We were able to confirm that the intrinsic signature described previously [1
] exists in a broad population-based cohort of breast cancers. This is by far the largest validation and clinical characterization of the molecular subtypes to date. Previous characterization was given in [1
] – p53
mutation status on 69 cases – and in an unpublished report [22
], reviewed in [23
]. The different prognosis of the subtypes was already reflected in the tumor characteristics, although the patterns were complex. By performing a separate analysis of endocrine-treated and adjuvant-untreated patients we were also able to evaluate the differences between subtypes in terms of both prognosis and response to therapy.
Normal-like tumors were small, found in postmenopausal women, seldom of Elston grade III, mostly stable diploid and more prevalent among HRT users. They had a favorable prognosis under endocrine therapy. Luminal-A tumors closely resembled normal-like tumors but were not associated with HRT use and had a superior prognosis regardless of adjuvant therapy. In contrast, the ERBB2 tumors showed a 50% relapse-free 5-year survival not influenced by therapy. The poor prognosis was reflected in large tumor size, high proportion of Elston grade III and unstable aneuploids, and few ER-positive tumors. Luminal B tumors had a clearly worse prognosis than luminal A ones, and were apparently less responsive to the hormonal therapy than the basal-like and normal-like tumors [2
]. In comparison with the luminal A group, this group had a higher proportion of Elston grade III, p53 mutation and lymph-node positive tumours, confirming the hypothesis of a more proliferative profile [23
]. Surprisingly, although the basal-like tumors had even fewer ER-positive tumors and more high-grade tumors, they had better relapse-free survival than the ERBB2 tumors, especially among the endocrine-treated patients. It is worth noting that about one-third of the patients with basal-like tumors used HRT at the time of diagnosis.
The intrinsic signature set of approximately 500 genes was originally selected on the basis of their stable expression between pairs of samples taken from the same tumors, before therapy and after 15 weeks of neoadjuvant therapy [1
]. This implies that the set is enriched in genes whose expression patterns are characteristic for individual tumors as opposed to those that vary as a function of tissue sampling, and hence would be ideally suited for classification. Another recent study [24
] showed that the subtypes were also observed among Asian-Chinese patients, although this study performed only the hierarchical clustering and identified only three subtypes.
The discordant subtyping between the centroid prediction and k-means clustering indicates some overlapping characteristics between subtypes, and potentially reveals a biological heterogeneity more complex than the subtypes alone. Discordances occurred most frequently between the luminal A and luminal B groups and between the luminal B and ERBB2 groups. Not surprisingly, despite differences in receptor status, luminal B and ERBB2 share similar clinical – such as tumor grade – and prognostic characteristics. This suggests that there must in fact be grade-associated and outcome-associated genes in the intrinsic gene set, and that these two subtypes have similar expression in these genes. A similar argument can also be made from the large confusion of luminal A and luminal B, namely that this must have been driven by ER-associated genes, because these two subtypes have similar receptor characteristics. Having the fewest discordances, the normal-like subtype seemed to have the most distinctive profile. One of the most striking characteristics was the high proportion of HRT users within this subtype.
Although the intrinsic signature is expected to reveal fundamental tissue properties, such as origin (basal-like versus luminal), our study supports the hypothesis that the subtypes are distinct biological entities with distinct clinical characteristics. Our study showed how they differed in progression-related characteristics, such as tumor grade, p53
mutation and genomic instability. This means that there must be a strong correlation between the intrinsic gene set and other progression-related signatures based on p53
transcription profile [4
], proliferation or genomic instability, and possibly an inherent capability to metastasize [25
]. The extent to which the different signatures contribute independent information toward clinical progression or prognosis requires further study.
Our findings further support the hypothesis of a reduced response to hormone-based treatments for ERBB2 and luminal B tumors [23
]. Whereas luminal A tumors seem to have a good prognosis overall, luminal B tumors had a poor prognosis regardless of endocrine therapy, indicating some resistance to therapy. This may be explained at least partly by a worse clinical profile, with a higher tumor grading and a higher proportion of p53
mutation. In contrast, in the endocrine-treated group the basal-like subtype seemed to survive better than expected considering its poor clinical characteristics including its high proportion of Elston grade III (75%) and few ER-positive (46%) tumors. The ERBB2 tumors had the lowest relapse-free survival: approximately 50% of the patients had either died from breast cancer or experienced distant metastases. Overexpression of the human transmembrane tyrosine kinase growth factor receptor (HER-2), characteristic of the ERBB2 group, has been associated with more aggressive forms of tumor and with resistance to endocrine therapies [26