Genomic characterization of HER2-positive breast cancer and response to neoadjuvant trastuzumab and chemotherapy—results from the ACOSOG Z1041 (Alliance) trial

Abstract

Background

HER2 (ERBB2) gene amplification and its corresponding overexpression are present in 15–30% of invasive breast cancers. While HER2-targeted agents are effective treatments, resistance remains a major cause of death. The American College of Surgeons Oncology Group Z1041 trial (NCT00513292) was designed to compare the pathologic complete response (pCR) rate of distinct regimens of neoadjuvant chemotherapy and trastuzumab, but ultimately identified no difference.

Patients and methods

In supplement to tissues from 37 Z1041 cases, 11 similarly treated cases were obtained from a single institution study (NCT00353483). We have extracted genomic DNA from both pre-treatment tumor biopsies and blood of these 48 cases, and performed whole genome (WGS) and exome sequencing. Coincident with these efforts, we have generated RNA-seq profiles from 42 of the tumor biopsies. Among patients in this cohort, 24 (50%) achieved a pCR.

Results

We have characterized the genomic landscape of HER2-positive breast cancer and investigated associations between genomic features and pCR. Cases assigned to the HER2-enriched subtype by RNA-seq analysis were more likely to achieve a pCR compared to the luminal, basal-like, or normal-like subtypes (19/27 versus 3/15; P = 0.0032). Mutational events led to the generation of putatively active neoantigens, but were overall not associated with pCR. ERBB2 and GRB7 were the genes most commonly observed in fusion events, and genomic copy number analysis of the ERBB2 locus indicated that cases with either no observable or low-level ERBB2 amplification were less likely to achieve a pCR (7/8 versus 17/40; P = 0.048). Moreover, among cases that achieved a pCR, tumors consistently expressed immune signatures that may contribute to therapeutic response.

Conclusion

The identification of these features suggests that it may be possible to predict, at the time of diagnosis, those HER2-positive breast cancer patients who will not respond to treatment with chemotherapy and trastuzumab.

ClinicalTrials.gov identifiers

NCT00513292, NCT00353483

Introduction

In breast cancer, ERBB2 gene amplification, and expression of its corresponding protein, HER2, is observed in ∼15–30% of invasive tumors [1, 2], and is associated with poor prognosis [3]. In these cases, the addition of HER2-targeted agents (e.g. trastuzumab, pertuzumab, and lapatinib) to neoadjuvant chemotherapy increases pathological complete response (pCR) rates [4], which in turn reduces the risk of recurrence and death [5, 6]. Nonetheless, therapeutic resistance remains a concern, with ∼40–50% of patients having residual disease (RD) after neoadjuvant treatment with chemotherapy and trastuzumab [4, 7].

The American College of Surgeons Oncology Group Z1041 trial (NCT00513292) was designed to compare the pCR rate of HER2-positive breast cancer patients that received either a sequential or concurrent regimen of chemotherapy and trastuzumab (Methods). Initial results from the trial identified no significant difference between the arms [8], suggesting that concurrent administration of trastuzumab with anthracyclines offers no additional benefit to patients.

As a monoclonal antibody, trastuzumab relies in part on the immune system to recognize and target HER2-expressing cells. However, there are conflicting reports of whether germline polymorphisms in Fc-gamma receptors, found on the surface of various immune cells and involved in the response to antibodies including trastuzumab, are associated with different outcomes in HER2-positive breast cancer cases that received adjuvant chemotherapy and trastuzumab [9–11]. Moreover, germline polymorphisms in ABCB1, an efflux pump that transports chemotherapeutic agents out of cells, have been identified as prognostic markers in HER2-positive patients [10]. Additionally, in the neoadjuvant setting it has been reported that somatic events are associated with differential response rates to chemotherapy and trastuzumab. Breast cancers assigned to the HER2-enriched subtype have been reported as having higher pCR rates than those assigned to the luminal A or luminal B subtypes [12]. Similarly, expression of the estrogen receptor alpha (ER) has been associated with higher RD rates [12]. While it has been reported that cases with PIK3CA mutations have higher rates of RD [13–15], these findings have not been universally observed [12].

Overall however, mechanisms of resistance remain poorly understood, and without a comprehensive molecular understanding of these mechanisms, therapeutic advances will be delayed. We have therefore performed massively parallel sequencing on 48 HER2-positive breast tumor biopsies and matched blood accrued from either the Z1041 trial or a local neoadjuvant tissue collection study (NCT00353483). All cases were identified as HER2-positive at diagnosis and received neoadjuvant chemotherapy and trastuzumab. We subsequently analysed the genomic landscape that defines HER2-positive breast cancer, and furthermore, identified molecular features that differentiate those cases that achieved a pCR from those that had RD.

Methods

37 cases were accessed from the American College of Surgeons Oncology Group Z1041 trial (NCT00513292) that compared the pCR rate of patients with HER2-positive breast cancer. Patients were randomly assigned to one of two arms; the first received fluorouracil 500 mg/m², epirubicin 75 mg/m², and cyclophosphamide 500 mg/m² (FEC-75) on day 1 of a 21-day cycle for four cycles followed by paclitaxel 80 mg/m² and trastuzumab 2 mg/kg (after a 4-mg/kg loading dose) once per week for 12 weeks, while the second group received paclitaxel and trastuzumab once per week for 12 weeks followed by four cycles of FEC-75 (on day 1 of each 21-day cycle) and once-weekly trastuzumab, in the same doses as the first group. In addition to the patients enrolled in the Z1041 trial, 11 HER2-positive breast cancer cases included in this study were enrolled in a local trial at Washington University School of Medicine (NCT00353483) and received neoadjuvant treatment with trastuzumab in combination with other chemotherapies. All cases were females with invasive breast cancer, not pregnant, and ≥ 18 years of age.

Institutional review boards of the participating institutions approved the study. All participants gave written informed consent.

For all cases, pre-treatment snap-frozen tumor tissue and normal blood were obtained. All cases analysed in this study were identified as HER2-positive by IHC and/or FISH.

All 48 pre-treatment tumor biopsy and blood pairs were assayed by whole genome sequencing (WGS) and whole exome sequencing (WES), and 42 of these tumor biopsies were additionally assayed by RNA sequencing (RNA-seq) (supplementary Figure S1, available at Annals of Oncology online). Alignment and variant calling were performed as previously described [16]. All sequence data have been deposited in the Database of Genotypes and Phenotypes (dbGAP) (accession: phs001291.v1.p1).

Additional details are provided in the Supplementary Methods, available at Annals of Oncology online.

Results

Patient characteristics

The patient cohort investigated in this study consists of 48 HER2-positive breast cancer cases treated with neoadjuvant chemotherapy and trastuzumab. Half of selected patients (n = 24) achieved a pCR after treatment, while the other half (n = 24) had RD (Figure 1 and Table 1; supplementary Table S1, available at Annals of Oncology online).

Figure 1.

REMARK diagram detailing the study cohort.

Table 1.

Summary of the clinical and molecular features of the cohort in this study

	Unstratified	pCR achieved	Residual disease
Number of cases	48	24	24
Age
Median	50	50	50
Range	36–70	39–70	36–67
Unknown	0 (0%)	0 (0%)	0 (0%)
Clinical primary tumor stage (T)
T1	2 (4.17%)	1 (4.17%)	1 (4.17%)
T2	24 (50%)	14 (58.3%)	10 (41.7%)
T3	16 (33.3%)	6 (25.0%)	10 (41.7%)
T4	6 (12.5%)	3 (12.5%)	3 (12.5%)
Clinical lymph node stage (N)
N0	16 (33.3%)	7 (29.2%)	9 (37.5%)
N1	25 (52.1%)	14 (58.3%)	11 (45.8%)
N2	6 (12.5%)	2 (8.33%)	4 (16.7%)
N3	1 (2.08%)	1 (4.17%)	0 (0%)
Unknown	0 (0%)	0 (0%)	0 (0%)
ER status prior to neoadjuvant treatment
Positive	28 (58.3%)	11 (45.8%)	17 (70.8%)
Negative	20 (41.7%)	13 (54.2%)	7 (29.2%)
Unknown	0 (0%)	0 (0%)	0 (0%)
PR status prior to neoadjuvant treatment
Positive	21 (43.8%)	7 (29.2%)	14 (58.3%)
Negative	26 (54.2%)	16 (66.7%)	10 (41.7%)
Unknown	1 (2.08%)	1 (4.17%)	0 (0%)
HER2 status prior to neoadjuvant treatment
Positive	48 (100%)	24 (100%)	24 (100%)
Negative	0 (0%)	0 (0%)	0 (0%)
Unknown	0 (0%)	0 (0%)	0 (0%)
Grade
1	1 (2.08%)	0 (0%)	1 (4.17%)
2	12 (25%)	7 (29.2%)	5 (20.8%)
3	34 (70.8%)	17 (70.8%)	17 (70.8%)
Unknown	1 (2.08%)	0 (0%)	1 (4.17%)
PAM50 subtype
Luminal A	12 (25%)	2 (8.33%)	10 (41.7%)
Luminal B	11 (22.9%)	8 (33.3%)	3 (12.5%)
HER2-enriched	14 (29.2%)	11 (45.8%)	3 (12.5%)
Basal-like	5 (10.4%)	1 (4.17%)	4 (16.7%)
Normal-like	0 (0%)	0 (0%)	0 (0%)
Unknown	6 (12.5%)	2 (8.33%)	4 (16.7%)
AIMS subtype
Luminal A	2 (4.17%)	0 (0%)	2 (8.33%)
Luminal B	8 (16.7%)	3 (12.5%)	5 (20.8%)
HER2-enriched	27 (56.2%)	19 (79.2%)	8 (33.3%)
Basal-like	4 (8.33%)	0 (0%)	4 (16.7%)
Normal-like	1 (2.08%)	0 (0%)	1 (4.17%)
Unknown	6 (12.5%)	2 (8.33%)	4 (16.7%)
Recurrence
Recurrence event	9 (18.8%)	1 (4.17%)	8 (33.3%)
No recurrence event	39 (81.2%)	23 (95.8%)	16 (66.7%)
Unknown	0 (0%)	0 (0%)	0 (0%)
Source of specimens
Z1041 trial (NCT00513292)	37 (77.1%)	21 (87.5%)	16 (66.7%)
WashU study (NCT00353483)	11 (22.9%)	3 (12.5%)	8 (33.3%)

In our study cohort, patients that achieved a pCR tended to have smaller tumors, clinical node negative disease, and be assigned to the HER2-enriched subtype. Although ER status, PR status, and grade were not associated with pCR, PAM50 subtype, AIMS subtype, and disease recurrence were significantly associated (supplementary Table S2, available at Annals of Oncology online). This trend was reduced after stratification on ER status, although the association between PAM50 and pCR as well as AIMS and pCR was significant within the ER-negative set, while only the association between PAM50 and pCR was significant in the ER-positive set. Among all cases, those assigned to the HER2-enriched subtype by RNA-seq analysis were significantly more likely to achieve a pCR compared to tumors assigned to luminal A, luminal B, basal-like, and normal-like subtypes (PAM50: 11/14 versus 11/28, P = 0.023; AIMS: 19/27 versus 3/15, P = 0.0032). About 57% of all cases were assigned to the same subtype in both the PAM50 and AIMS schemes, which is lower than the 77% overlap previously reported for these schemes [17].

Germline polymorphisms

In order to investigate the landscape of germline variants in HER2-positive breast cancer, we analysed the normal genomes of each patient for known pathogenic variants. We identified 196 pathogenic germline variants (482 total events) as well as 19 likely pathogenic germline variants (27 total events), with a median of 10 and 0 variants per patient, respectively (supplementary Table S3, available at Annals of Oncology online). Affected genes include those involved in DNA repair (ATM, BRCA2, MUTYH), cell cycle (FANCA), the Wnt pathway (LRP5, WNT10A), the PI3K pathway, (PIK3R5), and telomere maintenance (TERT). For each variant, the proportion of cases with that variant did not differ between pCR and RD groups (FDR > 0.2). Similarly, no significant differences were observed in the number of variants in a given gene between the two patient groups (FDR > 0.2).

Previous studies have reported germline polymorphisms as predictive of prognostic outcome in HER2-positive breast cancer patients treated with trastuzumab and chemotherapy [10, 11]. These polymorphisms are found in genes including P-glycoprotein (ABCB1) and Fc gamma receptors (FCGR2A, FCGR2B, FCGR3A). In our study, however, the rates of alternate alleles at these polymorphisms did not differ between those patients that achieved a pCR and those with RD (P > 0.05, supplementary Table S4, available at Annals of Oncology online).

The somatic variant landscape of HER2-positive breast cancer

Across all 48 cases, 6685 somatic events including single nucleotide variants (SNVs) and small insertions/deletions (indels) were identified in coding regions, including 5751 missense, 531 nonsense, 20 nonstop, and 165 frameshift insertions or deletions (Figure 2A and supplementary Table S5, available at Annals of Oncology online). TP53, PIK3CA, and MT-ND2 were the only genes mutated at significantly higher rates than expected by chance (supplementary Table S6, available at Annals of Oncology online, FDR < 0.2). No significant differences were observed in the mutation rates for any gene between cases that achieved a pCR versus those that had RD (P > 0.05).

Figure 2.

(A) Somatic variants and small indels are visually represented across the most recurrently mutated genes. Only TP53, PIK3CA, and MT-ND5 were mutated at rates significantly higher than random across the entire cohort (FDR < 0.2, marked in bold with an asterisk). No genes were mutated at significantly different rates between cases that achieved a pCR versus those that had residual disease. (B) Amino acid locations affected by somatic mutations are shown for TP53, PIK3CA, and ERBB2. Individual mutations are colored by whether they achieved a pCR or had residual disease, and are labeled according to their case identifier. Mutations in TP53 were spread across the protein, but PIK3CA mutations in the adaptor-binding, C2, and Ras-binding domains were only observed among cases that had residual disease. Two of the three mutations in ERBB2 were only predicted to be found in a short isoform.

27 somatic mutations were observed in TP53, the most frequently mutated gene (Figure 2B). Mutations were highly variable across p53 protein domains, and we did not observe apparent differences in the domain-specific mutation rates or predicted effect of mutations between cases that achieved a pCR versus those that had RD. In contrast, somatic mutations in the adaptor-binding, C2, and Ras-binding domains of PIK3CA were only observed among cases that had RD, indicating that there may be functional differences in the activity and regulation of the protein (Figure 2B). Somatic mutations in ERBB2 were identified in three cases, all of which failed to achieve pCR (Figure 2B). BRC254 was observed to have a nonsense mutation (p.W89*) that is expected to result in early truncation of the gene. However, this truncation was observed to affect a short isoform of the gene (ENST00000578502) that only contains part of the corresponding extracellular protein domains. Similarly, BRC258 was observed to have a missense mutation (p.S5F) that affected the same short isoform, causing an amino acid substitution within the insulin-like growth factor binding domain. Finally, BRC294 was observed to have a missense mutation (p.R1153Q) causing an amino acid substitution within the terminus of the intracellular portion of the protein, but not targeting known protein domains. To our knowledge, these mutations have not been previously reported, and none of the mutations were found in mutational ‘hot spots’.

We further investigated whether molecular pathways were enriched for somatic SNVs and indels. Various pathways were observed to be significantly enriched in all 48 cases, including those related to cell cycle, PI3K-Akt signaling, p53 signaling, Wnt signaling, focal adhesion, TGF-beta signaling, ErbB signaling, and immune function (supplementary Table S7, available at Annals of Oncology online). Moreover, those pathways found to be significant among cases that achieved a pCR tended to also be significant among cases that had RD. This observation is in line with our previous finding that no gene was mutated at significantly different rates between these two groups.

More broadly, using genome-wide somatic mutations we predicted the clonal architecture of individual cases. The number of subclones ranged from two to seven across all cases, but the subclone rates were not significantly different between cases that achieved a pCR versus those with RD (P = 0.356). Genome-wide mutations were dominated by C > T and C > G changes (or equivalently, G > A and G > C changes). These mutations were characterized as being predominantly attributed to mutational signature 1 (associated with age), and to a lesser extent signatures 2 and 13 (associated with the APOBEC family of cytidine deaminases), as well as signature 3 (associated with dysfunctional BRCA repair) [18, 19] (supplementary Table S8, available at Annals of Oncology online). These observations were mostly consistent with previous findings, although we did not observe contributions from signature 5 (aetiology unknown), which has been reported to be a prominent signature among HER2-positive cases [19]. Interestingly, we identified examples where the contribution of mutational signatures to the somatic landscape of tumors was significantly different between cases that achieved a pCR versus those that had RD. Signature 13 contributed significantly more to the landscape of cases with RD (P = 0.0401), and signature 3 contributed more to the landscape of cases that achieved a pCR (P = 0.0540).

Copy number changes and RNA expression associate with therapeutic response

Copy number aberrations previously characterized in breast cancer were commonly observed in our cohort (supplementary Figures S2 and S3, available at Annals of Oncology online). In particular, we observed focal amplifications in 17q, whole arm gains in 1q, 8q, and 20q, and whole arm deletions in 1p, 8p, and 17p. Similar patterns were observed between cases that achieved a pCR and those with RD; in a genome-wide analysis, no regions were found to have copy number gains or losses at significantly different rates (FDR > 0.2).

Most tumors in our cohort had observable copy number gain of the genomic region surrounding ERBB2 (Figure 3A). Consistent with what has been reported previously, nearly all these gains were focal amplifications rather than whole arm, although the copy number and size of the amplified region varied within both patient groups. There were no cases where we observed copy number gains of both alleles, indicating that possible germline variant heterozygosity of the HER2 protein was not a determinant of therapeutic response. However, we observed eight cases with either no observable (BRC259, BRC267, BRC274, BRC286) or low-level (BRC265, BRC283, BRC285, BRC289) ERBB2 copy number gain. These cases were significantly less likely to achieve pCR, with seven out of the eight cases having RD (P = 0.048). Interestingly, all four cases with no observable gain were confirmed to have a score of 3+ for HER2 IHC, and case BRC259 was additionally observed to have an average of >20 copies of ERBB2 per nucleus assessed by Fluorescent In Situ Hybridization (FISH). All eight cases had excellent WGS tumor coverage (median mean coverage = 53.5×; mean coverage range = 42×–59×), and high maximum WGS variant allele frequencies (median 67%, range 46–95%), indicating that these samples all had high tumor purity.

Figure 3.

(A) DNA copy number aberrations are shown for all cases across chromosome 17 and the ERBB2 amplicon. Predicted copy number gains are shown in red and losses in blue at 10 000 bp windows across the genome. The height of each panel was separately scaled for each case according to the maximum copy number value observed. Cases are ordered by whether they had residual disease or achieved a pCR. (B) ERBB2 RNA FPKM expression values and DNA copy number values are plotted against each other for the 42 cases where RNA sequencing was performed. The values were highly correlated (Spearman R²=0.769). Cases with low ERBB2 levels (bottom left corner of each plot) tended to have a residual disease and to not be classified as HER2-enriched.

Across the cohort, ERBB2 RNA-seq (FPKM) and copy number values were highly correlated (Spearman R²=0.769, Figure 3B). Although BRC286 achieved a pCR, the tumor from this case was observed to have copy number loss of ERBB2 and a relatively lower level of ERBB2 expression within the bulk profile (FPKM = 30.066).

More broadly within the RNA-seq data, no genes or distinct transcript isoforms were found to be univariately differentially expressed between cases that achieved pCR versus those with RD (FDR > 0.2). At the multivariate level, however, we did observe significant differences in the expression of gene signatures. About 1531 out of 10 959 signatures tested were enriched in cases that achieved a pCR (supplementary Table S9, available at Annals of Oncology online, P < 0.05 and FDR < 0.2). These signatures were predominantly related to immune categories (P = 8.69×10 ^− 42), with T-Cell (P = 5.49×10 ^− 21), B-Cell (P = 1.17×10 ^− 14), and inflammatory (P = 0.0203) signatures all significantly enriched. In contrast, no signatures were enriched in the RD group, indicating that these cases have greater heterogeneity than those that achieved a pCR.

A total of 311 gene fusion events were observed in the cohort, with a median of 6.5 and range of 1–27 events (supplementary Table S10, available at Annals of Oncology online). No recurrent fusion involving the same two genes was observed, however 62 genes were involved in a fusion in at least two cases. ERBB2 was the gene most frequently involved in fusion events, with five events observed in four cases (BRC260, BRC272, BRC287, BRC295), followed by GRB7 with four events in three cases (BRC251, BRC262, BRC295). Two of the four cases with ERBB2 gene fusions achieved a pCR, while all of the three cases with GRB7 gene fusions achieved a pCR. These results suggest that genomic instability in the ERBB2 amplicon is the predominant source for gene fusions in HER2-positive breast cancer, although such fusions do not lead to therapeutic resistance.

Neoantigen profiles of HER2-positive breast cancer

Given the observed role for T-Cell immune response in patients that achieved a pCR, we predicted in silico the tumor neoantigens and binding affinity scores for each case (supplementary Figure S4, available at Annals of Oncology online). Among neoantigens with binding affinity scores low enough to putatively elicit an immune response, loads ranged from 0 to 2584 epitopes across all cases. Cases that achieved a pCR had nonsignificantly higher loads compared to those with RD (P > 0.05). The genes that most frequently produced putatively active neoantigens were similar to those that were most frequently mutated in our cohort, including TP53 and PIK3CA (supplementary Table S11, available at Annals of Oncology online). Most peptides predicted to elicit an immune response were only observed in a single case, and none were observed in more than two cases (supplementary Table S12, available at Annals of Oncology online).

Prediction of therapeutic response using combined genomic data levels

In order to identify whether interactions between different data levels of genomic features are informative of response to chemotherapy and trastuzumab, we computationally generated predictors of pCR using copy number, gene mutation, and gene isoform expression data. Using these data, we were able to achieve an area under the curve (AUC) of 0.720 for predicting which cases achieved a pCR or had RD (supplementary Figure S5, available at Annals of Oncology online). Transcriptional features were consistently selected as the most influential in making these predictions (supplementary Table S13, available at Annals of Oncology online), and included isoforms of genes such as NOP56, SCARB1, LAD1, ETFA, and LTBP3. This result is consistent with our previous observations that RNA features are able to differentiate these two classes, while copy number and mutational data have a limited ability to do so. Similar results were observed when we constructed predictors of therapeutic response using only the transcriptional, copy number, or mutational data. A successful predictor was trained using the transcriptional data (AUC = 0.725) and to a lesser extent using the mutational data (AUC = 0.663), but the copy number data was not informative (AUC = 0.550).

Discussion

Here we have characterized the genomic landscape of HER2-positive breast cancer, and further identified a set of recurrent somatic events in HER2-positive breast cancer that may aid in the identification of patients, at the time of diagnosis, who will not have a pCR after treatment with chemotherapy and trastuzumab.

It is interesting to note that while all 48 cases were clinically classified as HER2-positive, we identified four cases with no observable ERBB2 copy number gains, and an additional four with very low level, whole chromosome arm gains. Although all eight cases were scored as 3+ by IHC, seven of them had RD, together indicating that massively parallel sequencing technologies provide predictive information in complement to clinical IHC/FISH. Moreover, we observed three cases with somatic mutations in ERBB2; all three cases had RD. Although the mutations did not directly confer obvious resistance by alteration of domains within the full-length protein, it is possible that these somatic events led to the generation of neoantigens, thereby allowing the host immune system to target the cancer cells harboring these mutations.

In addition to somatic events that directly alter ERBB2, we identified broad patterns at the RNA level that were associated with response to chemotherapy and trastuzumab. This is consistent with a previous report that transcriptional features are more predictive than other data types of drug response in breast cancer cell lines [20]. In particular, cases that achieved a pCR had evidence of an activated immune response, including T-Cell, B-Cell, and inflammatory signatures. In contrast, cases with RD displayed evidence of greater heterogeneity as no RNA signatures were enriched among this group. This hypothesis is further reinforced by the observation that cases classified as HER2-enriched achieved significantly higher rates of pCR compared to those classified as luminal A, luminal B, basal-like, or normal-like.