Part VI · Heterogeneity, evolution, and metastatic biology · Chapter 28
HER2 heterogeneity
One word on a report, four different claims underneath it.
1 · Definitions and thresholds, spatial, clonal, and the ten percent convention
A pathology report that says heterogeneous is making one of several different claims. It may mean that amplified and non-amplified cells sit in separate regions of the section. It may mean they are interspersed. It may mean that a minority of cells cross the amplification threshold while the tumour as a whole does not. It may mean that two biopsies from the same patient disagree. These are different findings with different causes and different consequences.
The working convention counts cells. A proportion of tumour nuclei is scored as amplified, and a threshold is applied to that proportion. The convention exists so that two observers can produce the same answer from the same slide. It is a counting rule.
It is worth being blunt about what a counting rule is not. A threshold at ten percent does not assert that ten percent is where biology changes. Nobody has shown that. The number was chosen so that the category could be assigned reproducibly, and reproducibility is a property of the assay rather than of the tumour. Most of the published disagreement in this field comes from reading a counting rule as a biological claim.
The obvious reading of a heterogeneity threshold is that it marks a point where the disease behaves differently. It does not. It marks a point where observers can agree. Those two statements support very different conclusions about what to do next.
2 · Spatial patterns, clustered, mosaic, and scattered
The same overall proportion of amplified cells can be arranged in quite different ways. Amplified cells may form a discrete cluster, occupying one region and leaving the rest of the tumour unamplified. They may be mosaic, with amplified and unamplified populations interleaved at the scale of small groups. They may be scattered as single cells within an unamplified background.
The distinction matters for two separate reasons, and they should not be run together. The first is sampling. A clustered pattern can be missed entirely by a core that happens to come from the wrong region, and it can be over-called by a core that comes from the right one. A scattered pattern is less sensitive to where the needle went. The second reason is delivery. A drug that kills neighbouring cells has a different problem with a two-centimetre amplified cluster than with single amplified cells distributed through the tumour.
A report that gives a percentage without describing the arrangement has discarded the information that distinguishes these two problems.
3 · Low-level amplification and its association with heterogeneity
Tumours with ratios close to the amplification threshold are enriched for heterogeneity. This is partly real and partly arithmetic. The arithmetic part is straightforward. A whole-section ratio is an average over cells. An average lands near the threshold either because most cells sit near it, or because a strongly amplified minority is diluted by an unamplified majority. Those two tumours are not the same, and a single ratio cannot separate them.
The real part is that low-level amplification is often the visible signature of an unstable amplification mechanism rather than a stable intermediate state. That argument is developed in Amplicon topology, extrachromosomal DNA, and the mechanistic origin of instability.
4 · Genetic and protein-level heterogeneity and their imperfect correlation
Amplification and overexpression are measured by different assays, and the assays fail in different ways. In situ hybridisation counts signals in nuclei and reports a copy number relationship. Immunohistochemistry reports the amount of protein at the membrane, through an antibody, a detection chemistry and a human eye. The analytic validity framework for both is set out in Receptor assessment and the measurement problem and is not restated here.
The two forms of heterogeneity are correlated, and the correlation is far from complete. A tumour can be uniformly amplified and variably stained. It can carry amplified subclones whose protein output is not high enough to register. Protein level is downstream of copy number, and several things intervene, including transcription, translation, receptor turnover and the fixation the specimen happened to receive.
Keeping the two apart is the single most useful discipline in this chapter. A claim about amplification is a claim about DNA. A claim about staining is a claim about protein measured by a particular assay on a particular block. Much of the literature reports one and concludes about the other.
5 · Amplicon topology, extrachromosomal DNA, and the mechanistic origin of instability
Heterogeneity in most of this book is described. Here it can be explained, which is why the chapter sits where it does.
An amplified gene has to live somewhere. It may be amplified in place on its chromosome, as tandem repeats within a chromosomal arm. It may be carried on extrachromosomal DNA, as circular elements with no centromere.
The absence of a centromere is the whole argument. Chromosomal amplicons segregate through the mitotic spindle, and both daughter cells inherit approximately what the parent had. Circular elements without centromeres are distributed to daughter cells without a mechanism that enforces equality. Copy number therefore drifts between daughters at every division. Under selection for high copy number the population shifts up. Remove that selection and it shifts down.
This gives heterogeneity a generator rather than a description. A tumour carrying its ERBB2 amplification extrachromosomally does not merely happen to be heterogeneous at the moment it was biopsied. It produces heterogeneity continuously, and it will keep producing it under treatment. The consequence for temporal evolution is taken up in Temporal heterogeneity and clonal evolution.
6 · Breakage-fusion-bridge and extrachromosomal amplification behave differently
Breakage-fusion-bridge cycles amplify a region through repeated breakage, fusion of sister chromatids and asymmetric division of the resulting bridge. The product is intrachromosomal, usually arranged head to head, and often accompanied by amplification of neighbouring genes that were never under selection.
Extrachromosomal amplification produces a different signature. Copy number is typically higher, the amplified segment is more sharply bounded, and the variance between cells is wider.
The two mechanisms are worth telling apart because they predict different behaviour under treatment. Intrachromosomal amplification is comparatively stable once established, and removing the drug does not readily remove the amplicon. Extrachromosomal amplification is reversible in the population sense, because unequal segregation works in both directions. Preclinical models of HER2 heterogeneity show subclonal dynamics of exactly this kind during the evolution of resistance to HER2-directed therapy, with high-expressing and low-expressing populations behaving as interacting compartments rather than as a single drifting average1.
7 · Reported frequency and its dependence on detection method and sampling
Published estimates of how often HER2-positive breast cancer is heterogeneous vary widely. The variation is mostly methodological, and the sources are worth naming individually.
The first is which assay was used. Immunohistochemistry and in situ hybridisation report different quantities, as set out in Genetic and protein-level heterogeneity and their imperfect correlation.
The second is how many cells were counted. Any proportion estimated from twenty nuclei carries wide uncertainty, and thresholds near the low end are especially unstable at small counts.
The third is which threshold was applied, and whether the denominator was all tumour cells or only the cells in a selected field.
The fourth is what tissue was examined. A single core, a whole section and multiple regions of a resection give systematically different answers, and the difference increases with the size of the tumour.
A frequency quoted without those four facts is not comparable to another frequency quoted without them. This is the same problem as the discordance literature in Receptor discordance and conversion, arriving through a different door.
8 · Consequences for response to antibodies and tyrosine kinase inhibitors
An antibody against HER2 acts on cells that display HER2. A tumour in which a substantial subpopulation does not display it contains, at the outset, a compartment on which the antibody has no direct effect. Nothing about this is subtle. It predicts incomplete response rather than no response, and it predicts that the residual population will be enriched for the unamplified compartment.
Small-molecule inhibitors of the receptor face the same arithmetic, with one difference. They are not restricted to the cell surface and they do not depend on immune effector engagement, so their failure mode against a low-expressing compartment is driven by dependence rather than by access. A cell that does not rely on HER2 signalling is unaffected by blocking it, whatever reaches the cytoplasm.
9 · Consequences for pathologic complete response in the neoadjuvant setting
The neoadjuvant setting gives the cleanest available test, because response is measured in tissue rather than inferred from imaging.
In a phase II neoadjuvant study of T-DM1 with pertuzumab, no patient whose tumour was heterogeneous for HER2 achieved a pathologic complete response. Among patients whose tumours were not heterogeneous, 55% did2. A separation of that size, in that direction, from a prospectively defined assessment, is the strongest single argument in this chapter.
Two cautions belong with it. The heterogeneous group was small, and a proportion of zero has wide confidence limits. And the finding is about a conjugate given with an antibody, so it should not be transferred unexamined to every HER2-directed regimen.
This is where heterogeneity stops being a descriptive property and becomes a predictive one. Mean expression and spatial variance are separate axes. A tumour with adequate average expression and a large unamplified compartment is not the same clinical problem as a tumour with uniformly moderate expression, even though a single ratio may report them identically.
10 · HER2 loss as an acquired resistance mechanism
Heterogeneity at diagnosis and antigen loss under treatment are the same phenomenon observed at two times. If a tumour contains a compartment that does not depend on HER2, treatment that works on the dependent compartment will leave the other one behind. What is then sampled at progression is a population selected for low expression.
This is now recognised as a route to resistance against HER2-directed conjugates and not only against antibodies. Preclinical models carrying mixed high and low expressing populations reproduce it, and they show the two populations cooperating rather than simply competing1.
The practical consequence is that a HER2-negative result on a biopsy taken at progression should not be read as a laboratory error, nor as proof that the original diagnosis was wrong. It may be an accurate measurement of what selection has left.
11 · Implications for conjugate payload delivery and the bystander effect
A conjugate whose payload crosses membranes can kill cells adjacent to the ones that internalised it. That is the reason a heterogeneous tumour can respond to a conjugate when it would not respond to a naked antibody. The chemistry that permits it is covered in Antibody-drug conjugates and targeted delivery.
The clinical evidence is more interesting than the mechanism alone would predict. In DAISY, trastuzumab deruxtecan produced confirmed objective responses in 70.6% of patients with HER2-overexpressing disease, 37.5% with HER2-low disease, and 29.7% with HER2 non-expressing disease, the last of these in 40 patients scored immunohistochemistry 03. Activity declined with expression, which is expected. Activity at immunohistochemistry 0 was not near zero, which is not.
The same study pressed further, and the result deserves attention. Among patients whose tumours scored immunohistochemistry 0, response was no more frequent in those with ERBB2 messenger RNA above the median than in those below it3. Within that group, the amount of target did not predict who responded.
A simple dose-of-antigen model predicts that more target means more response, all the way down. Within immunohistochemistry 0 disease that prediction failed. Either the assay is not measuring the quantity that matters at low levels, or something other than antigen abundance is governing response. The distinction is unresolved, and it is the reason this section does not end with a threshold.
12 · HER2-low and ultralow reframed as a heterogeneity problem
HER2-low is a category defined by an assay boundary, and the boundary was inherited rather than designed. The immunohistochemistry assay in routine use was validated to separate tumours with receptor overexpression from tumours without it. It is now being asked to separate 0 from 1+, a distinction it was never validated to make.
The reproducibility data are not reassuring. In a multi-institutional study, 18 breast pathologists from 15 institutions scored 170 breast cancer biopsies. On cases scored immunohistochemistry 0, overall agreement was 25%, with a Fleiss kappa of 0.494. Agreement across the full four-category system was 28.8%. Collapsing to three categories raised it to 46.5%. On the separate question of whether a case is 3+ or not 3+, agreement plateaued at 87.1%. The assay is doing what it was built to do and failing at what it has been repurposed to do.
The guideline position is unusually candid about this. The 2023 ASCO and College of American Pathologists update affirmed the existing recommendations and declined to create new result categories, stating that it is premature to do so, while acknowledging that the 0 against 1+ distinction is now clinically relevant because of the entry criteria of the trial that supported regulatory approval5. DESTINY-Breast04 defined eligibility as immunohistochemistry 1+, or 2+ with negative in situ hybridisation, and excluded immunohistochemistry 06.
Read together with Implications for conjugate payload delivery and the bystander effect, the position is coherent. A treatment decision now rests on a boundary that trained observers reproduce a quarter of the time, in a setting where the evidence that the boundary marks a difference in drug behaviour is weak. The category is an eligibility construct. Calling it a biological entity is a separate claim, and it has not been established.
13 · HER2 heterogeneity and hormone receptor signaling in triple-positive disease
Tumours that are both hormone receptor positive and HER2 positive carry two dependencies, and neither is uniform across the tumour. The compartment that escapes HER2 blockade is not necessarily the compartment that escapes endocrine therapy. Cross-talk between the two pathways is developed in Estrogen receptor signaling and endocrine resistance and HER family signaling.
The point specific to this chapter is that heterogeneity in one receptor and heterogeneity in the other are separately measured and separately consequential. A report describing both as positive has averaged twice.
14 · Trial design implications, eligibility, stratification, and endpoints
If heterogeneity predicts response, a trial that does not measure it is randomising an unmeasured prognostic variable. Three design choices follow, and they are not interchangeable.
Heterogeneity can be an eligibility criterion, which produces a clean population and a result that does not generalise to the patients who were excluded. It can be a stratification factor, which preserves generalisability and requires the assessment to be reproducible enough to stratify on. It can be a prespecified secondary analysis, which is the weakest option and the only one available retrospectively.
The obstacle to the first two is the reproducibility documented in HER2-low and ultralow reframed as a heterogeneity problem. A variable that observers assign inconsistently cannot carry an eligibility boundary without importing that inconsistency into the trial population. This is the same constraint that governs any assay-defined enrolment criterion, and it is developed in Clinical trials and evidence interpretation.
15 · What to do with a heterogeneous HER2 report in clinic today
Most of this chapter is biology. This section is not.
Read the report for what it actually says. A percentage of amplified cells, a spatial description, and the assay used are three different facts. If the report gives only one, the others are not absent from the tumour. They are absent from the report, and they can be asked for.
Do not downgrade HER2-directed treatment on the basis of heterogeneity. There is no prospective evidence that a heterogeneous tumour should be treated less intensively. The evidence points the other way, because the heterogeneous group responded worse rather than better2.
Use it to calibrate expectation in the neoadjuvant setting. A heterogeneous tumour is less likely to reach a pathologic complete response. That changes what a residual finding means, and it changes the conversation held before treatment starts rather than after.
Rebiopsy at progression, and interpret a fall in HER2 as possible selection rather than as error. HER2 loss as an acquired resistance mechanism gives the reasoning.
Do not reclassify a patient out of HER2-positive disease on a single heterogeneous result. A minority amplified population in a tumour that is otherwise unamplified is a different finding from a tumour that has lost amplification, and one biopsy often cannot distinguish them.
The honest summary is narrow. Heterogeneity is measurable, it carries prognostic weight in at least one prospective neoadjuvant dataset, and it has no established role in selecting therapy. The gap between those statements is where the next decade of work in this area sits.
References
- 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
- 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
- Robbins CJ, Fernandez AI, Han G, et al. Multi-institutional assessment of pathologist scoring HER2 immunohistochemistry. Mod Pathol 2023 36:100032. PMID 36788069
- Wolff AC, Somerfield MR, Dowsett M, et al. Human epidermal growth factor receptor 2 testing in breast cancer: ASCO-College of American Pathologists guideline update. J Clin Oncol 2023 41:3867-3872. PMID 37284804
- Modi S, Jacot W, Yamashita T, et al. Trastuzumab deruxtecan in previously treated HER2-low advanced breast cancer. N Engl J Med 2022 387:9-20. PMID 35665782