Complex rearrangements fuel ER+ and HER2+ breast tumours

The IC subtypes have distinct CNA landscapes (Extended Data Fig. 1j), but their SV landscape and evolution have not been investigated. Leveraging ENiClust, we found that the IC-subgroup-specific genomic landscape of breast cancer is consistent throughout disease progression despite an increased burden of alterations^10,13,16,17 (Fig. 2a and Extended Data Fig. 2a,b). Both HER2⁺ and ER⁺ high-risk primary and metastatic tumours exhibit characteristic sharp increases in SV burden at their respective recurrently amplified loci (IC5: 17q12; IC6: 8p11; IC2: 11q13; IC1: 17q23). The peak of SV burden at 17q12 (ERBB2) suggests that ERBB2 amplification is fuelled by complex alterations, such as ecDNA¹⁸. The mutational burden in primary ER⁺ typical-risk tumours was minimal (Supplementary Fig. 1i) but increased in metastatic disease (Fig. 2a), in part owing to treatment (Extended Data Fig. 2c). IC10 and IC4ER⁻ tumours exhibit diffuse genome-wide instability with an increased SV burden, although the latter show an attenuated pattern and harbour fewer pathogenic SVs and alterations in DNA repair pathways, confirming previous reports¹⁹ (Extended Data Fig. 2d,e). Across metastatic sites, the cumulative burden of alterations was higher in lung and subcutaneous metastases and lower in soft-tissue and in-breast recurrences (Extended Data Fig. 2f). These subgroup-specific alterations were seen in DCIS (Extended Data Fig. 2a), emphasizing early oncogene addiction and mechanisms of malignant transformation.

a, IC group-level CNA profile (shaded area; dark denotes amplification, light denotes deletion) with SV burden (line) as overlay and total alteration burden in primary and metastatic samples. b, Pareto front projection on ternary plot of CNA and SV signature profiles from primary (left) and metastatic (right) tumours independently, resulting in three genomic archetypes. Each plotted circle represents a tumour. c, Lollipop plots illustrating the correlation between mutational features and the distance to each archetype. amp., amplification; BFB, breakage–fusion–bridge; TIC, templated insertion chain; LOH, loss of heterozygosity; WGD, whole-genome doubling; FGA, fraction of genome altered.

Next we characterized CNA and SV signatures in 702 primary breast tumours, replicating the 24 CNA²⁰ and 6 rearrangement^8,21 signatures (RSs) previously reported (Supplementary Fig. 2a–c). RS3, RS5 (associated with homologous repair deficiency (HRD); Supplementary Fig. 2d) and CN17 were enriched in IC10 tumours, whereas RS4, RS6 (associated with complex amplifications) and CN7 were enriched in ER⁺ high-risk and HER2⁺ tumours (Extended Data Fig. 2g,h and Supplementary Fig. 2e–g). ER⁺ typical-risk tumours were enriched for CN1 (associated with diploid genomes; Supplementary Fig. 2d,e).

Projected on a two-dimensional plane (Supplementary Fig. 3a,b), the architectural profiles follow a continuum and form a polyhedron reminiscent of Pareto optimum theory, which illustrates trade-offs between biological tasks²². Primary breast cancers map onto three dominant genomic archetypes (Supplementary Fig. 3c–f): TNBC-enriched, ER⁺ typical-risk-enriched and ER⁺ high-risk + HER2⁺-enriched. Tumours dominated by a single mutational process are proximal to a vertex, whereas those characterized by multiple processes cluster at the centre (Fig. 2b and Extended Data Fig. 2i). The TNBC-enriched archetype was positively correlated with genomic instability, HRD and APOBEC-editing SNVs (Fig. 2c and Supplementary Fig. 3g). Compared to ER⁺ high-risk tumours, HER2⁺ tumours were enriched for tyfonas (Extended Data Fig. 2j). The ER⁺ high-risk + HER2⁺-enriched archetype was positively correlated with complex amplifications, reactive oxygen species and APOBEC-associated SNVs harbouring co-amplification of multiple cytobands (Extended Data Fig. 3a). By contrast, the ER⁺ typical-enriched archetype negatively correlated with most genomic features.

Tumours predicted to be BRCA-like on the basis of germline or somatic genomic features²³ map to the TNBC-enriched archetype (Extended Data Fig. 3b). Indeed, both BRCA1-like and BRCA2-like ER⁺ and ER⁻ tumours demonstrated significantly higher TNBC-archetype scores than non-HRD tumours, and HRD-like ER⁺ high-risk tumours were closer to the TNBC-enriched archetype than their non-HRD-like counterparts (OR = 5.09; P = 6.5 × 10⁻⁴). Additionally, the mutational patterns of BRCA1-like and BRCA2-like ER⁻ and ER⁺ tumours were highly concordant (Supplementary Fig. 3h,i). Notably, whereas 43.6% of TNBC tumours were HRD-like, 13.2% of ER⁺ high-risk tumours were also predicted to be HRD-like, with most being ER⁺ high-risk IC1 or IC9 (OR = 4.43; P = 0.03; Extended Data Fig. 3c and Supplementary Fig. 3j). Indeed, although foldback inversions and pyrgos were enriched in TNBC (foldback inversion: 17.3%, P = 2.00 × 10⁻³; pyrgos: 18.8%, P = 9.33 × 10⁻⁴), these mutational events were also observed in ER⁺ tumours (5.1% and 4.1%, respectively; Extended Data Fig. 3d). These data reinforce multiple mechanisms of genome instability in TNBC²⁴ that also affect a subset of ER⁺ tumours.

The three genomic archetypes replicated in an independent cohort of 2,229 primary tumours from Genomics England²¹ (Extended Data Fig. 3e). Overall, the genomic landscape of primary breast tumours falls along a continuum with mutational patterns captured by three main genomic archetypes, namely, genome-stable, diploid genomes (ER⁺ typical-risk-enriched), genome-wide instability (TNBC-enriched) and focal, complex amplifications (ER⁺ high-risk + HER2⁺-enriched).

Metastatic lesions exhibit increased SNV and SV burdens compared to unpaired primary tumours, probably owing to therapy, as we and others have shown^13,17. Using the above approach, we identified six de novo SV signatures in metastases that correlated with those in primary tumours (Supplementary Fig. 4a,b) and showed similar subgroup-specific enrichment patterns (Extended Data Fig. 3f). Two-dimensional projection again revealed three dominant archetypes (Supplementary Fig. 4c) that overlap with those in primary tumours (Fig. 2b,c, Extended Data Fig. 3g and Supplementary Fig. 4d). Our results were robust to choice of dimensionality reduction algorithm (Supplementary Fig. 4e–g). Thus, the three genomic archetypes of breast cancer are conserved in metastatic disease.

SV signatures were generally conserved, although increased, in metastatic tumours except for RS4 and RS6 in ER⁺ high-risk and HER2⁺ tumours, respectively, which were stable (Extended Data Fig. 3h). These data support the early occurrence of complex rearrangements and their persistence through metastasis. Although the distribution of CNA signatures mirrored primary tumours, the Pareto front revealed increased alteration burden and more intermixed profiles in metastasis, consistent with increased whole-genome doubling and genomic instability¹⁷ (Extended Data Fig. 3i and Supplementary Fig. 4h,i). Thus, metastatic tumours retain the scars of subgroup-specific mutational processes operative in early-stage disease.

Although ER⁺ typical-risk tumours have a favourable prognosis, 29% of patients experience distant relapse⁴. We investigated whether the genomic archetypes improve risk stratification. Mapping METABRIC onto the Pareto front (Methods, Extended Data Fig. 3j and Supplementary Fig. 4j–l), the position of ER⁺ typical-risk tumours was predictive of relapse, with recurrent tumours mapping closer to the ER⁺ high-risk + HER2⁺ archetype (Extended Data Fig. 3k,l) accompanied by a higher HRD loss-of-heterozygosity score, invasive lobular carcinoma (ILC) histology and increased proliferation.

In METABRIC, ILCs were enriched in ER⁺ typical-risk tumours (OR = 2.20, P = 2.27 × 10⁻³, Fisher’s exact test; Supplementary Fig. 4m). Within ER⁺ high-risk tumours, ILCs exhibited a higher 5-year recurrence risk (39% versus 30%) and cumulative recurrence risk (62% versus 54% at 20 years; Extended Data Fig. 3m). This difference was more marked among ER⁺ typical-risk tumours (55% versus 37% at 20 years). ILCs were closer to the ER⁺ typical-risk archetype than their invasive ductal carcinoma (IDC) counterparts (P = 2.10 × 10⁻⁵; Extended Data Fig. 3n,o) given their lower levels of whole-genome doubling, ploidy and fraction of genome altered. Thus, given comparable genomic architectures, lobular histology remains a high-risk feature.