TRAF3 is genetically inactivated in a substantial fraction of cBCLs.
Focal genetic loss of TRAF3 is recurrent in human DLBCLs.
Non-Hodgkin lymphomas (NHLs) are the most common cancer to affect pet dogs. In contrast to the many genes whose mutation contributes to lymphomagenesis in humans, relatively little is known about the acquired genetic alterations that lead to canine B-cell lymphomas (cBCLs). We performed a survey of 84 canine NHL tumors to identify genes affected by somatic point mutations. We found mutations affecting TRAF3, which encodes a negative regulator of nuclear factor (NF)-κB, to be a common feature of cBCLs, with mutations observed in 44% of tumors including a combination of somatic and rare germ-line variants. Overall, 30% of the tumors contained ≥1 somatic TRAF3 mutation. The majority of mutations are predicted to cause loss of TRAF3 protein including those impacting reading frame and splicing. To determine whether TRAF3 loss might be relevant to human NHL, we also analyzed 148 human diffuse large B-cell lymphoma (DLBCL) tumors and identified loss of TRAF3 as a common event, affecting ∼9% of DLBCLs, and reduced expression of TRAF3 among deleted cases. This study implicates mutations affecting NF-κB activity as a novel genetic commonality between human and canine NHLs and supports the potential utility of cBCLs with mutated TRAF3 as a model of the more aggressive activated B-cell subgroup of DLBCL.
Domestic dogs have potential as a clinical model for a variety of human cancers. In contrast to transgenic laboratory animal models in which cancers have been induced, dogs spontaneously develop tumors at a rate comparable to humans.1 Some of the tumors that commonly arise in canines appear histopathologically similar to those of humans, suggesting that they may also share similar genetic features to their human counterparts. Additional benefits of dogs as models of human cancer include a larger body size than other model organisms, a shared living environment with humans, and a higher sequence homology of known cancer genes with humans relative to mice.2 Dogs also age five- to eightfold faster than humans, and this accelerated lifespan affords the opportunity to observe more rapid response to experimental cancer treatments.3 Before canine B-cell lymphoma (cBCL) can be considered as a relevant model of human non-Hodgkin lymphoma (NHL) for the purpose of testing investigative compounds, it is important to capture the genetic commonalities and differences between these diseases and particularly any similarities in genetic pathways that are targeted by emerging therapeutics.
NHLs collectively represent the seventh most common group of cancers among humans in the United States, and their incidence continues to rise.4,5 NHLs are the most common cancer to afflict dogs, with a strong enrichment of certain malignancies in individual breeds.6,7 Human diffuse large B-cell lymphoma (hDLBCL) is an aggressive type of NHL that can be subclassified into 2 molecular subgroups with different prognoses, each thought to derive from a distinct cell of origin (COO), namely the germinal center B-cell (GCB) and activated B-cell lymphoma (ABC) subgroups.8 The former is characterized by a gene expression profile reminiscent of cells in the germinal center, whereas the latter (more aggressive) subgroup, ABC-DLBCL, resemble B cells that are poised to exit the germinal center including a signature of constitutive nuclear factor (NF)-κB activity.9 The most common form of cBCL is a disease that bears a histopathological resemblance to hDLBCL, and recent reports have indicated the possibility of deregulated NF-κB activity as a feature of cBCL.3,10,11 These include a recent study in which gene expression analysis demonstrated widespread heterogeneity across tumors and a potential indication of 2 groups with distinct prognostic features that may parallel the COO classification used in hDLBCL.10
High-throughput DNA sequencing has enabled the discovery of numerous genes whose mutation contributes to the onset of hDLBCL with significant genetic differences observed between the GCB and ABC subgroups. For example, mutations affecting multiple histone-modifying genes are a feature of this disease with some genes, such as EZH2, exclusively mutated in GCB DLBCL.12 Other histone-modifying enzymes such as KMT2D (formerly MLL2), CREBBP, and EP300 are frequently inactivated by mutations and structural variations across both hDLBCL subgroups.13,14 Consistent with its gene expression characteristics, among the genes that are commonly mutated in ABC-DLBCL are those responsible for regulating the activity of NF-κB. These include deletions and loss-of-function mutations affecting its negative regulator TNFAIP3 and activating mutations targeting the positive regulator CARD11, and less frequently affecting TRAF2, TRAF5, MAP3K7, and TNFRSF11A.15 More recently, genetic alterations that result in constitutive B-cell receptor signaling or other modes of deregulating NF-κB activity have been identified, such as mutations affecting CD79B and MYD88.16,17 Guided by the biology of ABC-DLBCL, compounds that inhibit NF-κB either directly or indirectly show promise as therapeutics for this subtype of hDLBCL.18,19
In contrast to human aggressive NHLs, relatively little is known regarding the genetic basis of cBCLs, with only a small number of genes having been reported to be significantly mutated in these cancers. The TP53 gene has been previously described as a target of mutations in cBCL,20 and recently, the tumor-suppressor gene TFPI-2 was shown to be epigenetically silenced in a high proportion of canine DLBCL cases.21 Extensive copy number alterations have also been documented in cBCL, namely the amplification of chromosome 13, containing MYC and KIT, and the loss of regions of chromosomes 21, containing RB1, and 26, containing PTEN.22⇓⇓-25 In addition, although deregulated NF-κB activity appears to be a potential feature of a subset of cBCLs,11 specific genetic (or epigenetic) alterations affecting genes in this pathway have not been reported.
In an effort to identify genes that are frequent targets of somatic point mutations, we describe the results of applying global and targeted massively parallel sequencing to canine lymphoma tumors and matched normal tissue from multiple breeds including golden retrievers (30), Labrador retrievers (21), German shepherds (11), rottweilers (6), cocker spaniels (5), poodles (5), basset hounds (5), and one Doberman pinscher (supplemental Table 1). Although mutations affecting orthologs of genes previously implicated in hDLBCL were rare among the samples analyzed by RNA-seq and exome sequencing, we report a preponderance of mutations affecting the TRAF3 gene, which encodes a negative regulator of the noncanonical NF-κB pathway. We observed a combination of rare germ-line mutations and somatic mutations affecting TRAF3, mainly representing variants predicted to cause premature truncation of the protein. The abundance and nature of mutations affecting this gene is consistent with the notion that NF-κB activity is deregulated in a subset of cBCLs and offers a genetic mechanism for this observation in some cases. Our results reveal a commonality between a subset of cBCLs and ABC-DLBCL, supporting the potential use of the former as a clinical model of the latter and suggest utility in targeting components of the noncanonical NF-κB pathway for treatment of canine lymphoma.
Materials and methods
Eighty-four canine lymphoma tumor and matched normal tissues (skin) samples were purchased from the Canine Comparative Oncology Genomic Consortium.26 Immunophenotyping confirmed B-cell lineage for 63 of the samples. Immunophenotype information was unavailable for an additional 21 samples. TRAF3 mutations detected in the entire cohort are shown in Figure 2 and in supplemental Table 6. Table 1 reflects only the tumors with immunophenotype information available. Clinical details available for the individual samples are included in supplemental Table 1. All sample collection and data production/analysis activities adhered to procedures as approved by the Simon Fraser University Research Ethics Board, in accordance with the Declaration of Helsinki.
DNA and RNA extractions
Nucleic acid extractions were performed using 20 to 30 mg of tissue. DNA and total RNA extraction was performed on 14 tumor samples using the AllPrep DNA/RNA Universal Kit (Qiagen, Hilden, Germany). For the remaining samples, genomic DNA was obtained using the DNeasy Blood and Tissue Kit (Qiagen). All extractions were performed to the manufacturer’s specifications and automated on the QIAcube (Qiagen). DNA concentrations were determined using the Qubit dsDNA High Sensitivity Assay Kit performed with the Qubit 2.0 Fluorometer (Life Technologies, Carlsbad, CA). RNA was quantified and assessed for quality and integrity using the Agilent RNA 6000 Nano Kit on the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA).
RNA-seq and exome library preparation and sequencing
The RNA-seq and exome data produced for this study are publicly available in the sequence read archive under accession no. SRP050276 and BioProject PRJNA268531. Only RNA extractions exceeding an RNA integrity number of 8 were selected for RNA-seq library preparation. Libraries were produced from 14 of the tumor RNA extractions. Using 1 µg of total RNA as input, mRNA was enriched using the NEB magnetic mRNA isolation kit and RNA-seq libraries were prepared using the NEBNext Ultra RNA Library Prep Kit for Illumina (New England BioLabs, Ipswich, MA) according to the manufacturer’s specifications. The prepared libraries were cleaned with Agencourt Ampure XP beads (Beckman Coulter, Brea, CA).
Exome libraries were prepared from tumor and matched normal samples from 5 of the dogs chosen for RNA sequencing. Although canine-specific exome kits are being developed and validated,27 at the time of these experiments, exome kits specific to canine sequences were not commercially available. In an attempt to enrich genomic libraries for exonic regions sufficiently conserved between human and dogs, exome libraries were prepared using the Nextera human Exome Enrichment Kit (Illumina, San Diego, CA) according to the manufacturer’s specifications. Quality control of all libraries was ensured with an Agilent High Sensitivity Bioanalyzer DNA Kit (Agilent Technologies). The libraries had an average final size of 500 bp.
Seven of the RNA-seq libraries were sequenced on a MiSeq (Illumina). Libraries were sequenced individually on 150 cycle runs except for CCB010048-0100 and CCB010214-0100, which were pooled in equimolar amounts and sequenced on a single run. The remaining 7 RNA-seq libraries were pooled in equimolar amounts and sequenced with a read length of 75 bp on 1 lane of the Illumina HiSeq 2000. The exome libraries were pooled and sequenced with a read length of 125 bp on 2 lanes of the HiSeq 2000.
To detect single nucleotide variants (SNVs) and indels, FASTQ data generated from sequencing were first aligned to the canFam3 canine reference genome using Tophat228 and Star.29 SNVMix2 was used to identify potential SNVs in the RNA-seq libraries.30 GATK version 3.2 was used to identify putative variants and indels as according to GATK Best practices for RNA-seq data (http://www.broadinstitute.org/gatk/guide/best-practices?bpm=RNAseq).31⇓-33 Somatic SNVs and indels were detected in paired exome data using Strelka.34
Targeted sequencing of TRAF3 exons
Using the exon sequences of the TRAF3 transcript (ENSCAFT00000028719), available on the Ensembl genome browser (http://www.ensembl.org/index.html), and the upstream and downstream 100 bp of sequence based on the reference genome (canFam 3.1), primers were designed, using Primer3,35 to amplify each coding exon in the TRAF3 gene including canonical splice sites. All forward primers were tailed with the sequence CGCTCTTCCGATCTCTGNNNN, and all reverse primers were tailed with the sequence TGCTCTTCCGATCTGACNNNN. Thermal cycling conditions were as follows: initial incubation at 98°C for 30 seconds, followed by 32 cycles at 98°C for 10 seconds, 63°C for 30 seconds, and 72°C for 30 seconds. Additionally, a final extension step at 72°C for 2 min followed the last cycle. For each sample, all of the first-round amplification products were pooled and cleaned with Agencourt Ampure XP beads (Beckman Coulter). These resulting pooled products then served as the template for amplicon library preparation. Nested polymerase chain reaction on the amplicons was performed using primers with these sequences: forward, AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGAC GCTCTTCCGATCTCT; reverse, CAAGCAGAAGACGGCATACGAGATXXXXXXGTGACTGGAGTTCAGACGTGTGCTCTTCCG (with the region denoted as XXXXXX reserved for sample indexes used in multiplexing). Thermal cycling conditions were as follows: initial incubation at 98°C for 30 seconds, followed by 6 cycles of 98°C for 10 seconds, 65°C for 30 seconds, and 72°C for 30 seconds, and then a final extension step at 72°C for 5 minutes. The amplicon libraries were cleaned with Agencourt Ampure XP beads. The TRAF3 exon amplicon libraries were sequenced on a MiSeq (Illumina).
Following sequencing, FASTQ data for exon libraries were aligned to the reference genome using the Burrows Wheeler Aligner36 and further processed using Samtools.37 SNV calling was performed with Strelka34 using a quasi-normal set of sequences sampled from all libraries in place of true matched normal sequences. Custom scripts were used to filter out known single nucleotide polymorphisms (SNPs) and recurrent mutations present in >4 dogs that were likely to be uncharacterized SNPs, and the remaining SNVs were manually curated with the Integrative Genomics Viewer.38 Polymerase chain reactions were performed on tumor and matched normal tissues to validate putative mutations and determine if they were somatic or germ line in origin. These validation amplicon libraries were prepared as described above and analyzed manually with Integrative Genomics Viewer.
Human patient cohort copy number analysis
From the tissue repository of the Center for Lymphoid Cancer at the BC Cancer Agency, a cohort of 148 diagnostic DLBCL samples was assembled. A total of 148 Affymetrix SNP 6.0 microarrays were used to interrogate the copy number (CN) landscape. The PennCNV-affy39 protocol was used for preprocessing, followed by OncoSNP40 for predicting CN segments along with states and logR values. These CN data were then projected onto genes (Ensembl version 72).
Fluorescence in situ hybridization experiments
Using fixed cytogenetic cell suspensions (methanol/acetic acid) from formalin-fixed paraffin-embedded DLBCL samples, fluorescence in situ hybridization (FISH) was performed with in-house bacterial artificial chromosomes (BAC) using standard protocols.41 The hg19 genome was used for selecting BAC clones that were directly labeled by Nick translation (Abbott Molecular Nick translation kit). To verify the chromosomal localization of a BAC, they were hybridized to normal metaphases. The cutoff value for calling a deletion was set at >5% scoring a minimum of 200 interphase cells per case. The Carl Zeiss Axio lmager Z2 microscope equipped with a Plan Apochromat 100×/1.4 oil objective were used to analyze the slides, and images were collected acquired using a CCD camera and Metasystems software (version 5.5.1). FISH validation of TRAF3-spanning deletions was successfully performed in the 5 patients (patient identifiers: 02-16987, 04-39156, 05-19287, 99-22226, and 99-25549).
Gene expression analysis
Ninety-one of the DLBCL samples had matching RNA-seq libraries13 and were used to ascertain the gene expression profile. Specifically, GSNAP42 was used for alignment followed by the generation of the metric reads/kilobase of transcript/million mapped reads. These data were then merged to form a gene expression matrix followed by log2 transformation and quantile normalization.
We and others have previously used a combination of RNA-seq and exome sequencing for the discovery of recurrent oncogenic mutations in human cancers including DLBCL.12,13,43 To screen cBCL for mutations affecting the coding regions of genes actively expressed in tumor cells, we sequenced 14 tumor tissue samples using RNA-seq, generating between 14 and 128 million reads per sample (supplemental Table 1). For 5 of these cases, we also prepared sequencing libraries from tumor and matched constitutional genomic DNA and enriched these libraries for exonic sequences and generated an average of 57 million paired 125 nt reads for each library (supplemental Table 2). We searched the aligned reads for single nucleotide variants affecting coding regions. Because matched nonmalignant tissue was not sequenced by RNA-seq, the bulk of variants detected were likely to represent polymorphisms. To enrich these variants for potential mutations, we removed known polymorphisms identified by the dog genome project,44 as well as any variant observed in ≥3 libraries. We also selected a set of these variants for validation in tumor and matched normal DNA as detailed below.
Given the substantial number of genes known to be commonly mutated in the human B-cell lymphomas,45 we first sought to ascertain whether cBCL harbors common mutations in the orthologs of genes mutated in hDLBCL or other B-cell cancers. We compiled a list of 98 such genes based on published studies13,46⇓-48 and tabulated the nonrecurrent variants found in these genes among our discovery cohort (supplemental Table 3). Analysis of our RNA-seq data yielded few candidate mutations in the canine orthologs of these genes with a few exceptions. A small number of candidate mutations were found in TP53, MEF2C, TBL1XR1, KDM6A, CREBBP, NOTCH1, and NOTCH2. By sequencing these in tumor and matched constitutional DNA, 3 somatic mutations in TP53 and 1 in each of MEF2C, TBL1XR1, and KDM6A were confirmed, and the bulk of the remaining variants were found to be sequencing artifacts or germ line in origin (supplemental Table 3). To demonstrate we achieved sufficient sequence depth to detect mutations in orthologs of other NHL-related genes, the average coverage attained by RNA-seq for each of these genes is also indicated. Although a limiting number of tumors were sampled, this observation suggests that the driver mutations relevant to this disease may affect a largely distinct set of genes than those implicated in human malignancy, but further exploration of this question is still warranted.
To identify other genes that may be relevant to cBCL, we tabulated the SNVs and indels across all genes. We first systematically searched for potential mutation hot spots by identifying codons affected by distinct SNVs in >1 patient. A combination of manual review and targeted sequencing eliminated all potential hot spots as artifacts or germ-line variants. Genes with patterns of nonrecurrent mutations leading to premature truncations were also sought, revealing a single gene, TRAF3, harboring somatic mutations in 5 of the 14 cases, including 2 SNVs predicted to have a truncating effect on the gene product (Figure 1A). A single canonical splice site SNV was also detected in TRAF3 with effects on splicing evident in the RNA-seq data from this sample (Figure 1B). Overall, mutations predicted to alter TRAF3 function were observed in 4 (28.6%) of the cases (supplemental Table 4). These observations together implicate TRAF3 as a candidate target of recurrent mutation in cBCL but do not directly indicate whether individual variants were acquired or inherited in individual cases.
Paired exome sequencing of tumor and matched normal DNA provides a direct assessment of the somatic mutations affecting protein-coding regions. We performed exome sequencing to directly detect somatic mutations in 5 cBCL tumors, and the complete list of mutations detected by this analysis is included in supplemental Table 5. Confirming that TRAF3 mutations were somatic in nature, protein-altering variants were detected in 2 of the 5 tumors analyzed, and each variant was absent in the matched constitutional DNA (supplemental Table 5). To ascertain the prevalence of mutations affecting TRAF3, we amplified and sequenced the coding exons of TRAF3 in 50 additional cBCL cases of varied breeds, as well as an additional 21 canine lymphoma cases whose cell lineage was not confirmed. A minimum coverage of 500× was achieved for all amplicons. For variants detected using RNA-seq, exome, or targeted sequencing, subsequent amplicon sequencing was performed on the tumor and matched normal samples to validate and determine the origin of each (supplemental Figure 1). Overall, somatic TRAF3 mutations were detected in 19 of the 63 cases confirmed by immunophenotype (30.2%) and 23 of total 84 (27.4%) (Figure 2; Table 1; supplemental Table 6). Of the 27 mutations affecting these cases, 13 were SNVs predicted to result in premature truncation of the protein, and 13 were indels. Interestingly, 24 rare germ-line variants affecting TRAF3 were found in 21 samples (Figure 2; Table 1; supplemental Table 6), including 1 truncating SNV and 7 nonsynonymous SNVs. Germ-line indels causing a frameshift were also found in 10 cases. If both somatic and rare germ-line variants are included, a total of 28 (44.4%) confirmed cBCL cases and 36 (42.9%) of our total samples had ≥1 mutation affecting TRAF3.
Genetic loss of TRAF3 has been implicated in other human cancers including multiple myeloma,49,50 but the potential relevance of this gene in B-cell lymphomas has not been firmly established. To determine whether TRAF3 mutation or deletion may also be a feature of hDLBCL, we searched this gene for evidence of somatic point mutations in all publically available genome and exome sequence data13,14,47,48,51 and confirmed an absence of somatic mutations across all samples with matched constitutional DNA sequenced. We further analyzed 148 DLBCL cases for copy number alterations using the Affymetrix SNP 6.0 platform and identified 13 cases (8.9%) with deletions affecting TRAF3 and a small minimal common region of deletion affecting a limited number of genes (Figure 3A). A separate study of a larger DLBCL cohort (609 cases) confirms that this locus is a significant target of deletion.52 Using FISH, we confirmed the presence of a somatic deletion in five cases with apparent loss of this region. These include examples in which TRAF3 but not the neighboring IGH locus was lost in tumor cells (Figure 3B). Using available RNA-seq data from 91 of the samples in our cohort, we next confirmed that genetic loss of TRAF3 is associated with a significant reduction in mRNA level (Figure 3C; P = .003914059, Wilcox test). Taken together, these data strongly support the notion that TRAF3 loss, resulting in decreased gene expression, is also feature of a subset of hDLBCLs.
Murine models of cancer have been an invaluable tool for clinical scientists. The tumors in these mice, however, do not arise spontaneously but rather are induced through the alteration of specific genes or xenotransplanted human tumors into an immunocompromised background.3 As such, these models fail to replicate the vast complexity of tumors found in humans. One limitation is that these models often fail to simulate important aspects of de novo tumor development such as the complex interplay with the tumor microenvironment and immune system.53 Further, tumors that are genetically induced in animal models typically possess significantly reduced intratumoral heterogeneity relative to naturally arising neoplasms.53 The use of domesticated dogs as clinical models addresses these concerns because dogs are prone to the spontaneous development of cancers including NHL. To better exploit the potential of cBCL as a model of human cancer, similarities in the genetic underpinnings of the disease in both species must first be identified.
TRAF3 has been implicated as a tumor suppressor gene in numerous human hematologic cancers including classical Hodgkin lymphoma54 and splenic marginal zone lymphoma,55 as well as multiple myeloma and other cancers.50 Biallelic loss of TRAF3 has also been noted in a limited number of B-cell lymphomas but the prevalence of these mutations was not ascertained.56 We describe, for the first time, the frequent inactivation of TRAF3 in canine B-cell lymphomas. A combination of somatic frameshift mutations and truncating SNVs were present in 30.2% of the cBCL tumors in our cohort. Additionally, germ-line mutations affecting TRAF3 were identified in 11/63 (17.5%) of these cases, 9 (14.2%) of which lacked somatic mutations in TRAF3, possibly indicating that inherited mutations in this gene contribute to genetic predisposition to cBCL. We found no clear association between somatic or inherited TRAF3 mutations and any specific breed. Prompted by these results, we confirmed the broader relevance of TRAF3 in human malignancy by establishing the frequency of mutations affecting this gene in hDLBCL. Although we observed no point mutations in TRAF3 in hDLBCL, our data show that TRAF3 is recurrently deleted and concomitantly downregulated at the mRNA level in a subset of cases.
In the noncanonical NF-κB pathway, TRAF3 targets NF-κB-inducing kinase (NIK) for constant ubiquitination and degradation, thus serving as a negative regulator of NF-κB activity by maintaining a low baseline level of NIK.57 TRAF3 induces NIK degradation by recruiting NIK to a TRAF2-cIAP1/2 ubiquitin ligase complex.58 Consequently, downregulation of TRAF3 results in NIK stabilization and constitutive NF-κB activity via the alternative pathway. Given the established role of NF-κB in the ABC subtype of hDLBCL, the discovery of mutations affecting genes involved in this pathway was not entirely surprising. Mutations in TRAF3, however, were wholly unexpected, as this gene is not generally considered a common mutation target in hDLBCL. Notably, however, TRAF3 loss has been identified as a common feature of multiple myeloma, a cancer deriving from more mature B cells and also characterized by deregulated NF-κB activity.49
The NF-κB pathway has long been sought as a therapeutic target in human cancer,50 and our findings and the growing evidence that constitutive NF-κB activity is found in a subset of cBCLs supports that canines may be a suitable animal model for therapeutics targeting this pathway. This study supports that canine lymphomas, particularly those harboring mutations in TRAF3, may be a suitable model for early clinical evaluation of such therapies. The primary intention of this study was to systematically evaluate the potential of cBCL as a relevant clinical model for hDLBCL by searching for genetic similarities underpinning both diseases. Nonetheless, the benefits of this finding may also extend to improvements in veterinary medicine. For example, the proteasome inhibitor bortezomib, which is a common treatment of human multiple myeloma, shows a response rate of 90% in patients with low levels of TRAF3.49 Our data suggest a potential therapeutic utility of proteasome inhibitors such as bortezomib in the substantial fraction of dogs whose tumors harbor inactivating mutations in TRAF3, and this possibility should be pursued in future studies.
This study has identified a novel gene targeted in cBCL and ascertained a genetic link between human ABC-DLBCL and cBCLs. Future studies evaluating the prevalence of germ-line mutations in TRAF3 in dogs and the impact on the risk of developing cBCL in carriers of these mutations are also warranted. The detection of mutations in KDM6A (a histone demethylase that functionally opposes EZH2) and MEF2C, a gene commonly mutated in GCB DLBCL, leaves open the possibility of parallels between GCB-DLBCL and a subset of cBCLs. Additional sequencing efforts are clearly needed to determine the full extent of genetic similarities and differences between these diseases and such efforts are likely to yield further insights into the applicability of canines as clinical models for hDLBCL.
Contribution: R.D.M., K.R.B., F.C.C., C.S., and R.D.G. conceived and designed experiments; K.R.B., Y.K., A.J., M.A., S.A., S.B.-N., and D.F. performed experiments; Y.K., B.M.G., D.F., and R.D.M. performed the analysis and visualization of data; and K.R.B., Y.K., F.C.C., and R.D.M. prepared the manuscript.
Conflict-of-interest disclosure: The authors declare no competing financial interests.
Correspondence: Ryan D. Morin, Simon Fraser University, 8888 University Dr, Burnaby, BC, Canada V5A 1S6; e-mail:.
The authors thank the Canine Comparative Oncology Genomics Consortium for providing the canine tissue samples.
This work was supported by a Discovery Grant from the Natural Sciences and Engineering Research Council of Canada (NSERC; R-RGPIN-435398) (R.D.M.), a Genome Canada contract (#5125), and start-up funds provided by the BC Cancer Foundation (R.D.M.). K.R.B. is supported by a Canadian Institutes of Health Research Canada Graduate Scholarship (Master's) award. Y.K. is supported by a NSERC Undergraduate Student Research Award. R.D.M. is the recipient of a Canadian Institutes of Health Research New Investigator award. C.S. is the recipient of a Career Investigator award from the Michael Smith Foundation for Health Research. R.D.G. and C.S. are supported by a Terry Fox Program Project (#1023).
The online version of this article contains a data supplement.
The publication costs of this article were defrayed in part by page charge payment. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.
- Submitted October 1, 2014.
- Accepted November 24, 2014.
- © 2015 by The American Society of Hematology