Relapsed Classic E-Cadherin (CDH1)–Mutated Invasive Lobular Breast Cancer Shows a High Frequency of HER2 (ERBB2) Gene Mutations

This article is featured in Highlights of This Issue, p. 2593

Substantial progress has been made over the past 3 decades in our understanding of the epidemiology, clinical course, and basic biology of breast cancer and the integration of routine and molecular biomarkers into patient management. Modern techniques designed to detect the disease at an earlier stage combined with new methods of determining risk assessment and more optimized combined modality treatment that have enhanced our ability to manage, and in some cases, achieve a cure for the disease. The morphologic classification of invasive carcinomas of the breast has had a significant role in managing individual patients but has not been a major driver of therapy development and clinical trials. In addition, the application of standard slide-based biomarker status for the estrogen receptor (ER), progesterone receptor (PR), and HER2 gene copy number (FISH) and/or protein expression [immunohistochemistry (IHC)] currently play a more significant role in clinical trial design than routine histologic subtyping (1). This is also true when the mRNA expression profile-based molecular portraits classification of invasive breast cancer including the luminal A, luminal B, normal, HER2-positive, and basal-like (typically “triple negative”) nomenclature has been used to classify the tumors (2). In drug development, strategies have approached invasive ductal and invasive lobular breast carcinomas as essentially the same disease (3–6).

Most published studies have found that the overall prognosis for invasive lobular carcinoma (ILC) of the breast is similar to that of invasive ductal carcinoma (IDC; refs. 7–9). However, most of these outcome studies have excluded the less common, but more aggressive pleomorphic variant of ILC in the clinical outcome data analysis (10), which could overestimate the prognosis of patients with ILC. In addition, although ER-positive ILC of the breast generally has a favorable prognosis, relapsed lobular breast cancer originally treated only with hormonal therapy may, on occasion, follow an aggressive clinical course (7–9). All of these observations favor the study of ILC as a separate entity from the more frequent invasive ductal breast carcinoma.

Recent studies suggest that lobular breast cancer, which makes up 10% of all invasive breast cancers and approximately 20,000 new cases in the United States each year, may be well characterized by a distinct genotype featuring a mutation in CDH1 that encodes the E-cadherin protein product, in contrast with the far more frequent ductal breast cancer, which typically has an unaltered CDH1 (11–15). In this study, restricted to 22 CDH1-mutated ILC from patients with relapsed disease, we queried whether a next-generation sequencing (NGS)–based assay could identify novel and unanticipated targets of therapy for these patients with relapsed, metastatic, and therapy-resistant disease.

We reviewed the records of a series of 308 invasive breast cancers submitted for NGS-based diagnostic testing at Foundation Medicine, Inc. and identified 22 CDH1-mutated ILCs all of which had relapsed after primary surgical and one or more systemic treatment approaches. NGS was conducted on hybridization-captured, adaptor ligation–based libraries using DNA extracted from 4 formalin-fixed paraffin-embedded sections cut at 10 μm. The pathologic diagnosis of each case was confirmed on routine hematoxylin and eosin (H&E)–stained slides and all samples forwarded for DNA extraction contained a minimum of 20% tumor cells. DNA sequencing was conducted for 3,320 exons of 182 cancer-related genes and 37 introns of 14 genes frequently rearranged in cancer (1.14 million total base pairs) on indexed, adaptor-ligated, hybridization-captured libraries (Agilent SureSelect Custom Kit) and fully sequenced using 49-bp paired reads on the Illumina HiSeq 2000 to at an average depth of 877× and evaluated for genomic alterations including base substitutions, insertions, deletions, copy number alterations (CNA; amplifications and homozygous deletions), and select gene fusions/rearrangements as previously described (16). To maximize mutation-detection accuracy (sensitivity and specificity) in impure clinical specimens, the test was optimized and validated to detect base substitutions at a 5% or more mutant allele frequency (MAF) and indels with a 10% or more MAF with 99% or more accuracy. To validate base substitution detection, 2 pools of 10 normal cell lines were used each containing a total of 2,057 known base substitutions representing a broad range of allele frequencies. We compared the alterations detected with those expected from base substitutions present in individual cell line constituents as previously described (17–20). To validate indel detection, 28 tumor cell lines were used containing a total of 47 known somatic indel alterations in 22 genes to generate 41 pools of 2 to 10 cell lines each creating a test set of 227 indels spanning a wide range of allele frequencies and indel lengths (17–20). To validate CNA detection, 7 tumor cell lines bearing 19 focal gene amplifications (6–15 copies, 15 genes) and 9 homozygous gene deletions (6 genes) were pooled with their matched normal cell lines (thereby maintaining consistent genotypes) in 5 ratios ranging from low to high tumor content (20%–75%), creating a test set of 210 CNAs (17–20).

CNAs had a validated accuracy of more than 95%. Actionable genomic alterations were defined as those identifying anticancer drugs in the market or available in registered clinical trials, the mechanism of action of which was predicted to be relevant based on the genomic alteration identified. Local site permissions to use clinical samples were used for this study.

The mean age of the female patients in this study was 56 years (range, 44–74). Tissue samples obtained from 22 cases of ILC originate from 22 unique patients. Either the primary tumor or a recurrent/metastatic tumor sample was available for each patient and histologically graded as follows: 1 (5%) grade 1, 16 (73%) grade 2, and 5 (23%) grade 3 tumors. Six ILC (27%) were stage III and 16 (73%) were stage IV at the time of sequencing (Table 1). Of the cases for which hormone receptor and HER2 slide-based test results were available, 15 of 19 (88%) were ER-positive; 11 of 18 (61%) were PR-positive; and 1 of 19 (5%) was HER2-positive by either IHC or FISH. Sequencing was conducted on the primary ILC in n = 9 (41%) cases and biopsies of metastatic sites in n = 13 (59%) cases including lymph node metastases in n = 3 (14%) cases, liver metastases in n = 3 (14%) cases, pleural fluid cell blocks and bone metastases in n = 2 (9%) cases each, and brain metastases in n = 1 (5%) cases.

Among the 22 ILCs, we identified 75 genomic alterations with an average of 3.4 alterations per tumor (range, 1–6), of which 35 were considered actionable for an average of 1.59 actionable alterations per patient (range, 0–3; Fig. 1; Table 1). Nineteen of 22 (86%) of the ILC harbored at least 1 actionable alteration potentially associated with clinical benefit of targeted therapies.

As a required study entry criterion, all 22 (100%) of the ILC featured a mutation in CDH1. There was a wide variety of CDH1 base substitutions and splice site mutations identified with no repeated pattern of sequence alteration in any of the 22 CDH1-mutated ILC cases in this study (Table 1).

Six (27%) of the CDH1-mutated ILC cases featured alterations in ERRB2 including 4 (18%) with ERBB2 mutations, 1 (5%) with an ERBB2–GRB7 gene fusion, and 1 (5%) with an ERBB2 copy number gain (amplification). There was a trend for high-grade tumors to harbor ERBB2 alterations with 3 (60%) of 5 grade 3 tumors having either ERBB2 mutations or amplification. Of the 9 primary tumors evaluated, 4 (44%) had an ERBB2 alteration (3 cases with mutations and 1 case with amplification) compared with 2 (15%) of 13 metastatic cases (1 case with an ERBB2 mutation and 1 case with an ERBB2 fusion). The 4 cases with ERBB2 mutations included study case number 7, a grade 3 tumor with a L755S ERBB2 mutation in a liver metastasis sample from an ER⁻/PR⁻/HER2 FISH⁻ 63-year-old patient (Fig. 2A). This tumor also featured an amplification of ESR1 as well as amplification in MYC and mutations in RUNX1 and TP53. Case study number 9, a primary tumor sequence from an ER⁻/PR⁻/HER2 IHC⁻ ILC harbored 2 ERBB2 mutations: a A775_G776insYVMA mutation and a R1153* truncation and no additional genomic alterations (Fig. 2B). Case study number 17 showed a P780_Y781insGSP ERBB2 mutation in the primary tumor from a 65-year-old patient. In addition to the ERBB2 mutation, this tumor also featured a splice site mutation in the MAP2K4 gene and amplification of MCL1 and RICTOR, respectively (Fig. 2C). In case number 22, a woman with relapsed metastatic ILC, her primary tumor was ER⁺/PR⁻/HER2⁻ (FISH). When this primary tumor was sequenced, 2 discrete ERBB2 mutations were found (S310F and V842l) as well as mutations in the MEN1 and SMAD4 genes (Fig. 2D). In case study number 18, a ER⁺/PR⁺/HER2⁻ bone metastasis from a 62-year-old patient, a novel ERBB2–GRB7 fusion was identified (Figs. 2E and 3). This tumor also had an amplification of the CCND1 gene and a H1047R PIK3CA mutation. There was insufficient archived extracted DNA or tumor tissue remaining on this sample for further DNA and RNA-based characterization of this gene fusion. Finally, in case study number 21, a CDH1-mutated stage IV high-grade ER⁻/PR⁻ pleomorphic breast cancer with papillary features, an ERBB2 gene copy number gain was detected by NGS, which correlated with the positive HER2 overexpression on immunohistochemical staining and gene amplification detected by FISH testing of the same specimen. This tumor also had amplifications in the CCND1 and MCL1 genes and a TP53 mutation. Only this ERBB2 amplification case also had a positive HER2 slide-based test result (both IHC and FISH) and the 5 ILC cases with either a mutation or fusion of ERBB2 were uniformly HER2-negative by IHC and/or FISH testing.

Although all ILCs in this study ultimately relapsed, 9 of the ILCs sequenced in this study were archived specimens of the primary disease. The incidence of genomic alterations in the primary tumors and metastatic site biopsies were virtually identical. When the original primary tumor was used for sequence analysis in this cohort of advanced and metastatic ILC, the 9 tumors yielded 31 total genomic alterations and 3.44 alterations per tumor (Fig. 1). In comparison, the 13 ILC cases where a metastatic site provided the sample for the NGS assay yielded 43 total alterations and 3.31 alterations per tumor. Interestingly, the incidence of actionable alterations was higher in the primary tumor sample group (2.0 actionable alterations per tumor) than in the metastatic site biopsy group (1.23 actionable alterations per tumor). However, given the relatively small numbers of cases in these 2 groups, this observation is preliminary and requires validation on a larger series of cases.

Additional genomic alterations identified in the cohort included mutations in PIK3CA (8 cases; 36%), TP53 (6 cases; 27%), RB1 (2 cases; 9%), KRAS (2 cases; 9%), AKT1 (2 cases 9%), and ERBB4 (1 case; 5%). The PIK3CA mutation frequency was higher in the 9 cases where the primary tumors were sequenced (56%) than in the metastatic site biopsy samples (23%), whereas the TP53 mutation frequency was higher in the metastatic lesions (31%) than for the cases where the primary tumor samples (11%) was used for genomic analysis. Mutations in PIK3CA and TP53 have been reported in 41% and 50% of ILC, respectively [Catalogue of Somatic Mutations in Cancer (COSMIC); Oct 2012]. Whether the observed differences between the 2 groups are significant requires validation on a larger series of cases. DNA copy number gains were seen in CCND1 (6 cases; 27%), FGFR1 (3 cases; 14%), MYC (2 cases; 9%), and ESR1 (1 case; 5%).

Finally, in order to assess clonality, the mutant allele frequencies of the 43 reported mutations in the 4 most commonly mutated genes in the study, CDH1, TP53, PIK3CA, and ERBB2, were assessed. Considering tumor suppressors (CDH1, TP53) and oncogenes (PIK3CA, ERBB2) separately, we observed mean mutant allele frequencies of 39.7% and 18.5%, respectively (Supplementary Table S2). Taken in the context of an average estimated tumoral purity of approximately 40% (range, 10%–95%) in the study, these mutant allele frequencies are consistent, on average, with a clonal homozygous model in the case of tumor suppressor mutations and the clonal heterozygous model in the case of oncogene mutations.

To define a more genetically homogenous population of ILC cases, patient entry into this study required the presence of a CDH1 mutation (7, 21). Using this approach, we found that a striking 19 of 22 (86%) ILC cases had one or more actionable alterations including changes in PIK3CA, CCND1, ERBB2, FGFR1, MCL1, KRAS, NF1, AKT1, and BRCA2. The incidence of CDH1 mutations in ILC varies significantly in public databases and literature from a low of 31% (COSMIC; July 2012) to 62% (22). Most investigators consider a mutation in CDH1 to represent a genomic alteration characteristic of the lobular subtype of breast cancer (7, 21). Other mechanisms that lead to loss of E-cadherin function include methylation of CDH1 (15). The CDH1 mutations in the 22 ILC cases included in this study were all unique and involved point mutations in 20 (91%) cases and splice site mutations/deletions in 2 (9%) cases. In triple-negative IDC of the breast, loss of E-cadherin expression has been linked to adverse clinical outcome (23). Finally, neither CDH1 mutation nor loss of E-cadherin function by other mechanisms, are considered targets for anticancer therapy at this time (24). Six (27%) of the CDH1-mutated ILC cases featured alterations in ERRB2 including 4 cases with mutations (L755S in 1 case, A775_G776 and insYVMA in 1 case, R1153* in 1 case, P780_Y781insGSP in 1 case, and S310F and V842l in 1 case), 1 case with an ERBB2–GRB7 fusion and 1 case with ERBB2 amplification. For ILC, HER2 gene amplification and protein overexpression are generally considered to be rare events typically below a 5% incidence in most published studies (25–27). ERBB2 mutations are rare in breast cancer and were not listed in a recently published major DNA sequence analysis study of more than 100 cases that also included whole-genome analysis in 22 cases (13). In the COSMIC database, 14 (1.52%) of 919 listed breast cancers have an ERBB2 mutation that alters the Her2 protein (COSMIC; Jan 2013). Similarly, in a currently unpublished prior assessment of a database of 308 clinical breast cancer samples of all types sequenced at Foundation Medicine, Inc. from which the 5 ILC with ERBB2 mutations that are also included in the current study were identified, only 10 (3.25%) total invasive breast cancers (5 ductal and 5 CDH1 mutated lobular) featured ERBB2 mutations. When the 5 of 22 (23%) CDH1-mutated ILC with ERBB2 mutation/fusion are compared with the 5 of 286 (2%) of the CDH1 WT (nonmutated) invasive breast cancers with ERBB2 mutations, this difference is statistically significant (P = 0.0006). When the ERBB2–GRB7 fusion case is excluded from analysis and only ERBB2-mutated cases are included, the enrichment of ERBB2 alterations in CDH1-mutated ILC remains significant (P = 0.003). Thus, the filtering of breast cancers by combining lobular histologic phenotype with CDH1 mutation significantly enriches for the presence of ERBB2 mutations. Given that the 5 ERBB2-mutated CDH1 WT tumors did not display a lobular histologic phenotype, these data indicate that it is unlikely that subsets of ILC in which E-cadherin is inactivated by other mechanisms such as methylation are also enriched for ERBB2 mutations.

Interestingly, ERBB2 mutations have been reported more frequently in lung cancer than in breast cancer reaching as high as 10% in the adenocarcinoma subtype (28, 29). The mutational spectrum of ERBB2 in breast cancer identified in this study is distinct from the mutations described for lung cancer, which has also recently been reported by others (30). The L755S mutation identified in case 7 of this series of ILC is located in the kinase domain of Erbb2, and has been shown to be an activating mutation, capable of inducing oncogenic transformation in cell culture (31, 32). Mutations in exon 20 of ERBB2, such as A775_G776insYVMA found in case 9 of this study, have been associated with Erbb2 activation and sensitivity to Erbb2 inhibitors and dual inhibitors of Erbb2 and Egfr (29, 33–36). Preclinical studies have shown that the L755S ERBB2 mutation is resistant to the reversible dual Egfr/Erbb2 inhibitor lapatinib but may be sensitive to irreversible dual Egfr/Erbb2 inhibitors, such as afatinib, which are under investigation in clinical trials (37). The P780_Y781insGSP mutation is located at exon 20 of ERBB2 and has also been linked to Erbb2 activation and sensitivity to Erbb2 inhibitors and dual inhibitors of Erbb2 and Egfr (33–36). The monoclonal antibodies trastuzumab and pertuzumab have been approved for use in ERBB2-amplified breast cancer, but their efficacy in the context of the L755S mutation is unknown.

The ERBB2–GRB7 putative fusion seen in study case 18 has not been previously reported. The fusion retains the kinase domain of ERBB2 (uniprot.org), which suggests that it could result in ERBB2 activation. The 17q12-21 amplicon, which includes both ERBB2 and GRB7 is frequently amplified in breast cancer and preclinical studies suggest that it may be a recombination hotspot (38). An expression screening study has reported that GRB7 can function as an ERBB2-dependent oncogene (39). Interestingly, 2 previous studies focused on breast cancer rearrangements did not report this fusion although the number of CDH1-mutated lobular cancers in these studies may have been limited (40, 41). GRB7 encodes an adaptor protein that interacts with ERBB2 and has been shown in a preclinical study to enhance its transformative capacity and increase ERBB2 phosphorylation in fibroblasts (41). In that the sample that yielded the ERBB2–GRB7 fusion was a small bone metastasis biopsy, there was insufficient remaining material available to further characterize the alteration at the RNA level. However, this fusion protein has not been functionally characterized, and its effect is therefore unknown and the relevance to the efficacy of clinical use of anti-HER2–targeted therapies is unclear.

The concept of targeting of ERBB2 mutations for anti-HER2 therapies has recently been expanded to breast cancer (30). Investigators identified 25 patient samples with ERBB2 mutations from a series of 8 independent breast cancer-sequencing projects. Using in vivo kinase assays, analysis of protein structure and animal models, the investigators were able to identify 7 activating ERBB2 mutations, G309A, D769H, D769Y, V777L, P780ins, V842I, and R896C that were sensitive to the kinase inhibitor neratinib (30). The P780_Y781insGSP mutation of ERBB2 in study case 17 is included in this list of activating ERBB2 alterations. Neither the A775_G776insYVMA, R1153*mutation identified in case 9 nor the novel ERBB2–GRB7 fusion from study case 18 were reported by these investigators. The L755S mutation found in case 7 of the current report was considered not to be an activating mutation by these investigators but conferred resistance to the drug lapatinib in their experimental systems. The authors concluded that ERBB2 (HER2) somatic mutation was capable of activating the gene in breast cancer and concluded that ERBB2 somatic mutations are potential drug targets for breast cancer treatment (30).

Several questions are raised by this study. Are all lobular breast cancers, or just CDH1-mutated ILCs enriched for ERBB2 mutations? If this relationship holds true, other CDH1-mutated cancers such as diffuse gastric carcinoma may also have a disproportionate enrichment of ERBB2 mutation. Moreover, the biologic mechanism for this apparent enrichment of ERBB2 alterations in CDH1 mutated ILCs is not known at this time. Interestingly, however, both the loss of cell adhesion associated with CDH1 mutation and the well-documented enhancement of cell motility attributed to ERBB2 activation (42, 43) could conceivably synergize to drive invasion and metastasis in these patients with difficult to treat, relapsed ILC. Finally, is ERBB2 mutation simply associated with relapsed breast cancer, rather than specifically being enriched in relapsed ILC? At this time, is not known whether a similar enrichment of ERBB2 alterations in CDH1-mutated ILC would be seen in cases that were successfully treated in the primary setting and, unlike the ILC in the current study, did not relapse or metastasize. This may explain whether ERBB2 mutation is typically present at the early stages of cancer for the patient or is an additionally acquired mutation as the cancer evolves, perhaps under selection pressure. Understanding the chronologic nature of this mutation could also explain distinct spectrums of ERBB2 mutation between lung and breast cancer.

Clinical trials focused on ERBB2 mutations and fusions in breast cancer using both on the market antibody therapeutics and tyrosine kinase inhibitors as well as new agents in late stages of development seem highly warranted. In conclusion, deep sequencing of genomic DNA using an optimized clinical grade diagnostic assay can provide a broad cancer-related gene survey at a depth of coverage that provides sensitive detection for all classes of cancer-related genomic alterations, and when applied to patients with ILC can reveal actionable alterations that have the potential to inform treatment decisions in the majority of patients. Wider use of this testing strategy could accelerate accrual to therapeutic trials of agents targeting these (and other mutations) and allow more efficient identification of more active single agents or combinations.

J.S. Ross, K. Wang, G. Otto, S.R. Downing, J. Sun, J. He, J.A. Curran, S. Ali, R. Yelensky, D. Lipson, G. Palmer, V.A. Miller, and P.J. Stephens are employees and stock owners of Foundation Medicine, Inc. V.A. Miller is a consultant/advisory board member of Boehringer Ingelheim. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.S. Ross, R. Yelensky

Development of methodology: J.S. Ross, G. Otto, S.R. Downing, R. Yelensky

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.S. Ross, C.E. Sheehan, A.B. Boguniewicz, S.R. Downing, J.A. Curran

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.S. Ross, K. Wang, A.B. Boguniewicz, S.R. Downing, J. Sun, J. He, J.A. Curran, R. Yelensky, D. Lipson, V.A. Miller, P.J. Stephens

Writing, review, and/or revision of the manuscript: J.S. Ross, K. Wang, C.E. Sheehan, A.B. Boguniewicz, S.R. Downing, S. Ali, R. Yelensky, G. Palmer, V.A. Miller

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.S. Ross, C.E. Sheehan, P.J. Stephens

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.