Synthesis: Intrinsic subtypes and PAM50 history

Evidence grades (GRADE-adapted): A high — multiple well-conducted RCTs or systematic reviews converge. B moderate — single pivotal RCT or consistent observational evidence. C limited — single observational study, mechanistic, or expert consensus. D preclinical / hypothesis-generating.

Why intrinsic subtypes

Before 2000, breast cancer was classified primarily by histology (ductal vs lobular), grade, and ER/PR status. These categories were and remain useful, but they obscured significant biological heterogeneity within each: two ER− ductal grade-3 tumors could behave very differently and could have very different responses to the same chemotherapy. The genomics revolution of the late 1990s made it possible to ask whether unsupervised clustering on whole-transcriptome data would reveal categories that mapped more cleanly to biology. The answer was yes — and the resulting taxonomy has defined how breast cancer is discussed for two decades.

Perou 2000: the original discovery

Perou and colleagues, working at Stanford under David Botstein, applied hierarchical clustering to cDNA microarray expression data from 65 breast tumors and identified five intrinsic subtypes plus a normal-tissue cluster^[1]A. The "intrinsic" terminology reflected the methodology: genes were selected based on consistent expression within paired samples from the same tumor (i.e., intrinsic to the tumor) and variable across different tumors. The five subtypes:

• Luminal A. ER-positive expression pattern reminiscent of luminal cells of the normal mammary epithelium. Lower proliferation; best prognosis. Most-common subtype overall.
• Luminal B. ER-positive but with higher proliferation and worse prognosis than luminal A. The luminal A vs luminal B distinction reflects clinically meaningful biology.
• HER2-enriched. High expression of ERBB2 (HER2) and adjacent 17q12 amplicon genes. Overlaps but is not identical to HER2-amplified disease by IHC/ISH.
• Basal-like. Expression pattern reminiscent of basal/myoepithelial cells of normal mammary epithelium: high cytokeratin 5/6, 14, 17; EGFR; low luminal cytokeratins. Predominantly ER−/PR−/HER2−.
• Normal-like. Expression pattern resembling normal breast tissue; possibly an artefact of high stromal content or sampling bias. Clinically less well-characterized.

Sørlie and colleagues, working at Norwegian Radium Hospital, validated the prognostic significance of these subtypes in a separate cohort of 49 tumors in 2001 and refined the basal-like description^[2]A. Sørlie 2003 extended this to 115 tumors with 51 paired observations, confirming subtype stability across primary-vs-recurrence pairs^[3]A. By 2005 the five-subtype framework was the de facto language of molecular breast oncology.

PAM50: making the framework clinical

The original Perou/Sørlie classifiers used hundreds-to-thousands of genes from cDNA microarray data — impractical for routine clinical samples, especially FFPE tissue. Parker and colleagues at the University of North Carolina addressed this in 2009 with PAM50: a 50-gene signature that classified samples into the five intrinsic subtypes using qRT-PCR or NanoString platforms compatible with FFPE input^[4]A. The reduction from hundreds of genes to 50 came from feature-selection using prediction-strength scoring; the 50 genes retain the discriminative power of the larger signature with minimal loss of accuracy in cross-cohort validation.

PAM50 was commercialized as the Prosigna Breast Cancer Prognostic Gene Signature Assay (NanoString Technologies, now Veracyte), and was FDA-cleared in 2013 for risk-of-recurrence prediction in postmenopausal node-negative HR+ early-stage breast cancer. The assay also reports intrinsic subtype as part of its output. EMA approval followed in 2014. Prosigna and the related Oncotype DX and MammaPrint assays are the three principal molecular tests used in HR+ breast cancer to guide adjuvant chemotherapy decisions.

The TNBC ↔ basal-like relationship

A persistent source of confusion in the literature: TNBC and basal-like are not synonyms but they overlap substantially. The relationship has been characterized in multiple cohorts^[5]A:

• Approximately 75–85% of TNBC tumors are basal-like by PAM50. The exact percentage varies by cohort (Western series often report ~80%; Asian series sometimes report higher).
• Approximately 15–25% of TNBC tumors are non-basal-like by PAM50 — classified instead as HER2-enriched (5–10%), luminal A (5–10%), luminal B (uncommon), or normal-like (uncommon).
• Approximately 10–15% of basal-like tumors are NOT triple-negative by IHC — typically because they have low-level ER or PR expression at the 1–9% range, or because they are HER2 IHC 1+ or 2+.

The take-away: TNBC is a clinical category defined by absence of three IHC markers; basal-like is a molecular category defined by a transcriptomic signature; they overlap substantially but distinctly. Treating them as synonymous in clinical contexts is approximately right but introduces ~20% misclassification in either direction. For most clinical decision-making (chemotherapy choice, biomarker-based therapy selection), the IHC-defined TNBC category is the operative one; for biology-and-mechanism research, basal-like is often the more biologically homogeneous group.

Intrinsic-subtype biology within TNBC

Among the ~75–85% of TNBC that is basal-like, the dominant biology features:

• High proliferation (high MKI67, MYC, AURKB, TOP2A expression)
• High genomic instability (BRCA1/2 pathway dysfunction, p53 mutations, complex copy-number profiles)
• Basal/myoepithelial differentiation markers (cytokeratin 5, 14, 17; EGFR; vimentin in some subsets)
• Frequent immune infiltration (variable; sometimes high)
• Sensitivity to platinum and DNA-damaging agents

Among the ~15–25% of non-basal-like TNBC:

• HER2-enriched non-amplified. These tumors show HER2-enriched transcriptomic profile without IHC/ISH amplification — possibly indicating chromosomal-level changes at 17q12 short of formal amplification. Limited clinical-decision impact currently.
• Luminal A or B with ER<1%. Tumors that have a luminal molecular profile but fall below the 1% ER threshold for IHC positivity. Whether these should be treated with endocrine therapy is unresolved; in retrospective series they sometimes have modest response to endocrine treatment.
• Normal-like. Rare in TNBC; possibly enriched in special histologies (apocrine, secretory).

The Lehmann/Pietenpol and Burstein TNBC-specific subtypes

Because PAM50 lumps approximately 80% of TNBC into a single category (basal-like), several groups have proposed TNBC-specific molecular subtypes that subdivide the basal-like majority further. The two most-cited TNBC-specific frameworks are the Lehmann/Pietenpol six-subtype (later four-subtype) classification and the Burstein four-subtype classification. See the Lehmann/Pietenpol subtypes synthesis and the Burstein subtypes synthesis for detailed coverage; this page focuses on the foundational intrinsic-subtype framework that those TNBC-specific systems sit on top of.

Current clinical use of intrinsic-subtype testing

Intrinsic-subtype testing via PAM50 (Prosigna) or research-grade equivalents is well-established in HR+ breast cancer for adjuvant chemotherapy decision-making in postmenopausal node-negative or low-node-positive disease. The Prosigna report provides:

• Intrinsic-subtype call (luminal A, luminal B, HER2-enriched, basal-like)
• Risk-of-recurrence (ROR) score — a continuous numerical risk assessment
• 10-year risk-of-distant-recurrence estimate

In TNBC, intrinsic-subtype testing is not routine clinical practice and is not gated by Prosigna labeling. Reasons:

• Approximately 80% of TNBC is basal-like — a positive call doesn't change clinical management materially because there's no basal-like-specific clinical option.
• The ~20% of non-basal-like TNBC (HER2-enriched non-amplified, luminal-with-low-ER, normal-like) has not yet been associated with clinically actionable distinct treatment paths.
• Standard IHC-defined TNBC management already incorporates the actionable biomarkers (PD-L1 CPS, germline BRCA, HER2-low) needed for current treatment decisions.

Intrinsic-subtype testing in TNBC remains a research tool, used in clinical trial stratification (e.g., KEYNOTE-522 sub-analyses), in mechanism-of-resistance studies, and in retrospective cohort characterizations.

Evidence table — intrinsic-subtype framework key publications

Open questions and active investigation

• Subtype calls on single-cell data. Bulk-tumor PAM50 produces a single subtype assignment per tumor. Single-cell transcriptomics reveals substantial intra-tumoral subtype heterogeneity (a "basal-like" tumor by bulk may contain populations with luminal expression patterns). Whether single-cell-resolution subtype heterogeneity has clinical implications for treatment selection or resistance prediction is an active area.
• Spatial subtype heterogeneity. Spatial transcriptomics methods reveal that subtype features can vary across tumor regions; the basal-like portion may be embedded with subdomains of normal-like or HER2-enriched expression. The clinical-decision implications are not yet established.
• The HER2-enriched non-amplified subset. Approximately 5–10% of TNBC and 5–10% of HR+ disease have a HER2-enriched intrinsic-subtype call without HER2 IHC/ISH amplification. Whether these tumors respond to HER2-directed therapy (T-DXd? Trastuzumab in combination?) is being tested in small phase II trials.
• Long-term stability of intrinsic subtype across treatment and metastatic progression. Sørlie 2003 demonstrated stability across primary-recurrence pairs in a small cohort. Whether intrinsic subtype reliably stays the same after chemotherapy + IO and through multi-line metastatic treatment is less well-characterized.
• Adoption of intrinsic-subtype testing in TNBC clinical practice. Could expand if (a) a clinically actionable intervention emerges for a specific TNBC intrinsic-subtype subset (e.g., HER2-enriched-non-amplified getting T-DXd benefit), or (b) intrinsic-subtype call becomes a useful biomarker for residual-disease-decision-making post-KEYNOTE-522.

For TNBC-specific subdivisions of the basal-like majority, see the Lehmann/Pietenpol subtypes synthesis and the Burstein subtypes synthesis. For the overview of TNBC biology and clinical course, see What is TNBC? (overview).

References

Each citation links to the original publication via DOI. The same records are searchable in the evidence library by title or DOI.

1. Perou CM, Sørlie T, Eisen MB, et al. Molecular portraits of human breast tumours. Nature. 2000;406(6797):747–752. doi:10.1038/35021093. ↩
2. Sørlie T, Perou CM, Tibshirani R, et al. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci USA. 2001;98(19):10869–10874. doi:10.1073/pnas.191367098. ↩
3. Sørlie T, Tibshirani R, Parker J, et al. Repeated observation of breast tumor subtypes in independent gene expression data sets. Proc Natl Acad Sci USA. 2003;100(14):8418–8423. doi:10.1073/pnas.0932692100. ↩
4. Parker JS, Mullins M, Cheang MCU, et al. Supervised risk predictor of breast cancer based on intrinsic subtypes. J Clin Oncol. 2009;27(8):1160–1167. doi:10.1200/JCO.2008.18.1370. ↩
5. Prat A, Adamo B, Cheang MCU, Anders CK, Carey LA, Perou CM. Molecular characterization of basal-like and non-basal-like triple-negative breast cancer. Oncologist. 2013;18(2):123–133. doi:10.1634/theoncologist.2012-0397. ↩
6. Cancer Genome Atlas Network. Comprehensive molecular portraits of human breast tumours. Nature. 2012;490(7418):61–70. doi:10.1038/nature11412. ↩

Last reviewed: 2026-06-04. Researcher-layer synthesis page. Evidence grades follow the GRADE-adapted rubric defined at the top of this page.