Part VI · Heterogeneity, evolution, and metastatic biology · Chapter 30

Temporal heterogeneity and clonal evolution

Treatment does not only fail. It selects, and what it selects is predictable by class.

1 · Therapy as selective pressure, the central premise

A tumour under treatment is a population under selection. Effective therapy removes the cells that depend on what the drug blocks. It does not remove the cells that do not. What remains is enriched for whatever allowed survival, and that enrichment is the mechanism of progression as often as any new mutation is.

The premise has a practical consequence that runs through the rest of this chapter. Resistance is frequently not an event that happens to a tumour. It is a composition that treatment produced. A clone that was present at one percent before treatment and at eighty percent after it has not acquired anything. It has been selected.

This distinction matters because the two situations imply different actions. A newly acquired alteration may be targetable once it appears. A selected pre-existing population was present the whole time, and the opportunity to act on it was earlier.

2 · Natural history and clonal selection before any treatment

Selection does not begin with the first prescription. A tumour reaching clinical attention has already passed through bottlenecks, including the transition from in situ to invasive growth and the establishment of any metastatic deposit. The subclonal structure of an untreated primary is covered in The somatic genomic landscape, and the diversification detectable in in situ disease in Breast carcinogenesis and precursor states.

What matters here is the baseline. The diversity present at diagnosis sets the material available to later selection. A tumour that is subclonally complex before treatment has more routes available to it than one that is not, and the therapeutic pressure that follows does not create those routes so much as choose among them.

3 · Endocrine therapy and the emergence of ESR1 alterations

ESR1 mutations are the clearest worked example in breast oncology of an alteration that treatment produces rather than finds.

In a study of 171 women with advanced breast cancer using high-sensitivity digital polymerase chain reaction on circulating tumour DNA, ESR1 mutations were detected exclusively in patients with oestrogen receptor positive disease who had previously been exposed to an aromatase inhibitor¹. Prevalence differed markedly by the setting in which that aromatase inhibitor was first given. Where it was first given as adjuvant treatment, prevalence was 5.8%, 3 of 52. Where it was first given for metastatic disease, it was 36.4%, 16 of 44, P=0.0002. The micrometastatic and the overt metastatic settings therefore select differently, which is the paper's own conclusion. Patients carrying the mutation had substantially shorter progression-free survival on subsequent aromatase inhibitor based therapy than those without it, with a hazard ratio of 3.1, 95% confidence interval 1.9 to 23.1, P=0.0041.

Three things follow, and they should be held separately.

The alteration is acquired under a specific pressure, so the question of when to test is a question about exposure rather than about tumour type. Testing an endocrine-naive patient for ESR1 is testing before the selection has occurred.

The alteration is a mechanism of escape from oestrogen deprivation specifically. It confers ligand-independent receptor activity, which is why continued oestrogen deprivation stops working and why receptor-directed strategies remain relevant. The biology is developed in Estrogen receptor signaling and endocrine resistance.

The mutation is a marker of a population rather than of a tumour. It is frequently polyclonal, with several different ESR1 alleles detectable in one patient. That is convergent evolution, and it is discussed in Convergent and divergent evolution and what each implies for sequencing.

In EMERALD, elacestrant improved progression-free survival against standard endocrine therapy in patients whose tumours carried ESR1 mutations, having required prior exposure to a CDK4/6 inhibitor². The treatment landscape that follows from ESR1 detection is developed in Endocrine resistance, and the presentations the sequence does not fit.

4 · CDK4/6 inhibition and the evolution of cell cycle escape

The PALOMA-3 circulating tumour DNA analysis is the cleanest demonstration available that a drug class selects for a specific escape route.

Paired samples were taken at baseline and at the end of treatment from 195 patients randomised to palbociclib with fulvestrant or to placebo with fulvestrant. RB1 mutations emerged only in the palbociclib arm, in 6 of 127 patients, which is 4.7%³.

The comparison arm is what makes this informative. Had RB1 alterations appeared in both arms, they would be a feature of progressing disease. Appearing only where CDK4/6 was inhibited, they are a feature of that inhibition. Loss of retinoblastoma protein removes the node through which CDK4 and CDK6 act, so a cell without it does not need the kinase the drug blocks. The pathway biology is set out in Cell cycle regulation.

The frequency is worth stating plainly. Fewer than one patient in twenty progressing on palbociclib did so with a detectable RB1 mutation. RB1 loss is a real mechanism and it is not the common one. Most progression on CDK4/6 inhibition is not explained by it, and a chapter that presents RB1 as the answer has overstated a 4.7% finding³.

5 · Emergence of ESR1, PIK3CA, RB1, CCNE1, and FAT1 alterations on treatment

The same analysis separated what the endocrine backbone selects from what the added inhibitor selects, and the separation is the useful part.

New driver mutations in PIK3CA emerged after treatment in both arms. New ESR1 mutations emerged in both arms, in particular Y537S. RB1 emerged in the palbociclib arm alone³. The shared alterations track the shared exposure, which was fulvestrant on a background of prior endocrine therapy. The arm-specific alteration tracks the arm-specific drug.

A second finding in the same dataset is less often quoted and is arguably more useful. Evolution of driver mutations was uncommon in patients who progressed early and common in patients who progressed late³. Early and late progression are therefore not the same event observed at different times. Early progression looks like pre-existing insensitivity. Late progression looks like something the treatment produced.

That distinction should change what is expected from a biopsy. In a patient who progressed in four months, there may be nothing new to find. In a patient who progressed after three years, there frequently is.

The mutational process itself contributes. APOBEC3 mutagenesis generates the hotspot alterations that define this escape set, including PIK3CA E545K and E542K, ESR1 E380Q and RB1 loss, and APOBEC3-dominant tumours show shorter progression-free survival on endocrine therapy with CDK4/6 inhibition⁴. The signature described in The somatic genomic landscape is not only a record of past damage. It is an ongoing source of the specific variants that end the current line of therapy.

6 · HER2-directed therapy and loss of dependence

Selection under HER2 blockade acts on the compartment structure described in HER2 heterogeneity. A tumour containing cells that do not depend on HER2 signalling will retain those cells, and the population sampled at progression is enriched for them.

The mechanism is not uniform. Dependence can be lost by losing the receptor, by activating a pathway downstream of it, or by switching to a different receptor input. Only the first of these changes what a HER2 assay reports, which means that a stable HER2 result at progression does not establish that dependence was retained.

7 · Conjugate exposure, antigen loss, payload resistance, and transporter change

A conjugate has three separable points of failure, and treating them as one problem is the commonest error in this area.

The antibody has to find its target, so loss of antigen expression is one route. This is the route that HER2 loss as an acquired resistance mechanism describes, and preclinical models of mixed high-expressing and low-expressing HER2 populations reproduce it during the evolution of conjugate resistance⁵.

The payload has to remain cytotoxic to the cell that receives it, so resistance to the payload class is a second route. It is independent of the target and it is why switching to another conjugate carrying the same payload class may not recover activity.

The payload has to stay inside the cell, so efflux transporter activity is a third route.

The clinical consequence is that antigen expression measured at progression answers only the first of the three questions. A tumour that still expresses the target may have become resistant in either of the other two ways, and no routine assay currently distinguishes them. Sequencing between conjugates is discussed in Antibody-drug conjugates and targeted delivery.

8 · Immunotherapy and immunoediting over time

Immunoediting describes three phases: elimination of immunogenic clones, equilibrium in which outgrowth is restrained, and escape when a variant evades control⁶. The framework is a temporal one, which is why it belongs in this chapter rather than only in Part V.

The consequence for a treating clinician is that the immune phenotype of a tumour is a snapshot of an ongoing selection process, not a fixed property. A tumour sampled at diagnosis has already been edited by whatever immune response preceded it. Its remaining antigens are the ones that survived that editing, which is a systematic bias and not a random one.

Treatment accelerates the process. Checkpoint blockade applies selective pressure favouring clones that escape T-cell recognition, and the escape routes are the ones in MHC loss and antigen presentation defects. Loss of antigen presentation, loss of the specific neoantigens that drove the response, and upregulation of alternative inhibitory pathways are all documented mechanisms of acquired resistance.

Chemotherapy edits differently. Immunogenic cell death releases antigen and can broaden a response, while lymphodepletion removes the effectors that would act on it. Which effect dominates depends on agent, dose and schedule, and this is one of the arguments for the sequencing questions in Immunotherapy.

The measurement problem is severe. TIL density and PD-L1 expression are both measured once, usually before treatment, and both change under therapy. A biomarker that predicts response at baseline says little about the state of the tumour three cycles later, which is the same temporal instability this chapter documents for receptor status.

9 · Genetic evolution and reversible cell state change as different problems

Not everything that changes under treatment is a change in DNA. A cell can stop depending on a pathway by altering its transcriptional state, and it can revert when the pressure is removed.

The two are separated by a test rather than by inspection. A genetic change is heritable and persists through drug-free passage. A state change reverts. Populations that survive treatment without any detectable alteration, then regain sensitivity after an interval off the drug, are behaving in the second way. The biology is developed in Plasticity and non-genetic heterogeneity.

The distinction has a direct clinical implication. Rechallenge is a coherent strategy against a reversible state and an incoherent one against a fixed alteration. Because the two are indistinguishable at progression by any routine test, rechallenge decisions are currently made without knowing which situation applies.

10 · Evolution across successive lines and the cost of each switch

Each line of therapy applies its own selection, and the selections accumulate. A tumour at fourth line has been filtered by every preceding regimen, and the population that remains is the product of all of them rather than of the most recent one.

Two consequences are worth stating.

The first is that the yield of any single targeted strategy falls as lines accumulate, because the population has been enriched for independence from the pathways already attacked. ESR1 mutation prevalence, for instance, rises across successive lines of endocrine therapy. In a real-world Austrian cohort of 184 patients, mutations were found in 7.1% of samples taken before first-line treatment and 26.7% of samples taken before second line⁷. That is selection made visible, and the figures are set beside the trial cohorts in Estrogen receptor signaling and endocrine resistance.

The second is that every switch has a cost that is not only toxicity. A switch discards whatever control the current regimen retains, and it applies a new pressure whose escape route may be less tractable than the current one. This is the argument that makes treatment-free intervals and maintenance strategies more than tolerability decisions, and it is taken up in General principles of metastatic management.

11 · Convergent and divergent evolution and what each implies for sequencing

Different tumours reaching the same escape route is convergent evolution. Different lesions within one patient reaching different routes is divergent evolution. Both occur, and they carry opposite implications for what a single assay can tell you.

Convergence is the friendlier case. Where a class of therapy reliably selects one route, that route can be anticipated, monitored and targeted. The ESR1 example in Endocrine therapy and the emergence of ESR1 alterations is convergent at the level of the gene, and frequently convergent within a patient, where several independent mutant alleles arise in parallel.

Divergence is the harder case. Where separate metastatic deposits have taken separate routes, a biopsy of one lesion characterises that lesion. Treatment selected on that basis addresses the sampled deposit and may not address the others. This is the temporal counterpart of the sampling problem in Spatial heterogeneity, and it is why mixed response is a biologically expected finding rather than an anomaly.

12 · Evolvability itself as prognostic information

The preceding sections treat evolution as something that happens to a tumour. This section proposes that the capacity to evolve is itself a measurable tumour property with prognostic meaning, separate from any particular alteration.

The components are individually measurable. Subclonal diversity at diagnosis, estimated from multiregion sequencing or from clonal deconvolution of a single sample^8,9, describes how much variation selection has to work on. Mutational processes that remain active, such as ongoing APOBEC mutagenesis, describe the rate at which new variation is generated^10,11. Chromosomal instability describes the capacity for large-scale change, and the punctuated copy number evolution seen in triple-negative disease shows that this capacity is not uniform across subtypes¹². Topographic single-cell sequencing shows that multiple clones invade together rather than sequentially¹³, which means the diversity present at invasion is carried forward rather than bottlenecked.

The argument is that a tumour with high standing diversity and an active mutational process will generate resistance faster than a tumour with low diversity and a quiescent one, for any given treatment. Two tumours with identical current biomarker profiles may therefore have different trajectories, and the difference is not captured by anything currently reported.

The obstacles to using this are real. Estimating diversity from one core underestimates it, and the degree of underestimation is unknown for that tumour. There is no validated assay, no threshold and no prospective evidence that acting on such a measure improves outcomes.

13 · Longitudinal monitoring and the case for repeat biopsy

If the composition of a tumour changes under treatment, then a characterisation performed once is a characterisation of the past. Two questions follow. When should it be repeated, and with what.

Tissue and plasma answer different questions. A tissue biopsy characterises one deposit in detail. Circulating tumour DNA integrates across deposits and misses what is not shed, so it detects divergence poorly and convergence well. Neither is a substitute for the other, and the respective blind spots are set out in Liquid biopsy and longitudinal monitoring.

The timing question has a defensible general answer. Repeat characterisation is worth performing when the result could change the next decision, which in practice means at progression, and before a line of therapy whose selection is biomarker-defined. Repeating it when no available action depends on the answer generates information without a decision attached to it.

14 · Acting on evolution, intermittent dosing, adaptive therapy, pre-emptive switching

Everything above describes evolution. The harder question is whether it can be acted on, and until recently the honest answer was that it had not been shown.

That changed with pre-emptive switching. In SERENA-6, patients on first-line endocrine therapy with a CDK4/6 inhibitor were monitored for the emergence of ESR1 mutation in circulating tumour DNA, and those in whom it emerged were switched to camizestrant before any radiographic progression. Median progression-free survival was 16.8 months against 9.2 months for those who continued¹⁴.

The design is what matters, more than the agent. The switch was triggered by a molecular event rather than by imaging, it was made while the patient was still responding by conventional assessment, and the comparator received the same drug later. This is an evolutionary argument tested prospectively and supported.

Two limits should be stated alongside it. The result establishes the principle for one alteration detected by one assay and acted on with one class of drug. Whether pre-emptive switching generalises to other selected alterations is unknown. And the strategy requires serial monitoring of patients who are, at the moment of the switch, doing well, which has implications for cost, for access and for how the decision is explained.

References

All 14 references

1. 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
2. Bidard FC, Kaklamani VG, Neven P, et al. Elacestrant (oral selective estrogen receptor degrader) versus standard endocrine therapy for estrogen receptor-positive, HER2-negative advanced breast cancer: results from the randomized phase III EMERALD trial. J Clin Oncol 2022 40:3246-3256. PMID 35584336
3. 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
4. Gupta A, Gazzo A, Selenica P, et al. APOBEC3 mutagenesis drives therapy resistance in breast cancer. Nat Genet 2025 57:1452-1462. PMID 40379787
5. Goyette MA, Graser C, Seehawer M, et al. HER2 heterogeneous breast cancer models reveal novel therapeutic targets and subclonal dynamics during evolution to resistance to HER2-targeted therapies. Cancer Discov 2026 16:1691-1710. PMID 41925564
6. Schreiber RD, Old LJ, Smyth MJ. Cancer immunoediting: integrating immunity's roles in cancer suppression and promotion. Science 2011 331:1565-1570. PMID 21436444
7. 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
8. Yates LR, Gerstung M, Knappskog S, et al. Subclonal diversification of primary breast cancer revealed by multiregion sequencing. Nat Med 2015 21:751-759. PMID 26099045
9. Roth A, Khattra J, Yap D, et al. PyClone: statistical inference of clonal population structure in cancer. Nat Methods 2014 11:396-398. PMID 24633410
10. Alexandrov LB, Kim J, Haradhvala NJ, et al. The repertoire of mutational signatures in human cancer. Nature 2020 578:94-101. PMID 32025018
11. Petljak M, Dananberg A, Chu K, et al. Mechanisms of APOBEC3 mutagenesis in human cancer cells. Nature 2022 607:799-807. PMID 35859169
12. Gao R, Davis A, McDonald TO, et al. Punctuated copy number evolution and clonal stasis in triple-negative breast cancer. Nat Genet 2016 48:1119-1130. PMID 27526321
13. Casasent AK, Schalck A, Gao R, et al. Multiclonal invasion in breast tumors identified by topographic single cell sequencing. Cell 2018 172:205-217. PMID 29307488
14. 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