Part VI · Heterogeneity, evolution, and metastatic biology · Chapter 35

Mechanisms of therapeutic resistance

The chapter to open at progression. What failed, whether it was the drug, and what to measure next.

1 · A working taxonomy, intrinsic, adaptive, and acquired resistance

Three categories are worth keeping apart, and they are separated by timing and reversibility rather than by mechanism.

Intrinsic resistance was present before the first dose. The population that would have responded was never large enough, or the dependency the drug targets was never load-bearing. Adaptive resistance appears within days to weeks. It is a change of cell state rather than of genome, it does not require a new mutation, and it reverts when the drug is withdrawn. Acquired resistance emerges over months. It is heritable, it persists through drug-free passage, and it is what selection leaves behind.

The clinical course carries information about which applies. In the PALOMA-3 circulating tumour DNA analysis, evolution of driver mutations was uncommon in patients who progressed early and common in patients who progressed late1. Early progression looks like a population that was never sensitive. Late progression looks like something the treatment produced.

Each category implies a different move. Intrinsic resistance means the class was wrong, so change class. Adaptive resistance means the schedule or the combination was wrong, so change those. Acquired resistance means the target or the dependency has changed, so characterise it before choosing.

Caution

These labels describe a clinical course, not a cell. One tumour can hold an intrinsically insensitive compartment, an adaptively tolerant compartment and a selected resistant clone at the same time. A biopsy samples whichever of them was in the needle.

2 · Endocrine resistance, ESR1 alterations, ER loss, and bypass signaling

Endocrine resistance is four different failures under one name, and separating them changes what to do next.

The receptor can become ligand-independent. ESR1 mutations were detected in circulating tumour DNA exclusively in patients with oestrogen receptor positive disease previously exposed to an aromatase inhibitor. Carriers had a hazard ratio of 3.1 for progression on subsequent aromatase inhibitor based therapy2. Oestrogen deprivation stops working because the receptor no longer requires oestrogen. Receptor-directed therapy still has a target, which is the logic behind EMERALD3 and behind the pre-emptive switch tested in SERENA-64. The clinical pathway is developed in ESR1-mutant disease and receptor-degrading strategies.

The receptor can be lost. That is a change of lineage rather than a change of ligand dependence, and it removes the target rather than modifying it. The plasticity behind it is set out in Plasticity and non-genetic heterogeneity.

The receptor can be retained while ceasing to drive. Growth factor receptor signalling substitutes for it, and the classical demonstration is that tamoxifen-resistant tumours show growth factor receptor signalling with repression of classical oestrogen receptor genomic function5. HER2 overexpression was identified as an independent marker of endocrine resistance long before it was targetable6. Hyperactivation of EGFR and HER2 mediates resistance both to endocrine therapy and to CDK4/6 inhibition7.

The transcriptional machinery around the receptor can be rewired. PI3K pathway activity controls oestrogen receptor dependent transcription through chromatin regulators8. Pathway detail lives in Estrogen receptor signaling and endocrine resistance and PI3K, AKT, mTOR, and related networks and is not repeated here.

The measurement caveat is specific. An ESR1 mutation found in plasma is a property of the shed population. Its absence is weak evidence, and its presence says nothing about which deposit contributed it.

3 · CDK4/6 inhibitor resistance, Rb loss, cyclin E, and bypass pathways

Loss of retinoblastoma protein is the mechanism everyone names and it is not the common one. In the PALOMA-3 paired analysis, RB1 mutations emerged only in the palbociclib arm, in 6 of 127 patients, which is 4.7%1. Fewer than one progressing patient in twenty had a detectable RB1 mutation. Retinoblastoma biology is also less binary than the tumour suppressor label suggests, and Rb-driven transcription can itself limit its suppressive effect9.

Cyclin E1 is the better-supported predictive marker and it points at the same node from the other side. Tumour tissue was analysed in 302 patients in PALOMA-3, 194 in the palbociclib arm and 108 in the placebo arm. In the palbociclib arm, median progression-free survival was 7.6 months with high cyclin E1 messenger RNA against 14.1 months with low10. In the placebo arm the corresponding figures were 4.0 and 4.8 months. The small separation in the placebo arm is what makes the large one interpretable. Cyclin E1 marks a cell cycle that does not depend on the kinase being blocked.

One detail in that analysis is a sampling result rather than a biological one. Cyclin E1 messenger RNA was more predictive when measured in metastatic tissue than in archival primary tissue10. The marker changes over time, so the block the laboratory happens to hold may answer a question about the tumour of several years ago.

The mutational process supplies the variants. APOBEC3 mutagenesis generates the hotspot alterations that define this escape set, and APOBEC3-dominant tumours show shorter progression-free survival on endocrine therapy with CDK4/6 inhibition11. Cell cycle detail is in Cell cycle regulation, and the treatment consequences in Metastatic hormone receptor-positive, HER2-negative disease.

The honest summary is uncomfortable. Most progression on CDK4/6 inhibition is not explained by any mechanism that can currently be measured in the clinic.

4 · HER2-directed therapy resistance across antibodies and kinase inhibitors

Three separable things have to hold for HER2-directed therapy to work. The drug has to reach the target, the cell has to depend on the pathway, and for antibodies the immune effector arm has to function. Resistance is the failure of any one of them, and only the first changes what a HER2 assay reports.

Access fails when the antigen is not there. That is the compartment structure described in HER2 heterogeneity, observed after selection rather than before it, and preclinical models with mixed high and low expressing populations reproduce it during the evolution of resistance12.

Dependence fails when signalling is supplied from elsewhere. Oestrogen receptor signalling is upregulated as an adaptive survival mechanism in HER2-positive tumours under anti-HER2 therapy13. Co-occurring activating mutations in HER2 and HER3 modulate both dimer activation and sensitivity to HER2 inhibition14. Truncated receptor forms lacking the antibody binding site are a further route, and the immune consequences of that biology sit in HER2-driven evasion, CD47 and p95HER2.

The two drug classes do not fail in the same way. Mechanisms of resistance to trastuzumab and to lapatinib differ, with distinct contributions from oestrogen receptor signalling and from HER2 reactivation15. The practical consequence is that failure of an antibody is a weak predictor of failure of a kinase inhibitor, and the converse also holds. Sequencing decisions belong to Sequencing HER2-directed agents after first progression, and receptor pathway detail to HER family signaling.

A stable HER2 result at progression does not establish that dependence was retained. It establishes that the antigen is still present.

5 · Conjugate resistance, antigen loss, payload efflux, and repair capacity

This is the least settled section in the chapter, and it should be read as such.

A conjugate has to bind, internalise, traffic to the lysosome, release payload, retain that payload and then kill with it. Six steps, five of which no clinical assay measures. Antigen expression answers the first one.

The experimental evidence that these are genuinely separate failures is direct. Two breast cancer cell line models selected for resistance to trastuzumab emtansine failed by different routes. One showed reduced HER2 with upregulation of the efflux transporter MDR1. The other showed loss of the lysosomal transporter SLC46A3 together with PTEN deficiency. An MDR1 inhibitor reversed the first and a PI3K inhibitor reversed the second16. Same drug, same endpoint, unrelated mechanisms and non-interchangeable rescues.

Antigen loss is real and is covered in HER2 loss as an acquired resistance mechanism. Subclonal structure complicates it, because high-expressing and low-expressing populations behave as interacting compartments rather than as one drifting average12.

Antigen abundance is not the whole story even before resistance develops. In DAISY, responses occurred in patients scored immunohistochemistry 0, and within that group response was no more frequent above the median ERBB2 messenger RNA than below it17. Whatever governs response at low expression is not the amount of target.

Payload handling is the part with clinical signal and no assay. Eighty-five patients received both trastuzumab deruxtecan and sacituzumab govitecan. PTEN loss was associated with de novo resistance to trastuzumab deruxtecan, with a hazard ratio of 3.20 and a 95% confidence interval of 1.47 to 6.97. For sacituzumab govitecan the hazard ratio was 1.18, with an interval of 0.54 to 2.5618. The two conjugates carry the same payload class, and the same genomic lesion affected only one of them.

Repair capacity should matter for topoisomerase-I payloads, because the lesion those payloads create is a DNA lesion. The preclinical argument is coherent. The clinical evidence that repair capacity predicts conjugate benefit in breast cancer is not yet there. Conjugate design and the bystander argument are developed in Antibody-drug conjugates and targeted delivery.

Caution

Measuring antigen expression at progression answers one of six questions. A tumour that still expresses the target may have become resistant through trafficking, efflux, payload tolerance or loss of a downstream dependency, and no routine assay separates those. Reporting retained expression as evidence that the conjugate should still work is not supported.

6 · PARP inhibitor and platinum resistance, including restoration of repair

Restoration of homologous recombination is the mechanism with the clearest evidence and the cleanest logic. Secondary intragenic mutations that restore the wild-type BRCA2 reading frame were found in three settings. They appeared in a cisplatin-resistant breast cancer cell line, in five independent clones of a pancreatic line selected under cisplatin, and in a recurrent ovarian carcinoma that had acquired cisplatin resistance19. Clones carrying those reversions were resistant both to cisplatin and to PARP inhibition. The deficiency that created the vulnerability was repaired, and the vulnerability went with it.

Cross-resistance between platinum and PARP inhibition follows directly, because both exploit the same defect. That is a mechanistic prediction and it is borne out.

Routes that do not involve reversion matter as much in practice. Loss of PARP trapping, drug efflux, restoration of end resection through loss of 53BP1 or shieldin components, and protection of stalled replication forks all restore tolerance without restoring the gene. The mechanisms are set out in PARP trapping and synthetic lethality and Reversion mutations and other routes to PARP inhibitor resistance, and the adjuvant setting in OlympiA.

Caution

A homologous recombination deficiency score is a record of past instability. Genomic scars do not disappear when repair function returns. A tumour can therefore score Homologous recombination deficiency high and have working homologous recombination at the moment of testing. The score is a statement about history, and the drug needs a statement about the present.

7 · Immunotherapy resistance, primary and acquired

Primary resistance is the ordinary case in breast cancer, and the reason is the starting immune context rather than an escape mechanism. Most hormone receptor positive disease is not inflamed, and the subtype distribution of immune infiltration is set out in Immune contexture by subtype and the luminal deficit. A checkpoint inhibitor releases a brake on a response that has to already exist.

Acquired resistance has an identified mechanistic core, and the clearest human evidence comes from melanoma rather than from breast cancer. Four patients relapsed after an initial response to pembrolizumab, and whole-exome sequencing of paired lesions was performed. Two carried loss-of-function mutations in JAK1 or JAK2 with deletion of the wild-type allele, and a third carried a truncating mutation in B2M20. The first two lost the ability to respond to interferon gamma. The third lost surface class I major histocompatibility complex. Interferon signalling and antigen presentation are the two axes, and the breast cancer versions of both are developed in MHC loss and antigen presentation defects and Primary and acquired resistance to checkpoint blockade.

The setting where checkpoint blockade is established in this disease is early triple-negative disease treated with chemotherapy, as in KEYNOTE-522. Agent-level detail belongs to Immunotherapy and the metastatic sequencing question to Metastatic triple-negative disease.

One measurement point governs the whole section. A PD-L1 result depends on the antibody clone, the scoring algorithm and the cut-point, and those differ between assays. PD-L1 positive is the output of a specific assay, not a property of a tumour, and the predictive biomarker problem is developed in Predictive biomarkers and patient selection.

8 · Cross-resistance between classes and what it means for sequencing

Cross-resistance follows shared dependence. It does not follow shared indication, and treating two drugs as interchangeable because they treat the same disease is the recurring error here.

Where the dependence is genuinely shared, cross-resistance is real. Platinum and PARP inhibition both exploit defective homologous recombination, and a reversion that restores the gene removes both19. An ESR1 mutation confers ligand independence, so it predicts failure of oestrogen deprivation as a class while leaving a receptor-directed drug with a target3.

Where the failure modes differ, the prediction is weak. Trastuzumab and lapatinib are escaped by different routes15, so progression on one is poor evidence about the other.

Conjugates are the unsolved case, and they should be named as such. Among 85 patients treated with both trastuzumab deruxtecan and sacituzumab govitecan, progression-free survival on the second conjugate was shorter than on the first in 75.2%18. Activity was retained, so the second conjugate is a reasonable option, and the expected magnitude is smaller. Longer progression-free survival on the first conjugate predicted longer survival on the second, which is what partial rather than complete cross-resistance looks like. A genomic lesion affected one of the two and not the other despite the shared payload class.

Three things are therefore separable in a conjugate, and each carries its own cross-resistance: the antibody target, the payload class and the cell's handling of that payload. Sequencing logic is developed in Payload class, cross-resistance, and sequencing between conjugates and the clinical ordering in General principles of metastatic management.

9 · Distinguishing resistance from under-dosing, non-adherence, and pseudo-progression

Resistance is a diagnosis of exclusion and it is rarely treated as one. Four alternatives should be excluded before the word is used, because three of them are cheaper to fix than a new line of therapy.

The first is that the drug was not taken. Adherence to oral endocrine therapy is worse than most clinics assume. In a cohort of 8,769 women with hormone-sensitive early breast cancer, 32% had discontinued by 4.5 years, and among those who continued, 72% were fully adherent21. Only 49% took adjuvant endocrine therapy for the full duration on the intended schedule. Women under 40 had the highest risk of discontinuation. These are adjuvant figures and they should not be transferred unexamined to the metastatic setting. They do establish that non-adherence is common enough to be a standing hypothesis rather than a rare one.

The second is that the drug was taken and not delivered. Dose reductions made during a toxicity episode often persist after the toxicity resolves. Cycles are missed for unrelated admissions. Absorption of several oral agents depends on gastric pH or on food, and enzyme-inducing co-medication lowers exposure silently. The question to ask is what dose was actually received over the last three months, not what was prescribed.

The third is that the measurement is misleading. Tumour burden can increase before it falls under immunotherapy. That is why the response framework for these agents requires confirmation of progression on a subsequent scan, rather than treating a single assessment as definitive22. Bone disease produces the classic trap in breast cancer. Healing lytic lesions become sclerotic, and new sclerotic lesions can appear on computed tomography as previously invisible deposits mineralise. That is a picture of response reported as new lesions. Increased radiotracer uptake on bone scintigraphy in a responding patient carries the same ambiguity, since it reflects osteoblastic activity rather than tumour cell number.

The fourth is that nothing was measured. A rising tumour marker with stable imaging and a stable patient is not progression. It is a reason to shorten the interval to the next assessment.

In practice

Before recording progression, answer these questions in order.

What dose was actually taken, over what period. Ask the patient directly and ask about missed doses without implying a right answer.

What dose was actually delivered. Check reductions, delays, interacting medication and anything affecting absorption.

Is the change outside measurement error. A small increase across one interval, on a scan performed with a different technique or at a different centre, is frequently not a change.

Is the new finding a picture of response. New sclerosis in bone, and increased uptake in known lesions, both occur during successful treatment.

Is the patient actually worse. Symptoms, performance status and laboratory trajectory are three independent readings, and disagreement between them and the scan is informative rather than inconvenient.

Caution

Calling a failure resistance has consequences beyond the label. It ends the current line, applies a new selective pressure described in Temporal heterogeneity and clonal evolution, and forecloses the cheapest available intervention, which is fixing delivery. An adherence problem treated as biology produces a second failure for the same reason as the first.

10 · What to measure at progression and what it will change

The test for ordering anything is whether a different result would produce a different action available now. Applied honestly, that leaves a short list.

In practice

Worth ordering at progression in hormone receptor positive disease.

Plasma ESR1 after aromatase inhibitor exposure. A positive result changes the class of endocrine agent2,3.

PIK3CA, AKT1 and PTEN status, in tissue or plasma. Each has a targeted option attached, and the pathway selection is developed in PI3K, AKT, and mTOR directed therapy and its toxicity burden.

Germline BRCA1 and BRCA2 if never tested. The result changes eligibility and it also changes family counselling.

Worth ordering when the clinical picture changed.

Receptor status on a new biopsy of a lesion that is behaving differently from the rest. The interpretation, and the distinction between true conversion and assay variation, is in Receptor discordance and conversion.

HER2 status on recent tissue before a HER2-directed conjugate. Expression changes under treatment, and the eligibility boundary is an assay boundary HER2-low and ultralow reframed as a heterogeneity problem.

Time to progression is free information and it is underused. Early progression on endocrine therapy with a CDK4/6 inhibitor is more likely to reflect a population that was never sensitive. Late progression is more likely to have produced something findable1. The expected yield of a biopsy therefore depends on how long the last line worked.

Two technical requirements apply to any plasma test ordered here. Matched white blood cell sequencing is needed to keep clonal haematopoiesis out of the result23. And a negative plasma result in low-volume or central nervous system disease is uninformative rather than reassuring, for the shedding reasons set out in Liquid biopsy and longitudinal monitoring.

What is not worth ordering is easier to state. Anything whose result has no available action. Anything whose observer agreement is too poor to carry a decision, which is the finding that governs the HER2 0 against 1+ boundary24. And a repeat of a test whose answer was already obtained under the same selective pressure.

Interplay

Resistance is where the heterogeneity argument becomes a bedside argument. A tumour that was heterogeneous at diagnosis contains the population that will be selected, so the composition at progression was partly determined before the first dose. The taxonomy in this chapter and the evolutionary account in Temporal heterogeneity and clonal evolution describe one process at two time points, and the assay chosen at progression determines which of the two is visible.

See Integrative biological interplay

References

  1. 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
  2. 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
  3. 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
  4. 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
  5. Massarweh S, et al. Tamoxifen resistance in breast tumors is driven by growth factor receptor signaling with repression of classic estrogen receptor genomic function. Cancer Res 2008. PMID 18245484
  6. Houston SJ, et al. Overexpression of c-erbB2 is an independent marker of resistance to endocrine therapy in advanced breast cancer. Br J Cancer 1999. PMID 10098763
  7. Belli S, et al. EGFR and HER2 hyper-activation mediates resistance to endocrine therapy and CDK4/6 inhibitors in ER+ breast cancer. Cancer Lett 2024
  8. Toska E, et al. PI3K pathway regulates ER-dependent transcription in breast cancer through the epigenetic regulator KMT2D. Science 2017. PMID 28336670
  9. Watt AC, et al. Rb-driven transcription limits its tumour-suppressive effects in breast cancer. Nature 2026. PMID 42587147
  10. Turner NC, Liu Y, Zhu Z, et al. Cyclin E1 expression and palbociclib efficacy in previously treated hormone receptor-positive metastatic breast cancer. J Clin Oncol 2019 37:1169-1178. PMID 30807234
  11. Gupta A, Gazzo A, Selenica P, et al. APOBEC3 mutagenesis drives therapy resistance in breast cancer. Nat Genet 2025 57:1452-1462. doi:10.1038/s41588-025-02187-1
  12. 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
  13. Giuliano M, et al. Upregulation of ER signaling as an adaptive mechanism of cell survival in HER2-positive breast tumors treated with anti-HER2 therapy. Clin Cancer Res 2015. PMID 26015514
  14. Hanker AB, et al. Co-occurring gain-of-function mutations in HER2 and HER3 modulate HER2/HER3 activation, oncogenesis and HER2 inhibitor sensitivity. Cancer Cell 2021. PMID 34171264
  15. Wang YC, et al. Different mechanisms for resistance to trastuzumab versus lapatinib in HER2-positive breast cancers — role of estrogen receptor and HER2 reactivation. Breast Cancer Res 2011. PMID 22123186
  16. Li G, Guo J, Shen BQ, et al. Mechanisms of acquired resistance to trastuzumab emtansine in breast cancer cells. Mol Cancer Ther 2018 17:1441-1453. PMID 29695635
  17. Mosele F, Deluche E, Lusque A, et al. Trastuzumab deruxtecan in metastatic breast cancer with variable HER2 expression: the phase 2 DAISY trial. Nat Med 2023 29:2110-2120. PMID 37488289
  18. Mai N, Klar Lieberman MM, Ferraro E, et al. Sequential antibody-drug conjugate therapy in patients with metastatic breast cancer treated with sacituzumab govitecan and trastuzumab deruxtecan. JCO Precis Oncol 2025 9:e2400898. PMID 40440568
  19. Sakai W, Swisher EM, Karlan BY, et al. Secondary mutations as a mechanism of cisplatin resistance in BRCA2-mutated cancers. Nature 2008 451:1116-1120. PMID 18264087
  20. Zaretsky JM, Garcia-Diaz A, Shin DS, et al. Mutations associated with acquired resistance to PD-1 blockade in melanoma. N Engl J Med 2016 375:819-829. PMID 27433843
  21. Hershman DL, Kushi LH, Shao T, et al. Early discontinuation and nonadherence to adjuvant hormonal therapy in a cohort of 8,769 early-stage breast cancer patients. J Clin Oncol 2010 28:4120-4128. PMID 20585090
  22. Seymour L, Bogaerts J, Perrone A, et al. iRECIST: guidelines for response criteria for use in trials testing immunotherapeutics. Lancet Oncol 2017 18:e143-e152. PMID 28271869
  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. Robbins CJ, Fernandez AI, Han G, et al. Multi-institutional assessment of pathologist scoring HER2 immunohistochemistry. Mod Pathol 2023 36:100032. PMID 36788069