Part VI · Heterogeneity, evolution, and metastatic biology · Chapter 36
Integrative biological interplay
Twelve pairings. Each one joins two arguments the book has already made separately.
1 · HER2 heterogeneity and hormone receptor signaling
The two receptor systems are each other's escape route, and in a tumour that carries both, the compartments are not the same compartments.
Under HER2 blockade, oestrogen receptor signalling is upregulated as an adaptive survival mechanism1. Under endocrine therapy the traffic runs the other way, with HER2 protein rising after short-term neoadjuvant endocrine exposure in hormone receptor positive HER2-negative disease2. Truncated HER2 forms down-modulate the oestrogen receptor, so the relationship is not a simple see-saw3. What the tumour holds is two dependencies, each partly able to cover for the other.
Heterogeneity makes this concrete rather than abstract. HER2 heterogeneity establishes that HER2 status is a proportion of cells and not a property of a tumour. Hormone receptor status is a proportion too. The cells that escape HER2 blockade are therefore not necessarily the cells that escape oestrogen deprivation. A report saying both receptors are positive has averaged twice, once per assay, and possibly over different populations.
The consequence for treatment is that dual blockade is not redundancy. It closes a route rather than duplicating an effect, which is why the endocrine question in HER2-positive disease is a biological question and not only a tolerability one Hormone receptor-positive HER2-positive disease and endocrine integration.
Receptor cross-talk is usually taught as a signalling fact about a cell. It is also a compositional fact about a tumour, because the two dependencies are distributed unevenly across it. Cross-talk and heterogeneity are the same problem measured with different assays.
2 · HER2 expression, payload sensitivity, and bystander killing
A bystander-capable payload converts a proportion problem into a distance problem. The quantity that matters stops being the fraction of antigen-positive cells and becomes the distance from an antigen-positive cell to an antigen-negative one, relative to how far the released payload travels.
Two datasets are the two halves of this argument. In a neoadjuvant study of T-DM1 with pertuzumab, no patient with a HER2-heterogeneous tumour achieved a pathologic complete response, against 55% of patients whose tumours were not heterogeneous4. T-DM1 carries a payload that does not readily cross membranes. In DAISY, trastuzumab deruxtecan produced confirmed responses in 29.7% of a cohort of 40 patients with HER2 non-expressing disease scored immunohistochemistry 05. That payload does cross.
The finding that discriminates the two models sits inside the second study. Among patients scored immunohistochemistry 0, response was no more frequent above the median ERBB2 messenger RNA than below it5. A dose-of-antigen model predicts a gradient there. A distance model does not require one.
Chemistry is in Bystander killing and why it matters in heterogeneous tumors and the eligibility lesson in Antigen thresholds as eligibility criteria, the general lesson. The pairing here is that the spatial arrangement described in HER2 heterogeneity is the variable that payload chemistry acts on.
Spatial pattern and payload design are a matched pair. Clustered antigen loss defeats a non-diffusing payload and a scattered pattern does not, so the same tumour is a different problem for two conjugates that share a target.
See Antibody-drug conjugates and targeted delivery
3 · Tumor genotype and the immune microenvironment
Genotype sets the antigen supply. It does not set the immune context, and the gap between those two statements is where most disappointing biomarker results in this disease have come from.
The supply side is real. Mutational processes generate the variants a T cell could see, and the process that generates therapy-resistance hotspots in this disease is the same one that generates much of its mutational load6. Genomic instability also feeds innate sensing through the cytosolic DNA pathway, which is the bridge built in Genomic instability, cGAS-STING, and the bridge to immune recognition.
The context side is not predicted by the genome. Integrated spatial proteomics across 280 tumour regions found that proteomic heterogeneity rose with tumour progression independently of genomic heterogeneity, and tracked instead with microenvironmental differences7. The same work found macrophages and T cells infiltrating higher-grade regions while anti-inflammatory pathways were more highly expressed in exactly those infiltrated regions.
That last observation is the pairing in one sentence. Infiltration and effective immunity are different variables, and a genotype-derived prediction of immunogenicity is a prediction about one of them. Antigen also has to survive presentation, which fails independently MHC loss and antigen presentation defects.
Tumour mutational burden and immune contexture are measured on the same tumour and are not the same axis. Breast cancer is where that dissociation is most visible, because its most mutagenic processes sit in subtypes with the least productive immune infiltrate.
4 · Clonal evolution and treatment sequencing
A treatment sequence is a series of selections, and each line hands the next one its starting population. Sequencing is therefore an evolutionary decision whether or not it is made as one.
The escape routes are known by class. Aromatase inhibition selects ESR1 mutation8. CDK4/6 inhibition selects loss of the node the drug acts through9. HER2-directed therapy selects the compartment that does not display or does not depend on the receptor10. Each of those is a prediction available before the line starts, which is what makes the class more informative about mechanism than the imaging pattern is.
SERENA-6 showed that the selection can be acted on before it becomes radiographic, with the switch triggered by a molecular event in a patient still responding by conventional assessment11. The same argument applied to conjugates is unresolved, because the second conjugate in a sequence performs worse than the first in most patients without failing outright12.
The pairing is that Temporal heterogeneity and clonal evolution describes what each class selects for and Mechanisms of therapeutic resistance describes what that selected population then resists. Order determines which dependencies are still standing when the options narrow.
Sequencing decisions are usually justified by what a trial showed in an unselected population. The evolutionary version of the question is different and better posed. What did the last line select for, and does the next agent depend on something that population still has.
5 · Organ microenvironment and metastatic phenotype
The same clone does not behave the same way in bone, liver, lung and brain. A phenotype measured at one site is a joint property of the clone and the site, and separating those two contributions is rarely possible from one biopsy.
Three consequences run through the book. Receptor discordance between deposits is partly a site effect rather than a clonal difference, which is why Discordance across metastatic sites within the same patient and Rebiopsy, when, which lesion, and when the result should change management treat lesion choice as part of the assay. Drug access differs by site, and the microenvironmental determinants of that are set out in Vascular permeability, interstitial pressure, and drug delivery. Shedding into plasma differs by site, so a plasma result systematically under-represents intracranial disease Cerebrospinal fluid analysis in leptomeningeal disease.
A correlative study makes the delivery point concretely. In 24 metastases from patients receiving T-DM1, extracellular matrix organisation pathways were enriched in lesions with low zirconium-89 trastuzumab uptake, and hypoxia and matrix processes were enriched in lesions without metabolic response13. The sample is small and the design is hypothesis-generating. It is also a direct observation that the stroma around a lesion tracks with whether an antibody reached it.
Organ-specific biology is developed in Metastatic dissemination and organ tropism and the clinical management of each site in Organ-specific metastases and complications. The pairing is that a biomarker result carries an unstated location.
Organotropism is usually framed as which organs a tumour reaches. The more useful framing for a biomarker reader is that the organ then changes what the tumour is, so the site of a biopsy is a variable in every result taken from it.
6 · Spatial heterogeneity, biopsy selection, and biomarker interpretation
Which lesion is biopsied is a biomarker decision made before any assay runs. It is almost never recorded as one, and it constrains the result more tightly than most assay parameters do.
The magnitude is measurable. Multi-region sequencing of 50 primary breast cancers across 303 samples found potentially targetable mutations to be subclonal in 13 of the 5014. Spatial proteomics across 280 regions found progression-associated heterogeneity that single-region sampling cannot see7. Sampling scheme is therefore a covariate of the reported result, not a detail of methods.
One asymmetry follows and it is useful at the bench and the bedside. A positive result cannot be produced by sampling the wrong region, so a positive finding is robust to sampling. A negative result can be produced that way, so a negative finding on a single core is weaker evidence than its report suggests. Assay validity as a separate problem is set out in Receptor assessment and the measurement problem.
The pairing joins Spatial heterogeneity, which establishes that receptors vary across space, to Prognostic and predictive assays and somatic profiling, which assumes a single representative sample. Both are correct within their own frame. The inconsistency is only visible when they are read together.
Sampling is treated as a source of noise. It is better treated as a parameter of the measurement, because it has a direction. Undersampling loses positives and does not manufacture them.
7 · Host biology, treatment exposure, and tumor adaptation
The dose a tumour experiences is a host variable, and resistance is defined against an exposure rather than against a drug.
Several host factors set that exposure. Adherence is the largest and most neglected, with only 49% of 8,769 women taking adjuvant endocrine therapy for the full duration on the intended schedule15. Hepatic and renal handling, drug interactions, body composition and gastric pH all move exposure without anyone recording that it moved. Host factors that change efficacy and toxicity covers the host determinants of efficacy and toxicity directly.
The interaction with tumour biology is the part worth stating carefully. Sub-therapeutic exposure is not a smaller version of treatment. It applies selection without applying eradication, which is the condition under which resistant populations expand fastest. An adherence problem and a biological resistance problem therefore produce the same clinical picture through opposite mechanisms, and the distinction is drawn in Distinguishing resistance from under-dosing, non-adherence, and pseudo-progression.
Host biology also changes the environment the tumour adapts within. Chronic inflammation, adrenergic signalling and metabolic context are developed across Host and tumor interactions, and they are inputs to adaptation rather than background.
Host factors are usually filed under tolerability. They belong equally under resistance, because they determine the exposure against which sensitivity is defined and the environment in which adaptation happens.
8 · Dormancy, immune control, and the timing of intervention
Dormancy is not absence of disease. It is a controlled state, and immune surveillance is one of the controls maintaining it. That single reframing changes what an intervention is trying to do.
Two windows exist and they call for opposite properties in a drug. Acting during dormancy means acting on quiescent cells, which most cytotoxics cannot reach because they depend on proliferation. Acting at escape means acting on a growing population, which is tractable pharmacologically and later than anyone would like. The dormancy machinery itself, including the signalling balance that holds the state, is developed in Dormancy, residual disease, and late recurrence.
Molecular residual disease defines the interval where the question is live. In 55 patients with early breast cancer, mutation tracking in plasma predicted relapse with a median lead time over clinical relapse of 7.9 months16. Whether acting inside that interval improves an outcome is the open question owned by Lead time and the unresolved question of whether acting early helps, and it should not be presented as settled.
The pairing is that immune control is a mechanism of dormancy, so immune-directed intervention is a candidate for the dormant window rather than only for the measurable one. The natural killer cell biology behind that is in Immune-mediated dormancy and natural killer cell control.
Late recurrence in hormone receptor positive disease is usually discussed as a duration-of-therapy question. It is also an immunology question, because something maintained control for a decade and then stopped. Annual hazard of recurrence and the long tail of hormone receptor-positive disease gives the hazard, and Breast cancer immunology gives one candidate for what changed.
9 · Endocrine resistance and immune evasion as one program
Hormone receptor positive disease is simultaneously the most endocrine-dependent and the least immune-infiltrated category in this book. Treating that as a coincidence has been costly.
The subtype-level observation is established, and Immune contexture by subtype and the luminal deficit documents the luminal immune deficit directly. The mechanistic claim goes further. Oestrogen receptor signalling participates in local immunosuppression and in the control of antigen presentation, which is the argument set out in Hormone receptor driven immune evasion and antigen presentation. On that reading, endocrine dependence and immune quiescence are two readings of one transcriptional program rather than two properties that happen to co-occur.
If the reading is right, two predictions follow. Relieving endocrine dependence should change immune context, and the endocrine-resistant state should not inherit the immune quiescence automatically. Both are testable and neither is established.
The honest limit belongs in the text rather than in a footnote. There is no established clinical evidence that endocrine manipulation improves benefit from checkpoint blockade in luminal breast cancer. The combination rationale is developed in Combination with chemotherapy and targeted agents, and the reason response rates in this disease sit where they do in Why response rates in breast cancer sit where they do.
The luminal immune deficit is normally presented as a reason not to use immunotherapy in this subtype. The pairing suggests a different question, which is whether the deficit is a consequence of the dependency being treated and therefore modifiable by treating it.
10 · The immune, stromal, and tumor triad as a single system
Three compartments, each heterogeneous, each measured by a different assay, all interacting within millimetres. Any single score for any one of them is an average over a system whose components vary at a finer scale than the score.
The measurement evidence is direct. Spatial proteomics found immune infiltration rising with grade while anti-inflammatory pathways were most expressed in the infiltrated regions themselves7. A stromal reading and an immune reading taken from the same region can therefore point in opposite directions without either being wrong.
The stroma is not a passive third party. In 24 metastases, extracellular matrix pathways were enriched in lesions with low uptake of labelled antibody and in lesions without metabolic response to T-DM113. Cellular architecture is in Cellular architecture of the tumor microenvironment, functional states in Functional states of the microenvironment, and what spatial resolution adds over bulk profiling in What spatial data adds over bulk profiling, and at what cost.
The consequence for A framework for heterogeneity is that heterogeneity is not one quantity with several measurements. It is several quantities, and the tumour compartment is only one of them. A study that measures tumour cell heterogeneity and concludes about tumour heterogeneity has dropped two thirds of the system.
Stromal composition is usually reported as a prognostic average for a tumour. Treating it as a local variable changes what it predicts, because drug delivery and immune access are determined lesion-by-lesion and region-by-region rather than patient-by-patient.
11 · Ancestry, environment, and biological embedding
Three variables are routinely collapsed into one, and the collapse causes most of the confusion in this literature. Genetic ancestry indexes allele frequencies. Race indexes social experience and exposure. Place indexes environment. Race, ethnicity, and ancestry as three different variables separates them properly and this section assumes that separation.
Biological embedding is the mechanism by which the second and third become measurable in tissue. Chronic stress, allostatic load and adrenergic signalling have tissue-level consequences, and Social adversity as biology, allostatic load and the bridge to Part XIII is where the book brings that into biology rather than leaving it as an addendum. The resulting differences are biological, and they are not therefore ancestral.
The inference rule follows and it is worth stating plainly. A tumour-biology difference observed between groups defined by race is not evidence of an ancestry-linked genetic difference. Designs that can separate the two are set out in Decomposing an outcome gap into biology, access, and delivery and Ancestry-associated tumor biology, what replicates and what is confounded. Where an ancestry-associated difference in tumour biology does replicate, its clinical use depends on whether the associated therapy was tested in the population in question, which is the trial representation problem in Clinical trial representation and the mechanisms of exclusion.
The pairing matters for this part specifically. Interpatient heterogeneity is the first row of the taxonomy in A framework for heterogeneity, and it is the row where biological and social explanations are most often confused for one another.
Disparities enter this book as biology rather than as an appendix. The bridge is that social exposure has measurable tissue consequences, so a difference can be both socially produced and biologically real without being genetically determined.
See Disparities across the breast cancer continuum
12 · Bystander-capable modalities as a heterogeneity-tolerant strategy
This is the part's constructive answer, and it is narrower than it first appears.
If heterogeneity is the problem, one class of solution does not require every cell to carry the target. Three such modalities exist. A conjugate with a membrane-permeable payload kills neighbours of the cell that internalised it Bystander killing and why it matters in heterogeneous tumors. Immune effector mechanisms kill by recognising a context rather than a receptor, and redirected T cell approaches extend that further Cellular therapies and immune-engaging antibodies. Radiation kills within a field rather than by target engagement.
The empirical anchor is that a bystander-capable conjugate produced responses in disease scored immunohistochemistry 0, where a target-abundance model predicts almost none5.
Two limits keep this from becoming a slogan. Heterogeneity-tolerant is not heterogeneity-proof. A conjugate carrying a non-diffusing payload failed precisely where heterogeneity was present4, so the property belongs to the payload rather than to conjugates as a class. Tolerance also degrades on repeat exposure, since the second conjugate in a sequence underperformed the first in 75.2% of 85 patients12.
The design consequence is the one Antigen thresholds as eligibility criteria, the general lesson owns. An antigen expression threshold used as an eligibility criterion assumes a dose-of-antigen model. For a bystander-capable agent that assumption is not safe, and the reproducibility of the boundary being used is itself poor17.
The honest summary of this chapter is the same as the honest summary of the part. Heterogeneity is measurable, it has prognostic weight in several prospective datasets, and exactly one modality class currently exploits it rather than merely suffering from it.
The framework chapter asks when heterogeneity changes a decision. This is the clearest answer the book can give. It changes the decision when a bystander-capable modality is available. That is the one situation in which the spatial arrangement of the target, rather than its average abundance, predicts what the drug will do.
References
- 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
- Chaudhary LN, et al. Frequent upregulation of HER2 protein in hormone-receptor-positive HER2-negative breast cancer after short-term neoadjuvant endocrine therapy. Breast Cancer Res Treat 2023. PMID 37460683
- Parra-Palau JL, et al. A major role of p95/611-CTF, a carboxy-terminal fragment of HER2, in the down-modulation of the estrogen receptor in HER2-positive breast cancers. Cancer Res 2010. PMID 20978202
- Metzger Filho O, Viale G, Stein S, et al. Impact of HER2 heterogeneity on treatment response of early-stage HER2-positive breast cancer: phase II neoadjuvant clinical trial of T-DM1 combined with pertuzumab. Cancer Discov 2021 11:2474-2487. PMID 33941592
- 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
- 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
- Mardamshina M, Karagach S, Mohan V, et al. Integrated spatial proteomic analysis of breast cancer heterogeneity unravels cancer cell phenotypic plasticity. Nat Commun 2025 16:10482. PMID 41290667
- 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
- 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
- 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
- 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
- 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
- Rediti M, Fimereli D, Mileva M, et al. Integrating molecular imaging and transcriptomic profiling in advanced HER2-positive breast cancer receiving trastuzumab emtansine: an analysis of the ZEPHIR clinical trial. Clin Cancer Res 2025 31:110-121. PMID 39470686
- 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
- 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
- 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
- Robbins CJ, Fernandez AI, Han G, et al. Multi-institutional assessment of pathologist scoring HER2 immunohistochemistry. Mod Pathol 2023 36:100032. PMID 36788069