Genetic heterogeneity and actionable mutations in HER2-positive primary breast cancers and their brain metastases

Brain metastases constitute a challenge in the management of patients with HER2-positive breast cancer treated with anti-HER2 systemic therapies. Here we sought to define the repertoire of mutations private to or enriched for in HER2-positive brain metastases. Massively parallel sequencing targeting all exons of 254 genes frequently mutated in breast cancers and/or related to DNA repair was used to characterize the spatial and temporal heterogeneity of HER2-positive breast cancers and their brain metastases in six patients. Data were analyzed with state-of-the-art bioinformatics algorithms and selected mutations were validated with orthogonal methods. Spatial and temporal inter-lesion genetic heterogeneity was observed in the HER2-positive brain metastases from an index patient subjected to a rapid autopsy. Genetic alterations restricted to the brain metastases included mutations in cancer genes FGFR2, PIK3CA and ATR, homozygous deletion in CDKN2A and amplification in KRAS. Shifts in clonal composition and the acquisition of additional mutations in the progression from primary HER2-positive breast cancer to brain metastases following anti-HER2 therapy were investigated in additional five patients. Likely pathogenic mutations private to or enriched in the brain lesions affected cancer and clinically actionable genes, including ATR, BRAF, FGFR2, MAP2K4, PIK3CA, RAF1 and TP53. Changes in clonal composition and the acquisition of additional mutations in brain metastases may affect potentially actionable genes in HER2-positive breast cancers. Our observations have potential clinical implications, given that treatment decisions for patients with brain metastatic disease are still mainly based on biomarkers assessed in the primary tumor.


HER2 immunohistochemistry and fluorescence in situ hybridization
Selected primary breast cancers and brain metastases with available tumor tissue were subjected to re-assessment of HER2 status using IHC (anti-HER-2/neu (4B5) Rabbit Monoclonal Primary Antibody kit, Ventana Medical Systems) and/or FISH. FISH for HER2 was performed on 4 μm-thick sections using the Food and Drug Administration (FDA)-approved Vysis PathVysion HER-2 DNA Probe Kit (Abbott Molecular Inc, Des Plaines IL, USA) according to manufacturer's instructions. Normal tissues including vessels, fibroblasts, lymphocytes or nontumor breast tissues served as internal controls. Tumor tissues were evaluated using a 100× objective.

Targeted capture massively parallel sequencing
Tumor and normal DNA samples were subjected to targeted capture massively parallel sequencing at the Memorial Sloan Kettering Cancer Center (MSKCC) Integrated Genomics Operation (IGO). In this study, we employed a previously described customized breast cancer panel targeting all exons of 254 genes recurrently mutated in breast cancer and DNA repair-related genes (Supplementary Table 1) using custom oligonucleotides (NimblegenSeqCap) as described [2,3]. Barcoded sequence libraries were prepared (New England Biolabs, KapaBiosystems) using 50 ng-250 ng of DNA. Massive parallel sequencing was performed on an Illumina HiSeq2000 (San Diego, CA), and bioinformatics analyses were performed as previously described [2,3]. Allele-specific copy number alterations (CNAs) and loss of heterozygosity (LOH) of the wild-type allele in genes harboring a somatic mutation were inferred from massively parallel sequencing data using FACETS [4].

Validation of mutations by amplicon sequencing
Selected mutations found by targeted sequencing (n = 108, consisting of 104 unique mutations) were subjected to orthogonal validation using amplicon resequencing in all samples for a given patient, where genomic DNA was available in the MSKCC IGO (Supplementary Table 2). For amplicon resequencing, 5 ng of genomic DNA was amplified with primers specific for a given mutation using the AmpliTaq gold 360 master mix (Invitrogen Life Science Technologies). Amplicons were bead-purified, quantified and pooled. Pooled amplicons were subjected to standard protocol of Illumina library preparation sequencing (MiSeq). Reads were aligned to the reference human genome GRCh37 using BWA (v0.6.2) [5], and local realignment was performed using GATK (v3.1.1) [6]. Pileup files were generated using SAMtools [7]. Mutations with > 1% of mutant allele frequency (MAF) and covered by at least 50 reads were considered validated. The validation rate of the somatic mutations with sufficient coverage was 93% (100/108). False-positive variants were excluded from further analyses. Given the high validation rate, untested mutations were included in further analyses. In addition, of the 70 mutations that were not detected in a sample of a given case based on the initial targeted sequencing, 11 (16%) were subsequently detected by amplicon sequencing and were included in the analyses, with the remaining confirmed to be genuinely absent even at a median depth of 10564× (range 272×-49532×).

Inference of cancer cell fraction (CCF)
ABSOLUTE (v1.0.6) [8] was used to infer the cancer cell fraction (CCF) of each mutation using the number of reads supporting the reference and the alternate alleles and the segmented Log 2 ratio from targeted capture massively parallel sequencing. A mutation was classified as clonal if its probability of being clonal was > 50% [9] or if the lower bound of the 95% confidence interval of its CCF was > 90%. Mutations that did not meet the above criteria were considered subclonal.

Identification of potentially pathogenic mutations
A combination of MutationTaster [10], CHASM (breast) [11] and FATHMM [12] was used to define the potential functional effect of each missense somatic single nucleotide variant (SNV). Missense SNVs defined as non-deleterious/passenger by both MutationTaster [10] and CHASM (breast) [11], a combination of mutation function predictors shown to have a high negative predictive value [13], were considered likely passenger alterations. The remaining missense SNVs were defined as likely pathogenic if they were predicted to be "driver" and/ or "cancer" by CHASM (breast classifier) and/ or FATHMM [12], respectively. Frameshift, splice-site and nonsense mutations were considered likely pathogenic if they were targeted by loss of the wild-type allele or affected haploinsufficient genes [14]. SNVs, including missense and nonsense SNVs, affecting hotspot residues [15] were also considered likely pathogenic and were separately annotated. Mutations were also annotated if they affected genes included in the cancer gene lists described by Kandoth et al. (127 significantly mutated genes) [16], the Cancer Gene Census [17] or Lawrence et al. (Cancer5000-S gene set) [18]. Mutations that were neither likely pathogenic nor likely passenger were considered of indeterminate pathogenicity.

Actionable somatic genetic alterations
The DGIdb database [19] defines a list of 402 known and potential actionable genes (Supplementary Table 5). The clinically actionability of DGIdb is defined based on Bader Lab Genes, Caris Molecular Intelligence, Foundation One Genes, GO, Guide To Pharmacology Genes, Hopkins Groom, MSK-Impact, Russ Lampel and dGene.
OncoKB [20] is a comprehensive and curated precision oncology database with more than 3,000 unique mutations, fusions, and copy number alterations in 418 cancer-associated genes. It annotates the biologic and oncogenic effects, and provides the level of evidence that a specific molecular alteration is predictive of drug response on the basis of US Food and Drug Administration labeling, National Comprehensive Cancer Network guidelines, disease-focused expert group recommendations, and scientific literature.