Identification of Recurrent Activating HER2 Mutations in Primary Canine Pulmonary Adenocarcinoma

This article is featured in Highlights of This Issue, p. 5727

Naturally occurring primary canine lung cancer is clinically challenging (1), with a disease course and underlying biology that resemble human lung cancer in never-smokers. Human never-smoker lung cancer accounts for 10% to 25% of lung cancers, causes approximately 26,000 deaths annually, and has a high incidence of erb-B family gene mutations such as those affecting EGFR. Although the incidence of smoking-related lung cancer is decreasing, lung cancer incidence in never-smokers is increasing (2). Never-smoker lung cancer is primarily non–small cell lung cancer (NSCLC), arising from lung tissue, as opposed to small cell lung cancer arising in bronchi of smokers. NSCLC histologies include adenocarcinoma (AC) and squamous cell carcinoma (SCC). The etiology of never-smoker lung cancer is also distinct from that of smokers. It is associated with factors including environmental exposures (secondhand smoke, radon, asbestos, arsenic, silica, and pollution) as well as age, sex, family history, and genetic loci (3). Unique genomic characteristics of human never-smoker lung cancer include low somatic mutation burden, enrichment for C:G>T:A transitions, and somatic activating point mutations or fusions affecting EGFR (45%), ALK (5%–11%), ROS (1.5%–6%), HER2 (3%–5%), and RET (2%; ref. 4). The five-year overall survival is estimated at 23%, but outcomes are dependent on molecular subtype and treatment regimen. For example, EGFR inhibitors can improve outcomes in EGFR-mutant lung cancers; however, 85% of never-smoker lung AC and SCC cases are EGFR wild-type (WT) in the United States. Clinical trials of immune-checkpoint inhibitors have recently shown improved outcomes for human lung cancers, but analysis of large phase II immunotherapy trials suggests that benefits are limited in low-mutation-burden (≤10 mutations/Mb) cases such as those found in never-smokers (5). A need exists for improved biological understanding and development of new models to fuel translational research in never-smoker lung cancer.

Lung cancer in pet dogs has limited standard of care beyond surgery, and little is known of its molecular underpinnings (1). Primary lung tumors typically arise in older dogs (11 years) and resemble human NSCLC histotypes including canine pulmonary adenocarcinoma (cPAC), adenosquamous carcinoma (cPASC), and SCC (cPSCC). These subtypes collectively represent 13% to 15% of primary lung tumors (6, 7). Patients are often diagnosed late with lesions incidentally discovered during routine geriatric evaluation or due to nonspecific symptoms including dyspnea (6%–24%) and cough (52%–93%) that do not manifest until the tumor is more than 3 cm. The detection of canine lung cancers has significantly increased over the past 20 years not only because of improved animal healthcare and diagnostics, but also possibly due to increased companion animal exposures to pollutants. These tumors can be diagnostically challenging. Rates at which ultrasound or CT-guided fine-needle aspirates of the pulmonary mass provide cytologic diagnosis range from 38% to 90% of cases, varying broadly based on tumor accessibility and aspirate quality. At diagnosis, 71% of malignant canine lung tumors show signs of invasion and 23% show distant metastasis. Partial or complete lung lobectomy is standard of care, dependent on the extent of disease spread. Median survival is 345 days for localized disease without nodal involvement where surgical remission can be achieved, but only 60 days when nodes are involved. Responses to cytotoxic chemotherapy (cisplatin, vindesine, doxorubicin, and mitoxantrone) in the setting of disseminated disease are limited. Targeted small molecules and immune-checkpoint inhibitors have not been extensively studied in part because the molecular underpinnings of canine lung cancer remain largely unknown. In naturally occurring canine NSCLC, although comprehensive genomic profiling has been limited, KRAS hotspot mutation prevalence estimates from targeted studies have varied from 0% to 25% (7–9). We have previously shown that EGFR mutation, overexpression, or phosphorylation is rare in cPAC compared with matched nonaffected chemotherapy-naïve lung tissue whereas significant overexpression and/or phosphorylation of PDGFRα, ALK, and HER2 are present (10). We now describe for the first time the detailed genetic underpinnings of primary canine lung cancers through multiplatform next-generation sequencing of 88 cases.

Tumors and cell lines from 89 dogs from the Canine Comparative Oncology and Genomics Consortium (CCOGC; ref. 11) and The Ohio State University (OSU) College of Veterinary Medicine Biospecimen Repository were included. Veterinary pathologists board certified by the American College of Veterinary Pathologists (ACVP) confirmed tumor diagnosis based on histopathology. This study was approved by The OSU Institutional Animal Care Use Committee (2010A0015-R2). Tumor and normal tissue samples were flash frozen in liquid nitrogen or formalin-fixed and paraffin-embedded (FFPE). Cell lines were received between May 29, 2017 (all sequenced lines), and May 18, 2018 (OSUK9PAPADRiley), maintained in RPMI-1640 with GlutaMAX (Gibco, Thermo Fisher Scientific, #61870036) supplemented with 10% heat-inactivated fetal bovine serum at 37°C and 5% CO₂ and passaged at 90% confluence. Cell lines with known passage data (except BACA) were sequenced within the first eight passages of derivation, subsequently expanded for phenotypic studies, and authenticated by IDEXX BioResearch using the Cell Check Canine STR Profile and Interspecies Contamination Test on January 30, 2018, or by NkX2 (or TTF-1) RT-PCR. Mycoplasma testing was performed in all lines at time of arrival and every 3 months of continuous harvest by using the Mycoplasma Detection Kit (ATCC, #30-1012K). All samples tested negative at all times with final testing performed on June 21, 2019, for Riley and June 15, 2017, for all other lines). Human cell lines included BT474 (ATCC, #HTB20, HER2 focal amplification) and H1781 (ATCC #CRL-5894, HER2^G776insV_G/C). Blood for cell-free DNA and germline DNA extraction was collected in 10 mL K₂ EDTA Blood tubes (Thermo Fisher Scientific, #22-253-145). Plasma separation was performed at room temperature within 1 hour (2× serial centrifugation at 2000 rpm × 15 minutes). Plasma aliquots were stored frozen at −80°C. DNA extraction from plasma, white blood cells, and tissue was performed with MagMAX Cell-Free DNA Isolation Kit (Thermo Fisher Scientific, #A29319), DNeasy Blood and Tissue Kit (QIAGEN, #69504), and Qiagen AllPrep DNA/RNA Mini Kit (QIAGEN, #80204), respectively.

Informatic tools, versions, and flags are shown in Supplementary Table S1. We utilized a custom Agilent SureSelect canine exome capture kit with 982,789 probes covering 19,459 genes. Exome libraries were sequenced on the Illumina HiSeq2000 producing paired-end reads of 85 bp. FASTQ files were aligned to the canine genome (CanFam3.1) using BWA v0.7.8. Aligned BAM files were realigned and recalibrated using GATK v3.3.0, and duplicate pairs were marked with Picard v1.128 (http://broadinstitute.github.io/picard). Somatic copy-number variants (CNV) and structural variants (SV) were called with tCoNutT (https://github.com/tgen/tCoNuT) and DELLY v0.76, respectively. Somatic single-nucleotide variants (SNV) were identified from two or more of the following callers: Seurat v2.6, Strelka v1.0.13 and MuTect v1.1.4. Germline SNVs were called using Haplotype Caller (GATK v3.3.0), Freebayes, and samtools-Mpileup. Variant annotation was performed with SnpEff v3.5. The SomaticSignatures R package was used to identify mutation signatures.

We developed a custom canine cancer amplicon sequencing panel consisting of 281 amplicons targeting exons and hotspot regions in 57 genes, with amplicon sizes ranging from 91 to 271 bp (Supplementary Table S7). We pooled primers in two multiplexed pools to separate adjacent amplicons and any amplicons that showed cross-amplification using in silico PCR. We prepared sequencing libraries using digital PCR amplification following the manufacturer's protocols for the ThunderBolts Cancer Panel (RainDance Technologies) with modifications as previously described (12). Sequencing was performed on the Illumina MiSeq generating paired-end 275 bp reads. Sequencing reads were demultiplexed and extracted using Picardtools. Sequencing adapters were trimmed using ea-utils, and fastq files were assessed for quality using FASTQC. Sequencing reads were aligned to CanFam3.1 using bwamem-MEM (13). Custom in-house scripts based on samtools were used to create pileups for every sample. Pileups were analyzed in R to call SNVs and indels. For each potential nonreference allele at each targeted locus in a sample, we evaluated the distribution of background noise across all other sequenced samples. To call a variant, we required the observed nonreference allele to be an outlier within the background distribution with a Z-score > 5. In addition, we required tumor depth ≥100×, allele frequency ≥10%, number of reads supporting the variation ≥10, and allele fraction in the germline sample <1%. Finally, variant calls were manually curated by visualization in IGV v2.3.71. All next-generation sequencing data (exome and amplicon) have been deposited in the NCBI Sequence Read Archive repository under accession number PRJNA523624.

Twenty-three primer pairs covering all exons of HER2 were designed using Primer 3 (http://bioinfo.ut.ee/primer3-0.4.0/) including a universal M13 tag. Amplicons were Sanger sequenced at the DNASU sequencing facility at Arizona State University on an ABI 3730XL (Applied Biosystems) and analyzed with Mutation Surveyor DNA Variant Analysis Software (SoftGenetics).

The HER2 inhibitors lapatinib (Selleckchem, #S2111), neratinib (Puma Biotechnology, Los Angeles, CA), and trastuzumab (Selleckchem, #A2007), as well as the EGFR inhibitor erlotinib (Selleckchem, #S1023), were assessed in 10–16-point 72 hours drug-dose response screens (from 2 × 10⁻⁷ nmol/L to 100 μmol/L for small molecules and from 1 × 10⁻⁶ μg/mL to 1,000 μg/mL for trastuzumab) with CellTiter-Glo luminescent cell viability assay (Promega, #G7570) endpoints. Luminescence was read using a Synergy Mx (Biotek) plate reader. Six replicates were performed for each dose. Curve-fitting and IC₅₀ calculations were performed using GraphPad Prism v7.00 (GraphPad Software).

HER2^V659E genotyping was performed on tumor samples and plasma cell-free DNA with droplet digital PCR (ddPCR). PCR amplification was performed as follows: 1 cycle 3 minutes at 95°C, 50 cycles 15 seconds at 95°C and 1 minute at 60°C with a 0.5°C/second ramping from 95°C to 60°C, 1 cycle 98°C for 10 minutes and hold at 4°C. Droplet fluorescence was measured using RainDrop Digital PCR Sense instrument and analyzed using accompanying RainDrop Analyst II Software v.1.0 (RainDance). Primer and probe sequences used for HER2^V659E detection in ctDNA were forward: 5′-CCCACGACCACAGCCA-3′, reverse: 5′-CCCTGTGACATCCATCATTGC-3′, and probe: 5′-CAGAATGCCC(T/A)CCACAGC-3′.

cDNA was obtained by reverse transcription with iScript (Bio-Rad, #1708891) and samples were subjected to HER2 (target) and HPRT1 (reference) amplification in a QuantStudio 6 Flex Real-Time PCR System under standard conditions with Syber Green technology (KapaBiosystems, #KK4602). Primer sequences were: HER2-forward: 5′-CATCTGCACCATTGATGTCTA-3′, HER2-reverse: 5′-GGCCCAAGTCTTCATTCTGA-3′, HPRT1-forward: 5′GCAGCCCCAGCGTCGTGATT-3′, HPRT1-reverse: 5′CATCTCGAGCAAGCCGCTCAGT-3′. Data were analyzed with Quantstudio Real-Time PCR software v1.1. Values for ΔCt, ΔΔCt, and fold changes were calculated as follows: ΔCt = Ct HER2 − Ct HPRT1; ΔΔCt = ΔCt tumor sample – ΔCt average of normal samples; and fold change = 2^(−ΔΔCt).

HER2 protein expression was evaluated on FFPE sections (4 μm) of normal lung and tumor mounted on SuperFrost Plus glass slides (Fisher Scientific, #12-550-15). Slides were deparaffinized in xylene and rehydrated in an ethanol gradient. Antigen retrieval was performed with 1 mmol/L EDTA adjusted to pH 9.0. An autostainer (Dako, model S3400) was used to carry out immunostaining. A HER2 rabbit monoclonal antibody (Cell Signaling Technology, #4290) was used at 1:400 dilution followed by secondary biotinylated rabbit anti-goat IgG (Vector Laboratories, BA-1000) diluted 1:200. Detection was performed with VECTASTAIN Elite ABC System (#PK-6100). IHC-positive controls for HER2 tyrosine kinase receptor expression were single tissue samples of two canine complex mammary carcinomas (14). Negative controls were performed on all tissues using a universal rabbit negative isotype control not directed against any known antigen (Dako, #IR600).

Immunostained and control 1 × 3-inch microscope slides were scanned at 40× on a high-resolution Scanscope XT (Leica Biosystems) at The OSU Comparative Pathology and Mouse Phenotyping Shared Resource. For the quantification of immunoreactivity, images were imported into Visiopharm Image Analysis software (Visiopharm, Hørsholm, Denmark version 2017.27.0.3313), segmented into areas of tumor, necrosis, and normal lung tissue using color labels for each tissue type. HER2 connectivity was scored using the modified 10007-HER2, Breast Cancer APP (Visiopharm). Thresholds were adjusted to match specimen HER2 stain intensities for accurate scoring. Area (μm²) was quantified for each tissue type, and percentages were derived from specimen total tissue area. Tumor areas were further segmented into staining and nonstaining categories, and their percentages were calculated based on the total tumor area in μm². Maximum, mean, and minimum intensities were also quantified using a built-in software calculation. Staining is expressed as a percentage of stain present with 100% equal to black (maximum dark brown) and 0% equal to white (no stain present). Initial thresholds and tissue types were established, and mark-ups were reviewed in consultation with a pathologist board-certified by the ACVP to ensure accurate measurements and to differentiate between tissue types.

Subconfluent cells were serum starved overnight, and then treated with 20 nmol/L neuregulin for 15 minutes prior to harvest. Cells were lysed in RIPA buffer with cOmplete Mini Protease Inhibitor (Roche, #11836153001) and PhosSTOP (Roche, #4906845001) and loaded in Laemmli buffer at 1 μg/μL. Samples were separated on 4% to 15% SDS-PAGE Criterion Gels (Bio-Rad, #5671085) and transferred to Immobilon-FL PVDF membrane (MilliporeSigma, #IEVH7850). Membranes were blocked for 1 hour in LI-COR blocking buffer and incubated with primary antibody at 4°C overnight, followed by fluorescence-conjugated secondary antibodies. Membranes were scanned using the LI-COR Odyssey CLx instrument. Primary antibodies were AKT (CST #4691S, 1:1,000), phospho-AKT (CST #4060P, 1:1,000), and β-actin (CST #4970S, 1:1,000).

In order to map the genomic landscape of primary canine lung cancer, we undertook multiplatform next-generation sequencing of 88 NSCLC cases, including 78 tumor/normal matched pairs and 10 cell lines (Table 1). The cohort included 74 cPAC, 11 cPASC, and 3 cPSCC. Labrador retrievers represented the most commonly affected pure breed dog (21%) with mixed breeds (25%) and multiple other single pure breeds. The predominant cPAC subtype was papillary adenocarcinoma (62%). The cohort was gender balanced (52% females) and primarily neutered/spayed (92%) with a median age at diagnosis of 11 years. Smoking status in the pet's household was not available. Extended clinical annotation is shown in Supplementary Table S2 and Supplementary Fig. S1.

To identify somatic point mutations, copy-number changes, and translocations, we first performed exome sequencing of five cPAC tumors and matching normal samples with a mean target coverage of 298× and 263×, respectively (99% of target bases covered ≥ 40×, Supplementary Table S3). A total of 648 high-confidence somatic SNVs (median 64; range, 37–406), 165 focal CNVs (median 19; range, 0–116), and 3 SVs (median 1; range, 0–1) were identified (Supplementary Tables S4–S6; Fig. 1A–C). The average tumor mutation burden (TMB) for somatic point mutations per haploid callable megabase (Mb) in these cases was 2.04 mutations/Mb (range, 0.58–6.38). Among these somatic variants, we identified mutations in genes whose human orthologs have been implicated in human cancer according to COSMIC (15) tiers 1 and 2, including somatic SNVs (Supplementary Table S4), focal CNVs (Fig. 1B; Supplementary Table S5), SVs, and numerical chromosomal changes (whole chromosome or arm-level gains/losses; Supplementary Fig. S2). The sole gene bearing recurrent nonsynonymous SNVs was HER2 (80%), with the missense mutation V659E occurring in three cases and the missense mutation K676E in a fourth case (Fig. 1D). No HER2 amplifications were detected in these five tumors. We also assessed somatic point mutation signatures according to their trinucleotide context (Supplementary Fig. S3; refs. 16, 17). The most common signature in these five cases was the age-associated COSMIC Signature 1A in four of five (80%) and COSMIC signature 2, associated with APOBEC cytidine deaminase activity, in two cases.

To identify somatic point mutations across a broader cohort of canine lung cancers, we used a custom canine cancer amplicon-based sequencing panel (Supplementary Table S7) in 73 additional lung tumors (61 cPAC, 10 cPASC, and 2 cPSCC), two previously exome-sequenced tumors with matched normal tissue, and 10 cell lines (8 cPAC, 1 cPASC, and 1 cPSCC). These cases were sequenced to an average depth of 3,383× (Supplementary Table S8). A median of 1 somatic coding point mutation (range, 0–3) within sequenced panel regions was identified across all cases. Likely pathogenic recurrent point mutations included HER2 V659E (29.8%), KRAS G12D/V (3.4%), SMAD4 D351Y/G (3.4%), and TP53 R239Q/G (2.2%; Supplementary Table S9). Two additional somatic missense mutations in HER2 were identified in single cases (Fig. 1D). Overall, recurrently mutated genes containing somatic potentially pathogenic SNVs included TP53 (12.5%), PTEN (5.7%), SMAD4 (4.5%), KRAS (4.5%), VHL (3.4%) and HRAS (2.3%). Finally, based on both exome and amplicon sequencing, we evaluated germline SNPs to identify putatively pathogenic rare variants (i.e., those not previously identified in dogs based on review of presence in dbSNP 151, ref. 18; and/or ≥10% frequency in DogSD, ref. 19) in 81 genes potentially associated with susceptibility to human lung cancer (20). We identified nine rare putatively pathogenic SNPs in five dogs in the genes CHRNA3, CYP1B1, DNAH11, and HER2 (Supplementary Table S10). Of these SNPs, the only variant with an equivalent in its human orthologous gene was DNAH11 R1460W corresponding to human DNAH11 R1444W (rs1035326227, Minor Allele Frequency < 0.01%). The human SNP has not been associated with disease. HER2 V1189I variants occurred in two cases without somatic HER2 tumor mutations. The human orthologous position V1184 has not shown human variation. The canine variant has been identified in 4% of cases in DogSD and based on functional effect prediction (FATHMM), it is likely neutral. None of the genes bearing rare SNPs showed second hits in tumor tissue.

We additionally performed matched tumor/normal amplicon sequencing to evaluate the genomic landscapes of 11 cPASC and 3 cPSCCs, subtypes that are understudied entities in dogs and humans, especially in never-smokers (Fig. 1A; Supplementary Table S9). In contrast with cPAC, no HER2 mutations were identified in these tumors. In cPASC, HRAS Q61L and KRAS Q61K each occurred in one case. Thus, 18% of cases bore RAS hotspot mutations. PTEN stop gains additionally occurred in two of 11 (18%) of cases at high tumor allele frequencies and were exclusive with RAS mutations. Additional likely pathogenic somatic mutations also occurred in a single cancer gene in a single tumor each including EGFR A726T, MET M1269V, TP53 R147C, and VHL P97L. Finally, although no recurrent mutations were identified in the three cPSCCs, we identified one case with a somatic BRAF V588E and another bearing PTPN11 G503V.

HER2 was the most frequently mutated gene in our multiplatform next-generation sequencing cohort, with missense mutations occurring exclusively in cPACs (27/74, 36.5%, two mutations occurring in a single patient; Fig. 1A). No HER2 insertions were identified. We additionally identified a HER2 mutation in the cell line OSUK9PAPADRiley solely by Sanger sequencing of the codon 659 locus. We thus identified 29 total HER2 mutations overall (Fig. 1D). In 24 cases, the HER2 variants were evaluated on at least two platforms, including exome sequencing, amplicon sequencing, Sanger sequencing, and/or ddPCR (Supplementary Table S11). The HER2 variant tumor allele fraction (AF) median by amplicon sequencing was 21.3% (range, 8.4%–51.9%). All low AF (<20%) cases identified by amplicon sequencing were also validated by Sanger and/or ddPCR. Notably, one cell line, OSUK9PAPADOscar, contained a low AF HER2 V659E variant (AFs of 15% by amplicon and 16% by ddPCR) during early short-term culture (passage 4) that was no longer detectable by Sanger sequencing or ddPCR in later passages (passage 15). Importantly, passage 15 was utilized for all functional studies described below, and it was thus considered HER2^WT in this setting. Overall, V659E missense mutations located in the HER2 TMD occurred in 93.3% of HER2-mutant cases. Additional HER2 mutations included A664T (OSU419040) and K676E (CCB050354), which have not been previously described in orthologous human HER2 regions. In some cases, HER2 mutations co-occurred with mutations in TP53, SMAD4, PTEN, VHL, AKT1, or KDR.

Cell-free tumor DNA (ctDNA) in plasma has been increasingly used for noninvasive genotyping in human cancer patients (21). To develop a canine blood test that could rapidly identify dogs with HER2-mutant lung cancer, we investigated whether cPAC HER2 hotspot mutations are detectable in ctDNA. We evaluated plasma from 11 dogs—5 with HER2^WT tumors and 6 with HER2^V659E tumors—using ddPCR (Supplementary Table S12). In order to evaluate assay performance (specificity), we first analyzed WT tumor samples, plasma DNA from unrelated commercially available canine plasma samples, and template-free controls (BioreclamationIVT, #BGLPLEDTA-100-P-m). Using uniform gating for all experiments, we found one of seven template-free samples showed one WT droplet and no samples showed any evidence of mutant DNA amplification. In WT tumor and plasma DNA samples, two of eight samples showed one mutant droplet each. Based on these results, we required at least three mutant droplets to confidently detect HER2^V659E. To confirm mutation detection and quantitative performance, we analyzed and detected HER2^V659E in six of six positive control tumor DNA samples where we had previously identified V659E mutations using amplicon sequencing. In these samples, we observed a high correlation between AFs measured using amplicon sequencing and ddPCR (Pearson r 0.976, P = 0.0008; Supplementary Fig. S4). In 11 plasma samples from dogs with cPAC tested using ddPCR, the median total cell-free DNA (cfDNA) concentration was 3.7 ng/mL plasma (range, 0.7–23.0; Supplementary Table S12). Requiring at least three mutant droplets to support mutation detection and testing cfDNA equivalent to 435 μL plasma, the median limit of detection for mutation allele fraction was calculated at 0.61% (range, 0.10%–3.11%). HER2^V659E mutations were detected in two of six plasma samples from dogs with HER2^V659E-positive tumors at 1.9% and 2.3% allele fractions. HER2^V659E was not detected in any plasma samples from dogs with HER2^WT tumors, confirming assay specificity (Fig. 1E). Sensitivity for mutation detection in this cohort may be limited due to low total cfDNA concentration and amounts analyzed.

In human cancers, HER2 bears activating point mutations, copy-number gains/amplifications, and RNA and protein overexpression. Amplification and overexpression are typically mutually exclusive with point mutations. HER2 copy number was first determined in the five exome-sequenced cases (Supplementary Fig. S2; Supplementary Table S5). No numerical CFA9 gains or focal HER2 amplifications were detected. However, these cases predominantly bore somatic putatively activating point mutations and might not be expected to contain concomitant gains. Therefore, we evaluated HER2 RNA and protein expression by qRT-PCR and IHC. RNA samples from 49 lung tumors (nine HER2 mutant) were evaluated alongside 14 normal lung tissue samples distal to tumor areas but from the same lung lobe. Median HER2 expression fold change relative to expression of the housekeeping gene HPRT in normal lung samples was 1.06 (range, 0.28–4.11) and in tumors was 0.85 (range, 0.07–4.50; Supplementary Fig. S5). No significant difference in relative HER2 expression was observed between tumor and normal or HER2-mutant and HER2^WT groups.

Additionally, in order to quantify HER2 protein expression in cPAC, digital image analysis was performed on eight tumors from FFPE. Three of the samples bore the HER2^V659E hotspot mutations, one bore HER2^A664T, and four were HER2^WT. All cases were positive for HER2 staining with homogeneous and diffuse staining of tumor cell cytoplasm and cell membrane, but no staining in adjacent stroma or vasculature (Supplementary Fig. S6). Positive staining was observed in bronchial epithelium of the adjacent nonaffected lung in all cases. Consistent with the absence of observed HER2 amplifications, no significant differences (mean ± SEM) were detected in the tumor positivity percentage for HER2 (47 ± 5.4 and 35 ± 5.1) between the WT and HER2-mutant groups, respectively. No significant differences in HER2 staining were present for percent minimum (51 ± 2.9 WT vs. 55 ± 5.1, HER2 mutations) or percent maximum (97 ± 0.31 WT vs. 96 ± 0.47 HER2 mutations) stain intensity (Supplementary Table S13). Overall, most tumors showed moderate expression of HER2 based on qRT-PCR and IHC with some variability, but levels were typically consistent with those seen in normal tissue and did not vary based on HER2 mutation status.

HER2 is a transmembrane receptor tyrosine kinase typically activated by homodimerization or heterodimerization with other HER family receptors. HER2 mutations or overexpression drive constitutive downstream signaling. In human cancers, HER2 V659E mutations stabilize dimers to increase HER2 autophosphorylation, EGFR phosphorylation, activation of phosphatidyl-inositol-3-kinase (PI3K), and activation of mitogen-activated-protein-kinase (MAPK) prosurvival signaling pathway members (e.g., AKT and ERK) relative to WT HER2 (22, 23). To determine whether HER2^V659E constitutively activates downstream signaling in cPAC, we first validated HER2 genotype in seven canine lung cancer cell lines through amplicon sequencing, ddPCR, or Sanger sequencing of the V659 locus, Sanger sequencing of all HER2 coding regions, and/or aCGH to determine HER2 copy-number status as previously published (Supplementary Table S11; ref. 24). One cell line, OSUK9PAPADOscar, bore a low allele frequency HER2^V659E mutation when sequenced by an amplicon panel at low passage (passage 4) as a primary culture, but had lost this allele in later established passages (passage 15) characterized by Sanger sequencing and ddPCR. The latter passages were utilized for functional studies. We thus evaluated HER2 activation in one HER2^V659E cPAC cell line, OSUK9PAPADRex, and three HER2^WT cell lines (two cPAC: CLAC and OSUK9PAPAPADOscar; and one cPASC: OSUK9PADSQ) by Western blotting for total and phospho-AKT in the presence and absence of the ErbB ligand, neuregulin (hNRG) post-serum starvation. Only the HER2^V659E line OSUK9PAPADRex showed constitutively high AKT phosphorylation post-starvation even in the absence of hNRG stimulation (Fig. 2A).

To determine the potential efficacy of anti-HER2 agents for treatment of HER2^V659E cPAC, we performed dose–response studies of selected tyrosine kinase inhibitors (TKI: neratinib, lapatinib, and erlotinib) and a humanized HER2 recombinant monoclonal antibody (mAb), trastuzumab, which binds the extracellular juxtamembrane domain IV of HER2. We first assessed the differential sensitivity of HER2^V659E and HER2^WT canine lung cancer cell lines to lapatinib (HER2 and EGFR inhibitor) and neratinib (HER2, HER4, and EGFR inhibitor). Five cPAC cell lines (two HER2^V659E and three HER2^WT) were treated with neratinib and four (one HER2^V659E and three HER2^WT) with lapatinib for 72 hours. Two HER2-mutant human cancer cell lines—BT474 (HER2^AMP) and H1781 (kinase domain HER2^G776insV_G/C)—were treated as positive drug controls (refs. 22, 25; Fig. 2B; Supplementary Fig. S7). Significant differences in viability were observed between HER2^V659E and HER2^WT cPAC cell lines for both TKIs (IC₅₀ < 200 nmol/L in HER2^V659E vs. IC₅₀ > 2,500 nmol/L in HER2^WT). All HER2-mutant cell lines were sensitive to neratinib with IC₅₀ < 50 nmol/L (Fig. 2B), significantly lower than those observed for HER2^WT cells (IC₅₀ > 2.7 μmol/L, P = 0.0079). We additionally observed a neratinib dose-dependent decrease in p-AKT in the HER2-mutant cell lines OSUK9PAPADRiley (HER2^V659E) and BT474 (HER2^amp), whereas p-AKT levels in OSUK9PAPADOscar (HER2^WT) were low at all treatment levels (Fig. 2C). Given that HER2 receptors dimerize with EGFR, we then evaluated differential sensitivity of the five canine cell lines and the human BT474 control line to the EGFR inhibitor erlotinib. Not surprisingly, HER2^V659E IC₅₀ values were greater than those of neratinib at 220 nmol/L and 1.17 μmol/L. HER2^WT IC₅₀ values ranged from 1.10 to 15.38 μmol/L (Supplementary Fig. S8). Finally, we evaluated the HER2-mutation–dependent effects of trastuzumab in the five cPAC cell lines described above (two HER2^V659E and three HER2^WT) and the positive control cell line BT474 at doses ranging from 1 × 10⁻⁶ μg/mL to 1 mg/mL for 72 hours. Although trastuzumab did not decrease viability below 50% for any of the cell lines at 72 hours, the greatest dose-dependent responses were observed in the BT474 control line (65% viability at 10 μg/mL, consistent with prior publications) and in the HER2^V659E line OSUK9PAPADRex (74% viability at 10 μg/mL). More limited responses were observed in OSUK9PAPADOscar (HER2^V659E) and OSUK9PADSnickers (HER2^WT) and no responses at any dose in Riley (HER2^V659E) and BACA (HER2^WT; Supplementary Fig. S9). Overall, these studies support that HER2^V659E in cPAC is an activating event that stabilizes HER2 homodimers and heterodimers, confers dependency on downstream signaling, and confers sensitivity to targeted HER2 tyrosine kinase inhibition.

Through multiplatform next-generation sequencing of 88 naturally occurring primary canine NSCLC cases (77 tumors and 11 cell lines), we describe for the first time the detailed genomic underpinnings of this cancer. The cohort included major NSCLC subtypes occurring in dogs and humans: cPAC (n = 74, cPASC (n = 11), and cPSCC (n = 3; Supplementary Table S2; Supplementary Fig. S1). Although lung cancer may be overrepresented in Doberman pinschers, Australian shepherds, Irish setters, and Bernese mountain dogs (7), Labrador retrievers comprised the largest pure breed in this cohort (21%) followed by mixed breeds (25%). The cohort was gender balanced (52% females), primarily neutered/spayed (92%), and bore a median age at diagnosis of 11 years. Given that dogs are companion and service animals that typically share the same environment with humans, they may have a role to play as sentinels for human lung cancer environmental risk factors. Some data suggest that environmental risks are shared across species. For example, an increased risk of developing cPAC (OR: 2.4; CI 95%, 0.7–7.8; P = not given) trends toward association with having a smoker in the home in dogs with short (brachycephalic) or medium length (mesocephalic) noses, such as Labrador retrievers (26). Although secondhand smoke exposure in the dogs in our cohort is possible given that exposure was not recorded, genomic landscapes of human lung cancers in never-smokers have not been shown to differ based on exposure to secondhand smoke (27). Exposure to other environmental carcinogens such as air pollutants may also play a role in the development of lung cancers. For example, increased lung cancer risk may be present in dogs with higher amounts of carbon deposits known as anthracosis (OR: 2.1; CI 95%, 1.20–3.70; P < 0.01; ref. 28), although in humans anthracosis has been commonly observed in normal lungs as well as tumors and lymph nodes. In this cohort, anthracosis was recorded in 15 cases and pneumoconiosis (lung disease associated with pollutant exposure) in one case. However, no associations between anthracosis annotation and genetic features of these cases were observed. Overall, our studies included broad representation of lung cancer across histologic subtypes, breeds, ages, and pollutant exposures reflective of primary canine lung cancer diversity seen in the clinical setting in the United States. Overall, support exists for shared etiologies between canine and human never-smoker lung cancer, including secondhand smoke, organic dusts, and outdoor and indoor air pollution, suggesting that study of canine lung cancer can be informative for understanding human lung cancer risks and etiologies. Genomic characterization of canine lung cancers is a first major step toward understanding variables influencing lung cancer development in pet dogs.

Unique genomic characteristics of human never-smoker lung cancer include low somatic mutation burden, C:G>T:A enrichment, and activating mutations or fusions impacting EGFR (45%), ALK (5%–11%), ROS (1.5%–6%), HER2 (3%–5%), and RET (2%; refs. 4, 29). Here, we also observed a low somatic burden of SNVs, CNVs, and SVs through exome sequencing in five matched tumor/normal cPAC pairs. We additionally observed that the most common mutation signature in these five cases was the age-associated COSMIC signature 1A in four of five (80%) similar to the enrichment seen in human NSCLC (Supplementary Fig. S3). This signature is associated with age in many human cancers, putatively the result of spontaneous deamination of 5-methyl-cytosine. COSMIC signature 2, associated with APOBEC cytidine deaminase activity, was also present in two cases. This signature is most prominently associated with cervical and bladder cancers, but is also commonly found in lung adenocarcinoma and SCC. Although these signatures are sometimes associated with APOBEC gene variants in human cancers (30), no putatively pathogenic germline or somatic APOBEC mutations were observed in this canine cohort. Based on our studies, primary canine lung cancers bear a low mutation burden (TMB mean of 2.04 mutations/Mb) and mutation signatures reflective of those seen in human never-smoker lung cancers.

The most common recurrently mutated genes containing somatic potentially pathogenic SNVs in the full cohort included HER2 (31.5%), TP53 (12.5%), PTEN (5.7%), SMAD4 (4.5%), KRAS (4.5%), VHL (3.4%), and HRAS (2.3%). Recurrent CDKN2A/B focal deletions were also observed in two of five (40%) cases (Fig. 1A and B) along with a homozygous missense mutation, G50R, equivalent to human codon G101 mutations. CDKN2A deletions were the most common alteration by frequency, occurring at rates comparable with those in human NSCLC. Two focal deletions were observed out of five exome-sequenced cases, with signs of larger-scale CFA11 losses in remaining cases (Supplementary Fig. S2). CDKN2A is mutated in ∼30% of all human NSCLC, primarily via homozygous deletion, and this number is reduced to around 25% in never-smokers. The next most common alterations after CDKN2A and HER2 were TP53 missense and truncating mutations comparable with DNA binding domain mutations in human TP53. Similar to human NSCLC, we observed a reduced burden of TP53 mutations (12.4%, two stop gains and nine likely pathogenic missense mutations) relative to human smoker NSCLC in which more than half of tumors are mutated. PTEN mutations were the next most common at 5.6%. PTEN is mutated in ∼9% of human NSCLC, but only ∼2% of never-smoker NSCLC. We additionally identified four somatic mutations in the tumor suppressor SMAD4, mutated in ∼5% of human NSCLC at comparable rates in smoker and never-smoker cancer. KRAS mutations are the most common oncogenic mutations in human smoker NSCLC (∼30%–40% of cases), but occur at reduced frequencies in never-smoker lung cancer (0%–7%). KRAS mutations in our cohort were rare (2 G12V, 1 G12D, and 1 Q61K), but comparable to human hotspots. Canine HRAS missense mutations were also located in human-equivalent hotspots (Q61L and F78S). Additional likely pathogenic somatic mutations included individual cases of AKT1 amplification, KIT/KDR amplification, EGFR A726T (human A755), MET M1269V (human M1268), and VHL P97L (human P97). WWTR1, the only COSMIC gene bearing a somatic translocation in exome-sequenced cases, has been shown to undergo translocation with CAMTA1 in human epithelioid hemangioendothelioma (31). We identified a WWTR1 translocation of unknown consequence with ATP5F1. Although we identified translocations occurring in coding regions in five exome-sequenced tumors, it remains possible that, as in human never-smoker lung cancer, EML4–ALK fusions, ROS1 fusions, RET fusions, and other fusions may also be present in canine lung cancer.

In addition to charting the landscape of cPAC, we have found recurrent KRAS and TP53 mutations in cPASC and provide a view of possible drivers in cPSCC. In cPASC, HRAS Q61L and KRAS Q61K each occurred in one case. Finally, although no recurrent mutations were identified in the three cPSCCs, we identified one case with somatic BRAF V588E (equivalent to the human V600E hotspot) and another bearing PTPN11 G503V (equivalent to the human G503V hotspot).

HER2 contained the most somatic mutations with hotspot mutations occurring solely in cPAC (37.8%). HER2 is a well-characterized human oncogene and drug target mutated in ∼6% of all cancers based on cBioPortal query of 10,967 cases in the TCGA pan-cancer atlas (32, 33). Most alterations are focal amplifications, but activating point mutations and insertions are also common. In human NSCLC, HER2 mutations are oncogenic drivers in ∼1% to 4% of cases with mutations and insertions mostly in exon 20 at codon 776 resulting in constitutive HER2 kinase domain activation and downstream signaling through the PI3K and MAPK pathways (29, 34, 35). HER2 may also be more commonly mutated in human never-smoker lung cancer, with point mutations at frequencies reported at 3% to 5% (36), predominately in female never-smokers who carry a median OS of ∼2 years (35). HER2 TMD polar mutations (HER2^V659E/D and HER2^G660D) are present in 0.18% of human lung adenocarcinomas and are exclusive with HER2 kinase domain mutations (37). Amplicon analysis capable of identification of point mutations and small insertions or deletions covered canine HER2 exons 8 and 17 to 22 including transmembrane and kinase domains. Additionally, Sanger sequencing of all exons in five canine cell lines with WT HER2 based on amplicon sequencing (OSUK9PAD, BACA, CLAC, K9PADSQ, and OSULSCC1) found no somatic HER2 mutations in other sites (Supplementary Table S11). It is nonetheless possible that somatic mutations occurring in other regions of HER2 were not identified in amplicon-sequenced samples even though data facilitating functional interpretation of these variants would be limited.

In addition to point mutations, HER2 amplification has also been identified in ∼1% of human NSCLC (29), with enrichment in EGFR-inhibitor–resistant tumors (38). Protein overexpression is reported in 6% to 35% of tumors, including up to 42% of adenocarcinomas, and correlates with poor prognosis (39–42). We detected no somatic HER2 focal amplifications or numerical CFA9 gains in five exome-sequenced cases (Fig. 1B; Supplementary Fig. S2) or two previously aCGH-profiled cell lines. However, four of these seven cases contained somatic, putatively activating HER2 SNVs. Given that HER2 amplification/overexpression and SNVs are typically mutually exclusive, it remains possible that our broader amplicon cohort contained undetected HER2 amplifications. We therefore utilized qRT-PCR and IHC studies to more broadly assess HER2 overexpression and did not find evidence for significant tumor-specific HER2 overexpression (Supplementary Figs. S5 and S6; Supplementary Table S13). Thus, it is unlikely that HER2 is frequently amplified in canine lung cancer.

Overall, though we observed a similar mutation spectrum in canine lung cancer relative to human never-smoker NSCLC, the notable exception is abundance of HER2 mutations and lack of EGFR mutations. EGFR mutations occur at low frequency in human smoker lung cancers (0%–7%), but are enriched in human never-smokers (∼45%). Canine HER2 shares normal and oncogenic roles with human HER2 based on sequence conservation (92.2% protein identity) and prior study of its role in canine cell signaling. HER2^V659E occurs at a highly conserved residue (100% identity in the TMD from amino acids 654–674) and to the neu (rat HER2) variant identified in a rat glioblastoma cell line that originally led to the discovery of HER2's oncogene status (43). HER2 has previously been implicated in canine cancers via overexpression by IHC and qRT-PCR in canine mammary tumors (44), through its utility as a vaccine target in canine osteosarcoma (45), and through downstream signaling activation in canine lung cancer (10). Thus, HER2 sequence and pathway biology is conserved, so the predominance of HER2 mutations as erbB signaling activators in lieu of EGFR mutations in cPAC may be the result of cell-of-origin and genetic background influences. Cell-of-origin determination in canine lung cancers is challenging because the pulmonary adenocarcinoma diagnosis includes tumors arising from primary, secondary, and tertiary bronchioles and thus topographic origin can be difficult to determine. However, evidence supports that HER2 is broadly important for canine pulmonary epithelium. For example, neuregulin-stimulated HER2 increases proliferation in pulmonary epithelial cells by activation of the JAK–STAT pathway. Further, when HER2 activation is blocked via antibodies to neuregulin or HER2 in a scratch wound-healing assay of pulmonary epithelial cells, wound closure is significantly delayed, suggesting HER2 activation is necessary for epithelial proliferation (46). We have also found that IHC of canine normal lung showed stronger HER2 staining of all bronchioalveolar regions when compared with EGFR staining of normal adult canine lung. These data suggest that HER2 may play a more central role than EGFR in canine alveolar and airway epithelial cells during chronic lung injury and for general proliferative processes. Prolonged activation could lead to cellular transformation and neoplasia. Further, EGFR mutations have been associated with particular histotypes; i.e., they are frequent in lepidic and acinar patterns and infrequent in mucinous patterns in female Asian never-smoker PAC. In this population, the most frequent adenocarcinoma histotype was acinar (142 cases, 71.7%), followed by papillary (18 cases, 9.1%), solid (17 cases, 8.6%), lepidic (9 cases, 4.5%), and micropapillary (1 case, 0.5%). Interestingly, our canine cohort had predominantly papillary morphology (69%), with only 5% acinar (5%). Therefore, differences in cell of origin in both species could account for the differences in EGFR mutation frequencies. Background genetic context likely also plays a primary role in shaping enrichment for HER2 mutations in cPAC. This is also true in human lung cancer where EGFR mutation frequency varies by more than 3-fold between different human populations. The Asian population has a very high rate of EGFR mutation among the never-smoking population, up to 51.4% overall and as high as 64% in some populations such as the Kinh, versus about 20% in Caucasians (47–49). These differences in human populations suggest a sensitivity of EGFR mutations to genetic context. Conversely, HER2 mutations are found in all human populations at about the same frequency, suggesting that HER2 mutations in humans may not be as sensitive to genetic background.

We have additionally shown that HER2 hotspot mutations can be detected in the plasma of dogs bearing HER2^V659E cPACs even at early disease stages (Fig. 1E; Supplementary Table S12). In human NSCLC, ctDNA has been shown to be significantly enriched in plasma relative to controls with key genetic features identifiable via liquid biopsy. Associations have been found between ctDNA levels and tumor stage, grade, lymph node status, metastatic sites, response, and survival (50, 51). The first FDA-approved liquid biopsy test was the cobas EGFR Mutation Test v2, a real-time PCR assay utilized in NSCLC for the detection of EGFR exon 18 to 21 mutations in tissue or plasma to guide EGFR inhibitor treatment assignment (52, 53). Our proof-of-principle study supports that ctDNA is also detectable in primary canine lung cancer patient plasma. A noninvasive HER2^V659E assay will enable genotyping patients when tumor tissue is limited and may have a role in treatment monitoring or detection of minimal residual disease. This assay will also facilitate prospective analysis of HER2^V659E's diagnostic and prognostic value.

In human cancers, HER2 TMD mutations constitutively activate prosurvival HER2 signaling (37) and are associated with HER2 inhibitor responses (23). We have confirmed in this study that, similar to human HER2 TMD mutants, canine HER2^V659E cell lines constitutively activate downstream signaling through AKT and are selectively sensitive to the HER2 TKI inhibitors neratinib and lapatinib in vitro (Fig. 2; Supplementary Fig. S7). In order to further assess the role of dimerization in HER2 activation in cPAC, we also performed drug-dose–response studies for erlotinib and trastuzumab (Supplementary Figs. S8 and S9). One of two HER2-mutant cell lines showed erlotinib sensitivity. Trastuzumab responses were poor overall and did not correlate with HER2 status, although dose–response relationships were observed in three of five cell lines. Trastuzumab's human binding site is highly conserved in canine (only a single amino acid difference) and trastuzumab has been shown to bind canine HER2 and inhibit proliferation of HER2-overexpressing canine cancer cell lines (54). However, even the human HER2-amplified cell line BT474 did not show viability reduction below 50% in our hands. It is likely that the effects of trastuzumab on CellTiter-Glo viability are broadly muted at the 72-hour time points we utilized. Overall, these studies indicate that HER2^V659E in cPAC is an activating event that stabilizes HER2 homodimers and heterodimers, confers dependency on downstream signaling, and confers sensitivity to targeted HER2 tyrosine kinase inhibition. We have charted the genomic landscape of primary canine lung cancers including the NSCLC subtypes cPAC, cPASC, and cPSCC. We have identified recurrent HER2 mutations in these cancers and present, to our knowledge, the first complete suite of evidence supporting an oncogenic role for and dependency on constitutively activating mutations in HER2 in a canine cancer. Further work is needed to exhaustively profile these tumors, particularly according to variation across breeds and through integration of additional data types including epigenomics, RNA-seq, and proteomics. However, these data nonetheless offer significant immediate diagnostic and therapeutic opportunities for dogs with primary lung cancer and aid comparative understanding of never-smoker and HER2-mutant lung cancer. These findings set the stage for HER2 inhibitor toxicity, dose-finding, and efficacy studies in dogs that will guide utilization of HER2 inhibitors in the veterinary clinic.

S.L. Sinicropi-Yao is an employee of Sanofi. M. Murtaza reports receiving commercial research grants from Ethos Discovery. W.P.D. Hendricks is a consultant/advisory board member for The One Health Company. No potential conflicts of interest were disclosed by the other authors.

Conception and design: G. Lorch, M. Murtaza, J. Trent, D.P. Carbone, W.P.D. Hendricks

Development of methodology: G. Lorch, K. Sivaprakasam, V. Zismann, M.J. Koenig, M. Murtaza, W.P.D. Hendricks

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G. Lorch, V. Zismann, N. Perdigones, S. Facista, S. Wong, W.S. Liang, J.M. Amann, S.L. Sinicropi-Yao, M.J. Koenig, K.L. Perle, M. Murtaza

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. Lorch, K. Sivaprakasam, N. Perdigones, A. Nazareno, S. Wong, K. Drenner, M.J. Koenig, K.L. Perle, T.G. Whitsett, M. Murtaza, D.P. Carbone

Writing, review, and/or revision of the manuscript: G. Lorch, K. Sivaprakasam, V. Zismann, N. Perdigones, J.M. Amann, K.L. Perle, T.G. Whitsett, M. Murtaza, J. Trent, D.P. Carbone, W.P.D. Hendricks

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G. Lorch, N. Perdigones, J. Trent, W.P.D. Hendricks

Study supervision: G. Lorch, V. Zismann, J. Trent, D.P. Carbone, W.P.D. Hendricks

Other (running a subset of the samples that generated data for this publication): T. Contente-Cuomo

This study was supported by National Canine Cancer Foundation GL150H-005 (G. Lorch), the Bisgrove Scholars Program (M. Murtaza), philanthropic support to the TGen Foundation (J. Trent and W. Hendricks), CTSA UL1TR001070 (G. Lorch), NCI P30 CA016058 (G. Lorch, J. Trent, and W. Hendricks), Brooke's Blossoming Hope (J. Trent and W. Hendricks). We thank Dr. Amy LeBlanc, Director of the NCI's Comparative Oncology Program (COP), and Christina Mazcko, COP Program Manager, for their assistance with CCOGC sample procurement and annotation. Additionally, we thank Dr. Holly Borghese of the OSU CVM Veterinary Biospecimen Repository for her expertise in sample procurement and Dr. Kurtis Yearsley of the OSU Pathology Imaging Core for his expertise in quantitative image analysis. We thank Drs. Alshad Lalani, Irmina Diala, and Lisa Eli of Puma Biotechnology for fruitful discussion of HER2 inhibition and provision of neratinib for this study.

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.