Molecular Subtypes

How a microarray changed a word

For most of the twentieth century, breast cancer was classified by what pathologists could see: tumour size, lymph node involvement, the degree of cellular disorder called grade, and whether certain proteins were present on the cell surface. These were powerful tools, but they described how a tumour looked, not what it was doing at the level of the gene.

That changed in 2000, when Charles Perou, Therese Sørlie and colleagues at Stanford University published work in Nature using DNA microarrays — chips that measure the activity of thousands of genes simultaneously. They profiled a set of breast tumours and found that the oestrogen receptor status, long treated as a single binary variable, was in fact a marker of something deeper: tumours did not merely have the receptor or lack it; they clustered into groups with coherent, reproducible patterns of gene expression that predicted behaviour more sharply than any single marker alone.

The taxonomy that emerged from that work and its successors divided the disease into subtypes. Luminal A tumours — typically oestrogen-receptor positive, low-grade, slow-growing — carried the best prognosis. Luminal B tumours shared the receptor but grew faster and sometimes expressed HER2. The HER2-enriched subtype, defined by amplification of the ERBB2 gene, was strongly associated with worse outcomes before targeted treatment existed. Basal-like tumours, largely receptor-negative and now closely overlapping the clinical category called triple-negative (negative for oestrogen receptor, progesterone receptor and HER2), carried the most aggressive biology and the fewest treatment targets.

The overlap between these intrinsic subtypes and clinical measurements is real but imperfect. Triple-negative is not synonymous with basal-like; luminal B is not a single thing. Subsequent platforms — PAM50, which gives the commercially licensed Prosigna score, and the Oncotype DX recurrence score — distilled the principle into clinically deployable assays, applying it to the question of which patients with receptor-positive disease would benefit from cytotoxic chemotherapy and which would not. The MINDACT trial in Europe and the TAILORx trial in the United States, both reported in the late 2010s, showed that a substantial proportion of women previously treated with chemotherapy could safely forgo it on the basis of their genomic score — results summarised in the TAILORx publication in the New England Journal of Medicine (2018).

The underlying point is taxonomic: what the word breast cancer refers to was redefined, not by revising anatomy or staging systems, but by listening to the genome's own account of each tumour's character, which turned out to be louder and more specific than anything a microscope slide alone could say.