Genomic profiling reveals extensive heterogeneity in somaticDNA copy number aberrations of canine hemangiosarcoma

Rachael Thomas & Luke Borst & Daniel Rotroff &Alison Motsinger-Reif & Kerstin Lindblad-Toh &

Jaime F. Modiano & Matthew Breen

Received: 11 December 2013 /Revised: 22 January 2014 /Accepted: 23 January 2014# Springer Science+Business Media Dordrecht 2014

Abstract Canine hemangiosarcoma is a highly aggres-sive vascular neoplasm associated with extensive clini-cal and anatomical heterogeneity and a grave prognosis.Comprehensive molecular characterization ofhemangiosarcoma may identify novel therapeutic tar-gets and advanced clinical management strategies, butthere are no published reports of tumor-associatedgenome instability and disrupted gene dosage in thiscancer. We performed genome-wide microarray-based

somatic DNA copy number profiling of 75 primaryintra-abdominal hemangiosarcomas from five populardog breeds that are highly predisposed to this disease.The cohort exhibited limited global genomic instability,compared to other canine sarcomas studied to date, andDNA copy number aberrations (CNAs) were predomi-nantly of low amplitude. Recurrent imbalances of sev-eral key cancer-associated genes were evident; however,the global penetrance of any single CNAwas low and no

Chromosome ResDOI 10.1007/s10577-014-9406-z

Responsible Editor: Conly Rieder

Electronic supplementary material The online version of thisarticle (doi:10.1007/s10577-014-9406-z) contains supplementarymaterial, which is available to authorized users.

R. Thomas :M. Breen (*)Department of Molecular Biomedical Sciences, NorthCarolina State University College of Veterinary Medicine,1060 William Moore Drive, Raleigh, NC 27607, USAe-mail: Matthew_Breen@ncsu.edu

R. Thomas : L. Borst :A. Motsinger-Reif :M. BreenCenter for ComparativeMedicine and Translational Research,North Carolina State University,Raleigh, NC, USA

L. BorstDepartment of Population Health and Pathobiology, NorthCarolina State University College of Veterinary Medicine,Raleigh, NC, USA

D. Rotroff :A. Motsinger-ReifBioinformatics Research Center, Department of Statistics,North Carolina State University,Raleigh, NC, USA

K. Lindblad-TohBroad Institute, Cambridge, MA, USA

K. Lindblad-TohScience for Life Laboratory, Department of MedicalBiochemistry and Microbiology, Uppsala University,Uppsala, Sweden

J. F. ModianoMasonic Cancer Center, University of Minnesota,Minneapolis, MN, USA

J. F. ModianoDepartment of Veterinary Clinical Sciences, College ofVeterinary Medicine, University of Minnesota,Saint Paul, MN, USA

M. BreenCancer Genetics Program, University of North CarolinaLineberger Comprehensive Cancer Center,Chapel Hill, NC, USA

distinct hallmark aberrations were evident. Copy num-ber gains of dog chromosomes 13, 24, and 31, and lossof chromosome 16, were the most recurrent CNAsinvolving large chromosome regions, but their relativedistribution within and between cases suggests theymost likely represent passenger aberrations. CNAsinvolving CDKN2A, VEGFA, and the SKI oncogenewere identified as potential driver aberrations ofhemangiosarcoma development, highlighting potentialtargets for therapeutic modulation. CNA profiles werebroadly conserved between the five breeds, althoughsubregional variation was evident, including a near two-fold lower incidence of VEGFA gain in GoldenRetrievers versus other breeds (22 versus 40 %). Theseobservations support prior transcriptional studies sug-gesting that the clinical heterogeneity of this cancer mayreflect the existence of multiple, molecularly distinctsubtypes of canine hemangiosarcoma.

Keywords canine . chromosome . hemangiosarcoma .

comparative genomic hybridization (CGH)

Abbreviations

ASD Australian Shepherd DogBAC bacterial artificial chromosomeBMD Bernese Mountain DogCFA Canis familiarisCNA copy number aberrationCNV copy number variantFCR Flat-Coated RetrieverFFPE formalin-fixed, paraffin-embeddedFISH fluorescence in situ hybridizationGR Golden RetrieverGSD German Shepherd DogH&E hematoxylin and eosinoaCGH oligonucleotide array comparative genomic

hybridizationSSC saline–sodium citrate

Introduction

Canine hemangiosarcoma is a highly aggressive malig-nant neoplasm arising from cells involved in bloodvessel formation, characterized by extensive clinicaland anatomical heterogeneity, local invasiveness, and a

high incidence of local recurrence and metastatic dis-ease. Hemangiosarcoma represents up to 7 % of allmalignant tumors of the domestic dog, translating tomore than 50,000 new diagnoses in the USA each year(Vail and MacEwen 2000; Tamburini et al. 2009), mostcommonly arising in the spleen, right atrium, liver, andsubcutis. While cutaneous hemangiosarcoma has a rel-atively favorable prognosis, intra-abdominal masses fre-quently remain undetected until presentation as acute,late-stage disease with extensive necrosis, coagulopa-thies, and untreatable internal hemorrhage secondary torupture of the tumor. In the most common splenic form,median survival is typically shorter than 6 months withstandard-of-care surgical resection and intensive chemo-therapy (reviewed in Bergman 2010).

Improved strategies for early detection and effectiveclinical management are an acutely high priority for thosebreeds that are highly predisposed to visceralhemangiosarcoma. Represented among these are severalof the most popular family-owned pure breeds, includingthe Golden Retriever (GR), Labrador Retriever (LR), andGerman Shepherd Dog (GSD) (Schultheiss 2004;Thamm 2007; Bergman 2010; American Kennel Club2012). These breeds also are among the ten most com-mon contributors to the genetic ancestry of mixed-breeddogs, which themselves account for 53 % of all family-owned pet dogs within the USA (Mars Veterinary 2011).Consequently, advances in clinical management ofhemangiosarcoma may have tremendous impact acrossthe full breadth of the pet dog population. The strongassociation between genetic background andhemangiosarcoma susceptibility suggests the presenceof breed-associated heritable risk factors and molecularsignatures reflecting biologically significant characteris-tics of their underlying pathogenesis. The lifetime risk forhemangiosarcoma in the GR is remarkably high, withestimates of greater than one in five (Glickman et al.2000; Schultheiss 2004; Shearin and Ostrander 2010).The genetic history of this breed is well documented(American Kennel Club 2006), and for over 30 yearsthe GR has constituted one of the five most popularfamily-owned breeds in the USA, yielding an estimated50,000 new American Kennel Club registrations eachyear (American Kennel Club 2012). The GR thereforeoffers a comprehensive resource of naturally occurringclinical specimens and epidemiologic data in a populationwith restricted heterogeneity and well-defined ancestry,conferring vast potential for identifying tumor-associatedrisk factors and therapeutic targets for hemangiosarcoma.

R. Thomas et al.

Prior studies have described tumor-associated disrup-tion of global transcriptional profiles in caninehemangiosarcoma, and gene-inactivating sequencemutations of key pathways associated with angiogenesis(Dickerson et al. 2005; Tamburini et al. 2009, 2010).Germline risk factors are now being recognized that mayexplain the elevated susceptibility of the purebred GRpopulation to this disease (Tonomura et al., submitted).To date, however, there remains no published knowledgeof global somatic genome instability in the form of genedosage imbalance in this cancer as a potential mecha-nism for transcriptional dysregulation. We used oligonu-cleotide array comparative genomic hybridization(oaCGH) analysis to develop the first description of thelandscape of non-random tumor-associated DNA copynumber aberrations (CNAs) within primary visceralhemangiosarcoma. We explored the diversity of CNAprofiles within and between genetically distinct popula-tions through parallel evaluation of 40GR tumors and 35tumors from four additional breeds that exhibit an ele-vated incidence of this disease (Australian Shepherd Dog[ASD], Bernese Mountain Dog [BMD], Flat CoatedRetriever [FCR], and German Shepherd Dog [GSD]).In combination these findings identify the subset ofsomatic hemangiosarcoma-associated CNAs that areshared across breeds, defining genomic intervals andgenes that may play a fundamental role in disease path-ogenesis and/or represent effective therapeutic targets.Additionally, we highlight breed-associated signaturesconsistent with differential gene dosage effects involvingsignaling pathways that are intimately involved inangiogenesis. Finally, we draw comparisons withrecently reported profiles of chromosomal instability inangiosarcoma (Italiano et al. 2012b), a rare andunderstudied human counterpart for which advances inclinical management may be aided greatly by the avail-ability of a pertinent canine model.

Materials and methods

Case recruitment and histological evaluation

Clinical specimens of primary visceral hemangiosarcomawere acquired from family-owned dogs between 2003and 2012 from community or institutional veterinarypractices distributed broadly across the continentalUSA, during routine diagnostic procedures, underapproved institutional protocols and with informed client

consent. Representative biopsies were formalin fixedand paraffin embedded (FFPE) for histological evalu-ation. A diagnosis of hemangiosarcoma was con-firmed by examination of hematoxylin and eosin(H&E)-stained sections. All histological specimenswere reviewed independently by two or more board-certified veterinary pathologists, one of whom (LB)was common to all cases. Consecutive 25 μm sec-tions from each hemangiosarcoma tissue block weremacrodissected to eliminate gross regions of healthytissue by reference to the corresponding H&E slide,using dedicated equipment for each case to preventcross-contamination. Tumor DNA was isolated fromthe fixed tissue sections using a QIAamp DNA FFPETissue Kit (Qiagen, Valencia, CA) for use in oaCGHanalysis. FFPE tissue processing (based on the rec-ommendations of van Essen and Ylstra (2012))included an extended treatment with Qiagen ProteinaseK (4 days at 56 °C) and a 1-h incubation in aguanidine hydrochloride-based lysis buffer (QiagenBuffer ATL) at 90 °C. These procedures aid removalof DNA–protein cross-links created during formalinfixation, yielding DNA templates suitable for high-resolution dual-channel Agilent oaCGH platformswithout need for amplification or further fragmenta-tion (Krijgsman et al. 2012; supplementary figure 1).

Identification of tumor-associated DNA copy numberaberrations

oaCGH analysis was performed as described previously(Thomas et al. 2011) using a ∼180,000 feature microar-ray (Agilent Technologies, Santa Clara, CA) comprisingrepeat-masked ∼60-mer oligonucleotides distributed atapproximately 13 kb intervals throughout the doggenome sequence assembly (canFam version 2.0, May2005, Lindblad-Toh et al. 2005). To maximize continu-ity and aid data integration, tumor DNA from each casewas hybridized against common reference DNA sam-ples comprising equimolar quantities of constitutionalDNA from ten clinical healthy GRs of the same genderas the patient. Array image files were processed usingFeature Extraction version 10.10 and Genomic Work-bench version 7 (Agilent Technologies, Santa Clara,CA) and were then imported into Nexus Copy Numberversion 7 (Biodiscovery, El Segundo, CA). Raw datawere filtered to exclude probes exhibiting non-uniformhybridization or signal saturation. Recurrent CNAswithin each tumor were defined using the FASST2

Genomic profiling of canine hemangiosarcoma

segmentation algorithm in Nexus Copy Number, basedon a minimum of three consecutive probes with log2tumor: reference values ≥0.201 (copy number gain) or≤−0.234 (copy number loss), resulting in an effectiveresolution of ∼26 kb (two intervals of ∼13 kb). Highamplitude gains and losses within individual tumorswere defined using default log2 tumor: reference valuesof ≥1.14 and ≤−1.1, respectively.

oaCGH data for all cases were compiled using athreshold of 20 % penetrance of the cohort to definethe broadest regions of recurrent genomic imbalance.Genes and uncharacterized coding sequences withinrecurrent CNAs were defined using the UCSC caninegenome sequence browser (http://genome.ucsc.edu/)and the Gene database at www.ncbi.nlm.gov/gene.These data were filtered to flag regions of knownnatural copy number polymorphism defined by priorstudies (Chen et al. 2009; Nicholas et al. 2009). Signif-icant differences between pairwise comparisons of dis-crete subpopulations were defined as regions exhibiting≥20 % relative difference in CNA frequency withp values <0.05 based on a two-tailed Fisher’s Exact test.The ‘comparisons’ tool of Nexus was used to identifybreed-associated aberrations by sequential comparisonof genome-wide CNA penetrance in each breed againstthe mean CNA penetrance for all other breeds. TheGISTIC algorithm (Beroukhim et al. 2007) was thenused to identify genomic regions for which the frequen-cy and magnitude of genomic gain and loss was signif-icantly increased relative to the ‘background’ level,indicating CNAs that are unlikely to occur by chancealone (G score<0.05). Benjamini-Hochberg false dis-covery rate correction for multiple testing was applied(Benjamini and Hochberg 1995), yielding a peak regionof highest significance with minimum Q-bound value,flanked by a broader region with lower significance.

Discrete genomic regions are herein denoted accord-ing to their cytogenetic location and then their Mb posi-tion on that chromosome, based on the canFam v2.0(Lindblad-Toh et al. 2005) dog genome sequence assem-bly. The absence of data for the dogY chromosome in thefemale dog genome sequence assembly precluded theinclusion of data analysis for this chromosome.

Targeted fluorescence in situ hybridization analysis

Cases exhibiting discrete high level amplifications ofthe vascular endothelial growth factor A gene (VEGFA)in oaCGH were evaluated further by targeted

fluorescence in situ hybridization (FISH) analysis. FISHanalysis was performed using bacterial artificial chro-mosome (BAC) clones from the CHORI-82 canineBAC library (http://bacpac.chori.org, BACPACResources, Children’s Hospital Oakland ResearchInstitute, Oakland, CA), containing the full codingsequence of the VEGFA gene (clone 152L05) andassociated receptor (VEGFR2/KDR, clone 34E11).FISH probes for VEGFA and VEGFR2/KDR weregenerated by incorporation of fluorescent-conjugatednucleotides into BAC DNA using nick translation, andtheir unique chromosomal location was verified by con-ventional FISH analysis of chromosome preparationsfrom clinically healthy dogs (Breen et al. 2004). Thin(5 micron) sections from FFPE tissue blocks werefloated onto positively charged slides and deparaffinizedwith xylene. Specimens were then exposed to 1 h enzy-matic treatments at 37 °C with hyaluronidase type VIII(45 U/μl in 250 mM TRIS-buffered saline, pH 7.4,Sigma-Aldrich, St. Louis, MO) followed by collagenase(236 U/μl each of collagenases I and II, and 0.4 U/μl ofcollagenase III [Invitrogen/Life Technologies, Carlsbad,CA] in Hank’s Balanced Salt Solution supplementedwith 2.4 μM CaCl2). Specimens were then exposed toAbbott VP2000 Pre-Treatment Reagent (Abbott Labo-ratories, Chicago, IL) for 1 h at 80 °C, and then subjectedto further digestion for 1 h with Abbott Protease IISolution at 37 °C. Labeled probes (100 ng each) werecombined with 25 μg of sonicated dog total genomicDNA and applied to processed FFPE sections anddenatured in situ at 80 °C for 5 min. Hybridizationwas performed for 16 h at 38 °C, followed by a 2-minwash in 0.4× saline–sodium citrate buffer (SSC)/0.3 %IGEPAL (pH 7) at 73 °C and a 1 min wash in 2× SSC/0.1 % IGEPAL (pH 7) at ambient temperature. Imageanalysis was performed as described previously (Breenet al. 2004), using incremental capture (Smart Capture 3,Digital Scientific, Cambridge, UK) of successive verti-cal planes to permit visualization of probe signalsthroughout the three-dimensional depth of each cell.

Results

Genome-wide profiles of recurrent CNAs in caninehemangiosarcoma

The cohort of 75 hemangiosarcomas exhibited limitedglobal DNA copy number instability, ranging from low

R. Thomas et al.

amplitude imbalances of large chromosomal regionsand entire chromosomes to infrequent, high amplitudeaberrations of focal subchromosomal regions suggestiveof structural rearrangement. A genome-wide penetranceplot of the frequency and relative distribution of CNAswithin the cohort indicated that the global incidence ofany single CNA was low (Fig. 1a). Of 317 discreteCNAs that were classed as recurrent (shared by ≥20 %of cases), none were common to more than 45 % of thecohort, and only ten were evident in ≥25 % of cases(Supplementary Table 1). These 317 regions (164 gainsand 153 losses) ranged in length from 4.5 kb to 19.6Mb,but were heavily skewed toward small intervals (mean654 kb, median 241 kb), with the vast majority (88 %,278/317 regions) smaller than 1 Mb in size. Of the 317recurrent CNAs, 125 overlap with natural copy numbervariants (CNVs) identified in prior surveys of healthydogs (Chen et al. 2009; Nicholas et al. 2009), includingall ten CNAs that were common to more than 25 % ofhemangiosarcoma cases. No recurrent high-level copynumber amplifications or apparent homozygous dele-tions were detected within the cohort of 75 cases.

The most physical extensive CNAs present in ≥20 %of the cohort were copy number gain along the fulllength of dog chromosomes 13 (CFA13), 24, and 31,and loss of CFA16 (Fig. 1a). These events resulted inconcomitant copy number imbalance of several cancer-associated genes with distinct biological relevance tohemangiosarcoma. Three such genes lie consecutivelywithin a 754-kb region of CFA13, namely the platelet-derived growth factor receptor gene PDGFRA

(CFA13q21.3:49.7 Mb, gained in 20 % of cases), theKIToncogene (CFA13q21.3:50.1Mb, gained in 27% ofcases), and the KDR receptor tyrosine kinase gene (alsoknown as FLK1/VEGFR2, CFA13q21.3:50.3 Mb, gainedin 28 % of cases). The angiopoetin 1 gene ANGPT1,located proximally at CFA13q12.1:11.2 Mb, was gainedin 20 % of cases. Additional members of the angiopoetinfamily were also contained within recurrent CNAs, specif-ically angiopoetin 2 (ANGPT2, CFA16q25.2:61.4 Mb,20 % loss ) and ang iopoe t in 4 (ANGPT4 ,CFA24q21:23.0 Mb, 31 % gain), and the angiopoetin-like 6 gene (ANGPTL6, CFA20q16:54 Mb, 24 % gain).A distinct peak of recurrent gain was evident onCFA12q13, corresponding to increased copy number ofthe vascular endothelial growth factor gene VEGFA(CFA12q13:15.2 Mb) in 29 % of cases, of which fourcases (5 % of the cohort) were consistent with a high levelamplification. oaCGH detected deletion of theCDKN2A/Btumor suppressor gene region (CFA11q16:44.3 Mb) in28% of cases, while theCDKN2A interacting protein geneCDKN2AIP (CFA16q24:49.9) was deleted in 20 % ofcases. CNAs of other well-described tumor-associatedgenes including PDGFRB (CFA4q32), TP53(CFA5q22), MYC (CFA13q13), and PTEN (CFA26q25)were not recurrent, occurring with <20 % penetrancewithin the cohort.

These data were interrogated further using GISTICanalysis to distinguish CNAs containing candidategenes as potential drivers of tumor pathogenesis fromthe background of generalized ‘passenger’ aberrationsoccurring at random. GISTIC identified 89 discrete

% o

f cas

es

100%

100%

50%

50%

20%20%

All cases (n = 75)

1.0e -10

1.0e -10

1.0e -05

1.0e -05

1.0e +00

All cases (n = 75)

G-s

core

A

B

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 X

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 X

Fig. 1 a Penetrance plot of recurrent CNAs identified within thecohort of 75 hemangiosarcoma cases. Genomic locations areplotted along the x-axis. The y-axis indicates the percentage ofthe corresponding cohort that demonstrated either copy numbergain (shown in blue above the midline) or loss (shown in red belowthe midline) of the corresponding chromosome region. The hori-zontal bars immediately above and below the midline indicate the20 % cutoff used to define recurrent CNAs. b GISTIC analysis of

CGH profiles from all 75 cases identified 89 discrete CNAs whoseamplitude and relative distribution was significantly increasedabove background levels (G-score>1.0, Q-bound>0.05). Thepeak region of significance is highlighted in dark grey, flankedby a broader region of reduced significance shown in pale grey.Genomic locations are plotted along the x-axis, and G-scores forstatistical significance are shown on the y-axis (larger bars indi-cate higher significance)

Genomic profiling of canine hemangiosarcoma

CNAs whose amplitude and relative distribution indi-cated they were significantly unlikely to occur bychance alone (30 gains and 59 losses, G-score >1.0,Q-bound value <0.05, Fig. 1b and Supplementarytable 2). The broad genomic distribution of these regionssuggests that the extensive recurrent gains of CFA13,24, and 31, and losses of CFA16, are more consistentwith passenger rather than driver aberrations. Of the 16regions identified by GISTIC that share no apparentoverlap with known natural CNVs, the most significantwas a region on CFA12q13:15.1 Mb that flanks theVEGFA gene (CFA12q13:15.2 Mb, G-score=13.9,Q-bound score=5.83×10−7). The second most signifi-cant (G-score=13.6, Q-bound score=2.44×10−5) wasdeletion of a region onCFA11q16:44.3Mb that containsthe MTAP, CDKN2A, and CDKN2B loci, with the peakof significance localized at the CDKN2A locus. Also ofnote was a significant peak of gain on CFA5q32:60 Mb(G-score=10.9, Q-bound score=1.48×10−5), detectedin 21 % of cases, which coincides with the SKI gene(v-ski sarcoma viral oncogene homolog). The com-bination of their penetrance in the population,accompanied by their significantly non-random rela-tive distribution and amplitude and their biological func-tion, supports the intimate association of disrupted dos-age of VEGFA, CDKN2A, and SKI genes with the dis-ease phenotype.

FISH analysis of cases exhibiting high levelamplification of VEGFA showed extensive hetero-geneity in the number and distribution of VEGFAand VEGFR2/KDR probe signals within histologi-cally confirmed regions of tumor. The VEGFAprobe frequently showed clusters of intense fluo-rescent signal suggestive of tandem duplication,while the VEGFR2/KDR probe showed only mod-est increase in copy number in cases showing gain ofthis region in oaCGH (Fig. 2). Aberrant cells weretypically interspersed with pockets of cells with grosslynormal copy number, indicative of a highly heteroge-neous cell population.

Comparison of CNA profiles in GRs with other breeds

oaCGH data were subcategorized by breed as either‘GR’ (n=40) or ‘other’ (n=35) for initial assessmentof evidence for breed-associated CNAs. This analysisdemonstrated that global CNA profiles were broadlyconserved in both groups as generalized characteristicsof hemangiosarcoma, including gains of CFA13, 24,

and 31, and loss of CFA16 (Fig. 3a). The relative distri-bution of these CNAs showed subregional variation;for example, GRs exhibited recurrent deletion alongthe full length of CFA16, while in other breeds thiswas restricted primarily to the distal 75 % of this chro-mosome. Similarly, while GR cases showed broad gainalong both CFA20 and 31, in other breeds this wasrestricted to the distal 25 % of these chromosomes.The peak in incidence of gain on CFA24 was proximalin GR cases (located at ∼16 Mb) versus distal in otherbreeds (∼39 Mb). Furthermore, the peak of recurrentgain on CFA12 around the VEGFA gene was diminishedby almost half in GR cases (22 % of cases) compared tothose of other breeds (40 %). Annotated below the CNApenetrance plots in Fig. 3a are the 188 discrete regionsof significant difference between the two categories(Supplementary Table 3). Of these 188 regions, 50coincide with known natural canine copy number vari-ants (CNVs, and none overlap with cancer-associatedgenes shown to be involved in recurrent CNAswithin the full cohort of 75 cases. The majorityof these regions represented small genomic inter-vals (mean=221 kb, with 181/188 regions <1 Mbin size and the remaining seven regions <2.4 Mbin size) and were distributed widely throughout thegenome. Figure 3a shows, however, that the inci-dence of CFA31 gain was significantly elevated in

Fig. 2 Hemangiosarcoma cases exhibited VEGFA amplificationdetected by oaCGH and FISH. a Genome-wide oaCGH profilesfor each of four cases with VEGFA amplification (denoted ithrough iv) demonstrating the extensive variation in the number,size, and distribution of CNAs between cases, which are indicatedby horizontal lines above (gain) or below (loss) the midline. Thechromosomal locations of the VEGFA and VEGFR2/KDR genesare indicated by yellow and red arrows, respectively. All four casesshow a discrete amplification at the VEGFA locus on CFA12q13.The VEGFR2/KDR locus is balanced in all cases except case [iv],which shows a low-amplitude copy number gain. Case [iii] exhib-ited the sole example ofMYC amplification, evident as the spike ofhigh amplitude copy number increase on CFA13q13, proximal tothe VEGFR2/KDR locus. b Enlarged view of the CGH profile ofCFA12 in each case, indicating VEGFA amplification (dotted line).Additional disruption in copy number along the length of thechromosome is suggestive of multiple structural rearrangements.c FISH analysis revealed clusters of probe signal consistent withtandem duplication of the VEGFA locus (yellow signal). Thenumber and distribution of VEGFA probe signals varied evenwithin the same field of view, indicative of extensive heterogeneityin cell populations and stromal contamination. FISH analysis ofVEGFR2/KDR (red signal) was grossly normal in all but case [iv,pictured], which showed a minor increase in copy number consis-tent with oaCGH analysis

b

R. Thomas et al.

GRs within the region spanning CFA31q11–q15.1(4.3–28.4 Mb), reaching a maximum of 45 % ofGRs versus 20 % of other breeds. An additional

cluster of significant regions reflected the increasedincidence of deletion on CFA16qprox in GRs versusother breeds.

0

1

23

-1

-2-3lo

g2

tum

or:

refe

ren

ce

0

1

23

-1

-2-3lo

g2

tum

or:

refe

ren

ce

0

1

23

-1

-2-3lo

g2

tum

or:

refe

ren

ce

0

1

23

-1

-2-3lo

g2

tum

or:

refe

ren

ce

i

ii

iv

0

1

23

-1

-2-3lo

g2

tum

or:

refe

ren

ce

0

1

23

-1

-2-3lo

g2

tum

or:

refe

ren

ce

0

1

23

-1

-2-3lo

g2

tum

or:

refe

ren

ce

0

1

23

-1

-2-3lo

g2

tum

or:

refe

ren

ce

i

ii

iii

iv

A

B C10µm

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 3233

3435

3637

38 X

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 3233

3435

3637

38 X

iii

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 3233

3435

3637

38 X

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 3233

3435

3637

38 X

Genomic profiling of canine hemangiosarcoma

Assessment of breed-associated CNA profiles

Similar principles were then applied to genome-wideCNA profiles of the ‘other breeds’ category, by exami-nation of the oaCGH data contributed by each of thefour component breeds (Fig. 3b). The most remarkabledifferences were evident from comparison of ASDtumors against all other cases. Hemangiosarcoma casesof this breed showed significantly elevated gain of bothCFA24 (including ANGPT4) and 35, and loss of CFA33and 37. Partial deletion of the distal half of CFA16(including the CDKN2A - interact ing proteinCDKN2AIP), and gain of CFA31qdist (including thetranscription factor gene RUNX1) and subregions ofCFA13 (including KIT, KDR, and PDGFRA), were also

significantly elevated in this breed. Interestingly, ASDcases exhibited significantly increased incidence of gainboth of the platelet-derived growth factor gene PDGFA(CFA6q15, 30 % of cases) and of the receptor tyrosinekinase gene PDGFRA (CFA13q21.3, 50 % of cases),relative to other breeds. The incidence of PDGFA gainwas also significantly elevated in the GSD, along withsubregional gain along CFA6qprox and CFA20qdist,and partial deletions of CFA7 and 38. FCR cases exhib-ited a significant peak of gain flanking VEGFA onCFA12q13. While BMD cases also showed significant-ly elevated CFA12qprox gain, this region was located∼4 Mb proximal to the VEGFA locus; however, thesecases showed a significant association with deletion ofCFA5qprox and gain of CFA36qdist. Figure 4

100%

100%

100%

0%

0%

% o

f ca

ses

GR cases (n = 40)

Other breeds (n = 35)

100%

100%

0%

100%

100%

0%

% o

f ca

ses

100%

100%

0%

ASD (n = 10)

BMD (n = 6)

FCR (n = 9)

100%

100%

0%

GSD (n = 10)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 X

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 X

A

B

Fig. 3 a Penetrance plot of recurrent CNAs inGRhemangiosarcoma(n=40) compared with ‘other breeds’ (n=35). Comparison of thesedata identifies discrete regions of significant difference in their CNAprofiles. These regions are denoted by red or blue bars below theupper panel that shows the relative difference in CNA frequencybetween the two groups. b CNA penetrance plots of each non-GR

breed group, annotated to show regions of significantly different CNAidentified within each breed when compared sequentially against themean penetrance within all other cases. The highly penetrant gain onCFA8 coincides with the dog T cell receptor alpha chain immuno-globulin domain, and is evident as an apparent CNA due to therearrangement of this region in the blood-derived reference DNA

R. Thomas et al.

summarizes the CNA frequencies for genes discussed inthe text both within the full cohort and in each breed,and additional details are provided in SupplementaryTable 4.

Discussion

Genome-wide profiles of copy number imbalancein canine hemangiosarcoma

Somatic DNA copy number profiling of 75 caninehemangiosarcomas revealed no highly recurrent CNAsignatures that might be considered hallmarks of thiscancer. Rather, hemangiosarcoma exhibits a relativelylow level of global CNA compared to the extensiveaneuploidy evident in other canine sarcomas evaluatedusing similar techniques (Angstadt et al. 2011; Hedanet al. 2011; Thomas et al. 2011), with remarkably fewaberrations exceeding 20 % penetrance. This is likely aconsequence of the high degree of cellular heterogeneityassociated with hemangiosarcoma, which was evidentfrom FISH analysis in the form of discrete pockets oftumor interspersed with extensive regions of grossly

normal stromal tissue. Targeted FISH analysis of high-level amplification events also demonstrated substantialcell-to-cell variation in both probe copy number anddistribution within individual cases, suggestive of vari-able, complex structural rearrangement throughmultiplerounds of chromosome breakage and fusion, as is typi-cal of solid tumors (for example, Thompson andCompton 2011).

Despite macrodissection of fixed tissue specimens toenrich for malignant cell populations it is highly likely thatresidual microscopic heterogeneity would result in atten-uation of CNA penetrance due to contaminating stromalDNA. The variable degree of aneuploidy apparent be-tween individual cases may be associated in part withdifferential progression rates and stromal involvement;however, the extensive range in the number and distribu-tion of CNAs evident within the cohort, despite theirhistomorphologic characteristics, supports prior transcrip-tional studies suggesting the existence of several relatedbut molecularly distinct forms of hemangiosarcoma(Dickerson et al. 2005; Tamburini et al. 2009, 2010).

Recurrent imbalance of extensive genomic regions inhemangiosarcoma was restricted primarily to gain ofCFA13, 24, and 31, and loss of CFA16. Increased copy

Fig. 4 Forest plot summarizing penetrance data for genesinvolved in recurrent CNAs identified within the cohort of75 hemangiosarcoma cases, and within each breed. Genesdescribed in the text are ordered on the vertical axis according totheir physical location in the dog genome. The frequency of gain

and loss of each locus in each breed is shown on the horizontalaxis as a percentage of that population, in addition to the frequencyin the entire cohort. Red circles indicate CNA penetrance valuesthat deviate significantly in one breed compared to all other breeds(Fishers Exact test, p<0.05)

Genomic profiling of canine hemangiosarcoma

number of CFA13 is recurrent in a diverse range ofcanine cancers, including non-Hodgkin’s lymphoma,appendicular osteosarcoma, leukemia, glioma, and his-tiocytic sarcoma (Thomas et al. 2009; Angstadt et al.2011; Becker et al. 2011; Hedan et al. 2011; Thomaset al. 2011). Similarly, the incidence of CFA31 gain inhemangiosarcoma was comparable to that of a broadrange of canine cancers studied to date, includingappendicular osteosarcoma (20–35 % of cases)(Angstadt et al. 2012; Karlsson et al. 2013), glioma(35 % of cases) (Thomas et al. 2009), and B-cell non-Hodgkins lymphoma (∼20 % of cases) (Thomas et al.2011). The present data therefore add to evidence thatgain of CFA31, and of CFA13 in particular, both repre-sent generalized CNAs rather than hallmarks of specifictumor subtypes. CFA24 gain is less widely evident inprior reports of canine cancers, but occurs inhemangiosarcoma at similar frequency to that of appen-dicular osteosarcoma (30–45 % of cases, Angstadt et al.2012; Karlsson et al. 2013), histiocytic sarcoma (20–30% of cases, Hedan et al. 2011), and glioma (∼30% ofcases, Thomas et al. 2009). In contrast, among publishedstudies to date, recurrent loss of CFA16 has been report-ed only in histiocytic sarcoma (∼70 % of cases, Hedanet al. 2011), T-cell non-Hodgkins lymphoma (∼20 % ofcases, Thomas et al. 2011), and appendicular osteosar-coma (50–70 % of cases, Angstadt et al. 2012; Karlssonet al. 2013), suggesting that this CNA may be a charac-teristic of the broader category of solid tumors definedas sarcomas.

GISTIC analysis revealed a significant peak of dele-tion in hemangiosarcoma on CFA11q16 encompassingCDKN2A/B and MTAP, with maximum significancecoinciding with the CDKN2A gene, highlighting thislocus as a potential driver of tumor development. Sev-eral additional dog cancers with recurrent CFA11 dele-tion also show a focal peak in penetrance at CFA11q16,similarly indicative of a fundamental role for this locusin tumor pathogenesis (for example, T-cell non-Hodgkins lymphoma (Thomas et al. 2011), appendicu-lar osteosarcoma (Angstadt et al. 2012; Karlsson et al.2013), and histiocytic sarcoma (Hedan et al. 2011). Likehemangiosarcoma, canine histiocytic sarcoma is a high-ly aggressive and locally invasive cancer with rapidlyfatal clinical course, frequently manifesting in a dissem-inated formwithmultiple organ involvement, and with astrong breed predilection. Interestingly the MTAP andCDKN2A/B region at CFA11q16 lies within the majorgermline risk haplotype for histiocytic sarcoma in the

BMD (Shearin et al. 2012). Furthermore, a germline riskhaplotype for appendicular osteosarcoma has now beenmapped to the same region in the Greyhound (Karlssonet al. 2013). This direct association between germlinerisk and somatic aberration provides strong support forthe fundamental involvement of MTAP and/orCDKN2A/B dysregulation as a driver of these diseasephenotypes. Interestingly, however, despite recurrentdeletion of the CFA11q16 locus in hemangiosarcoma(28 % of all cases, and 30 % of all GRs), a recentgenome-wide association study of germline risk factorsfor this disease in the GR (Tonomura et al., submitted)shows no significant overlap with somatic CNA.Although 21/75 hemangiosarcoma cases exhibiteddeletion of the CDKN2A/B region, only two cases wereconsistent with homozygous deletion. While this maybe a consequence of stromal contamination, the relativepeak in CNA penetrance at this site relative to itsflanking regions was less distinct in hemangiosarcomathan in osteosarcoma, histiocytic sarcoma, or T-celllymphoma, in which homozygous deletion is frequent(Angstadt et al. 2012;Hedan et al. 2011; Thomas et al.2011). This suggests that CFA11 may harbor othermajor gene targets for hemangiosarcoma pathogenesis.The extension of these genome-wide association studiesto additional purebred dog populations will revealwhether these risk haplotypes are more closely linkedto the disease phenotype or to genetic ancestry. More-over, many recurrent CNAs overlap with CNVs identi-fied in prior studies of clinically healthy dogs (forexample Chen et al. 2009; Nicholas et al. 2009). Thereis growing evidence for the involvement of these naturalpolymorphisms in a variety of human diseases, includ-ing cancer (for example, Fanciulli et al. 2010; Shlien andMalkin 2010). CNVs are less well characterized in dogs,however, and thus at present there is insufficient evi-dence to permit any association between specific naturalpolymorphisms and the disease phenotype.

GISTIC also identified VEGFA as a likely target forthe recurrent subregional gain on CFA12q13, consistentwith prior transcriptional studies supporting the role ofthis growth factor and associated receptors in malignantproliferation of hemangiosarcoma (Yonemaru et al.2007;Tamburini et al. 2009;Tamburini et al. 2010). Aprevious study (Yonemaru et al. 2006) demonstratedelevated mRNA and protein expression of VEGFA andits receptors VEGFR1/Flt-1 and VEGFR2/Flk-1/KDR inmalignant cells of hemangiosarcoma, compared tobenign hemangioma. In the present study genomic gain

R. Thomas et al.

of VEGFR2/Flk-1/KDR occurred in 28% of all cases, incontrast to VEGFR1/Flt-1 gain in only 8 % of cases.This is consistent with the generalized description of theformer as the major mediator of the angiogenic effects ofVEGFA, although the distinction may be somewhat morecomplex (Tamburini et al. 2009). Aberrant upregulation ofVEGFA in hemangiosarcoma has also been associatedwith the presence of inactivating point mutations anddeletions within the PTEN tumor suppressor gene onCFA26q25, resulting in unregulated endothelial cellgrowth and angiogenesis (Dickerson et al. 2005). WhilePTEN deletion is highly recurrent in other dog sarcomas(Angstadt et al. 2011; Hedan et al. 2011; Thomas et al.2011), only 5%of hemangiosarcoma cases (4/75) showedcopy number loss of this locus, suggesting that genomicdeletion does not constitute an alternative mechanism forloss of PTEN function in this disease. GISTIC analysisidentified the SKI oncogene as a strong candidate for thetarget of the CFA5q32 gain in 21 % of hemangiosarcomacases. Upregulation of SKI, a negative regulator of theTGFβ signaling pathway, has been reported in a diverserange of human cancers, promoting tumor growth andinduction of angiogenesis and conferring a negative prog-nosis (reviewed in Bonnon and Atanasoski 2012). More-over, elevated SKI expression has been identified inhuman pediatric hemangiomas, with the strongest expres-sion evident in the most actively proliferating of thesebenign endothelial cell tumors (O et al. 2009). Transcrip-tional regulation of the human SKI gene is not yet wellunderstood, and the mechanism by which SKI dysregula-tion occurs in malignant cells is also unclear (Bonnon andAtanasoski 2012). To our knowledge there are no priorstudies of the SKI gene in canine tumors, and definition ofits potential relevance will therefore require focused atten-tion in future investigations.

Assessment of differential CNA profiles in GRhemangiosarcoma

Subclassification of the cohort by breed permits the firstinsight into the relationship between somatic DNA copynumber aberrations and genetic background in caninehemangiosarcoma. Comparison of all GR cases againstthose of the four other selected breeds showed that CNAprofiles are broadly conserved between these subcate-gories, but revealed that GRs exhibit significantly morefrequent gain of CFA31 and subregional loss onCFA16qprox, as well as a contrasting profile of copynumber increase on CFA24. Of particular note was the

reduced incidence of VEGFA gain in GRs (22 % of GRsversus 40 % of cases from the four other breeds). Priorstudies have indicated that transcriptional profiles ofhemangiosarcoma in the GR are distinct from those ofother breeds, highlighting differential involvement ofmembers of the VEGF pathway. Tamburini et al.(2009) reported elevated mRNA and protein expressionof VEGFR1/Flt-1 in GR hemangiosarcoma, not as aconsequence of gross elevated gene expression but rath-er through enrichment of VEGFR1/Flt-1 in conjunctionwith other functionally related signaling molecules.Targeted inhibition of VEGFR1/Flt-1 increased tumorproliferation in GR hemangiosarcoma, but was ineffectivein other breeds. Conversely, immunohistochemical stain-ing of hemangiosarcoma cell lines from non-GRs showedelevated VEGFR2/Flk-1/KDR expression compared toGR cells (Tamburini et al. 2009). These observations werenot reinforced by significant differences in the DNA copynumber profiles of eitherVEGF receptor when comparingGR tumors with other breeds in the present study. Theidentification of VEGFA as a potential driver of tumordevelopment, coupled with the almost twofold reductionin VEGFA gain in GR tumors and the highly variableincidence of both VEGFA and VEGFR2/Flk-1/KDR gainin other breeds, however, continues to highlight the fun-damental relevance of this pathway. In recent years, crit-ical advances in human medicine have yielded severalmonoclonal antibodies and tyrosine kinase inhibitors ofthe VEGF pathway, with others showing promise inclinical trials (Saharinen et al. 2011). Of particular note,a recent phase II trial showed bevacizumab, a recombinanthumanized antibody against VEGF, to be safe and effec-tive for the treatment of angiosarcoma (Agulnik et al.2012). The increasing volume of evidence for an intimateassociation between hemangiosarcoma pathogenesis andthe VEGF pathway reinforces the urgency for furtherinvestigation of therapeutic options in veterinary medi-cine, and raises the possibility for breed-associated strate-gies for clinical management.

Emerging evidence for breed-associated CNA profilesin hemangiosarcoma

The inclusion of cases from four non-GR breeds enablesthe first glimpse into the relationship between geneticancestry and somatic DNA copy number imbalance inhemangiosarcoma. FCR cases showed the most com-pelling evidence for intimate association betweenVEGFA copy number gain and tumor pathogenesis,

Genomic profiling of canine hemangiosarcoma

but were otherwise globally consistent with observa-tions across the entire cohort. Significantly elevatedCFA20qdist gain was revealed in GSD cases, whileBMD cases exhibited elevated CFA5 imbalance. ASDcases demonstrated the most complex aberration pro-files, with a series of marked deviations away from theglobal penetrance profiles of the full cohort. The globalrecurrence of CFA16 deletions and gains of CFA24shown in Fig. 1a are strongly driven by the ASD, whichin turn exhibits significantly elevated incidence of severalgenes that are fundamental to tumor pathogenesis(Fig. 4). Among these are members of the angiopoietingene family, which are central to the regulation of angio-genesis, endothelial cell survival, proliferation, andmigration, and which have been implicated inangiosarcoma development in human patients (Brownet al. 2000). The products of the ANGPT1 (CFA13q12.1)and ANGPT4 (CFA24q21) genes act as ligands for theangiopoetin tyrosine kinase receptor Tie2, which isexpressed on vascular endothelium, promoting endothe-lial cell survival and proliferation and blood vessel sta-bility. Conversely, the ANGPT2 gene product(CFA16q25.2) generally acts as an antagonist, promotingendothelial cell death and regression of blood vessels(Cascone and Heymach 2012). It is interesting to note,therefore, that the pro-angiogenic factors ANGPT1 and 4each lie on chromosomes exhibiting recurrent genomicgain in hemangiosarcoma (CFA13 and 24), with a signif-icantly increased incidence of ANGPT4 in the ASD (70 %of ASD cases versus a mean of 31 % in other breeds). Incontrast, the typically anti-angiogenic ANGPT2 ligand lieson CFA16, the chromosome with highest incidence ofglobal deletion, but showed a low incidence of copynumber loss across the cohort (20 % of all cases). Prom-ising results are now emerging using various inhibitoryfactors for regulation of angiopoetins in a wide range ofhuman cancers (Cascone and Heymach 2012). Theresults of the present study suggest that these principlesmay be applicable to hemangiosarcoma, potentially withbreed-associated efficacy. Despite global conservation intheir CNA profiles, there is, therefore, emerging evidencealso for additional breed-associated CNAs inhemangiosarcoma outwith the GR.

Canine hemangiosarcoma as a potential modelfor human angiosarcoma

Several reports have proposed hemangiosarcoma as anaturally occurring model of human angiosarcoma,

principally based on commonalities in their clinicopath-ologic features (for example, Tamburini et al. 2010;Andersen et al. 2013; Dickerson et al. 2013).Angiosarcoma, a highly vascular endothelial neoplasm,commonly arises de novo, but may also manifest as asecondary tumor subsequent to radiation therapy or inassociation with chronic lymphedema (for example Royet al. 2004). Both forms share grossly comparable clin-ical behavior and histomorphology (Billings et al.2004). As with canine hemangiosarcoma, options foreffective clinical management of angiosarcoma are lim-ited, and the prognosis is typically poor (∼30 % overall5-year survival (Fury et al. 2005)). The genomic basis ofangiosarcoma remains poorly understood, in large partdue to its relative rarity, comprising just 1–2 % of the∼10,500 new soft-tissue sarcoma cases reported eachyear in the USA (Baumhoer et al. 2005; Antonescu et al.2009; Jemal et al. 2010; Lahat et al. 2010). The relativerarity of angiosarcoma will inevitably restrict resourcesfor evaluating the efficacy of novel therapeutic agents inhuman subjects and for clinically predictive patientstratification. The extensive repository of clinical mate-rials and data associated with the high incidence ofhemangiosarcoma in defined purebred dog populationscould represent an opportunity to address limitationsresulting from the infrequency of this cancer in humanpopulations. This concept has been supported by severalstudies that demonstrate related tumor-associated tran-scriptional profiles and gene-inactivating sequencemutations of key pathways associated with angiogenesisin angiosarcoma and hemangiosarcoma, and common-alities in the potential involvement of specific micro-RNA elements (Dickerson et al. 2005; Dews et al. 2006;Yonemaru et al. 2006, 2007; Antonescu et al. 2009;Tamburini et al. 2009, 2010; Sarver et al. 2010; Guoet al. 2011; Taylor et al. 2011; Italiano et al. 2012a, b).The findings of the present study afford the first oppor-tunities to establish whether the similarities betweenangiosarcoma and hemangiosarcoma extend to theirCNA profiles. Comparison of these canine data with18 angiosarcoma profiles generated using an oaCGHplatform with comparable resolution (Italiano et al.2012b) demonstrates that both tumors present with arelatively low degree of aneuploidy, with few individualCNAs exhibiting >20 % penetrance. High-level ampli-fication of the MYC locus on human chromosome8q24.21:128.8 Mb occurs in 50–60 % of angiosarcomacases, representing one of few consensus chromosomalaberrations to be identified in this cancer. This CNA is

R. Thomas et al.

associated with concomitant transcriptional upregula-tion of the MYC oncogene, resulting in downstreaminhibition of angiogenesis (Dews et al. 2006; Manneret al. 2010; Guo et al. 2011; Italiano et al. 2012b). In thepresent study, increased copy number ofMYC occurredin only 17 % of canine hemangiosarcoma cases, andexceeded 20 % penetrance in only two breeds, the ASD(30 %) and BMD (33 %). This event was associatedalmost exclusively with gain of a broad interval onCFA13, and only a single case exhibited high level focalMYC amplification. This may in part reflect the nature ofthe canine cohort as primary tumors, since MYC ampli-fication has historically been regarded as a hallmark ofsecondary angiosarcoma (Manner et al. 2010; Guo et al.2011). A more recent, high-resolution oaCGH study hassince demonstrated that MYC amplification is alsorecurrent in de novo human angiosarcoma, and revealedsignificant conservation of genomic profiles in primaryand secondary cases, blurring their distinction at amolecular level (Italiano et al. 2012b). Aside from lackof conservation in MYC copy number status, globalcomparison of data from the present study with humanangiosarcoma data from (Italiano et al. 2012b) indicatesnegligible overlap in their CNA profiles. It seems likelythat the apparent contrast in the molecular characteristicsof hemangiosarcoma and angiosarcoma may be drivenby the relative anatomical distribution of tumors in thesepopulations. Angiosarcoma arises in a variety of skin,soft tissue, or visceral locations but occurs most com-monly as a cutaneous lesion (Lucas 2009) while themajority of canine cases manifest in the abdominal orthoracic cavities (Bergman 2010). The 18 angiosarcomacases evaluated by Italiano et al. (2012b) represent acombination of both cutaneous and soft-tissue tumorsfrom a broad range of anatomical locations, which pre-cludes the ability to perform more focused comparisonswith canine genomic data. Comparisons of histopatho-logic morphology showed that the visceralhemangiosarcoma specimens in the present study weregrossly indistinguishable from human anaplasticangiosarcoma arising from non-cutaneous sites, whileangiosarcoma arising from the skin or breast parenchy-ma typically exhibited more well-differentiated features(Italiano et al. 2012b). It is therefore possible that greatercross-species conservation in CNA profiles will beevident through comparison of canine cutaneoushemangiosarcoma with human angiosarcoma. Cutane-ous hemangiosarcoma lesions, while globally infre-quent, are prevalent in several dog breeds with short

hair and limited or variable skin pigmentation, such asthe Whippet and Dalmation (Hargis et al. 1992;Schultheiss 2004). The extension of cytogenomic stud-ies to cases from these breeds will clarify the influenceof genetic background on molecular profiles inhemangiosarcoma. In the meantime, the results of thepresent study indicate the need for more extensiveassessment of the global validity of the canine systemas a relevant model for human angiosarcoma biology.

Conclusion

Visceral hemangiosarcoma constitutes a tremendouschallenge for veterinary practitioners due to its aggres-sive and rapid clinical course, limited treatment options,and heterogeneity in anatomical location andintratumoral histomorphology. Through genomic profil-ing of 75 primary cases we demonstrate that the clini-copathologic heterogeneity of canine hemangiosarcomais recapitulated by extensive variation in the distributionof non-random somatic DNA copy number aberrations,both within and between individuals and distinct geneticpopulations. These data reveal recurrent gene dosageimbalances with direct biological relevance to tumorpathogenesis that offer potential for therapeutic modu-lation of discrete pathways involved in tumor develop-ment. This in turn provides the first opportunity forcomparative assessment of genome-wide somaticDNA copy numbe r abe r r a t i on s i n c an i nehemangiosarcoma and clinicopathologically similarhuman angiosarcoma. Both present with limited aneu-ploidy but there is little conservation in their genomicprofiles, indicating the need for further assessment of therelevance of the dog as a model for the human disease.Our findings provide an insight into the relationshipbetween somatic DNA copy number profiles and geneticancestry through parallel evaluation of cases from fivedog breeds with elevated predisposi t ion tohemangiosarcoma. The high degree of intra- andintertumoral heterogeneity in CNA profiles indicates aneed for extension to larger cohorts and additionalbreeds for refining evidence for the existence of discretemolecular subtypes.

These findings contribute toward an expandingcatalogue of complementary genomic resources directedtowards improved strategies for early detectionand effective clinical management of caninehemangiosarcoma, which will also aid comprehension

Genomic profiling of canine hemangiosarcoma

of its underlying pathogenesis and validity as a modelfor human angiosarcoma.

Acknowledgments This study was supported by grant D10CA-501 (MB, JFM, KLT) from the Golden Retriever Foundation andMorris Animal Foundation, and grant 1131 from the AmericanKennel Club Canine Health Foundation (JFM). We thank SandraHorton and staff members of the North Carolina State UniversityHistopathology laboratory for their expertise in the preparationand diagnostic evaluation of clinical specimens. Additionalpathology support was provided by Drs Gerry O’Sullivan andSandra Yi, and the Comparative Pathology Shared Resource Coreof theMasonic Cancer Center, University ofMinnesota, supportedby grant P30CA077598 from the National Institutes of Health.Wethank Mitzi Lewellyn for co-ordination of clinical specimen col-lection, Katie Kennedy for assistance with FISH analysis and KateKelley, AlexHanes Sparrow,Milcah Scott, andMegan Duckett fortechnical assistance. Enrollment and diagnostic evaluation of casesrecruited by North Carolina State University was supported by theCanine Cancer Genomics Fund. Cases recruited through Univer-sity of Minnesota were supported in part by the Starlight Fund,The Land of PureGold Foundation, theWillPower Fund, and otherphilanthropic funding at the University of Minnesota AnimalCancer Care and Research Program. We gratefully acknowledgethe many dog owners, breeders, and veterinarians who have sup-ported this study through provision of clinical specimens andpatient data.

Ethical standards Experiments described in this manuscriptcomply with the current laws of the country in which they wereperformed (USA). All institutional and national guidelines for thecare and use of laboratory animals were followed.

Conflict of interest The authors declare that they have no con-flict of interest.

References

Agulnik M, Yarber JL, Okuno SH et al (2012) An open-label,multicenter, phase II study of bevacizumab for the treatmentof angiosarcoma and epithelioid hemangioendotheliomas.Ann Oncol 24:257–263

American Kennel Club (2012) American Kennel Club 2012 DogRegistration Statistics (http://www.akc.org/reg/dogreg_stats.cfm, accessed November 24th 2013)

Andersen NJ, Nickoloff BJ, Dykema KJ et al (2013)Pharmacologic inhibition of MEK signaling prevents growthof canine hemangiosarcoma. Mol Cancer Ther 12:1701–1714

Angstadt AY, Motsinger-Reif A, Thomas R et al (2011)Characterization of canine osteosarcoma by array compara-tive genomic hybridization and RT-qPCR: signatures ofgenomic imbalance in canine osteosarcoma parallel thehuman counterpart. Genes Chromosomes Cancer 50:859–874

Angstadt AY, Thayanithy V, Subramanian S, Modiano JF, BreenM (2012) A genome-wide approach to comparative

oncology: high-resolution oligonucleotide aCGH of canineand human osteosarcoma pinpoints shared microaberrations.Cancer Genet 205:572–587

Antonescu CR, Yoshida A, Guo T et al (2009) KDR activatingmutations in human angiosarcomas are sensitive to specifickinase inhibitors. Cancer Res 69:7175–7179

Baumhoer D, Gunawan B, Becker H, Fuzesi L (2005)Comparative genomic hybridization in four angiosarcomasof the female breast. Gynecol Oncol 97:348–352

Becker SE, Thomas R, Trifonov VA et al (2011) Anchoring thedog to its relatives reveals new evolutionary breakpointsacross 11 species of the Canidae and provides new clues forthe role of B chromosomes. Chromosome Res 19:685–708

Benjamini Y, Hochberg Y (1995) Controlling the false discoveryrate: a practical and powerful approach to multiple testing. JR Stat Soc Ser B Methodol 57:289–300

Bergman P (2010) Hemangiosarcoma. In: Ettinger S, Feldman E(eds) Textbook of veterinary internal medicine, 7th edn.Elsevier, St Louis, pp 2175–2180

Beroukhim R, Getz G, Nghiemphu L et al (2007) Assessing thesignificance of chromosomal aberrations in cancer: method-ology and application to glioma. Proc Natl Acad Sci U S A104:20007–20012

Billings SD, McKenney JK, Folpe AL, Hardacre MC, Weiss SW(2004) Cutaneous angiosarcoma following breast-conservingsurgery and radiation: an analysis of 27 cases. Am J SurgPathol 28:781–788

Bonnon C, Atanasoski S (2012) c-Ski in health and disease. CellTissue Res 347:51–64

Breen M, Hitte C, Lorentzen TD et al (2004) An integrated 4249marker FISH/RH map of the canine genome. BMCGenomics 5:65

Brown LF, Dezube BJ, Tognazzi K, Dvorak HF, Yancopoulos GD(2000) Expression of Tie1, Tie2, and angiopoietins 1, 2, and4 in Kaposi’s sarcoma and cutaneous angiosarcoma. Am JPathol 156:2179–2183

Cascone T, Heymach JV (2012) Targeting the angiopoietin/Tie2pathway: cutting tumor vessels with a double-edged sword? JClin Oncol 30:441–444

Chen WK, Swartz JD, Rush LJ, Alvarez CE (2009) MappingDNA structural variation in dogs. Genome Res 19:500–509

Club AK (2006) The complete dog book, 20th edn. BallantineBooks, New York

DewsM,Homayouni A, Yu D et al (2006) Augmentation of tumorangiogenesis by a Myc-activated microRNA cluster. NatGenet 38:1060–1065

Dickerson EB, Thomas R, Fosmire SP et al (2005) Mutations ofphosphatase and tensin homolog deleted from chromosome10 in canine hemangiosarcoma. Vet Pathol 42:618–632

Dickerson EB, Marley K, Edris W et al (2013) Imatinib andDasatinib inhibit hemangiosarcoma and implicate PDGFR-beta and Src in tumor growth. Transl Oncol 6:158–168

Fanciulli M, Petretto E, Aitman TJ (2010) Gene copy numbervariation and common human disease. Clin Genet 77:201–213

Fury MG, Antonescu CR, Van Zee KJ, Brennan MF, Maki RG(2005) A 14-year retrospective review of angiosarcoma:clinical characteristics, prognostic factors, and treatmentoutcomes with surgery and chemotherapy. Cancer J 11:241–247

Glickman L, Glickman N, Thorpe R (2000) The Golden RetrieverClub of America National Health Survey http://www.grca.

R. Thomas et al.

org/pdf/health/healthsurvey.pdf, accessed 24th November2013). Golden Retriever Club of America

Guo T, Zhang L, Chang NE et al (2011) Consistent MYC andFLT4 gene amplification in radiation-induced angiosarcomabut not in other radiation-associated atypical vascular lesions.Genes Chromosomes Cancer 50:25–33

Hargis AM, Ihrke PJ, Spangler WL, Stannard AA (1992) Aretrospective clinicopathologic study of 212 dogs with cuta-neous hemangiomas and hemangiosarcomas. Vet Pathol 29:316–328

Hedan B, Thomas R, Motsinger-Reif A et al (2011) Molecularcytogenetic characterization of canine histiocytic sarcoma: aspontaneous model for human histiocytic cancer identifiesdeletion of tumor suppressor genes and highlights influence ofgenetic background on tumor behavior. BMC Cancer 11:201

Italiano A, Chen CL, Thomas R et al (2012a) Alterations of thep53 and PIK3CA/AKT/mTOR pathways in angiosarcomas:a pattern distinct from other sarcomas with complex geno-mics. Cancer 118:5878–5887

Italiano A, Thomas R, Breen M et al (2012b) The miR-17-92cluster and its target THBS1 are differentially expressed inangiosarcomas dependent on MYC amplification. GenesChromosomes Cancer 51:569–578

Jemal A, Siegel R, Xu J, Ward E (2010) Cancer statistics. CACancer J Clin 60:277–300

Karlsson E, Sigurdsson S, Ivansson E et al (2013) Genome-wideanalyses implicate 33 loci in heritable dog osteosarcoma,including regulatory variants near CDKN2A/B. GenomeBiol.14:R132

Krijgsman O, Israeli D, Haan JC et al (2012) CGH arrays comparedfor DNA isolated from formalin-fixed, paraffin-embeddedmaterial. Genes Chromosomes Cancer 51:344–352

Lahat G, DhukaAR, Hallevi H et al (2010) Angiosarcoma: clinicaland molecular insights. Ann Surg 251:1098–1106

Lindblad-Toh K, Wade CM, Mikkelsen TS et al (2005) Genomesequence, comparative analysis and haplotype structure ofthe domestic dog. Nature 438:803–819

Lucas DR (2009) Angiosarcoma, radiation-associatedangiosarcoma, and atypical vascular lesion. Arch PatholLab Med 133:1804–1809

Manner J, Radlwimmer B, Hohenberger P et al (2010) MYC highlevel gene amplification is a distinctive feature ofangiosarcomas after irradiation or chronic lymphedema.Am J Pathol 176:34–39

Mars Veterinary M (2011) Results of nation’s first ever mutt censusreveal paw print Retrieved from http://www.marsveterinary.com/blogs/news/resultsofnationsfirstevermuttcensusrevealpawprint.html Accessed 30 Nov 2013

Nicholas TJ, Cheng Z, Ventura M et al (2009) The genomicarchitecture of segmental duplications and associated copynumber variants in dogs. Genome Res 19:491–499

Roy P, ClarkMA, Thomas JM (2004) Stewart-Treves syndrome—treatment and outcome in six patients from a single centre.Eur J Surg Oncol 30:982–986

Saharinen P, Eklund L, Pulkki K, Bono P, Alitalo K (2011) VEGFand angiopoietin signaling in tumor angiogenesis and metas-tasis. Trends Mol Med 17:347–362

Sarver AL, Phalak R, Thayanithy V, Subramanian S (2010)S-MED: sarcoma microRNA expression database. LabInvestig 90:753–761

Schultheiss PC (2004) A retrospective study of visceral andnonvisceral hemangiosarcoma and hemangiomas in domesticanimals. J Vet Diagn Investig 16:522–526

Shearin AL, Ostrander EA (2010) Leading the way: caninemodelsof genomics and disease. Dis Model Mech 3:27–34

Shearin AL, Hedan B, Cadieu E et al (2012) The MTAP-CDKN2A locus confers susceptibility to a naturally occur-ring canine cancer. Cancer Epidemiol Biomarkers Prev 21:1019–1027

Shlien A, Malkin D (2010) Copy number variations and cancersusceptibility. Curr Opin Oncol 22:55–63

Tamburini BA, Trapp S, Phang TL et al (2009) Gene expressionprofiles of sporadic canine hemangiosarcoma are uniquelyassociated with breed. PLoS One 4:e5549

Tamburini BA, Phang TL, Fosmire SP et al (2010) Gene expres-sion profiling identifies inflammation and angiogenesis asdistinguishing features of canine hemangiosarcoma. BMCCancer 10:619

Taylor BS, Barretina J, Maki RG et al (2011) Advances in sarcomagenomics and new therapeutic targets. Nat Rev Cancer 11:541–557

Thamm D (2007) Hemangiosarcoma. In: Withrow S, Vail D (eds)Small animal clinical oncology, 4th edn. WB SaundersElsevier, St Louis, pp 785–795

Thomas R, Duke SE, Wang HJ et al (2009) ‘Putting our headstogether’: insights into genomic conservation betweenhuman and canine intracranial tumors. J Neurooncol 94:333–349

Thomas R, Seiser EL, Motsinger-Reif A et al (2011) Refiningtumor-associated aneuploidy through ‘genomic recoding’ ofrecurrent DNA copy number aberrations in 150 canine non-Hodgkin lymphomas. Leuk Lymphoma 52:1321–1335

Thompson SL, Compton DA (2011) Chromosomes and cancercells. Chromosome Res 19:433–444

TM O, Tan M, Tarango M et al (2009) Differential expression ofSKI oncogene protein in hemangiomas. Otolaryngol HeadNeck Surg 141:213–218

Tonomura N, Thomas R, Karlsson EK et al. Genome-wide asso-ciation study identifies shared risk loci common to twomalignancies in golden retrievers. Nat Commun. (submitted)

Vail DM, MacEwen EG (2000) Spontaneously occurring tumorsof companion animals as models for human cancer. CancerInvest 18:781–792

van Essen HF, Ylstra B (2012) High-resolution copy numberprofiling by array CGH using DNA isolated from formalin-fixed, paraffin-embedded tissues. Methods Mol Biol 838:329–341

Yonemaru K, Sakai H, Murakami M, Yanai T, Masegi T (2006)Expression of vascular endothelial growth factor, basic fibro-blast growth factor, and their receptors (flt-1, flk-1, and flg-1)in canine vascular tumors. Vet Pathol 43:971–980

YonemaruK, Sakai H,MurakamiM et al (2007) The significance ofp53 and retinoblastoma pathways in canine hemangiosarcoma.J Vet Med Sci 69:271–278

Genomic profiling of canine hemangiosarcoma