Part VI · Heterogeneity, evolution, and metastatic biology · Chapter 33

Dormancy, residual disease, and late recurrence

Early and late recurrence are not one process observed at two times. The hazard curve says so, and it is why the word cure is hard to use in hormone receptor-positive disease.

1 · Disseminated tumor cells and where they reside

Disseminated tumour cells are the physical object this chapter is about. They are detectable, they carry prognostic weight, and they are almost never looked for in practice.

A pooled analysis of individual data from nine studies and 4,703 patients with stage I, II or III breast cancer detected bone marrow micrometastasis at diagnosis in 30.6%¹. Detection was associated with larger tumours, higher grade, nodal involvement and hormone receptor-negative disease. Univariate mortality ratios against patients without micrometastasis were 2.15 for overall survival and 2.44 for breast cancer-specific survival. Micrometastasis remained an independent predictor of poor outcome in multivariable analysis, and among patients with tumours 2 cm or smaller, node-negative, who received no systemic adjuvant therapy, the mortality ratio was 3.65.

Where those cells sit has been mapped in models. Dormant cells occupy the microvasculature of lung, bone marrow and brain². Within marrow specifically, quiescent cells concentrate in perisinusoidal regions rich in E-selectin and CXCL12, with E-selectin required for entry into the marrow and CXCL12 signalling through CXCR4 anchoring them there³. A different marrow compartment, one with osteogenic features, supports progression from single cells to micrometastases through heterotypic adherens junctions and mTOR activation⁴.

2 · Cellular quiescence programs and the dormant state

Dormancy names at least three different situations, and treating them as one is the commonest error in this area.

Tumour mass dormancy describes a deposit in which proliferation and death are balanced, so the lesion persists without growing. It has two recognised forms, angiogenic dormancy limited by vascular supply and immune-mediated dormancy limited by cytotoxic control⁵.

Cellular dormancy describes a single cell arrested in G0 or G1, not dividing at all.

The distinction is not academic, because the two predict opposite responses to the same treatment.

An antiproliferative agent can shrink a mass-dormant lesion, because cells within it are cycling. It cannot kill a cell that is not cycling, because there is no cycle to interrupt. That single sentence explains why adjuvant cytotoxic therapy reduces recurrence without abolishing it, and why the residual population it leaves is enriched for exactly the cells least susceptible to what was given.

It also explains the shape of the problem in Annual hazard of recurrence and the long tail of hormone receptor-positive disease. A recurrence at fifteen years is not evidence of slow growth over fifteen years. A deposit doubling slowly enough to take fifteen years to become detectable would be an unusual biology. A cell that stopped and later restarted is the more parsimonious account, and it is the account the dormancy literature supports.

Quiescence here is not senescence. The distinction, and the ways incomplete arrest is confused with both, is set out in Senescence, mitotic slippage, and incomplete arrest.

3 · p38 and ERK signaling balance, NR2F1, and the dormancy program

Dormancy has a signalling logic, and it is a ratio rather than a switch.

The original work identified urokinase plasminogen activator receptor as the upstream regulator. High receptor expression drives frequent activating interactions with alpha5beta1 integrin, which sustains high ERK activity and permits fibronectin fibril assembly, and the fibrils in turn suppress p38⁶. Low-receptor derivatives of the same line are growth arrested in vivo and carry a high p38 to ERK ratio. Inhibiting p38 in those arrested cells reactivated ERK, restored receptor expression and interrupted dormancy.

The ratio then proved general. Across breast, prostate, melanoma and fibrosarcoma lines, the level of active phospho-ERK and the ERK to p38 activity ratio predicted in vivo behaviour in approximately 90% of lines tested⁷. Pharmacological or genetic modulation of the ratio shifted the phenotype in the predicted direction. High ERK relative to p38 favours growth. High p38 relative to ERK produces arrest.

NR2F1 is the transcriptional output. It is epigenetically upregulated in experimental dormancy models and in disseminated cells from prostate cancer patients who carried dormant disease for 7 to 18 years⁸. NR2F1-induced quiescence depends on SOX9, RAR-beta and CDK inhibitors, and NR2F1 also induces global chromatin repression and the pluripotency gene NANOG. Blocking NR2F1 in vivo interrupted either the growth arrest or the survival of dormant cells, depending on the organ.

The program has an upstream trigger in the primary tumour. Hypoxic microenvironments in breast and head and neck primaries upregulated NR2F1, DEC2 and p27, and disseminated cells that had passed through hypoxia were frequently dormant and evaded chemotherapy⁹. NR2F1 and HIF1-alpha were both required for p27 induction in those cells.

The clinical translation exists and it is small. Extra cytospins from 114 bone marrow samples from 86 selected patients already known to be positive for disseminated tumour cells were scored by double immunofluorescence¹⁰. Of patients with detectable cells, 27% had at least 50% NR2F1-high cells, a cut-off chosen in advance as a dormant profile. Every patient with systemic relapse within 12 months of aspiration carried 1% or fewer NR2F1-high cells, including patients who had carried NR2F1-expressing cells in earlier samples. Among 18 relapse-free patients at their last positive aspiration, distant disease-free interval favoured the at least 50% group over the predominantly NR2F1-low group, with a log-rank P of 0.007. Classifying the same patients by Ki-67-expressing disseminated cells produced no survival difference, with a P of 0.520.

4 · Immune-mediated dormancy and natural killer cell control

Dormancy is partly enforced from outside the cell, and the enforcing arm is not the one most often measured.

Latency-competent cells isolated from early-stage human lung and breast carcinoma lines express SOX2 and SOX9, and those factors are required for their survival in host organs under immune surveillance¹¹. The mechanism is self-imposed. By expressing the WNT inhibitor DKK1, these cells enter a slow-cycling state with broad downregulation of ULBP ligands for natural killer cells, which lets them evade natural killer-mediated clearance. Silencing WNT is how they become both quiet and invisible.

The complementary experiment showed the host side. In mouse liver, the dormant milieu carried a selective increase in natural killer cells¹². Adjuvant interleukin-15-based immunotherapy maintained that pool and sustained dormancy through interferon-gamma signalling. It prevented hepatic metastases and prolonged survival. Exit from dormancy followed contraction of the natural killer compartment together with accumulation of activated hepatic stellate cells, whose secreted CXCL12 drives natural killer cells into quiescence via CXCR4. CXCL12 expression and activated stellate cell abundance correlated with each other in patients with liver metastases.

Two things follow that are worth separating.

The immune control that restrains dormant cells in these models is innate rather than T cell mediated. The biology is developed in Innate immunity, natural killer cells, and antibody-dependent cytotoxicity. Checkpoint blockade acts on a different arm, which is one reason the immunotherapy logic of Immunotherapy does not transfer straightforwardly to residual disease.

Anything that suppresses innate immunity is, on this mechanism, a reactivation risk. That is a mechanistic statement rather than a clinical one, and no clinical intervention currently follows from it.

5 · Matrix, stromal, and niche control of dormancy

The niche does not merely house a dormant cell. It instructs it, and the instruction changes when the niche changes.

The vascular instruction is the best characterised. Endothelial-derived thrombospondin-1 from stable microvasculature induces sustained quiescence in breast cancer cells². That suppressive cue is lost in sprouting neovasculature. Sprouting vessels do not merely permit outgrowth, they accelerate it, through active TGF-beta1 and periostin derived from endothelial tip cells. The same tissue is a dormancy niche when quiet and a growth niche when it remodels.

The marrow instruction has two arms that work in opposite directions. E-selectin interactions are required for breast cancer cells to enter marrow through the sinusoidal niche, and the CXCL12 and CXCR4 interaction then anchors them to that microenvironment³. Inhibiting CXCR4 mobilises dormant micrometastases back into the circulation. A separate osteogenic compartment does the reverse, supporting the transition from single cell to micrometastasis through E-cadherin and N-cadherin junctions that activate mTOR⁴.

Matrix is a signal rather than a scaffold, and proteolysis rewrites the signal. Laminin sequentially cleaved by neutrophil elastase and MMP9 acquires the ability to induce proliferation in dormant cells through integrin alpha3beta1, and antibodies raised against the remodelled laminin prevented awakening¹³. The composition and mechanics of the matrix are developed in Extracellular matrix composition and tissue mechanics.

6 · Metabolic control of dormancy and reactivation

A dormant cell is not metabolically idle. It is running a specific program, and that program is the most tractable target in this chapter.

Downregulation of Her2 in breast cancer cells produced changes in cellular metabolism that culminated in oxidative stress, with compensatory upregulation of the antioxidant transcription factor NRF2¹⁴. NRF2 was activated during dormancy and in recurrent tumours in animal models, and in breast cancer patients with poor prognosis. The relationship was causal in the models. Constitutive NRF2 activation accelerated recurrence and suppressing NRF2 impaired it.

What NRF2 does in recurrent tumours is a transcriptional metabolic reprogramming that re-establishes redox homeostasis and upregulates de novo nucleotide synthesis. That combination is what a cell needs to restart division after a long arrest, which is why it is a reactivation program rather than a survival program.

It also creates a dependency. The NRF2-driven state rendered recurrent tumour cells sensitive to glutaminase inhibition, which prevented reactivation of dormant cells in vitro¹⁴.

The general principle is the same one developed in Targeting the plastic state rather than the resistant clone for drug-tolerant states. A cell in an unusual state has unusual requirements, and those requirements are targets that the bulk population does not share. The metabolic constraints imposed by the surrounding tissue are set out in Hypoxia, acidosis, and nutrient competition.

7 · Mechanisms of dormancy escape, including inflammation and aging

Escape is where the mechanistic literature is richest and the clinical translation is thinnest.

Inflammation is the best-supported trigger. Sustained lung inflammation caused by tobacco smoke exposure or by nasal instillation of lipopolysaccharide converted disseminated dormant cancer cells into aggressively growing metastases in mice¹³. The requirement was neutrophil extracellular traps. Two trap-associated proteases, neutrophil elastase and MMP9, sequentially cleaved laminin, and the proteolytically remodelled laminin induced proliferation through integrin alpha3beta1. Antibodies against the remodelled laminin prevented awakening. The trap biology is developed in Neutrophil extracellular traps and their role in awakening tumor cells.

Ageing of the destination organ is a second trigger, and the evidence is currently strongest outside breast cancer. In melanoma, age-induced reprogramming of lung fibroblasts increased their secretion of the WNT antagonist sFRP1¹⁵. That antagonist inhibited WNT5A in disseminated melanoma cells and enabled metastatic outgrowth. The same work identified AXL and MER as a dormancy-to-reactivation axis. The host ageing context is developed in Aging, immunosenescence, and clonal hematopoiesis.

Loss of immune restraint is a third, and it has the clearest mechanism¹².

8 · Minimal residual disease as a measurable state

Molecular residual disease turns dormancy from a concept into a variable, and two assays measure different parts of it.

Circulating tumour DNA measures shedding. In a prospective cohort of 55 patients with early breast cancer receiving neoadjuvant chemotherapy, detection of ctDNA after apparently curative treatment predicted metastatic relapse¹⁶. At a single postsurgical time point the hazard ratio for relapse was 25.1, with a 95% confidence interval of 4.08 to 130.5. With serial follow-up sampling it was 12.0, with a 95% confidence interval of 3.36 to 43.07. Serial mutation tracking gave a median lead time of 7.9 months over clinical relapse. Sequencing the residual disease predicted the genetic events of the subsequent metastatic relapse more accurately than sequencing the primary tumour did.

Marrow aspiration measures presence¹.

Those are not interchangeable, and the reason is the biology in Cellular quiescence programs and the dormant state. A cell arrested in G0, not dividing and not dying, releases little DNA. It is exactly the cell that a shedding-based assay is least likely to see and a cell-based assay might. The analytic detail of tumour-informed and tumour-naive approaches is set out in Molecular residual disease, tumor-informed and tumor-naive approaches, and the sources of false positive results in Clonal hematopoiesis and other false positives.

Two statements should be held apart when reporting a result to a patient.

A positive ctDNA result after curative-intent treatment is strong evidence that residual disease is present, on the effect sizes above.

A negative result is evidence that no shedding was detected at that sensitivity, at that time point. It is not evidence that no residual disease exists, and given the dormancy biology it is specifically weak evidence against the quiet population that matters most.

9 · Early and late recurrence as different biological problems

This is the chapter's central claim, and the strongest support for it is a crossover in a hazard curve rather than any mechanism.

In International Breast Cancer Study Group trials I to V, 4,105 patients were randomised between 1978 and 1985 and followed for a median of 24 years¹⁷. For the whole group, the annualised hazard of recurrence was highest in the first 5 years at 10.4%, peaking between years 1 and 2 at 15.2%.

Within those first 5 years, patients with oestrogen receptor-positive disease had a lower annualised hazard than those with receptor-negative disease, 9.9% against 11.5%.

Beyond 5 years the ordering reverses. From 5 to 10 years the annualised hazards were 5.4% for receptor-positive against 3.3% for receptor-negative disease. From 10 to 15 years, 2.9% against 1.3%. From 15 to 20 years, 2.8% against 1.2%.

That inversion is the argument. If early and late recurrence were one process sampled at two times, the rank order of risk by receptor status would be preserved and only the magnitude would fall. It is not preserved. It flips, and it flips durably.

The mechanistic reading follows naturally. Early recurrence is best understood as outgrowth of disease that was already proliferating and was never controlled, which is why it tracks with the proliferative, receptor-negative phenotype. Late recurrence requires a population that survived, stopped, and later restarted, which is the dormancy machinery in p38 and ERK signaling balance, NR2F1, and the dormancy program through Mechanisms of dormancy escape, including inflammation and aging.

The same shape appears in the metastatic setting, which supports the generality of the argument. Evolution of driver mutations in PALOMA-3 was uncommon in patients who progressed early and common in patients who progressed late¹⁸. Early failure looked like pre-existing insensitivity. Late failure looked like something the interval produced.

10 · Annual hazard of recurrence and the long tail of hormone receptor-positive disease

This is the number that makes the word cure difficult, and it deserves to be given exactly.

A meta-analysis of 88 trials involving 62,923 women with oestrogen receptor-positive breast cancer who were disease-free after 5 years of scheduled endocrine therapy followed outcomes from year 5 to year 20¹⁹. Recurrences occurred at a steady rate throughout that period. They did not taper towards zero.

The absolute risk of distant recurrence between year 5 and year 20 depended on the original tumour size and nodal status. For T1 disease it was 13% with no nodal involvement, 20% with one to three nodes involved and 34% with four to nine nodes involved. For T2 disease it was 19%, 26% and 41% respectively. Within T1N0 disease, grade separated those risks further, at 10% for low grade, 13% for moderate and 17% for high. The corresponding risks of any recurrence or a contralateral breast cancer in T1N0 disease were 17%, 22% and 26%.

Two negative findings from the same analysis are as useful as the positive ones. Given tumour and nodal status, progesterone receptor status in 54,115 patients and HER2 status in 15,418 patients from trials without trastuzumab were not predictive of distant recurrence in that window. Grade and Ki-67 carried only moderate independent predictive value.

The annualised view says the same thing. In receptor-positive node-negative disease, annualised hazards were 2.0%, 2.1% and 1.1% for years 10 to 15, 15 to 20 and 20 to 25¹⁷. With one to three positive nodes they were 3.0%, 3.5% and 1.5%. An older analysis of Eastern Cooperative Oncology Group adjuvant trials found a mean annual hazard of recurrence of 4.3% between years 5 and 12 after surgery²⁰.

Now the word. Cure asserts that the hazard reached zero. What these datasets show is that the hazard became small and then stayed roughly flat for as long as anyone has followed these women. A woman who is well fifteen years after a T2 node-positive receptor-positive tumour still carries a measurable annual risk of distant recurrence. It is of the order of a few percent per year rather than zero.

Those are different statements, and the difference is the honest answer to the question raised in Evidence, uncertainty, and what the word cure is doing. It is also why extended endocrine therapy is a dormancy intervention rather than a consolidation, which is the framing that belongs in the discussion at Extended endocrine therapy and duration.

11 · Strategies to detect and intercept residual disease

Interception divides into three approaches, and they differ in how much evidence stands behind them.

Change the soil. Adjuvant bisphosphonates are the only strategy in this chapter that is established practice, and the meta-analysis is instructive about the mechanism. Across 18,766 women, overall reductions in recurrence and breast cancer mortality were of borderline significance, while the reduction in bone recurrence was more definite²¹. Among 11,767 postmenopausal women the effects were clear. The rate ratio for recurrence was 0.86, with a 95% confidence interval of 0.78 to 0.94. For distant recurrence it was 0.82, 0.74 to 0.92. For bone recurrence it was 0.72, 0.60 to 0.86. For breast cancer mortality it was 0.82, 0.73 to 0.93. Among premenopausal women there was no apparent effect on any outcome. Nothing about the tumour was targeted. The bone niche was, and the site-specific benefit follows the niche biology in Matrix, stromal, and niche control of dormancy. The menopausal-status interaction remains unexplained.

Target the dormant cell directly. A randomised phase 2 trial enrolled 51 breast cancer survivors within 5 years of diagnosis who had detectable disseminated tumour cells on bone marrow aspirate²². They were allocated to hydroxychloroquine in 15 patients, everolimus in 15, or both in 21. Treatment was feasible and tolerable, with one early discontinuation for grade 3 toxicity. At a median follow-up of 42 months, landmark 3-year recurrence-free survival was 91.7%, 92.9% and 100% respectively, and estimated reductions in disseminated cells were 80%, 78% and 87%. This is proof of concept that the target is modifiable in patients. It is not evidence of benefit. There was no observation arm randomised for survival. The hazard ratio comparing patients who cleared cells against those who did not was 0.21, with a 95% confidence interval of 0.01 to 3.4. That interval excludes almost nothing.

Act on a molecular signal. DARE, LEADER and SURVIVE each test a version of the design in which a ctDNA result rather than an image triggers an intervention. The principle has been established once in the metastatic setting by SERENA-6²³. Whether it transfers to a disease-free population is the open question, and it is the question of Lead time and the unresolved question of whether acting early helps rather than of this chapter.

References

1. Braun S, Vogl FD, Naume B, et al. A pooled analysis of bone marrow micrometastasis in breast cancer. N Engl J Med 2005 353:793-802. PMID 16120859
2. Ghajar CM, Peinado H, Mori H, et al. The perivascular niche regulates breast tumour dormancy. Nat Cell Biol 2013 15:807-817. PMID 23728425
3. Price TT, Burness ML, Sivan A, et al. Dormant breast cancer micrometastases reside in specific bone marrow niches that regulate their transit to and from bone. Sci Transl Med 2016 8:340ra73. PMID 27225183
4. Wang H, Yu C, Gao X, et al. The osteogenic niche promotes early-stage bone colonization of disseminated breast cancer cells. Cancer Cell 2015 27:193-210. PMID 25600338
5. Aguirre-Ghiso JA. Models, mechanisms and clinical evidence for cancer dormancy. Nat Rev Cancer 2007 7:834-846. PMID 17957189
6. Aguirre-Ghiso JA, Liu D, Mignatti A, Kovalski K, Ossowski L. Urokinase receptor and fibronectin regulate the ERK(MAPK) to p38(MAPK) activity ratios that determine carcinoma cell proliferation or dormancy in vivo. Mol Biol Cell 2001 12:863-879. PMID 11294892
7. Aguirre-Ghiso JA, Estrada Y, Liu D, Ossowski L. ERK(MAPK) activity as a determinant of tumor growth and dormancy; regulation by p38(SAPK). Cancer Res 2003 63:1684-1695. PMID 12670923
8. Sosa MS, Parikh F, Maia AG, et al. NR2F1 controls tumour cell dormancy via SOX9- and RARbeta-driven quiescence programmes. Nat Commun 2015 6:6170. PMID 25636082
9. Fluegen G, Avivar-Valderas A, Wang Y, et al. Phenotypic heterogeneity of disseminated tumour cells is preset by primary tumour hypoxic microenvironments. Nat Cell Biol 2017 19:120-132. PMID 28114271
10. Borgen E, Rypdal MC, Sosa MS, et al. NR2F1 stratifies dormant disseminated tumor cells in breast cancer patients. Breast Cancer Res 2018 20:120. PMID 30322396
11. Malladi S, Macalinao DG, Jin X, et al. Metastatic latency and immune evasion through autocrine inhibition of WNT. Cell 2016 165:45-60. PMID 27015306
12. Correia AL, Guimaraes JC, Auf der Maur P, et al. Hepatic stellate cells suppress NK cell-sustained breast cancer dormancy. Nature 2021 594:566-571. PMID 34079127
13. Albrengues J, Shields MA, Ng D, et al. Neutrophil extracellular traps produced during inflammation awaken dormant cancer cells in mice. Science 2018 361:eaao4227. PMID 30262472
14. Fox DB, Garcia NMG, McKinney BJ, et al. NRF2 activation promotes the recurrence of dormant tumour cells through regulation of redox and nucleotide metabolism. Nat Metab 2020 2:318-334. PMID 32691018
15. Fane ME, Chhabra Y, Alicea GM, et al. Stromal changes in the aged lung induce an emergence from melanoma dormancy. Nature 2022 606:396-405. PMID 35650435
16. 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
17. Colleoni M, Sun Z, Price KN, et al. Annual hazard rates of recurrence for breast cancer during 24 years of follow-up: results from the International Breast Cancer Study Group trials I to V. J Clin Oncol 2016 34:927-935. PMID 26786933
18. 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
19. Pan H, Gray R, Braybrooke J, et al. 20-year risks of breast-cancer recurrence after stopping endocrine therapy at 5 years. N Engl J Med 2017 377:1836-1846. PMID 29117498
20. Saphner T, Tormey DC, Gray R. Annual hazard rates of recurrence for breast cancer after primary therapy. J Clin Oncol 1996 14:2738-2746. PMID 8874335
21. Early Breast Cancer Trialists' Collaborative Group (EBCTCG). Adjuvant bisphosphonate treatment in early breast cancer: meta-analyses of individual patient data from randomised trials. Lancet 2015 386:1353-1361. PMID 26211824
22. DeMichele A, Clark AS, Shea E, et al. Targeting dormant tumor cells to prevent recurrent breast cancer: a randomized phase 2 trial. Nat Med 2025 31:3464-3474. PMID 40897974
23. Turner NC, Mayer EL, Park YH, et al. Switching to camizestrant at ESR1 mutation emergence before disease progression during first-line treatment of hormone receptor-positive advanced breast cancer (SERENA-6). Lancet Oncol 2026. PMID 42442380