Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Novel origins of copy number variation in the dog genome

Novel origins of copy number variation in the dog genome Background: Copy number variants (CNVs) account for substantial variation between genomes and are a major source of normal and pathogenic phenotypic differences. The dog is an ideal model to investigate mutational mechanisms that generate CNVs as its genome lacks a functional ortholog of the PRDM9 gene implicated in recombination and CNV formation in humans. Here we comprehensively assay CNVs using high-density array comparative genomic hybridization in 50 dogs from 17 dog breeds and 3 gray wolves. Results: We use a stringent new method to identify a total of 430 high-confidence CNV loci, which range in size from 9 kb to 1.6 Mb and span 26.4 Mb, or 1.08%, of the assayed dog genome, overlapping 413 annotated genes. Of CNVs observed in each breed, 98% are also observed in multiple breeds. CNVs predicted to disrupt gene function are significantly less common than expected by chance. We identify a significant overrepresentation of peaks of GC content, previously shown to be enriched in dog recombination hotspots, in the vicinity of CNV breakpoints. Conclusions: A number of the CNVs identified by this study are candidates for generating breed-specific phenotypes. Purifying selection seems to be a major factor shaping structural variation in the dog genome, suggesting that many CNVs are deleterious. Localized peaks of GC content appear to be novel sites of CNV formation in the dog genome by non-allelic homologous recombination, potentially activated by the loss of PRDM9. These sequence features may have driven genome instability and chromosomal rearrangements throughout canid evolution. Background [6], glomerulonephritis [7] and systemic lupus erythema- The fraction of genomic variation attributable to copy tosus [8]. However, there are also a small number of number variants (CNVs) is larger than single nucleotide examples of CNVs that may be beneficial, such as adap- polymorphisms (SNPs) and yet the full extent of such tive variation in copy number of the amylase gene in structural variation is still relatively unexplored [1,2]. response to diet [9], and variation in HIV/AIDS suscept- CNVs involve duplications, deletions or insertions of ibility [10]. DNA segments up to several megabases in length and are A variety of mechanisms are thought to give rise to responsible for significant phenotypic variation [3]. In CNVs [11]. A major source of structural variation is non- humans, the frequency distribution of CNVs shows sig- allelic homologous recombination (NAHR), which occurs nals of purifying selection, suggesting that a significant due to aberrant pairing of regions of extended homology. proportion of CNVs have harmful phenotypic effects [1]. Other mechanisms involve re-joining of breaks in DNA CNVs are associated with a number of genetic disorders, but do not require extensive homology. In addition to including Crohn’s disease [4], psoriasis [5], osteoporosis this, errors in replication, such as slippage at variable number of tandem repeat (VNTR) loci or insertion of * Correspondence: [email protected] transposable elements, also generate variation in copy Science for Life Laboratory, Department of Medical Biochemistry and number. CNV formation appears to occur at higher rates Microbiology, Uppsala University, Box 582, SE-751 23, Uppsala, Sweden in certain genomic regions termed rearrangement Full list of author information is available at the end of the article © 2012 Berglund et al.; licensee BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Berglund et al. Genome Biology 2012, 13:R73 Page 2 of 17 http://genomebiology.com/2012/13/8/R73 hotspots. In particular, CNVs associated with NAHR Chen et al. [29] used a 385,000 oligo array on nine dogs tend to be clustered in the genome, and CNVs are from different breed groups. They discovered 155 high enriched in the vicinity of segmental duplications. This confidence CNVs in 60 CNV regions. Nicholas et al. suggests regions of local sequence homology are hotspots [30] focused on areas of segmental duplications (SDs) of CNV formation by NAHR [12-14]. In humans, the using single dogs from 17 breeds and a gray wolf and initiation of meiotic double-stranded breaks (DSBs) is identified approximately 3,600 CNVs in approximately 700 overlapping regions found in two or more samples. thought to begin with the binding of the protein PRDM9 A subsequent study [31] used aCGH with 2.1 million to a degenerate 13-bp sequence motif [15-17]. This motif probes with an average density of 1 kb in nine dogs and is also enriched in CNV breakpoints [2], including several involved in disease [18], which implicates DSBs formed in one wolf sample and identified 403 CNVs. As expected, this way in CNV formation by NAHR. CNVs were found to be enriched in SDs. It was also Domestic dogs harbor an astonishing level of phenoty- shown that CNVs not associated with SDs were more pic variation, which is mostly apportioned into distinct likely to be present only once or at lower frequencies in breeds. The hundreds of dog breeds recognized today the dataset. An additional population genetic analysis on were formed by population bottlenecks accompanied by a set of these CNVs revealed some with divergent pat- strong artificial selection, which has led to both their terns of fixation in different breeds, which could be unique collections of characteristics and an increased responsible for breed-specific traits. prevalence of genetic disease. This makes the dog an Despite extensive efforts to type CNVs in dogs, several ideal genetic model for uncovering the genetic basis of questions remain about their mechanisms and effects. normal and pathogenic phenotypic variation [19]. Many There is evidence that meiotic DSBs localize to different traits have now been mapped in the dog genome using a sites in dogs than other mammals [27]; does this cause variety of approaches [19-22], and structural variation is a different distribution of CNVs in the genome? Is there implicated in a number of these. For example, a duplica- evidence of fixed CNVs in certain breeds that may lead tion of three fibroblast growth factor (FGF) genes causes to breed-specific phenotypes? Can variation at CNV loci the dorsal hair ridge in Rhodesian and Thai Ridgeback be used to delineate different breeds? Besides these dogs and predisposes to dermoid sinus [23], a duplication questions, a comprehensive catalogue of dog CNVs upstream of Hyaluronic acid synthase 2 (HAS2) is would be useful to aid gene-mapping studies. Here we responsiblefor thecharacteristicwrinkledskinof present the most comprehensive CNV discovery effort in dogs to date. We use a 2.1 million-probe array, as in Chinese Shar-Pei dogs and predisposes to periodic fever Nicholas et al. [31], with probes spaced on average of 1 syndrome[24],and an insertionof anFGF4retrogeneis responsible for chondrodysplasia typical of certain breeds kb. We investigate the genome-wide extent and charac- [25]. As in humans, much phenotypic variation is likely teristicsofCNVsin50 dogs from17breeds and 3 to be attributable to CNVs, which makes investigating wolves. This enables us to examine sequence features in them important for uncovering the genetic basis of phe- breakpoints at high resolution and determine patterns of notypic variation in dogs. fixation. There is a possibility that the genomic features that promote CNV formation in dogs differ from other mam- Results and discussion mals. The dog genome differs from the majority of other CNV discovery, genotyping and validation mammals in that it lacks an active copy of PRDM9, We performed aCGH analysis using a 2.1M probe tech- which suggests that formation of meiotic DSBs is con- nology platform spanning the assayable portion of the dog trolled differently in dogs [26]. A fine scale analysis of reference genome with a median spacing of 1 kb. We recombination rate variation in the dog genome indicated restricted our analysis to identifying high-confidence that, like in humans, recombination is clustered into hot- CNVs containing at least 10 probes, which allows identifi- spots but that, unlike in humans, these regions were cation of CNVs down to approximately 9 kb in length. We strongly enriched for short regions (approximately 1 kb) assayed CNVs in 53 samples, comprising purebred dogs of highly elevated GC content (GC peaks) [27]. This sug- from 17 breeds plus 3 wolf samples. Two breeds were gests GC peaks may be targets of meiotic DSBs. Interest- represented by ten unrelated individuals each to enable ingly, GC rich regions also seem to be involved in CNVs segregating at lower frequencies in these breeds to genome rearrangements during canid genome evolution, be identified, whereas the other breeds were represented where they have relocated to telomeric regions [28]. This by two individuals to maximize coverage of different could indicate that GC peaks are important targets of breeds. A male Boxer was used as the reference sample. NAHR and often involved in rearrangements. CNVs were analyzed in autosomes and × chromosome by Three studies have identified canine CNVs using array comparing the ratio of signal intensities between test sam- comparative genomic hybridization (aCGH) [29-31]. ples and the reference. Berglund et al. Genome Biology 2012, 13:R73 Page 3 of 17 http://genomebiology.com/2012/13/8/R73 We identified CNVs using a three-stage procedure In total, this procedure detected 430 CNV loci distribu- comprising segmentation, identification of CNV loci, and ted along the chromosomes (Figure 1). Out of these loci, genotype calling. We first performed a comprehensive 226 were classified as deletions (53%) and 104 as duplica- comparison of five segmentation algorithms, including tions (24%), meaning that the only variants identified at NimbleGen, DNAcopy, Ultrasome, pennCNV and these loci were deletions or duplications, respectively, cghFLasso, and selected the algorithm most robust to relative to the reference, whereas 100 CNV loci exhibited both deletions and duplications (23%) among samples. In noise (see Materials and methods and Supplementary addition, 77 of the loci exhibited a complex deviation methods in Additional file 1). This comparison identified from the reference in at least one sample (Table 1). cghFLasso as the most accurate method, while other methods resulted in excessive segmentation of the signal Across all calls at all loci, 70.2% matched the reference, intensity ratio in samples with noisy data, large discre- whereas 28.4% exhibited a deviation consistent with a pancies in CNV numbers between samples and lack of a single deletion or duplication (17.7% single deletions and strong correlation between levels of variation in CNVs 10.7% single duplications). Only 1.4% of calls were of compared with SNPs (Supplementary methods, Tables greater magnitude (Figure S6 in Additional file 1). The S1 andS2and FiguresS1toS5inAdditionalfile 1).We finding that deletions are more numerous than duplica- therefore used this algorithm to perform segmentation of tions is generally observed in studies using aCGH signal intensity ratios compared to a reference for sam- [1,12,29-32]. This may reflect the greater relative diffi- ples in our dataset. culty of identifying duplications due to the smaller rela- Identification of CNV loci was performed with a tive change in copy number (3:2 versus 2:1) and also the method to estimate the absolute copy number compared fact that insertions of sequence not present in the refer- to the reference based on fixed thresholds and chromo- ence will not be detected. Also concordant with recent some-specific variance in each sample. The principle of studies, duplications were found to be larger than dele- our method is that fixed thresholds are used as a baseline tions with a median size of 30 kb versus 19 kb (Table 1). cutoff, but samples with very high variance use a higher This could suggest that duplications are less likely to be threshold whereas samples with low variance use lower severely deleterious than deletions and therefore less thresholds (Materials and methods). We first identified likely to be purged by purifying selection. Alternatively, it segments falling above a stringent threshold, designed to could reflect a bias against detecting small duplications. exclude false positives, to identify CNV loci in single But despite deletions being smaller, their higher inci- dence indicates that deletions and duplications affect samples. The final set of CNV loci is the union of indivi- similar proportions of the genome. dual calls at each locus, which were merged across sam- ples into asingleCNV.The breakpointsweredefined by On average, 130.9 loci (30%) differ from the reference the outermost boundaries of all individual CNV calls at per sample, and among breeds with two samples an each locus. After identification of loci, the genotype of average of 172.4 loci (40%) differ from the reference in each individual was inferred at each locus using less one or both of the samples (Table 1). In total 50 (12%) stringent criteria to determine the most likely state of CNVs were detected only once in the dataset. This is an each sample (described in Materials and methods ). This average of one singleton per sample, and is consistent method allowed us to distinguish which samples exhib- across samples from breeds with ten and two samples. ited deletions and duplications relative to the reference. We observe one CNV that differs from the reference in Each change was then categorized as simple or complex all samples, indicating that it is a singleton in the refer- based on variance between samples: changes where the ence (chromosome 15, 57.86 to 57.88 Mb). This corre- mean value exceeds the threshold were considered sim- sponds well with the number of singletons found in ple, whereas changes where the mean value is less than other samples, suggesting these numbers are accurate. the threshold but the mean deviation from zero exceeds However, this contrasts with a recent study by Nicholas the threshold were considered complex. This happens et al.[31] using thesamearray,whichidentified403 especially when the segmentation algorithm fails to dis- CNV loci in nine dogs and a wolf, of which 260 (65%) cern two adjacent CNVs, of which one is a deletion and were detected in a single sample (an average of 26 per the other is a duplication (where mean value is close to sample). This previous study used only one sample per zero, but mean deviation from zero is large). Loci where breed, which makes identifying CNV loci as singletons at least one sample exhibited a complex change from the more likely. However, the fact that we also find that most CNV loci are shared between breeds suggests that reference were defined as complex. It is important to differences in CNV calling contribute to differences in note that it is not possible to discern how the copies are CNVs found in each study. distributed between alleles using aCGH on diploid sam- ples, that is, distinguish if a CNV is heterozygous or We validated our set of CNVs using two complemen- homozygous. tary methods: quantitative PCR (qPCR) and analysis on Berglund et al. Genome Biology 2012, 13:R73 Page 4 of 17 http://genomebiology.com/2012/13/8/R73 Figure 1 The genomic architecture of CNVs. Black lines represent all 38 canine autosomes and the × chromosome. Deletions are plotted as red rectangles below each chromosome and duplications are plotted as blue rectangles above each chromosome. the CanineHD high-density SNP genotyping array. A total of 53 sample-locus combinations were tested Because of the high overlap (50%) with previously iden- and 3 tests did not match the state predicted from tified and confirmed CNVs (see ‘Distribution and geno- aCGH (1 false positive and 2 false negatives; 94% mic effects of CNVs’), we focused our validation on concordance). novel CNVs detected only in our study, and performed All of the samples used for aCGH were also geno- qPCR on four loci ranging in size from 17 to 80 kb. typed on the Illumina CanineHD array, which contains Table 1 Number of CNVs identified Deletions Duplications Both All Total CNV loci (complex) 226 (19) 104 (16) 100 (42) 430 (77) Mean CNV loci per breed 100.5 68.5 3.4 172.4 Mean CNV loci per sample 77 53.8 0 130.9 Median size (kb) 19 29.5 27.8 24.3 Deletions are defined as CNV loci that only exhibit reduction in copy number relative to the reference assembly. Duplications are loci that only exhibit increased copy number. ‘Both’ refers to loci that exhibit both deletions and duplications. Based on 15 breeds with 2 samples. Berglund et al. Genome Biology 2012, 13:R73 Page 5 of 17 http://genomebiology.com/2012/13/8/R73 >174,000 probes designed for assaying SNPs with an the RNA-binding region gene RNPC3 and upstream of the average spacing of 13 kb [33]. The probe density of this collagen genes COL11A1 and COL5A1. more than ten times sparser array permits identification The second largest locus is a complex CNV on chromo- of larger CNVs (greater than approximately 100 kb with some 9 between 20.1 and 21.6 Mb. There are no RefSeq our filters) compared to the aCGH chip. In total, 13 annotated dog genes in this region, although it is ortholo- CNV loci passed our filtering procedure and had gous to the human genes VPS13D, RDM1, ARHGAP27 enough probes on the SNP array to be used in the vali- and members of the LRRC37 family. These genes are not co-located in human, so this region must represent syn- dation procedure (Table S3 in Additional file 1). These teny to multiple loci. RDM1 is near members of the 13 CNVs, with an average size of 266 kb, included 8 sin- gletons, 3 two-sample CNVs and 2 CNVs at midrange TBC1D3 family, which shows primate-specific expansion frequency. In total there were 6 mismatches between via SDs on chromosome 17, and is present in a majority of calling on the aCGH and CanineHD SNP array among human-specific breakpoints of conserved synteny to 689 genotypes (1 false positive and 5 false negatives), mouse [36]. Both LRRC37 and ARHGAP27 are ortholo- which is a correspondence of >99% between individual gous to regions flanking an inversion in the human gen- genotypes. This suggests that the genotyping error is ome at 17q21.31, thought to have arisen through NAHR smaller than the 6% estimated from the qPCR, at least between large blocks of flanking SDs, which are distribu- for large CNVs. ted throughout chromosome 17 and contain the LRRC37 gene family [37]. This region is associated with a micro- Distribution and genomic effects of CNVs deletion leading to mental retardation and has undergone We compared our dataset to CNVs previously identified multiple complex rearrangements during primate evolu- in dog in multiple studies merged and augmented by tion. The finding that this region is also a large and com- Nicholas et al. [31]. We identify 216 overlaps, which con- plex CNV in dogs suggests that it may be a region of tain 196 (of 615) of previously identified CNVs and 213 instability across a wide range of mammalian evolution. (of 430) of ours (Figure 2). This overlap is highly signifi- Both of the two discussed loci were also present in pre- cant compared to random redistributions of CNVs in the vious dog aCGH reports [29-31]. genome (P < 0.001). The reason that some CNVs do not CNVs overlapping SDs occur at higher frequencies in overlap between studies is likely to be a combination of the population (P <0.001), aremorelikely tobecom- differences in breed selection, sample size, array resolu- plex (P < 0.001), tend to be longer (P < 0.001), and are tion, genotyping algorithms and errors. Furthermore, we more likely to overlap genes (P <0.001) than CNVsnot associated with SDs (bootstrapping used for all signifi- generated a canine segmental duplication map using a cance tests), which confirms previous observations [31]. modified version of the method of Bailey et al.[34] (described in [35]), to which both datasets were mapped. Both singletons (80%) and breed-specific CNVs (85%) The Venn diagram in Figure 2 shows that more than a are more likely to fall in the category of CNVs not over- third (165) of our CNVs overlap these SDs and more lapping SDs. Non-SD CNVs have an average frequency than half (319) of Nicholas’ CNVs overlap SDs, which is of 0.24 (0.33 of breeds), whereas CNVs inside of SDs a highly significant overlap (P < 0.001, random redistribu- have an average frequency of 0.4 (0.56 of breeds), which tion test) considering <5% of the dog genome is com- is almost twice as high (P < 0.001). Complex CNVs are posed of recent SDs. The high proportion of CNVs in preferentially observed in SDs, where every third CNV SDs together with our non-targeted approach indicates a is complex, compared to less than 10% complex CNVs large involvement of SDs in CNV formation as indicated outside SDs. These observations may reflect a higher by previous studies. However, it should also be noted mutation rate of CNVs in SDs, with recurrent events that a small proportion of SDs are likely not fixed in the around the same genomic location leading to both dog genome and may actually be CNVs. higher frequencies and more complexity, but could also The largest CNV locus we identified in the dog genome reflect the involvement of SDs in more complex rearran- is located at 48.5 to 50.0 Mb on chromosome 6. The pat- gements. CNVs in SDs are also larger than non-SD tern of variation in this region is consistent with a deletion CNVs; with a median size of 40 versus 20 kb. The in all samples except both Labradors and one Boxer, increased frequency and complexity of CNVs in SDs which match the reference sequence. Inspection of probe may reflect the dynamic nature of SDs, and that these intensities in this region on the CanineHD array shows a CNVs have arisen from overlapping but distinct events. pattern consistent with the presence of duplicated We next examined the functional effects of CNVs by identifying genes they overlap. In order to use a high- sequence in these Labrador and Boxer samples. Consider- confidence gene set, 24,232 dog genes annotated under ing the high frequency of the deletion allele, it is most Ensembl ID were filtered with the g:Orth tool from the likely that the duplication allele is the derived state. The CNV encompasses a gene-sparse region downstream of g:Profiler website to extract only human-dog 1:1 Berglund et al. Genome Biology 2012, 13:R73 Page 6 of 17 http://genomebiology.com/2012/13/8/R73 Figure 2 Comparison of LUPA dataset with a summary of CNVs presented by Nicholas et al. [31]and a list of segmental duplications in dog [53]. The Venn diagram shows the number of overlapping CNVs from the datasets. Since the datasets contain different entries, numbers are colored according to which dataset the counted entries belong to. For example, 196 of Nicholas et al.’s CNVs overlap with 213 of our (LUPA) CNVs, and 120 and 132, respectively, of these also overlap with segmental duplications. orthologs. Out of 15,258 1:1 orthologous genes, 130 In addition to their paucity, CNVs overlapping genes are genes (0.85%) are overlapped either completely or par- characterized by a much higher proportion of duplications tially by 90 CNVs (out of 430 total CNVs). This number than deletions (Table 2). In intergenic CNVs, deletions are is significantly less than predicted by chance (P <0.001, more than three times as common as duplications, random redistributions per chromosome). A great pro- whereas within genes they occur at a proportion of two- -8 portion of the affected genes, 53 (41%), had their entire thirds compared with duplications (P = 3.1 × 10 ; Fisher’s coding sequence covered by a CNV (Table 2). This exact test). This pattern is even more pronounced in intra- could suggest that CNVs overlapping genes are more genic CNVs that are predicted to remove a stop codon, likely to have deleterious effects. A test of the size distri- where deletions occur at a proportion of 0.2 relative to bution of CNVs affecting genes revealed that they are duplications, which is significantly lower than other intra- larger, with median size of 36 versus 24 kb (bootstrap- genic CNVs (P = 0.009; Fisher’s exact test). This tendency ping, P < 0.005). They also show a slight tendency for duplication enrichment among stop codons has pre- towards lower frequencies, although the difference is viously been detected in humans [1], and suggests a strong not significant (bootstrapping, P < 0.1). deleterious effect of removal of stop codons. Berglund et al. Genome Biology 2012, 13:R73 Page 7 of 17 http://genomebiology.com/2012/13/8/R73 Table 2 Number of CNV loci covering genomic regions Intragenic Partial gene Total Intergenic Total Whole gene Total Stop codon Total 430 340 90 31 67 29 Deletion 226 198 28 8 21 4 Duplication 104 62 42 13 33 19 Both 100 80 20 10 12 6 Deletion:duplication ratio 2.17 3.08 0.67 0.62 0.64 0.21 All groups are non-overlapping. Subcategories are not additive as a CNV can span several genes and harbor both deletions and duplications in different samples. The set of 130 1:1 orthologous genes overlapping CNVs genes they overlap: no functional categories are inferred to were scanned for enrichment of gene ontology (GO) cate- be enriched in both deletions and duplications. For exam- gories against a background of all 15,258 1:1 human-dog ple, the ‘homophilic cell adhesion’ category is only enriched in deletions, whereas ‘olfactory receptor activity’ orthologs (Table 3). This gene set was chosen because of the higher accuracy of annotations for human genes. For is only enriched in duplications. The enrichment of speci- this purpose the g:GOSt tool from g:Profiler website was fic GO categories could reflect changes in patterns of used. The most significantly enriched term in each domain selective constraint in dogs [38], positive selection for dog- was the biological process ‘homophilic cell adhesion’ (P = specific traits, or tolerance of certain gene categories to -13 7.31 × 10 ), the cellular component ‘integral to mem- deletions or duplications. -5 brane’ (P = 6.07 × 10 ) and the molecular function ‘olfac- -6 tory receptor activity’ (P = 6.22 × 10 ). Cell adhesion also Mechanisms of CNV formation appears as a strongly enriched category among human We searched for repeats that were enriched close to break- CNVs [1,32]. In our dataset, CNVs are also enriched for points of CNVs, calculating the observed to expected ratio genes involved in the Kyoto Encyclopedia of Genes and to identify over-represented motifs. There is uncertainty in Genomes (KEGG) pathway ‘olfactory transduction’ (P = precisely defining breakpoint location because of smooth- -3 5.04 × 10 ). Analyzing simple deletion and simple dupli- ing of probe intensities and experimental noise. Each cation loci separately reveals a difference in the kind of breakpoint location was defined as a 10 kb window to Table 3 Enriched gene ontology categories in genic CNVs P-value Number of genes All CNVs Deletions Duplications GO ID Domain All Deletions Duplications Gene ontology term 9.48e-02 - 3.12e-03 GO:0032787 BP 5 0 5 1. Monocarboxylic acid metabolic process 5.00e-06 6.32e-04 1 GO:0022610 BP 14 7 1 1. Biological adhesion 5.00e-06 6.32e-04 1 GO:0007155 BP 14 7 1 2. Cell adhesion 2.67e-11 1.36e-06 1 GO:0016337 BP 14 7 1 3. Cell-cell adhesion 7.31e-13 2.37e-07 1 GO:0007156 BP 14 7 1 4. Homophilic cell adhesion 3.56e-03 6.29e-02 1 GO:0071944 CC 18 8 2 1. Cell periphery 2.68e-04 1.76e-01 1 GO:0016020 CC 47 16 20 1. Membrane 2.15e-03 4.88e-02 1 GO:0005886 CC 18 8 2 2. Plasma membrane 3.37e-04 6.08e-01 3.64e-01 GO:0044425 CC 40 13 19 1. Membrane part 8.74e-05 2.49e-01 3.45e-01 GO:0031224 CC 39 13 18 2. Intrinsic to membrane 6.07e-05 2.17e-01 2.92e-01 GO:0016021 CC 39 13 18 3. Integral to membrane 5.97e-02 1 3.82e-02 GO:0004872 MF 16 4 12 1. Receptor activity -1 2.66e-02 GO:0038023 MF 18 3 11 2. Signaling receptor activity 1.96e-02 1 1.27e-02 GO:0004888 MF 16 3 11 3. Transmembrane signaling receptor activity 1.49e-03 1 5.21e-04 GO:0004930 MF 15 3 11 4. G-protein coupled receptor activity 6.22e-06 1 9.88e-06 GO:0004984 MF 12 2 9 4. Olfactory receptor activity 3.18e-04 7.11e-05 1 GO:0005509 MF 17 10 1 1. Calcium ion binding 5.20e-01 - 9.98e-03 GO:0005506 MF 6 0 6 1. Iron ion binding 5.04e-03 1 6.44e-03 KEGG:04740 ke 7 1 5 1. Olfactory transduction Significant P-values are in bold. P-values were determined using a hierarchical multiple testing procedure. Indented numbers before the GO terms indicate relative depth of the term in the hierarchy. BP, biological process; CC, cellular component; MF, molecular function; ke, KEGG pathway. Berglund et al. Genome Biology 2012, 13:R73 Page 8 of 17 http://genomebiology.com/2012/13/8/R73 account for this imprecision. The list of known repeats to analyze enrichment of GC content in breakpoints. We was downloaded from the RepeatMasker track of the found a more than two-fold enrichment in CNV break- UCSC genome browser. Nearly all CNV breakpoints over- points (P < 0.001), which rapidly decays with increasing lap some repeat family, with L1, ERV[1/L], RNA and satel- distance from the breakpoint, both within and outside of lite DNA being overrepresented (Table S4 in Additional the CNV (Figure 3). We also found an enrichment of file 1). The last two types involve fewer than ten break- CpG islands and gaps in CNV breakpoints (P < 0.001; Table S5 in Additional file 1). CpG islands are usually points, and cannot be considered a significant contribution GC-rich, and the great overlap with GC peaks is some- to CNV formation. However, >96% of the CNV break- what expected. Gaps in the reference genome assembly points contain LINE elements, an excess of 36% compared with expected coverage, and almost 65% of the CNV have shown an extremely high likelihood of being asso- breakpoints contain a long terminal repeat (includes the ciated with CNVs in human [32]. In dog, gaps tend to endogenous retrovirus (ERV) family) entry. We found a correlate with high GC content because these regions are 1.5-fold excess of L1 elements in CNV breakpoints. Inter- less well captured during the genome sequencing, and estingly this excess seems to be most prominent for many of them could qualify as GC peaks if they were younger L1 repeats (Table 4). This follows the pattern of fully characterized. The association between CNV break- CNVs and SDs detected in humans [32], where the likeli- points and GC peaks in dogs suggests that CNV break- hood of a SD being associated with a CNV was highly cor- points may often occur at recombination hotspots, where related to its sequence similarity to the duplicated copy, as DSBs have a tendency to form, followed by their repair younger L1 elements are likely to have increased levels of by NAHR. sequence homology with each other. The association with GC peaks and CNVs has not been The pattern of L1 enrichment together with the signifi- reported in other species. In humans, a 13-bp sequence cant overlap between SDs and CNVs is consistent with a motif targeted by the PRDM9 protein is strongly asso- large contribution of NAHR, which is known to operate ciated with recombination hotspots, which also associates on highly similar copies, and supports the generation of with CNVs [18]. In dogs, the PRDM9 gene is inactive and CNVs in duplicated regions and regions with mobile ele- recombination hotspots are strongly associated with GC ment insertions. L1 transduction, where additional 3’ peaks. The association between GC peaks and CNV flanking sequence is transferred to a new genomic loca- breakpoints may, therefore, be an additional consequence tion together with an L1 insertion, may also contribute to of the death of PRDM9 [27]. There is a strong overlap this enrichment. We do not find any enrichment of CNV between CpG islands and GC peaks, and both are enriched close to CNV breakpoints. This could indicate breakpoints around SINEs. This is in concordance with some studies in humans [39,40] that find associations that non-methylated DNA promotes DSB formation that with L1 but not Alu at CNV breakpoints. This may sug- leadstostructuralvariation.However,itisalsopossible gest that LINE elements promote structural variation that it is simply GC-richness that promotes recombina- throughNAHRmorestronglythanSINEs inthedog tion, or that the peaks of high GC content are mainly a genome. consequence of elevated recombination rate due to GC- As GC peaks are enriched in recombination hotspots in biased gene conversion rather than its cause, which are the dog genome and may be important for formation of detected as CpG islands regardless of methylation [41]. DSBs [27], we searched for enrichment of this sequence These results are particularly interesting in light of the feature within CNV breakpoints. We first defined GC suggestion that GC-rich regions have acted as novel tar- peaksasinAxelsson et al. [27], where a GC peak is get sites of chromosomal fissions during canid evolution recorded when a 500-bp sliding window centered in a [28]. 10 kb background sliding window has a 1.5-fold increase We also searched for regions of extended perfect in GC content. We then performed randomization tests homology between pairs of breakpoints flanking CNV Table 4 Excess of L1 repeats in CNV breakpoints a b Name Excess Divergence Length Number of repeats in Number of repeats in Number of breakpoints with (bp) genome breakpoints repeats L1_Cf 3.92 0.035 1,192 16,127 178 137 L1_Canis 2.54 0.088 799 72,587 678 309 L1_Canid 1.87 0.144 534 42,070 254 135 L1_Carn 1.25 0.168 525 132,806 652 275 Eutheria 0.91 0.246 359 589,155 2046 590 a b The proportion of repeat coverage in breakpoints versus whole genome. Average divergence of the repeat from its reference in RepBase. Berglund et al. Genome Biology 2012, 13:R73 Page 9 of 17 http://genomebiology.com/2012/13/8/R73 Figure 3 Enrichment of GC peaks around CNV breakpoints. The strongest excess of GC peaks compared to random expectations is observed close to CNV breakpoints and decays with increasing distance from the breakpoint. The midpoint of the breakpoint is at position 0, the shaded area to the left of this represents points within the CNV, and the area to the right represents flanking sequence. loci. Of the 430 CNV loci, 86 have runs of perfect (133.1 bp compared with 116.6 bp). This suggests that sequence homology greater than 75 bp and 11 have NAHR is not restricted to segmental duplications and stretches longer than 1 kb. The mean length of runs of may be equally or more common outside of them. perfect homology between breakpoints is 126.8 bp. We tested the significance of these stretches of homology by CNV distribution among breeds and samples simulating 1,000 datasets with the same number of loci We next analyzed the distribution of CNVs among breeds and distance between breakpoints, located at randomly (Table 5). The majority (341, 79%) of CNVs were found in chosen positions on the same chromosomes. The mean several breeds, while 89 CNVs (21%) were breed-specific. length of homology in the simulated dataset was 35.0 The breed-specific CNVs are larger (32 kb) than CNVs bp, with none of the simulated means exceeding the found in multiple breeds (22 kb) (P < 0.005, bootstrap- observed one (P < 0.001). These findings provide addi- ping). Some variation in the number of CNVs between tional support for an important role of NAHR in dog samples was seen both within and between breeds. Nota- CNV formation. Surprisingly, the mean length of bly, the total number of CNVs identified in Boxers was homology between breakpoints of CNVs overlapping lower than in any other breed, with an average of 64.5 loci SDs is slightly shorter than those outside of SDs different from the reference per sample, largely due to the Berglund et al. Genome Biology 2012, 13:R73 Page 10 of 17 http://genomebiology.com/2012/13/8/R73 Table 5 CNVs identified in each breed and sample compared to the reference genome Total CNV loci per breed Average CNV loci per sample Breed Samples Total MB BS Total MB BS Border Terrier (BTe) 2 156 154 2 128.5 126.5 2 Boxer (Box) 2 103 102 1 64.5 64 0.5 Cavalier King Charles Spaniel (CCS) 2 189 182 7 148.5 143.5 5 Chihuahua (Chi) 2 181 178 3 142 140.5 1.5 Dachshund (Dac) 2 180 172 8 131 127 4 English Cocker Spaniel (ECS) 2 166 165 1 129 128.5 0.5 English Springer Spaniel (ESS) 2 214 207 7 160 156 4 Finnish Spitz (FSp) 2 186 176 10 148.5 140.5 8 German Shepherd (GSh) 2 163 160 3 126 123.5 2.5 Labrador Retriever (LRe) 2 170 168 2 129 128 1 Nova Scotia Duck Tolling Retriever (NSD) 2 193 190 3 147.5 145.5 2 Poodle (Pdl) 2 182 180 2 130 129 1 Sarloos (Sar) 2 175 168 7 135 130 5 Schnauzer (Sch) 2 169 163 6 127 123.5 3.5 Swedish Elkhound (Elk) 2 159 158 1 116.5 115.5 1 Average (2-sample breeds) 2 172.4 168.2 4.2 130.9 128.1 2.8 Golden Retriever (GRe) 10 264 255 9 121.6 118.8 2.8 Irish Wolfhound (IrW) 10 229 215 14 136.2 127.7 8.5 Average (10-sample breeds) 10 246.5 235 11.5 128.9 123.3 5.7 Wolf (Wlf) 3 163 160 3 96.3 95.3 1 MB, multi-breed (CNVs found in more than one breed); BS, breed-specific (CNVs found in only one breed). reference being a Boxer. Of the remaining breeds, the are present per breed, making it less likely that three or average number of CNVs per sample varied from 116.5 more samples share a CNV. A broadly similar frequency (Swedish Elkhound) to 160 (English Springer Spaniel). On distribution is seen within breeds. Figure 4b shows the average, a sample differs from the reference at 130.9 CNV allele frequency distribution with ten genotyped indivi- loci, of which 2.8 are specific to that breed. Fewer than 6% duals analyzed on a breed basis (identity of the minor allele is defined from the two-sample breeds of the entire of CNVs found in any one breed are specific to that breed. These patterns are broadly consistent in the wolf samples, dataset). which exhibit a slightly lower than average number of The two breeds for which ten individuals were analyzed CNVs per sample. On average, 2 dogs from the same were selected for their large (Golden Retriever) and small breed differ at 83.1 CNV loci whereas 2 dogs from differ- (Irish Wolfhound) population sizes, respectively. We first ent breeds differ at 103 CNV loci. attempted to see how many CNVs were present in all ten A more detailed picture of polymorphism in breeds with dogs of a breed, suggesting fixation (Table 7). For Golden two samples is given in Table 6. We find that an average Retrievers, 20 CNVs were present in all dogs, and for of 172.4 CNV loci (40%) are observed in one or both of Irish Wolfhounds, 38 CNVs appeared fixed, possibly the samples in any one breed and an average of 4.2 (2.4%) reflecting the slightly higher degree of inbreeding in Irish are only found in that breed (private CNVs). Similar num- Wolfhounds. A much smaller number of breed-specific bers of CNVs are found in one sample (polymorphic) or loci were identified, and no cases of breed-specific fixed both samples (fixed) of these breeds. An average of one CNVs were identified in these deeply sampled breeds. private CNV is fixed (found in both samples) in the The relative lack of breed-specific fixed CNVs suggests breeds, although with a small sample size, the majority of that instances of those involved in breed-specific pheno- these are likely to be polymorphic. types must be rare. Overall, the CNV frequency distribu- To assess the frequency distribution of CNVs among tions appear qualitatively similar to those expected for samples, we used their presence in all samples to build a neutral polymorphisms. site frequency spectrum. Figure 4a shows the minor allele We scanned our dataset for CNVs that were fixed for frequency distribution across all samples in two-sample one allele in some two-sample breeds and another allele in all other two-sample breeds (that is, are not poly- breeds. There is a marked drop in frequency above two morphic in any breed). There are 24 such CNVs, of samples, which can be attributed to the higher within compared to between breed fixations, where two samples which all but one is specific to a single breed. This CNV Berglund et al. Genome Biology 2012, 13:R73 Page 11 of 17 http://genomebiology.com/2012/13/8/R73 Table 6 Polymorphic CNVs in breeds with two samples Loci matching Shared loci between breeds Private loci single breeds Breed Samples reference Total Polymorphic Fixed Total Polymorphic Fixed Border Terrier 2 274 154 55 99 2 0 2 Boxer 2 327 102 76 26 1 1 0 Cavalier King Charles Spaniel 2 241 182 77 105 7 4 3 Chihuahua 2 249 178 75 103 3 3 0 Dachshund 2 250 172 90 82 8 8 0 English Cocker Spaniel 2 264 165 73 92 1 1 0 English Springer Spaniel 2 216 207 102 105 7 6 1 Finnish Spitz 2 244 176 71 105 10 4 6 German Shepherd 2 267 160 73 87 3 1 2 Labrador Retriever 2 260 168 80 88 2 2 0 Nova Scotia Duck Tolling Retriever 2 237 190 89 101 3 2 1 Poodle 2 248 180 102 78 2 2 0 Sarloos 2 255 168 76 92 7 4 3 Schnauzer 2 261 163 79 84 6 5 1 Swedish Elkhound 2 271 158 85 73 1 0 1 Average 2 257.6 168.2 80.2 88 4.2 2.9 1.3 Wolf 3 267 160 121 39 3 3 0 Total, number of CNVs observed in the breed; Polymorphic, average number of polymorphic CNV sites between samples of the breed; Fixed, number of CNVs found in both samples of the breed. is a 12.7 kb deletion (located at 55.5 Mb on chromo- clusters on a branch leading to Sarloos (a wolf hybrid) some 6) shared by Cavalier King Charles Spaniel and and German Shepherd, as observed in a previous SNP English Springer Spaniel, which overlaps the gene analysis [33]. There is also some evidence for clustering DPYD, which encodes the enzyme dihydropyrimidine by breed type as demonstrated by vonHoldt et al. [42]. dehydrogenase (DPD) involved in pyrimidine catabolism. One cluster contains spaniels (English Cocker Spaniel, Examples of breed-specific fixations include a 32.7 kb English Springer Spaniel, Cavalier King Charles Spaniel), deletion on chromosome 28 in German Shepherds scent hounds (Dachshund) and toy dogs (Chihuahua), downstream of DUSP10, which is involved in immune and another contains retrievers and terriers (Golden responses and mediates various physiological processes, Retriever, Labrador Retriever, Border Terrier). However, and a 18.1 kb deletion on chromosome 21 in Finnish there are exceptions present, and in general, this analysis Spitz immediately downstream of CYP2R1 (cytochrome is limited by a smaller number of loci and less precise p450, family 2, subfamily R, polypeptide 1) and overlap- calling than the studies based on SNP arrays. ping PDE3B (phosphodiesterase 3B, cGMP-inhibited). These regions are good candidates for governing breed- Conclusions specific characteristics, although further investigation is This study provides insights into the mutational mechan- necessary to determine if they are really fixed, or have isms and functional effects of CNVs in the dog genome. any phenotypic effect. Our results suggest that many CNVs are generated by NAHR events directed towards peaks of GC content, Breed relationships which is consistent with observations that these sequence We explored the extent to which patterns of CNV varia- features are also enriched in dog, but not human, recombi- tion can be used to infer population structure between nation hotspots. Hence, GC peaks may represent novel breeds. Based on the proportion of shared CNVs between sites of elevated recombination and genome instability in each pair of samples, a neighbor-joining phylogeny was dogs. This shift in recombinational activity towards GC constructed (Figure 5). With a few exceptions, all samples peaks in dogs is likely to be due to the lack of a functional clustered together with samples from the same breed. copy of the PRDM9 gene, which initiates recombination at This indicates that the CNVs have a strong phylogenetic separate specific sequence motifs in humans. In support of signal, grouping dogs into breeds as shown by large-scale a strong role of NAHR in dog CNV formation, we also SNP analyses [33,42]. Ability to construct this tree identify associations between CNV breakpoints and L1 demonstrates the high accuracy of dataset and high con- elements and long stretches of sequence homology. We gruency with SNP data. Notably, one of the wolf samples also show that dog CNVs are affected by the signal of Berglund et al. Genome Biology 2012, 13:R73 Page 12 of 17 http://genomebiology.com/2012/13/8/R73 (a) (b) Figure 4 Minor allele frequency distribution of CNVs. (a) Minor allele frequency distribution of CNVs compared across all 15 breeds with a sample size of two. (b) Minor allele frequency distribution of CNVs within the two breeds with a sample size of ten. purifying selection and identify candidate CNVs for invol- Materials and methods vement in breed-specific characteristics. This comprehen- Sample collection sive catalogue of CNVs will be useful for future studies to EDTA-blood was collected as part of the LUPA project uncover the genetic basis of complex traits in dogs. [43] from pedigree dogs around Europe and USA with Berglund et al. Genome Biology 2012, 13:R73 Page 13 of 17 http://genomebiology.com/2012/13/8/R73 Table 7 Fixed CNVs in breeds with ten samples Loci matching Shared loci between breeds Private loci in single breed Breed reference Total Polymorphic Fixed Total Polymorphic Fixed Golden Retriever 166 255 235 20 9 9 0 Irish Wolfhound 203 215 177 38 14 14 0 Average 184.5 235 206 29 11.5 11.5 0 Total, number of CNVs observed in the breed; Polymorphic, average number of polymorphic CNV sites between samples of the breed; Fixed, number of CNVs found in both samples of the breed. owners’ consents: a total of 50 dogs from 17 breeds Cocker Spaniel, Finnish Spitz, German Shepherd, Labra- including 2 unrelated individuals from the breeds Border dor Retriever, Nova Scotia Duck Tolling Retriever, Poo- Terrier, Boxer, Cavalier King Charles Spaniel, Chihua- dle, Sarloos Wolfhound, Schnauzer and Swedish hua, Dachshund, English Springer Spaniel, English Elkhound; and 10 unrelated individuals from Golden Figure 5 Neighbor-joining tree based on allele sharing at CNV loci. Box, Boxer; BTe, Border Terrier; CCS, Cavalier King Charles Spaniel; Chi, Chihuahua; Dac, Dachshund; ECS, English Cocker Spaniel; Elk, Swedish Elkhound; ESS, English Springer Spaniel; Fsp, Finnish Spitz; GRe, Golden Retriever; GSh, German Shepherd; IrW, Irish Wolfhound; LRe, Labrador Retriever; NSD, Nova Scotia Duck Tolling Retriever; Pdl, Poodle; Sar, Sarloos; Sch, Schnauzer; Wlf, Wolf. Berglund et al. Genome Biology 2012, 13:R73 Page 14 of 17 http://genomebiology.com/2012/13/8/R73 Retriever and Irish Wolfhound breeds. In addition, three The fixed genotyping ratio was set to correspond to a gray wolves from Belarus, Spain and Finland, respec- deviation of 0.5 copies from the reference, while the tively, were included. A Finnish male Boxer was used as fixed discovery ratio was set to the slightly higher 0.65 the reference sample. Genomic DNA was purified using copies deviation. The standard deviation genotyping commercial purification kits and the quality of the DNA value was chosen to include 5% of the data points, while was analyzed by spectrophotometry and agarose gel the standard deviation discovery value was chosen to electrophoresis prior to the hybridization experiment. include 0.1% of the data points, from the theoretical normal distribution (any copy number variants should be clearly distinct from the distribution). These were Discovery and genotyping aCGH was used to detect DNA copy number alterations chosen so that a similar number of samples used the using NimbleGen’s canFam2 Whole Genome CGH oligo fixed values and the standard deviation values. The dis- array platform with 2.1 million probes on a single slide covery threshold was picked as the maximum of these and a median probe spacing of 1 kb. The array design is two numbers, while the genotyping threshold was based on the annotated CanFam2.0 genome sequence of picked as the minimum of the two numbers. This a female Boxer genome. The isothermal 50-75mer means that no CNVs are identified in the first stage if probes were evenly distributed throughout the unique the deviation from the reference is below 0.65 copies, sequence of the genome. The genomic DNA samples andinthefinalstage allCNVswithadeviationabove were sent to NimbleGen’s service facility where the 0.5 copies from the reference are identified as CNVs. hybridizations were performed in a two-color format according to Selzer et al. [44]. Copy number was quanti- Validation fied from the fluorescence ratios of the two dyes. Nim- aCGH results were experimentally validated by qPCR in bleGen conducted the initial data processing from randomly selected CNV loci. Relative copy numbers of normalization to signal calling and quantification. Ratios the selected regions were determined in comparison to were log2 transformed, and positive log2 ratios indicate the reference sample (Finnish male Boxer). Regions gains and negative log2 ratios indicate loss of copy num- were selected based on the aCGH profiles across breeds. ber. Copy number for each called CNV was calculated qPCR experiments were performed on ABI Prism 7500 (1+mean log2ratio) as 2 and rounded to whole numbers. Fast instrument (Applied Biosystems, Stockholm, Swe- CNV calling was conducted in three stages: smooth- den) using SYBR Green detection chemistry according ing, segmentation and selection. Prior to segmentation to the manufacturer’s instructions. Primers (available and selection, triangular smoothing was done on the upon request) were designed inside CNVs using Primer3 ratios, which implements an 11-point weighted smooth- and NCBI primer design programs. Each assay was per- ing function along the chromosomes. This is equivalent formed in triplicate using 20 μl reactions containing 10 to several passes of fewer-point rectangular smoothings μl of qPCR master mix (Roche, Stockholm, Sweden), 10 (unweighted sliding-average), and is more effective at nM concentration of dNTPs, 250 nM concentration of reducing high-frequency noise in the signal. Segmenta- forward and reverse primer and 10 ng of genomic DNA. tion was done with the R package cghFLasso to segment Amplification was performed under the following condi- a continuous distribution of intensity ratios into discrete tions: one cycle at 50°C for 2 minutes, one cycle at 95°C regions of consecutively different ratios. From these for 10 minutes, 40 cycles at 95°C for 15 seconds and putative CNV segments, those with significant deviation 62°C for 45 seconds. Beta-actin and GAP175 were used are selected as the final set of CNVs. The selection was as controls. Serial dilutions were performed for each performed in three stages: first, ascertain CNV locations assay to estimate the PCR efficiency (E) prior to analysis. per sample using a stringent threshold (discovery The ddC method was used for the quantification of threshold); second, consolidate the calls from all indivi- copy numbers in test individuals relative to the same duals by mapping them onto one ‘master genome’ to get reference Boxer sample used in the aCGH experiments. merged general CNV locations; and third, decide the The C values for each set of triplicates were averaged CT state of the general CNVs in each breed using a less and adjusted for PCR efficiency (E) as log2(E ). The stringent threshold (genotyping threshold) to assign C values were then normalized against the control pri- individuals to diploid copy number classes. Noisy data mers. The relative copy number for each site was calcu- -(t-r) were handled by recalculating the ratios into values of lated as 2 , where t = normalized C for the test standard deviation from a theoretical normal distribu- sample and r = normalized C for the reference sample. tion of ratios. These were used to allow samples with lit- Large CNVs, by virtue of their potential functional tle noise to utilize lower thresholds. The two thresholds impact, were validated in an additional validation step. were then individually chosen on chromosome basis The same set of individuals was genotyped according to from a fixed log2 ratio and a standard deviation value. manufacturer’s instructions on the CanineHD 170K SNP Berglund et al. Genome Biology 2012, 13:R73 Page 15 of 17 http://genomebiology.com/2012/13/8/R73 array (Illumina) with a resolution of 1 SNP per 13 kb. fold increase in GC content compared to background, a The GenomeStudio V2010.3 software package (Illumina) GC peak was marked for all base pairs in the window. was used to obtain normalized total signal intensity, Log The program Neighbour from the Phylip package was R ratio, and B allele frequencies for all SNPs according to used to construct the breed phylogeny from a distance the manual and Peiffer et al. [45]. Exported Log R ratio matrix containing the pairwise differences between the and B allele frequencies for every SNP were used in the breeds for all CNV loci. subsequent CNV calling. The CNV calling was GO analysis was made with the web-based tool g: GOSt Gene Group Functional Profiling provided by g: performed using QuantiSNP [46]. Default settings in Profiler (formerly known as GOSt, Gene Ontology Sta- QuantiSNP were used, that is, L = 2M, expectation-maxi- mization iterations =10, and the parameter file levels-hd. tistics), [47], with default parameters. This tool not only dat. Samples having a standard deviation of the Log R searches for enrichment in GO categories, it also looks ratio above 0.35 were removed from the subsequent ana- for enrichment in KEGG/REACTOME pathways and lysis (three samples). Furthermore, CNVs having less TRANSFAC regulatory motifs/MicroCosm microRNA than five SNPs or a Log Bayes factor <10 were removed. sitesaswellasHuman PhenotypeOntologyand Bio- The Log Bayes factor is a score that represents the sup- GRID protein-protein interaction networks. The set of port for the existence of the CNV and a Log Bayes factor orthologous genes was extracted with the tool g:Orth of at least 10 will result in up to 10% false positive calls. from the same project, which maps orthologous genes in related organisms using Ensembl alignments. Statistical and population genetics analysis Statistical significance of overlap between genomic Prior to analysis, all chromosomes were centered on their features was assessed by estimating the distribution of mean log2 ratio to remove potential chip biases. The × sample means. This was done by either random redistri- chromosome was treated differently in this manner, since butions per chromosome or bootstrapping. Chromo- the pseudo-autosomal regions were centered separate to some-wise redistributions were done by repeatedly and the rest of the chromosome, which differs in relative randomly redistributing all features on the same chro- copy number to a male reference. To identify the mosome to get a distribution of means from which sig- pseudo-autosomal regions, the raw log2 ratios from all nificance can be assessed. Bootstrapping was done by female samples were used to determine the average posi- re-sampling the individual observations with replace- tion of the probe where copy number changed from ment from the population of CNVs to get an empirical diploidy. Since no CNVs were successively called in that bootstrap distribution from which a bootstrap confi- dence interval could be derived to infer statistical particular region, the position seems to be accurate. significance. The metric used to call CNVs from the aCGH data was the log2 ratio, which is the log (base 2) ratio of the R was used to smooth the aCGH data and the R pack- observed normalized R-value for a signal intensity age cghFLasso was used to assess copy number altera- divided by the mean signal of a reference sample. The tions. For genome annotation the University of Santa threshold of log2 ratio was defined as in the discovery Cruz (UCSC) Genome Browser [48] and Ensembl Gen- andgenotypingsectionto defineatruecopynumber ome Browser [49] were used. Gene and exon coordi- change that presents a pattern consistently different from nates were downloaded from Ensembl website version a diploid region. CNVs were defined as |log2 ratio| 65. Disease statuses of genes were obtained from the >threshold, where deletions are indicated by negative Online Mendelian Inheritance in Man database [50]. log2 ratios and duplications are indicated by positive log2 ratios. Deletion and duplication variants are special cases Access to data of multi-allelic CNVs. To minimize the risk of false posi- A database of genomic copy number variants, tives, we required each locus to be targeted by at least DoG_CNV, in the dog genome has been developed to ten consecutive probes to call a CNV, resulting in a final provide the annotation of CNVs discovered in this study resolution of approximately 9 kb. and a useful resource to assist with the assessment of Segmental duplications, defined as regions at least 1 kb CNVs in the contexts of canine variation and disease in length, at least 90% identical at two or more loci, and susceptibility [51]. In addition, the raw array data have not consisting entirely of mobile elements, were identi- been submitted to the Gene Expression Omnibus at fied by self-alignment of the genome as described in NCBI [52]. Zody et al. [35]. In the GC peak analysis we identified GC peaks by sliding two windows along the genome; a 10 Additional material kb window for background rate (same size as a CNV breakpoint) and centered in this a 500 bp window for Additional file 1: Supplementary methods, tables and figures. peak discovery. When the peak window showed a 1.5- Berglund et al. Genome Biology 2012, 13:R73 Page 16 of 17 http://genomebiology.com/2012/13/8/R73 6. Yang TL, Chen XD, Guo Y, Lei SF, Wang JT, Zhou Q, Pan F, Chen Y, Abbreviations Zhang ZX, Dong SS, Xu XH, Yan H, Liu X, Qiu C, Zhu XZ, Chen T, Li M, aCGH: array comparative genomic hybridization; CNV: copy number variant; Zhang H, Zhang L, Drees BM, Hamilton JJ, Papasian CJ, Recker RR, Song XP, DSB: double-stranded break; GO: gene ontology; KEGG: Kyoto Encyclopedia Cheng J, Deng HW: Genome-wide copy-number-variation study of Genes and Genomes; LINE: long interspersed nuclear element; NAHR: identified a susceptibility gene, UGT2B17, for osteoporosis. Am J Hum non-allelic homologous recombination; qPCR: quantitative polymerase chain Genet 2008, 83:663-674. reaction; SD: segmental duplication; SINE: short interspersed nuclear element; 7. Aitman TJ, Dong R, Vyse TJ, Norsworthy PJ, Johnson MD, Smith J, SNP: single-nucleotide polymorphism. Mangion J, Roberton-Lowe C, Marshall AJ, Petretto E, Hodges MD, Bhangal G, Patel SG, Sheehan-Rooney K, Duda M, Cook PR, Evans DJ, Acknowledgements Domin J, Flint J, Boyle JJ, Pusey CD, Cook HT: Copy number polymorphism This study was supported by the European Commission FP7 project LUPA- in Fcgr3 predisposes to glomerulonephritis in rats and humans. Nature GA201370 [43], the Swedish Research Council Formas and a European Science Foundation EURYI award to KLT. We thank all of the dog owners 2006, 439:851-855. who contributed samples used in this project. 8. Yang Y, Chung EK, Wu YL, Savelli SL, Nagaraja HN, Zhou B, Hebert M, Jones KN, Shu Y, Kitzmiller K, Blanchong CA, McBride KL, Higgins GC, Author details Rennebohm RM, Rice RR, Hackshaw KV, Roubey RA, Grossman JM, Tsao BP, Science for Life Laboratory, Department of Medical Biochemistry and Birmingham DJ, Rovin BH, Hebert LA, Yu CY: Gene copy-number variation Microbiology, Uppsala University, Box 582, SE-751 23, Uppsala, Sweden. and associated polymorphisms of complement component C4 in human Department of Basic Veterinary Sciences, Department of Medical Genetics, systemic lupus erythematosus (SLE): low copy number is a risk factor for Program in Molecular Medicine, Folkhälsan Institute of Genetics, Biomedicum and high copy number is a protective factor against SLE susceptibility in Helsinki, University of Helsinki, PO Box 63, 00014 Helsinki, Finland. European Americans. Am J Hum Genet 2007, 80:1037-1054. Department of Animal Breeding and Genetics, Swedish University of 9. Perry GH, Dominy NJ, Claw KG, Lee AS, Fiegler H, Redon R, Werner J, Agricultural Sciences, Box 597, SE-751 24 Uppsala, Sweden. Broad Institute Villanea FA, Mountain JL, Misra R, Carter NP, Lee C, Stone AC: Diet and the of Harvard and Massachusetts Institute of Technology, 7 Cambridge Center, evolution of human amylase gene copy number variation. Nat Genet 5 6 Cambridge, Massachusetts 02142, USA. www.eurolupa.org. Institut de 2007, 39:1256-1260. Génétique et Développement de Rennes, CNRS-UMR6290, Université de 10. Gonzalez E, Kulkarni H, Bolivar H, Mangano A, Sanchez R, Catano G, Rennes 1, Rennes, France. Nibbs RJ, Freedman BI, Quinones MP, Bamshad MJ, Murthy KK, Rovin BH, Bradley W, Clark RA, Anderson SA, O’Connell RJ, Agan BK, Ahuja SS, Authors’ contributions Bologna R, Sen L, Dolan MJ, Ahuja SK: The influence of CCL3L1 gene- JB performed the majority of the bioinformatics analyses and drafted the containing segmental duplications on HIV-1/AIDS susceptibility. Science 2005, 307:1434-1440. manuscript. EMN performed the array experiments. A-MM analyzed the SNP array data. MP helped to coordinate the array experiments. CA participated 11. Hastings PJ, Lupski JR, Rosenberg SM, Ira G: Mechanisms of change in gene copy number. Nat Rev Genet 2009, 10:551-564. in useful discussions. MCZ and TS identified segmental duplications. CH 12. Graubert TA, Cahan P, Edwin D, Selzer RR, Richmond TA, Eis PS, contributed bioinformatics analyses and constructed the website. KL-T and Shannon WD, Li X, McLeod HL, Cheverud JM, Ley TJ: A high-resolution HL conceived of the study and participated in its design and coordination. map of segmental DNA copy number variation in the mouse genome. MTW conceived and coordinated the bioinformatics analyses and wrote the PLoS Genet 2007, 3:e3. paper. All authors read and approved the final version. 13. Sharp AJ, Locke DP, McGrath SD, Cheng Z, Bailey JA, Vallente RU, Pertz LM, Clark RA, Schwartz S, Segraves R, Oseroff VV, Albertson DG, Pinkel D, Competing interests Eichler EE: Segmental duplications and copy-number variation in the The authors declare that they have no competing interests. human genome. Am J Hum Genet 2005, 77:78-88. 14. She X, Cheng Z, Zollner S, Church DM, Eichler EE: Mouse segmental Received: 23 April 2012 Revised: 27 June 2012 Accepted: 23 August 2012 Published: 23 August 2012 duplication and copy number variation. Nat Genet 2008, 40:909-914. 15. Myers S, Bowden R, Tumian A, Bontrop RE, Freeman C, Macfie TS, McVean G, Donnelly P: Drive against hotspot motifs in primates References implicates the PRDM9 gene in meiotic recombination. Science 2010, 1. Conrad DF, Pinto D, Redon R, Feuk L, Gokcumen O, Zhang Y, Aerts J, 327:876-879. Andrews TD, Barnes C, Campbell P, Fitzgerald T, Hu M, Ihm CH, 16. Baudat F, Buard J, Grey C, Fledel-Alon A, Ober C, Przeworski M, Coop G, de Kristiansson K, Macarthur DG, Macdonald JR, Onyiah I, Pang AW, Robson S, Massy B: PRDM9 is a major determinant of meiotic recombination Stirrups K, Valsesia A, Walter K, Wei J, Tyler-Smith C, Carter NP, Lee C, hotspots in humans and mice. Science 2010, 327:836-840. Scherer SW, Hurles ME: Origins and functional impact of copy number 17. Parvanov ED, Petkov PM, Paigen K: Prdm9 controls activation of variation in the human genome. Nature 2010, 464:704-712. mammalian recombination hotspots. Science 2010, 327:835-835. 2. Mills RE, Walter K, Stewart C, Handsaker RE, Chen K, Alkan C, Abyzov A, 18. Myers S, Freeman C, Auton A, Donnelly P, McVean G: A common sequence Yoon SC, Ye K, Cheetham RK, Chinwalla A, Conrad DF, Fu Y, Grubert F, motif associated with recombination hot spots and genome instability Hajirasouliha I, Hormozdiari F, Iakoucheva LM, Iqbal Z, Kang S, Kidd JM, in humans. Nat Genet 2008, 40:1124-1129. Konkel MK, Korn J, Khurana E, Kural D, Lam HY, Leng J, Li R, Li Y, Lin CY, 19. Karlsson EK, Lindblad-Toh K: Leader of the pack: gene mapping in dogs Luo R, Mu XJ, et al: Mapping copy number variation by population-scale and other model organisms. Nat Rev Genet 2008, 9:713-725. genome sequencing. Nature 2011, 470:59-65. 20. Karlsson EK, Baranowska I, Wade CM, Salmon Hillbertz NH, Zody MC, 3. Hurles ME, Dermitzakis ET, Tyler-Smith C: The functional impact of Anderson N, Biagi TM, Patterson N, Pielberg GR, Kulbokas EJ, Comstock KE, structural variation in humans. Trends Genet 2008, 24:238-245. Keller ET, Mesirov JP, von Euler H, Kampe O, Hedhammar A, Lander ES, 4. McCarroll SA, Huett A, Kuballa P, Chilewski SD, Landry A, Goyette P, Andersson G, Andersson L, Lindblad-Toh K: Efficient mapping of Zody MC, Hall JL, Brant SR, Cho JH, Duerr RH, Silverberg MS, Taylor KD, mendelian traits in dogs through genome-wide association. Nat Genet Rioux JD, Altshuler D, Daly MJ, Xavier RJ: Deletion polymorphism 2007, 39:1321-1328. upstream of IRGM associated with altered IRGM expression and Crohn’s 21. Parker HG, Shearin AL, Ostrander EA: Man’s best friend becomes biology’s disease. Nat Genet 2008, 40:1107-1112. best in show: genome analyses in the domestic dog. Annu Rev Genet 5. de Cid R, Riveira-Munoz E, Zeeuwen PL, Robarge J, Liao W, Dannhauser EN, 2010, 44:309-336. Giardina E, Stuart PE, Nair R, Helms C, Escaramis G, Ballana E, Martin- 22. Sutter NB, Bustamante CD, Chase K, Gray MM, Zhao K, Zhu L, Ezquerra G, den Heijer M, Kamsteeg M, Joosten I, Eichler EE, Lazaro C, Padhukasahasram B, Karlins E, Davis S, Jones PG, Quignon P, Johnson GS, Pujol RM, Armengol L, Abecasis G, Elder JT, Novelli G, Armour JA, Kwok PY, Parker HG, Fretwell N, Mosher DS, Lawler DF, Satyaraj E, Nordborg M, Bowcock A, Schalkwijk J, Estivill X: Deletion of the late cornified envelope Lark KG, Wayne RK, Ostrander EA: A single IGF1 allele is a major LCE3B and LCE3C genes as a susceptibility factor for psoriasis. Nat Genet determinant of small size in dogs. Science 2007, 316:112-115. 2009, 41:211-215. Berglund et al. Genome Biology 2012, 13:R73 Page 17 of 17 http://genomebiology.com/2012/13/8/R73 23. Salmon Hillbertz NH, Isaksson M, Karlsson EK, Hellmen E, Pielberg GR, 38. Cruz F, Vila C, Webster MT: The legacy of domestication: accumulation of Savolainen P, Wade CM, von Euler H, Gustafson U, Hedhammar A, deleterious mutations in the dog genome. Mol Biol Evol 2008, Nilsson M, Lindblad-Toh K, Andersson L, Andersson G: Duplication of FGF3, 25:2331-2336. FGF4, FGF19 and ORAOV1 causes hair ridge and predisposition to 39. Kim PM, Lam HY, Urban AE, Korbel JO, Affourtit J, Grubert F, Chen X, dermoid sinus in Ridgeback dogs. Nat Genet 2007, 39:1318-1320. Weissman S, Snyder M, Gerstein MB: Analysis of copy number variants 24. Olsson M, Meadows JR, Truve K, Rosengren Pielberg G, Puppo F, Mauceli E, and segmental duplications in the human genome: Evidence for a Quilez J, Tonomura N, Zanna G, Docampo MJ, Bassols A, Avery AC, change in the process of formation in recent evolutionary history. Karlsson EK, Thomas A, Kastner DL, Bongcam-Rudloff E, Webster MT, Genome Res 2008, 18:1865-1874. Sanchez A, Hedhammar A, Remmers EF, Andersson L, Ferrer L, Tintle L, 40. Korbel JO, Urban AE, Affourtit JP, Godwin B, Grubert F, Simons JF, Kim PM, Lindblad-Toh K: A novel unstable duplication upstream of HAS2 Palejev D, Carriero NJ, Du L, Taillon BE, Chen ZT, Tanzer A, Saunders ACE, predisposes to a breed-defining skin phenotype and a periodic fever Chi JX, Yang FT, Carter NP, Hurles ME, Weissman SM, Harkins TT, syndrome in Chinese Shar-Pei dogs. PLoS Genet 2011, 7:e1001332. Gerstein MB, Egholm M, Snyder M: Paired-end mapping reveals extensive 25. Parker HG, VonHoldt BM, Quignon P, Margulies EH, Shao S, Mosher DS, structural variation in the human genome. Science 2007, 318:420-426. Spady TC, Elkahloun A, Cargill M, Jones PG, Maslen CL, Acland GM, 41. Duret L, Galtier N: The covariation between TpA deficiency, CpG Sutter NB, Kuroki K, Bustamante CD, Wayne RK, Ostrander EA: An expressed deficiency, and G + C content of human isochores is due to a fgf4 retrogene is associated with breed-defining chondrodysplasia in mathematical artifact. Mol Biol Evol 2000, 17:1620-1625. domestic dogs. Science 2009, 325:995-998. 42. vonHoldt BM, Pollinger JP, Lohmueller KE, Han EJ, Parker HG, Quignon P, 26. Oliver PL, Goodstadt L, Bayes JJ, Birtle Z, Roach KC, Phadnis N, Beatson SA, Degenhardt JD, Boyko AR, Earl DA, Auton A, Reynolds A, Bryc K, Brisbin A, Lunter G, Malik HS, Ponting CP: Accelerated evolution of the Prdm9 Knowles JC, Mosher DS, Spady TC, Elkahloun A, Geffen E, Pilot M, speciation gene across diverse metazoan taxa. PLoS Genet 2009, 5: Jedrzejewski W, Greco C, Randi E, Bannasch D, Wilton A, Shearman J, e1000753. Musiani M, Cargill M, Jones PG, Qian ZW, Huang W, Ding ZL, et al: 27. Axelsson E, Webster MT, Ratnakumar A, Consortium L, Ponting CP, Genome-wide SNP and haplotype analyses reveal a rich history Lindblad-Toh K: Death of PRDM9 coincides with stabilization of the underlying dog domestication. Nature 2010, 464:898-902. recombination landscape in the dog genome. Genome Res 2012, 22:51-63. 43. LUPA.. [http://www.eurolupa.org/]. 28. Webber C, Ponting CP: Hotspots of mutation and breakage in dog and 44. Selzer RR, Richmond TA, Pofahl NJ, Green RD, Eis PS, Nair P, Brothman AR, human chromosomes. Genome Res 2005, 15:1787-1797. Stallings RL: Analysis of chromosome breakpoints in neuroblastoma at 29. Chen WK, Swartz JD, Rush LJ, Alvarez CE: Mapping DNA structural sub-kilobase resolution using fine-tiling oligonucleotide array CGH. Gene variation in dogs. Genome Res 2009, 19:500-509. Chromosome Canc 2005, 44:305-319. 30. Nicholas TJ, Cheng Z, Ventura M, Mealey K, Eichler EE, Akey JM: The 45. Peiffer DA, Le JM, Steemers FJ, Chang W, Jenniges T, Garcia F, Haden K, Li J, genomic architecture of segmental duplications and associated copy Shaw CA, Belmont J, Cheung SW, Shen RM, Barker DL, Gunderson KL: High- number variants in dogs. Genome Res 2009, 19:491-499. resolution genomic profiling of chromosomal aberrations using Infinium 31. Nicholas TJ, Baker C, Eichler EE, Akey JM: A high-resolution integrated map whole-genome genotyping. Genome Res 2006, 16:1136-1148. of copy number polymorphisms within and between breeds of the 46. Colella S, Yau C, Taylor JM, Mirza G, Butler H, Clouston P, Bassett AS, modern domesticated dog. BMC Genomics 2011, 12:414. Seller A, Holmes CC, Ragoussis J: QuantiSNP: an Objective Bayes Hidden- 32. Redon R, Ishikawa S, Fitch KR, Feuk L, Perry GH, Andrews TD, Fiegler H, Markov Model to detect and accurately map copy number variation Shapero MH, Carson AR, Chen W, Cho EK, Dallaire S, Freeman JL, using SNP genotyping data. Nucleic Acids Res 2007, 35:2013-2025. Gonzalez JR, Gratacos M, Huang J, Kalaitzopoulos D, Komura D, 47. g:Profiler.. [http://biit.cs.ut.ee/gprofiler/index.cgi]. MacDonald JR, Marshall CR, Mei R, Montgomery L, Nishimura K, Okamura K, 48. UCSC Genome Bioinformatics.. [http://genome.ucsc.edu/]. Shen F, Somerville MJ, Tchinda J, Valsesia A, Woodwark C, Yang F, Zhang J, 49. ENSMBL.. [http://www.ensembl.org/]. et al: Global variation in copy number in the human genome. Nature 50. OMIM.. [http://www.omim.org/]. 2006, 444:444-454. 51. DoG_CNV.. [http://dogs.genouest.org/LUPA.dir/CNV.html]. 33. Vaysse A, Ratnakumar A, Derrien T, Axelsson E, Rosengren Pielberg G, 52. Gene Expression Omnibus.. [http://www.ncbi.nlm.nih.gov/geo/]. Sigurdsson S, Fall T, Seppala EH, Hansen MS, Lawley CT, Karlsson EK, 53. Lindblad-Toh K, Wade CM, Mikkelsen TS, Karlsson EK, Jaffe DB, Kamal M, Bannasch D, Vila C, Lohi H, Galibert F, Fredholm M, Haggstrom J, Clamp M, Chang JL, Kulbokas EJ, Zody MC, Mauceli E, Xie X, Breen M, Hedhammar A, Andre C, Lindblad-Toh K, Hitte C, Webster MT: Wayne RK, Ostrander EA, Ponting CP, Galibert F, Smith DR, DeJong PJ, Identification of genomic regions associated with phenotypic variation Kirkness E, Alvarez P, Biagi T, Brockman W, Butler J, Chin CW, Cook A, Cuff J, between dog breeds using selection mapping. PLoS Genet 2011, 7: Daly MJ, DeCaprio D, Gnerre S, Grabherr M, et al: Genome sequence, e1002316. comparative analysis and haplotype structure of the domestic dog. 34. Bailey JA, Yavor AM, Massa HF, Trask BJ, Eichler EE: Segmental duplications: Nature 2005, 438:803-819. organization and impact within the current human genome project doi:10.1186/gb-2012-13-8-r73 assembly. Genome Res 2001, 11:1005-1017. Cite this article as: Berglund et al.: Novel origins of copy number 35. Zody MC, Garber M, Sharpe T, Young SK, Rowen L, O’Neill K, Whittaker CA, variation in the dog genome. Genome Biology 2012 13:R73. Kamal M, Chang JL, Cuomo CA, Dewar K, FitzGerald MG, Kodira CD, Madan A, Qin S, Yang X, Abbasi N, Abouelleil A, Arachchi HM, Baradarani L, Birditt B, Bloom S, Bloom T, Borowsky ML, Burke J, Butler J, Cook A, DeArellano K, DeCaprio D, Dorris L, Dors M, et al: Analysis of the DNA sequence and duplication history of human chromosome 15. Nature 2006, 440:671-675. 36. Zody MC, Garber M, Adams DJ, Sharpe T, Harrow J, Lupski JR, Nicholson C, Submit your next manuscript to BioMed Central Searle SM, Wilming L, Young SK, Abouelleil A, Allen NR, Bi W, Bloom T, and take full advantage of: Borowsky ML, Bugalter BE, Butler J, Chang JL, Chen CK, Cook A, Corum B, Cuomo CA, de Jong PJ, DeCaprio D, Dewar K, FitzGerald M, Gilbert J, • Convenient online submission Gibson R, Gnerre S, Goldstein S, Grafham DV, et al: DNA sequence of human chromosome 17 and analysis of rearrangement in the human • Thorough peer review lineage. Nature 2006, 440:1045-1049. • No space constraints or color figure charges 37. Zody MC, Jiang Z, Fung HC, Antonacci F, Hillier LW, Cardone MF, Graves TA, • Immediate publication on acceptance Kidd JM, Cheng Z, Abouelleil A, Chen L, Wallis J, Glasscock J, Wilson RK, Reily AD, Duckworth J, Ventura M, Hardy J, Warren WC, Eichler EE: • Inclusion in PubMed, CAS, Scopus and Google Scholar Evolutionary toggling of the MAPT 17q21.31 inversion region. Nat Genet • Research which is freely available for redistribution 2008, 40:1076-1083. Submit your manuscript at www.biomedcentral.com/submit http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Genome Biology Springer Journals

Novel origins of copy number variation in the dog genome

Download PDF
18 pages

Loading next page...
 
/lp/springer-journals/novel-origins-of-copy-number-variation-in-the-dog-genome-1l1QMVLgga

References (105)

Publisher
Springer Journals
Copyright
2012 Berglund et al.; licensee BioMed Central Ltd.
eISSN
1474-760X
DOI
10.1186/gb-2012-13-8-r73
Publisher site
See Article on Publisher Site

Abstract

Background: Copy number variants (CNVs) account for substantial variation between genomes and are a major source of normal and pathogenic phenotypic differences. The dog is an ideal model to investigate mutational mechanisms that generate CNVs as its genome lacks a functional ortholog of the PRDM9 gene implicated in recombination and CNV formation in humans. Here we comprehensively assay CNVs using high-density array comparative genomic hybridization in 50 dogs from 17 dog breeds and 3 gray wolves. Results: We use a stringent new method to identify a total of 430 high-confidence CNV loci, which range in size from 9 kb to 1.6 Mb and span 26.4 Mb, or 1.08%, of the assayed dog genome, overlapping 413 annotated genes. Of CNVs observed in each breed, 98% are also observed in multiple breeds. CNVs predicted to disrupt gene function are significantly less common than expected by chance. We identify a significant overrepresentation of peaks of GC content, previously shown to be enriched in dog recombination hotspots, in the vicinity of CNV breakpoints. Conclusions: A number of the CNVs identified by this study are candidates for generating breed-specific phenotypes. Purifying selection seems to be a major factor shaping structural variation in the dog genome, suggesting that many CNVs are deleterious. Localized peaks of GC content appear to be novel sites of CNV formation in the dog genome by non-allelic homologous recombination, potentially activated by the loss of PRDM9. These sequence features may have driven genome instability and chromosomal rearrangements throughout canid evolution. Background [6], glomerulonephritis [7] and systemic lupus erythema- The fraction of genomic variation attributable to copy tosus [8]. However, there are also a small number of number variants (CNVs) is larger than single nucleotide examples of CNVs that may be beneficial, such as adap- polymorphisms (SNPs) and yet the full extent of such tive variation in copy number of the amylase gene in structural variation is still relatively unexplored [1,2]. response to diet [9], and variation in HIV/AIDS suscept- CNVs involve duplications, deletions or insertions of ibility [10]. DNA segments up to several megabases in length and are A variety of mechanisms are thought to give rise to responsible for significant phenotypic variation [3]. In CNVs [11]. A major source of structural variation is non- humans, the frequency distribution of CNVs shows sig- allelic homologous recombination (NAHR), which occurs nals of purifying selection, suggesting that a significant due to aberrant pairing of regions of extended homology. proportion of CNVs have harmful phenotypic effects [1]. Other mechanisms involve re-joining of breaks in DNA CNVs are associated with a number of genetic disorders, but do not require extensive homology. In addition to including Crohn’s disease [4], psoriasis [5], osteoporosis this, errors in replication, such as slippage at variable number of tandem repeat (VNTR) loci or insertion of * Correspondence: [email protected] transposable elements, also generate variation in copy Science for Life Laboratory, Department of Medical Biochemistry and number. CNV formation appears to occur at higher rates Microbiology, Uppsala University, Box 582, SE-751 23, Uppsala, Sweden in certain genomic regions termed rearrangement Full list of author information is available at the end of the article © 2012 Berglund et al.; licensee BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Berglund et al. Genome Biology 2012, 13:R73 Page 2 of 17 http://genomebiology.com/2012/13/8/R73 hotspots. In particular, CNVs associated with NAHR Chen et al. [29] used a 385,000 oligo array on nine dogs tend to be clustered in the genome, and CNVs are from different breed groups. They discovered 155 high enriched in the vicinity of segmental duplications. This confidence CNVs in 60 CNV regions. Nicholas et al. suggests regions of local sequence homology are hotspots [30] focused on areas of segmental duplications (SDs) of CNV formation by NAHR [12-14]. In humans, the using single dogs from 17 breeds and a gray wolf and initiation of meiotic double-stranded breaks (DSBs) is identified approximately 3,600 CNVs in approximately 700 overlapping regions found in two or more samples. thought to begin with the binding of the protein PRDM9 A subsequent study [31] used aCGH with 2.1 million to a degenerate 13-bp sequence motif [15-17]. This motif probes with an average density of 1 kb in nine dogs and is also enriched in CNV breakpoints [2], including several involved in disease [18], which implicates DSBs formed in one wolf sample and identified 403 CNVs. As expected, this way in CNV formation by NAHR. CNVs were found to be enriched in SDs. It was also Domestic dogs harbor an astonishing level of phenoty- shown that CNVs not associated with SDs were more pic variation, which is mostly apportioned into distinct likely to be present only once or at lower frequencies in breeds. The hundreds of dog breeds recognized today the dataset. An additional population genetic analysis on were formed by population bottlenecks accompanied by a set of these CNVs revealed some with divergent pat- strong artificial selection, which has led to both their terns of fixation in different breeds, which could be unique collections of characteristics and an increased responsible for breed-specific traits. prevalence of genetic disease. This makes the dog an Despite extensive efforts to type CNVs in dogs, several ideal genetic model for uncovering the genetic basis of questions remain about their mechanisms and effects. normal and pathogenic phenotypic variation [19]. Many There is evidence that meiotic DSBs localize to different traits have now been mapped in the dog genome using a sites in dogs than other mammals [27]; does this cause variety of approaches [19-22], and structural variation is a different distribution of CNVs in the genome? Is there implicated in a number of these. For example, a duplica- evidence of fixed CNVs in certain breeds that may lead tion of three fibroblast growth factor (FGF) genes causes to breed-specific phenotypes? Can variation at CNV loci the dorsal hair ridge in Rhodesian and Thai Ridgeback be used to delineate different breeds? Besides these dogs and predisposes to dermoid sinus [23], a duplication questions, a comprehensive catalogue of dog CNVs upstream of Hyaluronic acid synthase 2 (HAS2) is would be useful to aid gene-mapping studies. Here we responsiblefor thecharacteristicwrinkledskinof present the most comprehensive CNV discovery effort in dogs to date. We use a 2.1 million-probe array, as in Chinese Shar-Pei dogs and predisposes to periodic fever Nicholas et al. [31], with probes spaced on average of 1 syndrome[24],and an insertionof anFGF4retrogeneis responsible for chondrodysplasia typical of certain breeds kb. We investigate the genome-wide extent and charac- [25]. As in humans, much phenotypic variation is likely teristicsofCNVsin50 dogs from17breeds and 3 to be attributable to CNVs, which makes investigating wolves. This enables us to examine sequence features in them important for uncovering the genetic basis of phe- breakpoints at high resolution and determine patterns of notypic variation in dogs. fixation. There is a possibility that the genomic features that promote CNV formation in dogs differ from other mam- Results and discussion mals. The dog genome differs from the majority of other CNV discovery, genotyping and validation mammals in that it lacks an active copy of PRDM9, We performed aCGH analysis using a 2.1M probe tech- which suggests that formation of meiotic DSBs is con- nology platform spanning the assayable portion of the dog trolled differently in dogs [26]. A fine scale analysis of reference genome with a median spacing of 1 kb. We recombination rate variation in the dog genome indicated restricted our analysis to identifying high-confidence that, like in humans, recombination is clustered into hot- CNVs containing at least 10 probes, which allows identifi- spots but that, unlike in humans, these regions were cation of CNVs down to approximately 9 kb in length. We strongly enriched for short regions (approximately 1 kb) assayed CNVs in 53 samples, comprising purebred dogs of highly elevated GC content (GC peaks) [27]. This sug- from 17 breeds plus 3 wolf samples. Two breeds were gests GC peaks may be targets of meiotic DSBs. Interest- represented by ten unrelated individuals each to enable ingly, GC rich regions also seem to be involved in CNVs segregating at lower frequencies in these breeds to genome rearrangements during canid genome evolution, be identified, whereas the other breeds were represented where they have relocated to telomeric regions [28]. This by two individuals to maximize coverage of different could indicate that GC peaks are important targets of breeds. A male Boxer was used as the reference sample. NAHR and often involved in rearrangements. CNVs were analyzed in autosomes and × chromosome by Three studies have identified canine CNVs using array comparing the ratio of signal intensities between test sam- comparative genomic hybridization (aCGH) [29-31]. ples and the reference. Berglund et al. Genome Biology 2012, 13:R73 Page 3 of 17 http://genomebiology.com/2012/13/8/R73 We identified CNVs using a three-stage procedure In total, this procedure detected 430 CNV loci distribu- comprising segmentation, identification of CNV loci, and ted along the chromosomes (Figure 1). Out of these loci, genotype calling. We first performed a comprehensive 226 were classified as deletions (53%) and 104 as duplica- comparison of five segmentation algorithms, including tions (24%), meaning that the only variants identified at NimbleGen, DNAcopy, Ultrasome, pennCNV and these loci were deletions or duplications, respectively, cghFLasso, and selected the algorithm most robust to relative to the reference, whereas 100 CNV loci exhibited both deletions and duplications (23%) among samples. In noise (see Materials and methods and Supplementary addition, 77 of the loci exhibited a complex deviation methods in Additional file 1). This comparison identified from the reference in at least one sample (Table 1). cghFLasso as the most accurate method, while other methods resulted in excessive segmentation of the signal Across all calls at all loci, 70.2% matched the reference, intensity ratio in samples with noisy data, large discre- whereas 28.4% exhibited a deviation consistent with a pancies in CNV numbers between samples and lack of a single deletion or duplication (17.7% single deletions and strong correlation between levels of variation in CNVs 10.7% single duplications). Only 1.4% of calls were of compared with SNPs (Supplementary methods, Tables greater magnitude (Figure S6 in Additional file 1). The S1 andS2and FiguresS1toS5inAdditionalfile 1).We finding that deletions are more numerous than duplica- therefore used this algorithm to perform segmentation of tions is generally observed in studies using aCGH signal intensity ratios compared to a reference for sam- [1,12,29-32]. This may reflect the greater relative diffi- ples in our dataset. culty of identifying duplications due to the smaller rela- Identification of CNV loci was performed with a tive change in copy number (3:2 versus 2:1) and also the method to estimate the absolute copy number compared fact that insertions of sequence not present in the refer- to the reference based on fixed thresholds and chromo- ence will not be detected. Also concordant with recent some-specific variance in each sample. The principle of studies, duplications were found to be larger than dele- our method is that fixed thresholds are used as a baseline tions with a median size of 30 kb versus 19 kb (Table 1). cutoff, but samples with very high variance use a higher This could suggest that duplications are less likely to be threshold whereas samples with low variance use lower severely deleterious than deletions and therefore less thresholds (Materials and methods). We first identified likely to be purged by purifying selection. Alternatively, it segments falling above a stringent threshold, designed to could reflect a bias against detecting small duplications. exclude false positives, to identify CNV loci in single But despite deletions being smaller, their higher inci- dence indicates that deletions and duplications affect samples. The final set of CNV loci is the union of indivi- similar proportions of the genome. dual calls at each locus, which were merged across sam- ples into asingleCNV.The breakpointsweredefined by On average, 130.9 loci (30%) differ from the reference the outermost boundaries of all individual CNV calls at per sample, and among breeds with two samples an each locus. After identification of loci, the genotype of average of 172.4 loci (40%) differ from the reference in each individual was inferred at each locus using less one or both of the samples (Table 1). In total 50 (12%) stringent criteria to determine the most likely state of CNVs were detected only once in the dataset. This is an each sample (described in Materials and methods ). This average of one singleton per sample, and is consistent method allowed us to distinguish which samples exhib- across samples from breeds with ten and two samples. ited deletions and duplications relative to the reference. We observe one CNV that differs from the reference in Each change was then categorized as simple or complex all samples, indicating that it is a singleton in the refer- based on variance between samples: changes where the ence (chromosome 15, 57.86 to 57.88 Mb). This corre- mean value exceeds the threshold were considered sim- sponds well with the number of singletons found in ple, whereas changes where the mean value is less than other samples, suggesting these numbers are accurate. the threshold but the mean deviation from zero exceeds However, this contrasts with a recent study by Nicholas the threshold were considered complex. This happens et al.[31] using thesamearray,whichidentified403 especially when the segmentation algorithm fails to dis- CNV loci in nine dogs and a wolf, of which 260 (65%) cern two adjacent CNVs, of which one is a deletion and were detected in a single sample (an average of 26 per the other is a duplication (where mean value is close to sample). This previous study used only one sample per zero, but mean deviation from zero is large). Loci where breed, which makes identifying CNV loci as singletons at least one sample exhibited a complex change from the more likely. However, the fact that we also find that most CNV loci are shared between breeds suggests that reference were defined as complex. It is important to differences in CNV calling contribute to differences in note that it is not possible to discern how the copies are CNVs found in each study. distributed between alleles using aCGH on diploid sam- ples, that is, distinguish if a CNV is heterozygous or We validated our set of CNVs using two complemen- homozygous. tary methods: quantitative PCR (qPCR) and analysis on Berglund et al. Genome Biology 2012, 13:R73 Page 4 of 17 http://genomebiology.com/2012/13/8/R73 Figure 1 The genomic architecture of CNVs. Black lines represent all 38 canine autosomes and the × chromosome. Deletions are plotted as red rectangles below each chromosome and duplications are plotted as blue rectangles above each chromosome. the CanineHD high-density SNP genotyping array. A total of 53 sample-locus combinations were tested Because of the high overlap (50%) with previously iden- and 3 tests did not match the state predicted from tified and confirmed CNVs (see ‘Distribution and geno- aCGH (1 false positive and 2 false negatives; 94% mic effects of CNVs’), we focused our validation on concordance). novel CNVs detected only in our study, and performed All of the samples used for aCGH were also geno- qPCR on four loci ranging in size from 17 to 80 kb. typed on the Illumina CanineHD array, which contains Table 1 Number of CNVs identified Deletions Duplications Both All Total CNV loci (complex) 226 (19) 104 (16) 100 (42) 430 (77) Mean CNV loci per breed 100.5 68.5 3.4 172.4 Mean CNV loci per sample 77 53.8 0 130.9 Median size (kb) 19 29.5 27.8 24.3 Deletions are defined as CNV loci that only exhibit reduction in copy number relative to the reference assembly. Duplications are loci that only exhibit increased copy number. ‘Both’ refers to loci that exhibit both deletions and duplications. Based on 15 breeds with 2 samples. Berglund et al. Genome Biology 2012, 13:R73 Page 5 of 17 http://genomebiology.com/2012/13/8/R73 >174,000 probes designed for assaying SNPs with an the RNA-binding region gene RNPC3 and upstream of the average spacing of 13 kb [33]. The probe density of this collagen genes COL11A1 and COL5A1. more than ten times sparser array permits identification The second largest locus is a complex CNV on chromo- of larger CNVs (greater than approximately 100 kb with some 9 between 20.1 and 21.6 Mb. There are no RefSeq our filters) compared to the aCGH chip. In total, 13 annotated dog genes in this region, although it is ortholo- CNV loci passed our filtering procedure and had gous to the human genes VPS13D, RDM1, ARHGAP27 enough probes on the SNP array to be used in the vali- and members of the LRRC37 family. These genes are not co-located in human, so this region must represent syn- dation procedure (Table S3 in Additional file 1). These teny to multiple loci. RDM1 is near members of the 13 CNVs, with an average size of 266 kb, included 8 sin- gletons, 3 two-sample CNVs and 2 CNVs at midrange TBC1D3 family, which shows primate-specific expansion frequency. In total there were 6 mismatches between via SDs on chromosome 17, and is present in a majority of calling on the aCGH and CanineHD SNP array among human-specific breakpoints of conserved synteny to 689 genotypes (1 false positive and 5 false negatives), mouse [36]. Both LRRC37 and ARHGAP27 are ortholo- which is a correspondence of >99% between individual gous to regions flanking an inversion in the human gen- genotypes. This suggests that the genotyping error is ome at 17q21.31, thought to have arisen through NAHR smaller than the 6% estimated from the qPCR, at least between large blocks of flanking SDs, which are distribu- for large CNVs. ted throughout chromosome 17 and contain the LRRC37 gene family [37]. This region is associated with a micro- Distribution and genomic effects of CNVs deletion leading to mental retardation and has undergone We compared our dataset to CNVs previously identified multiple complex rearrangements during primate evolu- in dog in multiple studies merged and augmented by tion. The finding that this region is also a large and com- Nicholas et al. [31]. We identify 216 overlaps, which con- plex CNV in dogs suggests that it may be a region of tain 196 (of 615) of previously identified CNVs and 213 instability across a wide range of mammalian evolution. (of 430) of ours (Figure 2). This overlap is highly signifi- Both of the two discussed loci were also present in pre- cant compared to random redistributions of CNVs in the vious dog aCGH reports [29-31]. genome (P < 0.001). The reason that some CNVs do not CNVs overlapping SDs occur at higher frequencies in overlap between studies is likely to be a combination of the population (P <0.001), aremorelikely tobecom- differences in breed selection, sample size, array resolu- plex (P < 0.001), tend to be longer (P < 0.001), and are tion, genotyping algorithms and errors. Furthermore, we more likely to overlap genes (P <0.001) than CNVsnot associated with SDs (bootstrapping used for all signifi- generated a canine segmental duplication map using a cance tests), which confirms previous observations [31]. modified version of the method of Bailey et al.[34] (described in [35]), to which both datasets were mapped. Both singletons (80%) and breed-specific CNVs (85%) The Venn diagram in Figure 2 shows that more than a are more likely to fall in the category of CNVs not over- third (165) of our CNVs overlap these SDs and more lapping SDs. Non-SD CNVs have an average frequency than half (319) of Nicholas’ CNVs overlap SDs, which is of 0.24 (0.33 of breeds), whereas CNVs inside of SDs a highly significant overlap (P < 0.001, random redistribu- have an average frequency of 0.4 (0.56 of breeds), which tion test) considering <5% of the dog genome is com- is almost twice as high (P < 0.001). Complex CNVs are posed of recent SDs. The high proportion of CNVs in preferentially observed in SDs, where every third CNV SDs together with our non-targeted approach indicates a is complex, compared to less than 10% complex CNVs large involvement of SDs in CNV formation as indicated outside SDs. These observations may reflect a higher by previous studies. However, it should also be noted mutation rate of CNVs in SDs, with recurrent events that a small proportion of SDs are likely not fixed in the around the same genomic location leading to both dog genome and may actually be CNVs. higher frequencies and more complexity, but could also The largest CNV locus we identified in the dog genome reflect the involvement of SDs in more complex rearran- is located at 48.5 to 50.0 Mb on chromosome 6. The pat- gements. CNVs in SDs are also larger than non-SD tern of variation in this region is consistent with a deletion CNVs; with a median size of 40 versus 20 kb. The in all samples except both Labradors and one Boxer, increased frequency and complexity of CNVs in SDs which match the reference sequence. Inspection of probe may reflect the dynamic nature of SDs, and that these intensities in this region on the CanineHD array shows a CNVs have arisen from overlapping but distinct events. pattern consistent with the presence of duplicated We next examined the functional effects of CNVs by identifying genes they overlap. In order to use a high- sequence in these Labrador and Boxer samples. Consider- confidence gene set, 24,232 dog genes annotated under ing the high frequency of the deletion allele, it is most Ensembl ID were filtered with the g:Orth tool from the likely that the duplication allele is the derived state. The CNV encompasses a gene-sparse region downstream of g:Profiler website to extract only human-dog 1:1 Berglund et al. Genome Biology 2012, 13:R73 Page 6 of 17 http://genomebiology.com/2012/13/8/R73 Figure 2 Comparison of LUPA dataset with a summary of CNVs presented by Nicholas et al. [31]and a list of segmental duplications in dog [53]. The Venn diagram shows the number of overlapping CNVs from the datasets. Since the datasets contain different entries, numbers are colored according to which dataset the counted entries belong to. For example, 196 of Nicholas et al.’s CNVs overlap with 213 of our (LUPA) CNVs, and 120 and 132, respectively, of these also overlap with segmental duplications. orthologs. Out of 15,258 1:1 orthologous genes, 130 In addition to their paucity, CNVs overlapping genes are genes (0.85%) are overlapped either completely or par- characterized by a much higher proportion of duplications tially by 90 CNVs (out of 430 total CNVs). This number than deletions (Table 2). In intergenic CNVs, deletions are is significantly less than predicted by chance (P <0.001, more than three times as common as duplications, random redistributions per chromosome). A great pro- whereas within genes they occur at a proportion of two- -8 portion of the affected genes, 53 (41%), had their entire thirds compared with duplications (P = 3.1 × 10 ; Fisher’s coding sequence covered by a CNV (Table 2). This exact test). This pattern is even more pronounced in intra- could suggest that CNVs overlapping genes are more genic CNVs that are predicted to remove a stop codon, likely to have deleterious effects. A test of the size distri- where deletions occur at a proportion of 0.2 relative to bution of CNVs affecting genes revealed that they are duplications, which is significantly lower than other intra- larger, with median size of 36 versus 24 kb (bootstrap- genic CNVs (P = 0.009; Fisher’s exact test). This tendency ping, P < 0.005). They also show a slight tendency for duplication enrichment among stop codons has pre- towards lower frequencies, although the difference is viously been detected in humans [1], and suggests a strong not significant (bootstrapping, P < 0.1). deleterious effect of removal of stop codons. Berglund et al. Genome Biology 2012, 13:R73 Page 7 of 17 http://genomebiology.com/2012/13/8/R73 Table 2 Number of CNV loci covering genomic regions Intragenic Partial gene Total Intergenic Total Whole gene Total Stop codon Total 430 340 90 31 67 29 Deletion 226 198 28 8 21 4 Duplication 104 62 42 13 33 19 Both 100 80 20 10 12 6 Deletion:duplication ratio 2.17 3.08 0.67 0.62 0.64 0.21 All groups are non-overlapping. Subcategories are not additive as a CNV can span several genes and harbor both deletions and duplications in different samples. The set of 130 1:1 orthologous genes overlapping CNVs genes they overlap: no functional categories are inferred to were scanned for enrichment of gene ontology (GO) cate- be enriched in both deletions and duplications. For exam- gories against a background of all 15,258 1:1 human-dog ple, the ‘homophilic cell adhesion’ category is only enriched in deletions, whereas ‘olfactory receptor activity’ orthologs (Table 3). This gene set was chosen because of the higher accuracy of annotations for human genes. For is only enriched in duplications. The enrichment of speci- this purpose the g:GOSt tool from g:Profiler website was fic GO categories could reflect changes in patterns of used. The most significantly enriched term in each domain selective constraint in dogs [38], positive selection for dog- was the biological process ‘homophilic cell adhesion’ (P = specific traits, or tolerance of certain gene categories to -13 7.31 × 10 ), the cellular component ‘integral to mem- deletions or duplications. -5 brane’ (P = 6.07 × 10 ) and the molecular function ‘olfac- -6 tory receptor activity’ (P = 6.22 × 10 ). Cell adhesion also Mechanisms of CNV formation appears as a strongly enriched category among human We searched for repeats that were enriched close to break- CNVs [1,32]. In our dataset, CNVs are also enriched for points of CNVs, calculating the observed to expected ratio genes involved in the Kyoto Encyclopedia of Genes and to identify over-represented motifs. There is uncertainty in Genomes (KEGG) pathway ‘olfactory transduction’ (P = precisely defining breakpoint location because of smooth- -3 5.04 × 10 ). Analyzing simple deletion and simple dupli- ing of probe intensities and experimental noise. Each cation loci separately reveals a difference in the kind of breakpoint location was defined as a 10 kb window to Table 3 Enriched gene ontology categories in genic CNVs P-value Number of genes All CNVs Deletions Duplications GO ID Domain All Deletions Duplications Gene ontology term 9.48e-02 - 3.12e-03 GO:0032787 BP 5 0 5 1. Monocarboxylic acid metabolic process 5.00e-06 6.32e-04 1 GO:0022610 BP 14 7 1 1. Biological adhesion 5.00e-06 6.32e-04 1 GO:0007155 BP 14 7 1 2. Cell adhesion 2.67e-11 1.36e-06 1 GO:0016337 BP 14 7 1 3. Cell-cell adhesion 7.31e-13 2.37e-07 1 GO:0007156 BP 14 7 1 4. Homophilic cell adhesion 3.56e-03 6.29e-02 1 GO:0071944 CC 18 8 2 1. Cell periphery 2.68e-04 1.76e-01 1 GO:0016020 CC 47 16 20 1. Membrane 2.15e-03 4.88e-02 1 GO:0005886 CC 18 8 2 2. Plasma membrane 3.37e-04 6.08e-01 3.64e-01 GO:0044425 CC 40 13 19 1. Membrane part 8.74e-05 2.49e-01 3.45e-01 GO:0031224 CC 39 13 18 2. Intrinsic to membrane 6.07e-05 2.17e-01 2.92e-01 GO:0016021 CC 39 13 18 3. Integral to membrane 5.97e-02 1 3.82e-02 GO:0004872 MF 16 4 12 1. Receptor activity -1 2.66e-02 GO:0038023 MF 18 3 11 2. Signaling receptor activity 1.96e-02 1 1.27e-02 GO:0004888 MF 16 3 11 3. Transmembrane signaling receptor activity 1.49e-03 1 5.21e-04 GO:0004930 MF 15 3 11 4. G-protein coupled receptor activity 6.22e-06 1 9.88e-06 GO:0004984 MF 12 2 9 4. Olfactory receptor activity 3.18e-04 7.11e-05 1 GO:0005509 MF 17 10 1 1. Calcium ion binding 5.20e-01 - 9.98e-03 GO:0005506 MF 6 0 6 1. Iron ion binding 5.04e-03 1 6.44e-03 KEGG:04740 ke 7 1 5 1. Olfactory transduction Significant P-values are in bold. P-values were determined using a hierarchical multiple testing procedure. Indented numbers before the GO terms indicate relative depth of the term in the hierarchy. BP, biological process; CC, cellular component; MF, molecular function; ke, KEGG pathway. Berglund et al. Genome Biology 2012, 13:R73 Page 8 of 17 http://genomebiology.com/2012/13/8/R73 account for this imprecision. The list of known repeats to analyze enrichment of GC content in breakpoints. We was downloaded from the RepeatMasker track of the found a more than two-fold enrichment in CNV break- UCSC genome browser. Nearly all CNV breakpoints over- points (P < 0.001), which rapidly decays with increasing lap some repeat family, with L1, ERV[1/L], RNA and satel- distance from the breakpoint, both within and outside of lite DNA being overrepresented (Table S4 in Additional the CNV (Figure 3). We also found an enrichment of file 1). The last two types involve fewer than ten break- CpG islands and gaps in CNV breakpoints (P < 0.001; Table S5 in Additional file 1). CpG islands are usually points, and cannot be considered a significant contribution GC-rich, and the great overlap with GC peaks is some- to CNV formation. However, >96% of the CNV break- what expected. Gaps in the reference genome assembly points contain LINE elements, an excess of 36% compared with expected coverage, and almost 65% of the CNV have shown an extremely high likelihood of being asso- breakpoints contain a long terminal repeat (includes the ciated with CNVs in human [32]. In dog, gaps tend to endogenous retrovirus (ERV) family) entry. We found a correlate with high GC content because these regions are 1.5-fold excess of L1 elements in CNV breakpoints. Inter- less well captured during the genome sequencing, and estingly this excess seems to be most prominent for many of them could qualify as GC peaks if they were younger L1 repeats (Table 4). This follows the pattern of fully characterized. The association between CNV break- CNVs and SDs detected in humans [32], where the likeli- points and GC peaks in dogs suggests that CNV break- hood of a SD being associated with a CNV was highly cor- points may often occur at recombination hotspots, where related to its sequence similarity to the duplicated copy, as DSBs have a tendency to form, followed by their repair younger L1 elements are likely to have increased levels of by NAHR. sequence homology with each other. The association with GC peaks and CNVs has not been The pattern of L1 enrichment together with the signifi- reported in other species. In humans, a 13-bp sequence cant overlap between SDs and CNVs is consistent with a motif targeted by the PRDM9 protein is strongly asso- large contribution of NAHR, which is known to operate ciated with recombination hotspots, which also associates on highly similar copies, and supports the generation of with CNVs [18]. In dogs, the PRDM9 gene is inactive and CNVs in duplicated regions and regions with mobile ele- recombination hotspots are strongly associated with GC ment insertions. L1 transduction, where additional 3’ peaks. The association between GC peaks and CNV flanking sequence is transferred to a new genomic loca- breakpoints may, therefore, be an additional consequence tion together with an L1 insertion, may also contribute to of the death of PRDM9 [27]. There is a strong overlap this enrichment. We do not find any enrichment of CNV between CpG islands and GC peaks, and both are enriched close to CNV breakpoints. This could indicate breakpoints around SINEs. This is in concordance with some studies in humans [39,40] that find associations that non-methylated DNA promotes DSB formation that with L1 but not Alu at CNV breakpoints. This may sug- leadstostructuralvariation.However,itisalsopossible gest that LINE elements promote structural variation that it is simply GC-richness that promotes recombina- throughNAHRmorestronglythanSINEs inthedog tion, or that the peaks of high GC content are mainly a genome. consequence of elevated recombination rate due to GC- As GC peaks are enriched in recombination hotspots in biased gene conversion rather than its cause, which are the dog genome and may be important for formation of detected as CpG islands regardless of methylation [41]. DSBs [27], we searched for enrichment of this sequence These results are particularly interesting in light of the feature within CNV breakpoints. We first defined GC suggestion that GC-rich regions have acted as novel tar- peaksasinAxelsson et al. [27], where a GC peak is get sites of chromosomal fissions during canid evolution recorded when a 500-bp sliding window centered in a [28]. 10 kb background sliding window has a 1.5-fold increase We also searched for regions of extended perfect in GC content. We then performed randomization tests homology between pairs of breakpoints flanking CNV Table 4 Excess of L1 repeats in CNV breakpoints a b Name Excess Divergence Length Number of repeats in Number of repeats in Number of breakpoints with (bp) genome breakpoints repeats L1_Cf 3.92 0.035 1,192 16,127 178 137 L1_Canis 2.54 0.088 799 72,587 678 309 L1_Canid 1.87 0.144 534 42,070 254 135 L1_Carn 1.25 0.168 525 132,806 652 275 Eutheria 0.91 0.246 359 589,155 2046 590 a b The proportion of repeat coverage in breakpoints versus whole genome. Average divergence of the repeat from its reference in RepBase. Berglund et al. Genome Biology 2012, 13:R73 Page 9 of 17 http://genomebiology.com/2012/13/8/R73 Figure 3 Enrichment of GC peaks around CNV breakpoints. The strongest excess of GC peaks compared to random expectations is observed close to CNV breakpoints and decays with increasing distance from the breakpoint. The midpoint of the breakpoint is at position 0, the shaded area to the left of this represents points within the CNV, and the area to the right represents flanking sequence. loci. Of the 430 CNV loci, 86 have runs of perfect (133.1 bp compared with 116.6 bp). This suggests that sequence homology greater than 75 bp and 11 have NAHR is not restricted to segmental duplications and stretches longer than 1 kb. The mean length of runs of may be equally or more common outside of them. perfect homology between breakpoints is 126.8 bp. We tested the significance of these stretches of homology by CNV distribution among breeds and samples simulating 1,000 datasets with the same number of loci We next analyzed the distribution of CNVs among breeds and distance between breakpoints, located at randomly (Table 5). The majority (341, 79%) of CNVs were found in chosen positions on the same chromosomes. The mean several breeds, while 89 CNVs (21%) were breed-specific. length of homology in the simulated dataset was 35.0 The breed-specific CNVs are larger (32 kb) than CNVs bp, with none of the simulated means exceeding the found in multiple breeds (22 kb) (P < 0.005, bootstrap- observed one (P < 0.001). These findings provide addi- ping). Some variation in the number of CNVs between tional support for an important role of NAHR in dog samples was seen both within and between breeds. Nota- CNV formation. Surprisingly, the mean length of bly, the total number of CNVs identified in Boxers was homology between breakpoints of CNVs overlapping lower than in any other breed, with an average of 64.5 loci SDs is slightly shorter than those outside of SDs different from the reference per sample, largely due to the Berglund et al. Genome Biology 2012, 13:R73 Page 10 of 17 http://genomebiology.com/2012/13/8/R73 Table 5 CNVs identified in each breed and sample compared to the reference genome Total CNV loci per breed Average CNV loci per sample Breed Samples Total MB BS Total MB BS Border Terrier (BTe) 2 156 154 2 128.5 126.5 2 Boxer (Box) 2 103 102 1 64.5 64 0.5 Cavalier King Charles Spaniel (CCS) 2 189 182 7 148.5 143.5 5 Chihuahua (Chi) 2 181 178 3 142 140.5 1.5 Dachshund (Dac) 2 180 172 8 131 127 4 English Cocker Spaniel (ECS) 2 166 165 1 129 128.5 0.5 English Springer Spaniel (ESS) 2 214 207 7 160 156 4 Finnish Spitz (FSp) 2 186 176 10 148.5 140.5 8 German Shepherd (GSh) 2 163 160 3 126 123.5 2.5 Labrador Retriever (LRe) 2 170 168 2 129 128 1 Nova Scotia Duck Tolling Retriever (NSD) 2 193 190 3 147.5 145.5 2 Poodle (Pdl) 2 182 180 2 130 129 1 Sarloos (Sar) 2 175 168 7 135 130 5 Schnauzer (Sch) 2 169 163 6 127 123.5 3.5 Swedish Elkhound (Elk) 2 159 158 1 116.5 115.5 1 Average (2-sample breeds) 2 172.4 168.2 4.2 130.9 128.1 2.8 Golden Retriever (GRe) 10 264 255 9 121.6 118.8 2.8 Irish Wolfhound (IrW) 10 229 215 14 136.2 127.7 8.5 Average (10-sample breeds) 10 246.5 235 11.5 128.9 123.3 5.7 Wolf (Wlf) 3 163 160 3 96.3 95.3 1 MB, multi-breed (CNVs found in more than one breed); BS, breed-specific (CNVs found in only one breed). reference being a Boxer. Of the remaining breeds, the are present per breed, making it less likely that three or average number of CNVs per sample varied from 116.5 more samples share a CNV. A broadly similar frequency (Swedish Elkhound) to 160 (English Springer Spaniel). On distribution is seen within breeds. Figure 4b shows the average, a sample differs from the reference at 130.9 CNV allele frequency distribution with ten genotyped indivi- loci, of which 2.8 are specific to that breed. Fewer than 6% duals analyzed on a breed basis (identity of the minor allele is defined from the two-sample breeds of the entire of CNVs found in any one breed are specific to that breed. These patterns are broadly consistent in the wolf samples, dataset). which exhibit a slightly lower than average number of The two breeds for which ten individuals were analyzed CNVs per sample. On average, 2 dogs from the same were selected for their large (Golden Retriever) and small breed differ at 83.1 CNV loci whereas 2 dogs from differ- (Irish Wolfhound) population sizes, respectively. We first ent breeds differ at 103 CNV loci. attempted to see how many CNVs were present in all ten A more detailed picture of polymorphism in breeds with dogs of a breed, suggesting fixation (Table 7). For Golden two samples is given in Table 6. We find that an average Retrievers, 20 CNVs were present in all dogs, and for of 172.4 CNV loci (40%) are observed in one or both of Irish Wolfhounds, 38 CNVs appeared fixed, possibly the samples in any one breed and an average of 4.2 (2.4%) reflecting the slightly higher degree of inbreeding in Irish are only found in that breed (private CNVs). Similar num- Wolfhounds. A much smaller number of breed-specific bers of CNVs are found in one sample (polymorphic) or loci were identified, and no cases of breed-specific fixed both samples (fixed) of these breeds. An average of one CNVs were identified in these deeply sampled breeds. private CNV is fixed (found in both samples) in the The relative lack of breed-specific fixed CNVs suggests breeds, although with a small sample size, the majority of that instances of those involved in breed-specific pheno- these are likely to be polymorphic. types must be rare. Overall, the CNV frequency distribu- To assess the frequency distribution of CNVs among tions appear qualitatively similar to those expected for samples, we used their presence in all samples to build a neutral polymorphisms. site frequency spectrum. Figure 4a shows the minor allele We scanned our dataset for CNVs that were fixed for frequency distribution across all samples in two-sample one allele in some two-sample breeds and another allele in all other two-sample breeds (that is, are not poly- breeds. There is a marked drop in frequency above two morphic in any breed). There are 24 such CNVs, of samples, which can be attributed to the higher within compared to between breed fixations, where two samples which all but one is specific to a single breed. This CNV Berglund et al. Genome Biology 2012, 13:R73 Page 11 of 17 http://genomebiology.com/2012/13/8/R73 Table 6 Polymorphic CNVs in breeds with two samples Loci matching Shared loci between breeds Private loci single breeds Breed Samples reference Total Polymorphic Fixed Total Polymorphic Fixed Border Terrier 2 274 154 55 99 2 0 2 Boxer 2 327 102 76 26 1 1 0 Cavalier King Charles Spaniel 2 241 182 77 105 7 4 3 Chihuahua 2 249 178 75 103 3 3 0 Dachshund 2 250 172 90 82 8 8 0 English Cocker Spaniel 2 264 165 73 92 1 1 0 English Springer Spaniel 2 216 207 102 105 7 6 1 Finnish Spitz 2 244 176 71 105 10 4 6 German Shepherd 2 267 160 73 87 3 1 2 Labrador Retriever 2 260 168 80 88 2 2 0 Nova Scotia Duck Tolling Retriever 2 237 190 89 101 3 2 1 Poodle 2 248 180 102 78 2 2 0 Sarloos 2 255 168 76 92 7 4 3 Schnauzer 2 261 163 79 84 6 5 1 Swedish Elkhound 2 271 158 85 73 1 0 1 Average 2 257.6 168.2 80.2 88 4.2 2.9 1.3 Wolf 3 267 160 121 39 3 3 0 Total, number of CNVs observed in the breed; Polymorphic, average number of polymorphic CNV sites between samples of the breed; Fixed, number of CNVs found in both samples of the breed. is a 12.7 kb deletion (located at 55.5 Mb on chromo- clusters on a branch leading to Sarloos (a wolf hybrid) some 6) shared by Cavalier King Charles Spaniel and and German Shepherd, as observed in a previous SNP English Springer Spaniel, which overlaps the gene analysis [33]. There is also some evidence for clustering DPYD, which encodes the enzyme dihydropyrimidine by breed type as demonstrated by vonHoldt et al. [42]. dehydrogenase (DPD) involved in pyrimidine catabolism. One cluster contains spaniels (English Cocker Spaniel, Examples of breed-specific fixations include a 32.7 kb English Springer Spaniel, Cavalier King Charles Spaniel), deletion on chromosome 28 in German Shepherds scent hounds (Dachshund) and toy dogs (Chihuahua), downstream of DUSP10, which is involved in immune and another contains retrievers and terriers (Golden responses and mediates various physiological processes, Retriever, Labrador Retriever, Border Terrier). However, and a 18.1 kb deletion on chromosome 21 in Finnish there are exceptions present, and in general, this analysis Spitz immediately downstream of CYP2R1 (cytochrome is limited by a smaller number of loci and less precise p450, family 2, subfamily R, polypeptide 1) and overlap- calling than the studies based on SNP arrays. ping PDE3B (phosphodiesterase 3B, cGMP-inhibited). These regions are good candidates for governing breed- Conclusions specific characteristics, although further investigation is This study provides insights into the mutational mechan- necessary to determine if they are really fixed, or have isms and functional effects of CNVs in the dog genome. any phenotypic effect. Our results suggest that many CNVs are generated by NAHR events directed towards peaks of GC content, Breed relationships which is consistent with observations that these sequence We explored the extent to which patterns of CNV varia- features are also enriched in dog, but not human, recombi- tion can be used to infer population structure between nation hotspots. Hence, GC peaks may represent novel breeds. Based on the proportion of shared CNVs between sites of elevated recombination and genome instability in each pair of samples, a neighbor-joining phylogeny was dogs. This shift in recombinational activity towards GC constructed (Figure 5). With a few exceptions, all samples peaks in dogs is likely to be due to the lack of a functional clustered together with samples from the same breed. copy of the PRDM9 gene, which initiates recombination at This indicates that the CNVs have a strong phylogenetic separate specific sequence motifs in humans. In support of signal, grouping dogs into breeds as shown by large-scale a strong role of NAHR in dog CNV formation, we also SNP analyses [33,42]. Ability to construct this tree identify associations between CNV breakpoints and L1 demonstrates the high accuracy of dataset and high con- elements and long stretches of sequence homology. We gruency with SNP data. Notably, one of the wolf samples also show that dog CNVs are affected by the signal of Berglund et al. Genome Biology 2012, 13:R73 Page 12 of 17 http://genomebiology.com/2012/13/8/R73 (a) (b) Figure 4 Minor allele frequency distribution of CNVs. (a) Minor allele frequency distribution of CNVs compared across all 15 breeds with a sample size of two. (b) Minor allele frequency distribution of CNVs within the two breeds with a sample size of ten. purifying selection and identify candidate CNVs for invol- Materials and methods vement in breed-specific characteristics. This comprehen- Sample collection sive catalogue of CNVs will be useful for future studies to EDTA-blood was collected as part of the LUPA project uncover the genetic basis of complex traits in dogs. [43] from pedigree dogs around Europe and USA with Berglund et al. Genome Biology 2012, 13:R73 Page 13 of 17 http://genomebiology.com/2012/13/8/R73 Table 7 Fixed CNVs in breeds with ten samples Loci matching Shared loci between breeds Private loci in single breed Breed reference Total Polymorphic Fixed Total Polymorphic Fixed Golden Retriever 166 255 235 20 9 9 0 Irish Wolfhound 203 215 177 38 14 14 0 Average 184.5 235 206 29 11.5 11.5 0 Total, number of CNVs observed in the breed; Polymorphic, average number of polymorphic CNV sites between samples of the breed; Fixed, number of CNVs found in both samples of the breed. owners’ consents: a total of 50 dogs from 17 breeds Cocker Spaniel, Finnish Spitz, German Shepherd, Labra- including 2 unrelated individuals from the breeds Border dor Retriever, Nova Scotia Duck Tolling Retriever, Poo- Terrier, Boxer, Cavalier King Charles Spaniel, Chihua- dle, Sarloos Wolfhound, Schnauzer and Swedish hua, Dachshund, English Springer Spaniel, English Elkhound; and 10 unrelated individuals from Golden Figure 5 Neighbor-joining tree based on allele sharing at CNV loci. Box, Boxer; BTe, Border Terrier; CCS, Cavalier King Charles Spaniel; Chi, Chihuahua; Dac, Dachshund; ECS, English Cocker Spaniel; Elk, Swedish Elkhound; ESS, English Springer Spaniel; Fsp, Finnish Spitz; GRe, Golden Retriever; GSh, German Shepherd; IrW, Irish Wolfhound; LRe, Labrador Retriever; NSD, Nova Scotia Duck Tolling Retriever; Pdl, Poodle; Sar, Sarloos; Sch, Schnauzer; Wlf, Wolf. Berglund et al. Genome Biology 2012, 13:R73 Page 14 of 17 http://genomebiology.com/2012/13/8/R73 Retriever and Irish Wolfhound breeds. In addition, three The fixed genotyping ratio was set to correspond to a gray wolves from Belarus, Spain and Finland, respec- deviation of 0.5 copies from the reference, while the tively, were included. A Finnish male Boxer was used as fixed discovery ratio was set to the slightly higher 0.65 the reference sample. Genomic DNA was purified using copies deviation. The standard deviation genotyping commercial purification kits and the quality of the DNA value was chosen to include 5% of the data points, while was analyzed by spectrophotometry and agarose gel the standard deviation discovery value was chosen to electrophoresis prior to the hybridization experiment. include 0.1% of the data points, from the theoretical normal distribution (any copy number variants should be clearly distinct from the distribution). These were Discovery and genotyping aCGH was used to detect DNA copy number alterations chosen so that a similar number of samples used the using NimbleGen’s canFam2 Whole Genome CGH oligo fixed values and the standard deviation values. The dis- array platform with 2.1 million probes on a single slide covery threshold was picked as the maximum of these and a median probe spacing of 1 kb. The array design is two numbers, while the genotyping threshold was based on the annotated CanFam2.0 genome sequence of picked as the minimum of the two numbers. This a female Boxer genome. The isothermal 50-75mer means that no CNVs are identified in the first stage if probes were evenly distributed throughout the unique the deviation from the reference is below 0.65 copies, sequence of the genome. The genomic DNA samples andinthefinalstage allCNVswithadeviationabove were sent to NimbleGen’s service facility where the 0.5 copies from the reference are identified as CNVs. hybridizations were performed in a two-color format according to Selzer et al. [44]. Copy number was quanti- Validation fied from the fluorescence ratios of the two dyes. Nim- aCGH results were experimentally validated by qPCR in bleGen conducted the initial data processing from randomly selected CNV loci. Relative copy numbers of normalization to signal calling and quantification. Ratios the selected regions were determined in comparison to were log2 transformed, and positive log2 ratios indicate the reference sample (Finnish male Boxer). Regions gains and negative log2 ratios indicate loss of copy num- were selected based on the aCGH profiles across breeds. ber. Copy number for each called CNV was calculated qPCR experiments were performed on ABI Prism 7500 (1+mean log2ratio) as 2 and rounded to whole numbers. Fast instrument (Applied Biosystems, Stockholm, Swe- CNV calling was conducted in three stages: smooth- den) using SYBR Green detection chemistry according ing, segmentation and selection. Prior to segmentation to the manufacturer’s instructions. Primers (available and selection, triangular smoothing was done on the upon request) were designed inside CNVs using Primer3 ratios, which implements an 11-point weighted smooth- and NCBI primer design programs. Each assay was per- ing function along the chromosomes. This is equivalent formed in triplicate using 20 μl reactions containing 10 to several passes of fewer-point rectangular smoothings μl of qPCR master mix (Roche, Stockholm, Sweden), 10 (unweighted sliding-average), and is more effective at nM concentration of dNTPs, 250 nM concentration of reducing high-frequency noise in the signal. Segmenta- forward and reverse primer and 10 ng of genomic DNA. tion was done with the R package cghFLasso to segment Amplification was performed under the following condi- a continuous distribution of intensity ratios into discrete tions: one cycle at 50°C for 2 minutes, one cycle at 95°C regions of consecutively different ratios. From these for 10 minutes, 40 cycles at 95°C for 15 seconds and putative CNV segments, those with significant deviation 62°C for 45 seconds. Beta-actin and GAP175 were used are selected as the final set of CNVs. The selection was as controls. Serial dilutions were performed for each performed in three stages: first, ascertain CNV locations assay to estimate the PCR efficiency (E) prior to analysis. per sample using a stringent threshold (discovery The ddC method was used for the quantification of threshold); second, consolidate the calls from all indivi- copy numbers in test individuals relative to the same duals by mapping them onto one ‘master genome’ to get reference Boxer sample used in the aCGH experiments. merged general CNV locations; and third, decide the The C values for each set of triplicates were averaged CT state of the general CNVs in each breed using a less and adjusted for PCR efficiency (E) as log2(E ). The stringent threshold (genotyping threshold) to assign C values were then normalized against the control pri- individuals to diploid copy number classes. Noisy data mers. The relative copy number for each site was calcu- -(t-r) were handled by recalculating the ratios into values of lated as 2 , where t = normalized C for the test standard deviation from a theoretical normal distribu- sample and r = normalized C for the reference sample. tion of ratios. These were used to allow samples with lit- Large CNVs, by virtue of their potential functional tle noise to utilize lower thresholds. The two thresholds impact, were validated in an additional validation step. were then individually chosen on chromosome basis The same set of individuals was genotyped according to from a fixed log2 ratio and a standard deviation value. manufacturer’s instructions on the CanineHD 170K SNP Berglund et al. Genome Biology 2012, 13:R73 Page 15 of 17 http://genomebiology.com/2012/13/8/R73 array (Illumina) with a resolution of 1 SNP per 13 kb. fold increase in GC content compared to background, a The GenomeStudio V2010.3 software package (Illumina) GC peak was marked for all base pairs in the window. was used to obtain normalized total signal intensity, Log The program Neighbour from the Phylip package was R ratio, and B allele frequencies for all SNPs according to used to construct the breed phylogeny from a distance the manual and Peiffer et al. [45]. Exported Log R ratio matrix containing the pairwise differences between the and B allele frequencies for every SNP were used in the breeds for all CNV loci. subsequent CNV calling. The CNV calling was GO analysis was made with the web-based tool g: GOSt Gene Group Functional Profiling provided by g: performed using QuantiSNP [46]. Default settings in Profiler (formerly known as GOSt, Gene Ontology Sta- QuantiSNP were used, that is, L = 2M, expectation-maxi- mization iterations =10, and the parameter file levels-hd. tistics), [47], with default parameters. This tool not only dat. Samples having a standard deviation of the Log R searches for enrichment in GO categories, it also looks ratio above 0.35 were removed from the subsequent ana- for enrichment in KEGG/REACTOME pathways and lysis (three samples). Furthermore, CNVs having less TRANSFAC regulatory motifs/MicroCosm microRNA than five SNPs or a Log Bayes factor <10 were removed. sitesaswellasHuman PhenotypeOntologyand Bio- The Log Bayes factor is a score that represents the sup- GRID protein-protein interaction networks. The set of port for the existence of the CNV and a Log Bayes factor orthologous genes was extracted with the tool g:Orth of at least 10 will result in up to 10% false positive calls. from the same project, which maps orthologous genes in related organisms using Ensembl alignments. Statistical and population genetics analysis Statistical significance of overlap between genomic Prior to analysis, all chromosomes were centered on their features was assessed by estimating the distribution of mean log2 ratio to remove potential chip biases. The × sample means. This was done by either random redistri- chromosome was treated differently in this manner, since butions per chromosome or bootstrapping. Chromo- the pseudo-autosomal regions were centered separate to some-wise redistributions were done by repeatedly and the rest of the chromosome, which differs in relative randomly redistributing all features on the same chro- copy number to a male reference. To identify the mosome to get a distribution of means from which sig- pseudo-autosomal regions, the raw log2 ratios from all nificance can be assessed. Bootstrapping was done by female samples were used to determine the average posi- re-sampling the individual observations with replace- tion of the probe where copy number changed from ment from the population of CNVs to get an empirical diploidy. Since no CNVs were successively called in that bootstrap distribution from which a bootstrap confi- dence interval could be derived to infer statistical particular region, the position seems to be accurate. significance. The metric used to call CNVs from the aCGH data was the log2 ratio, which is the log (base 2) ratio of the R was used to smooth the aCGH data and the R pack- observed normalized R-value for a signal intensity age cghFLasso was used to assess copy number altera- divided by the mean signal of a reference sample. The tions. For genome annotation the University of Santa threshold of log2 ratio was defined as in the discovery Cruz (UCSC) Genome Browser [48] and Ensembl Gen- andgenotypingsectionto defineatruecopynumber ome Browser [49] were used. Gene and exon coordi- change that presents a pattern consistently different from nates were downloaded from Ensembl website version a diploid region. CNVs were defined as |log2 ratio| 65. Disease statuses of genes were obtained from the >threshold, where deletions are indicated by negative Online Mendelian Inheritance in Man database [50]. log2 ratios and duplications are indicated by positive log2 ratios. Deletion and duplication variants are special cases Access to data of multi-allelic CNVs. To minimize the risk of false posi- A database of genomic copy number variants, tives, we required each locus to be targeted by at least DoG_CNV, in the dog genome has been developed to ten consecutive probes to call a CNV, resulting in a final provide the annotation of CNVs discovered in this study resolution of approximately 9 kb. and a useful resource to assist with the assessment of Segmental duplications, defined as regions at least 1 kb CNVs in the contexts of canine variation and disease in length, at least 90% identical at two or more loci, and susceptibility [51]. In addition, the raw array data have not consisting entirely of mobile elements, were identi- been submitted to the Gene Expression Omnibus at fied by self-alignment of the genome as described in NCBI [52]. Zody et al. [35]. In the GC peak analysis we identified GC peaks by sliding two windows along the genome; a 10 Additional material kb window for background rate (same size as a CNV breakpoint) and centered in this a 500 bp window for Additional file 1: Supplementary methods, tables and figures. peak discovery. When the peak window showed a 1.5- Berglund et al. Genome Biology 2012, 13:R73 Page 16 of 17 http://genomebiology.com/2012/13/8/R73 6. Yang TL, Chen XD, Guo Y, Lei SF, Wang JT, Zhou Q, Pan F, Chen Y, Abbreviations Zhang ZX, Dong SS, Xu XH, Yan H, Liu X, Qiu C, Zhu XZ, Chen T, Li M, aCGH: array comparative genomic hybridization; CNV: copy number variant; Zhang H, Zhang L, Drees BM, Hamilton JJ, Papasian CJ, Recker RR, Song XP, DSB: double-stranded break; GO: gene ontology; KEGG: Kyoto Encyclopedia Cheng J, Deng HW: Genome-wide copy-number-variation study of Genes and Genomes; LINE: long interspersed nuclear element; NAHR: identified a susceptibility gene, UGT2B17, for osteoporosis. Am J Hum non-allelic homologous recombination; qPCR: quantitative polymerase chain Genet 2008, 83:663-674. reaction; SD: segmental duplication; SINE: short interspersed nuclear element; 7. Aitman TJ, Dong R, Vyse TJ, Norsworthy PJ, Johnson MD, Smith J, SNP: single-nucleotide polymorphism. Mangion J, Roberton-Lowe C, Marshall AJ, Petretto E, Hodges MD, Bhangal G, Patel SG, Sheehan-Rooney K, Duda M, Cook PR, Evans DJ, Acknowledgements Domin J, Flint J, Boyle JJ, Pusey CD, Cook HT: Copy number polymorphism This study was supported by the European Commission FP7 project LUPA- in Fcgr3 predisposes to glomerulonephritis in rats and humans. Nature GA201370 [43], the Swedish Research Council Formas and a European Science Foundation EURYI award to KLT. We thank all of the dog owners 2006, 439:851-855. who contributed samples used in this project. 8. Yang Y, Chung EK, Wu YL, Savelli SL, Nagaraja HN, Zhou B, Hebert M, Jones KN, Shu Y, Kitzmiller K, Blanchong CA, McBride KL, Higgins GC, Author details Rennebohm RM, Rice RR, Hackshaw KV, Roubey RA, Grossman JM, Tsao BP, Science for Life Laboratory, Department of Medical Biochemistry and Birmingham DJ, Rovin BH, Hebert LA, Yu CY: Gene copy-number variation Microbiology, Uppsala University, Box 582, SE-751 23, Uppsala, Sweden. and associated polymorphisms of complement component C4 in human Department of Basic Veterinary Sciences, Department of Medical Genetics, systemic lupus erythematosus (SLE): low copy number is a risk factor for Program in Molecular Medicine, Folkhälsan Institute of Genetics, Biomedicum and high copy number is a protective factor against SLE susceptibility in Helsinki, University of Helsinki, PO Box 63, 00014 Helsinki, Finland. European Americans. Am J Hum Genet 2007, 80:1037-1054. Department of Animal Breeding and Genetics, Swedish University of 9. Perry GH, Dominy NJ, Claw KG, Lee AS, Fiegler H, Redon R, Werner J, Agricultural Sciences, Box 597, SE-751 24 Uppsala, Sweden. Broad Institute Villanea FA, Mountain JL, Misra R, Carter NP, Lee C, Stone AC: Diet and the of Harvard and Massachusetts Institute of Technology, 7 Cambridge Center, evolution of human amylase gene copy number variation. Nat Genet 5 6 Cambridge, Massachusetts 02142, USA. www.eurolupa.org. Institut de 2007, 39:1256-1260. Génétique et Développement de Rennes, CNRS-UMR6290, Université de 10. Gonzalez E, Kulkarni H, Bolivar H, Mangano A, Sanchez R, Catano G, Rennes 1, Rennes, France. Nibbs RJ, Freedman BI, Quinones MP, Bamshad MJ, Murthy KK, Rovin BH, Bradley W, Clark RA, Anderson SA, O’Connell RJ, Agan BK, Ahuja SS, Authors’ contributions Bologna R, Sen L, Dolan MJ, Ahuja SK: The influence of CCL3L1 gene- JB performed the majority of the bioinformatics analyses and drafted the containing segmental duplications on HIV-1/AIDS susceptibility. Science 2005, 307:1434-1440. manuscript. EMN performed the array experiments. A-MM analyzed the SNP array data. MP helped to coordinate the array experiments. CA participated 11. Hastings PJ, Lupski JR, Rosenberg SM, Ira G: Mechanisms of change in gene copy number. Nat Rev Genet 2009, 10:551-564. in useful discussions. MCZ and TS identified segmental duplications. CH 12. Graubert TA, Cahan P, Edwin D, Selzer RR, Richmond TA, Eis PS, contributed bioinformatics analyses and constructed the website. KL-T and Shannon WD, Li X, McLeod HL, Cheverud JM, Ley TJ: A high-resolution HL conceived of the study and participated in its design and coordination. map of segmental DNA copy number variation in the mouse genome. MTW conceived and coordinated the bioinformatics analyses and wrote the PLoS Genet 2007, 3:e3. paper. All authors read and approved the final version. 13. Sharp AJ, Locke DP, McGrath SD, Cheng Z, Bailey JA, Vallente RU, Pertz LM, Clark RA, Schwartz S, Segraves R, Oseroff VV, Albertson DG, Pinkel D, Competing interests Eichler EE: Segmental duplications and copy-number variation in the The authors declare that they have no competing interests. human genome. Am J Hum Genet 2005, 77:78-88. 14. She X, Cheng Z, Zollner S, Church DM, Eichler EE: Mouse segmental Received: 23 April 2012 Revised: 27 June 2012 Accepted: 23 August 2012 Published: 23 August 2012 duplication and copy number variation. Nat Genet 2008, 40:909-914. 15. Myers S, Bowden R, Tumian A, Bontrop RE, Freeman C, Macfie TS, McVean G, Donnelly P: Drive against hotspot motifs in primates References implicates the PRDM9 gene in meiotic recombination. Science 2010, 1. Conrad DF, Pinto D, Redon R, Feuk L, Gokcumen O, Zhang Y, Aerts J, 327:876-879. Andrews TD, Barnes C, Campbell P, Fitzgerald T, Hu M, Ihm CH, 16. Baudat F, Buard J, Grey C, Fledel-Alon A, Ober C, Przeworski M, Coop G, de Kristiansson K, Macarthur DG, Macdonald JR, Onyiah I, Pang AW, Robson S, Massy B: PRDM9 is a major determinant of meiotic recombination Stirrups K, Valsesia A, Walter K, Wei J, Tyler-Smith C, Carter NP, Lee C, hotspots in humans and mice. Science 2010, 327:836-840. Scherer SW, Hurles ME: Origins and functional impact of copy number 17. Parvanov ED, Petkov PM, Paigen K: Prdm9 controls activation of variation in the human genome. Nature 2010, 464:704-712. mammalian recombination hotspots. Science 2010, 327:835-835. 2. Mills RE, Walter K, Stewart C, Handsaker RE, Chen K, Alkan C, Abyzov A, 18. Myers S, Freeman C, Auton A, Donnelly P, McVean G: A common sequence Yoon SC, Ye K, Cheetham RK, Chinwalla A, Conrad DF, Fu Y, Grubert F, motif associated with recombination hot spots and genome instability Hajirasouliha I, Hormozdiari F, Iakoucheva LM, Iqbal Z, Kang S, Kidd JM, in humans. Nat Genet 2008, 40:1124-1129. Konkel MK, Korn J, Khurana E, Kural D, Lam HY, Leng J, Li R, Li Y, Lin CY, 19. Karlsson EK, Lindblad-Toh K: Leader of the pack: gene mapping in dogs Luo R, Mu XJ, et al: Mapping copy number variation by population-scale and other model organisms. Nat Rev Genet 2008, 9:713-725. genome sequencing. Nature 2011, 470:59-65. 20. Karlsson EK, Baranowska I, Wade CM, Salmon Hillbertz NH, Zody MC, 3. Hurles ME, Dermitzakis ET, Tyler-Smith C: The functional impact of Anderson N, Biagi TM, Patterson N, Pielberg GR, Kulbokas EJ, Comstock KE, structural variation in humans. Trends Genet 2008, 24:238-245. Keller ET, Mesirov JP, von Euler H, Kampe O, Hedhammar A, Lander ES, 4. McCarroll SA, Huett A, Kuballa P, Chilewski SD, Landry A, Goyette P, Andersson G, Andersson L, Lindblad-Toh K: Efficient mapping of Zody MC, Hall JL, Brant SR, Cho JH, Duerr RH, Silverberg MS, Taylor KD, mendelian traits in dogs through genome-wide association. Nat Genet Rioux JD, Altshuler D, Daly MJ, Xavier RJ: Deletion polymorphism 2007, 39:1321-1328. upstream of IRGM associated with altered IRGM expression and Crohn’s 21. Parker HG, Shearin AL, Ostrander EA: Man’s best friend becomes biology’s disease. Nat Genet 2008, 40:1107-1112. best in show: genome analyses in the domestic dog. Annu Rev Genet 5. de Cid R, Riveira-Munoz E, Zeeuwen PL, Robarge J, Liao W, Dannhauser EN, 2010, 44:309-336. Giardina E, Stuart PE, Nair R, Helms C, Escaramis G, Ballana E, Martin- 22. Sutter NB, Bustamante CD, Chase K, Gray MM, Zhao K, Zhu L, Ezquerra G, den Heijer M, Kamsteeg M, Joosten I, Eichler EE, Lazaro C, Padhukasahasram B, Karlins E, Davis S, Jones PG, Quignon P, Johnson GS, Pujol RM, Armengol L, Abecasis G, Elder JT, Novelli G, Armour JA, Kwok PY, Parker HG, Fretwell N, Mosher DS, Lawler DF, Satyaraj E, Nordborg M, Bowcock A, Schalkwijk J, Estivill X: Deletion of the late cornified envelope Lark KG, Wayne RK, Ostrander EA: A single IGF1 allele is a major LCE3B and LCE3C genes as a susceptibility factor for psoriasis. Nat Genet determinant of small size in dogs. Science 2007, 316:112-115. 2009, 41:211-215. Berglund et al. Genome Biology 2012, 13:R73 Page 17 of 17 http://genomebiology.com/2012/13/8/R73 23. Salmon Hillbertz NH, Isaksson M, Karlsson EK, Hellmen E, Pielberg GR, 38. Cruz F, Vila C, Webster MT: The legacy of domestication: accumulation of Savolainen P, Wade CM, von Euler H, Gustafson U, Hedhammar A, deleterious mutations in the dog genome. Mol Biol Evol 2008, Nilsson M, Lindblad-Toh K, Andersson L, Andersson G: Duplication of FGF3, 25:2331-2336. FGF4, FGF19 and ORAOV1 causes hair ridge and predisposition to 39. Kim PM, Lam HY, Urban AE, Korbel JO, Affourtit J, Grubert F, Chen X, dermoid sinus in Ridgeback dogs. Nat Genet 2007, 39:1318-1320. Weissman S, Snyder M, Gerstein MB: Analysis of copy number variants 24. Olsson M, Meadows JR, Truve K, Rosengren Pielberg G, Puppo F, Mauceli E, and segmental duplications in the human genome: Evidence for a Quilez J, Tonomura N, Zanna G, Docampo MJ, Bassols A, Avery AC, change in the process of formation in recent evolutionary history. Karlsson EK, Thomas A, Kastner DL, Bongcam-Rudloff E, Webster MT, Genome Res 2008, 18:1865-1874. Sanchez A, Hedhammar A, Remmers EF, Andersson L, Ferrer L, Tintle L, 40. Korbel JO, Urban AE, Affourtit JP, Godwin B, Grubert F, Simons JF, Kim PM, Lindblad-Toh K: A novel unstable duplication upstream of HAS2 Palejev D, Carriero NJ, Du L, Taillon BE, Chen ZT, Tanzer A, Saunders ACE, predisposes to a breed-defining skin phenotype and a periodic fever Chi JX, Yang FT, Carter NP, Hurles ME, Weissman SM, Harkins TT, syndrome in Chinese Shar-Pei dogs. PLoS Genet 2011, 7:e1001332. Gerstein MB, Egholm M, Snyder M: Paired-end mapping reveals extensive 25. Parker HG, VonHoldt BM, Quignon P, Margulies EH, Shao S, Mosher DS, structural variation in the human genome. Science 2007, 318:420-426. Spady TC, Elkahloun A, Cargill M, Jones PG, Maslen CL, Acland GM, 41. Duret L, Galtier N: The covariation between TpA deficiency, CpG Sutter NB, Kuroki K, Bustamante CD, Wayne RK, Ostrander EA: An expressed deficiency, and G + C content of human isochores is due to a fgf4 retrogene is associated with breed-defining chondrodysplasia in mathematical artifact. Mol Biol Evol 2000, 17:1620-1625. domestic dogs. Science 2009, 325:995-998. 42. vonHoldt BM, Pollinger JP, Lohmueller KE, Han EJ, Parker HG, Quignon P, 26. Oliver PL, Goodstadt L, Bayes JJ, Birtle Z, Roach KC, Phadnis N, Beatson SA, Degenhardt JD, Boyko AR, Earl DA, Auton A, Reynolds A, Bryc K, Brisbin A, Lunter G, Malik HS, Ponting CP: Accelerated evolution of the Prdm9 Knowles JC, Mosher DS, Spady TC, Elkahloun A, Geffen E, Pilot M, speciation gene across diverse metazoan taxa. PLoS Genet 2009, 5: Jedrzejewski W, Greco C, Randi E, Bannasch D, Wilton A, Shearman J, e1000753. Musiani M, Cargill M, Jones PG, Qian ZW, Huang W, Ding ZL, et al: 27. Axelsson E, Webster MT, Ratnakumar A, Consortium L, Ponting CP, Genome-wide SNP and haplotype analyses reveal a rich history Lindblad-Toh K: Death of PRDM9 coincides with stabilization of the underlying dog domestication. Nature 2010, 464:898-902. recombination landscape in the dog genome. Genome Res 2012, 22:51-63. 43. LUPA.. [http://www.eurolupa.org/]. 28. Webber C, Ponting CP: Hotspots of mutation and breakage in dog and 44. Selzer RR, Richmond TA, Pofahl NJ, Green RD, Eis PS, Nair P, Brothman AR, human chromosomes. Genome Res 2005, 15:1787-1797. Stallings RL: Analysis of chromosome breakpoints in neuroblastoma at 29. Chen WK, Swartz JD, Rush LJ, Alvarez CE: Mapping DNA structural sub-kilobase resolution using fine-tiling oligonucleotide array CGH. Gene variation in dogs. Genome Res 2009, 19:500-509. Chromosome Canc 2005, 44:305-319. 30. Nicholas TJ, Cheng Z, Ventura M, Mealey K, Eichler EE, Akey JM: The 45. Peiffer DA, Le JM, Steemers FJ, Chang W, Jenniges T, Garcia F, Haden K, Li J, genomic architecture of segmental duplications and associated copy Shaw CA, Belmont J, Cheung SW, Shen RM, Barker DL, Gunderson KL: High- number variants in dogs. Genome Res 2009, 19:491-499. resolution genomic profiling of chromosomal aberrations using Infinium 31. Nicholas TJ, Baker C, Eichler EE, Akey JM: A high-resolution integrated map whole-genome genotyping. Genome Res 2006, 16:1136-1148. of copy number polymorphisms within and between breeds of the 46. Colella S, Yau C, Taylor JM, Mirza G, Butler H, Clouston P, Bassett AS, modern domesticated dog. BMC Genomics 2011, 12:414. Seller A, Holmes CC, Ragoussis J: QuantiSNP: an Objective Bayes Hidden- 32. Redon R, Ishikawa S, Fitch KR, Feuk L, Perry GH, Andrews TD, Fiegler H, Markov Model to detect and accurately map copy number variation Shapero MH, Carson AR, Chen W, Cho EK, Dallaire S, Freeman JL, using SNP genotyping data. Nucleic Acids Res 2007, 35:2013-2025. Gonzalez JR, Gratacos M, Huang J, Kalaitzopoulos D, Komura D, 47. g:Profiler.. [http://biit.cs.ut.ee/gprofiler/index.cgi]. MacDonald JR, Marshall CR, Mei R, Montgomery L, Nishimura K, Okamura K, 48. UCSC Genome Bioinformatics.. [http://genome.ucsc.edu/]. Shen F, Somerville MJ, Tchinda J, Valsesia A, Woodwark C, Yang F, Zhang J, 49. ENSMBL.. [http://www.ensembl.org/]. et al: Global variation in copy number in the human genome. Nature 50. OMIM.. [http://www.omim.org/]. 2006, 444:444-454. 51. DoG_CNV.. [http://dogs.genouest.org/LUPA.dir/CNV.html]. 33. Vaysse A, Ratnakumar A, Derrien T, Axelsson E, Rosengren Pielberg G, 52. Gene Expression Omnibus.. [http://www.ncbi.nlm.nih.gov/geo/]. Sigurdsson S, Fall T, Seppala EH, Hansen MS, Lawley CT, Karlsson EK, 53. Lindblad-Toh K, Wade CM, Mikkelsen TS, Karlsson EK, Jaffe DB, Kamal M, Bannasch D, Vila C, Lohi H, Galibert F, Fredholm M, Haggstrom J, Clamp M, Chang JL, Kulbokas EJ, Zody MC, Mauceli E, Xie X, Breen M, Hedhammar A, Andre C, Lindblad-Toh K, Hitte C, Webster MT: Wayne RK, Ostrander EA, Ponting CP, Galibert F, Smith DR, DeJong PJ, Identification of genomic regions associated with phenotypic variation Kirkness E, Alvarez P, Biagi T, Brockman W, Butler J, Chin CW, Cook A, Cuff J, between dog breeds using selection mapping. PLoS Genet 2011, 7: Daly MJ, DeCaprio D, Gnerre S, Grabherr M, et al: Genome sequence, e1002316. comparative analysis and haplotype structure of the domestic dog. 34. Bailey JA, Yavor AM, Massa HF, Trask BJ, Eichler EE: Segmental duplications: Nature 2005, 438:803-819. organization and impact within the current human genome project doi:10.1186/gb-2012-13-8-r73 assembly. Genome Res 2001, 11:1005-1017. Cite this article as: Berglund et al.: Novel origins of copy number 35. Zody MC, Garber M, Sharpe T, Young SK, Rowen L, O’Neill K, Whittaker CA, variation in the dog genome. Genome Biology 2012 13:R73. Kamal M, Chang JL, Cuomo CA, Dewar K, FitzGerald MG, Kodira CD, Madan A, Qin S, Yang X, Abbasi N, Abouelleil A, Arachchi HM, Baradarani L, Birditt B, Bloom S, Bloom T, Borowsky ML, Burke J, Butler J, Cook A, DeArellano K, DeCaprio D, Dorris L, Dors M, et al: Analysis of the DNA sequence and duplication history of human chromosome 15. Nature 2006, 440:671-675. 36. Zody MC, Garber M, Adams DJ, Sharpe T, Harrow J, Lupski JR, Nicholson C, Submit your next manuscript to BioMed Central Searle SM, Wilming L, Young SK, Abouelleil A, Allen NR, Bi W, Bloom T, and take full advantage of: Borowsky ML, Bugalter BE, Butler J, Chang JL, Chen CK, Cook A, Corum B, Cuomo CA, de Jong PJ, DeCaprio D, Dewar K, FitzGerald M, Gilbert J, • Convenient online submission Gibson R, Gnerre S, Goldstein S, Grafham DV, et al: DNA sequence of human chromosome 17 and analysis of rearrangement in the human • Thorough peer review lineage. Nature 2006, 440:1045-1049. • No space constraints or color figure charges 37. Zody MC, Jiang Z, Fung HC, Antonacci F, Hillier LW, Cardone MF, Graves TA, • Immediate publication on acceptance Kidd JM, Cheng Z, Abouelleil A, Chen L, Wallis J, Glasscock J, Wilson RK, Reily AD, Duckworth J, Ventura M, Hardy J, Warren WC, Eichler EE: • Inclusion in PubMed, CAS, Scopus and Google Scholar Evolutionary toggling of the MAPT 17q21.31 inversion region. Nat Genet • Research which is freely available for redistribution 2008, 40:1076-1083. Submit your manuscript at www.biomedcentral.com/submit

Journal

Genome Biology – Springer Journals

Published: Aug 1, 2012

Keywords: Animal Genetics and Genomics; Human Genetics; Plant Genetics and Genomics; Microbial Genetics and Genomics; Bioinformatics; Evolutionary Biology

There are no references for this article.