Abstract 92P
Background
The HER2+ subtype is characterized by the overexpression of the HER2 oncoprotein based on immunohistochemistry (IHC) or by ERBB2 gene amplification using in situ hybridization (ISH) techniques. Although all patients are eligible for target therapy, most don't benefit from it in neoadjuvant (NAC) and adjuvant settings. Identifying predictive factors of response to therapy is thus crucial for optimizing treatment protocol.
Methods
We performed a comprehensive integrative analysis based on clinical, molecular, and genetic profiles from publicly available datasets (TCGA, METABRIC, and SPY). We selected all samples classified as HER2 positive by IHC or ISH. Unsupervised hierarchical clustering was used to infer distinct molecular profiles. The higher informative genes (HIG) were selected by differential expression comparisons and discriminative selection based on principal component analysis (PCA) in the discovery dataset (TCGA). Survival analysis was performed on the METABRIC dataset, and Bayesian inference was used to compare the enhancement regarding response to NAC prediction in SPY.
Results
The unsupervised clustering identified 3 clusters among 182 HER2+ TCGA samples. Cluster 1 (C1) consisted of 40% ER-negative and 60% ER-positive samples, while C2 (93) was composed of 98% ER-negative samples. C3 was the smallest cluster (10) and consisted mainly of triple-negative (TNBC) tumors. We identified the 17q12-q21 amplicon as completely distinct among the clusters. C1 samples displayed a solid core amplification (r> 0.9) around the ERBB2 gene composed of 9 genes (NEUROD2, PPP1R1B, STARD3, TCAP, PNMT, PGAP3, MIR4728, MIEN1, GRB7, IKZF3). This pattern was not observed in C2 and C3. The hierarchical clustering in the METABRIC and SPY datasets based on the expression of 44 HIGs identified 3 clusters with similar IHC profiles. In METABRIC, the risk of death was significantly higher in C1 versus C2 (p= 0.01). In the SPY dataset, response to NACT in C1 was 59.2% compared to 23% in C2 (p< 0.0001).
Conclusions
We have demonstrated that molecular heterogeneity in HER2+ breast cancer may be highly predictive in selecting patients who benefit from target therapy. Table: 92P
| C1 | C2 | |
| pCR (all) | 87/147 (59) | 18/78 (23) |
| pCR (ER+) | 39/76 (51) | 18/78 (23) |
| pCR (ER-) | 48/71 (67.6) | NA |
Clinical trial identification
| C1 | C2 | |
| pCR (all) | 87/147 (59) | 18/78 (23) |
| pCR (ER+) | 39/76 (51) | 18/78 (23) |
| pCR (ER-) | 48/71 (67.6) | NA |
Editorial acknowledgement
Legal entity responsible for the study
The authors.
Funding
Has not received any funding.
Disclosure
All authors have declared no conflicts of interest.