Skip to main content
Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Actionable mutations in canine hemangiosarcoma

  • Guannan Wang,

    Roles Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Resources, Supervision, Validation, Writing – review & editing

    Affiliation Department of Pathology and Laboratory Medicine, Raymond and Ruth Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, United States of America

  • Ming Wu,

    Roles Formal analysis, Methodology, Software

    Affiliation Illumina, San Diego, CA, United States of America

  • Martha A. Maloneyhuss,

    Roles Resources

    Affiliation Department of Clinical Studies, School of Veterinary Medicine, University of Pennsylvania, Philadelphia, PA, United States of America

  • John Wojcik,

    Roles Resources

    Affiliation Department of Pathology and Laboratory Medicine, Raymond and Ruth Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, United States of America

  • Amy C. Durham,

    Roles Resources, Writing – review & editing

    Affiliation Department of Clinical Studies, School of Veterinary Medicine, University of Pennsylvania, Philadelphia, PA, United States of America

  • Nicola J. Mason,

    Roles Resources, Writing – review & editing

    Affiliations Department of Clinical Studies, School of Veterinary Medicine, University of Pennsylvania, Philadelphia, PA, United States of America, Department of Pathobiology, School of Veterinary Medicine, University of Pennsylvania, Philadelphia, PA, United States of America

  • David B. Roth

    Roles Conceptualization, Funding acquisition, Project administration, Supervision, Writing – original draft

    Affiliation Department of Pathology and Laboratory Medicine, Raymond and Ruth Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, United States of America



Angiosarcomas (AS) are rare in humans, but they are a deadly subtype of soft tissue sarcoma. Discovery sequencing in AS, especially the visceral form, is hampered by the rarity of cases. Most diagnostic material exists as archival formalin fixed, paraffin embedded tissue which serves as a poor source of high quality DNA for genome-wide sequencing. We approached this problem through comparative genomics. We hypothesized that exome sequencing a histologically similar tumor, hemangiosarcoma (HSA), that occurs in approximately 50,000 dogs per year, may lead to the identification of potential oncogenic drivers and druggable targets that could also occur in angiosarcoma.


Splenic hemangiosarcomas are common in dogs, which allowed us to collect a cohort of archived matched tumor and normal tissue samples suitable for whole exome sequencing. Mapping of the reads to the latest canine reference genome (Canfam3) demonstrated that >99% of the targeted exomal regions were covered, with >80% at 20X coverage and >90% at 10X coverage.

Results and conclusions

Sequence analysis of 20 samples identified somatic mutations in PIK3CA, TP53, PTEN, and PLCG1, all of which correspond to well-known tumor drivers in human cancer, in more than half of the cases. In one case, we identified a mutation in PLCG1 identical to a mutation observed previously in this gene in human visceral AS. Activating PIK3CA mutations present novel therapeutic targets, and clinical trials of targeted inhibitors are underway in human cancers. Our results lay a foundation for similar clinical trials in canine HSA, enabling a precision medicine approach to this disease.


Angiosarcomas (AS) are rare malignant tumors of the endothelium of blood vessels, comprising up to 1–2% of sarcomas in humans (Sarcoma Foundation of America, AS is deadly, with a 5 year survival of less than 30%[1]. Disease pathogenesis is poorly understood and there are no effective therapies. AS presents in three clinical situations: post irradiation therapy, sun exposed skin in older individuals, and, less commonly, spontaneous disease in visceral organs. As a result of its rarity, little is known about potential driver mutations that could serve as therapeutic targets. There is some evidence that post irradiation AS have a different mutational spectrum than cutaneous AS [2]. Visceral AS have not been well studied at the molecular level. The rarity of AS has impeded efforts to identify driver mutations that could present therapeutic targets, and the prospects for accruing patients to clinical trials are poor. To address this problem, some investigators are using "crowdsourcing" to identify patients with this disease (@ASCaProject, Angiosarcoma Project, A search of the Hospital of the University of Pennsylvania archives spanning several decades revealed only a few cases, and the quality of the DNA we extracted from archived paraffin blocks was too low to perform exome sequencing. We turned to comparative genomics, taking advantage of the observation that a histologically similar malignant tumor, hemangiosarcoma (HSA), is common in dogs, which would enable discovery (exome) sequencing of paired tumor-normal tissue samples.

Canine hemangiosarcoma (HSA) is a highly aggressive, fatal malignancy characterized by neoplastic cells of endothelial origin that form vascular structures histologically similar to human AS. HSA affects approximately 50,000 dogs per year in the United States[3,4]. As in humans, both visceral and cutaneous forms of the disease are recognized, but in dogs the visceral form, specifically splenic, is more common. Canine patients with visceral HSA usually present with hemoabdomen following rupture of the splenic mass. Despite emergency splenectomy and adjuvant chemotherapy, most dogs die within 6 months from pulmonary metastatic disease.

Based on the clinical and histopathological similarities to human visceral AS (S1 Fig), we hypothesized that dogs with HSA might serve as a tractable, spontaneous model system to identify potential therapeutic targets and to carry out clinical trials. If successful, this approach could provide key information applicable to human AS, and would also provide new tools to address this common, deadly disease in dogs. Our hypothesis that AS and HSA might share pathogenic features is supported by previous studies that have identified mutations and deletions in the same genes that are shared between HSA and AS, such as TP53 and PTEN [57]. More recently, genome-wide expression profiling identified three distinct tumor subtypes associated with angiogenesis (group 1), inflammation (group 2), and adipogenesis (group 3). These data suggest that a common HSA progenitor may differentiate into the three different tumor subtypes and raise the intriguing question of the dynamic tumor evolution in HSA[8]. Genome-wide array-based comparative genomic hybridization (aCGH) revealed a relatively low rate of copy number abnormalities with small amplitude in HSA[9]. These data suggest that, in contrast to some of the other sarcomas, the driving force behind HSA might not involve global, high-grade alterations at the chromosomal level. Rather, small insertions/deletions and single nucleotide variants may drive tumorigenesis. Genome-wide association studies of HSA identified two predisposing loci on chromosome 5 but failed to find coding changes[10]. Thus, neither the somatic driver mutations nor the nature of the pathways leading to oncogenesis in this common and deadly disease are known.


Case collection and exome sequencing

We collected archived paired tumor and normal tissue samples from patients representing a variety of breeds, with a slight over-representation of Golden Retrievers, German Shepherd dogs and Labradors (S1 Table). We chose twenty-one cases for exome sequencing. DNA was extracted from microdissected tumor and matched normal formalin fixed, paraffin embedded (FFPE) tissue samples for whole exome sequencing (WES). We analyzed WES data following a bioinformatics workflow we customized for canine genomics, including mapping, mutation calling, and annotation (Fig 1; Materials and Methods). Mapping of the reads to the latest canine reference genome (Canfam3) demonstrated that >99% of the targeted exomal regions were covered, with >80% at 20X coverage and >90% at 10X coverage. Sequencing runs were designed to cover tumor at 2X the coverage of normal control DNA, as reflected by our data showing mean coverage of normal and tumor samples was 37X and 74X, respectively (S2 Fig). Sequencing data were filtered as described in Materials and Methods (Fig 2). One case (P2) was eliminated from the analysis because further review of stained slides from the dissected tumor revealed heavy contamination of the tumor sample by normal tissue. As expected, we observed no somatic mutations in that case.

Fig 1. Whole exome sequencing workflow.

Case identification and sequencing: Splenectomy samples (tumor and normal) were obtained from specimens derived from patients undergoing standard clinical care at our veterinary hospital. Genomic DNA were extracted and DNA quality was assessed via three QC steps. Exome libraries were prepared according to manufacturer’s protocol and sequenced on a Nextseq 500 at the Perelman School of Medicine sequencing core. Primary data processing: Reads were aligned to canine reference genome CanFam3 and PCR duplicates were removed. Variant calling and annotation: Sequence variants were detected, filtered and annotated by in-house pipeline. Candidate driver mutations were identified as summarized in Fig 2.

Fig 2. Decision tree for filtering and curation of variants.

Single nucleotide variants (SNVs) and small insertion-deletion variants (indels) were filtered as shown, on the left and right sides of the Fig, respectively. The final filtered SNVs (167) and indels (32) were subjected to various annotation steps, such as pathway analysis, comparative genomics analysis and candidate gene comparison, as schematized at the bottom of the Figure.

From these 20 cases, we identified 171 nonsynonymous somatic mutations: 153 single nucleotide variants (SNVs) and 18 insertions or deletions (indels). On average, we detected 8.6 missense somatic mutations per case (range, 0–21; Fig 3), and 9 or fewer missense mutations in more than half of our cases (Fig 3). Totaling the number of both nonsynonymous and synonymous mutations, the calculated HSA mutation burden is estimated to be 0.1–2.1 mutations per megabase, which falls on the low end of the mutation burden observed in human tumors (ranging 0.001–400 mutations per megabase), and is comparable to human tumors with low mutational burdens such as ovarian cancer and sarcomas [11,12].

Fig 3. Number of mutations in each tumor.

Nonsynonymous somatic mutation loads in canine HSA exomes: the numbers of somatic nonsynonymous mutations is displayed for each case of 20 HSA cases, ranked from left to right by the total number of mutations. Sequence coverage is listed under each case. Note that the number of mutations per case is not correlated with sequence coverage.

Mutations corresponding to drivers of human cancer occur commonly in HSA

The relatively low mutation burden simplified our search for candidate driver mutations. We examined somatic mutations in depth using three approaches: pathway analysis as described in Materials and Methods (Fig 2), comparative genomics (comparing to well-curated human data), and a candidate gene analysis in which we compared mutations with an in-house curated cancer gene list including COSMIC consensus cancer genes, genes with previously described mutations in human AS and canine HSA, as well as genes residing in pathways that are affected in vascular cancers[2,13,14].

Most informative was the cross-species comparative genomics approach, which allowed us to annotate canine mutations using human cancer-associated phenotypic data such as COSMIC, ClinVar, and HGMD (after converting to human amino acid coordinates as described in Materials and Methods). In over half of the cases (12/20), we identified somatic mutations predicted to have significant functional consequences in 4 genes previously established as human cancer drivers. We found PIK3CA mutations in 9 cases (45%), TP53 mutations in 7 cases (35%), PTEN mutations in 2 cases (10%), and a PLCG1 mutation in one case (5%), Fig 4A.

Fig 4. Candidate driver mutations and recurrent mutations.

Each row represents a mutated gene and each column represents an individual tumor. Shown in green blocks are non-synonymous, non truncating variants which could represent gain-of-function mutations. Shown in black are predicted inactivating mutations, including truncating mutations, essential splice-site variants, and nonsense mutations. A: Candidate driver mutations, B: additional recurrent mutations.

We next took a complementary approach to identify candidate driver mutations, which was to identify all recurrently mutated genes in our cohort, based on the hypothesis that some recurrently mutated genes might play a role in pathogenesis (Fig 4A and 4B). This analysis re-identified our strong driver candidates (PIK3CA mutations in 9 cases; TP53 mutations in 7 cases; PTEN mutations in 2 cases). We also identified additional recurrently mutated genes: two mutations in two uncharacterized genes We also identified additional recurrently mutated genes: two mutations in two uncharacterized genes (ENSCAFG00000023562 and ENSCAFG00000031493), two mutations in TTN (likely passenger mutations [15], and two mutations in DSCAM. DSCAM, a large protein (1776 aa), is a negative regulator of cell adhesion and plays a role in neuronal self-avoidance, acting in central and peripheral nervous system development. Germline mutations in DSCAM have been reported in Down's syndrome and in congenital cardiac malformations [16,17]. We identified a DSCAM mutation not previously reported in cancer (Gly1679Ser; human: Gly1915Ser) and an Arg1562* mutation, whose human homologue, Arg1798*, has been identified in esophageal cancer [18]. Further investigation will be required to determine whether these mutations contribute to pathogenesis.

Mutations in the PIK3CA pathway are likely drivers of HSA

The most frequent potential driver mutations affected the PIK3CA gene (9 cases; 45%). The human and canine PIK3CA genes share 99.8% sequence identity, and in 8 of our 9 PIK3CA mutant cases, the missense mutation affected amino acid 1047, the same position most commonly mutated in human cancers [19,20]. Indeed, six cases bear an H1047R substitution, a well-established driver mutation in human cancers (Fig 5). Two other cases have another nonconservative substitution at the same position (H1047L). In humans, H1047R is found in at least 4% of all cancers (COSMIC,, and accounts for almost 40% of all PIK3CA coding mutations (COSMIC, The mutant protein shows increased enzymatic activity and activates downstream signaling, resulting in oncogenic transformation in vitro and tumorigenesis in vivo [1922]. H1047L is also observed in human cancers, and both H1047R and H1047L are associated with a poor prognosis in adenocarcinoma of the lung [23]. Furthermore, clinical trials show that these pulmonary adenocarcinomas respond to PIK3CA inhibitors [24]. These data strongly suggest that the most prevalent mutations in PIK3CA we observed, H1047R and H1047I, function as cancer drivers in HSA.

Fig 5. Distribution of alterations in the PIK3CA gene.

(A) The table shows the distribution of PIK3CA mutations in our cohort of canine HSA cases, the corresponding human amino acid change, and the functional impacts of the mutants based on previous human studies. "GOF" denotes gain-of-function. (B) Schematic diagram of homologous mutations in human PIK3CA. The x axis represents amino acid positions and domain structure of human PIK3CA, and the y axis represents the number of times each mutation observed in our cohort.

In one case, we identified a PIK3CA D350G mutation, which has been identified in human breast, endometrial, pancreatic, and colorectal carcinomas [25,26]. Structural modeling of the human protein predicts that D350G disrupts interactions with PI3K signaling pathway inhibitors, and is therefore a gain-of-function mutation [27], suggesting that this mutation could drive canine HSA. Another indicator that the PIK3CA mutations we identified drive the development of HSA is that 7 of the 9 cases bearing PIK3CA mutations have fewer than 10 additional mutations (apart from those found in PIK3CA and TP53), and these are not obvious cancer drivers. Indeed, in one case (P16) bearing a presumed PIK3CA activating mutation (H1047R), only two additional mutations were identified, both in uncharacterized genes.

Our hypothesis that activation of the PIK3CA drives the development of HSA is further supported by the presence of other mutations affecting the PI3K signaling pathway in our cohort. PTEN, a negative regulator of the PIK3CA pathway, is mutated in two cases that conspicuously lack PIK3CA mutations. One PTEN mutation, shown in black in Fig 4A, is a frameshift mutation that encodes a truncated protein (Thr56fs). This heterozygous loss-of-function mutation is likely to have functional consequences, given that haploinsufficiency of the PTEN tumor suppressor gene promotes prostate cancer progression [28]. The second case, shown in green in Fig 4A, bears a missense mutation (Ile79Thr), affecting amino acid 79, which corresponds to amino acid 101 in the human protein. Although functional data are lacking, Ile101Thr is predicted to be highly pathogenic by FATHMM. In humans, Ile101 is mutated in several human tumors, primarily gliomas [29], suggesting that the canine PTEN Ile79Thr mutation may play a role in the pathogenesis of HSA.

In another case, we observed a truncating mutation in FoxO3 (Glu93fs), which appears to be a loss of function (P8; Fig 6). This case lacks PIK3CA and PTEN mutations. Previous work has shown, however, that activating PI3K signaling could negatively regulate Foxo3, leading to its cytoplasmic sequestration and degradation [30]. Thus, a loss of function mutation in FoxO3 could activate downstream PI3K signaling, and may function in the same way as activation of PIK3CA or inactivation of PTEN.

Fig 6. Summary of somatic mutations in HSA cases lacking strong driver mutation candidates.

Green blocks denote missense mutations. Shown in black are predicted loss-of-function mutations (essential splice-site variants, nonsense mutations, and frameshift indels).

TP53 is frequently mutated in HSA

We identified TP53 mutations in 7 cases (35%), making TP53 the second most frequently mutated gene in our cohort (8 mutations in 7 tumors). All TP53 mutations are predicted to have functional consequences, with truncating (1), missense (5), and essential splice site (2) mutations, all of which affect the protein's DNA binding domain (Fig 7). This is reminiscent of human cancer, in which more than 80% of TP53 mutations localize to the DNA binding domain. Indeed, all of the canine TP53 mutations we identified have been observed in human cancers, and some of them have been studied extensively. For clarity, we refer to amino acid positions in the human protein in the following discussion. In addition to the classic loss-of-function mutants (Leu111 nonsense mutation, Y126N), we also observed a dominant negative mutation (V274L) and identified several gain-of-function mutants. H179Y produces a growth advantage by upregulating Cyclin A1 and Cdk4 [31]. S241F acts as both a dominant negative and a gain-of-function mutant, inhibiting p21 expression and modifying chromatin remodeling patterns [32], and R248W induces genomic instability by inactivating ATM and also alters the epigenetic landscape [33,34]. Determining the mechanisms underlying the effects of these mutants requires further investigation.

Fig 7. Distribution of alterations in the TP53 gene.

(A) Table shows the distribution of TP53 mutations in our cohort, the mutation types, the corresponding human amino acid change, and the predicted functional impact of each mutation based on previous studies. (B) Schematic diagram of homologue mutations in human TP53. The x axis represents amino acid positions and domain structure of human TP53, and the y axis represents the number of times each mutation was identified in our cohort.

A predicted activating mutation in PLCG1 is homologous to a recurrent mutation in human AS.

We observed a missense mutation in PLCG1 (S273F) in one tumor. S273 is in the kinase domain of PLCG1 and is highly conserved among species. The corresponding human mutant, S345F, exhibits enzyme activation and increased downstream signaling, such as T cell receptor signaling and NF-kB signaling [35]. S345F drives tumorigenesis in human nodal peripheral T-cell lymphomas and cutaneous T-cell lymphomas [35,36], and this mutation was recently identified in human visceral AS [37]. These data suggest that canine PLCG1 S273F contributes to the pathogenesis of HSA, and we speculate that this case might represent a subgroup of HSA with molecular similarity to human AS, especially the visceral form.

In conclusion, we identified mutations in 12 cases of canine HSA that, supported by abundant data from human cancers, likely play a role in oncogenesis in canine HSA. Apart from these strong driver candidate mutations, we identified a few mutations in genes potentially involved in cell growth, differentiation, or apoptosis (S2 and S3 Tables). However, none of the other mutations are found in human cancer databases or predicted to be involved in tumorigenesis. At this point, the data do not warrant implicating these mutations in HSA pathogenesis.

HSA cases lacking obvious driver mutations

Eight cases lacked obvious candidate driver mutations. As noted above, analysis of recurrently mutated genes or comparative genomics failed to identify additional candidate driver mutations. All 36 mutated genes identified in these cases are schematized in Fig 6 and S2 Table. Our analysis failed to reveal candidates with sufficient evidence in the literature to support a clear role in pathogenesis of HSA. Clarification of the pathogenic drivers in these cases, which could involve structural variants or epigenetic changes not identified by exome sequence analysis, must await further investigation.


The recent development of canine exome capture reagents has opened the door for detailed profiling of canine cancer genomes. Our study represents the first use of this technology, as far as we are aware, to examine somatic genomic alterations in canine HSA. Sequencing a cohort of 20 HSA cases, we identified recurrent mutations in more than half of the cases whose human homologues are well-established cancer drivers, strongly suggesting that these candidates indeed drive canine HSA. Only one of our cases contained an apparent driver mutation (in PLCG1) which corresponds to mutations identified in human AS by targeted sequencing.

Molecular pathways driving the pathogenesis of canine HSA

Our exome sequencing approach revealed that many cases of canine HSA harbor somatic point mutations that are likely to upregulate the PI3K pathway, which is well-established in human carcinogenesis. Our hypothesis that these mutations drive HSA pathogenesis is substantiated by previous work that demonstrated constitutive activation of the mTORC2/Akt/4E-BP1 pathway (by Western blot and immunohistochemistry) in newly derived canine HSA cell lines [38,39]. Indeed, viability of cell lines derived from HSA was significantly reduced by PI3K/AKT/mTOR inhibitors [40], and a PI3K inhibitor slows growth of primary cells derived from canine visceral, cutaneous, and cardiac HSA [41].

Pathogenic PIK3CA mutations also underlie human segmental hypertrophy syndromes, which commonly involve proliferative vascular malformations. For example, somatic, mosaic PIK3CA activating mutations are present in CLOVES (Congenital Lipomatous Overgrowth with Vascular, Epidermal, and Skeletal anomalies) Syndrome [42]. Analysis of 6 patients revealed the following PIK3CA mutations: p.His1047Arg, p.Glu542Lys, and p.Cys420Arg [42]. Activating PIK3CA mutations are also present in syndromes that now fall under the umbrella of PIK3CA-related overgrowth spectrum (PROS), in which vascular malformations are often a key feature [43,44]. Inhibiting the PI3K pathway reduces proliferation of primary cells from these patients in culture, linking PIK3CA activation to pathogenesis [43]. The presence of activating PIK3CA mutations in these syndromes and in HSA suggests that these disorders may share mechanistic similarities.

Although the most common candidate driver mutations in canine visceral HSA in our study affected the PI3K pathway, we also identified one case in which a PLCG1 mutation is a likely driver candidate. This corresponds to mutations identified in a few cases of human visceral AS [37]. Additional work is required to determine whether PLCG1 mutations are a common feature of canine HSA, and whether this potential subtype has a distinct biological phenotype. Finally, our analysis failed to identify obvious driver candidates in 40% of the cases. This may be due to insufficient sequence coverage (failing to detect low allele frequency events). Alternatively, this situation might reflect technical limitations of exome sequencing approach, which does not readily detect structural variants (gene fusions, deletions, gene amplification) that are commonly found in other sarcomas, but which are not suspected to play a large role in HSA [9]. Although further work is required to identify potential drivers in these cases, our data raise the intriguing possibility that the clinic-pathologic entity known as canine HSA may consist of multiple molecular subtypes, which may require different therapeutic approaches.

The high frequency mutations affecting the PI3K pathway in our cohort suggests the possibility that clinical trials in canine patients could be useful in determining whether inhibition of this pathway provides clinical benefit. Such trials are feasible: the disease is common (easy to fully enroll a multi-armed study), and rapidly fatal (allows expeditious determination of efficacy of experimental therapies).

Is canine HSA a potential model of human visceral AS?

The PLCG1 mutation (canine S273F, corresponding to human S345F) observed in our cohort was recently reported in human visceral AS [37] in a study of a large cohort of 120 well-characterized human AS from different anatomic sites, including several cases of visceral AS, studied by targeted hotspot sequencing. Furthermore, another activating PLCG1 mutation, Arg707Gln, has been observed in human AS, primarily secondary and cutaneous forms [2,37,45]. We also observed mutations in TP53 with likely functional consequences in canine HSA, which have also been observed in human visceral AS [46]. These data suggest that subgroups of canine HSA might serve as a natural model for human visceral AS.

We did not, however, identify some known genetic alterations found in human AS, such as MYC and KLT4 amplifications and inactivating PTPRB mutations. Furthermore, we did not observe the high frequency of genetic events affecting the MAPK pathway suggested to play a central role in human AS [41]. Likewise, no PIK3CA mutations have been reported in human AS, either by NGS panel sequencing [37,45] or by PCR screening [47]. These discrepancies could reflect technical limitations of the various studies, but may also be due to differences in the biology of the various subtypes of human AS. For example, human AS is most commonly seen secondary to therapeutic irradiation (e.g., for breast cancer) or chronic sun exposure (cutaneous AS), both situations in which a higher mutational burden is expected.

Taken together, our results advance diagnosis and potential therapy of both canine HSA and potentially human AS in several ways. First, our analysis provides candidates for targeted therapy in canine HSA. Second, our results provide a basis for rational construction of "hotspot", panel-based NGS sequencing approaches to facilitate classification of HSA cases in clinical trials of targeted therapeutics. Third, our identification of likely driver mutations highly homologous to those found in humans in a subset of HSA cases could facilitate development of panel-based sequencing approaches to allow interrogation of human visceral AS cases from archival FFPE tissue, perhaps ultimately resulting in targeted therapies.


Canine HSA cases and whole exome sequencing

Forty-two HSA tumor and matched normal formalin fixed paraffin embedded (FFPE) tissue blocks were collected at PennVet between 2015 and 2016 as part of standard clinical care. Microscopic slides were reviewed and confirmed by a veterinary pathologist (ACD); tumor and normal tissue for dissection were identified. FFPE tissue cores were harvested from tumor and normal tissue blocks using a 3 mm punch tool. Genomic DNA (gDNA)) was extracted using GeneRead DNA FFPE Kit (Qiagen) and was quantified using a Qubit 2.0 Fluorometer. We then accessed the quality of extracted gDNA with a TapeStation 2200 system. We prepared libraries from 23 HSA cases whose gDNA was of sufficient quality and quantity. Whole exome libraries were prepared for tumor and matched normal samples using Canine All Exon kit (Agilent technologies). Paired-end sequencing (2X75) was performed on Illumina NextSeq500 platform in the Next Generation Sequencing Core at the Perelman School of Medicine (NGSC, Sequence data were acquired and processed per NGSC instructions followed by uploading to Illumina BaseSpace sequence hub for downstream analysis. Sequence reads were aligned to the latest canine reference genome, CanFam3. Reads with a mapping quality less than 20 were filtered out, unmapped reads and PCR-derived duplicated reads were also eliminated using Picard (version:1.139)

Variant detection

Somatic variants, both single nucleotide variants (SNVs) and small insertions/deletions (indels), were identified using VarScan (version: 2.3.9) with standard settings. We used SnpEff (version: 4.2) to annotate the variants for primary gene and protein information, as well as to predict basic functional impacts.

To identify somatic variants that are robust and functionally important, we implemented the following filters: 1) variants at the intergenic, intragenic, intronic, UTRs regions, as well as synonymous variants; 2) variants referenced in Dog SNPs (Broad Institute) and DoGSD: the Dog Genome SNP Database (, applying a minor allele frequency threshold of 0.05; 3) variants found in any normal cases in this study 4) variants with read depth fewer than 10X. Finally, resulted variants were subjected to manual read validation and review using Integrative Genomics Viewer (IGV) to eliminate artifacts from the sequencing workflow.

Variant annotation

We developed a pipeline to convert the canine mutations to human orthologues. First, we retrieved sequences of the altered canine proteins as well as human homologous proteins from ensemble, then we used BLASTp with smith-waterman algorithm to map the homologous proteins, with these parameters applied to identify orthologues: 1) e-value 0.0001 as the threshold to identify significant matches; 2) Gene symbols match between canFam and hg19; 3) more than 70% identity rate for sequence similarity. Finally, the mutated positions in the canine samples were mapped to the corresponding positions in the human gene using a custom script. We then performed downstream analysis with the human orthologues.

Pathway analysis.

We applied pathway analysis to all pass-filter variants in order to find biological pathways (co-functioning genes/mutations sets) that are over-represented in the WES data, using major web-based databases, such as KEGG, Panther pathways, DAVID Bioinformatics database, ConsensusPathDB, etc. In addition, we also manually curated the mutated genes by ranking keywords extracted from Gene Ontology (GO) terms and from publications. In general, we found that the top-ranked enriched pathways are consistent between these two approaches.

Supporting information

S1 Table. Clinical features of canine hemangiosarcoma patients.


S2 Table. Additional somatic mutations in cases that have strong driver mutation candidates (shown in Fig 4).

Highlighted in different colors are groups of mutated genes that might be involved in HSA pathogenesis.


S3 Table. Additional somatic mutations in cases lacking strong driver mutations (shown in Fig 4).

Highlighted in different colors are groups of mutated genes that might be involved in HSA pathogenesis.


S1 Fig. Morphologic and immunohistologic similarities between canine HSA and human AS.

Canine Visceral (Splenic) HSA and Human Visceral (Retroperitoneal) AS. Neoplasms share similar histologic features and are composed of plump polygonal cells, which line vascular spaces to form blood-filled clefts and cavities. There is marked nuclear and cellular pleomorphism and mitotic activity in both neoplasms.


S2 Fig. Mean sequencing coverage of tumors and matched normal tissue.



We acknowledge the generous support of a grant from the Abramson Cancer Center's sarcoma program. We are grateful to Dr. Rebecca Linn (Children's Hospital of Philadelphia) for observing that PIK3CA mutations are found in human segmental overgrowth syndromes.

We are grateful to Dr. Jennifer Morrissette (Department of Pathology and Laboratory Medicine) for helpful discussions and for reviewing a draft of the manuscript. Dr. Taehyong Kim and Dr. Paul Wang, from Penn's Institute for Biomedical Informatics, helped us develop a custom script to convert canine amino acid position to human amino acid position.

JW is supported by the Abramson Family Cancer Center Paul Calabresi Career Development Award for Clinical Oncology (K12; PAR-13-201).


  1. 1. Naka N, Ohsawa M, Tomita Y, Kanno H, Uchida A, Myoui A, et al. Prognostic factors in angiosarcoma: A multivariate analysis of 55 cases. J Surg Oncol. 1996;61: 170–176. pmid:8637202
  2. 2. Behjati S, Tarpey PS, Sheldon H, Martincorena I, Van Loo P, Gundem G, et al. Recurrent PTPRB and PLCG1 mutations in angiosarcoma. Nat Genet. 2014;46: 376–379. pmid:24633157
  3. 3. Tamburini BA, Trapp S, Phang TL, Schappa JT, Hunter LE, Modiano JF. Gene expression profiles of sporadic canine hemangiosarcoma are uniquely associated with breed. PloS One. 2009;4: e5549. pmid:19461996
  4. 4. Vail DM, MacEwen EG. Spontaneously occurring tumors of companion animals as models for human cancer. Cancer Invest. 2000;18: 781–792. pmid:11107448
  5. 5. Dickerson EB, Thomas R, Fosmire SP, Lamerato-Kozicki AR, Bianco SR, Wojcieszyn JW, et al. Mutations of phosphatase and tensin homolog deleted from chromosome 10 in canine hemangiosarcoma. Vet Pathol. 2005;42: 618–632. pmid:16145208
  6. 6. Murai A, Asa SA, Kodama A, Hirata A, Yanai T, Sakai H. Constitutive phosphorylation of the mTORC2/Akt/4E-BP1 pathway in newly derived canine hemangiosarcoma cell lines. BMC Vet Res. 2012;8: 128. pmid:22839755
  7. 7. Tate G, Suzuki T, Mitsuya T. Mutation of the PTEN gene in a human hepatic angiosarcoma. Cancer Genet Cytogenet. 2007;178: 160–162. pmid:17954274
  8. 8. Gorden BH, Kim J-H, Sarver AL, Frantz AM, Breen M, Lindblad-Toh K, et al. Identification of three molecular and functional subtypes in canine hemangiosarcoma through gene expression profiling and progenitor cell characterization. Am J Pathol. 2014;184: 985–995. pmid:24525151
  9. 9. Thomas R, Borst L, Rotroff D, Motsinger-Reif A, Lindblad-Toh K, Modiano JF, et al. Genomic profiling reveals extensive heterogeneity in somatic DNA copy number aberrations of canine hemangiosarcoma. Chromosome Res Int J Mol Supramol Evol Asp Chromosome Biol. 2014;22: 305–319. pmid:24599718
  10. 10. Tonomura N, Elvers I, Thomas R, Megquier K, Turner-Maier J, Howald C, et al. Genome-wide Association Study Identifies Shared Risk Loci Common to Two Malignancies in Golden Retrievers. PLOS Genet. 2015;11: e1004922. pmid:25642983
  11. 11. Martin SD, Brown SD, Wick DA, Nielsen JS, Kroeger DR, Twumasi-Boateng K, et al. Low Mutation Burden in Ovarian Cancer May Limit the Utility of Neoantigen-Targeted Vaccines. PLOS ONE. 2016;11: e0155189. pmid:27192170
  12. 12. Chalmers ZR, Connelly CF, Fabrizio D, Gay L, Ali SM, Ennis R, et al. Analysis of 100,000 human cancer genomes reveals the landscape of tumor mutational burden. Genome Med. 2017;9: 34. pmid:28420421
  13. 13. Antonescu C. Malignant vascular tumors—an update. Mod Pathol. 2014;27: S30–S38. pmid:24384851
  14. 14. Ghilardi C, Chiorino G, Dossi R, Nagy Z, Giavazzi R, Bani M. Identification of novel vascular markers through gene expression profiling of tumor-derived endothelium. BMC Genomics. 2008;9: 201. pmid:18447899
  15. 15. Greenman C, Stephens P, Smith R, Dalgliesh GL, Hunter C, Bignell G, et al. Patterns of somatic mutation in human cancer genomes. Nature. 2007;446: 153–158. pmid:17344846
  16. 16. Grossman TR, Gamliel A, Wessells RJ, Taghli-Lamallem O, Jepsen K, Ocorr K, et al. Over-expression of DSCAM and COL6A2 cooperatively generates congenital heart defects. PLoS Genet. 2011;7: e1002344. pmid:22072978
  17. 17. Yamakawa K, Huot YK, Haendelt MA, Hubert R, Chen XN, Lyons GE, et al. DSCAM: a novel member of the immunoglobulin superfamily maps in a Down syndrome region and is involved in the development of the nervous system. Hum Mol Genet. 1998;7: 227–237. pmid:9426258
  18. 18. Dulak AM, Stojanov P, Peng S, Lawrence MS, Fox C, Stewart C, et al. Exome and whole-genome sequencing of esophageal adenocarcinoma identifies recurrent driver events and mutational complexity. Nat Genet. 2013;45: 478–486. pmid:23525077
  19. 19. Gkeka P, Evangelidis T, Pavlaki M, Lazani V, Christoforidis S, Agianian B, et al. Investigating the Structure and Dynamics of the PIK3CA Wild-Type and H1047R Oncogenic Mutant. PLoS Comput Biol. 2014;10. pmid:25340423
  20. 20. Bader AG, Kang S, Vogt PK. Cancer-specific mutations in PIK3CA are oncogenic in vivo. Proc Natl Acad Sci U S A. 2006;103: 1475–1479. pmid:16432179
  21. 21. Nisa L, Häfliger P, Poliaková M, Giger R, Francica P, Aebersold DM, et al. PIK3CA hotspot mutations differentially impact responses to MET targeting in MET-driven and non-driven preclinical cancer models. Mol Cancer. 2017;16: 93. pmid:28532501
  22. 22. Samuels Y, Diaz LA, Schmidt-Kittler O, Cummins JM, Delong L, Cheong I, et al. Mutant PIK3CA promotes cell growth and invasion of human cancer cells. Cancer Cell. 2005;7: 561–573. pmid:15950905
  23. 23. Zhang L, Shi L, Zhao X, Wang Y, Yue W. PIK3CA gene mutation associated with poor prognosis of lung adenocarcinoma. OncoTargets Ther. 2013;6: 497–502. pmid:23674897
  24. 24. Janku F, Wheler JJ, Naing A, Falchook GS, Hong DS, Stepanek VM, et al. PIK3CA mutation H1047R is associated with response to PI3K/AKT/mTOR signaling pathway inhibitors in early phase clinical trials. Cancer Res. 2013;73: 276–284. pmid:23066039
  25. 25. Cancer Genome Atlas Network. Comprehensive molecular characterization of human colon and rectal cancer. Nature. 2012;487: 330–337. pmid:22810696
  26. 26. Wang L, Tsutsumi S, Kawaguchi T, Nagasaki K, Tatsuno K, Yamamoto S, et al. Whole-exome sequencing of human pancreatic cancers and characterization of genomic instability caused by MLH1 haploinsufficiency and complete deficiency. Genome Res. 2012;22: 208–219. pmid:22156295
  27. 27. Shi H, Hugo W, Kong X, Hong A, Koya RC, Moriceau G, et al. Acquired Resistance and Clonal Evolution in Melanoma during BRAF Inhibitor Therapy. Cancer Discov. 2014;4: 80–93. pmid:24265155
  28. 28. Kwabi-Addo B, Giri D, Schmidt K, Podsypanina K, Parsons R, Greenberg N, et al. Haploinsufficiency of the Pten tumor suppressor gene promotes prostate cancer progression. Proc Natl Acad Sci U S A. 2001;98: 11563–11568. pmid:11553783
  29. 29. James CD, Galanis E, Frederick L, Kimmel DW, Cunningham JM, Atherton-Skaff PJ, et al. Tumor suppressor gene alterations in malignant gliomas: histopathological associations and prognostic evaluation. Int J Oncol. 1999;15: 547–553. pmid:10427138
  30. 30. Chen J, Gomes AR, Monteiro LJ, Wong SY, Wu LH, Ng T-T, et al. Constitutively Nuclear FOXO3a Localization Predicts Poor Survival and Promotes Akt Phosphorylation in Breast Cancer. PLoS ONE. 2010;5. pmid:20808831
  31. 31. Yang D, Qi Y, Chen Q, Wang Z, Jin X, Gao J, et al. The over-expression of p53 H179Y residue mutation causes the increase of cyclin A1 and Cdk4 expression in HELF cells. Mol Cell Biochem. 2007;304: 219–226. pmid:17530187
  32. 32. Mullany LK, Wong K-K, Marciano DC, Katsonis P, King-Crane ER, Ren YA, et al. Specific TP53 Mutants Overrepresented in Ovarian Cancer Impact CNV, TP53 Activity, Responses to Nutlin-3a, and Cell Survival. Neoplasia N Y N. 2015;17: 789–803. pmid:26585234
  33. 33. Liu D, Song H, Xu Y. A common Gain of function of p53 cancer mutants in inducing genetic instability. Oncogene. 2010;29: 949–956. pmid:19881536
  34. 34. Song H, Hollstein M, Xu Y. p53 gain-of-function cancer mutants induce genetic instability by inactivating ATM. Nat Cell Biol. 2007;9: 573–580. pmid:17417627
  35. 35. Manso R, Rodríguez-Pinilla SM, González-Rincón J, Gómez S, Monsalvo S, Llamas P, et al. Recurrent presence of the PLCG1 S345F mutation in nodal peripheral T-cell lymphomas. Haematologica. 2015;100: e25–27. pmid:25304611
  36. 36. Vaqué JP, Gómez-López G, Monsálvez V, Varela I, Martínez N, Pérez C, et al. PLCG1 mutations in cutaneous T-cell lymphomas. Blood. 2014;123: 2034–2043. pmid:24497536
  37. 37. Huang S-C, Zhang L, Sung Y-S, Chen C-L, Kao Y-C, Agaram NP, et al. Recurrent CIC Gene Abnormalities in Angiosarcomas: A Molecular Study of 120 Cases With Concurrent Investigation of PLCG1, KDR, MYC, and FLT4 Gene Alterations. Am J Surg Pathol. 2016;40: 645–655. pmid:26735859
  38. 38. Murai A, Asa SA, Kodama A, Hirata A, Yanai T, Sakai H. Constitutive phosphorylation of the mTORC2/Akt/4E-BP1 pathway in newly derived canine hemangiosarcoma cell lines. BMC Vet Res. 2012;8: 128. pmid:22839755
  39. 39. Murai A, Abou Asa S, Kodama A, Sakai H, Hirata A, Yanai T. Immunohistochemical analysis of the Akt/mTOR/4E-BP1 signalling pathway in canine haemangiomas and haemangiosarcomas. J Comp Pathol. 2012;147: 430–440. pmid:22789858
  40. 40. Adachi M, Hoshino Y, Izumi Y, Sakai H, Takagi S. Effects of inhibitors of vascular endothelial growth factor receptor 2 and downstream pathways of receptor tyrosine kinases involving phosphatidylinositol 3-kinase/Akt/mammalian target of rapamycin or mitogen-activated protein kinase in canine hemangiosarcoma cell lines. Can J Vet Res. 2016;80: 209–216. pmid:27408334
  41. 41. ANDERSEN NJ, BOGUSLAWSKI EB, KUK CY, CHAMBERS CM, DUESBERY NS. Combined inhibition of MEK and mTOR has a synergic effect on angiosarcoma tumorgrafts. Int J Oncol. 2015;47: 71–80. pmid:25955301
  42. 42. Somatic mosaic activating mutations in PIK3CA cause CLOVES syndrome.—PubMed—NCBI [Internet]. [cited 8 Sep 2017]. Available:
  43. 43. Loconte DC, Grossi V, Bozzao C, Forte G, Bagnulo R, Stella A, et al. Molecular and Functional Characterization of Three Different Postzygotic Mutations in PIK3CA-Related Overgrowth Spectrum (PROS) Patients: Effects on PI3K/AKT/mTOR Signaling and Sensitivity to PIK3 Inhibitors. PloS One. 2015;10: e0123092. pmid:25915946
  44. 44. Keppler-Noreuil KM, Parker VER, Darling TN, Martinez-Agosto JA. Somatic overgrowth disorders of the PI3K/AKT/mTOR pathway & therapeutic strategies. Am J Med Genet C Semin Med Genet. 2016;172: 402–421. pmid:27860216
  45. 45. Murali R, Chandramohan R, Möller I, Scholz SL, Berger M, Huberman K, et al. Targeted massively parallel sequencing of angiosarcomas reveals frequent activation of the mitogen activated protein kinase pathway. Oncotarget. 2015;6: 36041–36052. pmid:26440310
  46. 46. Naka N, Tomita Y, Nakanishi H, Araki N, Hongyo T, Ochi T, et al. Mutations of p53 tumor-suppressor gene in angiosarcoma. Int J Cancer. 1997;71: 952–955. pmid:9185695
  47. 47. Italiano A, Chen C-L, Thomas R, Breen M, Bonnet F, Sevenet N, et al. Alterations of the p53 and PIK3CA/AKT/mTOR pathways in angiosarcomas. Cancer. 2012;118: 5878–5887. pmid:22648906