Part III · Pathology and classification · Chapter 12

Triple-negative and other biologically diverse phenotypes

No subtyping scheme for this disease is used in routine care, and the reasons are worth knowing.

1 · Molecular diversity within triple-negative disease

Triple-negative disease is defined by three absences measured by two assays. It is not a positive biological claim about anything. A tumour enters the category by failing to reach the oestrogen receptor threshold, the progesterone receptor threshold and the HER2 threshold.

That has a consequence people underuse. The membership of the category depends on where the thresholds sit. The current oestrogen receptor boundary is 1% of tumour nuclei1. At a boundary of 10% the category would be larger, and the tumours added would resemble the ones already in it. Among cancers staining 1% to 9%, 48% are basal-like by PAM50 and only 24% are positive by ESR1 messenger RNA2. Adjuvant endocrine therapy showed no association with time to recurrence in that band3. The measurement argument is in Receptor assessment and the measurement problem.

What sits inside the category is genuinely diverse. Whole-genome sequencing of 560 breast cancers identified 93 protein-coding genes carrying probable driver mutations, twelve base substitution signatures and six rearrangement signatures4. Three of the rearrangement signatures appear associated with defective homologous-recombination-based repair, one of them with deficient BRCA1 function specifically. Those processes are not distributed evenly across triple-negative tumours, and the differences between them are what the subtyping schemes are trying to capture.

The recurring axes are basal-like differentiation, homologous recombination deficiency, androgen receptor signalling, immune infiltration and mesenchymal or metaplastic states. Several schemes have been built on those axes. None of them is used to select treatment in routine care, and Why TNBC classifiers disagree with one another sets out why.

2 · Lehmann subtypes and their subsequent refinements

The original analysis clustered gene expression across 21 public data sets and produced six subtypes, namely basal-like 1, basal-like 2, immunomodulatory, mesenchymal, mesenchymal stem-like and luminal androgen receptor5. Each was matched to preclinical models so that candidate targeted agents could be tested. The scheme was built to nominate drugs, which is worth remembering when it is used for prognosis.

Retrospective response data followed. Among 130 evaluable patients given standard neoadjuvant chemotherapy, pathological complete response reached 52% in basal-like 1, 0% in basal-like 2 and 10% in luminal androgen receptor6. Subtype was an independent predictor of response in that series. The per-subtype numbers are small and the confidence intervals are wide.

The refinement is the most instructive part of the story. Histopathological quantification and laser-capture microdissection showed that the transcripts defining the immunomodulatory subtype came from infiltrating lymphocytes, and that those defining the mesenchymal stem-like subtype came from tumour-associated stroma7. Both were dropped. Four tumour-intrinsic classes remained, namely basal-like 1, basal-like 2, mesenchymal and luminal androgen receptor. In over 300 retrospectively subtyped patients, pathological complete response was 41% in basal-like 1, 18% in basal-like 2 and 29% in luminal androgen receptor.

Two of six original subtypes were properties of what else was in the sample. That is not a flaw in the original work. It is what bulk expression measurement does, and it is the same composition problem as in Single-cell and spatial data as a challenge to bulk-defined subtypes.

Other groups reached different partitions of the same disease. Analysis of 198 triple-negative tumours produced four subtypes, namely luminal androgen receptor, mesenchymal, basal-like immunosuppressed and basal-like immune-activated, with the immunosuppressed group carrying the worst survival and the immune-activated group the best8. A cohort of 465 primary Chinese triple-negative tumours produced four again, namely luminal androgen receptor, immunomodulatory, basal-like immune-suppressed and mesenchymal-like9.

3 · Basal-like and non-basal-like biology

Basal-like is the one intrinsic class that survives a change of classifier. Across three published single-sample predictors applied to four cohorts, basal-like was the only subtype with almost perfect agreement, above a kappa of 0.81210. The proportion of tumours called basal-like was stable whichever predictor was used. Nothing else in the intrinsic scheme behaves that well.

Basal-like and triple-negative overlap without coinciding, and the two words are measured by different things. Basal-like is an expression class assigned by a classifier. Triple-negative is a pair of immunohistochemical results below threshold. Most basal-like tumours are triple negative and most triple-negative tumours are basal-like, and neither statement is an identity.

Claudin-low illustrates how a class can dissolve under better data. It was originally described as an additional intrinsic subtype, characterised by low expression of luminal differentiation markers, enrichment for epithelial-to-mesenchymal transition genes, immune response genes and stem-cell-like features11. A later reanalysis argued that claudin-low is not analogous to the intrinsic subtypes at all12. It behaves instead as a phenotype that can permeate tumours of several intrinsic subtypes, distinguished by low genomic instability, low mutational burden, low proliferation and high immune and stromal infiltration.

That reanalysis is worth reading as a template. A class defined by low expression of a marker set is vulnerable to being a statement about what else is in the sample.

4 · DNA repair deficiency and genomic scarring

Three things get called the same name here, and separating them removes most of the confusion.

The cause is loss of a repair gene. That may be a germline or somatic BRCA1 or BRCA2 alteration, or silencing of BRCA1 by promoter hypermethylation, which produces a molecular profile closely resembling the mutated case13.

The phenotype is homologous recombination deficiency, which is a functional state of the cell at a given moment.

The measurement is a scar score, a mutational signature or a functional assay, and each measures the phenotype indirectly. The HRD score combines loss of heterozygosity, telomeric allelic imbalance and large-scale state transitions, and it predicted response to platinum-containing neoadjuvant chemotherapy in triple-negative disease14. HRDetect uses whole-genome mutational signatures to predict BRCA1 and BRCA2 deficiency15. Rearrangement signatures from whole-genome data track the same deficiency through a different feature set4.

The critical property of a scar is temporal. A scar records that repair was deficient while the genome was being damaged. It does not establish that repair is deficient now. A reversion mutation restores function and leaves every existing scar in place. So scar-based assays have better negative than positive predictive value for current sensitivity to a PARP inhibitor.

The clinical anchors remain narrower than the biology. Olaparib and talazoparib have established benefit in germline BRCA1 or BRCA2 carriers with metastatic disease16,17, and adjuvant olaparib in the high-risk germline setting18. TBCRC 048 extended activity to germline PALB2 and somatic BRCA alterations19. No scar score currently carries a licensed indication in breast cancer. The repair biology is in DNA repair, replication stress, and genomic instability and the assays in Prognostic and predictive assays and somatic profiling.

5 · Androgen receptor driven and luminal androgen receptor disease

Luminal androgen receptor is the most reproducible triple-negative subtype. It appears under essentially the same name in three independently derived schemes5,8,9. That reproducibility is a clue about its nature, because it is driven by a transcriptional programme that is either running or not.

Its genomics differ from the rest of the category. Among 550 triple-negative tumours classified from METABRIC and the Cancer Genome Atlas, the luminal androgen receptor group carried PIK3CA mutations in 55%, AKT1 in 13% and CDH1 in 13%20. In that multivariable analysis it was the subtype associated with the worst prognosis, and the immunomodulatory subtype with the best. In the Chinese cohort, the same subtype showed more ERBB2 somatic mutations, infrequent mutational signature 3 and frequent CDKN2A loss9.

Androgen blockade has been tested twice, and the results are modest. Bicalutamide was given to androgen receptor positive, ER and PR negative metastatic patients21. Of 424 screened, 12% were androgen receptor positive at a threshold of more than 10% nuclear staining. The six-month clinical benefit rate was 19%, with a 95% confidence interval of 7% to 39%. Enzalutamide enrolled 118 patients with androgen receptor positive triple-negative disease, using a threshold of any nuclear staining above 0%22. Clinical benefit rate at 16 weeks was 25% in the intention-to-treat population. It was 33% in the evaluable subgroup of 78 patients, defined by at least 10% nuclear androgen receptor staining plus a postbaseline assessment. Median progression-free survival was 2.9 months and 3.3 months respectively.

The measurement problem is visible on the surface of those two trials. They used different androgen receptor thresholds, so they did not enrol the same population. No androgen-directed therapy is approved in this setting.

6 · Immune-enriched and immune-excluded phenotypes

Tumour-infiltrating lymphocytes are the one immune measurement with an agreed scoring method and a large evidence base. The International TILs Working Group set out how to score stromal lymphocytes reproducibly23, and later documented the pitfalls that remain24. Incidence and magnitude differ systematically by subtype25. In a pooled analysis of 3771 patients treated with neoadjuvant therapy, higher lymphocytic infiltration was associated with response and with outcome, and the association differed by subtype26. A pooled individual patient analysis in early triple-negative disease confirmed prognostic value27, and a separate series showed it in patients who received no adjuvant chemotherapy at all28.

Density and arrangement are different measurements, and this is where the subtyping schemes diverge from each other. Spatially distinct immune microenvironments stratify triple-negative tumours into groups that a single percentage cannot separate29. Spatial immunophenotypes predict response to PD-1 blockade and capture distinct routes of T cell evasion30. Multiplexed ion beam imaging showed that triple-negative tumours are either compartmentalised or mixed, with immune and tumour cells segregated or interleaved, and that the distinction carries prognostic weight31.

Two tumours can report the same lymphocyte percentage and have opposite architecture. An average density discards the arrangement, in exactly the way that a HER2 percentage discards it in HER2 heterogeneity.

The clinical test of the immune classifier that exists is instructive. In KEYNOTE-522, adding pembrolizumab to neoadjuvant chemotherapy improved pathological complete response in early triple-negative disease regardless of PD-L1 status32. The biomarker did not select the population that benefited. The immunology is developed in Breast cancer immunology and the spatial argument in Spatial organization and what it adds.

7 · Mesenchymal, metaplastic, and rare biological states

Metaplastic carcinoma is a rare special histological type within triple-negative disease, defined by differentiation towards squamous epithelium or mesenchymal elements. Copy number profiling of 17 cases found only limited differences between the histological subgroups33. Expression profiling separated them more clearly, with spindle cell tumours differing from chondroid and squamous tumours in epithelial-to-mesenchymal transition genes. The histological patterns were judged unlikely to be underpinned by a recurrent expressed fusion gene or a pathognomonic expressed mutation. PTEN loss or mutations affecting PIK3CA or PIK3R1 were present in 8 of 17 cases.

So the histological subgroups of metaplastic carcinoma are transcriptionally distinct and are not defined by a specific genetic lesion. That is a different situation from most special types.

The mesenchymal signal in bulk data has a second source that is not the tumour. Part of it came from tumour-associated stroma, which is why the mesenchymal stem-like subtype was withdrawn7. A mesenchymal call on a bulk sample may be a statement about the specimen.

Epithelial-to-mesenchymal transition itself is not binary. Multiple intermediate states exist and they differ in tumour-initiating capacity and metastatic behaviour34. Differentiation-state plasticity in basal-like disease behaves as a resistance mechanism rather than as a fixed subtype35. A transcriptional state that can change under pressure will not stay put long enough to be a classification.

Not every triple-negative tumour behaves badly. Several rare histological types, including adenoid cystic and secretory carcinoma, are usually triple negative by immunohistochemistry and follow an indolent course36. The immunophenotype is not the prognosis, and morphology outperforms it in these cases. The histology is developed in Histopathology and the spectrum of breast neoplasia.

8 · Why TNBC classifiers disagree with one another

No triple-negative subtyping system is used to select treatment in routine care anywhere. The reasons divide into four, and they are not equally fixable.

The first is input material. Bulk expression measures a mixture of tumour, immune and stromal cells. Two of the six original subtypes were shown to come from lymphocytes and stroma rather than from the tumour7. Any scheme built on bulk data inherits that problem.

The second is cohort. The Chinese cohort found PIK3CA mutations and copy number gains at 22q11 more frequent than in the Cancer Genome Atlas9. A classifier trained on one population does not automatically transfer to another.

The third is algorithm and class number. The schemes propose six, four and four classes on overlapping biology. Nearest-centroid assignment forces every sample into a class, and the instability that causes is measurable even for the intrinsic subtypes, where only basal-like agreed across predictors10.

The fourth is purpose. The Lehmann scheme was built to nominate preclinical models and candidate targets5. The Burstein scheme was built around prognosis8. They are not competing answers to one question.

Is the disagreement biological or methodological? Mostly methodological, with one real biological component underneath it. The underlying variation is continuous, and every scheme imposes discrete classes on a continuum. Luminal androgen receptor is the most reproducible class precisely because it is the most nearly discrete, being driven by a transcriptional programme that is switched on or off. Immune and mesenchymal classes are the least reproducible because they sit on gradients, and because they are partly contributed by cells that are not the tumour.

The deciding argument is the one in Analytical validity, clinical validity, and clinical utility defined. These schemes have shown clinical validity in retrospective series with small per-subtype numbers6,7. None has been tested for clinical utility, which would require randomising a treatment decision on the subtype call. Until that is done, a subtype is a research finding with a prognosis attached.

Caution

The natural reading of four competing classifiers is that one of them is right. That reading is almost certainly wrong. They are four discretisations of a continuum, built for different purposes on different material. Asking which is correct is less useful than asking which axis a particular patient's tumour actually sits on, and whether anything can be done about it.

Interplay

Triple-negative disease is where the heterogeneity argument and the classification argument collide hardest. The category is defined by measurement, its internal structure is continuous, and its most reproducible divisions are the ones contributed by non-tumour cells. A classifier that assigns one label per tumour is answering a question this disease does not pose in that form.

See Integrative biological interplay

References

All 36 references
  1. Allison KH, Hammond MEH, Dowsett M, et al. Estrogen and progesterone receptor testing in breast cancer: ASCO/CAP guideline update. J Clin Oncol 2020 38:1346-1366. PMID 31928404
  2. Iwamoto T, Booser D, Valero V, et al. Estrogen receptor mRNA and ER-related gene expression in breast cancers that are 1% to 10% ER-positive by immunohistochemistry. J Clin Oncol 2012 30:729-734. PMID 22291085
  3. Fujii T, Kogawa T, Dong W, et al. Revisiting the definition of estrogen receptor positivity in HER2-negative primary breast cancer. Ann Oncol 2017 28:2420-2428. PMID 28961844
  4. Nik-Zainal S, Davies H, Staaf J, et al. Landscape of somatic mutations in 560 breast cancer whole-genome sequences. Nature 2016 534:47-54. PMID 27135926
  5. Lehmann BD, Bauer JA, Chen X, et al. Identification of human triple-negative breast cancer subtypes and preclinical models for selection of targeted therapies. J Clin Invest 2011 121:2750-2767. PMID 21633166
  6. Masuda H, Baggerly KA, Wang Y, et al. Differential response to neoadjuvant chemotherapy among 7 triple-negative breast cancer molecular subtypes. Clin Cancer Res 2013 19:5533-5540. PMID 23948975
  7. Lehmann BD, Jovanovic B, Chen X, et al. Refinement of triple-negative breast cancer molecular subtypes: implications for neoadjuvant chemotherapy selection. PLoS One 2016 11:e0157368. PMID 27310713
  8. Burstein MD, Tsimelzon A, Poage GM, et al. Comprehensive genomic analysis identifies novel subtypes and targets of triple-negative breast cancer. Clin Cancer Res 2015 21:1688-1698. PMID 25208879
  9. Jiang YZ, Ma D, Suo C, et al. Genomic and transcriptomic landscape of triple-negative breast cancers: subtypes and treatment strategies. Cancer Cell 2019 35:428-440. PMID 30853353
  10. Weigelt B, Mackay A, A'hern R, et al. Breast cancer molecular profiling with single sample predictors: a retrospective analysis. Lancet Oncol 2010 11:339-349. PMID 20181526
  11. Prat A, Parker JS, Karginova O, et al. Phenotypic and molecular characterization of the claudin-low intrinsic subtype of breast cancer. Breast Cancer Res 2010 12:R68. PMID 20813035
  12. Fougner C, Bergholtz H, Norum JH, Sorlie T. Re-definition of claudin-low as a breast cancer phenotype. Nat Commun 2020 11:1787. PMID 32286297
  13. Glodzik D, Bosch A, Hartman J, et al. Comprehensive molecular comparison of BRCA1 hypermethylated and BRCA1 mutated triple negative breast cancers. Nat Commun 2020 11:3747. PMID 32719340
  14. 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
  15. Davies H, Glodzik D, Morganella S, et al. HRDetect is a predictor of BRCA1 and BRCA2 deficiency based on mutational signatures. Nat Med 2017 23:517-525. PMID 28288110
  16. 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
  17. 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
  18. Tutt ANJ, Garber JE, Kaufman B, et al. Adjuvant olaparib for patients with BRCA1- or BRCA2-mutated breast cancer. N Engl J Med 2021 384:2394-2405. PMID 34081848
  19. 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
  20. Bareche Y, Venet D, Ignatiadis M, et al. Unravelling triple-negative breast cancer molecular heterogeneity using an integrative multiomic analysis. Ann Oncol 2018 29:895-902. PMID 29365031
  21. Gucalp A, Tolaney S, Isakoff SJ, et al. Phase II trial of bicalutamide in patients with androgen receptor-positive, estrogen receptor-negative metastatic breast cancer. Clin Cancer Res 2013 19:5505-5512. PMID 23965901
  22. Traina TA, Miller K, Yardley DA, et al. Enzalutamide for the treatment of androgen receptor-expressing triple-negative breast cancer. J Clin Oncol 2018 36:884-890. PMID 29373071
  23. Salgado R, Denkert C, Demaria S, et al. The evaluation of tumor-infiltrating lymphocytes (TILs) in breast cancer: recommendations by an International TILs Working Group 2014. Ann Oncol 2015 26:259-271. PMID 25214542
  24. Kos Z, Roblin E, Kim RS, et al. Pitfalls in assessing stromal tumor infiltrating lymphocytes (sTILs) in breast cancer. NPJ Breast Cancer 2020 6:17. PMID 32411819
  25. Stanton SE, Adams S, Disis ML. Variation in the incidence and magnitude of tumor-infiltrating lymphocytes in breast cancer subtypes: a systematic review. JAMA Oncol 2016 2:1354-1360. PMID 27355489
  26. Denkert C, von Minckwitz G, Darb-Esfahani S, et al. Tumour-infiltrating lymphocytes and prognosis in different subtypes of breast cancer: a pooled analysis of 3771 patients treated with neoadjuvant therapy. Lancet Oncol 2018 19:40-50. PMID 29233559
  27. Loi S, Drubay D, Adams S, et al. Tumor-infiltrating lymphocytes and prognosis: a pooled individual patient analysis of early-stage triple-negative breast cancers. J Clin Oncol 2019 37:559-569. PMID 30650045
  28. Park JH, Jonas SF, Bataillon G, et al. Prognostic value of tumor-infiltrating lymphocytes in patients with early-stage triple-negative breast cancers (TNBC) who did not receive adjuvant chemotherapy. Ann Oncol 2019 30:1941-1949. PMID 31566659
  29. Gruosso T, Gigoux M, Manem VSK, et al. Spatially distinct tumor immune microenvironments stratify triple-negative breast cancers. J Clin Invest 2019 129:1785-1800. PMID 30753167
  30. Hammerl D, Martens JWM, Timmermans M, et al. Spatial immunophenotypes predict response to anti-PD1 treatment and capture distinct paths of T cell evasion in triple negative breast cancer. Nat Commun 2021 12:5668. PMID 34580291
  31. Keren L, Bosse M, Marquez D, et al. A structured tumor-immune microenvironment in triple negative breast cancer revealed by multiplexed ion beam imaging. Cell 2018 174:1373-1387. PMID 30193111
  32. Schmid P, Cortes J, Pusztai L, et al. Pembrolizumab for early triple-negative breast cancer. N Engl J Med 2020 382:810-821. PMID 32101663
  33. Piscuoglio S, Ng CKY, Geyer FC, et al. Genomic and transcriptomic heterogeneity in metaplastic carcinomas of the breast. NPJ Breast Cancer 2017 3:48. PMID 29214215
  34. Pastushenko I, Brisebarre A, Sifrim A, et al. Identification of the tumour transition states occurring during EMT. Nature 2018 556:463-468. PMID 29670281
  35. Risom T, Langer EM, Chapman MP, et al. Differentiation-state plasticity is a targetable resistance mechanism in basal-like breast cancer. Nat Commun 2018 9:3815. PMID 30232459
  36. Loibl S, Poortmans P, Morrow M, Denkert C, Curigliano G. Breast cancer. Lancet 2021 397:1750-1769. PMID 33812473