Emerging concepts in liquid biopsies

Here, we refer to technologies as “current” if they can be viewed as established approaches reflected in several publications describing their applicability. In contrast, “emerging technologies” are novel ideas and concepts for which proof-of-concepts or only a few applications have been published. Current technologies applied in liquid biopsy research have been extensively reviewed [9‐12] and we have therefore only briefly summarized them herein.

A great promise attributed to liquid biopsies is their possible potential to detect cancer early or even to detect precursor lesions before clinical signs occur or before sophisticated imaging systems are able to detect them. However, a major problem is the number of somatic mutations that occur in healthy individuals.

The question regarding what constitutes typical somatic variation and to what extent does it take form in terms of phenotype has gained attention by recent landmark large-scale studies [51, 52]. Interestingly, it is possible for healthy individuals to harbor disadvantageous variants without exhibiting any apparent disease phenotype [51, 52]. In fact, the identification of rare homozygotes that predicted loss of function genotypes revealed that loss of most proteins is relatively harmless to the individual [52]. The Exome Aggregation Consortium study analyzed high-quality exome sequencing data from 60,706 individuals of diverse geographic ancestry and identified 3230 genes highly intolerant to loss-of-function. Interestingly, 72% of these genes have no established human disease phenotype yet [51]. Thus, despite our growing knowledge about the human genome, identified variants require cautious interpretation with respect to the potential consequences for the phenotype.

In the context of cancer and according to the somatic mutation theory of cancer [53], malignant diseases are, to a large extent, the result of acquired genetic and epigenetic changes, which has now been extensively confirmed by NGS technologies [54, 55]. However, a tremendous challenge is the measurement of the somatic mutation rate in normal tissue and establishment of baseline values, i.e., what number of mutations is normal for a healthy person at a certain age. In general, somatic mutation rates are higher than germline mutation rates. For example, it is estimated that, in humans, the per generation rate in intestinal epithelium or fibroblasts/lymphocytes is approximately 13- and 5-fold, respectively, greater than in the germline [56].

As somatic mutations occur in individual cells, each mutation represents a low frequency event and special NGS methods for detection of such rare mutations are needed. Promising approaches include single cell genomic sequencing [6, 34, 57‐59] and applications of molecular barcodes [60, 61]. The bottleneck sequencing system is a novel technology that allows the quantification of somatic mutational load in normal human tissues, even at a genome-wide level [62]. The bottleneck is created by dilution of a sequencing library before PCR amplification, resulting in random sampling of double-stranded template molecules. This increases the signal of a rare mutation compared to wild-type sequences and thus enables the detection of mutations occurring at 6 × 10^–8 per base pair. With this approach, it was shown that, in normal colonic epithelium, mutation rates in individuals over 91 years of age had increased by an average of 30-fold in mitochondrial DNA and 6.1-fold in nuclear DNA [62]. Importantly, the spectra of rare mutations in normal colon and kidney tissues were similar to those of the corresponding cancer type [62], confirming previous reports that cancer-associated mutations may also occur in normal stem cells [63, 64].

Thus, direct measurements of mutations in adult stem cells are necessary, as the gradual accumulation of mutations in adult stem cells is thought to have an especially large impact on the mutational load of tissues due to their potential for self-renewal and capacity to propagate mutations to their daughter cells [63]. Indeed, statistical analyses have recently suggested that the total number of divisions of adult cells needed to maintain tissue homeostasis correlates with the lifetime risk of cancer [63]. However, these calculations could not exclude extrinsic risk factors as additional important determinants for cancer risk [65].

Measurement of the somatic mutation load in stem cells within various human tissues poses an immense technical problem. Blokzijl et al. [66] addressed this challenge by using cells capable of forming long-term organoid cultures. An organoid can be defined as a cellular structure containing several cell types that have developed from stem cells or organ progenitors that self-organize through cell sorting and spatially restricted lineage commitment [67]. Single adult stem cells from the small intestine, colon, and liver, tissues which differ greatly in proliferation rate and cancer risk, were expanded into epithelial organoids to obtain sufficient DNA for whole-genome sequencing. The donors ranged in age from 3 to 87 years and, not unexpectedly, it was found that stem cells accumulated mutations with age independent of tissue type [66]. The mutation rate, i.e., the increase in the number of somatic point mutations in each stem cell, was in the same range for all assessed tissues, at approximately 36 mutations per year, despite the large variation in cancer incidence among these tissues (Fig. 1a). Importantly, the findings suggested a universal genomic ageing mechanism, i.e., a chemical process acting on DNA molecules, independent of cellular function or proliferation rate. Furthermore, this intrinsic, unavoidable mutational process can cause the same types of mutations as those observed in cancer driver genes [66].

Given the high mutation rate in adult stem cells, it may be surprising that cancer incidence is not actually higher. According to the “Three Strikes and You’re Out” theory [68] (Fig. 1b), alterations in as few as three driver genes may be sufficient for a cell to evolve into an advanced cancer. However, several reasons may account for the relatively low cancer incidence. First, mutations in stem cells are non-randomly distributed and associated with depletion in exonic regions. Second, if a mutation occurs in an exonic region, it has to be in a cancer driver gene and only a small number of genes in the human genome have been shown to act as driver genes [69]. Third, the order in which driver gene mutations accumulate is important, meaning that mutations which are initiating events have to occur first [68]. Fourth, many of the initiating driver gene mutations are tissue specific; thus, the driver gene mutation must occur in the right gene and not in any driver gene.

In light of these findings, it is not surprising that cancer-associated mutations might be identified in plasma DNA from healthy persons. This was shown in a recent study that used an assay specifically designed to accurately detect TP53 mutations at very low allelic fractions, in which cfDNA TP53-mutated fragments were found in 11.4% of 123 matched non-cancer controls [70] (Fig. 1c). However, the detection of low-allele variants may be impeded by background errors occurring during library preparation and/or sequencing. To address this, approaches such as molecular barcoding and background reduction by sophisticated bioinformatics methods have been developed, as is discussed below.

In particular, a more mature understanding of the biology behind ctDNA, CTCs and exosomes will help us understand if the molecular profiles generated from these sources truly reflect the physiological disease state of the patient and if they can help physicians reliably detect and monitor the disease. In order to confirm this, we must uncover the origin and dynamics of these tumor parts in the circulation and furthermore, determine their biological significance and clinical relevance.

Although the exact mechanisms behind the release and dynamics of cfDNA remain unknown, several hypotheses exist to explain the existence of tumor DNA in the bloodstream. Perhaps the most widely accepted theory is that tumor cells release DNA via apoptosis, necrosis, or cell secretion in the tumor microenvironment [14, 93, 94]. Some cancer cases examined had detectable ctDNA levels but no detectable levels of CTCs [13]. Vice versa, a patient with an excessive number of CTCs of more than 100,000 was described, who, despite of progressive disease, had a low ctDNA allelic frequency in the range of merely 2–3% [26]. While in most patients CTC number and ctDNA levels are mutually correlated [26], such cases illustrate that exceptions exist and that the underlying biology of both CTC and ctDNA release is still poorly understood.

Other basic unknowns regarding liquid biopsy implementation in the clinic revolve around questions of whether or not ctDNA does actually indeed offer a full representation of a patient’s cancer, if all existing metastases contribute to the ctDNA, CTCs, and exosomes found in the bloodstream, or if all tumor cells release an equal amount of ctDNA into the circulation. In order to establish to what extent ctDNA represents metastatic heterogeneity, one study followed a patient with metastatic ER-positive and HER2-positive breast cancer over 3 years [95]. The genomic architecture of the disease was inferred from tumor biopsies and plasma samples and, indeed, mutation levels in the plasma samples suggested that ctDNA may allow real-time sampling of multifocal clonal evolution [95]. Conduction of warm autopsies, i.e., rapid tumor characterization within hours of death, might further help answer these questions more fully, as data derived from the tumor post-mortem could be compared to previously collected ctDNA from the patient [96].

It has furthermore been demonstrated that the percentage of ctDNA within total cfDNA can vary greatly between patients from less than 10% to greater than 50% or, as suggested more recently, can even be detected at fractions of 0.01% [13, 19, 97]. However, despite this high variability in ctDNA levels in different cancer patients, numerous studies have shown that intra-patient levels correlate with both tumor burden and progression of disease [14, 17‐20, 27, 29, 98‐102], giving evidence for the use of ctDNA levels as a proxy measurement of tumor progression and response to therapy. Accordingly, in colorectal cancer, ctDNA analyses revealed how the tumor genome adapts to a given drug schedule and liquid biopsies may therefore guide clinicians in their decision to re-challenge therapies based on the EGFR blockade [98]. For patients with NSCLC, the Food and Drug Administration approved the implementation of cfDNA in EGFR mutation analysis, through a test called the “cobas EGFR Mutation Test v2” (Roche), which serves as the first blood-based companion diagnostic to test which patients are potential candidates for the drug Tarceva (erlotinib). In a very recent study [103], this kit was used to confirm that patients treated with first-line EGFR tyrosine kinase inhibitor had acquired the EGFR T790M (p.Thr790Met) mutation, which confers resistance to first-generation EGFR tyrosine kinase inhibitors [103]. The authors then showed that NSCLC patients with this T790M mutation who were treated with osimertinib had better response rates and progression-free survival than patients treated with platinum therapy [103]. This is a beautiful example in which an invasive lung tissue biopsy was replaced by a plasma DNA-based blood test, i.e., a liquid biopsy, to identify a group of patients who could benefit from a specific treatment. This will likely propel development of further NGS-based EGFR mutation detection assays, which are of particular relevance for the Asian population in which EGFR mutation-positive lung cancers occur more frequently than in the Caucasian population [104].

However, before liquid biopsies can serve as viable diagnostic assays, pre-analytical steps, such as the collection of biofluid (e.g., blood, serum, plasma), centrifugation settings, isolation reagents, and storage conditions, must be standardized in order to ensure reproducible processing procedures. Furthermore, analytical steps, such as quantification of cfDNA and subsequent mutational analysis, i.e., the NGS assay and sequencing platform itself, must be validated to simulate clinical settings. In addition, sensitivities and specificities of the applied assays must be robust, reproducible, and have the appropriate internal and external quality controls [72]. Perhaps the most imperative step is the need to evaluate the clinical relevance of ctDNA at various time points depending on the application, such as patient stratification, evaluation of treatment response, efficacy, and resistance, as well as validating this data in large multicenter clinical studies [72]. Furthermore, the clinical performance of cfDNA assays must satisfy the requirements of the respective regulative agencies, such as Clinical Laboratory Improvement Amendments in the USA or the genetic testing practices in European countries. In Europe, efforts to harmonize liquid biopsy testing are supported by CANCER-ID, a European consortium supported by Europe’s Innovative Medicines Initiative, which aims at the establishment of standard protocols for and clinical validation of blood-based biomarkers (www.​cancer-id.​eu/​).