins are required for all processes within cells, and alterations in proteins caused by pathogenic mutations represent the link between genotype and phenotype. Furthermore, as almost all currently available targeted therapies are directed against proteins, measurements of circulating proteins may be valuable for biomarker discovery. To date, only a small number of established protein markers have been used in clinics, and, for most proteins, information about tissue specificity or cancer specificity is largely missing146. Prostate-specific antigen (PSA) testing is a controversial, yet commonly used, method for early detection of prostate cancer147. Tumour markers, such as CA125 or CA19-9, are less frequently used for screening purposes and their use for monitoring purposes is complicated by their long half-lives148. Recent key developments in multiplexed proteomic technologies, such as discovery proteomics to achieve a complete coverage of the proteome, targeted proteomics to acquire a subset of known peptides of interest and multiplexed fragmentation of all peptides to generate comprehensive protein maps for a sample, have transformed mass-spectrometry-based proteomics into a mainstream analytical tool149. Bioinformatics tools, such as machine learning, have enabled the generation of high-quality protein association maps and their organization into a modular proteome of the cell149. In fact, initial attempts to use circulating proteins in combination with other analytes have suggested that these approaches substantially improved liquid-biopsy-based detection of early-stage cancer6,7.


The development of a tumour is frequently accompanied by systemic changes that may influence physiology, particularly global metabolism. As cancer cells seem to have almost unlimited options to rewire their metabolic pathways, circulating metabolite levels may be suitable cancer biomarkers150. A prime example may be altered plasma levels of BCAAs, which could indicate disturbance of metabolic regulation in early-stage tumours. During early development of pancreatic cancer, currently unknown signals may induce breakdown of long-term protein stores, thereby resulting in an increase of BCAAs19. By contrast, NSCLCs may rapidly take up BCAAs, causing decreased BCAA plasma levels109. These data suggest that metabolites have the potential to become important constituents of the circulating analyte portfolio.

Viable CTCs

Recent improvements to single-cell analyses have enabled a shift from enumeration of CTCs to a more detailed analysis of the genomes151,152, transcriptomes153, proteins154 and epigenomes155 they contain. CTCs are either apoptotic or viable13, but only viable CTCs are relevant for forming metastases and are therefore of special interest. EPithelial ImmunoSPOT (EPISPOT) is a functional assay of CTC viability that detects both EpCAM-positive and EpCAM-negative CTCs by capturing and culturing them short term on a membrane coated with antibodies for tumour marker proteins156. An interesting feature of the EPISPOT assay is the option to measure the sensitivity of CTCs to certain drugs14.

CTCs can also provide patient-specific tumour models. The first xenograft CTC assay was derived from primary human luminal breast cancer CTCs and it enabled the identification of metastasis-initiating cells among CTCs157. CTC-derived explants (CDXs) from patients with small-cell lung cancer were proficient for therapy testing and detection of drug resistance mechanisms158. By contrast, CTCs are less prevalent in patients with NSCLC, which makes it more difficult to develop CDXs from these individuals158. Novel protocols have enabled permanent cell lines to be established from CTCs from patients with CRC159. However, in addition to the lack of cost-effective high-throughput approaches to capture their full heterogeneity160, the success rates for both ex vivo CTC cultures and CTC-derived xenografts are low and the time required to generate them often exceeds the lifespan of the donor patient161. Therefore, these approaches are not informative about treatment options of individual patients, and establishment of tumour models for preclinical drug development remains the main current use for CDXs162.

Future challenges for liquid biopsies

Despite recent progress, there are several important remaining challenges to the wider use of liquid biopsies: the composition of analytes in the peripheral blood in health and under particular physiologic conditions such as ageing needs to be better understood; tools need to be developed that enable the effects of the microenvironment and the immunologic response on tumorigenesis to be investigated in liquid biopsy samples; and further improvements to diagnostic tools are required to enable the detection of small amounts of tumour-derived components in the circulation. We argue that an improved understanding of the mechanisms involved in release of liquid biopsy components and adoption of emerging comprehensive multidimensional profiling strategies will be fundamental to solving these challenges.

Understanding the composition of plasma components

The nature and origin of cfDNA, and other plasma analytes, is still a matter of debate. For example, the likely apoptotic origin of cfDNA44,45 implies that ctDNA should be derived largely from dying tumour cells. However, recent ctDNA analyses have identified somatic alterations associated with aggressive, proliferative cancer cell clones49,163. Determining to what extent different tissues contribute to the composition of plasma in both health and disease will be critical to understanding the physiology of DNA release from cells into the circulation. To date, tissue contribution has been studied in healthy individuals, pregnant females and patients with cancer by the analysis of mRNA121,122,127, miRNA164,165, methylation markers116,117,166,167,168 and nucleosome patterns44. However, the majority of existing studies investigated tissue-specific components from only one or a few cell types127,164,165, and only a few studies have tried to establish the origin of plasma cfDNA in a genome-wide and tissue-wide manner (Fig. 3a). These studies have in common that they established white blood cells as the predominant contributor to circulating nucleic acids, albeit their percentage contribution ranged from 42%121 to 90%166,169 (Fig. 3b). Indeed, there is substantial discordance between estimates of tissue contribution generated using different analytical methodologies166,167,170 (Fig. 3b). The fact that a recent study found that erythroid DNA (a tissue source not analysed in previous studies) represented a median of 30.1% of cfDNA from healthy subjects168 reflects how little is known about plasma cfDNA composition.

Fig. 3: Determining the tissue of origin of nucleic acids in plasma.

a | Schematic illustration of the principle of tissue mapping in plasma. Determination of tissue of origin can be done by deconvolution of gene expression signatures obtained from microarray or RNA sequencing (RNA-seq) data, nucleosome occupancy patterns obtained from whole-genome sequencing (WGS) data or methylation patterns obtained from whole-genome bisulfite sequencing (WBS) data. b | The bar charts show the contributions of different tissues to plasma nucleic acids on the basis of data from four studies (study 1 (ref.121), study 2 (ref.170), study 3 (ref.166) and study 4 (ref.167), which all confirmed that white blood cells are the main contributors to the plasma cell-free DNA pool, followed by placenta (in pregnant females) and liver166,167. However, the relative contribution from other tissues differed substantially between studies. With the exception of study 2 (ref.170), in which only one individual was analysed, percentages for each bar correspond to the average values of the different cohorts. Values of n are calculated from all four studies and represent the total number of individuals (healthy individuals, pregnant women or patients with hepatocellular carcinoma (HCC)) for whom a contribution from the respective tissues was reported. Median percentages were calculated from all individuals; values in parentheses refer to the range. TSS, transcription start sites.

Liquid biopsies and immuno-oncology

The interplay between liquid biopsy components and anticancer immunity is a new and fascinating area of research171. Once a CTC leaves the primary tumour, it becomes exposed to a hostile environment with active immunosurveillance mechanisms and it needs to use a variety of immune-escape mechanisms to survive. The best understood CTC immune evasion mechanisms include altered major histocompatibility class (MHC) I presentation, altered expression of natural killer (NK) cell ligands, downregulation of FAS (also known as CD95) or expression of FASL (FAS ligand)171. Furthermore, the immune system likely has an impact on CTCs during some stages of tumour progression, such as dormancy and metastasis171. However, it is difficult to investigate the tumour microenvironment non-invasively; therefore, integrating liquid biopsies into immuno-oncology workflows may be more challenging than for tissue biopsies. Nevertheless, possible liquid biopsy biomarkers are beginning to be identified. For example, programmed cell death 1 (PD1) is a protein on the surface of immune cells that, together with its ligands (such as PD1 ligand 1 (PDL1)), negatively regulates immune response by switching off T cell activation. Under physiological conditions, PD1 protects tissues from damage during immune system stimulation; however, expression of PD1 in tumours prevents the immune system from destroying cancer cells. Antibodies that target and block the PD1–PDL1 pathway result in antitumour immune activity and are an emerging form of cancer therapy. PDL1 expression is usually measured by tissue staining of the tumour and its microenvironment. However, PDL1 expression was recently detected in CTCs with a PDL1-specific antibody172 and in EVs by western blotting173, suggesting that these analytes may potentially be used to stratify patients with cancer undergoing immune checkpoint blockade. Another predictor of response to immunotherapies is tumour mutational burden (TMB); higher mutational counts increase the chance that neoantigens will form, which results in increased responsiveness to immunotherapies174. In fact, a hypermutated state that correlated with response to checkpoint inhibitors could be established from ctDNA when panels consisting of 54–70 genes were used175. In Epstein–Barr virus (EBV)-positive gastric cancer, ctDNA TMB correlated with radiographic response and predicted PFS as well as tissue mutational load176. Moreover, early changes in ctDNA predicted response and PFS176. Another recent study established that a TMB cut-off of ≥16 derived from sequencing 1.1 Mb of coding region could identify patients who derived clinically significant improvements in PFS from atezolizumab in second-line and higher NSCLC177.

Measuring and integrating multiple parameters

A clear long-term goal of liquid biopsy approaches is to increase resolution to enable detection of minimal residual disease and, eventually, early detection of cancer. For some applications, assays using only one of the liquid biopsy analytes may be sufficient. For instance, a recent landmark paper described the analysis of EBV DNA in plasma to screen for early nasopharyngeal carcinoma in asymptomatic individuals8. Each nasopharyngeal carcinoma cell contains ~500 copies of the EBV target sequence, making its detection in the peripheral blood fairly straightforward8. However, successful application of just a single analyte for early cancer detection will likely remain an exception. Diagnostic sensitivity and diagnostic specificity are essential issues in cancer diagnostics, which raises the question of whether comprehensive molecular profiling, as performed by TCGA, would be superior for some applications.

The first attempts at multiparameter analyses focused on utilizing ctDNA and protein biomarkers. A combination of protein biomarkers (CA19-9, TIMP1 and LRG1) with ultra-sensitive KRAS mutation detection was capable of detecting the majority of patients with resectable pancreatic cancers and had significantly higher detection rates than ctDNA testing alone6. Another study also suggested that the presence or absence of CA19-9 in combination with ctDNA and CTCs may improve diagnosis of pancreatic cancer; positive tests for at least two of the three markers increased both sensitivity and specificity178.

A recent strategy for early cancer detection, named CancerSEEK, tested 1,005 participants with clinically detected non-metastatic forms of one of eight common cancer types (ovarian, liver, stomach, pancreatic, oesophageal, colorectal, lung and breast) for combined protein and genetic biomarkers7. The tests were positive for a median of 70% of these eight cancer types. The protein markers were most informative regarding candidate tissues of origin because driver gene mutations (used as genetic biomarkers) are usually not tissue-specific7. However, the sensitivity of the test was high for ovarian cancer but was only moderate for other common tumours, such as breast cancer. A key concern with the test is its achievable positive predictive value (PPV)179. The prevalence of the eight cancers in healthy individuals >64 years of age is ~1%: assuming that CancerSEEK could achieve a 99% sensitivity and 99% specificity, the resulting PPV would be only 50%; that is, 50% of test positives would be false positive179. Moreover, the median sensitivity of CancerSEEK was only 43% for stage I cancers, which would further decrease the effective PPV. Furthermore, although ctDNA genotyping and serum protein markers should be orthogonal markers, their relative contributions varied considerably, which suggested that the optimal set of protein markers has not yet been identified.

The same group that developed CancerSEEK has published two further assays for early cancer detection — UroSEEK and PapSEEK. UroSEEK180 assays urothelial cancer cells shed into urine with a ten-gene multiplex assay, a TERT singleplex assay and an aneuploidy assay and has a sensitivity of 75% for the detection of urothelial cancer180. PapSEEK181 is a multiplex-PCR-based test capable of detecting genetic alterations in Pap or Tao brush samples. In addition to endometrial cancers, the test also enabled the detection of a substantial fraction of ovarian cancer, for which the sensitivity could be increased from 43% to 63% by including ctDNA assays181. These first examples imply that combination strategies have the power to greatly improve liquid biopsy analyses, especially if orthogonal analytes are combined to improve signal. The early detection of pancreatic cancer, a notoriously lethal cancer when surgical resection is not possible, could become a leading example for such a multiparameter approach (Fig. 4). To date, several different liquid biopsy assays that each use different analytes have demonstrated great potential for the early detection of pancreatic cancer6,7,19,137. Elevated BCAA levels may precede diagnosis of pancreatic cancer by 2–5 years19. Combined blood tests for mutations and proteins could detect early-stage prostate cancer with sensitivities of 64% and ~70% and specificities of >99%6,7. Additionally, the early detection of pancreatic cancer was also reported by the assessment of a cell surface proteoglycan, glypican 1 (GPC1), in exosomes137. Hence, combining these orthogonal markers into a truly multianalyte blood test should improve diagnostic sensitivity and specificity and eventually enable early detection of pancreatic cancer (Fig. 4).

Fig. 4: Combination strategies for early detection of cancer from liquid biopsy samples.

Various tumour-specific circulating analytes, including circulating tumour cells (CTCs), circulating cell-free DNA (cfDNA) and/or circulating tumour DNA (ctDNA), circulating cell-free RNA, extracellular vesicles (EVs) and tumour-educated platelets (TEPs), can be used for liquid biopsy approaches. Each analyte can yield different information about the genome (including mutations and copy number alterations (CNAs)), the epigenome, the transcriptome, the proteome or the metabolome (such as branched-chain amino acids (BCAAs)). These multidimensional data could be combined in innovative ways and used for machine learning purposes. A canonical machine learning workflow to build classifiers that might enable a distinction between tumour and normal states comprises four key steps: data cleaning and pre-processing, featurization extraction, model fitting and evaluation of the classifier. miRNA, microRNA.

Novel computational and machine learning strategies

Analytical and computational aspects of multiparameter analyses are challenging, and inclusion of multiple analytes requires ever larger cohorts for meaningful analysis. Machine learning methods offer an opportunity to automatically discover and detect cancer-specific signatures from liquid biopsies and range from simple approaches, such as logistic regression and support vector machines, through to complex artificial neural networks with many hidden layers20. Some of the main machine learning breakthroughs of recent years have used deep artificial neural networks for both feature discovery and building classifiers168. The canonical machine learning workflow comprises four key steps: data cleaning and pre-processing, featurization, model fitting and evaluation169 (Fig. 4). Machine learning for cancer detection and/or classification may also involve the use of publicly available data sets such as TCGA to inform features, for simulation experiments and for testing on real data. The advent of large public initiatives to explore the interaction of genotype and phenotype has made multitissue and multi-individual training data available; however, tissue-specific somatic and germline data are more comprehensive than data for circulating analytes.

Nevertheless, machine learning has already been used to classify cancer-relevant signatures in analytes such as ctDNA, CTCs, EVs and proteins18,182,183,184. For example, to improve label-free classification of cells, integrative feature extraction of flow cytometry images was developed using cell morphological features or protein concentrations, among other characteristics. The resulting multivariate deep learning classifier was able to detect CTCs with superior sensitivity and specificity than was achieved using single features182. Neural-network-based analyses that were trained to identify miRNA patterns that distinguished between healthy women and those with ovarian cancer achieved 100% specificity183. Another algorithm, CancerLocator, sought to simultaneously infer the proportion of ctDNA in plasma and its tissue of origin using genome-wide DNA methylation data and, on the basis of maximum-likelihood estimation, could indicate whether probands had tumours and where they may be located185. These studies point to the potential to recover biologically meaningful features from liquid biopsy samples using deep networks.

An important challenge for machine learning methods in cancer detection is the ‘curse of dimensionality’, in which the number of independent observations (n) is substantially smaller than the number of potential features in the raw data sets (p)169. In the case of cancer detection from cfDNA, n represents the number of patient samples and is typically in the range of tens to thousands, whereas p, if taken at the base pair resolution level, could easily run into the tens of billions for even a single patient. In an ideal setting, n would be greater than p and preferably by several orders of magnitude; in image-recognition problems, where deep learning is perhaps most developed, it is typical to have over a million labelled samples (n) but only a few thousand potential features (p), such as the number of pixels in an image. Thus, overfitting, which occurs when a model fits to training data but fails to generalize to new cases, is likely to be a problem for cancer detection. There are several measures for mitigating this issue, including regularization and data augmentation (that is, the generation of new examples by modifying real training data), and seeking better solutions remains an active area of research for the machine learning community. For multiomic data from blood, featurization by way of feature selection (that is, choosing the features with the most influence on the target variable) and feature extraction =and/or engineering (that is, generation of new features from a set of existing ones, but optimized for input to learning algorithms) may be paramount for both reducing the dimensions of the input and readying the data for machine learning.

Machine learning methods may also offer opportunities to reduce the experimental cost of liquid biopsy assays. Many interesting signals in liquid biopsies (including cfDNA nucleosome positioning, protein and RNA concentration, and cell subpopulations) are continuous rather than discrete. As a consequence, they are intrinsically more difficult to estimate and typically require a high depth of sequencing. If machine learning approaches are able to impute the status of certain signals from the status of other signals, the depth of sequencing and the amount of reagents required could be reduced, which might also make the test more suitable for wide adoption.

However, the increasing use of deep learning programmes raises ethical and legal issues. Addressing these concerns will require either the development of privacy-oriented machine learning algorithms, such as systems that operate directly on encrypted data sets without exposing private data186, or the future availability of deep learning and/or machine learning programmes in many institutions to preclude the need to send data to external parties.

Conclusions and future perspectives

Liquid biopsy samples are increasingly being adopted for a wide variety of applications in oncology. However, the increasing use of these promising biomarkers precedes a core understanding of the mechanisms and dynamics underlying liquid analytes and the resolution of important technical issues. Additional preclinical studies addressing the biology of liquid biopsy analytes are needed. Furthermore, the majority of liquid biopsy assays lack evidence of clinical validity and, in particular, clinical utility; therefore, their use is confined to research purposes within clinical studies. In addition, most existing assays have focused on a single analyte; however, resolution could be improved and the range of suitable applications could be vastly extended by adopting multiparametric assays that incorporate data from the multiple analytes present in a single blood sample. Developments in assay technology must be accompanied by novel statistical tools that make use of high-dimensional machine learning approaches to integrate the large amounts of data obtained from single samples. Thus, future liquid biopsy developments are expected to span a range of disciplines, including basic biology and physiology, molecular biology and assay technology and statistics and machine learning. Critically, assay developers need to follow regulatory guidance to establish assay performance and analytical validity. However, it is only after clinical validity and clinical utility have been demonstrated that liquid biopsies will reach their full potential and have the expected impact on genomics-driven oncology and the clinical management of patients with cancer.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Related links


Blueprint Epigenome consortium:


CEN/TS 16835-3:2015:

International Human Epigenome Consortium (IHEC):

Standardization of generic pre-analytical procedures for in vitro diagnostics for personalized medicine (SPIDIA4P):

The European Committee for Standardization (CEN):


  1. 1.

    Ashley, E. A. Towards precision medicine. Nat. Rev. Genet. 17, 507–522 (2016).

  2. 2.

    Kumar-Sinha, C. & Chinnaiyan, A. M. Precision oncology in the age of integrative genomics. Nat. Biotechnol. 36, 46–60 (2018).

  3. 3.

    Moscow, J. A., Fojo, T. & Schilsky, R. L. The evidence framework for precision cancer medicine. Nat. Rev. Clin. Oncol. 15, 183–192 (2018).

  4. 4.

    Zehir, A. et al. Mutational landscape of metastatic cancer revealed from prospective clinical sequencing of 10,000 patients. Nat. Med. 23, 703–713 (2017).

  5. 5.

    Cancer Genome Atlas Research Network. The Cancer Genome Atlas Pan-Cancer analysis project. Nat. Genet. 45, 1113–1120 (2013).

  6. 6.

    Cohen, J. D. et al. Combined circulating tumor DNA and protein biomarker-based liquid biopsy for the earlier detection of pancreatic cancers. Proc. Natl Acad. Sci. USA 114, 10202–10207 (2017).

  7. 7.

    Cohen, J. D. et al. Detection and localization of surgically resectable cancers with a multi-analyte blood test. Science 359, 926–930 (2018). This paper shows that a blood test can detect eight common cancer types through assessment of the levels of circulating proteins and mutations in cell-free DNA with a median diagnostic sensitivity of 70%.

  8. 8.

    Chan, K. C. A. et al. Analysis of plasma Epstein-Barr virus DNA to screen for nasopharyngeal cancer. N. Engl. J. Med. 377, 513–522 (2017). This prospective study investigates the use of EBV DNA in plasma samples to screen for early nasopharyngeal carcinoma in asymptomatic persons and achieves a sensitivity and specificity of 97.1% and 98.6%, respectively.

  9. 9.

    Pantel, K. & Alix-Panabieres, C. Circulating tumour cells in cancer patients: challenges and perspectives. Trends Mol. Med. 16, 398–406 (2010).

  10. 10.

    Siravegna, G., Marsoni, S., Siena, S. & Bardelli, A. Integrating liquid biopsies into the management of cancer. Nat. Rev. Clin. Oncol. 14, 531–548 (2017).

  11. 11.

    Wan, J. C. M. et al. Liquid biopsies come of age: towards implementation of circulating tumour DNA. Nat. Rev. Cancer 17, 223–238 (2017).

  12. 12.

    Heitzer, E., Ulz, P. & Geigl, J. B. Circulating tumor DNA as a liquid biopsy for cancer. Clin. Chem. 61, 112–123 (2015).

  13. 13.

    Alix-Panabieres, C. & Pantel, K. Challenges in circulating tumour cell research. Nat. Rev. Cancer 14, 623–631 (2014).

  14. 14.

    Alix-Panabieres, C. & Pantel, K. Clinical applications of circulating tumor cells and circulating tumor dna as liquid biopsy. Cancer Discov. 6, 479–491 (2016).

  15. 15.

    Bardelli, A. & Pantel, K. Liquid biopsies, what we do not know (yet). Cancer Cell 31, 172–179 (2017).

  16. 16.

    Amorim, M. G. et al. A total transcriptome profiling method for plasma-derived extracellular vesicles: applications for liquid biopsies. Sci. Rep. 7, 14395 (2017).

  17. 17.

    Chan, K. C. et al. Noninvasive detection of cancer-associated genome-wide hypomethylation and copy number aberrations by plasma DNA bisulfite sequencing. Proc. Natl Acad. Sci. USA 110, 18761–18768 (2013).

  18. 18.

    Kim, Y. et al. Targeted proteomics identifies liquid-biopsy signatures for extracapsular prostate cancer. Nat. Commun. 7, 11906 (2016).

  19. 19.

    Mayers, J. R. et al. Elevation of circulating branched-chain amino acids is an early event in human pancreatic adenocarcinoma development. Nat. Med. 20, 1193–1198 (2014).

  20. 20.

    Ko, J. et al. Machine learning to detect signatures of disease in liquid biopsies — a user’s guide. Lab. Chip 18, 395–405 (2018).

  21. 21.

    Teutsch, S. M. et al. The Evaluation of Genomic Applications in Practice and Prevention (EGAPP) Initiative: methods of the EGAPP Working Group. Genet. Med. 11, 3–14 (2009).

  22. 22.

    Hayes, D. F. Biomarker validation and testing. Mol. Oncol. 9, 960–966 (2015).

  23. 23.

    Hayes, D. F. Precision medicine and testing for tumor biomarkers — are all tests born equal? JAMA Oncol. 4, 773–774 (2017).

  24. 24.

    Cristofanilli, M. et al. Circulating tumor cells, disease progression, and survival in metastatic breast cancer. N. Engl. J. Med. 351, 781–791 (2004).

  25. 25.

    Cristofanilli, M. et al. Circulating tumor cells: a novel prognostic factor for newly diagnosed metastatic breast cancer. J. Clin. Oncol. 23, 1420–1430 (2005).

  26. 26.

    Riethdorf, S. et al. Detection of circulating tumor cells in peripheral blood of patients with metastatic breast cancer: a validation study of the CellSearch system. Clin. Cancer Res. 13, 920–928 (2007).

  27. 27.

    Pantel, K., Brakenhoff, R. H. & Brandt, B. Detection, clinical relevance and specific biological properties of disseminating tumour cells. Nat. Rev. Cancer 8, 329–340 (2008).

  28. 28.

    Bidard, F. C. et al. Circulating tumor cells in breast cancer patients treated by neoadjuvant chemotherapy: a meta-analysis. J. Natl Cancer Inst. 110, 560–567 (2018).

  29. 29.

    Bidard, F. C. et al. Clinical validity of circulating tumour cells in patients with metastatic breast cancer: a pooled analysis of individual patient data. Lancet Oncol. 15, 406–414 (2014).

  30. 30.

    de Bono, J. S. et al. Circulating tumor cells predict survival benefit from treatment in metastatic castration-resistant prostate cancer. Clin. Cancer Res. 14, 6302–6309 (2008).

  31. 31.

    Antonarakis, E. S. et al. Clinical significance of androgen receptor splice variant-7 mRNA detection in circulating tumor cells of men with metastatic castration-resistant prostate cancer treated with first- and second-line abiraterone and enzalutamide. J. Clin. Oncol. 35, 2149–2156 (2017).

  32. 32.

    Lorente, D. et al. Decline in circulating tumor cell count and treatment outcome in advanced prostate cancer. Eur. Urol. 70, 985–992 (2016).

  33. 33.

    Heller, G. et al. Circulating tumor cell number as a response measure of prolonged survival for metastatic castration-resistant prostate cancer: a comparison with prostate-specific antigen across five randomized phase III clinical trials. J. Clin. Oncol. 36, 572–580 (2018).

  34. 34.

    Cohen, S. J. et al. Relationship of circulating tumor cells to tumor response, progression-free survival, and overall survival in patients with metastatic colorectal cancer. J. Clin. Oncol. 26, 3213–3221 (2008).

  35. 35.

    Huang, X. et al. Meta-analysis of the prognostic value of circulating tumor cells detected with the CellSearch System in colorectal cancer. BMC Cancer 15, 202 (2015).

  36. 36.

    Riethdorf, S., O’Flaherty, L., Hille, C. & Pantel, K. Clinical applications of the CellSearch platform in cancer patients. Adv. Drug. Deliv. Rev. 125, 102–121 (2018).

  37. 37.

    Sacher, A. G. et al. Prospective validation of rapid plasma genotyping for the detection of EGFR and KRAS mutations in advanced lung cancer. JAMA Oncol. 2, 1014–1022 (2016).

  38. 38.

    Leighl, N. B. et al. Molecular testing for selection of patients with lung cancer for epidermal growth factor receptor and anaplastic lymphoma kinase tyrosine kinase inhibitors: American Society of Clinical Oncology Endorsement of the College of American Pathologists/International Association for the Study of Lung Cancer/Association for Molecular Pathology Guideline. J. Clin. Oncol. 32, 3673–3679 (2014).

  39. 39.

    Warren, J. D. et al. Septin 9 methylated DNA is a sensitive and specific blood test for colorectal cancer. BMC Med. 9, 133 (2011).

  40. 40.

    Song, L., Jia, J., Peng, X., Xiao, W. & Li, Y. The performance of the SEPT9 gene methylation assay and a comparison with other CRC screening tests: a meta-analysis. Sci. Rep. 7, 3032 (2017).

  41. 41.

    Nian, J. et al. Diagnostic accuracy of methylated SEPT9 for blood-based colorectal cancer detection: a systematic review and meta-analysis. Clin. Transl Gastroenterol. 8, e216 (2017).

  42. 42.

    Wu, Y. L. et al. First-line erlotinib versus gemcitabine/cisplatin in patients with advanced EGFR mutation-positive non-small-cell lung cancer: analyses from the phase III, randomized, open-label, ENSURE study. Ann. Oncol. 26, 1883–1889 (2015).

  43. 43.

    Lindeman, N. I. et al. Updated molecular testing guideline for the selection of lung cancer patients for treatment with targeted tyrosine kinase inhibitors guideline from the College of American Pathologists, the International Association for the Study of Lung Cancer, and the Association for Molecular Pathology. Arch. Pathol. Lab. Med. 142, 321–346 (2018).

  44. 44.

    Snyder, M. W., Kircher, M., Hill, A. J., Daza, R. M. & Shendure, J. Cell-free DNA comprises an in vivo nucleosome footprint that informs its tissues-of-origin. Cell 164, 57–68 (2016).

  45. 45.

    Ulz, P. et al. Inferring expressed genes by whole-genome sequencing of plasma DNA. Nat. Genet. 48, 1273–1278 (2016). This paper reports the use of machine learning for gene classification based on differential read depth patterns at transcription start sites, which reflect nucleosome occupancy patterns.

  46. 46.

    De Mattos-Arruda, L. et al. Cerebrospinal fluid-derived circulating tumour DNA better represents the genomic alterations of brain tumours than plasma. Nat. Commun. 6, 8839 (2015).

  47. 47.

    De Mattos-Arruda, L. et al. Capturing intra-tumor genetic heterogeneity by de novo mutation profiling of circulating cell-free tumor DNA: a proof-of-principle. Ann. Oncol. 25, 1729–1735 (2014).

  48. 48.

    Jamal-Hanjani, M. et al. Detection of ubiquitous and heterogeneous mutations in cell-free DNA from patients with early-stage non-small-cell lung cancer. Ann. Oncol. 27, 862–867 (2016).

  49. 49.

    Murtaza, M. et al. Multifocal clonal evolution characterized using circulating tumour DNA in a case of metastatic breast cancer. Nat. Commun. 6, 8760 (2015).

  50. 50.

    Kuo, Y. B., Chen, J. S., Fan, C. W., Li, Y. S. & Chan, E. C. Comparison of KRAS mutation analysis of primary tumors and matched circulating cell-free DNA in plasmas of patients with colorectal cancer. Clin. Chim. Acta 433, 284–289 (2014).

  51. 51.

    Oxnard, G. R. et al. Association between plasma genotyping and outcomes of treatment with osimertinib (AZD9291) in advanced non-small-cell lung cancer. J. Clin. Oncol. 34, 3375–3382 (2016).

  52. 52.

    Siravegna, G. et al. Clonal evolution and resistance to EGFR blockade in the blood of colorectal cancer patients. Nat. Med. 21, 795 (2015).

  53. 53.

    Ulz, P. et al. Whole-genome plasma sequencing reveals focal amplifications as a driving force in metastatic prostate cancer. Nat. Commun. 7, 12008 (2016).

  54. 54.

    Bettegowda, C. et al. Detection of circulating tumor DNA in early- and late-stage human malignancies. Sci. Transl Med. 6, 224ra224 (2014). This study evaluates 640 patients with different cancer types and reports a high variability of ctDNA in patients with the same tumour entities and disease stages.

  55. 55.

    Carreira, S. et al. Tumor clone dynamics in lethal prostate cancer. Sci. Transl Med. 6, 254ra125 (2014).

  56. 56.

    Newman, A. M. et al. An ultrasensitive method for quantitating circulating tumor DNA with broad patient coverage. Nat. Med. 20, 548–554 (2014). This study demonstrates that the combination of UMIs with highly sophisticated bioinformatics algorithms can substantially improve the sensitivity of sequencing-based approaches.

  57. 57.

    Scherer, F. et al. Distinct biological subtypes and patterns of genome evolution in lymphoma revealed by circulating tumor DNA. Sci Transl Med. 8, 364ra155 (2016).

  58. 58.

    Yu, M. et al. Cancer therapy. Ex vivo culture of circulating breast tumor cells for individualized testing of drug susceptibility. Science 345, 216–220 (2014).

  59. 59.

    Abbosh, C. et al. Phylogenetic ctDNA analysis depicts early-stage lung cancer evolution. Nature 545, 446–451 (2017). This study uses a tumour-specific phylogenetic approach to track the subclonal nature of lung cancer relapse and metastasis. In addition, independent predictors of ctDNA release were identified and correlations of the tumour volume and ctDNA allele frequencies were described.

  60. 60.

    Parkinson, C. A. et al. Exploratory analysis of TP53 mutations in circulating tumour DNA as biomarkers of treatment response for patients with relapsed high-grade serous ovarian carcinoma: a retrospective study. PLOS Med. 13, e1002198 (2016). This study demonstrates that TP53 mutations can be used as personalized markers to monitor tumour burden and early changes as a predictor of response and time to progression in patients with HGSOC.

  61. 61.

    Chaudhuri, A. A. et al. Early detection of molecular residual disease in localized lung cancer by circulating tumor DNA profiling. Cancer Discov. 12, 1394–1403 (2017).

  62. 62.

    Tie, J. et al. Circulating tumor DNA as an early marker of therapeutic response in patients with metastatic colorectal cancer. Ann. Oncol. 26, 1715–1722 (2015).

  63. 63.

    Tie, J. et al. Circulating tumor DNA analysis detects minimal residual disease and predicts recurrence in patients with stage II colon cancer. Sci. Transl Med. 8, 346ra392 (2016).

  64. 64.

    Olsson, E. et al. Serial monitoring of circulating tumor DNA in patients with primary breast cancer for detection of occult metastatic disease. EMBO Mol. Med. 7, 1034–1047 (2015).

  65. 65.

    Diaz, L. A. Jr. et al. The molecular evolution of acquired resistance to targeted EGFR blockade in colorectal cancers. Nature 486, 537–540 (2012).

  66. 66.

    Misale, S., Di Nicolantonio, F., Sartore-Bianchi, A., Siena, S. & Bardelli, A. Resistance to anti-EGFR therapy in colorectal cancer: from heterogeneity to convergent evolution. Cancer Discov. 4, 1269–1280 (2014).

  67. 67.

    Misale, S. et al. Emergence of KRAS mutations and acquired resistance to anti-EGFR therapy in colorectal cancer. Nature 486, 532–U131 (2012).

  68. 68.

    Mohan, S. et al. Changes in colorectal carcinoma genomes under anti-EGFR therapy identified by whole-genome plasma DNA sequencing. PLOS Genet. 10, e1004271 (2014).

  69. 69.

    Beddowes, E., Sammut, S. J., Gao, M. & Caldas, C. Predicting treatment resistance and relapse through circulating DNA. Breast 34, S31–S35 (2017).

  70. 70.

    Murtaza, M. et al. Non-invasive analysis of acquired resistance to cancer therapy by sequencing of plasma DNA. Nature 497, 108–112 (2013).

  71. 71.

    Del Monte, U. Does the cell number 109 still really fit one gram of tumor tissue? Cell Cycle 8, 505–506 (2009).

  72. 72.

    Heidary, M. et al. The dynamic range of circulating tumor DNA in metastatic breast cancer. Breast Cancer Res. 16, 421 (2014).

  73. 73.

    Heitzer, E. et al. Establishment of tumor-specific copy number alterations from plasma DNA of patients with cancer. Int. J. Cancer 133, 346–356 (2013).

  74. 74.

    Newman, A. M. et al. Integrated digital error suppression for improved detection of circulating tumor DNA. Nat. Biotechnol. 34, 547–555 (2016).

  75. 75.

    Park, G. et al. Characterization of background noise in capture-based targeted sequencing data. Genome Biol. 18, 136 (2017).

  76. 76.

    Kinde, I., Wu, J., Papadopoulos, N., Kinzler, K. W. & Vogelstein, B. Detection and quantification of rare mutations with massively parallel sequencing. Proc. Natl Acad. Sci. USA 108, 9530–9535 (2011).

  77. 77.

    Forshew, T. et al. Noninvasive identification and monitoring of cancer mutations by targeted deep sequencing of plasma DNA. Sci. Transl Med. 4, 136ra168 (2012).

  78. 78.

    Schmitt, M. W. et al. Detection of ultra-rare mutations by next-generation sequencing. Proc. Natl Acad. Sci. USA 109, 14508–14513 (2012).

  79. 79.

    Phallen, J. et al. Direct detection of early-stage cancers using circulating tumor DNA. Sci. Transl Med. 9, eaan2415 (2017). In this study, high-resolution gene panel profiling (TEC-Seq) is employed for non-invasive detection of early-stage tumours in 200 patients with stage I or II colorectal, breast, lung or ovarian cancer.

  80. 80.

    Smith, T., Heger, A. & Sudbery, I. UMI-tools: modeling sequencing errors in unique molecular identifiers to improve quantification accuracy. Genome Res. 27, 491–499 (2017).

  81. 81.

    Jiang, P. et al. Lengthening and shortening of plasma DNA in hepatocellular carcinoma patients. Proc. Natl Acad. Sci. USA 112, E1317–E1325 (2015).

  82. 82.

    Underhill, H. R. et al. Fragment length of circulating tumor DNA. PLOS Genet. 12, e1006162 (2016).

  83. 83.

    Jiang, P. Y. & Lo, Y. M. D. The long and short of circulating cell-free DNA and the ins and outs of molecular diagnostics. Trends Genet. 32, 360–371 (2016).

  84. 84.

    Lo, Y. M. D. et al. Maternal plasma DNA sequencing reveals the genome-wide genetic and mutational profile of the fetus. Sci. Transl Med. 2, 61ra91 (2010).

  85. 85.

    Moser, T. et al. Single-stranded DNA library preparation does not preferentially enrich circulating tumor DNA. Clin. Chem. 63, 1656–1659 (2017).

  86. 86.

    Vong, J. S. L. et al. Single-stranded DNA library preparation preferentially enriches short maternal DNA in maternal plasma. Clin. Chem. 63, 1031–1037 (2017).

  87. 87.

    Sun, K. et al. Size-tagged preferred ends in maternal plasma DNA shed light on the production mechanism and show utility in noninvasive prenatal testing. Proc. Natl Acad. Sci. USA 115, E5106–E5114 (2018). The authors investigate whether preferred end sites might bear any relationship with fragment lengths of plasma DNA. Short and long plasma DNA molecules were associated with different preferred DNA end sites.

  88. 88.

    Blokzijl, F. et al. Tissue-specific mutation accumulation in human adult stem cells during life. Nature 538, 260–264 (2016).

  89. 89.

    Tomasetti, C. & Vogelstein, B. Variation in cancer risk among tissues can be explained by the number of stem cell divisions. Science 347, 78–81 (2015).

  90. 90.

    Adams, P. D., Jasper, H. & Rudolph, K. L. Aging-induced stem cell mutations as drivers for disease and cancer. Cell Stem Cell 16, 601–612 (2015).

  91. 91.

    Genovese, G. et al. Clonal hematopoiesis and blood-cancer risk inferred from blood DNA sequence. N. Engl. J. Med. 371, 2477–2487 (2014).

  92. 92.

    Jaiswal, S. et al. Age-related clonal hematopoiesis associated with adverse outcomes. N. Engl. J. Med. 371, 2488–2498 (2014).

  93. 93.

    Xie, M. C. et al. Age-related mutations associated with clonal hematopoietic expansion and malignancies. Nat. Med. 20, 1472–1478 (2014).

  94. 94.

    Ptashkin, R. N. et al. Prevalence of clonal hematopoiesis mutations in tumor-only clinical genomic profiling of solid tumors. JAMA Oncol. This study demonstrates how clonal haematopoiesis-derived mutations could lead to erroneous reporting and treatment recommendations when tumour-only sequencing is used.

  95. 95.

    Coombs, C. C. et al. Identification of clonal hematopoiesis mutations in solid tumor patients undergoing unpaired next-generation sequencing assays. Clin. Cancer Res. (2018).

  96. 96.

    Hu, Y. et al. False positive plasma genotyping due to clonal hematopoiesis. Clin. Cancer Res. (2018).

  97. 97.

    Jacobs, K. B. et al. Detectable clonal mosaicism and its relationship to aging and cancer. Nat. Genet. 44, 651–668 (2012).

  98. 98.

    O’Huallachain, M., Karczewski, K. J., Weissman, S. M., Urban, A. E. & Snyder, M. P. Extensive genetic variation in somatic human tissues. Proc. Natl Acad. Sci. USA 109, 18018–18023 (2012).

  99. 99.

    Behjati, S. et al. Genome sequencing of normal cells reveals developmental lineages and mutational processes. Nature 513, 422–425 (2014).

  100. 100.

    Krimmel, J. D. et al. Ultra-deep sequencing detects ovarian cancer cells in peritoneal fluid and reveals somatic TP53 mutations in noncancerous tissues. Proc. Natl Acad. Sci. USA 113, 6005–6010 (2016).

  101. 101.

    Fernandez-Cuesta, L. et al. Identification of circulating tumor DNA for the early detection of small-cell lung cancer. EBioMedicine 10, 117–123 (2016).

  102. 102.

    Odegaard, J. I. et al. Validation of a plasma-based comprehensive cancer genotyping assay utilizing orthogonal tissue- and plasma-based methodologies. Clin. Cancer Res. 24, 3539–3549 (2018).

  103. 103.

    Torga, G. & Pienta, K. J. Patient-paired sample congruence between 2 commercial liquid biopsy tests. JAMA Oncol. 4, 868–870 (2017).

  104. 104.

    Merker, J. D. et al. Circulating tumor DNA analysis in patients with cancer: American Society of Clinical Oncology and College of American Pathologists joint review. J. Clin. Oncol. 16, 1631–1641 (2018).

  105. 105.

    Grossman, R. L. et al. Collaborating to compete: Blood Profiling Atlas in Cancer (BloodPAC) Consortium. Clin. Pharmacol. Ther. 101, 589–592 (2017).

  106. 106.

    Inamdar, S., Nitiyanandan, R. & Rege, K. Emerging applications of exosomes in cancer therapeutics and diagnostics. Bioeng. Transl Med. 2, 70–80 (2017).

  107. 107.

    Best, M. G. et al. Swarm intelligence-enhanced detection of non-small-cell lung cancer using tumor-educated platelets. Cancer Cell 32, 238–252 (2017).

  108. 108.

    Best, M. G. et al. RNA-seq of tumor-educated platelets enables blood-based pan-cancer, multiclass, and molecular pathway cancer diagnostics. Cancer Cell 28, 666–676 (2015).

  109. 109.

    Mayers, J. R. et al. Tissue of origin dictates branched-chain amino acid metabolism in mutant Kras-driven cancers. Science 353, 1161–1165 (2016).

  110. 110.

    Hao, X. et al. DNA methylation markers for diagnosis and prognosis of common cancers. Proc. Natl Acad. Sci. USA 114, 7414–7419 (2017).

  111. 111.

    Baylin, S. B. & Jones, P. A. A decade of exploring the cancer epigenome — biological and translational implications. Nat. Rev. Cancer 11, 726–734 (2011).

  112. 112.

    Warton, K., Mahon, K. L. & Samimi, G. Methylated circulating tumor DNA in blood: power in cancer prognosis and response. Endocr. Relat. Cancer 23, R157–R171 (2016).

  113. 113.

    Widschwendter, M. et al. Methylation patterns in serum DNA for early identification of disseminated breast cancer. Genome Med. 9, 115 (2017).

  114. 114.

    Widschwendter, M. et al. The potential of circulating tumor DNA methylation analysis for the early detection and management of ovarian cancer. Genome Med. 9, 116 (2017).

  115. 115.

    Li, M. et al. Sensitive digital quantification of DNA methylation in clinical samples. Nat. Biotechnol. 27, 858–863 (2009).

  116. 116.

    Guo, S. et al. Identification of methylation haplotype blocks aids in deconvolution of heterogeneous tissue samples and tumor tissue-of-origin mapping from plasma DNA. Nat. Genet. 49, 635–642 (2017).

  117. 117.

    Lehmann-Werman, R. et al. Identification of tissue-specific cell death using methylation patterns of circulating DNA. Proc. Natl Acad. Sci. USA 113, E1826–E1834 (2016).

  118. 118.

    Pixberg, C. F., Schulz, W. A., Stoecklein, N. H. & Neves, R. P. Characterization of DNA methylation in circulating tumor cells. Genes 6, 1053–1075 (2015).

  119. 119.

    Mastoraki, S. et al. ESR1 methylation: a liquid biopsy-based epigenetic assay for the follow-up of patients with metastatic breast cancer receiving endocrine treatment. Clin. Cancer Res. 24, 1500–1510 (2018).

  120. 120.

    Kelsey, G., Stegle, O. & Reik, W. Single-cell epigenomics: recording the past and predicting the future. Science 358, 69–75 (2017).

  121. 121.

    Koh, W. et al. Noninvasive in vivo monitoring of tissue-specific global gene expression in humans. Proc. Natl Acad. Sci. USA 111, 7361–7366 (2014).

  122. 122.

    Tsui, N. B. et al. Maternal plasma RNA sequencing for genome-wide transcriptomic profiling and identification of pregnancy-associated transcripts. Clin. Chem. 60, 954–962 (2014).

  123. 123.

    Silva, J. M. et al. Detection of epithelial tumour RNA in the plasma of colon cancer patients is associated with advanced stages and circulating tumour cells. Gut 50, 530–534 (2002).

  124. 124.

    Collado, M. et al. Genomic profiling of circulating plasma RNA for the analysis of cancer. Clin. Chem. 53, 1860–1863 (2007).

  125. 125.

    Chen, X. et al. Characterization of microRNAs in serum: a novel class of biomarkers for diagnosis of cancer and other diseases. Cell Res. 18, 997–1006 (2008).

  126. 126.

    Montani, F. & Bianchi, F. Circulating cancer biomarkers: the macro-revolution of the micro-RNA. EBioMedicine 5, 4–6 (2016).

  127. 127.

    Ng, E. K. O. et al. mRNA of placental origin is readily detectable in maternal plasma. Proc. Natl Acad. Sci. USA 100, 4748–4753 (2003).

  128. 128.

    Montani, F. et al. miR-Test: a blood test for lung cancer early detection. J. Natl. Cancer Inst. 107, djv063 (2015).

  129. 129.

    Sozzi, G. et al. Clinical utility of a plasma-based miRNA signature classifier within computed tomography lung cancer screening: a correlative MILD trial study. J. Clin. Oncol. 32, 768–773 (2014).

  130. 130.

    Cortez, M. A. et al. MicroRNAs in body fluids-the mix of hormones and biomarkers. Nat. Rev. Clin. Oncol. 8, 467–477 (2011).

  131. 131.

    Schwarzenbach, H., da Silva, A. M., Calin, G. & Pantel, K. Data normalization strategies for microRNA quantification. Clin. Chem. 61, 1333–1342 (2015).

  132. 132.

    Schwarzenbach, H., Nishida, N., Calin, G. A. & Pantel, K. Clinical relevance of circulating cell-free microRNAs in cancer. Nat. Rev. Clin. Oncol. 11, 145–156 (2014).

  133. 133.

    Maas, S. L., Breakefield, X. O. & Weaver, A. M. Extracellular vesicles: unique intercellular delivery vehicles. Trends Cell. Biol. 27, 172–188 (2017).

  134. 134.

    Vlassov, A. V., Magdaleno, S., Setterquist, R. & Conrad, R. Exosomes: current knowledge of their composition, biological functions, and diagnostic and therapeutic potentials. Biochim. Biophys. Acta 1820, 940–948 (2012).

  135. 135.

    Yuan, T. Z. et al. Plasma extracellular RNA profiles in healthy and cancer patients. Sci. Rep. 6, 19413 (2016).

  136. 136.

    Khan, S. et al. Plasma-derived exosomal survivin, a plausible biomarker for early detection of prostate cancer. PLOS ONE 7, e46737 (2012).

  137. 137.

    Melo, S. A. et al. Glypican-1 identifies cancer exosomes and detects early pancreatic cancer. Nature 523, 177–182 (2015).

  138. 138.

    Rupp, A. K. et al. Loss of EpCAM expression in breast cancer derived serum exosomes: role of proteolytic cleavage. Gynecol. Oncol. 122, 437–446 (2011).

  139. 139.

    Hoshino, A. et al. Tumour exosome integrins determine organotropic metastasis. Nature 527, 329–335 (2015).

  140. 140.

    Peinado, H. et al. Pre-metastatic niches: organ-specific homes for metastases. Nat. Rev. Cancer 17, 302–317 (2017).

  141. 141.

    Thakur, B. K. et al. Double-stranded DNA in exosomes: a novel biomarker in cancer detection. Cell Res. 24, 766–769 (2014).

  142. 142.

    Allenson, K. et al. High prevalence of mutant KRAS in circulating exosome-derived DNA from early-stage pancreatic cancer patients. Ann. Oncol. 28, 741–747 (2017).

  143. 143.

    Kalluri, R. The biology and function of exosomes in cancer. J. Clin. Invest. 126, 1208–1215 (2016).

  144. 144.

    Im, H. et al. Label-free detection and molecular profiling of exosomes with a nano-plasmonic sensor. Nat. Biotechnol. 32, 490–495 (2014).

  145. 145.

    Abels, E. R. & Breakefield, X. O. Introduction to extracellular vesicles: biogenesis, RNA cargo selection, content, release, and uptake. Cell. Mol. Neurobiol. 36, 301–312 (2016).

  146. 146.

    Sharma, S. Tumor markers in clinical practice: general principles and guidelines. Indian J. Med. Paediatr. Oncol. 30, 1–8 (2009).

  147. 147.

    Attard, G. et al. Prostate cancer. Lancet 387, 70–82 (2016).

  148. 148.

    Aziz, D. C. Clinical use of tumor markers based on outcome analysis. Lab. Med. 27, 817–821 (1996).

  149. 149.

    Aebersold, R. & Mann, M. Mass-spectrometric exploration of proteome structure and function. Nature 537, 347–355 (2016).

  150. 150.

    Crutchfield, C. A., Thomas, S. N., Sokoll, L. J. & Chan, D. W. Advances in mass spectrometry-based clinical biomarker discovery. Clin. Proteom. 13, 1 (2016).

  151. 151.

    Zahn, H. et al. Scalable whole-genome single-cell library preparation without preamplification. Nat. Methods 14, 167–173 (2017).

  152. 152.

    Vitak, S. A. et al. Sequencing thousands of single-cell genomes with combinatorial indexing. Nat. Methods 14, 302–308 (2017).

  153. 153.

    Kalinich, M. et al. An RNA-based signature enables high specificity detection of circulating tumor cells in hepatocellular carcinoma. Proc. Natl Acad. Sci. USA 114, 1123–1128 (2017).

  154. 154.

    Alix-Panabieres, C. EPISPOT assay: detection of viable DTCs/CTCs in solid tumor patients. Recent Results Cancer Res. 195, 69–76 (2012).

  155. 155.

    Clark, S. J., Lee, H. J., Smallwood, S. A., Kelsey, G. & Reik, W. Single-cell epigenomics: powerful new methods for understanding gene regulation and cell identity. Genome Biol. 17, 72 (2016).

  156. 156.

    Alix-Panabieres, C. & Pantel, K. Liquid biopsy in cancer patients: advances in capturing viable CTCs for functional studies using the EPISPOT assay. Expert Rev. Mol. Diagn. 15, 1411–1417 (2015).

  157. 157.

    Baccelli, I. et al. Identification of a population of blood circulating tumor cells from breast cancer patients that initiates metastasis in a xenograft assay. Nat. Biotechnol. 31, 539–544 (2013).

  158. 158.

    Hodgkinson, C. L. et al. Tumorigenicity and genetic profiling of circulating tumor cells in small-cell lung cancer. Nat. Med. 20, 897–903 (2014).

  159. 159.

    Cayrefourcq, L. et al. Establishment and characterization of a cell line from human circulating colon cancer cells. Cancer Res. 75, 892–901 (2015).

  160. 160.

    Heitzer, E. et al. Complex tumor genomes inferred from single circulating tumor cells by array-CGH and next-generation sequencing. Cancer Res. 73, 2965–2975 (2013).

  161. 161.

    Malaney, P., Nicosia, S. V. & Dave, V. One mouse, one patient paradigm: new avatars of personalized cancer therapy. Cancer Lett. 344, 1–12 (2014).

  162. 162.

    Lallo, A., Schenk, M. W., Frese, K. K., Blackhall, F. & Dive, C. Circulating tumor cells and CDX models as a tool for preclinical drug development. Transl Lung Cancer Res. 6, 397–408 (2017).

  163. 163.

    Chicard, M. et al. Genomic copy number profiling using circulating free tumor DNA highlights heterogeneity in neuroblastoma. Clin. Cancer Res. 22, 5564–5573 (2016).

  164. 164.

    Chim, S. S. C. et al. Detection and characterization of placental MicroRNAs in maternal plasma. Clin. Chem. 54, 482–490 (2008).

  165. 165.

    Wang, K. et al. Circulating microRNAs, potential biomarkers for drug-induced liver injury. Proc. Natl Acad. Sci. USA 106, 4402–4407 (2009).

  166. 166.

    Sun, K. et al. Plasma DNA tissue mapping by genome-wide methylation sequencing for noninvasive prenatal, cancer, and transplantation assessments. Proc. Natl Acad. Sci. USA 112, E5503–E5512 (2015). In this study, the tissue of origin of circulating cell-free DNA is tracked using methylation deconvolution in different cohorts.

  167. 167.

    Wong, F. C. et al. Cell-free DNA in maternal plasma and serum: a comparison of quantity, quality and tissue origin using genomic and epigenomic approaches. Clin. Biochem. 49, 1379–1386 (2016).

  168. 168.

    Lam, W. K. J. et al. DNA of erythroid origin is present in human plasma and informs the types of anemia. Clin. Chem. 63, 1614–1623 (2017).

  169. 169.

    Lui, Y. N. Y. N. et al. Predominant hematopoietic origin of cell-free DNA in plasma and serum after sex-mismatched bone marrow transplantation. Clin. Chem. 48, 421–427 (2002).

  170. 170.

    Wu, D. C. & Lambowitz, A. M. Facile single-stranded DNA sequencing of human plasma DNA via thermostable group II intron reverse transcriptase template switching. Sci. Rep. 7, 8421 (2017).

  171. 171.

    Mohme, M., Riethdorf, S. & Pantel, K. Circulating and disseminated tumour cells — mechanisms of immune surveillance and escape. Nat. Rev. Clin. Oncol. 14, 155–167 (2017).

  172. 172.

    Mazel, M. et al. Frequent expression of PD-L1 on circulating breast cancer cells. Mol. Oncol. 9, 1773–1782 (2015).

  173. 173.

    Ricklefs, F. L. et al. Immune evasion mediated by PD-L1 on glioblastoma-derived extracellular vesicles. Sci. Adv. 4, eaar2766 (2018).

  174. 174.

    Gibney, G. T., Weiner, L. M. & Atkins, M. B. Predictive biomarkers for checkpoint inhibitor-based immunotherapy. Lancet Oncol. 17, E542–E551 (2016).

  175. 175.

    Khagi, Y. et al. Hypermutated circulating tumor DNA: correlation with response to checkpoint inhibitor-based immunotherapy. Clin. Cancer Res. 23, 5729–5736 (2017).

  176. 176.

    Kim, S. T. et al. Comprehensive molecular characterization of clinical responses to PD-1 inhibition in metastatic gastric cancer. Nat. Med. (2018).

  177. 177.

    Gandara, D. R. et al. Blood-based tumor mutational burden as a predictor of clinical benefit in non-small-cell lung cancer patients treated with atezolizumab. Nat. Med. 24, 1441–1448 (2018).

  178. 178.

    Sefrioui, D. et al. Diagnostic value of CA19.9, circulating tumour DNA and circulating tumour cells in patients with solid pancreatic tumours. Br. J. Cancer 117, 1017–1025 (2017).

  179. 179.

    Kalinich, M. & Haber, D. A. Cancer detection: Seeking signals in blood. Science 359, 866–867 (2018).

  180. 180.

    Springer, S. U. et al. Non-invasive detection of urothelial cancer through the analysis of driver gene mutations and aneuploidy. eLife 7, e32143 (2018).

  181. 181.

    Wang, Y. et al. Evaluation of liquid from the Papanicolaou test and other liquid biopsies for the detection of endometrial and ovarian cancers. Sci. Transl Med. 10, eaap8793 (2018).

  182. 182.

    Chen, C. L. et al. Deep learning in label-free cell classification. Sci. Rep. 6, 21471 (2016).

  183. 183.

    Elias, K. M. et al. Diagnostic potential for a serum miRNA neural network for detection of ovarian cancer. eLife 6, e28932 (2017).

  184. 184.

    Yuan, Y. et al. DeepGene: an advanced cancer type classifier based on deep learning and somatic point mutations. BMC Bioinformatics 17, 476 (2016).

  185. 185.

    Kang, S. L. et al. CancerLocator: non-invasive cancer diagnosis and tissue-of-origin prediction using methylation profiles of cell-free DNA. Genome Biol. 18, 53 (2017). This study exploits the diagnostic potential of genome-wide DNA methylation data from cfDNA to determine not only the presence but also the location of tumours. CancerLocator simultaneously infers the proportions and the tissue of origin of tumour-derived cell-free DNA in a blood sample.

  186. 186.

    Mohassel, P. SecureML: a system for scalable privacy-preserving machine learning. IEEE (2017).

Download references


The authors thank P. Ulz for his assistance in drafting the figures for this article and S. Perakis for her assistance in revising the text. The work in the authors’ laboratory is supported by CANCER-ID, a project funded by the Innovative Medicines Joint Undertaking; the Austrian National Bank (grant 16917); the Austrian Science Fund (grant P28949-B28); the BioTechMed-Graz flagship project ‘EPIAge’; and the Christian Doppler Research Fund for Liquid Biopsies for Early Detection of Cancer.

Reviewer information

Nature Reviews Genetics thanks P. Hofman and the other anonymous reviewer(s) for their contribution to the peer review of this work.

Author information

E.H. and M.R.S. researched data for the article, made substantial contributions to discussions of the content, wrote the article and reviewed and/or edited the manuscript before submission. I.S.H. and C.E.S.R. wrote the article and reviewed and/or edited the manuscript before submission.

Competing interests

I.S.H. and C.E.S.R. are employees of Freenome. Freenome is the industrial partner of the Christian Doppler Research Fund for Liquid Biopsies for Early Detection of Cancer, which is headed by E.H. The authors declare that no other competing interests exist.

Correspondence to Ellen Heitzer.


Precision oncology

Molecular profiling of a tumour with the aim to detect somatic alterations that can be targeted for therapy.

Next-generation sequencing

(NGS). A high-throughput method used to determine the nucleotide sequence of DNA or RNA.

The Cancer Genome Atlas

(TCGA). A comprehensive and coordinated effort to accelerate our understanding of the molecular basis of cancer through the application of genome analysis technologies.


A biochemical change in the genome, such as DNA methylation or histone modification, that does not alter the DNA sequence but may affect gene activity and expression.

Druggable targets

Somatic mutations involved in cancer development and progression that can be exploited with a therapeutic intent.

Pleural effusions

Excessive accumulations of fluid in the space surrounding the lung (pleural cavity).

Circulating tumour cells

(CTCs). Cells that have been shed into the vasculature or lymphatics from a primary tumour and/or metastasis and are carried around the body in the blood circulation.

Circulating cell-free DNA

(cfDNA). DNA circulating in the bloodstream that is not associated with cells.

Circulating tumour DNA

(ctDNA). Tumour-derived, cell-free DNA that is thought to be representative of the entire tumour genome.

Circulating cell-free RNA

(cfRNA). Circulating gene transcripts (mRNA and non-coding RNAs) that are partly protected from degradation by their packaging into exosomes.

Extracellular vesicles

(EVs). Generic term for vesicles, including exosomes, microvesicles or apoptotic bodies, that are secreted from all cells and carry complex cargoes such as proteins, lipids and nucleic acids across biological membranes.


Cell-derived vesicles likely present in all body fluids, which contain nucleic acids, lipids and metabolites and are involved in intercellular signalling and communication.

Tumour-educated platelets

(TEPs). Platelets with altered functions that interact with tumour cells via different signalling molecules, thereby promoting tumour cell survival and metastasis.

Copy number alterations

Loss (deletion) or gain (ranging from duplication to high-level amplification) of genomic regions resulting in a copy number that deviates from two.


The full range of mRNA molecules expressed in a cell, tissue or organism at a certain time.


The full complement of epigenetic marks within a genome, which helps to determine the activity of genes in any particular cell and its lineage. The epigenome is prone to change during ageing and in cancer cells.


The entire set of proteins expressed in a cell, tissue or organism at a certain time.


The complete set of small-molecule chemicals found within a cell, tissue or organism at a certain time.

Laboratory-developed tests

(LDTs). According to US Food and Drug Administration regulations, these tests are in vitro diagnostic tests designed, manufactured and used within a single laboratory.

Epithelial cell adhesion molecules

(EpCAMs). Transmembrane glycoproteins that are expressed solely in epithelia and epithelial-derived cancer and are commonly used as a diagnostic marker.

Variant allele frequency

(VAF). The frequency of a particular allele of a gene relative to all other alleles in a DNA sample.

Metabolic tumour volume

The proportion of a tumour, measured by volume, that is hypermetabolic. Often measured by positron emission tomography.

Minimal residual disease

Cancer cells that remain in the patient’s body during and after treatment, frequently escape detection by routine diagnostic procedures and are critically involved in relapse.

Genome equivalents

(GE). The amount of DNA that corresponds to the diploid genome of a single cell.

Sequencing sensitivity

The analytical sensitivity of a sequencing method; that is, the ability to detect a low concentration of a particular sequence in a biological sample. The sequencing sensitivity of an approach determines its ability to accurately measure the variant allele frequency of a variant.

Molecular barcodes

Degenerate sequence tags consisting of random or specified bases to label DNA fragments in order to track individual molecules after PCR; they are also referred to as unique molecular identifiers.

Cancer driver gene

A gene that, when mutated, is essential for cancer to develop and progress and that is under positive selection during tumour evolution.

Clonal expansion

A process in which acquisition of somatic mutations drives the production of daughter cells all arising from a single cell.


All nucleic acid methylation modifications in the genomes of cells.

Bisulfite sequencing

A method for detecting DNA methylation patterns. DNA is treated with bisulfite, which converts cytosine but not 5-methylcytosine to uracil, enabling methylated and unmethylated cytosines to be discriminated from sequencing data.


A means to group data into categories on the basis of certain characteristics, such as inherent similarity.

Tumour mutational burden

(TMB). A biomarker that measures the number of mutations present in a tumour of a patient with cancer. Usually, given as the number of coding somatic mutations per megabase of DNA.

Diagnostic sensitivity

The likelihood that a diagnostic test will be positive when testing a person with the disease.

Diagnostic specificity

The likelihood that a diagnostic test will be negative when testing a person without the disease.

Positive predictive value

(PPV). The proportion of positive results in diagnostic tests. PPV depends on the sensitivity and specificity of the test and on the prevalence of the disease within the general population.


The process of extracting cell type-specific information from heterogeneous samples. In liquid biopsies, this might involve resolving plasma DNA fragments into their constituent elements (for example, determining their tissue of origin on the basis of methylation markers).

Hidden layers

The intermediate layers of information generated when an artificial neural network breaks down and processes input data to generate the output.

Rights and permissions

To obtain permission to re-use content from this article visit RightsLink.

About this article

Further reading