Part VII · Detection, diagnosis, and biomarker assessment · Chapter 40

Liquid biopsy and longitudinal monitoring

Circulating tumour DNA tells you that residual disease is present. Nobody has yet shown that acting on it early makes a patient live longer.

1 · Biology of circulating tumor DNA, shedding, half-life, and detectability

Plasma always contains cell-free DNA. Most of it comes from haematopoietic cells. The tumour-derived fraction is a minority of that pool, and in early disease it is a very small minority.

Cell-free DNA has a short circulating half-life, measured in hours rather than days. That single property is what makes the assay useful. A plasma sample reports what is shedding now, not what was shedding last year. It is the opposite of an archival block.

Shedding is not proportional to tumour volume. It depends on cell turnover, on necrosis and apoptosis within the deposit, on vascular access, and on where the deposit sits. A proliferating liver metastasis contributes far more DNA per gram than a quiescent bone deposit does.

The site effect can be quantified. In a prospective cohort followed with serial plasma sampling, circulating tumour DNA detected 22 of 23 distant extracranial metastatic relapses, which is 96%. It detected 1 of 6 brain-only relapses, which is 17%¹. The central nervous system is a separate compartment, and Cerebrospinal fluid analysis in leptomeningeal disease takes that up.

Detectability is also a property of the assay, and the two should never be reported as one. An ultrasensitive whole-genome-based platform tracking a median of 1,451 variants per patient detected circulating tumour DNA before treatment in 49 of 50 patients with early breast cancer, which is 98%. Thirty-nine per cent of all its detections sat below 100 parts per million of the cell-free DNA pool². A less sensitive assay run on the same plasma would have called most of those patients negative.

2 · Circulating tumor cells and extracellular vesicles

Circulating tumour cells were the first blood-based prognostic marker in this disease, and their history is the cautionary tale that governs the rest of this chapter.

In 177 patients with measurable metastatic breast cancer starting a new line of treatment, a count of 5 or more cells per 7.5 mL of whole blood at baseline separated two populations. Median progression-free survival was 2.7 months against 7.0 months. Median overall survival was 10.1 months against more than 18 months³. The count was an independent predictor in multivariable analysis.

Then the strategy was tested rather than assumed. SWOG S0500 enrolled 595 eligible patients starting first-line chemotherapy. Those whose counts remained elevated after 21 days were randomised to continue or to switch to an alternative cytotoxic agent. Median overall survival was 10.7 months in the continuation arm and 12.5 months in the switch arm, with a P value of .98⁴. Across the three prognostic groups defined by the counts, median overall survival was 35, 23 and 13 months.

Read those two results together. The marker was strongly prognostic. Acting on it changed nothing. That is the exact shape of the question this chapter carries, and it was answered negatively for cells more than a decade before it was asked for DNA.

The cells themselves are biologically informative. Their epithelial and mesenchymal composition shifts dynamically during treatment⁵. Clusters are oligoclonal and carry greater metastatic potential than single cells⁶. The seeding biology belongs to Circulating tumor cells and the inefficiency of seeding, and the platform-dependence of counts to What is actionable today and what remains a research instrument.

Extracellular vesicles are a third analyte. Tumour-derived exosomes carry integrin repertoires that help determine which organ a tumour seeds⁷, which is developed in Premetastatic niche formation. No vesicle-based assay has established clinical validity in breast cancer, and none has a management decision attached to it.

3 · Molecular residual disease, tumor-informed and tumor-naive approaches

Molecular residual disease detection by plasma comes in two designs, and the difference is where the variant list comes from.

A tumour-informed assay sequences the resected primary, selects variants specific to that tumour, and builds a personalised assay to hunt for them. A tumour-naive or tissue-free assay uses a fixed panel or a methylation classifier, and needs no tumour tissue at all.

The tumour-informed evidence is the older and the larger. In 55 patients with early breast cancer receiving neoadjuvant chemotherapy, detection at a single postsurgical time point carried a hazard ratio for relapse of 25.1, with a 95% confidence interval of 4.08 to 130.5. Serial follow-up sampling gave a hazard ratio of 12.0, and a median lead time over clinical relapse of 7.9 months⁸. A larger validation cohort reproduced it. Among 101 women followed prospectively, detection during follow-up carried a hazard ratio for relapse of 25.2, with a 95% confidence interval of 6.7 to 95.6¹.

Sensitivity scales with how many variants are tracked. In 61 patients, a next-generation sequencing assay tracking up to 50 patient-specific variants gave a hazard ratio for relapse of 37.2, with a 95% confidence interval of 10.5 to 131.9. Median lead time over clinical relapse was 11.7 months⁹. A whole-genome-powered assay tracking a median of 1,451 variants identified residual disease in 11 of 11 patients who relapsed, and none of 64 patients in whom it was never detected relapsed during follow-up².

The tissue-free comparison has now been done head to head within one cohort. Across 1,026 plasma samples from 159 women with triple-negative disease, a methylation-based tissue-free assay detected circulating tumour DNA in 54 patients, which is 34.0%. The hazard ratio for recurrence was 27.2, with a 95% confidence interval of 13.7 to 54.2¹⁰. Against single-variant digital PCR the tissue-free assay was earlier, with median lead times of 7.9 against 5.8 months. Against a multivariant whole-exome-powered tumour-informed assay it was not, with lead times of 7.6 against 7.1 months. Among 41 patients detected by both, 12 were found earlier by the tumour-informed assay and 1 earlier by the tissue-free one.

The practical reading is narrow and useful. Tissue-free assays work when no tissue is available, and they beat a single-variant approach. The most analytically sensitive assays remain the ones that track many tumour-specific variants, and the number of variants tracked should be on the report.

4 · Lead time and the unresolved question of whether acting early helps

This is the section the chapter exists for, and the honest answer is uncomfortable.

Molecular residual disease detected in plasma is strongly prognostic across every subtype that has been examined. The hazard ratios in Molecular residual disease, tumor-informed and tumor-naive approaches are not marginal effects. They are among the largest prognostic separations reported for any biomarker in this disease.

Lead time is real and it is subtype-dependent. One multicentre study followed 83 patients with residual disease after neoadjuvant chemotherapy, using personalised assays against structural variants. Detection carried a hazard ratio for relapse-free interval of 29.2, with a 95% confidence interval of 6.9 to 124.0¹¹. Median lead time over clinical relapse was 13.9 months overall. It was 20.9 months in oestrogen receptor-positive HER2-negative disease, 14.2 months in HER2-positive disease, and 4.3 months in triple-negative disease. Sensitivity for locoregional or distant relapse was 25 of 27 patients, which is 92.6%. Among patients with a negative postsurgical sample, 5-year relapse-free interval was 72.6%, with a 95% confidence interval of 60.1% to 81.6%.

Those subtype differences are biology, not assay performance. Fast-growing disease gives less warning because the interval between detectable shedding and clinical relapse is shorter.

Now the negative. c-TRAK TN prospectively surveilled 161 women with moderate-risk and high-risk early triple-negative breast cancer, sampling three-monthly, and offered pembrolizumab to those who became positive. The detection rate by 12 months was 27.3%, which was 44 of 161, with a 95% confidence interval of 20.6% to 34.9%. Of the patients allocated to intervention, 72% already had metastases on staging imaging at the moment circulating tumour DNA was detected, which was 23 of 32. Of the five patients who actually started pembrolizumab, none achieved sustained clearance of circulating tumour DNA¹². Seven patients relapsed with no prior detection at all.

Three separate lessons sit inside that result. Surveillance started too late, after adjuvant therapy had finished. Sampling was too infrequent for a subtype whose lead time is around four months¹¹. And the intervention did not work on the disease it found.

In the advanced setting the same design has succeeded twice. PADA-1 monitored 1,017 women on first-line aromatase inhibitor and palbociclib for a rising ESR1 mutation in blood. Of those, 279 developed one, which is 27%, and 172 without synchronous progression were randomised. Median progression-free survival from randomisation was 11.9 months in those switched to fulvestrant and palbociclib, against 5.7 months in those continued. The stratified hazard ratio was 0.61, with a 95% confidence interval of 0.43 to 0.86 and a P value of 0.0040¹³. SERENA-6 then confirmed the principle with a next-generation degrader, and Acting on evolution, intermittent dosing, adaptive therapy, pre-emptive switching develops that result rather than repeating it here.

The two settings are not the same experiment, and conflating them is the error to avoid. In advanced disease the patient is already on treatment, the molecular signal triggers a switch, and an effective next agent exists. After curative-intent treatment the patient is on no therapy or on endocrine therapy alone, and no agent has been shown to eradicate molecularly detected residual disease. DARE, LEADER and SURVIVE are testing that, alongside TRAK-ER and PERSEVERE. None has reported a survival result.

Guidelines reflect this. ASCO concluded that there are insufficient data to recommend routine use of circulating tumour DNA or circulating tumour cells to monitor response to therapy in metastatic disease¹⁴. Its 2026 follow-up and surveillance update identified no randomised trial evaluating different follow-up approaches by risk of relapse, and issued consensus-based rather than evidence-based recommendations¹⁵. Molecular residual disease surveillance is not recommended outside a trial.

5 · Monitoring response and detecting emerging resistance

Serial plasma measures whether the treated population is shrinking. Whether that measurement is useful depends entirely on which variant is being tracked.

In PALOMA-3, the relative change in PIK3CA circulating tumour DNA after 15 days of treatment strongly predicted progression-free survival on palbociclib with fulvestrant. The hazard ratio was 3.94, with a log-rank P value of 0.0013. ESR1 dynamics in the same patients predicted clinical outcome only weakly¹⁶.

The explanation is clonal structure rather than assay failure. PIK3CA mutations in this population are typically truncal, so their abundance tracks the bulk of the disease. ESR1 mutations selected by prior aromatase inhibitor exposure are frequently subclonal, so their abundance tracks one compartment¹⁶. A falling ESR1 level can coexist with a growing ESR1 wild-type population.

The same samples show what the treatment selected. RB1 mutations emerged only in the palbociclib arm, in 6 of 127 patients¹⁷. Evolution of driver mutations was uncommon in patients who progressed early and common in those who progressed late. That argument belongs to CDK4/6 inhibition and the evolution of cell cycle escape and Emergence of ESR1, PIK3CA, RB1, CCNE1, and FAT1 alterations on treatment and is not repeated here.

Baseline plasma metrics carry independent prognostic weight in advanced hormone receptor-positive disease¹⁸. Prognosis is not the same claim as utility.

6 · Mutation tracking for ESR1 and PIK3CA to guide treatment change

Two alterations currently justify plasma genotyping in routine practice, and they justify it for different reasons.

ESR1 mutations are acquired under oestrogen deprivation. In 171 women with advanced disease, they were detected exclusively in oestrogen receptor-positive patients with prior aromatase inhibitor exposure. Patients with a detectable mutation had shorter progression-free survival on subsequent aromatase inhibitor based therapy than patients without one. The hazard ratio was 3.1, with a 95% confidence interval of 1.9 to 23.1 and a P value of 0.0041^19,20. That interval is wide enough that the point estimate should not be quoted alone. Testing before that exposure tests before the selection has happened. The biology is in Estrogen receptor signaling and endocrine resistance and the treatment consequences in Oral selective estrogen receptor degraders and their biomarker-defined benefit.

Plasma genotyping for these alterations is analytically mature. In plasmaMATCH, 1,034 of 1,051 registered patients returned a circulating tumour DNA result. Agreement between digital PCR and targeted sequencing was 96% to 99%, with kappa values of 0.89 to 0.93 across 800 samples. Sensitivity against mutations identified by tissue sequencing was 93%, with a 95% confidence interval of 83% to 98%²¹.

Finding the mutation is solved. What follows from finding it is not uniform. In the same trial, extended-dose fulvestrant in ESR1-mutant disease produced confirmed responses in 6 of 74 patients, which is 8%, and missed its target. Neratinib in HER2-mutant disease produced 5 responses in 20 patients, and capivasertib with fulvestrant in AKT1-mutant oestrogen receptor-positive disease produced 4 in 18²¹. The assay was equally good in all four cohorts. The drugs were not.

PIK3CA testing carries a specific asymmetry that should be stated explicitly. ASCO recommends testing tumour tissue or plasma to determine eligibility for a PI3K inhibitor, and recommends tissue testing if no mutation is found in plasma and tissue is available¹⁴. A positive plasma result is actionable. A negative plasma result is not conclusive, because it may only mean the tumour is not shedding.

7 · Cerebrospinal fluid analysis in leptomeningeal disease

The central nervous system is a separate compartment, and plasma reports it badly. That is a statement about anatomy rather than about assay quality.

Cerebrospinal fluid cytology is the conventional diagnostic standard, and its sensitivity is imperfect, which is why repeated lumbar puncture is often needed to confirm or exclude the diagnosis. Measuring the circulating tumour DNA fraction in cerebrospinal fluid by ultra-low-pass whole-genome sequencing performs better. In 30 patients under investigation for breast cancer leptomeningeal metastasis, a fraction of at least 0.10 was found in all 24 patients with confirmed disease, with a median fraction of 0.57. It was absent in the 6 patients without it, whose median fraction was 0.03, with a P value below 0.0001²². Detection held despite negative cytology or borderline magnetic resonance imaging.

The compartment separation is visible in the same study. Plasma circulating tumour DNA was detected only in patients with extracranial disease progression or previous whole-brain radiotherapy²². A patient with leptomeningeal disease and controlled systemic disease can have an undetectable plasma result.

Serial measurement behaved like a response marker. Among 12 patients monitored during intrathecal treatment, suppression of the cerebrospinal fluid fraction was associated with longer survival, with a P value of 0.034. A rising fraction was detectable up to 12 weeks before clinical progression²².

Clinical management of the syndrome itself belongs to Leptomeningeal disease and whether delivery has to be intrathecal.

8 · Discordance between blood, tissue, and imaging

Three assays sample three different things. Blood integrates every shedding deposit. A biopsy characterises one deposit in detail. A scan sees every deposit and characterises none of them. A disagreement between them usually identifies which sampling frame failed, not which result was wrong.

Most blood-tissue discordance is temporal. In plasmaMATCH, sensitivity of plasma digital PCR against mutations identified in tissue was 93% overall, and 98% when the tissue biopsy was contemporaneous, with a 95% confidence interval of 87% to 100%²¹. The gap between those two numbers is the archive. A mutation absent from plasma but present in an old block may have been selected out, and a mutation present in plasma but absent from the block may have been selected in.

Shedding produces the second class of discordance. Low-volume disease, bone-only disease and intracranial disease all contribute less DNA than their apparent burden implies^1,22. A negative plasma result in that setting is uninformative rather than reassuring.

Contamination from non-tumour clones produces the third class, and Clonal hematopoiesis and other false positives develops it²³.

Plasma reports DNA and only DNA. A receptor conversion driven by transcriptional or post-translational change is invisible to it, which is why Receptor discordance and conversion still requires tissue.

Imaging adds a fourth axis. In 56 patients with HER2-positive metastatic disease, pretreatment zirconium-89 trastuzumab positron emission tomography was negative in 29%, and a mixed pattern within a single patient was found in 46%²⁴. Nearly half of a population defined as HER2-positive by a tissue assay carried lesions that did not bind the antibody. Imaging-based heterogeneity, radiomics, PET metrics, and HER2 PET develops the imaging argument.

9 · Clonal hematopoiesis and other false positives

Most of the cell-free DNA in plasma comes from white blood cells, so most of the mutations found in plasma come from white blood cells too.

The magnitude was established by sequencing plasma alongside matched white blood cell DNA. Among 47 controls without cancer and 124 patients with metastatic cancer, 81.6% of the cell-free DNA mutations in controls and 53.2% of those in patients had features consistent with clonal haematopoiesis²³. More than half of what a plasma assay calls in a cancer patient can come from the blood rather than the tumour.

Clonal haematopoiesis is common, it increases with age, and it is associated with adverse outcomes independent of cancer^25,26. Cytotoxic chemotherapy and radiotherapy select for specific clones, particularly those carrying PPM1D and TP53 mutations²⁷. That makes the problem worse in exactly the population under surveillance, because these patients have just finished chemotherapy.

The control is matched white cell sequencing. Its absence from an assay is a design choice, and it should be visible on the report.

Tumour-informed assays are largely immune to this failure, and that is their strongest argument. Their variant list comes from the resected tumour, so a haematopoietic clone that was never in the tumour cannot generate a positive result. A tumour-naive panel has no such protection.

10 · Implementation thresholds, regulation, and reimbursement

A limit of detection is a property of an assay at a stated plasma input and a stated variant count. It is not a property of the patient. A result reported as negative means no signal was found at that sensitivity, from that volume, on that day.

That makes reporting standards the practical issue. A circulating tumour DNA report should state six things: the assay design, whether it is tumour-informed or tissue-free, how many variants were tracked, the plasma volume, the limit of detection, and whether matched white cell sequencing was performed. A result without those fields cannot be compared to a result from a different laboratory, and serial results from different platforms cannot be compared at all.

The evidence position and the commercial position have diverged, and the tension should be stated rather than smoothed. ASCO found insufficient data to recommend routine circulating tumour DNA monitoring of treatment response in metastatic disease¹⁴. Its 2026 surveillance guideline found no randomised trials comparing follow-up strategies by relapse risk, and produced consensus recommendations instead¹⁵. Coverage and reimbursement decisions are made by different bodies against different standards, and assays are available and paid for in settings where no trial has shown that acting on the result helps.

Both facts are true at once. The test performs as advertised. The action that should follow a positive result has not been established.

References

All 27 references

1. Garcia-Murillas I, Chopra N, Comino-Mendez I, et al. Assessment of molecular relapse detection in early-stage breast cancer. JAMA Oncol 2019 5:1473-1478. PMID 31369045
2. Garcia-Murillas I, Abbott CW, Cutts RJ, et al. Whole genome sequencing-powered ctDNA sequencing for breast cancer detection. Ann Oncol 2025 36:673-681. PMID 39914664
3. Cristofanilli M, Budd GT, Ellis MJ, et al. Circulating tumor cells, disease progression, and survival in metastatic breast cancer. N Engl J Med 2004 351:781-791. PMID 15317891
4. Smerage JB, Barlow WE, Hortobagyi GN, et al. Circulating tumor cells and response to chemotherapy in metastatic breast cancer: SWOG S0500. J Clin Oncol 2014 32:3483-3489. PMID 24888818
5. Yu M, Bardia A, Wittner BS, et al. Circulating breast tumor cells exhibit dynamic changes in epithelial and mesenchymal composition. Science 2013 339:580-584. PMID 23372014
6. Aceto N, Bardia A, Miyamoto DT, et al. Circulating tumor cell clusters are oligoclonal precursors of breast cancer metastasis. Cell 2014 158:1110-1122. PMID 25171411
7. Hoshino A, Costa-Silva B, Shen TL, et al. Tumour exosome integrins determine organotropic metastasis. Nature 2015 527:329-335. PMID 26524530
8. Garcia-Murillas I, Schiavon G, Weigelt B, et al. Mutation tracking in circulating tumor DNA predicts relapse in early breast cancer. Sci Transl Med 2015 7:302ra133. PMID 26311728
9. Garcia-Murillas I, Cutts RJ, Walsh-Crestani G, et al. Longitudinal monitoring of circulating tumor DNA to detect relapse early and predict outcome in early breast cancer. Breast Cancer Res Treat 2024 209:493-502. PMID 39424680
10. Cunningham N, Cutts RJ, Swift C, et al. Tissue-free vs tumor-informed ctDNA assays for molecular residual disease detection in early triple negative breast cancer. JAMA Oncol 2026. PMID 42593771
11. Blaye C, Bidard FC, Hue L, et al. Tracking structural variants in early breast cancer using a high-sensitivity ctDNA assay predicts relapse in the post-neoadjuvant setting: the multicenter ALIENOR trial. ESMO Open 2026 11:107766. PMID 42242004
12. Turner NC, Swift C, Jenkins B, et al. Results of the c-TRAK TN trial: ctDNA mutation tracking to detect molecular residual disease and trigger intervention in moderate- and high-risk early-stage triple-negative breast cancer. Ann Oncol 2023 34:200-211. PMID 36423745
13. Bidard FC, Hardy-Bessard AC, Dalenc F, et al. Switch to fulvestrant and palbociclib versus no switch in advanced breast cancer with rising ESR1 mutation during aromatase inhibitor and palbociclib therapy (PADA-1). Lancet Oncol 2022 23:1367-1377. PMID 36183733
14. Henry NL, Somerfield MR, Dayao Z, et al. Biomarkers for systemic therapy in metastatic breast cancer: ASCO guideline update. J Clin Oncol 2022 40:3205-3221. PMID 35759724
15. Nahleh Z, Alfano CM, Somerfield MR, et al. Breast cancer follow-up and surveillance after primary treatment: ASCO guideline update. J Clin Oncol 2026. PMID 42462191
16. O'Leary B, Hrebien S, Morden JP, et al. Early circulating tumor DNA dynamics and clonal selection with palbociclib and fulvestrant for breast cancer. Nat Commun 2018 9:896. PMID 29497091
17. O'Leary B, Cutts RJ, Liu Y, et al. The genetic landscape and clonal evolution of breast cancer resistance to palbociclib plus fulvestrant in the PALOMA-3 trial. Cancer Discov 2018 8:1390-1403. PMID 30206110
18. Dobric N, Hasenleithner SO, Suppan C, et al. Integrating baseline ctDNA-derived tumor metrics enhances risk stratification in HR-positive/HER2-negative advanced breast cancer: a real-world multicenter cohort study from Austria. ESMO Open 2026 11:106939. PMID 41922145
19. Schiavon G, Hrebien S, Garcia-Murillas I, et al. Analysis of ESR1 mutation in circulating tumor DNA demonstrates evolution during therapy for metastatic breast cancer. Sci Transl Med 2015 7:313ra182. PMID 26560360
20. Fribbens C, O'Leary B, Kilburn L, et al. Plasma ESR1 mutations and the treatment of estrogen receptor-positive advanced breast cancer. J Clin Oncol 2016 34:2961-2968. PMID 27269946
21. Turner NC, Kingston B, Kilburn LS, et al. Circulating tumour DNA analysis to direct therapy in advanced breast cancer (plasmaMATCH): a multicentre, multicohort, phase 2a, platform trial. Lancet Oncol 2020 21:1296-1308. PMID 32919527
22. Fitzpatrick A, Iravani M, Mills A, et al. Assessing CSF ctDNA to improve diagnostic accuracy and therapeutic monitoring in breast cancer leptomeningeal metastasis. Clin Cancer Res 2022 28:1180-1191. PMID 34921020
23. Razavi P, Li BT, Brown DN, et al. High-intensity sequencing reveals the sources of plasma circulating cell-free DNA variants. Nat Med 2019 25:1928-1937. PMID 31768066
24. Gebhart G, Lamberts LE, Wimana Z, et al. Molecular imaging as a tool to investigate heterogeneity of advanced HER2-positive breast cancer and to predict patient outcome under trastuzumab emtansine (T-DM1): the ZEPHIR trial. Ann Oncol 2016 27:619-624. PMID 26598545
25. Jaiswal S, Fontanillas P, Flannick J, et al. Age-related clonal hematopoiesis associated with adverse outcomes. N Engl J Med 2014 371:2488-2498. PMID 25426837
26. Coombs CC, Zehir A, Devlin SM, et al. Therapy-related clonal hematopoiesis in patients with non-hematologic cancers is common and associated with adverse clinical outcomes. Cell Stem Cell 2017 21:374-382. PMID 28803919
27. Bolton KL, Ptashkin RN, Gao T, et al. Cancer therapy shapes the fitness landscape of clonal hematopoiesis. Nat Genet 2020 52:1219-1226. PMID 33106634