How circulating tumor DNA results are quantified, cross-checked against tissue, and routed into molecular tumor board decisions
Every liquid biopsy begins with an ordinary blood draw — no needle biopsy, no imaging suite, no sedation. But turning 10 mL of whole blood into analyzable tumor genetic material requires careful preanalytical handling, because cell-free DNA (cfDNA) is a fragile, low-abundance, and easily contaminated analyte. Get the plasma separation wrong and the tumor signal is drowned out by DNA released from lysed white blood cells during shipping.
Cell-free DNA is not actively secreted — it is the wreckage of cell death. As cells throughout the body undergo apoptosis, and to a lesser extent necrosis, nucleases chew their chromatin into short fragments that leak into the bloodstream. Because DNA wrapped around a nucleosome is protected from nuclease digestion, the resulting fragments cluster tightly around ~166 bp (147 bp nucleosome core + linker), producing a characteristic fragmentation "fingerprint" that itself carries diagnostic information (fragmentomics).
In a cancer patient, a small fraction of this circulating cfDNA pool is shed specifically by tumor cells — this is circulating tumor DNA (ctDNA). Tumors shed more ctDNA than their cell count alone would predict, likely because malignant cells turn over faster, are more prone to apoptosis/necrosis in poorly vascularized regions, and because tumor-associated macrophages actively clear dying tumor cells and release their DNA. Highly vascularized, large, or necrotic tumors shed more; small, indolent, or poorly vascularized tumors (and some tumor types, like CNS malignancies behind the blood-brain barrier) shed very little — a key limitation of the entire liquid biopsy field.
The single greatest source of error in liquid biopsy is not the sequencing — it is what happens to the tube between the phlebotomist and the lab.
• Collection tubes: standard EDTA tubes must be processed within hours because white blood cells lyse over time, flooding the plasma with normal genomic DNA that dilutes the ctDNA fraction. Cell-stabilizing tubes (Streck Cell-Free DNA BCT, PAXgene) contain a fixative that keeps leukocytes intact for 7+ days at room temperature, enabling shipment to central labs.
• Double-spin plasma separation: blood is first centrifuged at low speed (~1,600×g) to pellet cells while keeping plasma on top, then the plasma is transferred and centrifuged again at high speed (~16,000×g) to remove residual cellular debris and platelets. Skipping the second spin is a common cause of falsely elevated background.
• Extraction: silica-membrane or magnetic-bead columns (QIAamp Circulating Nucleic Acid Kit, MagMAX) purify cfDNA from ~4–10 mL of plasma, typically yielding only 5–50 ng/mL of plasma — often just 10–100 ng total input for the entire downstream sequencing library. This scarcity is why ultra-sensitive library chemistry (unique molecular identifiers, error-corrected sequencing) matters so much more than in tissue NGS, where micrograms of DNA are routinely available.
A poorly handled tube — delayed processing, single-spin plasma, freeze-thaw of whole blood — can inflate background wild-type DNA by 10–100×, pushing a true low-level ctDNA signal below the assay's limit of detection. Preanalytical rigor is the invisible foundation of every ctDNA result a tumor board sees.
Isolated cfDNA is not pure tumor material — it is a mixture, typically overwhelmingly dominated by normal hematopoietic cfDNA shed from turning-over blood cells. Before any mutation can be interpreted clinically, the lab must estimate what fraction of the total cfDNA pool is tumor-derived. This "ctDNA fraction" (also called tumor fraction) is the single most important quality metric on a liquid biopsy report — it calibrates how much confidence to place in both positive and negative findings.
Variant allele fraction (VAF) approach: if a known somatic mutation is detected at, say, 2% VAF in plasma, and the tumor is roughly diploid at that locus, the ctDNA fraction is approximately 2× the VAF (accounting for one mutant allele out of two). This works well when a clonal, truncal mutation is already known (e.g., from prior tissue sequencing) but is confounded by tumor ploidy, subclonality, and copy number state at the mutated locus.
Genome-wide aneuploidy approach (e.g., ichorCNA): most solid tumors carry large-scale copy number alterations — whole chromosome arm gains/losses that are nearly universal in a given cancer's genome even without knowing any specific point mutation. By shallow whole-genome sequencing of cfDNA (~0.1× coverage, very cheap) and measuring genome-wide deviations in read depth across 1 Mb bins, algorithms like ichorCNA estimate tumor fraction independent of any single mutation — useful even when no driver mutation is yet known, and as an orthogonal cross-check on VAF-based estimates.
Commercial reports (Guardant360, FoundationOne Liquid CDx, Signatera) typically report both a per-variant VAF and an overall "maximum somatic allele frequency" (MSAF) as a proxy for tumor fraction, flagging low-MSAF samples as having reduced sensitivity for detecting all clinically relevant alterations.
Guardant360 CDx: 74-gene hybrid-capture panel (single nucleotide variants, indels, fusions, amplifications), FDA-approved as a companion diagnostic across several solid tumor indications, widely used for real-time treatment selection.
FoundationOne Liquid CDx: 324-gene panel, also profiles tumor mutational burden (bTMB) and microsatellite instability (MSI) directly from plasma, FDA-approved for multiple companion diagnostic claims across NSCLC, breast, prostate, and ovarian cancer.
Digital droplet PCR (ddPCR): partitions a sample into ~20,000 nanoliter droplets, each undergoing independent PCR; presence/absence of amplification per droplet gives absolute quantification of mutant vs. wild-type molecules. Far cheaper and faster than NGS panels, with sensitivity down to ~0.01% VAF, but limited to pre-specified hotspot mutations — ideal for serial monitoring of a single known resistance mutation once identified.
Unique molecular identifiers (UMIs): random barcode sequences ligated to each cfDNA molecule before PCR amplification, allowing bioinformatic collapse of PCR duplicates and distinguishing true low-frequency mutations from polymerase and sequencing errors — the key technology enabling NGS panels to approach ddPCR-level sensitivity across hundreds of genes simultaneously.
A molecular tumor board rarely sees a liquid biopsy result in isolation — it is almost always interpreted alongside the original tissue NGS report. Concordance between the two is reassuring and can substitute for a repeat tissue biopsy; discordance is not necessarily an error, and figuring out why the two disagree is often the most clinically important part of the discussion.
Discordance is the norm, not the exception, and stems from distinct sources on each side:
Tissue-only ("false negative" in plasma): the tumor is shedding too little ctDNA to reach a given mutation's detection threshold (low tumor fraction), the biopsied tumor region harbors a subclone not well represented in the shed DNA, or the mutation type (e.g., certain structural rearrangements, large indels) is poorly captured by the plasma assay's chemistry.
Plasma-only ("false positive" relative to tissue," but often real biology): the original tissue biopsy sampled only one region of a spatially heterogeneous tumor and missed a subclone that the whole-body ctDNA pool captures; the mutation arose after the tissue biopsy was taken (acquired resistance); or — critically — the variant is not tumor-derived at all.
Clonal hematopoiesis of indeterminate potential (CHIP): with age, hematopoietic stem cells acquire somatic mutations in genes like DNMT3A, TET2, ASXL1, and even TP53 — clonal blood cell populations that are not cancer but shed mutant DNA into plasma indistinguishable, on a plasma-only panel, from a tumor-derived variant. CHIP prevalence rises sharply with age (>10% of people over 70) and increases further after chemotherapy. The gold-standard mitigation is paired sequencing of the patient's white blood cells (buffy coat) alongside plasma — any variant present in both is CHIP, not cancer, and should be filtered from the tumor board discussion.
A canonical trap: TP53 mutation detected in plasma at low VAF, absent from the original tumor tissue report, in a 74-year-old post-chemotherapy patient. Before flagging this as tumor evolution or a resistance mechanism, the molecular tumor board must ask for matched white-blood-cell sequencing to rule out CHIP — otherwise a hematologic bystander gets mistaken for an oncologic actor.
Repeat tissue biopsies carry procedural risk, cost, and delay — and are sometimes technically infeasible (inaccessible lesion, poor performance status, insufficient tissue yield). When a plasma panel identifies an actionable variant that is either concordant with historical tissue or biologically plausible as tumor evolution (e.g., a known resistance mutation emerging under selective pressure from a targeted therapy), many tumor boards will act on the plasma result directly, particularly when tumor fraction is adequately high (reducing false-negative risk) and the mutation is a well-established, recurrent driver rather than a novel variant of uncertain significance.
Conversely, a plasma-negative result in a low-ctDNA-fraction sample is uninterpretable as a true negative — absence of evidence, given low tumor shedding, is not evidence of absence — and in that scenario tissue re-biopsy (or MRI-guided approaches) may still be recommended before ruling out an actionable alteration.
Unlike a single tissue biopsy — a snapshot frozen at one moment — serial liquid biopsies turn tumor genotyping into a movie. Repeated blood draws every 6–8 weeks during targeted therapy let clinicians watch a resistance mutation's variant allele fraction rise from undetectable to dominant, often long before a CT or MRI scan shows measurable tumor growth.
Targeted therapies apply strong selective pressure on a tumor's cell population. A pre-existing or newly acquired resistance mutation confers a survival advantage to the subclone carrying it, which then expands while drug-sensitive clones are suppressed. Because ctDNA is shed proportionally to the size and turnover of each subclone in the body, the VAF of a resistance mutation tracked over serial draws is effectively a real-time readout of that subclone's expanding population fraction — before it is large enough to be visible as a growing mass on imaging.
Canonical examples routinely discussed at molecular tumor boards:
• EGFR-mutant NSCLC on osimertinib (3rd-generation EGFR TKI): resistance can emerge via EGFR C797S (which, depending on whether it occurs in cis or trans with T790M, dictates sensitivity to different drug combinations), MET amplification, or histologic transformation to small-cell carcinoma. Serial ctDNA distinguishes these mechanisms without a repeat biopsy in many cases.
• ER+ breast cancer on aromatase inhibitors: ESR1 ligand-binding-domain mutations (e.g., Y537S, D538G) confer estrogen-independent receptor activation and resistance to aromatase inhibition; their emergence in serial ctDNA prompts a switch to selective estrogen receptor degraders (e.g., fulvestrant, elacestrant) that remain effective against mutant ESR1.
• Colorectal cancer on anti-EGFR therapy (cetuximab/panitumumab): RAS pathway mutations (KRAS, NRAS) frequently emerge as resistance, detectable in ctDNA weeks to months before progression — and importantly, some studies show these resistant clones can decay after anti-EGFR therapy is stopped, raising the possibility of ctDNA-guided re-challenge.
A rising resistance-mutation VAF on serial monitoring is a molecular early-warning signal, but acting on it requires clinical judgment, not just a lab value:
1. Confirm the trend: a single rising data point can reflect assay noise near the limit of detection; tumor boards generally want at least two consecutive rises, or a clear trajectory, before treating it as a true signal.
2. Correlate with clinical status: is the patient having new symptoms? Is imaging due soon? A molecular signal in an asymptomatic patient with stable imaging is treated differently than the same signal accompanying new symptoms.
3. Mechanism-match the next therapy: if a specific resistance mutation is identified (e.g., MET amplification under osimertinib), a molecularly matched combination (osimertinib + MET inhibitor) may be considered even before RECIST progression is documented — an increasingly studied strategy sometimes called "molecular progression"-triggered switching, still largely investigational outside of clinical trials.
4. Avoid over-treating noise: switching therapy based on a borderline VAF rise risks abandoning an effective regimen prematurely; most current guidelines still anchor major therapy changes to radiographic or clinical progression, with ctDNA trends used as a prompt for closer monitoring rather than an automatic trigger.
In the AURA3 and FLAURA osimertinib trials and subsequent real-world cohorts, detectable ctDNA-based resistance mutations (T790M loss, C797S gain, MET amplification) were frequently identified 2–6 months before RECIST-defined radiographic progression — a window during which a molecular tumor board can pre-plan the next line of therapy rather than reacting after the scan.
After a curative-intent surgery removes all visible tumor, the single biggest open question is whether microscopic disease remains, invisible to any scan. Minimal residual disease (MRD) ctDNA testing was built to answer exactly this question — and its result is now one of the most consequential single data points a molecular tumor board reviews, directly shaping whether a patient receives additional chemotherapy or is spared it.
Tumor-informed assays (e.g., Signatera, RaDaR): the patient's own resected tumor tissue is whole-exome or panel-sequenced first, identifying a set of ~16–50 somatic mutations unique to that individual's cancer. A patient-specific, multiplex PCR or hybrid-capture NGS panel is then custom-built to track exactly those mutations in subsequent plasma draws. Because the assay only looks for mutations already proven to exist in that patient's tumor, background noise (CHIP, sequencing error) is dramatically reduced, enabling detection sensitivity down to roughly one mutant molecule in 100,000–1,000,000 wild-type molecules (~0.01% VAF or lower).
Tumor-naive (fixed panel) assays (e.g., Guardant Reveal): use a pre-fixed gene panel plus epigenomic (methylation) signals to detect ctDNA without requiring prior tumor tissue sequencing. This trades some sensitivity for speed and lower up-front cost — clinically valuable when tissue is unavailable or a rapid turnaround is needed, but generally less sensitive than tumor-informed approaches, particularly at very low tumor fractions typical of the post-surgical, sub-clinical disease state.
MRD testing is typically first performed 2–4 weeks after surgery — early enough to inform adjuvant therapy timing, but late enough to let surgery-related cfDNA (released by the trauma of the operation itself) clear from circulation; testing too early risks a transient false positive from surgical inflammation rather than true residual tumor.
MRD-positive: detection of tumor-derived ctDNA after complete surgical resection implies that malignant cells persist somewhere in the body below the resolution of any imaging study. This is strongly associated with eventual clinical recurrence — hazard ratios of roughly 10 to 40-fold higher recurrence risk compared to MRD-negative patients have been reported across colorectal, breast, and lung cancer cohorts. At tumor board, an MRD-positive result typically supports escalating or continuing adjuvant chemotherapy, and prompts closer surveillance imaging intervals.
MRD-negative: no tumor-derived ctDNA detected. While not an absolute guarantee of cure — sensitivity is not 100%, and very small or slow-shedding residual disease can be missed — a persistently MRD-negative status is increasingly used to support de-escalation, i.e., withholding or shortening adjuvant chemotherapy in patients who would otherwise receive it "just in case." This is precisely the question being tested prospectively in trials such as DYNAMIC (stage II colon cancer) and GALAXY/CIRCULATE (stage II–III colorectal cancer), which are reshaping adjuvant treatment guidelines toward ctDNA-guided, rather than purely stage-guided, therapy.
In the GALAXY observational study (part of the CIRCULATE-Japan program, >2,000 colorectal cancer patients), ctDNA MRD-positive patients had a 3-year disease-free survival of roughly 30–40%, compared to over 90% in MRD-negative patients — and ctDNA clearance after adjuvant chemotherapy in initially MRD-positive patients was itself associated with dramatically better outcomes, establishing ctDNA dynamics as both a prognostic marker and a real-time readout of adjuvant treatment effectiveness.