Part X · Metastatic disease: principles and cross-cutting management · Chapter 55

Expression and alteration-defined treatment across subtypes

What each expression threshold and each alteration actually buys, measured against the treatment it displaces.

1 · HER2-low and HER2-ultralow disease

Three categories now sit inside what the guideline still calls HER2-negative disease, and all three are assay outputs rather than entities.

HER2-low means immunohistochemistry 1+, or 2+ with negative in situ hybridisation. HER2-ultralow means immunohistochemistry 0 with faint incomplete membrane staining in a small proportion of cells. HER2-negative without qualification means everything below that. The categories were created by trial eligibility criteria rather than by a finding about biology¹. HER2-low

Prevalence follows from where the lines were drawn. Roughly a third to a half of all breast cancers score HER2-low, and among HER2-negative tumours the proportion is higher in hormone receptor-positive than in triple-negative disease².

The question underneath all of this is whether the amount of antigen selects anyone, or whether the antigen only has to be there. The story below states what each group of trials was built to find out, and it carries no figures at all.

The studies in that table agree on one thing. Activity does not stop at the bottom of the expression scale. In both randomised comparisons trastuzumab deruxtecan outperformed the physician's choice of chemotherapy in hormone receptor-positive disease. In the single-arm cohort study it produced responses in tumours the assay scored as non-expressing⁴.

Where they diverge, the antigen is not the reason. DESTINY-Breast04 enrolled after one or two lines of chemotherapy⁵. DESTINY-Breast06 enrolled patients who had never had chemotherapy for metastatic disease, after endocrine therapy³. The same drug was therefore tested at two positions in the disease, against whatever a physician would choose at each. Each result describes the position it was measured at rather than the score that governed entry. DAISY differs again, in design rather than in population alone. It is single-arm, its endpoint is response, and its cohorts are expression strata⁴. TRADE-DXd enters patients already progressing on a conjugate, which is a fourth population and a different question.

The decision rests on DESTINY-Breast04. It is the randomised comparison against the treatment the clinic would otherwise give. It is also the only study here with overall survival on the record^5,6. DESTINY-Breast06 changed who is eligible and when the drug is given. Its survival data were immature at the primary analysis. Giving a conjugate before any chemotherapy rests on a progression-free survival result alone.

What nobody did is randomise anyone on the score. The ultralow group in DESTINY-Breast06 was an exploratory cohort recruited under a rule rather than an arm. Its result tracked the HER2-low population it sat beside³. DAISY's response falls as staining falls, but each cohort received the same drug with no comparator inside it. The gradient therefore cannot say whether the advantage over chemotherapy narrows as expression falls. That is the missing column. A patient and her clinician want the gain relative to her own alternative, and no design here measures it by expression.

The same question put to a different antigen came back differently. TROPiCS-02 and TROPION-Breast01 imposed no expression requirement at entry. Their results are argued in Conjugates and chemotherapy in later lines. Where TROP2 was measured three ways in SACI-IO HR+, by immunohistochemistry, by immunofluorescence and in plasma, no measure tracked outcome. An antigen acquires a threshold when somebody draws one, not when the drug demands it.

Toxicity belongs beside efficacy here, because the table has no column for it. Adjudicated interstitial lung disease or pneumonitis occurred in 11.3% of patients given trastuzumab deruxtecan in DESTINY-Breast06, against one patient given chemotherapy³. That is the quantity to weigh against a progression-free survival gain in a patient who has not yet had chemotherapy.

2 · Antigen thresholds as eligibility criteria, the general lesson

The structure of this problem recurs across the field, and it is worth setting out in the abstract before returning to HER2.

A drug's activity varies continuously with a target that is continuously distributed across tumours. A trial needs a binary rule to decide who enters. So a cut-point is placed on the continuum. Everything downstream inherits it. The label, the reimbursement decision, the reported efficacy, the patient's access, and the number a pathologist has to produce all rest on that one line.

Three separate questions then get collapsed into one.

Does the drug work better at higher target expression? For trastuzumab deruxtecan, response falls as the score falls, and the cohort gradient in the table above is the measurement⁴. That is a gradient in response rate across single-arm cohorts rather than a gradient in benefit over chemotherapy.

Is there a level below which it does not work? Not established. Activity at immunohistochemistry 0 is well above zero. Within that group the amount of ERBB2 expression did not separate responders from non-responders⁴.

Can observers reliably place a case on one side of the line? No. When 18 breast pathologists from 15 institutions scored 170 breast cancer biopsies, agreement on cases scored immunohistochemistry 0 was 25%, with a Fleiss kappa of 0.49. Agreement across the full four-category system was 28.8%. On the question the assay was built to answer, 3+ against not 3+, agreement reached 87.1%⁷.

That third answer is the one to say out loud rather than to imply. A treatment decision affecting a large number of patients now rests on an assay boundary that trained specialist observers reproduce about a quarter of the time. This is the clearest case in oncology of a therapeutic eligibility line drawn where reproducibility is poorest. The evidence that the line marks a real difference in drug behaviour is weak.

If the assay cannot place the case, the obvious response is to measure the target some other way. That has been tried, and what came back is argued in Heterogeneous or changing HER2 expression.

The guideline is unusually candid about the same problem. The 2023 ASCO and College of American Pathologists update affirmed the existing recommendations and declined to create new result categories, calling it premature to do so¹. It acknowledged in the same document that the distinction between 0 and 1+ is now clinically relevant, because of the entry criteria of the trial that supported regulatory approval. The analytic validity framework is in Receptor assessment and the measurement problem and HER2-low and HER2-ultralow, entity or assay artifact. The same problem read as a heterogeneity problem is HER2-low and ultralow reframed as a heterogeneity problem.

The lesson generalises to every threshold of this kind. PD-L1 combined positive score at 10 governs first-line immunotherapy in triple-negative disease First-line chemoimmunotherapy and the PD-L1 threshold. An androgen receptor percentage governed eligibility in a trial that set its own threshold Androgen receptor and other subtype-directed targets. Ki67 governs risk categories Ki-67, proliferation indices, and interobserver reproducibility. In each case the same three questions apply, and in each case the third is the one nobody asks.

Trop-2 is the instructive counterexample, because somebody checked. In the biomarker analysis of ASCENT, sacituzumab govitecan outperformed chemotherapy across high, medium and low Trop-2 expression, so no threshold was imposed⁸. An antigen whose benefit does not require a cut-point does not acquire one.

3 · Germline and somatic DNA repair alterations

Germline BRCA1 and BRCA2 pathogenic variants identify a population in which an oral PARP inhibitor outperforms chemotherapy. The two registration trials entered HER2-negative advanced disease and selected on the germline mutation alone.

The question that followed was how far out from those two genes the benefit extends. The story below is what the trials were built to settle, and it carries no figures.

OlympiAD and EMBRACA are read together and they agree^12,11. In germline BRCA1 or BRCA2 carriers with HER2-negative disease, an oral PARP inhibitor delays progression and produces responses more often than the single-agent chemotherapy a physician would otherwise choose. Neither converted that into longer overall survival^13,14. The EMBRACA investigators repeated the survival analysis with adjustment for later PARP inhibitor and platinum exposure, and the estimate moved towards benefit without excluding no difference. Crossover is a candidate explanation that the trials cannot test, and the general form of that problem is in PARP inhibition for germline carriers.

BROCADE3 reads as the weaker result in the table and is a different experiment⁹. Its control arm is carboplatin with paclitaxel. It therefore asks what veliparib adds to a platinum doublet, not whether PARP inhibition beats the alternative to chemotherapy. The dose was one a full-dose doublet tolerates, and it was raised only when chemotherapy stopped. A combination design of that shape trades target engagement for haematological headroom. Reading its hazard ratio beside OlympiAD's sets a drug-against-chemotherapy result next to a drug-added-to-chemotherapy result. The drug is not the variable that differs.

TBCRC 048 went outside germline BRCA, and the useful part of its answer is which genes did not respond^10,15. Confirmed responses were confined to germline PALB2 and to somatic BRCA1 or BRCA2. Carriers of ATM or CHEK2 alterations alone had none. Talazoparib Beyond BRCA gave talazoparib to patients wild-type for BRCA1 and BRCA2 who carried a mutation in another repair gene. Its activity again concentrated in germline PALB2¹⁶. A homologous recombination gene list is not one biomarker. It is several, and they do not behave alike.

Two cautions travel with that gradient. The cohorts hold a few patients per gene, so a gene that looks inert is as uncertain as a gene that looks active. And the TBCRC 048 row carries a BRCA subtype tag although most of its patients had hormone receptor-positive, HER2-negative disease. The row reads as a germline BRCA result. That is a filing convention rather than a population.

What no reported trial here has done is randomise a PARP inhibitor against chemotherapy outside germline BRCA. Somatic BRCA and germline PALB2 are treated on response rates from single arms, against a comparator that exists only in the germline trials. EvoPAR-Breast01 does not close that gap, and it asks the question the clinic now has instead. It sets a PARP1-selective inhibitor with an oral SERD against a CDK4/6 inhibitor with endocrine therapy in the first line, in disease carrying a germline or somatic BRCA1, BRCA2 or PALB2 alteration. That is the comparison which decides whether a repair defect displaces the endocrine sequence or waits behind it. It has not reported.

Homologous recombination deficiency scores are a different kind of measurement and a common source of error. They record genomic scarring accumulated in the past. Scars persist after repair function returns, so a tumour can score high while repairing normally at the moment of testing. The score predicts response to platinum-containing neoadjuvant chemotherapy in triple-negative disease¹⁷, which is a validated use. It is not a measurement of current repair capacity.

Resistance follows the same logic in reverse. Secondary intragenic mutations restoring the BRCA2 reading frame confer resistance to platinum and to PARP inhibition alike¹⁸, because both exploit the same defect.

Germline testing carries a second purpose that somatic testing does not, which is the implication for relatives. That is a reason to order it early rather than when a treatment decision forces it Inherited susceptibility and familial breast cancer.

4 · Rare actionable mutations and fusions

Activating HER2 mutations occur in tumours without HER2 amplification and are genuine oncogenic drivers¹⁹. This is a different disease from HER2-positive breast cancer and it is routinely filed under the same heading, which produces the wrong drug. The agent that works is an irreversible pan-HER kinase inhibitor rather than an antibody. In a basket study, neratinib produced responses across HER2-mutant and HER3-mutant cancers, with the rate depending on the specific allele²⁰.

Every cohort in that table selected on the mutation. None required amplification and none required protein overexpression. All three paired a HER2 inhibitor with endocrine therapy in hormone receptor-positive disease, and all three produced activity. Where response was the endpoint, it occurred across kinase-domain and extracellular-domain mutations²¹, and in lobular as well as ductal disease²². Selecting on the mutation finds a treatable population the amplification assay cannot see.

The result columns are not three measurements of the same object. MutHER reports a clinical benefit rate, which counts durable stable disease. Its cohorts are strata of oestrogen receptor status and previous fulvestrant exposure rather than of mutation²³. SUMMIT and SGNTUC-019 report objective response. SGNTUC-019 is one cohort of a solid-tumour basket and gives its intervals at 90% confidence, which is not the convention the rest of this registry uses²¹. Two of the three used neratinib and one used tucatinib. The drug is the smallest of the differences between them, and ranking the three by result column ranks three different things.

SUMMIT carries the decision, and not because of its response rate²². It is the only study here with anything randomised inside it. Neither fulvestrant alone nor fulvestrant with trastuzumab produced a response. Patients crossing over to the triplet responded. That is what attributes the effect to the kinase inhibitor rather than to endocrine therapy or to antibody blockade of an unamplified receptor. The randomised substudy is small and allowed crossover, so it fixes the direction and settles nothing about magnitude.

Nobody randomised any of this against the chemotherapy these patients would otherwise be offered. That is the comparison the table cannot show and the one the choice actually turns on. Mutations are pooled for the reported result within each trial. MutHER saw clinical benefit associated with lobular histology and against L755 alterations, in an exploratory analysis far too small to act on. It also found acquired HER2 mutations at progression²³. Allele-level selection is a hypothesis here rather than a rule. HER2-negative status was also assigned locally in these studies, so the population is defined by whichever laboratory reported the case.

NTRK fusions are the tumour-agnostic example. Larotrectinib produced an overall response rate of 75% by independent review, with a 95% confidence interval of 61% to 85%, in 55 patients with 17 different TRK fusion-positive tumour types²⁴. Entrectinib produced an objective response in 31 of 54 patients, 57%, in an integrated analysis of three early-phase trials, with a median duration of response of 10 months²⁵. Repotrectinib, a next-generation inhibitor, gave responses in 59% of 51 inhibitor-naive patients and 48% of 69 previously treated with a TRK inhibitor²⁶. The relevance to breast cancer is limited by prevalence rather than by the size of the effect. These fusions are very rare outside secretory carcinoma, so the pre-test probability determines whether looking for them is sensible.

PIK3CA, AKT1 and PTEN alterations form the largest actionable group in this disease and they belong to Endocrine resistance, and the presentations the sequence does not fit, where the trials and the sequencing are set out^27,28. ESR1 mutations belong there too, and the selection that produces them is described in Endocrine therapy and the emergence of ESR1 alterations.

The organising principle is that actionable is not a binary property. A tiering framework ranks alterations by the strength of the evidence linking them to benefit in a specific tumour type. The breast-specific application of that framework separates alterations with randomised support from those resting on case reports²⁹.

The practical rule follows from prevalence rather than from biology. A rare alteration is worth looking for when a panel is already being run for another reason, or when histology raises the pre-test probability, as lobular histology does for HER2 mutation Invasive lobular disease in the metastatic setting.

5 · Tumor-agnostic biomarkers

A tumour-agnostic approval says that a biomarker predicts benefit across tissue of origin. It does not say that the biomarker is common in any particular disease, and in breast cancer that distinction does most of the work.

Mismatch repair deficiency was the first. Objective radiographic responses occurred in 53% of patients with advanced mismatch repair-deficient cancers across 12 tumour types treated with PD-1 blockade, and complete responses in 21%³⁰. The proposed mechanism is that defective mismatch repair generates a large burden of mutant neoantigens. The United States approval of pembrolizumab for these tumours followed in May 2017, on an overall response rate of 39.6% in 149 patients with 15 tumour types from five single-arm trials³¹. It was the first approval of a cancer treatment defined by a biomarker rather than by the site of origin. Dostarlimab has since given a response rate of 44.0% in 327 patients with mismatch repair-deficient tumours, most of them endometrial or colorectal³².

High tumour mutational burden followed. A prospective biomarker analysis within a multi-cohort phase 2 study set the threshold at 10 mutations per megabase³³. Objective responses to pembrolizumab occurred in 30 of 102 patients above it, which is 29%, against 43 of 688 patients below it, which is 6%.

That study is worth reading for what it did not include. The eligible tumour types were anal, biliary, cervical, endometrial, mesothelioma, neuroendocrine, salivary, small-cell lung, thyroid and vulvar³³. Breast cancer was not among them. The tumour-agnostic claim reaches breast cancer by inference rather than by measurement, and that should be said when the option is offered to a patient.

The inference has since been tested in breast cancer directly. In the TAPUR basket trial, a breast cohort with high tumour mutational burden treated with pembrolizumab had disease control in 37% and objective response in 21% of 28 patients³⁴. The story below is what those studies set out to do, and it carries no figures.

The three studies behind that question do not point the same way, and the disagreement is about what each one selected on rather than about the drugs.

NIMBUS is the study built on the burden argument³⁵. It gave nivolumab with low-dose ipilimumab to HER2-negative disease entered on a tumour mutational burden above a fixed threshold, with a prespecified correlative analysis at a higher one. Responses occurred. They were commoner above the higher cut-off. Two things stop that from selecting a patient in clinic. Most of its patients had hormone receptor-positive disease, but its comparison by receptor status was exploratory, so no result for this subtype reads out of it. And the threshold analysis rests on a handful of patients and was a correlative outcome rather than a randomised comparison. It nominates a cut-off. It does not validate one.

SACI-IO HR+ is the randomised comparison, and it did not select on burden at all³⁶. Adding pembrolizumab to a TROP2 conjugate in hormone receptor-positive, HER2-negative disease did not improve progression-free survival. Its PD-L1-defined subgroup gives a direction rather than a result. That analysis was secondary, confined to the patients with evaluable tissue, and not significant. This is the load-bearing study here, because it is the only one that set checkpoint blockade against the treatment it was being added to.

DART / SWOG S1609 cohort 36 cuts the other way³⁷. A rare-histology cohort of a basket study produced few responses, and those it produced were durable. They arose in tumours with low burden, low PD-L1 and absent tumour-infiltrating lymphocytes, which is the opposite of what the burden argument predicts. The report describes most of its tumours as triple-negative, so the cohort cannot be placed in this subtype at all. Selection was on histology, and histology appears to be doing the work.

Burden is also a property of the panel that counts it. A figure expressed as mutations per megabase depends on the size of the panel, on what it calls a somatic variant, and on how germline variants are filtered out. It travels between assays less well than it reads on a report. A patient is not above or below ten on a biological scale. She is above or below ten on one laboratory's.

NTRK fusions complete the set²⁴, with the prevalence caveat in Rare actionable mutations and fusions. RET fusions joined them in September 2022, when selpercatinib received accelerated approval in the United States for fusion-positive solid tumours³⁸. In 41 evaluable patients with tumours other than lung or thyroid cancer, the response rate was 43.9%³⁹. Breast cancer was among the tumour types in which a response occurred³⁸.

The breast-specific arithmetic governs whether to test. High tumour mutational burden occurs in a small minority of breast cancers and is concentrated in particular contexts rather than distributed evenly⁴⁰. Mismatch repair deficiency is rarer still. A test with a low positive rate is worth running as part of a panel ordered for other reasons, and is hard to justify as a standalone investigation.

One reading habit is worth acquiring here. When a tumour-agnostic result is quoted, look for the number of patients with the disease in front of you. The per-disease denominator inside a basket trial is frequently in single figures, and a response rate pooled across tissue types is not a response rate in any one of them.

6 · Discordant or changing biomarker results

A result that disagrees with a previous result has three possible explanations and they call for different responses. True biological change. Assay variation, including antibody clone, platform, fixation and scoring. Sampling, because two specimens are two different pieces of tissue Receptor discordance and conversion.

HER2-low is the most unstable of the categories now governing eligibility. Across 547 patients with matched primary and relapse specimens, overall HER2 discordance was 38.0%. Movement from HER2-0 to HER2-low occurred in 15% of cases and from HER2-low to HER2-0 in 14%⁴¹. Two movements of similar size in opposite directions are what measurement variation looks like. Some of it is certainly real change, and that design cannot separate the two components.

Hormone receptor and HER2 conversion between primary and distant metastasis is the better-characterised version of the same problem⁴².

Plasma and tissue disagree for reasons that are not error. A lesion that does not shed is invisible to plasma. A plasma result integrates across lesions and cannot say which one it came from. And variants arising in blood cells rather than in tumour appear in plasma unless matched white cell sequencing removes them⁴³. The blind spots on each side are set out in Liquid biopsy and longitudinal monitoring.

The rule that follows is decision-theoretic rather than biological, and saying so is better than pretending otherwise. When two adequate specimens disagree and the result governs eligibility, acting on the one that opens a treatment option is usually right. The justification is asymmetric consequences, not a claim about which measurement was correct.

7 · Competing actionable targets and how to prioritize

The common situation is not an absence of targets. It is several at once. One report can carry a PIK3CA mutation, HER2-low expression, a germline BRCA variant and a fusion of unknown significance on the same page.

Five criteria order them, and applying them in sequence is better than weighting them informally.

Evidence level for that alteration in this disease. A tiering framework exists for exactly this purpose and it separates randomised support from anecdote²⁹.

Magnitude and durability of the expected benefit, with the comparator named. A hazard ratio against the physician's choice of chemotherapy is not measured on the same scale as a response rate in a single-arm cohort of germline PALB2 carriers. The tables under HER2-low and HER2-ultralow disease and Germline and somatic DNA repair alterations carry both quantities, and ranking one against the other requires saying which each is.

Whether the option is line-restricted. An agent available only before chemotherapy is lost by giving chemotherapy first, which is a reason to sequence it earlier rather than a reason to prefer it.

Toxicity against the patient's current function. Interstitial lung disease on a deruxtecan conjugate is the example that changes most decisions, and its frequency is argued in HER2-low and HER2-ultralow disease. That is a different proposition in a patient with baseline lung disease than in one without.

Whether the alteration is truncal or subclonal. A target present in a fraction of cells leaves a compartment untouched, which is the argument in HER2 heterogeneity and Temporal heterogeneity and clonal evolution. Variant allele fraction in plasma gives a crude indication and tissue gives a better one.

A molecular tumour board exists to do this ordering, and the reporting standards that make it possible are in Reporting standards and molecular tumor boards.

References

All 43 references

1. 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
2. Tarantino P, Hamilton E, Tolaney SM, et al. HER2-low breast cancer: pathological and clinical landscape. J Clin Oncol 2020 38:1951-1962. PMID 32330069
3. Bardia A, Hu X, Dent R, et al. Trastuzumab deruxtecan after endocrine therapy in metastatic breast cancer. N Engl J Med 2024 391:2110-2122. PMID 39282896
4. 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
5. 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
6. Modi S, et al. Trastuzumab deruxtecan in HER2-low metastatic breast cancer: long-term survival analysis of the randomized, phase 3 DESTINY-Breast04 trial. Nat Med 2025 31:4205-4213. PMID 41062831
7. Robbins CJ, Fernandez AI, Han G, et al. Multi-institutional assessment of pathologist scoring HER2 immunohistochemistry. Mod Pathol 2023 36:100032. PMID 36788069
8. Bardia A, Tolaney SM, Punie K, et al. Biomarker analyses in the phase III ASCENT study of sacituzumab govitecan versus chemotherapy in patients with metastatic triple-negative breast cancer. Ann Oncol 2021 32:1148-1156. PMID 34116144
9. Diéras V, et al. Veliparib with carboplatin and paclitaxel in BRCA-mutated advanced breast cancer (BROCADE3): a randomised, double-blind, placebo-controlled, phase 3 trial. Lancet Oncol 2020 21:1269-1282. PMID 32861273
10. Tung NM, Robson ME, Ventz S, et al. TBCRC 048: phase II study of olaparib for metastatic breast cancer and mutations in homologous recombination-related genes. J Clin Oncol 2020 38:4274-4282. PMID 33119476
11. Litton JK, Rugo HS, Ettl J, et al. Talazoparib in patients with advanced breast cancer and a germline BRCA mutation. N Engl J Med 2018 379:753-763. PMID 30110579
12. Robson M, Im SA, Senkus E, et al. Olaparib for metastatic breast cancer in patients with a germline BRCA mutation. N Engl J Med 2017 377:523-533. PMID 28578601
13. Robson ME, et al. OlympiAD final overall survival and tolerability results: Olaparib versus chemotherapy treatment of physician's choice in patients with a germline BRCA mutation and HER2-negative metastatic breast cancer. Ann Oncol 2019 30:558-566. PMID 30689707
14. Litton JK, et al. Talazoparib versus chemotherapy in patients with germline BRCA1/2-mutated HER2-negative advanced breast cancer: final overall survival results from the EMBRACA trial. Ann Oncol 2020 31:1526-1535. PMID 32828825
15. Tung NM, et al. TBCRC 048 (Olaparib Expanded) Expansion Cohorts: Phase II Study of Olaparib Monotherapy for Patients With Metastatic Breast Cancer With Germline Mutations inor Somatic Mutations inor. J Clin Oncol 2026 44:653-661. PMID 41604601
16. Gruber JJ, Afghahi A, Timms K, et al. A phase II study of talazoparib monotherapy in patients with wild-type BRCA1 and BRCA2 with a mutation in other homologous recombination genes. Nat Cancer 2022 3:1181-1191. PMID 36253484
17. Telli ML, Timms KM, Reid J, et al. Homologous recombination deficiency (HRD) score predicts response to platinum-containing neoadjuvant chemotherapy in patients with triple-negative breast cancer. Clin Cancer Res 2016 22:3764-3773. PMID 26957554
18. 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
19. Bose R, Kavuri SM, Searleman AC, et al. Activating HER2 mutations in HER2 gene amplification negative breast cancer. Cancer Discov 2013 3:224-237. PMID 23220880
20. Hyman DM, Piha-Paul SA, Won H, et al. HER kinase inhibition in patients with HER2- and HER3-mutant cancers. Nature 2018 554:189-194. PMID 29420467
21. Okines AFC, Curigliano G, Mizuno N, et al. Tucatinib and trastuzumab in HER2-mutated metastatic breast cancer: a phase 2 basket trial. Nat Med 2025 31:909-916. PMID 39825152
22. Jhaveri K, Eli LD, Wildiers H, et al. Neratinib + fulvestrant + trastuzumab for HR-positive, HER2-negative, HER2-mutant metastatic breast cancer: outcomes and biomarker analysis from the SUMMIT trial. Ann Oncol 2023 34:885-898. PMID 37597578
23. Ma CX, Luo J, Freedman RA, et al. The Phase II MutHER Study of Neratinib Alone and in Combination with Fulvestrant in HER2-Mutated, Non-amplified Metastatic Breast Cancer. Clin Cancer Res 2022 28:1258-1267. PMID 35046057
24. Drilon A, Laetsch TW, Kummar S, et al. Efficacy of larotrectinib in TRK fusion-positive cancers in adults and children. N Engl J Med 2018 378:731-739. PMID 29466156
25. Doebele RC, Drilon A, Paz-Ares L, et al. Entrectinib in patients with advanced or metastatic NTRK fusion-positive solid tumours: integrated analysis of three phase 1-2 trials. Lancet Oncol 2020 21:271-282. PMID 31838007
26. Besse B, Lin JJ, Bazhenova L, et al. Repotrectinib in NTRK fusion-positive advanced solid tumors: a phase 1/2 trial. Nat Med 2026 32:682-689. PMID 41639379
27. Andre F, Ciruelos E, Rubovszky G, et al. Alpelisib for PIK3CA-mutated, hormone receptor-positive advanced breast cancer. N Engl J Med 2019 380:1929-1940. PMID 31091374
28. Turner NC, Oliveira M, Howell SJ, et al. Capivasertib in hormone receptor-positive advanced breast cancer. N Engl J Med 2023 388:2058-2070. PMID 37256976
29. Condorelli R, Mosele F, Verret B, et al. Genomic alterations in breast cancer: level of evidence for actionability according to ESMO Scale for Clinical Actionability of molecular Targets. Ann Oncol 2019 30:365-373. PMID 30715161
30. Le DT, Durham JN, Smith KN, et al. Mismatch repair deficiency predicts response of solid tumors to PD-1 blockade. Science 2017 357:409-413. PMID 28596308
31. Marcus L, Lemery SJ, Keegan P, Pazdur R. FDA Approval Summary: Pembrolizumab for the Treatment of Microsatellite Instability-High Solid Tumors. Clin Cancer Res 2019 25:3753-3758. PMID 30787022
32. André T, Berton D, Curigliano G, et al. Antitumor Activity and Safety of Dostarlimab Monotherapy in Patients With Mismatch Repair Deficient Solid Tumors: A Nonrandomized Controlled Trial. JAMA Netw Open 2023 6:e2341165. PMID 37917058
33. Marabelle A, Fakih M, Lopez J, et al. Association of tumour mutational burden with outcomes in patients with advanced solid tumours treated with pembrolizumab: prospective biomarker analysis of the multicohort, open-label, phase 2 KEYNOTE-158 study. Lancet Oncol 2020 21:1353-1365. PMID 32919526
34. Alva AS, Mangat PK, Garrett-Mayer E, et al. Pembrolizumab in Patients With Metastatic Breast Cancer With High Tumor Mutational Burden: Results From the Targeted Agent and Profiling Utilization Registry (TAPUR) Study. J Clin Oncol 2021 39:2443-2451. PMID 33844595
35. Barroso-Sousa R, Zanudo JGT, Li T, et al. Nivolumab plus low-dose ipilimumab in hypermutated HER2-negative metastatic breast cancer: a phase II trial (NIMBUS). Nat Commun 2025 16:4430. PMID 40360544
36. Garrido-Castro AC, Kim SE, Li T, et al. SACI-IO HR+: a randomized phase II trial of sacituzumab govitecan with or without pembrolizumab in patients with metastatic hormone receptor-positive/HER2-negative breast cancer. Ann Oncol 2026. PMID 42323161
37. Adams S, Othus M, Patel SP, et al. A Multicenter Phase II Trial of Ipilimumab and Nivolumab in Unresectable or Metastatic Metaplastic Breast Cancer: Cohort 36 of Dual Anti-CTLA-4 and Anti-PD-1 Blockade in Rare Tumors (DART, SWOG S1609). Clin Cancer Res 2021 28:271-278. PMID 34716198
38. Duke ES, Bradford D, Marcovitz M, et al. FDA Approval Summary: Selpercatinib for the Treatment of Advanced RET Fusion-Positive Solid Tumors. Clin Cancer Res 2023 29:3573-3578. PMID 37265412
39. Subbiah V, Wolf J, Konda B, et al. Tumour-agnostic efficacy and safety of selpercatinib in patients with RET fusion-positive solid tumours other than lung or thyroid tumours (LIBRETTO-001): a phase 1/2, open-label, basket trial. Lancet Oncol 2022 23:1261-1273. PMID 36108661
40. Barroso-Sousa R, Jain E, Cohen O, et al. Prevalence and mutational determinants of high tumor mutation burden in breast cancer. Ann Oncol 2020 31:387-394. PMID 32067680
41. Miglietta F, Griguolo G, Bottosso M, et al. Evolution of HER2-low expression from primary to recurrent breast cancer. NPJ Breast Cancer 2021 7:137. PMID 34642348
42. Schrijver WAME, Suijkerbuijk KPM, van Gils CH, et al. Receptor conversion in distant breast cancer metastases: a systematic review and meta-analysis. J Natl Cancer Inst 2018 110:568-580. PMID 29315431
43. 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