Liquid Biopsy: Approaches to Dynamic Genotyping in Cancer

The majority of cancers contain numerous genetic variants that determine the oncogenic behavior of tumor cells [1, 2]. The knowledge of critical driver mutations and key pathways as rational drug targets has led to the development of molecular targeted cancer therapies including kinase inhibitors and immune checkpoint inhibitors. Consequently, cancer treatment is shifting from an entity-driven approach to molecular-guided individualized decision-making and patient care [3].

In contrast, the development of non-invasive methods to detect and monitor tumor growth and treatment response continues to be a major challenge in oncology. Response to therapy is evaluated based on diagnostic imaging and blood-based protein biomarkers. Current imaging techniques including positron emission tomography-computed tomography (PET-CT) display limited sensitivity and specificity [4, 5]. Pseudoprogressions seen with immune checkpoint inhibitors have challenged existing RECIST criteria and have led to the development of immune-related response criteria [6]. Tumor biomarkers are hampered by low sensitivity, low positive predictive value, and low specificity, since patients with low tumor load are often false-negative, and false-positive findings result from release by non-cancerous tissues and comorbid diseases. This includes well-established protein biomarkers such as prostate-specific antigen (PSA) in prostate cancer, carcinoembryonic antigen (CEA) in colorectal carcinoma (CRC) [7, 8], and S100 and lactate dehydrogenase (LDH) in melanoma [9, 10]. Thus, despite their widespread use, available protein biomarkers will often not detect relapse or progression of cancer. Consequently, treatment changes are often not instituted before clinical or radiologic progression is evident.

Tumors release tumor cells and nucleic acid content into the circulation [11] (fig. 1A). Circulating tumor cells (CTCs) and circulating tumor (ct)DNA can be detected and quantified (fig. 1B). Although the detection and quantification of ctDNA is hampered by its sometimes extremely low levels, ctDNA is unique in its tumor-specific origin: It contains tumor-specific mutations allowing strict discrimination from normal cell-free (cf)DNA, thus resulting in high-specificity detection, provided that accurate and sensitive assays are used. This opens up a broad range of novel applications (table 1) not covered by currently available biomarkers, including the possibility to depict changes in clonal genetic composition and to adapt treatment in a dynamic fashion (fig. 1B).

The presence of cfDNA in blood plasma and serum was first recognized in 1948 [12]. Cells undergoing necrosis or apoptosis release nucleic acid content into the bloodstream. Consequently, cfDNA is elevated after exercise, trauma, and inflammation, and in patients with cancer [13, 14, 15]. The presence of fetal DNA in maternal blood has recently allowed the development of commercial tests for prenatal assessment of germline fetal changes [16, 17]. DNA fragments released by tumor cells constitute cell-free ctDNA. Mechanisms of release by tumor cells include passive shedding by necrotic or apoptotic cells and possibly also active release [18] (fig. 1A). Fragments of ctDNA are often 160-200 bp in length, and the presence of nucleasomal laddering suggests the majority to be derived from cells undergoing apoptosis [19]. The amount of DNA released into the circulation correlates with cell turnover and hence tumor burden [19], and varies between tumor entities [20]. The fraction of cfDNA derived from a tumor varies between 0.01 and 90% [21, 22] and might depend on the size of the tumor and its location and vascularization [23]. Tumor nucleic acid content in blood samples can be accessed either by analyzing the ctDNA compartment or by genotyping CTCs [20, 24, 25].

The low abundance of ctDNA and the varying background of non-cancer-derived cfDNA hamper ctDNA monitoring and require highly sensitive and specific techniques for accurate detection and quantification. This is technically challenging, especially at the early tumor stages. Standard Sanger sequencing will not be sensitive enough in most cases. A number of different sequencing technologies have been demonstrated to specifically detect and quantify ctDNA with a sensitivity of 0.01% or lower, including digital polymerase chain reaction (dPCR) [26], beads, emulsion, amplification, and magnetics (BEAMing) - a modified dPCR technique [27] ligation (L)PCR [28], and recently targeted sequencing strategies such as amplicon deep sequencing [29] or hybrid-capture deep sequencing [30]. Using these technologies, ctDNA can be detected in 50% of early-stage cancers and in up to 100% of metastatic cancers [20, 21, 30]. In the last years, technologies based on dPCR and targeted sequencing have proven to be most suitable for liquid biopsy applications. Both technologies allow very sensitive detection of allele variants, down to a resolution of 0.005%. dPCR can simultaneously detect up to 4 different mutations [31]. Thus, dPCR might be most suitable as a robust point-of-care platform for the detection and monitoring of non-synonymous stereotypic mutations such as RAS in CRC or epidermal growth factor receptor (EGFR) mutations in lung cancer [31, 32]. However, each dPCR assay is limited to detecting 1 specific mutation site. In contrast, targeted sequencing of ctDNA will cover hundreds to thousands of amplicons in 1 assay read in an unbiased fashion [30, 33]. Advantages include detection of subclonal mutations and analysis of changes in clonal composition over time. However, targeted sequencing of ctDNA is expensive and time-consuming, and requires customized protocols to allow an assay sensitivity of below 1% and to avoid false-positives [34].

Currently, sequencing analyses of tissue biopsies are the standard for genetic treatment stratification. Examples include the use of BRAF inhibitors in BRAF-mutant melanoma [35], the use of KIT inhibitors in gastrointestinal stromal tumors (GIST) harboring activating KIT mutations [36], application of EGFR inhibitors in non-small-cell lung cancer (NSCLC) with activating EGFR mutations [37], or the negative predictive value of RAS mutations for EGFR antibodies in CRC [38]. However, tissue biopsies are invasive and bear the risk of complications [39], whereas liquid biopsy approaches are less invasive and thus suitable for repeated analyses. The use of tissue for cancer sequencing is limited by a variable content of tumor cellularity, and preservation techniques compromise the integrity of DNA usually extracted from formalin-fixed paraffin-embedded material [40]. While the tumor-derived DNA content and cfDNA fraction will also vary, blood-derived ctDNA can be freshly used at any time. In a blinded prospective study, the BRAF and KRAS mutation status in tissue and ctDNA showed a concordance rate of 96% with a sensitivity and specificity for ctDNA of 93 and 98%, respectively [22] (table 2). In a different study assessing ctDNA genotyping using hybrid-capture sequencing in patients with stage I-IV NSCLC, ctDNA was detected with a sensitivity of 100% in stage II-IV patients and 50% in stage I patients, with a specificity of 96% [30]. In a retrospective study comparing ctDNA samples from patients with NSCLC, GIST, CRC, and melanoma, analyzed using 6-gene panel sequencing with paired tissue samples and a Taqman-derived assay [41], the sensitivity and specificity of ctDNA detection relative to tumor tissue was 79 and 100%, respectively, with 96.8% concordance. These data suggest that ctDNA-based genotyping is feasible and competitive compared to tissue-based analysis.

Recently, it has become evident that cancers display a high level of clonal genetic heterogeneity between the primary and metastases, and even within specific lesions [42, 43]. While limited spatial coverage of a tissue biopsy might introduce a sampling bias, a liquid biopsy approach covering and integrating the genetic ‘output' of all coexisting lesions should be able to depict the predominating genetic fingerprint of the disease within 1 blood sample (fig. 1B). Moreover, it has become increasingly evident that the clonal genetic composition of a cancer is a highly dynamic process, especially if cancer cells are faced with the selection pressure of a molecular targeted treatment. Repeated blood sampling is minimally invasive and allows dynamic genotyping of the disease. Depicting changes in the mutational landscape over time might allow to adapt the treatment accordingly. Examples include selection of secondary EGFR kinase domain mutations mediating resistance to EGFR kinase inhibitors [44], or selection of RAS-mutant clones in CRC receiving EGFR antibodies [45].

A growing body of data supports the concept that changes in the amount of ctDNA correlate with tumor burden and may thus be used for disease monitoring. The short half-life of ctDNA of less than 2 h represents a challenge for sample processing unless collection tubes including additives are used; however, on the other side, it allows real-time assessment of tumor activity and genotype. Studies in CRC, breast and ovarian cancer, melanoma, GIST, and lymphoma have shown that levels of ctDNA correlate with the clinical course of the disease in individual patients and can precede changes seen in imaging studies by weeks or months [21, 28, 29, 46, 47, 48]. Thus, ctDNA monitoring might allow early assessment of response and prediction of progression. In a study of 18 patients with stage II-IV CRC, mutations in 4 genes (PIK3CA, APC, TP53, and KRAS) were quantified in the ctDNA in serial plasma samples using BEAMing [21]. The median ctDNA levels decreased by 96.7% only 24 h after surgery, and correlated with the clinical course thereafter. In 30 patients with metastatic breast cancer receiving systemic therapy, a prospective trial assessed the levels of ctDNA harboring TP53 and PIK3CA mutations using amplicon sequencing and dPCR [49]. In this analysis, the levels of mutant ctDNA correlated with disease status as measured by CT. Analysis of ctDNA displayed superior sensitivity when compared to levels of CTCs and carcinoma antigen (CA) 15-3 (table 2), and predicted response to treatment as assessed by CT. In 17 patients with localized and metastatic NSCLC, a prospective study assessed ctDNA using hybrid-capture sequencing with barcoding [30]. The levels of ctDNA correlated with tumor volume and predicted radiographic responses. Importantly, the assay design allowed the detection of single nucleotide variants, rearrangements of ALK, ROS1, and RET, insertions and deletions (indels), and copy number alterations. The detection limit of mutant allele fractions in ctDNA was 0.02%. Of note, analysis of ctDNA depicted clonal heterogeneity in individual patients such as co-occurrence of ROS and ALK fusions in 1 individual patient.

It has become impressively evident that therapeutic strategies and specific drugs are no longer specific for treatment of distinct entities but rather for particular molecular profiles across entities [50]. However, several additional layers of complexity aggravate simple 1-mutation/1-drug approaches. First, it is difficult to discriminate significant mutations and pathways from background and passenger mutations [51]. Second, the clonal genetic composition varies between patients, within 1 patient between lesions [42], and even within single lesions [43]. Finally, genetic composition is a highly dynamic process, especially if treatment leads to the selection of resistant subclones. In tissue, the analysis of the clonal composition of a tumor requires multiple single-region or single-cell samples [42, 52, 53]. Since ctDNA should be able to integrate DNA shed from different cancer lesions in 1 sample, it probably represents the most accurate strategy to query clonal heterogeneity and to study changes in clonal composition over time (fig. 1B). Consequently, ctDNA analysis has detected cancer mutations that were missed in corresponding tissue [45, 54], and ctDNA allele frequency can inform about clonal and subclonal mutations [55]. Recently, it has been shown that hybrid-capture sequencing of ctDNA can depict changes in clonal composition over time in breast, ovarian, and lung cancer, and thus can inform about mutations associated with acquired drug resistance [56]. In practice, the aim of serial genotyping of ctDNA would be to query known genetic changes that mediate resistance to treatment and/or predict response to treatment. In chronic myeloid leukemia, tumor cells are easily accessible, and monitoring of emerging disease subclones harboring secondary mutations in BCR-ABL that mediate resistance to ABL kinase inhibitors is routinely performed [57]. In solid tumors, ctDNA analysis might provide valuable information for dynamic treatment stratification. In lung adenocarcinoma, about half of the patients carry at least 1 driver mutation representing a rational target for therapeutic intervention [58]. Among those, activating EGFR mutations are predictive for response to EGFR inhibitors such as erlotinib and gefitinib [37]. Analysis of ctDNA has been demonstrated to detect EGFR exon 19 deletions and L858R mutations with a sensitivity of 82-87% and a specificity of 97-98% [54, 59], and thus might be a suitable alternative to tissue genotyping. Most patients receiving EGFR-inhibiting drugs will eventually experience disease progression within 2 years of treatment, and in 60% resistance to treatment is caused by selection of a disease clone harboring an EGFR T790M secondary mutation that blocks the access of the drug to the kinase [44]. EGFR T790M is predictive for response to osimertinib [60], and prescription requires a positive mutation test. In the phase I AURA study that assessed the safety, tolerability, and efficacy of osimertinib in patients with EGFR mutation-positive NSCLC who had progressed following prior therapy with an EGFR inhibitor, testing of ctDNA for T790M demonstrated a sensitivity of 70-81% with a specificity of 58-69% with tissue results as reference standard [54, 59]. The inferior results compared to the detection of exon 19 deletions and L858R in ctDNA probably reflect a more pronounced clonal heterogeneity in the resistant situation. Of note, plasma T790M-positive patients had a similar outcome to tissue T790M-positive patients. Thus, a T790M-positive plasma genotype may obviate the need for a biopsy. In contrast, plasma T790M-negative patients would require confirmatory tissue biopsy, as plasma-negative and tumor T790M-positive patients had a favorable outcome. Of note, plasma T790M-positive and tissue-negative patients had an outcome only slightly superior to double-negative patients, again reflecting the notion that ctDNA might detect a minor subclone [54].

Patients with CRC and KRAS mutations do not benefit from anti-EGFR therapy [61, 62, 63], and mutations in BRAF, NRAS, and PIK3CA may similarly confer resistance to such treatments [64, 65, 66]. In patients with metastatic RAS wild-type CRC, the acquisition of RAS and BRAF mutations in ctDNA was associated with and often preceded acquired treatment resistance [20, 38, 45, 67, 68]. Importantly, different mutations in ctDNA can coexist in individual patients, probably reflecting clonal divergence under treatment-induced selection pressure [20, 67, 68]. Mutant clones can decline upon withdrawal of EGFR-specific antibodies, and a treatment re-challenge may lead to responses in some cases [45]. These observations demonstrate a remarkable clonal plasticity of cancer, which can be tracked using ctDNA analyses, allowing dynamic treatment stratification and pulsatile treatment strategies.

Early detection of cancer allows curative intervention and thus might improve survival. Detection rates of ctDNA differ with entity and tumor stage at the time of sampling [20]. In a study of breast, colon, pancreatic, and gastroesophageal cancer, the detection rates of ctDNA varied between 49 and 78% in patients with localized disease, and 86 and 100% in patients with metastatic disease [20]. Of all patients with stage I cancer of any type, 47% were ctDNA-positive versus 82% with stage IV. In a different study, 32% of stage I/II melanoma versus 39% of stage III/IV patients were ctDNA-positive [47]. Thus, ctDNA detection rates are determined by biology (type of cancer, site, proliferation rate, vascularization), stage, pre-analytical factors including collection, processing, and extraction of samples and the volume of plasma used for ctDNA purification, and, finally, analytical factors and conditions including the sensitivity and specificity of the assay used. In early-stage breast cancer, detection of known mutations in ctDNA by dPCR showed a sensitivity of 93.3% [69]. In a study using hybrid-capture sequencing, ctDNA analysis in single patients with localized NSCLC after radiotherapy or chemoradiotherapy successfully distinguished between residual disease and treatment-related imaging changes [30], and adding integrated digital error suppression (iDES) allowed the detection of ctDNA even in stage I NSCLC in 3 out of 3 cases [34]. In practice, early-stage cancer might be successfully detected by sensitive assays such as dPCR and the targeting of known or recurrent hot-spot mutations. In contrast, while ultrasensitive targeted sequencing or even sequencing of the entire genome from ctDNA will allow detection of any cancer mutation, it will at the same time require extensive bioinformatic analyses (‘data polishing') to achieve a high diagnostic specificity, especially in early-stage cancer. Still, the presence of a low-level ‘cancer' mutation such as TP53 or RAS in cfDNA might not necessarily indicate the presence of cancer [34, 70, 71], and the management of ctDNA-positive patients with negative imaging results would remain to be evaluated in prospective studies.

During follow-up of patients in complete remission, ctDNA analysis might allow minimal residual disease (MRD) detection, and thus might complement the information provided by imaging studies. Several studies have shown that a rise in ctDNA can precede cancer relapse by months [28, 34, 72]. Assays detecting tumor-specific translocations might offer a particular high sensitivity and specificity [73], and a small study in neuroblastoma suggests that MRD measurement of structural changes in ctDNA might not only allow prediction of relapse but might also be of prognostic significance [74]. However, prospective data demonstrating the significance of ctDNA analyses in identifying patients who are at risk of relapse after treatment of localized cancer, or in detecting a relapse early during follow-up, are missing, and thus liquid biopsy assays cannot complement regular follow-up studies at this time. In a retrospective study of 18 patients who underwent surgery for CRC with curative intent, ctDNA positivity at first follow-up after resection was highly predictive for relapse (relapse in 16/20 cases) whereas none of the 4 subjects negative for ctDNA relapsed [21]. Of note, the reliable identification of patients at risk of relapse would allow interventions, such as the stratification of adjuvant treatment. Thus, unnecessary treatment and toxicities would be avoided. In addition, extensive imaging for routine monitoring and follow-up might be cut down and adapted to individual risk.

A recently published prospective trial evaluated multitarget stool DNA testing for CRC screening. In 9,989 participants with an average risk for CRC, DNA testing had a higher sensitivity for detecting CRC but produced more false-positive results compared to a standard hemoglobin immunoassay [75]. The number needed to screen to detect 1 cancer was 154 with colonoscopy, 166 with DNA testing, and 208 with the hemoglobin immunoassay, and the effects on outcomes and costs were not examined. Clinical testing of a ctDNA assay for cancer screening would have to follow a similar study design. The significance of detecting ctDNA for early detection and screening is unclear at this time since prospective data are missing.

The presence of CTCs constitutes a strong independent prognostic factor for overall survival (OS) in small-cell lung cancer [76], breast cancer [77, 78], CRC [79], prostate cancer [80], and in neuroendocrine tumors [81]. Of note, changing treatment in patients with CTC-positive metastatic breast cancer did not improve outcome [77]; hence the feature of shedding CTCs or ctDNA might well be an independent biologic property of a cancer correlated with clinical behavior and prognosis. A number of reports support the concept that ctDNA levels provide prognostic information. In patients with stage I-IV CRC, the detection of KRAS mutations in ctDNA predicted recurrence-free survival (RFS) and OS [82]. Later studies confirmed the predictive value of ctDNA levels for RFS and OS [20, 21, 48]. In addition, these studies indicated that the amount of ctDNA might be independent of established prognostic factors (age, ECOG performance status, and CEA) [20, 21], and that the slope of ctDNA after initiation of treatment adds prognostic information [48]. In a prospective study with 230 stage II CRC patients, subjects with positive ctDNA at first follow-up had a 3-year RFS of 0% in contrast to 90% in the ctDNA-negative patients [83], suggesting that ctDNA might in the future be used to stratify for adjuvant treatment. In stage IV BRAF V600 mutation-positive melanoma, the detection of ctDNA indicated inferior progression-free survival and OS [84]. In metastatic breast cancer, quantiles of ctDNA but not CA 15-3 levels were highly predictive for OS [49]. In addition, the kinetics of mutant clones as depicted in ctDNA can be prognostic. In stage IV CRC patients receiving chemotherapy and the EGFR-specific antibody cetuximab, a 45-205× ctDNA ‘explosion' was predictive for fast disease progression and death [67]. Thus, shedding of ctDNA seems to reflect an aggressive cancer biology and the dynamics of clonal expansion and progression. In fact, shedding of tumor content might determine the progression of cancer. Data from a melanoma mouse model suggest that microvesicles containing ctDNA and RNA might direct metastatic behavior of a cancer [85].

At present, the routine clinical use of ctDNA analysis is limited to commercially available (cobas® EGFR, Roche Molecular Systems, Pleasanton, CA, USA; therascreen® EGFR, Qiagen, Hilden, Germany) or validated non-commercial tests approved by the US Food and Drug Administration (FDA) and the European Medicines Agency (EMA) to select patients with NSCLC for the appropriate EGFR-directed therapy. However, in the event of a negative ctDNA analysis, a tissue biopsy will be necessary. Other tests and applications will follow, and, if validated in prospective studies, will enter routine diagnostic use. Testing for ctDNA will be a unique tool to monitor responses and to allow non-invasive serial genotyping. The emergence of mutant clones might be detected months prior to radiographic progression. This will allow early identification of resistance and inform about the appropriate next line of treatment.

Still, many applications for liquid biopsy technologies will be investigational, and many questions remain open. To date, mutation-based biomarkers in most cases are not informative for tumor type. Thus, liquid biopsy approaches will not replace tissue-based diagnostics at baseline. In addition, the absolute sensitivity and specificity of ctDNA detection might challenge the screening and early detection of cancer, and even in advanced-stage cancer it is currently unclear whether all tumor manifestations equally contribute to ctDNA composition. To implement liquid biopsy approaches into clinical routine, the next important step will be to compare ctDNA-guided treatment stratification with standard of care in prospective randomized studies.

The author would like to thank all members of the Plasma Laboratory Freiburg as well as J. Duyster, Department of Hematology, Oncology and Stem Cell Transplantation, Medical Center University of Freiburg for their support. He would like to thank U. Breschke (Ketchum Pleon, Dresden) for providing the images used in figure 1. The author would also like to acknowledge the support of the University of Freiburg Medical Faculty, the Comprehensive Cancer Centre Freiburg (CCCF), and the German Cancer Consortium (DKTK), and the Bundesministerium für Bildung und Forschung (BMBF; grant no. 71; to N.v.B.) and Novartis Germany for research support (to N.v.B.).

N.v.B. received honoraria from Novartis and BMS and research funding from Novartis.