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

Learn More →

Genomic characteristics of cattle copy number variations

Genomic characteristics of cattle copy number variations Background: Copy number variation (CNV) represents another important source of genetic variation complementary to single nucleotide polymorphism (SNP). High-density SNP array data have been routinely used to detect human CNVs, many of which have significant functional effects on gene expression and human diseases. In the dairy industry, a large quantity of SNP genotyping results are becoming available and can be used for CNV discovery to understand and accelerate genetic improvement for complex traits. Results: We performed a systematic analysis of CNV using the Bovine HapMap SNP genotyping data, including 539 animals of 21 modern cattle breeds and 6 outgroups. After correcting genomic waves and considering the pedigree information, we identified 682 candidate CNV regions, which represent 139.8 megabases (~4.60%) of the genome. Selected CNVs were further experimentally validated and we found that copy number “gain” CNVs were predominantly clustered in tandem rather than existing as interspersed duplications. Many CNV regions (~56%) overlap with cattle genes (1,263), which are significantly enriched for immunity, lactation, reproduction and rumination. The overlap of this new dataset and other published CNV studies was less than 40%; however, our discovery of large, high frequency (> 5% of animals surveyed) CNV regions showed 90% agreement with other studies. These results highlight the differences and commonalities between technical platforms. Conclusions: We present a comprehensive genomic analysis of cattle CNVs derived from SNP data which will be a valuable genomic variation resource. Combined with SNP detection assays, gene-containing CNV regions may help identify genes undergoing artificial selection in domesticated animals. Background genotyping assays, QTL distributions and artificial selec- With two cattle genome assemblies available (Btau_4 tion signatures in dairy cattle have been reported [9,10]. and UMD3) [1,2], the cattle research community has Copy Number Variation (CNV) represents another been focusing on single nucleotide polymorphisms important source of genetic variation that provides (SNPs) as the main source of genetic variation in cattle. genomic structural information complementary to SNP This effort led to the development of the cattle SNP data. Genomic structural variations ranging from 1 kb map [3] and the Illumina Bovine SNP50 (> 50,000 SNP to 5 Mb comprise mainly of CNVs in the form of large- probes) genotyping array [4,5]. Evaluations of genetic scale insertions and deletions, as well as inversions and merit based on SNPs became a reality in early 2009 translocations [11]. In humans, ~29,000 CNVs that cor- leading to an acceleration of improvements to dairy and respond to over 8,400 CNV regions have been identified, beef breed stocks [6-8]. Widespread use of the Bovi- and over 9,000 genes have been mapped within or near neSNP50 array has resulted in the availability of tens of regions of human structural variation [12,13]. Some of thousands of SNP genotyping results. Based on SNP these CNVs have been shown to be important in both normal phenotypic variability and disease susceptibility. * Correspondence: [email protected] Several recent publications have reviewed the effects of Bovine Functional Genomics Laboratory, ANRI, USDA-ARS, Beltsville, CNVs on gene expression and human diseases [14-17]. Maryland 20705, USA Due to their low cost and high-density, SNP arrays have Full list of author information is available at the end of the article © 2011 Hou 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. Hou et al. BMC Genomics 2011, 12:127 Page 2 of 11 http://www.biomedcentral.com/1471-2164/12/127 been routinely used for human CNV detection and ana- Results and Discussion lysis [13]. Compared to CGH arrays which only report Optimization of cattle CNV detection relative signal intensities, SNP arrays collect normalized A total of 58,336 markers were selected for the Bovi- total intensities (Log R ratio - LRR) and allelic intensity neSNP50 assay [4,5]. Except for 1,389 markers which ratios (B allele frequency - BAF) which represent overall failed to pass manufacturer assay production pipeline, we copy numbers and allelic contrasts [18]. Multiple algo- intentionally kept all remaining 56,947 markers without rithms have been developed to exploit SNP data to iden- any other filtering. These included 1,465 markers (2.57%) tify CNVs, including QuantiSNP [19], PennCNV [20], which had a call rate of 0. The markers with a call rate of Birdseye [21] and Cokgen [22]. Comparisons of the 0 are resistant to the default biallelic SNP clustering and strengths and weaknesses of these algorithms have been often fall in CNV regions. Compared to the standard published [23,24]. As one of the leading methods, BovineSNP50 Genotyping Beadchip v1 featuring 54,001 PennCNV incorporates multiple sources of information, SNP probes, 2,946 more SNPs were included in our ana- including total signal intensity and allelic intensity ratio lysis, of which, ~17% located in cattle segmental duplica- at each SNP marker, the distance between neighboring tion (SD) regions [32], ~9% overlapped with the CNVRs SNPs, and the allele frequency of SNPs. PennCNV also detected by array CGH method [29], and ~27% contribu- integrates a computational approach by fitting regres- ted to the CNVRs reported here. sion models with GC content to overcome “genomic We tested the cattle CNV calls made with or without waves” [25,26]. Furthermore, PennCNV is capable of the -gcmodel option on Batu_4.0 to identify the impact considering pedigree information (a parents-offspring of genomic waves on CNV calling. Agreeing with pre- trio) to improve call rates and accuracy of breakpoint viousresults [26],wefound thetotal CNVR counts prediction as well as to infer chromosome-specific SNP were higher without -gcmodel (719) than those with genotypes in CNVs [27]. -gcmodel enabled (682). However, only 86.80% (592/ Previous cattle studies have produced a number of CNV 682) of the gcmodel calls directly overlapped with datasets. For example, our earlier array CGH survey using 79.28% (570/719) calls made without gcmodel, revealing 3 Holstein bulls identified 25 germline CNVs [28]. a ~20% CNV discordancy rate. These discordant calls Recently, we reported a broader, systematic CNV survey were likely due to false positives called from the differ- in 90 cattle using array CGH [29]. We identified over 200 entiating signal intensities caused by “genomic waves” candidate CNV regions (CNVRs); some of which are likely rather than by real CNV events. This further demon- to underlie cattle domestication and breed formation. strated that genomic waves have a significant effect on Fadista et al. recently reported 304 CNV regions in 20 ani- this type of analysis. mals of 4 cattle breeds using high-density array CGH [30]. We also compared results of PennCNV using -test, Besides array CGH experiments, other evidences for cattle -trio and -joint options sequentially. In other words, we CNV came from SNP genotyping results, where a screen compared data resultant from not considering trio infor- of Bovine HapMap Consortium samples (over 500 animals mation (-test), considering trio information only after from multiple cattle breeds) identified 79 candidate dele- calling (-trio) and finally by considering trio information tions using an earlier version of cnvPartition [5]. However, in a simultaneous fashion during CNV calling (-joint) these results only included homozygous deletions which (Additional file 1: Table S5). Consistent with the earlier were validated by multiple observations. A recent paper comparisons using simulated and real SNP data [27,33], reported 368 unique CNV regions from 265 Korean Han- trio information significantly increased our CNV call woo cattle based on BovineSNP50 genotyping data; how- rates. The result of the -joint option (1276 calls) was sig- ever, during the PennCNV calling, the “genomic waves” nificantly higher than those of the other options: -test pattern was not discussed and pedigree information was (684 calls) and -trio (1019 calls). After merging overlap- not considered [31]. In this study, we reprocessed the pub- ping CNVs, ~87% of the 682 CNVRs deduced from the lished Bovine HapMap Consortium SNP genotyping -joint option overlapped with those deduced from the results using optimal settings for PennCNV by adjusting -test and -trio options (both with a total of 621 CNVRs). for “genomic waves” and utilizing trio/pedigree informa- Due to its improved call sensitivity and breakpoint infer- tion whenever possible. We identified 682 candidate CNV ence, the -joint option reported about 13% more CNVRs regions in a diverse panel of 521 animals from 21 different which were not detected by the -test or -trio options. breeds. We also included 18 animals from 6 outgroups to derive the ancestral states of CNVs. We then compared Cattle CNV discovery and distribution this CNV call set with the existing cattle CNV call sets, Due to issues regarding CNVR calls, we excluded chrX validated several novel CNVR calls and discussed the evo- and chrUn from our analysis. In our initial analysis of lutionary impact of cattle CNVs. chrX, it was found that almost half of the potential Hou et al. BMC Genomics 2011, 12:127 Page 3 of 11 http://www.biomedcentral.com/1471-2164/12/127 CNVRs were unreasonably large (> 1 Mb) and several were based on the same raw WGS reads. The most events were present in high frequencies (> 25%). This is obvious difference between the two assemblies is that likely due to the fact that PennCNV assumes two copies Btau_4.0 unplaced contigs are placed on UMD3. This of each SNP as the normal copy number state, which resulted in more markers that were on Btau_4.0 ChrUn was likely not the case within the pseudoautosomal contigs to be placed on UMD3 autosomes, which could region [34] and segmental duplications [32] on chrX. partially explain the increase in the CNVR counts. Since Additionally, since chrX sequence and annotation also themajorityofcattlegenomeannotations were per- formed on the Btau_4.0 assembly, we focused on further differ dramatically between Btau_4.0 and UMD3 builds, characterization of the 682 high-confidence CNV we considered the CNV calls on chrX as unreliable and excluded them from further analysis. Since chrUn only regions from Btau_4.0 autosomes. contains unassigned sequence contigs, it was not These 682 CNVRs include 370 loss, 216 gain and 96 included due to the lack of sequence and SNPs as well both (loss and gain within the same region) events, ran- as the SNP mapping uncertainty. ging from 32,566 to 5,569,091 bp with a mean or med- Within the placed autosomes, a total of 3,666 CNVs in ian of 204,965 or 131,179 bp, respectively (Additional 521 samples were detected and an average of 7.09 gain file 1: Table S2). Loss events are approximately 1.7-fold or loss events were evident in each sample (Table 1). more common than gain events, but have slightly smal- CNVRs were determined by aggregating overlapping ler sizes than gain regions on average. Furthermore, 278 CNVs identified across all samples, following previously CNVRswerefound in only onesample(Unique), 404 published protocols [13]. A total of 682 high-confidence CNVRs were present in two or more animals or breeds autosomal CNVRs were identified, covering 139.8 Mb of and 18 of 404 multiple events had a frequency >5% polymorphic sequence and corresponding to 5.49% of (Table 1 and Additional file 1: Table S2). These datasets the autosomal genome sequence (139.8/2,545.9 Mb) and confirm that segregating CNVs exist among these 21 4.60% of the whole cattle genome (139.8/3,036.6 Mb, cattle breeds and/or groups, which is consistent with Figure 1 and Additional file 1: Table S2). our earlier results based on array CGH [29]. In general, To test the stability of CNV calls with respect to dif- the number of CNVs identified in each sample is con- ferent genome builds and SNP mapping, we also sistent with SNP estimates of breed-specific founding migrated 56,408 out of 56,947 SNP markers from and effective population sizes and levels of polymorph- Btau_4.0 to UMD3 using the UCSC liftOver tool [35], ism based on ≥50,000 SNPs [5]. As shown in Table 1, and repeated the entire calling analyses to ensure con- more CNV events were detected in indicine (11.41 per sample) than in African groups (7.21 per sample) and sistency in calls (Additional file 1: Tables S3 and S4). Only 61 more CNVRs were identified on the UMD3 composite (7.17 per sample). The taurine breeds (6.23 assembly (making a total of 743 CNVRs). A simple com- per sample) had the fewest detected CNVs. While some parison indicated that the total coverage of variable of these differences could be related to the fact that the regions were 13.05% larger on UMD3 (158.0 Mb, SNP markers were designed based on the Btau_4.0 Additional file 1: Table S3) than on Btau_4.0. For all reference genome (which was derived from the three CNVR types (gain, loss and both), counts sequence of a Hereford cow of European origin; Domin- increased slightly. This was expected as both assemblies ette 01449), this observation is consistent with the Table 1 CNV events by species and breeds Btau_4.0 Sample Count Unique Gain Loss Gene Total Length Taurine 366 2,256(6.23) 239(0.66) 1,454(4.02) 802(2.22) 4,744(13.10) 373,001,599(165,337) Composite 46 330(7.17) 23(0.50) 224(4.87) 106(2.30) 651(14.15) 113,483,966(142,032) Indicine 70 799(11.41) 62(0.89) 401(5.73) 398(5.69) 1,464(20.91) 57,402,891(173,948) African Breeds 39 281(7.21) 38(0.97) 213(5.46) 68(1.74) 775(19.87) 54,728,022(194,761) CNV 521 3,666(7.09) 362(0.70) 2,292(4.43) 1,374(2.66) 7,634(14.77) 598,616,478(163,288) b c d d CNVR 521 682 278 216 370 1,263 139,786,166(204,965) Outgroup CNV 18 1,003(55.72) 284(15.78) 48(2.67) 955(53.06) 2,603(144.61) 442,235,607(440,912) CNVR 18 483 187 21 458 1,593 276,846,573(573,181) The numbers in parentheses are normalized by sample counts except that the lengths in parentheses are average lengths normalized by CNV counts. At sample level, each sample has 7.09 (3666/517) CNVRs averagely and 6.23 (2256/362) for Taurine, since there are 4 taurine individuals without identified CNVs; These numbers are nonredundent CNVR counts. 278 CNVRs are unique to one sample while 404 CNVRs are shared by at least 2 individuals or breeds and 18 of 404 d e multiple events have frequency >5%; Besides 370 loss and 216 gain CNVRs, there are 96 CNVRs containing both loss and gain events; Outgroup animals are not included in the total counts for CNV and CNVR. Hou et al. BMC Genomics 2011, 12:127 Page 4 of 11 http://www.biomedcentral.com/1471-2164/12/127 Figure 1 Genomic landscape of cattle copy number variations and segmental duplications. CNV regions (682 events, 139 Mb, ~4.60% of the bovine genome) reported by 521 SNP genotyped individuals are shown above the chromosomes in green (gain), red (loss) and dark blue (both), while below are the CNV regions (177 events, 28 Mb, ~1% of the bovine genome) reported by 90 array CGH experiments by Liu et al. The bar height represents their frequencies: short (appeared in 1 sample), median (≥2 samples) and tall (≥5 samples). Segmental duplications (94.4 Mb, 3.1% of the bovine genome) predicted by two independent computational approaches are illustrated on the chromosomes in red (WSSD), blue (WGAC) or purple (both). The patterns are depicted for all duplications for ≥5 kb in length and ≥90% sequence identity. The gaps in the assembly are represented on the chromosomes as white ticks. concept of subspecies divergence and supports the chrX and chrUn, pericentromeric and subtelomeric hypothesis of multiple independent domestications of regions each represent 3.42% of genomic sequence but cattle in the Fertile Crescent, Southwest Asia and likely show an enrichment of 1.5-2.4-fold more CNVRs (both Africa [36,37]. P values <0.001) and contain 7.78-12.54% of all poly- Cattle CNVs are distributed in a nonrandom fashion morphic sequence. at two different levels. First, CNV content varies signifi- cantly among different chromosomes. The proportion of Quality assessment of selected CNV Regions any given known chromosome susceptible to CNV The quality of our 682 CNV calls was assessed in mul- regions varies from 1.32-8.80% (Additional file 1: Table tiple ways, though our first assessment was a compari- S2). Chromosomes 1 and 6 show the greatest enrich- son against existing cattle CNV datasets (Table 2 and ment for CNV (Figure 1 and Additional file 1: Table S2) Figure 2). One of the earlier datasets included 79 fil- with almost two-fold of the variable content of the auto- tered deletion variants (representing 42 unique geno- somal average. It is interesting to note that these chro- mic loci and 9 single SNPs) reported earlier using the mosomes do not have the highest SD content [29,32]. Illumina genotyping software module cnvPartition Furthermore, similar to the human, mouse, rat and dog v1.0.2 [5]. Nineteen of our CNVRs overlapped with 11 genomes, there are a greater proportion of CNVs near of the deletion variants (21.57%) in that dataset (Table pericentromeric and subtelomeric regions. Excluding 2). We also identified 129 CNVRs (18.91%) in our Hou et al. BMC Genomics 2011, 12:127 Page 5 of 11 http://www.biomedcentral.com/1471-2164/12/127 Table 2 Summary of genome-wide studies of cattle copy number variations Study Assay Count CNVR Size Marker Sample Breed Type Count Range (kb) Median Mean Total (kb) (kb) (Mb) Matukumalli BovineSNP50 54,001 556 21 Deletion only 51 22.92- 394.87 960.67 49.0 et al. 2009 11,050.69 Liu et al. 2010 Array CGH ~385,000 90 17 Deletion, 163 18.00-1,261.90 86.19 153.75 25.1 insertion Bae et al. 2010 BovineSNP50 54,001 265 1 Deletion, 368 25.35-967.18 128.33 171.49 63.1 insertion Fadista et al. 2010 Array CGH ~6,300,000 20 4 Deletion, 254 1.72-2,031.34 15.51 62.26 15.8 insertion This study BovineSNP50 56,947 521 21 Deletion, 682 32.57-5,569.09 131.18 204.97 139.8 insertion a b This includes 9 independent SNPs and 42 CNVRs. The statistics are calculated for 42 CNVR excluding the 9 SNPs; This is the number excluding chrX and chrUn; This is the number excluding chrX, chrUn and mitochondrial sequence. dataset that overlapped with 128 CNVRs from a SNP- an arrayCGH-based studyof20cattlefromfour based CNV study on 265 Korean Hanwoo cattle [31] breeds [30]. Since our dataset excluded CNV calls in (Figure 2B). The Hanwoo CNV study identified 368 the chrX, chrUn and mitochondrial sequences, we CNVRs in total, so our dataset overlapped with 34.83% compared our autosomal CNVR calls (682 CNVRs) to of their calls [31]. We then compared our calls against the autosomal CNV calls of that study (254 CNVRs) Bae et al (2010) This study Merged This study dataset ~482 ~200 ~518 554 46.6 Mb 123.3 Mb 16.5 Mb 66.5 Mb 100 Mb 29.8 Mb A B Count Length This study 682 139.8 Mb This study ~600 126 Mb Merged dataset 718 99.3 Mb ~24 ~30 2.4 Mb 7.1 Mb 28 368 63.1 Mb Bae et al (2010) 4.3 Mb ~86 ~183 Fadista et al (2010) 254 15.8 Mb 7.1 Mb 11.7 Mb ~19 Fadista et al (2010) Liu et al (2010) 2 Mb 163 25.1 Mb Liu et al (2010) Figure 2 Comparisons between identified 682 CNVRs in this study and the other existing cattle CNVR datasets in terms of count and length. A, compared to the total nonredundant CNVR merged from existing published datasets; B, compared to CNVR derived from SNP array (Bae et al, 2010); C, compared to two CNVR datasets derived from array CGH studies (Liu et al,2010; Fadista et al, 2010); D, the summaries and legends of existing cattle CNVR datasets. Hou et al. BMC Genomics 2011, 12:127 Page 6 of 11 http://www.biomedcentral.com/1471-2164/12/127 [30].Only51ofour CNVRs(7.48%)directlyover- (50 animals), the overlap included 100% of our filtered lapped with 55 of their calls (21.65%, Figure 2C). Our dataset. This further demonstrated that large, common final comparison was against our previous array CGH- CNVRs can be reliably detected through using different based study of 90 animals from 14 breeds which detection technologies even when the majority of sam- resulted in 163 autosomal CNVR calls [29]. In this ples were different. For example, our current SNP array comparison, 57 of our SNP-based CNVR calls (8.36%) study identified most of the large, common CNVRs which were confirmed in our published results [29]. overlapped with 59 CNVRs derived from array CGH After comparison with other existing datasets, we (36.20%,Figure2C).Ifweonlyfocused onthe16Hap- foundthat~70% of our CNVRcalls were notreported Map samples which were assessed by both platforms (60 CNVRs derived from array CGH and 106 CNVRs in the literature. In order to confirm these novel reported by SNP array), there were 21 overlapping CNVRs, we performed 24 quantitative PCR (qPCR) CNVRs: 19 for array CGH (31.67%, 19/60), and 20 assays for 15 low frequency, novel CNVR calls spread for SNP array (18.87%, 20/106). When we merged among seven individuals (Additional file 1: Table S9). existing CNV datasets, a total of ~ 200 out of 682 Nine of the CNV regions had two target amplicons (about 30%) newly identified CNVRs overlapped with placed near two different SNP loci. Out of 24 total loca- them (Figure 2A). tions, 11 loci (~48%) were in agreement with CNV esti- It is expected that the variants identified in these stu- mates by PennCNV (Additional file 1: Table S9 and dies do not overlap, suggesting a vast amount of CNVs Figure S1). When counting the CNV regions, 9 out of exist in cattle population and saturation for this type of 15 (60%) CNV regions had positive qPCR confirmations variation has not yet been approached. It is likely that in at least one location. If CNVRs previously validated many thousands of more common structural variants in the literature [29] were also included, approximately may still remain undiscovered in the cattle genome. two third (30% + 70% × 48%) of our detected CNVRs A similar situation was encountered in human CNV stu- had positive confirmations. dies using the early version of SNP, CGH arrays and As expected, the Bovine SNP50 platform has a large detection methods [38,39]. For example, although resolution limit under the current PennCNV calling cri- cnvPartition detects CNVs by processing the similar raw teria. The size of the CNVRs detected ranged from 32.6 data as PennCNV (i.e. LRR and BAF), it is based on a kb to 5.6 Mb, with a median size of 131.2 kb (Additional different proprietary sliding window algorithm. Only file 1: Table S2). This is partially due to the fact that the Bovine SNP50 assay was originally developed for high- those homozygous deletion events segregating in differ- throughput SNP genotyping in association studies. ent animals were reported due to concerns with the quality of calls [5]. In the future, high-density SNP Although CNV detection is feasible with SNP arrays, it arrays combined with improved CNV calling algorithms is impaired by low density and non-uniform distribution could remedy these differences. of SNPs, especially in CNV and SD regions. Compared Besides the technology and detection method differ- to a CGH array, a SNP array lacks non-polymorphic ences, the following could also contribute to the probes designed specifically for CNV identification. observed differences: (1) sampling differences: 521 indi- Thus, only the large CNVRs are expected to be identi- viduals from 21 diverse breeds and/or groups were fied with the Bovine SNP50 assay. This explains the dif- included in our study; (2) genome coverage biases: ference in CNV length between our study and the 56,947 markers were included in our study rather than a earlier results. subset of “well-behaved” SNPs (54,001 markers) which The discrepancies between the qPCR and PennCNV exclude those SNPs in CNV-rich regions; (3) correction resultsmay representsmall CNVeventsthatwere of genomic waves in order to minimize false positive missed in the PennCNV calls, or instances where SNPs calls; and (4) trio/pedigree information was fully caused the qPCR reaction to fail or be suboptimal but explored in our study to improve the accuracy and call did not affect the SNP assay. Despite the fact that a rate of CNVs. When filtering criteria varying the CNVR two-copy state was assumed for test PCR loci in Domin- length and frequency were applied, we observed signifi- ette, smaller CNV events in Dominette may have evaded cant overlap within our 2 datasets derived from SNP detection by PennCNV. If our test primers amplified a arrays and array CGH (Additional file 1: Table S8). For small CNV event in Dominette, that would skew the example, when the large CNVRs (SNP count = 10, a relative copy number estimates of our qPCR reactions. Although qPCR primers were designed within 250 bp median ~574kb) were considered, the overlap reached around the target SNP positions, additional SNPs and 21.74%. When the CNVR frequency was increased to 1, small indels may have influenced the hybridization of 2, or 5% (animal count = 5, 10, or 25, respectively), the overlap increased to 89.47%. When we filtered the the qPCR primers in some animals, thereby reducing CNVR frequency to greater than 10% of our population primer efficiency. Other causes may also contribute to Hou et al. BMC Genomics 2011, 12:127 Page 7 of 11 http://www.biomedcentral.com/1471-2164/12/127 the discrepancy in CNVR validation by qPCR. The draft functions, and provides a rich resource for testing status of the cattle genome assembly and the low SNP hypotheses on the genetic basis of phenotypic variation densityoftheBovineSNPassaymakeitdifficult to within and among breeds. determine the real breakpoints of CNVRs. For example, Consistent with similar CNV analyses in other mam- multiple, neighboring, discrete CNV events could result mals (human, mouse and dog), several of these CNVs, in a larger call by PennCNV; therefore, giving an which are important in drug detoxification, defense/ over estimation of the CNV size. Therefore, it cannot innate and adaptive immunity and receptor and signal be ruled out that the qPCR primers used to confirm recognition, are also present in cattle. These gene families include olfactory receptors, ATP-binding cas- theCNVRs mayhavebeendesignedoutside the breakpoints. sette (ABC) transporters, Cytochrome P450, b-defensins, interleukins, the bovine MHC (BoLA) and multiple CNVs overlap with segmental duplications and other solute carrier family proteins which support the shared genomic features GO termsamong mammalsasshown in Additional file Following previous studies of other genomes, we 1: Table S10. For gene families that went through cattle- detected the association between CNVRs and SDs. specific gene duplication [32], such as interferon tau, Agreeing with previous predictions regarding cattle SDs pregnancy-associated glycoproteins, SCP2 and ULBP [32], a local tandem distribution pattern is predominant and WC1.1 subfamilies, we also detected marked varia- in our cattle CNVR dataset (Figure 1). It should be tion in copy number between individuals and across noted that about 25.66% (175/682) of CNV regions diverse cattle breeds and/or groups (Additional file 2: directly overlap with cattle SDs with an overlapping Table S6). It is intriguing to note that we also detected span of 16,283,071 bp (11.65% of the total 139,786,166 variations of TLR3 (toll-like receptor 3) and PPARA bp). Approximately 12.06% (356/2952) of the SDs (peroxisome proliferator-activated receptor alpha). This (excluding chrX and chrUn) identified by WGAC and current CNV survey further supports a hypothesis that WSSD [32] exhibit CNVs. In comparison, 58.90% of the the generation of new CNV insertions and deletions CNVRs (96/163) detected by using array CGH [29] may be a constant phenomenon in multiple cattle excluding X chromosome overlap with cattle SDs, corre- breeds/individuals [41]. sponding to 15,176,612 bp (60.56% of the total We also overlapped our CNVRs with two sets of 25,061,646 bp). The proportion of our new CNVR calls genomic regions under positive or balance selection (identified in this study) that overlap with SDs reaches detected by iHS and F using SNP data [3,10] (Addi- ST approximately 40% compared to 61% in our previous tional file 3: Table S11). By doing so, we have identified study. This lower overlap fraction probably reflects the CNV regions that may span potential cattle QTLs and fact that the BovineSNP50 array used in this study is human orthologous OMIM genes influencing disease biased against cattle SD regions. SNP density on the susceptibility (Additional file 3: Table S11). For instance, array drops by one-third (from 21 probes/Mb in unique multiple CNV regions directly overlap with QTLs for regions down to 14 probes/Mb) in SD regions. We also significant and typical economic traits and 87 out of 682 failed to detect any correlation between 682 CNV CNVRs correspond to loci known to cause disease in regions and evolutionary breakpoint regions (EBRs). humans. However, since the cattle genome and cattle Compared to the genomic averages, cattle-specific EBRs QTLs are less well defined, future study is warranted. and artiodactyl-specific EBRs do not show enrichments of CNV sequences [40]. This negative result is consent Cattle CNV frequency differences among breeds with the fact that EBRs have fewer overlaps with SD To highlight the potential evolutionary contributions of regions. these CNVs to cattle breed formation and adaptation, we queried 91 CNVRs that have breed-specific CNV fre- Gene Content of Cattle CNV regions quency differences (Additional file 1: Table S12). We Within autosomes, the 682 CNV regions overlap with only considered breeds that had at least 12 samples and 1,679 Ensembl peptides, corresponding to 1,263 unique any detectable variations must be identified in at least 3 Ensembl genes (Table 1 and Additional file 2: S6). individuals or 10% of samples (for Holstein, Angus and About 55.57% (379/682) of high-confidence CNVRs Limousin where n > 30). Fifty-eight of these CNVRs completely or partially span cattle Ensembl genes. We correspond to annotated genes or gene families; many assigned PANTHER accessions to a total of 1,263 over- of which were identified in other mammals as influen- lapping genes. Statistically significant over or under cing adaptation to the environment. Some of the anno- representations were observed for multiple categories tated genes are known to be important in cattle (Additional file 1: Table S10). This set of copy number adaptation including CNVR266(IFNT) in Brown Swiss, variable genes possess a wide spectrum of molecular CNVR122 (SCP2) in Hereford [32] and CNVR178 Hou et al. BMC Genomics 2011, 12:127 Page 8 of 11 http://www.biomedcentral.com/1471-2164/12/127 (Olfactory Receptors) in most of breeds [See also [42]]. results, including those collected for the Bovine Hap- The differences of CNV frequency among cattle breeds Map project [3] (Additional file 1: Table S1). PennCNV supported our earlier hypothesis that some cattle CNVs quality filters were applied after the CNV detection, are likely to arise independently in breeds, are likely to resulting in 521 distinct high quality genotyping results contribute to breed differences and are therefore related from the original 556 animals. This panel included 366 to the breed formation and adaptation. However, the animals from 14 taurine dairy and beef breeds, 70 ani- observed differences between breed variations could be mals from three breeds of predominantly indicine back- caused by both selection and genetic drift due to genetic ground, 46 animals from two breeds that are Taurine × bottlenecks for some breeds. Our findings, therefore, Indicine composites, and 39 animals from two African must be confirmed with an even larger sample size. groups, one of which (Sheko) is an ancient hybrid. It is worth to note that for many of the breeds, individuals CNVs in outgroup animals were sampled from more than one continent to repre- For the 18 individuals in outgroups, which were ana- sent the global cattle population. This panel contained lyzed similarly together with 521 modern breeds indivi- 39 trios where both parents and an offspring were geno- duals, 1003 CNVs and 483 CNVRs were detected, typed. Additionally, we included 18 animals from 6 out- covering 276.8 Mb base pairs, with 21 gain and signifi- groups (Bos gaurus -Gaur, Bos bison - North American cantly more (458) loss events (Table 1 and Additional Bison, Bubalus depressicornis- Lowland Anoa, Bos java- file 1: S7). About 34.60% of our current CNVRs (236/ nicus -Banteng, Bos grunniens -Yak,and Syncerus caf- 682) directly overlapped with 37.47% of these ancient fer -Cape Buffalo) with 1 trio information to derive the CNVRs (181/483) derived from ancient outgroups, ancestral states of CNVs. Among these outgroups, the which indicates over one third of CNVRs were likely average sample call rate was 89.91%, reflecting their ancestral. We suspected this observation of more loss divergent relationship from Bos Taurus. events than gain events was at least partially related to the high genetic divergences between these outgroup Identification of cattle CNVs animals and the cattle reference genome. With addi- PennCNV algorithm [20] was only applied to autosomes tional cattle, sheep, goat and pig CNV papers published (-lastchr 29) to detect cattle CNV in this study. In our recently [43-46], it will be interesting to look into the initial analysis, chrX (-chrx) was also considered sepa- evolutionary mechanism of CNVs within livestock rately from automosomes. PennCNV incorporates multi- animals. ple sources of information together, including LRR and BAF at each SNP marker, more realistic models for Conclusions state transition between different copy number states We have performed a comprehensive genomic analysis based on the distance between neighboring SNPs, popu- of cattle CNVs based on whole genome SNP genotyping lation frequency of B allele (PFB), the allele frequency of data, therefore providing a valuable genomic variation SNPs, and the pedigree information where available, resource. A total of 682 CNVRs were identified, cover- into a hidden Markov model (HMM). Both LRR and ing 139.8 megabases (~4.60%) of the genome. A subset BAF were exported from Illumina GenomeStudio Geno- of these CNVRs showed Mendelian inheritance and typing Module v1.0 software given the default clustering were also confirmed in other cattle CNV studies and file for each SNP. The PFB file was calculated based on other mammalian species. As high density cattle SNP the BAF of each marker in this population. Because genotyping data are becoming available, CNVs com- there were 153 out of 556 animals (~27.5%) with abso- binedwithSNPs, mayhelpidentify genes undergoing lute values of waviness factor larger than 0.04 in our ori- artificial selection in domesticated animals. ginal analysis, the genomic waves were adjusted using the -gcmodel option. The cattle gcmodel file was gener- Methods ated by calculating the GC content of the 1Mb genomic Selection of cattle breeds and animals region surrounding each marker (500kb each side). For It hasbeendemonstratedthatthe BovineSNP50geno- comparison, the analysis without considering gcmodel typing array provides a robust resource for genome- was also conducted. Three different PennCNV options wide, high-density SNP genotyping of cattle and for were performed wherever possible: 1) -test: the indivi- population genetic analyses of closely related artiodactyl dual-calling algorithm that treats family members as if species [4,47]. In which, less than 3% of markers had they were unrelated; 2) -trio: the posterior-calling algo- call rates below 99.94%, the average call rate for indivi- rithm which accommodates family information to dual samples was greater than 97.5% and 85% of sam- improve the accuracy of individual-based CNV calling ples had call rates above 98.8% [5]. Cattle CNVs in this and boundary prediction; 3) -joint: the joint-calling algorithmthatidentifiesCNVsusing family data study were detected by using the same SNP genotyping Hou et al. BMC Genomics 2011, 12:127 Page 9 of 11 http://www.biomedcentral.com/1471-2164/12/127 simultaneously. After CNV detection, filtering of low- RefSeq and in silico mapped human RefSeq (the UCSC quality samples was carried out with the default cutoffs: Genome Browser website at http://genome.ucsc.edu/). standard deviation (STD) of LRR as 0.30, BAF drift as We obtained a catalog of all bovine peptides from 0.01 and waviness factor as 0.05. The filtered results Ensembl. This yielded 26,271 peptides, 1,679 of which from the three algorithms were compared in terms of overlap with predicted 682 high-confidence CNV CNV numbers, lengths and number of SNP in CNVs regions, and correspond to 1,263 unique Ensembl genes. (Additional file 1: Table S5). The final CNVs set was the Using the PANTHER classification system, we tested the nonredundant combinationof CNVsfromthe -joint hypothesis that the PANTHER molecular function, bio- results for family trio members and the -test results for logical process and pathway terms were under- or over- unrelated individuals. For the outgroup animals, quality represented in CNV regions after Bonferroni corrections filtering was not performed due to their divergent rela- [32]. It is worth noting that a portion of the genes in tionship from Bos Taurus. CNVRs are determined by the bovine genome has not been annotated or has been aggregating overlapping CNVs identified across all sam- annotated with unknown function, which may influence ples [13]. the outcome of this analysis. CNV validation Web Site References array CGH experiments were performed as previously The Database of Genomic Variants: http://projects.tcag. described [11]. Primers were designed for qPCR valida- ca/variation/ tion using the Primer3 webtool http://frodo.wi.mit.edu/ Ensembl genes ftp://ftp.ensembl.org/pub/current_fasta/ primer3/ by limiting amplicon length to 150 bp to 250 bp bos_taurus/pep/ and by incorporating a GC clamp of 2. All other settings PANTHER http://www.pantherdb.org/ were left at the default. Primer information is shown in OMIM http://www.ncbi.nlm.nih.gov/omim/ Additional file 1: Table S9. Quantitative PCR experiments OMIA http://omia.angis.org.au/ were conducted using SYBR green chemistry in triplicate QTL http://www.animalgenome.org/ reactions, each with a reaction volume of 25 μl. All reac- tions were amplified on a BioRad MyIQ thermocycler. Additional material An intron-exon junction of the BTF3 gene was chosen as a reference location for all qPCR experiments. Analysis Additional file 1: Supplemental Material file. Table S1. Numbers of species, breeds, animals and trios used to call CNVs genotyped by of resultant crossing thresholds (Ct) was performed using BovineSNP50 assay. Table S2. Btau_4.0 cattle CNV regions and their the ΔΔCt method [48,49]. Calibration ΔCt values were frequencies. Table S3. Comparison of CNV regions identified on two derived from amplification of reference and test primers cattle genome assemblies. Table S4. UMD3 cattle CNV regions and their frequencies. Table S5. The comparison of CNVs from 39 trios using three on a genomic DNA template derived from the European CNV calling algorithms: individual-calling, posterior-calling and joint- Hereford, Dominette 01449. Since all reference and test calling. Table S7. Outgroup CNV regions and their frequencies. Table S8. primers did not overlap with any of Dominette’sCNV The effects of CNV length and frequency on calling consistances between CNV callings based on SNP array and aCGH. Table S9. qPCR regions, two-copy states were assumed for both ampli- Summary. Table S10. Over/Underrepresentation of PANTHER molecular cons in Dominette. function, biological process and pathway terms. Table S12. CNVR frequency differences among breeds. Figure S1. Illustration of a typical CNV call with qPCR validation. Cattle CNV distribution and association with Segmental Additional file 2: Supplemental Material file. Table S6. Gene contents Duplications and other features of cattle CNV regions. We investigated the genomic distribution of 682 CNVRs Additional file 3: Supplemental Material file. Table S11. Cattle CNV by testing the hypothesis that pericentromeric and sub- regions overlap with genomic regions under positive selection, human telomeric regions were enriched for CNVs as described orthologous OMIM genes and cattle QTLs. previously [32]. Briefly, all predicted variable bases that overlapped with these regions were totaled and chi- square tests were used to test the null hypothesis of no Acknowledgements enrichment. CNVRs were also overlapped with SD and We thank members of the Bovine HapMap Consortium for sharing their samples. We thank R. Anderson, J. Shaffer, D. Hebert, S. Schroeder and L. the other genomic features such as evolutionary break- Shade for technical assistance. This work was supported by in part by NRI point regions (ERBs), which were obtained from litera- grant No. 2007-35205-17869 from USDA CSREES (now NIFA), Project 1265- ture and public databases listed in web site references. 31000-098-00 from USDA-ARS and Research Grant Award No. IS-4201-09 from BARD, United States - Israel Binational Agricultural Research and Development Fund. Gene content Gene content of cattle CNV regions was assessed using Author details Ensembl genes ftp://ftp.ensembl.org/pub/current_fasta/ Bovine Functional Genomics Laboratory, ANRI, USDA-ARS, Beltsville, Maryland 20705, USA. Department of Animal and Avian Sciences, University bos_taurus/pep/, the Glean consensus gene set, cattle Hou et al. BMC Genomics 2011, 12:127 Page 10 of 11 http://www.biomedcentral.com/1471-2164/12/127 of Maryland, College Park, Maryland 20742, USA. Department of Genetics 19. Colella S, Yau C, Taylor JM, Mirza G, Butler H, Clouston P, et al: QuantiSNP: and Microbiology, University of Bari, Bari 70126, Italy. Center for Applied an Objective Bayes Hidden-Markov Model to detect and accurately map Genomics, Children’s Hospital of Philadelphia, Philadelphia, PA 19104, USA. copy number variation using SNP genotyping data. Nucleic Acids Res Bioinformatics and Computational Biology, George Mason University, 2007, 35:2013-2025. Manassas, VA 20110, USA. Animal Improvement Programs Laboratory, ANRI, 20. Wang K, Li M, Hadley D, Liu R, Glessner J, Grant SF, et al: PennCNV: an USDA-ARS, Beltsville, Maryland 20705, USA. integrated hidden Markov model designed for high-resolution copy number variation detection in whole-genome SNP genotyping data. Authors’ contributions Genome Res 2007, 17:1665-1674. YH, KW, EK, LKM, JS carried out computational analysis. DMB, MFC, MV 21. Korn JM, Kuruvilla FG, McCarroll SA, Wysoker A, Nemesh J, Cawley S, et al: carried out the experimental validations. PMV, TSS and CPVT participated in Integrated genotype calling and association analysis of SNPs, common data collection. GEL conceived of the study and led in its design and copy number polymorphisms and rare CNVs. Nat Genet 2008, coordination. All authors contributed to writing the manuscript, read and 40:1253-1260. approved the final manuscript. 22. Yavas G, Koyuturk M, Ozsoyoglu M, Gould MP, LaFramboise T: An optimization framework for unsupervised identification of rare copy Received: 10 September 2010 Accepted: 23 February 2011 number variation from SNP array data. Genome Biol 2009, 10:R119. Published: 23 February 2011 23. Winchester L, Yau C, Ragoussis J: Comparing CNV detection methods for SNP arrays. Brief Funct Genomic Proteomic 2009, 8:353-366. 24. LaFramboise T: Single nucleotide polymorphism arrays: a decade of References biological, computational and technological advances. Nucleic Acids Res 1. The Bovine Genome Sequencing and Analysis Consortium: The Genome 2009, 37:4181-4193. Sequence of Taurine Cattle: a window to ruminant biology and 25. Marioni JC, Thorne NP, Valsesia A, Fitzgerald T, Redon R, Fiegler H, et al: evolution. Science 2009, 324:522-528. Breaking the waves: improved detection of copy number variation from 2. Zimin AV, Delcher AL, Florea L, Kelley DR, Schatz MC, Puiu D, et al: A microarray-based comparative genomic hybridization. Genome Biol 2007, whole-genome assembly of the domestic cow, Bos taurus. Genome Biol 8:R228. 2009, 10:R42. 26. Diskin SJ, Li M, Hou C, Yang S, Glessner J, Hakonarson H, et al: Adjustment 3. The Bovine HapMap Consortium: Genome wide survey of SNP variation of genomic waves in signal intensities from whole-genome SNP uncovers the genetic structure of cattle breeds. Science 2009, genotyping platforms. Nucleic Acids Res 2008, 36:e126. 324:528-532. 27. Wang K, Chen Z, Tadesse MG, Glessner J, Grant SF, Hakonarson H, et al: 4. Van Tassell CP, Smith TP, Matukumalli LK, Taylor JF, Schnabel RD, Lawley CT, Modeling genetic inheritance of copy number variations. Nucleic Acids et al: SNP discovery and allele frequency estimation by deep sequencing Res 2008, 36:e138. of reduced representation libraries. Nat Methods 2008, 5:247-252. 28. Liu GE, Van Tassell CP, Sonstegard TS, Li RW, Alexander LJ, Keele JW, et al: 5. Matukumalli LK, Lawley CT, Schnabel RD, Taylor JF, Allan MF, Heaton MP, Detection of germline and somatic copy number variations in cattle. et al: Development and characterization of a high density SNP Developments in Biologicals 2008, 132:231-237. genotyping assay for cattle. PLoS ONE 2009, 4:e5350. 29. Liu GE, Hou Y, Zhu B, Cardone MF, Jiang L, Cellamare A, et al: Analysis of 6. VanRaden PM, Van Tassell CP, Wiggans GR, Sonstegard TS, Schnabel RD, copy number variations among diverse cattle breeds. Genome Res 2010, Taylor JF, et al: Invited review: reliability of genomic predictions for North 20:693-703. American Holstein bulls. J Dairy Sci 2009, 92:16-24. 30. Fadista J, Thomsen B, Holm LE, Bendixen C: Copy number variation in the 7. Hayes BJ, Bowman PJ, Chamberlain AJ, Goddard ME: Invited review: bovine genome. BMC Genomics 2010, 11:284. Genomic selection in dairy cattle: progress and challenges. J Dairy Sci 31. Bae JS, Cheong HS, Kim LH, NamGung S, Park TJ, Chun JY, et al: 2009, 92:433-443. Identification of copy number variations and common deletion 8. Wiggans GR, Sonstegard TS, VanRaden PM, Matukumalli LK, Schnabel RD, polymorphisms in cattle. BMC Genomics 2010, 11:232. Taylor JF, et al: Selection of single-nucleotide polymorphisms and quality 32. Liu GE, Ventura M, Cellamare A, Chen L, Cheng Z, Zhu B, et al: Analysis of of genotypes used in genomic evaluation of dairy cattle in the United recent segmental duplications in the bovine genome. BMC Genomics States and Canada. J Dairy Sci 2009, 92:3431-3436. 2009, 10:571. 9. Cole JB, VanRaden PM, O’Connell JR, Van Tassell CP, Sonstegard TS, 33. Marioni JC, White M, Tavare S, Lynch AG: Hidden copy number variation Schnabel RD, et al: Distribution and location of genetic effects for dairy in the HapMap population. Proc Natl Acad Sci USA 2008, 105:10067-10072. traits. J Dairy Sci 2009, 92:2931-2946. 34. Sonstegard TS, Barendse W, Bennett GL, Brockmann GA, Davis S, 10. Flori L, Fritz S, Jaffrezic F, Boussaha M, Gut I, Heath S, et al: The genome Droegemuller C, et al: Consensus and comprehensive linkage maps of response to artificial selection: a case study in dairy cattle. PLoS ONE the bovine sex chromosomes. Anim Genet 2001, 32:115-117. 2009, 4:e6595. 35. Rhead B, Karolchik D, Kuhn RM, Hinrichs AS, Zweig AS, Fujita PA, et al: The 11. Scherer SW, Lee C, Birney E, Altshuler DM, Eichler EE, Carter NP, et al: UCSC Genome Browser database: update 2010. Nucleic Acids Res 2010, 38: Challenges and standards in integrating surveys of structural variation. D613-D619. Nat Genet 2007, 39:S7-15. 36. Troy CS, Machugh DE, Bailey JF, Magee DA, Loftus RT, Cunningham P, et al: 12. Conrad DF, Pinto D, Redon R, Feuk L, Gokcumen O, Zhang Y, et al: Origins Genetic evidence for Near-Eastern origins of European cattle. Nature and functional impact of copy number variation in the human genome. 2001, 410:1088-1091. Nature 2009, 464:704-712. 37. Caramelli D: The Origins of Domesticated Cattle. Human Evolution 2006, 13. Redon R, Ishikawa S, Fitch KR, Feuk L, Perry GH, Andrews TD, et al: Global 21:107-122. variation in copy number in the human genome. Nature 2006, 38. Eichler EE: Widening the spectrum of human genetic variation. Nat Genet 444:444-454. 2006, 38:9-11. 14. Feuk L, Carson AR, Scherer SW: Structural variation in the human 39. Matsuzaki H, Wang PH, Hu J, Rava R, Fu GK: High resolution discovery and genome. Nat Rev Genet 2006, 7:85-97. confirmation of copy number variants in 90 Yoruba Nigerians. Genome 15. McCarroll SA, Altshuler DM: Copy-number variation and association Biol 2009, 10:R125. studies of human disease. Nat Genet 2007, 39:S37-S42. 40. Larkin DM, Pape G, Donthu R, Auvil L, Welge M, Lewin HA: Breakpoint 16. Marques-Bonet T, Girirajan S, Eichler EE: The origins and impact of primate regions and homologous synteny blocks in chromosomes have different segmental duplications. Trends Genet 2009, 25:443-454. evolutionary histories. Genome Res 2009, 19:770-777. 17. Zhang F, Gu W, Hurles ME, Lupski JR: Copy number variation in human 41. Nei M, Rooney AP: Concerted and birth-and-death evolution of health, disease, and evolution. Annu Rev Genomics Hum Genet 2009, multigene families. Annu Rev Genet 2005, 39:121-152. 10:451-481. 42. Liu GE, Li RW, Sonstegard TS, Matukumalli LK, Silva MV, Van Tassell CP: 18. Peiffer DA, Le JM, Steemers FJ, Chang W, Jenniges T, Garcia F, et al: High- Characterization of a novel microdeletion polymorphism on BTA5 in resolution genomic profiling of chromosomal aberrations using Infinium cattle. Anim Genet 2008, 39:655-658. whole-genome genotyping. Genome Res 2006, 16:1136-1148. Hou et al. BMC Genomics 2011, 12:127 Page 11 of 11 http://www.biomedcentral.com/1471-2164/12/127 43. Seroussi E, Glick G, Shirak A, Yakobson E, Weller JI, Ezra E, et al: Analysis of copy loss and gain variations in Holstein cattle autosomes using BeadChip SNPs. BMC Genomics 2010, 11:673. 44. Fontanesi L, Beretti F, Martelli PL, Colombo M, Dall’olio S, Occidente M, et al: A first comparative map of copy number variations in the sheep genome. Genomics 2010. 45. Fontanesi L, Martelli PL, Beretti F, Riggio V, Dall’olio S, Colombo M, et al: An initial comparative map of copy number variations in the goat (Capra hircus) genome. BMC Genomics 2010, 11:639. 46. Ramayo-Caldas Y, Castello A, Pena RN, Alves E, Mercade A, Souza CA, et al: Copy number variation in the porcine genome inferred from a 60 k SNP BeadChip. BMC Genomics 2010, 11:593. 47. Decker JE, Pires JC, Conant GC, McKay SD, Heaton MP, Chen K, et al: Resolving the evolution of extant and extinct ruminants with high- throughput phylogenomics. Proc Natl Acad Sci USA 2009, 106:18644-18649. 48. Livak KJ, Schmittgen TD: Analysis of relative gene expression data using real-time quantitative PCR and the 2(-Delta Delta C(T)) Method. Methods 2001, 25:402-408. 49. Graubert TA, Cahan P, Edwin D, Selzer RR, Richmond TA, Eis PS, et al: A high-resolution map of segmental DNA copy number variation in the mouse genome. PLoS Genet 2007, 3:e3. doi:10.1186/1471-2164-12-127 Cite this article as: Hou et al.: Genomic characteristics of cattle copy number variations. BMC Genomics 2011 12:127. Submit your next manuscript to BioMed Central and take full advantage of: • Convenient online submission • Thorough peer review • No space constraints or color figure charges • Immediate publication on acceptance • Inclusion in PubMed, CAS, Scopus and Google Scholar • Research which is freely available for redistribution Submit your manuscript at www.biomedcentral.com/submit http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png BMC Genomics Springer Journals

Genomic characteristics of cattle copy number variations

Download PDF
11 pages

Loading next page...
 
/lp/springer-journals/genomic-characteristics-of-cattle-copy-number-variations-KVzUT2gLfn

References (104)

Publisher
Springer Journals
Copyright
Copyright © 2011 by Hou et al; licensee BioMed Central Ltd.
Subject
Life Sciences; Life Sciences, general; Microarrays; Proteomics; Animal Genetics and Genomics; Microbial Genetics and Genomics; Plant Genetics & Genomics
eISSN
1471-2164
DOI
10.1186/1471-2164-12-127
pmid
21345189
Publisher site
See Article on Publisher Site

Abstract

Background: Copy number variation (CNV) represents another important source of genetic variation complementary to single nucleotide polymorphism (SNP). High-density SNP array data have been routinely used to detect human CNVs, many of which have significant functional effects on gene expression and human diseases. In the dairy industry, a large quantity of SNP genotyping results are becoming available and can be used for CNV discovery to understand and accelerate genetic improvement for complex traits. Results: We performed a systematic analysis of CNV using the Bovine HapMap SNP genotyping data, including 539 animals of 21 modern cattle breeds and 6 outgroups. After correcting genomic waves and considering the pedigree information, we identified 682 candidate CNV regions, which represent 139.8 megabases (~4.60%) of the genome. Selected CNVs were further experimentally validated and we found that copy number “gain” CNVs were predominantly clustered in tandem rather than existing as interspersed duplications. Many CNV regions (~56%) overlap with cattle genes (1,263), which are significantly enriched for immunity, lactation, reproduction and rumination. The overlap of this new dataset and other published CNV studies was less than 40%; however, our discovery of large, high frequency (> 5% of animals surveyed) CNV regions showed 90% agreement with other studies. These results highlight the differences and commonalities between technical platforms. Conclusions: We present a comprehensive genomic analysis of cattle CNVs derived from SNP data which will be a valuable genomic variation resource. Combined with SNP detection assays, gene-containing CNV regions may help identify genes undergoing artificial selection in domesticated animals. Background genotyping assays, QTL distributions and artificial selec- With two cattle genome assemblies available (Btau_4 tion signatures in dairy cattle have been reported [9,10]. and UMD3) [1,2], the cattle research community has Copy Number Variation (CNV) represents another been focusing on single nucleotide polymorphisms important source of genetic variation that provides (SNPs) as the main source of genetic variation in cattle. genomic structural information complementary to SNP This effort led to the development of the cattle SNP data. Genomic structural variations ranging from 1 kb map [3] and the Illumina Bovine SNP50 (> 50,000 SNP to 5 Mb comprise mainly of CNVs in the form of large- probes) genotyping array [4,5]. Evaluations of genetic scale insertions and deletions, as well as inversions and merit based on SNPs became a reality in early 2009 translocations [11]. In humans, ~29,000 CNVs that cor- leading to an acceleration of improvements to dairy and respond to over 8,400 CNV regions have been identified, beef breed stocks [6-8]. Widespread use of the Bovi- and over 9,000 genes have been mapped within or near neSNP50 array has resulted in the availability of tens of regions of human structural variation [12,13]. Some of thousands of SNP genotyping results. Based on SNP these CNVs have been shown to be important in both normal phenotypic variability and disease susceptibility. * Correspondence: [email protected] Several recent publications have reviewed the effects of Bovine Functional Genomics Laboratory, ANRI, USDA-ARS, Beltsville, CNVs on gene expression and human diseases [14-17]. Maryland 20705, USA Due to their low cost and high-density, SNP arrays have Full list of author information is available at the end of the article © 2011 Hou 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. Hou et al. BMC Genomics 2011, 12:127 Page 2 of 11 http://www.biomedcentral.com/1471-2164/12/127 been routinely used for human CNV detection and ana- Results and Discussion lysis [13]. Compared to CGH arrays which only report Optimization of cattle CNV detection relative signal intensities, SNP arrays collect normalized A total of 58,336 markers were selected for the Bovi- total intensities (Log R ratio - LRR) and allelic intensity neSNP50 assay [4,5]. Except for 1,389 markers which ratios (B allele frequency - BAF) which represent overall failed to pass manufacturer assay production pipeline, we copy numbers and allelic contrasts [18]. Multiple algo- intentionally kept all remaining 56,947 markers without rithms have been developed to exploit SNP data to iden- any other filtering. These included 1,465 markers (2.57%) tify CNVs, including QuantiSNP [19], PennCNV [20], which had a call rate of 0. The markers with a call rate of Birdseye [21] and Cokgen [22]. Comparisons of the 0 are resistant to the default biallelic SNP clustering and strengths and weaknesses of these algorithms have been often fall in CNV regions. Compared to the standard published [23,24]. As one of the leading methods, BovineSNP50 Genotyping Beadchip v1 featuring 54,001 PennCNV incorporates multiple sources of information, SNP probes, 2,946 more SNPs were included in our ana- including total signal intensity and allelic intensity ratio lysis, of which, ~17% located in cattle segmental duplica- at each SNP marker, the distance between neighboring tion (SD) regions [32], ~9% overlapped with the CNVRs SNPs, and the allele frequency of SNPs. PennCNV also detected by array CGH method [29], and ~27% contribu- integrates a computational approach by fitting regres- ted to the CNVRs reported here. sion models with GC content to overcome “genomic We tested the cattle CNV calls made with or without waves” [25,26]. Furthermore, PennCNV is capable of the -gcmodel option on Batu_4.0 to identify the impact considering pedigree information (a parents-offspring of genomic waves on CNV calling. Agreeing with pre- trio) to improve call rates and accuracy of breakpoint viousresults [26],wefound thetotal CNVR counts prediction as well as to infer chromosome-specific SNP were higher without -gcmodel (719) than those with genotypes in CNVs [27]. -gcmodel enabled (682). However, only 86.80% (592/ Previous cattle studies have produced a number of CNV 682) of the gcmodel calls directly overlapped with datasets. For example, our earlier array CGH survey using 79.28% (570/719) calls made without gcmodel, revealing 3 Holstein bulls identified 25 germline CNVs [28]. a ~20% CNV discordancy rate. These discordant calls Recently, we reported a broader, systematic CNV survey were likely due to false positives called from the differ- in 90 cattle using array CGH [29]. We identified over 200 entiating signal intensities caused by “genomic waves” candidate CNV regions (CNVRs); some of which are likely rather than by real CNV events. This further demon- to underlie cattle domestication and breed formation. strated that genomic waves have a significant effect on Fadista et al. recently reported 304 CNV regions in 20 ani- this type of analysis. mals of 4 cattle breeds using high-density array CGH [30]. We also compared results of PennCNV using -test, Besides array CGH experiments, other evidences for cattle -trio and -joint options sequentially. In other words, we CNV came from SNP genotyping results, where a screen compared data resultant from not considering trio infor- of Bovine HapMap Consortium samples (over 500 animals mation (-test), considering trio information only after from multiple cattle breeds) identified 79 candidate dele- calling (-trio) and finally by considering trio information tions using an earlier version of cnvPartition [5]. However, in a simultaneous fashion during CNV calling (-joint) these results only included homozygous deletions which (Additional file 1: Table S5). Consistent with the earlier were validated by multiple observations. A recent paper comparisons using simulated and real SNP data [27,33], reported 368 unique CNV regions from 265 Korean Han- trio information significantly increased our CNV call woo cattle based on BovineSNP50 genotyping data; how- rates. The result of the -joint option (1276 calls) was sig- ever, during the PennCNV calling, the “genomic waves” nificantly higher than those of the other options: -test pattern was not discussed and pedigree information was (684 calls) and -trio (1019 calls). After merging overlap- not considered [31]. In this study, we reprocessed the pub- ping CNVs, ~87% of the 682 CNVRs deduced from the lished Bovine HapMap Consortium SNP genotyping -joint option overlapped with those deduced from the results using optimal settings for PennCNV by adjusting -test and -trio options (both with a total of 621 CNVRs). for “genomic waves” and utilizing trio/pedigree informa- Due to its improved call sensitivity and breakpoint infer- tion whenever possible. We identified 682 candidate CNV ence, the -joint option reported about 13% more CNVRs regions in a diverse panel of 521 animals from 21 different which were not detected by the -test or -trio options. breeds. We also included 18 animals from 6 outgroups to derive the ancestral states of CNVs. We then compared Cattle CNV discovery and distribution this CNV call set with the existing cattle CNV call sets, Due to issues regarding CNVR calls, we excluded chrX validated several novel CNVR calls and discussed the evo- and chrUn from our analysis. In our initial analysis of lutionary impact of cattle CNVs. chrX, it was found that almost half of the potential Hou et al. BMC Genomics 2011, 12:127 Page 3 of 11 http://www.biomedcentral.com/1471-2164/12/127 CNVRs were unreasonably large (> 1 Mb) and several were based on the same raw WGS reads. The most events were present in high frequencies (> 25%). This is obvious difference between the two assemblies is that likely due to the fact that PennCNV assumes two copies Btau_4.0 unplaced contigs are placed on UMD3. This of each SNP as the normal copy number state, which resulted in more markers that were on Btau_4.0 ChrUn was likely not the case within the pseudoautosomal contigs to be placed on UMD3 autosomes, which could region [34] and segmental duplications [32] on chrX. partially explain the increase in the CNVR counts. Since Additionally, since chrX sequence and annotation also themajorityofcattlegenomeannotations were per- formed on the Btau_4.0 assembly, we focused on further differ dramatically between Btau_4.0 and UMD3 builds, characterization of the 682 high-confidence CNV we considered the CNV calls on chrX as unreliable and excluded them from further analysis. Since chrUn only regions from Btau_4.0 autosomes. contains unassigned sequence contigs, it was not These 682 CNVRs include 370 loss, 216 gain and 96 included due to the lack of sequence and SNPs as well both (loss and gain within the same region) events, ran- as the SNP mapping uncertainty. ging from 32,566 to 5,569,091 bp with a mean or med- Within the placed autosomes, a total of 3,666 CNVs in ian of 204,965 or 131,179 bp, respectively (Additional 521 samples were detected and an average of 7.09 gain file 1: Table S2). Loss events are approximately 1.7-fold or loss events were evident in each sample (Table 1). more common than gain events, but have slightly smal- CNVRs were determined by aggregating overlapping ler sizes than gain regions on average. Furthermore, 278 CNVs identified across all samples, following previously CNVRswerefound in only onesample(Unique), 404 published protocols [13]. A total of 682 high-confidence CNVRs were present in two or more animals or breeds autosomal CNVRs were identified, covering 139.8 Mb of and 18 of 404 multiple events had a frequency >5% polymorphic sequence and corresponding to 5.49% of (Table 1 and Additional file 1: Table S2). These datasets the autosomal genome sequence (139.8/2,545.9 Mb) and confirm that segregating CNVs exist among these 21 4.60% of the whole cattle genome (139.8/3,036.6 Mb, cattle breeds and/or groups, which is consistent with Figure 1 and Additional file 1: Table S2). our earlier results based on array CGH [29]. In general, To test the stability of CNV calls with respect to dif- the number of CNVs identified in each sample is con- ferent genome builds and SNP mapping, we also sistent with SNP estimates of breed-specific founding migrated 56,408 out of 56,947 SNP markers from and effective population sizes and levels of polymorph- Btau_4.0 to UMD3 using the UCSC liftOver tool [35], ism based on ≥50,000 SNPs [5]. As shown in Table 1, and repeated the entire calling analyses to ensure con- more CNV events were detected in indicine (11.41 per sample) than in African groups (7.21 per sample) and sistency in calls (Additional file 1: Tables S3 and S4). Only 61 more CNVRs were identified on the UMD3 composite (7.17 per sample). The taurine breeds (6.23 assembly (making a total of 743 CNVRs). A simple com- per sample) had the fewest detected CNVs. While some parison indicated that the total coverage of variable of these differences could be related to the fact that the regions were 13.05% larger on UMD3 (158.0 Mb, SNP markers were designed based on the Btau_4.0 Additional file 1: Table S3) than on Btau_4.0. For all reference genome (which was derived from the three CNVR types (gain, loss and both), counts sequence of a Hereford cow of European origin; Domin- increased slightly. This was expected as both assemblies ette 01449), this observation is consistent with the Table 1 CNV events by species and breeds Btau_4.0 Sample Count Unique Gain Loss Gene Total Length Taurine 366 2,256(6.23) 239(0.66) 1,454(4.02) 802(2.22) 4,744(13.10) 373,001,599(165,337) Composite 46 330(7.17) 23(0.50) 224(4.87) 106(2.30) 651(14.15) 113,483,966(142,032) Indicine 70 799(11.41) 62(0.89) 401(5.73) 398(5.69) 1,464(20.91) 57,402,891(173,948) African Breeds 39 281(7.21) 38(0.97) 213(5.46) 68(1.74) 775(19.87) 54,728,022(194,761) CNV 521 3,666(7.09) 362(0.70) 2,292(4.43) 1,374(2.66) 7,634(14.77) 598,616,478(163,288) b c d d CNVR 521 682 278 216 370 1,263 139,786,166(204,965) Outgroup CNV 18 1,003(55.72) 284(15.78) 48(2.67) 955(53.06) 2,603(144.61) 442,235,607(440,912) CNVR 18 483 187 21 458 1,593 276,846,573(573,181) The numbers in parentheses are normalized by sample counts except that the lengths in parentheses are average lengths normalized by CNV counts. At sample level, each sample has 7.09 (3666/517) CNVRs averagely and 6.23 (2256/362) for Taurine, since there are 4 taurine individuals without identified CNVs; These numbers are nonredundent CNVR counts. 278 CNVRs are unique to one sample while 404 CNVRs are shared by at least 2 individuals or breeds and 18 of 404 d e multiple events have frequency >5%; Besides 370 loss and 216 gain CNVRs, there are 96 CNVRs containing both loss and gain events; Outgroup animals are not included in the total counts for CNV and CNVR. Hou et al. BMC Genomics 2011, 12:127 Page 4 of 11 http://www.biomedcentral.com/1471-2164/12/127 Figure 1 Genomic landscape of cattle copy number variations and segmental duplications. CNV regions (682 events, 139 Mb, ~4.60% of the bovine genome) reported by 521 SNP genotyped individuals are shown above the chromosomes in green (gain), red (loss) and dark blue (both), while below are the CNV regions (177 events, 28 Mb, ~1% of the bovine genome) reported by 90 array CGH experiments by Liu et al. The bar height represents their frequencies: short (appeared in 1 sample), median (≥2 samples) and tall (≥5 samples). Segmental duplications (94.4 Mb, 3.1% of the bovine genome) predicted by two independent computational approaches are illustrated on the chromosomes in red (WSSD), blue (WGAC) or purple (both). The patterns are depicted for all duplications for ≥5 kb in length and ≥90% sequence identity. The gaps in the assembly are represented on the chromosomes as white ticks. concept of subspecies divergence and supports the chrX and chrUn, pericentromeric and subtelomeric hypothesis of multiple independent domestications of regions each represent 3.42% of genomic sequence but cattle in the Fertile Crescent, Southwest Asia and likely show an enrichment of 1.5-2.4-fold more CNVRs (both Africa [36,37]. P values <0.001) and contain 7.78-12.54% of all poly- Cattle CNVs are distributed in a nonrandom fashion morphic sequence. at two different levels. First, CNV content varies signifi- cantly among different chromosomes. The proportion of Quality assessment of selected CNV Regions any given known chromosome susceptible to CNV The quality of our 682 CNV calls was assessed in mul- regions varies from 1.32-8.80% (Additional file 1: Table tiple ways, though our first assessment was a compari- S2). Chromosomes 1 and 6 show the greatest enrich- son against existing cattle CNV datasets (Table 2 and ment for CNV (Figure 1 and Additional file 1: Table S2) Figure 2). One of the earlier datasets included 79 fil- with almost two-fold of the variable content of the auto- tered deletion variants (representing 42 unique geno- somal average. It is interesting to note that these chro- mic loci and 9 single SNPs) reported earlier using the mosomes do not have the highest SD content [29,32]. Illumina genotyping software module cnvPartition Furthermore, similar to the human, mouse, rat and dog v1.0.2 [5]. Nineteen of our CNVRs overlapped with 11 genomes, there are a greater proportion of CNVs near of the deletion variants (21.57%) in that dataset (Table pericentromeric and subtelomeric regions. Excluding 2). We also identified 129 CNVRs (18.91%) in our Hou et al. BMC Genomics 2011, 12:127 Page 5 of 11 http://www.biomedcentral.com/1471-2164/12/127 Table 2 Summary of genome-wide studies of cattle copy number variations Study Assay Count CNVR Size Marker Sample Breed Type Count Range (kb) Median Mean Total (kb) (kb) (Mb) Matukumalli BovineSNP50 54,001 556 21 Deletion only 51 22.92- 394.87 960.67 49.0 et al. 2009 11,050.69 Liu et al. 2010 Array CGH ~385,000 90 17 Deletion, 163 18.00-1,261.90 86.19 153.75 25.1 insertion Bae et al. 2010 BovineSNP50 54,001 265 1 Deletion, 368 25.35-967.18 128.33 171.49 63.1 insertion Fadista et al. 2010 Array CGH ~6,300,000 20 4 Deletion, 254 1.72-2,031.34 15.51 62.26 15.8 insertion This study BovineSNP50 56,947 521 21 Deletion, 682 32.57-5,569.09 131.18 204.97 139.8 insertion a b This includes 9 independent SNPs and 42 CNVRs. The statistics are calculated for 42 CNVR excluding the 9 SNPs; This is the number excluding chrX and chrUn; This is the number excluding chrX, chrUn and mitochondrial sequence. dataset that overlapped with 128 CNVRs from a SNP- an arrayCGH-based studyof20cattlefromfour based CNV study on 265 Korean Hanwoo cattle [31] breeds [30]. Since our dataset excluded CNV calls in (Figure 2B). The Hanwoo CNV study identified 368 the chrX, chrUn and mitochondrial sequences, we CNVRs in total, so our dataset overlapped with 34.83% compared our autosomal CNVR calls (682 CNVRs) to of their calls [31]. We then compared our calls against the autosomal CNV calls of that study (254 CNVRs) Bae et al (2010) This study Merged This study dataset ~482 ~200 ~518 554 46.6 Mb 123.3 Mb 16.5 Mb 66.5 Mb 100 Mb 29.8 Mb A B Count Length This study 682 139.8 Mb This study ~600 126 Mb Merged dataset 718 99.3 Mb ~24 ~30 2.4 Mb 7.1 Mb 28 368 63.1 Mb Bae et al (2010) 4.3 Mb ~86 ~183 Fadista et al (2010) 254 15.8 Mb 7.1 Mb 11.7 Mb ~19 Fadista et al (2010) Liu et al (2010) 2 Mb 163 25.1 Mb Liu et al (2010) Figure 2 Comparisons between identified 682 CNVRs in this study and the other existing cattle CNVR datasets in terms of count and length. A, compared to the total nonredundant CNVR merged from existing published datasets; B, compared to CNVR derived from SNP array (Bae et al, 2010); C, compared to two CNVR datasets derived from array CGH studies (Liu et al,2010; Fadista et al, 2010); D, the summaries and legends of existing cattle CNVR datasets. Hou et al. BMC Genomics 2011, 12:127 Page 6 of 11 http://www.biomedcentral.com/1471-2164/12/127 [30].Only51ofour CNVRs(7.48%)directlyover- (50 animals), the overlap included 100% of our filtered lapped with 55 of their calls (21.65%, Figure 2C). Our dataset. This further demonstrated that large, common final comparison was against our previous array CGH- CNVRs can be reliably detected through using different based study of 90 animals from 14 breeds which detection technologies even when the majority of sam- resulted in 163 autosomal CNVR calls [29]. In this ples were different. For example, our current SNP array comparison, 57 of our SNP-based CNVR calls (8.36%) study identified most of the large, common CNVRs which were confirmed in our published results [29]. overlapped with 59 CNVRs derived from array CGH After comparison with other existing datasets, we (36.20%,Figure2C).Ifweonlyfocused onthe16Hap- foundthat~70% of our CNVRcalls were notreported Map samples which were assessed by both platforms (60 CNVRs derived from array CGH and 106 CNVRs in the literature. In order to confirm these novel reported by SNP array), there were 21 overlapping CNVRs, we performed 24 quantitative PCR (qPCR) CNVRs: 19 for array CGH (31.67%, 19/60), and 20 assays for 15 low frequency, novel CNVR calls spread for SNP array (18.87%, 20/106). When we merged among seven individuals (Additional file 1: Table S9). existing CNV datasets, a total of ~ 200 out of 682 Nine of the CNV regions had two target amplicons (about 30%) newly identified CNVRs overlapped with placed near two different SNP loci. Out of 24 total loca- them (Figure 2A). tions, 11 loci (~48%) were in agreement with CNV esti- It is expected that the variants identified in these stu- mates by PennCNV (Additional file 1: Table S9 and dies do not overlap, suggesting a vast amount of CNVs Figure S1). When counting the CNV regions, 9 out of exist in cattle population and saturation for this type of 15 (60%) CNV regions had positive qPCR confirmations variation has not yet been approached. It is likely that in at least one location. If CNVRs previously validated many thousands of more common structural variants in the literature [29] were also included, approximately may still remain undiscovered in the cattle genome. two third (30% + 70% × 48%) of our detected CNVRs A similar situation was encountered in human CNV stu- had positive confirmations. dies using the early version of SNP, CGH arrays and As expected, the Bovine SNP50 platform has a large detection methods [38,39]. For example, although resolution limit under the current PennCNV calling cri- cnvPartition detects CNVs by processing the similar raw teria. The size of the CNVRs detected ranged from 32.6 data as PennCNV (i.e. LRR and BAF), it is based on a kb to 5.6 Mb, with a median size of 131.2 kb (Additional different proprietary sliding window algorithm. Only file 1: Table S2). This is partially due to the fact that the Bovine SNP50 assay was originally developed for high- those homozygous deletion events segregating in differ- throughput SNP genotyping in association studies. ent animals were reported due to concerns with the quality of calls [5]. In the future, high-density SNP Although CNV detection is feasible with SNP arrays, it arrays combined with improved CNV calling algorithms is impaired by low density and non-uniform distribution could remedy these differences. of SNPs, especially in CNV and SD regions. Compared Besides the technology and detection method differ- to a CGH array, a SNP array lacks non-polymorphic ences, the following could also contribute to the probes designed specifically for CNV identification. observed differences: (1) sampling differences: 521 indi- Thus, only the large CNVRs are expected to be identi- viduals from 21 diverse breeds and/or groups were fied with the Bovine SNP50 assay. This explains the dif- included in our study; (2) genome coverage biases: ference in CNV length between our study and the 56,947 markers were included in our study rather than a earlier results. subset of “well-behaved” SNPs (54,001 markers) which The discrepancies between the qPCR and PennCNV exclude those SNPs in CNV-rich regions; (3) correction resultsmay representsmall CNVeventsthatwere of genomic waves in order to minimize false positive missed in the PennCNV calls, or instances where SNPs calls; and (4) trio/pedigree information was fully caused the qPCR reaction to fail or be suboptimal but explored in our study to improve the accuracy and call did not affect the SNP assay. Despite the fact that a rate of CNVs. When filtering criteria varying the CNVR two-copy state was assumed for test PCR loci in Domin- length and frequency were applied, we observed signifi- ette, smaller CNV events in Dominette may have evaded cant overlap within our 2 datasets derived from SNP detection by PennCNV. If our test primers amplified a arrays and array CGH (Additional file 1: Table S8). For small CNV event in Dominette, that would skew the example, when the large CNVRs (SNP count = 10, a relative copy number estimates of our qPCR reactions. Although qPCR primers were designed within 250 bp median ~574kb) were considered, the overlap reached around the target SNP positions, additional SNPs and 21.74%. When the CNVR frequency was increased to 1, small indels may have influenced the hybridization of 2, or 5% (animal count = 5, 10, or 25, respectively), the overlap increased to 89.47%. When we filtered the the qPCR primers in some animals, thereby reducing CNVR frequency to greater than 10% of our population primer efficiency. Other causes may also contribute to Hou et al. BMC Genomics 2011, 12:127 Page 7 of 11 http://www.biomedcentral.com/1471-2164/12/127 the discrepancy in CNVR validation by qPCR. The draft functions, and provides a rich resource for testing status of the cattle genome assembly and the low SNP hypotheses on the genetic basis of phenotypic variation densityoftheBovineSNPassaymakeitdifficult to within and among breeds. determine the real breakpoints of CNVRs. For example, Consistent with similar CNV analyses in other mam- multiple, neighboring, discrete CNV events could result mals (human, mouse and dog), several of these CNVs, in a larger call by PennCNV; therefore, giving an which are important in drug detoxification, defense/ over estimation of the CNV size. Therefore, it cannot innate and adaptive immunity and receptor and signal be ruled out that the qPCR primers used to confirm recognition, are also present in cattle. These gene families include olfactory receptors, ATP-binding cas- theCNVRs mayhavebeendesignedoutside the breakpoints. sette (ABC) transporters, Cytochrome P450, b-defensins, interleukins, the bovine MHC (BoLA) and multiple CNVs overlap with segmental duplications and other solute carrier family proteins which support the shared genomic features GO termsamong mammalsasshown in Additional file Following previous studies of other genomes, we 1: Table S10. For gene families that went through cattle- detected the association between CNVRs and SDs. specific gene duplication [32], such as interferon tau, Agreeing with previous predictions regarding cattle SDs pregnancy-associated glycoproteins, SCP2 and ULBP [32], a local tandem distribution pattern is predominant and WC1.1 subfamilies, we also detected marked varia- in our cattle CNVR dataset (Figure 1). It should be tion in copy number between individuals and across noted that about 25.66% (175/682) of CNV regions diverse cattle breeds and/or groups (Additional file 2: directly overlap with cattle SDs with an overlapping Table S6). It is intriguing to note that we also detected span of 16,283,071 bp (11.65% of the total 139,786,166 variations of TLR3 (toll-like receptor 3) and PPARA bp). Approximately 12.06% (356/2952) of the SDs (peroxisome proliferator-activated receptor alpha). This (excluding chrX and chrUn) identified by WGAC and current CNV survey further supports a hypothesis that WSSD [32] exhibit CNVs. In comparison, 58.90% of the the generation of new CNV insertions and deletions CNVRs (96/163) detected by using array CGH [29] may be a constant phenomenon in multiple cattle excluding X chromosome overlap with cattle SDs, corre- breeds/individuals [41]. sponding to 15,176,612 bp (60.56% of the total We also overlapped our CNVRs with two sets of 25,061,646 bp). The proportion of our new CNVR calls genomic regions under positive or balance selection (identified in this study) that overlap with SDs reaches detected by iHS and F using SNP data [3,10] (Addi- ST approximately 40% compared to 61% in our previous tional file 3: Table S11). By doing so, we have identified study. This lower overlap fraction probably reflects the CNV regions that may span potential cattle QTLs and fact that the BovineSNP50 array used in this study is human orthologous OMIM genes influencing disease biased against cattle SD regions. SNP density on the susceptibility (Additional file 3: Table S11). For instance, array drops by one-third (from 21 probes/Mb in unique multiple CNV regions directly overlap with QTLs for regions down to 14 probes/Mb) in SD regions. We also significant and typical economic traits and 87 out of 682 failed to detect any correlation between 682 CNV CNVRs correspond to loci known to cause disease in regions and evolutionary breakpoint regions (EBRs). humans. However, since the cattle genome and cattle Compared to the genomic averages, cattle-specific EBRs QTLs are less well defined, future study is warranted. and artiodactyl-specific EBRs do not show enrichments of CNV sequences [40]. This negative result is consent Cattle CNV frequency differences among breeds with the fact that EBRs have fewer overlaps with SD To highlight the potential evolutionary contributions of regions. these CNVs to cattle breed formation and adaptation, we queried 91 CNVRs that have breed-specific CNV fre- Gene Content of Cattle CNV regions quency differences (Additional file 1: Table S12). We Within autosomes, the 682 CNV regions overlap with only considered breeds that had at least 12 samples and 1,679 Ensembl peptides, corresponding to 1,263 unique any detectable variations must be identified in at least 3 Ensembl genes (Table 1 and Additional file 2: S6). individuals or 10% of samples (for Holstein, Angus and About 55.57% (379/682) of high-confidence CNVRs Limousin where n > 30). Fifty-eight of these CNVRs completely or partially span cattle Ensembl genes. We correspond to annotated genes or gene families; many assigned PANTHER accessions to a total of 1,263 over- of which were identified in other mammals as influen- lapping genes. Statistically significant over or under cing adaptation to the environment. Some of the anno- representations were observed for multiple categories tated genes are known to be important in cattle (Additional file 1: Table S10). This set of copy number adaptation including CNVR266(IFNT) in Brown Swiss, variable genes possess a wide spectrum of molecular CNVR122 (SCP2) in Hereford [32] and CNVR178 Hou et al. BMC Genomics 2011, 12:127 Page 8 of 11 http://www.biomedcentral.com/1471-2164/12/127 (Olfactory Receptors) in most of breeds [See also [42]]. results, including those collected for the Bovine Hap- The differences of CNV frequency among cattle breeds Map project [3] (Additional file 1: Table S1). PennCNV supported our earlier hypothesis that some cattle CNVs quality filters were applied after the CNV detection, are likely to arise independently in breeds, are likely to resulting in 521 distinct high quality genotyping results contribute to breed differences and are therefore related from the original 556 animals. This panel included 366 to the breed formation and adaptation. However, the animals from 14 taurine dairy and beef breeds, 70 ani- observed differences between breed variations could be mals from three breeds of predominantly indicine back- caused by both selection and genetic drift due to genetic ground, 46 animals from two breeds that are Taurine × bottlenecks for some breeds. Our findings, therefore, Indicine composites, and 39 animals from two African must be confirmed with an even larger sample size. groups, one of which (Sheko) is an ancient hybrid. It is worth to note that for many of the breeds, individuals CNVs in outgroup animals were sampled from more than one continent to repre- For the 18 individuals in outgroups, which were ana- sent the global cattle population. This panel contained lyzed similarly together with 521 modern breeds indivi- 39 trios where both parents and an offspring were geno- duals, 1003 CNVs and 483 CNVRs were detected, typed. Additionally, we included 18 animals from 6 out- covering 276.8 Mb base pairs, with 21 gain and signifi- groups (Bos gaurus -Gaur, Bos bison - North American cantly more (458) loss events (Table 1 and Additional Bison, Bubalus depressicornis- Lowland Anoa, Bos java- file 1: S7). About 34.60% of our current CNVRs (236/ nicus -Banteng, Bos grunniens -Yak,and Syncerus caf- 682) directly overlapped with 37.47% of these ancient fer -Cape Buffalo) with 1 trio information to derive the CNVRs (181/483) derived from ancient outgroups, ancestral states of CNVs. Among these outgroups, the which indicates over one third of CNVRs were likely average sample call rate was 89.91%, reflecting their ancestral. We suspected this observation of more loss divergent relationship from Bos Taurus. events than gain events was at least partially related to the high genetic divergences between these outgroup Identification of cattle CNVs animals and the cattle reference genome. With addi- PennCNV algorithm [20] was only applied to autosomes tional cattle, sheep, goat and pig CNV papers published (-lastchr 29) to detect cattle CNV in this study. In our recently [43-46], it will be interesting to look into the initial analysis, chrX (-chrx) was also considered sepa- evolutionary mechanism of CNVs within livestock rately from automosomes. PennCNV incorporates multi- animals. ple sources of information together, including LRR and BAF at each SNP marker, more realistic models for Conclusions state transition between different copy number states We have performed a comprehensive genomic analysis based on the distance between neighboring SNPs, popu- of cattle CNVs based on whole genome SNP genotyping lation frequency of B allele (PFB), the allele frequency of data, therefore providing a valuable genomic variation SNPs, and the pedigree information where available, resource. A total of 682 CNVRs were identified, cover- into a hidden Markov model (HMM). Both LRR and ing 139.8 megabases (~4.60%) of the genome. A subset BAF were exported from Illumina GenomeStudio Geno- of these CNVRs showed Mendelian inheritance and typing Module v1.0 software given the default clustering were also confirmed in other cattle CNV studies and file for each SNP. The PFB file was calculated based on other mammalian species. As high density cattle SNP the BAF of each marker in this population. Because genotyping data are becoming available, CNVs com- there were 153 out of 556 animals (~27.5%) with abso- binedwithSNPs, mayhelpidentify genes undergoing lute values of waviness factor larger than 0.04 in our ori- artificial selection in domesticated animals. ginal analysis, the genomic waves were adjusted using the -gcmodel option. The cattle gcmodel file was gener- Methods ated by calculating the GC content of the 1Mb genomic Selection of cattle breeds and animals region surrounding each marker (500kb each side). For It hasbeendemonstratedthatthe BovineSNP50geno- comparison, the analysis without considering gcmodel typing array provides a robust resource for genome- was also conducted. Three different PennCNV options wide, high-density SNP genotyping of cattle and for were performed wherever possible: 1) -test: the indivi- population genetic analyses of closely related artiodactyl dual-calling algorithm that treats family members as if species [4,47]. In which, less than 3% of markers had they were unrelated; 2) -trio: the posterior-calling algo- call rates below 99.94%, the average call rate for indivi- rithm which accommodates family information to dual samples was greater than 97.5% and 85% of sam- improve the accuracy of individual-based CNV calling ples had call rates above 98.8% [5]. Cattle CNVs in this and boundary prediction; 3) -joint: the joint-calling algorithmthatidentifiesCNVsusing family data study were detected by using the same SNP genotyping Hou et al. BMC Genomics 2011, 12:127 Page 9 of 11 http://www.biomedcentral.com/1471-2164/12/127 simultaneously. After CNV detection, filtering of low- RefSeq and in silico mapped human RefSeq (the UCSC quality samples was carried out with the default cutoffs: Genome Browser website at http://genome.ucsc.edu/). standard deviation (STD) of LRR as 0.30, BAF drift as We obtained a catalog of all bovine peptides from 0.01 and waviness factor as 0.05. The filtered results Ensembl. This yielded 26,271 peptides, 1,679 of which from the three algorithms were compared in terms of overlap with predicted 682 high-confidence CNV CNV numbers, lengths and number of SNP in CNVs regions, and correspond to 1,263 unique Ensembl genes. (Additional file 1: Table S5). The final CNVs set was the Using the PANTHER classification system, we tested the nonredundant combinationof CNVsfromthe -joint hypothesis that the PANTHER molecular function, bio- results for family trio members and the -test results for logical process and pathway terms were under- or over- unrelated individuals. For the outgroup animals, quality represented in CNV regions after Bonferroni corrections filtering was not performed due to their divergent rela- [32]. It is worth noting that a portion of the genes in tionship from Bos Taurus. CNVRs are determined by the bovine genome has not been annotated or has been aggregating overlapping CNVs identified across all sam- annotated with unknown function, which may influence ples [13]. the outcome of this analysis. CNV validation Web Site References array CGH experiments were performed as previously The Database of Genomic Variants: http://projects.tcag. described [11]. Primers were designed for qPCR valida- ca/variation/ tion using the Primer3 webtool http://frodo.wi.mit.edu/ Ensembl genes ftp://ftp.ensembl.org/pub/current_fasta/ primer3/ by limiting amplicon length to 150 bp to 250 bp bos_taurus/pep/ and by incorporating a GC clamp of 2. All other settings PANTHER http://www.pantherdb.org/ were left at the default. Primer information is shown in OMIM http://www.ncbi.nlm.nih.gov/omim/ Additional file 1: Table S9. Quantitative PCR experiments OMIA http://omia.angis.org.au/ were conducted using SYBR green chemistry in triplicate QTL http://www.animalgenome.org/ reactions, each with a reaction volume of 25 μl. All reac- tions were amplified on a BioRad MyIQ thermocycler. Additional material An intron-exon junction of the BTF3 gene was chosen as a reference location for all qPCR experiments. Analysis Additional file 1: Supplemental Material file. Table S1. Numbers of species, breeds, animals and trios used to call CNVs genotyped by of resultant crossing thresholds (Ct) was performed using BovineSNP50 assay. Table S2. Btau_4.0 cattle CNV regions and their the ΔΔCt method [48,49]. Calibration ΔCt values were frequencies. Table S3. Comparison of CNV regions identified on two derived from amplification of reference and test primers cattle genome assemblies. Table S4. UMD3 cattle CNV regions and their frequencies. Table S5. The comparison of CNVs from 39 trios using three on a genomic DNA template derived from the European CNV calling algorithms: individual-calling, posterior-calling and joint- Hereford, Dominette 01449. Since all reference and test calling. Table S7. Outgroup CNV regions and their frequencies. Table S8. primers did not overlap with any of Dominette’sCNV The effects of CNV length and frequency on calling consistances between CNV callings based on SNP array and aCGH. Table S9. qPCR regions, two-copy states were assumed for both ampli- Summary. Table S10. Over/Underrepresentation of PANTHER molecular cons in Dominette. function, biological process and pathway terms. Table S12. CNVR frequency differences among breeds. Figure S1. Illustration of a typical CNV call with qPCR validation. Cattle CNV distribution and association with Segmental Additional file 2: Supplemental Material file. Table S6. Gene contents Duplications and other features of cattle CNV regions. We investigated the genomic distribution of 682 CNVRs Additional file 3: Supplemental Material file. Table S11. Cattle CNV by testing the hypothesis that pericentromeric and sub- regions overlap with genomic regions under positive selection, human telomeric regions were enriched for CNVs as described orthologous OMIM genes and cattle QTLs. previously [32]. Briefly, all predicted variable bases that overlapped with these regions were totaled and chi- square tests were used to test the null hypothesis of no Acknowledgements enrichment. CNVRs were also overlapped with SD and We thank members of the Bovine HapMap Consortium for sharing their samples. We thank R. Anderson, J. Shaffer, D. Hebert, S. Schroeder and L. the other genomic features such as evolutionary break- Shade for technical assistance. This work was supported by in part by NRI point regions (ERBs), which were obtained from litera- grant No. 2007-35205-17869 from USDA CSREES (now NIFA), Project 1265- ture and public databases listed in web site references. 31000-098-00 from USDA-ARS and Research Grant Award No. IS-4201-09 from BARD, United States - Israel Binational Agricultural Research and Development Fund. Gene content Gene content of cattle CNV regions was assessed using Author details Ensembl genes ftp://ftp.ensembl.org/pub/current_fasta/ Bovine Functional Genomics Laboratory, ANRI, USDA-ARS, Beltsville, Maryland 20705, USA. Department of Animal and Avian Sciences, University bos_taurus/pep/, the Glean consensus gene set, cattle Hou et al. BMC Genomics 2011, 12:127 Page 10 of 11 http://www.biomedcentral.com/1471-2164/12/127 of Maryland, College Park, Maryland 20742, USA. Department of Genetics 19. Colella S, Yau C, Taylor JM, Mirza G, Butler H, Clouston P, et al: QuantiSNP: and Microbiology, University of Bari, Bari 70126, Italy. Center for Applied an Objective Bayes Hidden-Markov Model to detect and accurately map Genomics, Children’s Hospital of Philadelphia, Philadelphia, PA 19104, USA. copy number variation using SNP genotyping data. Nucleic Acids Res Bioinformatics and Computational Biology, George Mason University, 2007, 35:2013-2025. Manassas, VA 20110, USA. Animal Improvement Programs Laboratory, ANRI, 20. Wang K, Li M, Hadley D, Liu R, Glessner J, Grant SF, et al: PennCNV: an USDA-ARS, Beltsville, Maryland 20705, USA. integrated hidden Markov model designed for high-resolution copy number variation detection in whole-genome SNP genotyping data. Authors’ contributions Genome Res 2007, 17:1665-1674. YH, KW, EK, LKM, JS carried out computational analysis. DMB, MFC, MV 21. Korn JM, Kuruvilla FG, McCarroll SA, Wysoker A, Nemesh J, Cawley S, et al: carried out the experimental validations. PMV, TSS and CPVT participated in Integrated genotype calling and association analysis of SNPs, common data collection. GEL conceived of the study and led in its design and copy number polymorphisms and rare CNVs. Nat Genet 2008, coordination. All authors contributed to writing the manuscript, read and 40:1253-1260. approved the final manuscript. 22. Yavas G, Koyuturk M, Ozsoyoglu M, Gould MP, LaFramboise T: An optimization framework for unsupervised identification of rare copy Received: 10 September 2010 Accepted: 23 February 2011 number variation from SNP array data. Genome Biol 2009, 10:R119. Published: 23 February 2011 23. Winchester L, Yau C, Ragoussis J: Comparing CNV detection methods for SNP arrays. Brief Funct Genomic Proteomic 2009, 8:353-366. 24. LaFramboise T: Single nucleotide polymorphism arrays: a decade of References biological, computational and technological advances. Nucleic Acids Res 1. The Bovine Genome Sequencing and Analysis Consortium: The Genome 2009, 37:4181-4193. Sequence of Taurine Cattle: a window to ruminant biology and 25. Marioni JC, Thorne NP, Valsesia A, Fitzgerald T, Redon R, Fiegler H, et al: evolution. Science 2009, 324:522-528. Breaking the waves: improved detection of copy number variation from 2. Zimin AV, Delcher AL, Florea L, Kelley DR, Schatz MC, Puiu D, et al: A microarray-based comparative genomic hybridization. Genome Biol 2007, whole-genome assembly of the domestic cow, Bos taurus. Genome Biol 8:R228. 2009, 10:R42. 26. Diskin SJ, Li M, Hou C, Yang S, Glessner J, Hakonarson H, et al: Adjustment 3. The Bovine HapMap Consortium: Genome wide survey of SNP variation of genomic waves in signal intensities from whole-genome SNP uncovers the genetic structure of cattle breeds. Science 2009, genotyping platforms. Nucleic Acids Res 2008, 36:e126. 324:528-532. 27. Wang K, Chen Z, Tadesse MG, Glessner J, Grant SF, Hakonarson H, et al: 4. Van Tassell CP, Smith TP, Matukumalli LK, Taylor JF, Schnabel RD, Lawley CT, Modeling genetic inheritance of copy number variations. Nucleic Acids et al: SNP discovery and allele frequency estimation by deep sequencing Res 2008, 36:e138. of reduced representation libraries. Nat Methods 2008, 5:247-252. 28. Liu GE, Van Tassell CP, Sonstegard TS, Li RW, Alexander LJ, Keele JW, et al: 5. Matukumalli LK, Lawley CT, Schnabel RD, Taylor JF, Allan MF, Heaton MP, Detection of germline and somatic copy number variations in cattle. et al: Development and characterization of a high density SNP Developments in Biologicals 2008, 132:231-237. genotyping assay for cattle. PLoS ONE 2009, 4:e5350. 29. Liu GE, Hou Y, Zhu B, Cardone MF, Jiang L, Cellamare A, et al: Analysis of 6. VanRaden PM, Van Tassell CP, Wiggans GR, Sonstegard TS, Schnabel RD, copy number variations among diverse cattle breeds. Genome Res 2010, Taylor JF, et al: Invited review: reliability of genomic predictions for North 20:693-703. American Holstein bulls. J Dairy Sci 2009, 92:16-24. 30. Fadista J, Thomsen B, Holm LE, Bendixen C: Copy number variation in the 7. Hayes BJ, Bowman PJ, Chamberlain AJ, Goddard ME: Invited review: bovine genome. BMC Genomics 2010, 11:284. Genomic selection in dairy cattle: progress and challenges. J Dairy Sci 31. Bae JS, Cheong HS, Kim LH, NamGung S, Park TJ, Chun JY, et al: 2009, 92:433-443. Identification of copy number variations and common deletion 8. Wiggans GR, Sonstegard TS, VanRaden PM, Matukumalli LK, Schnabel RD, polymorphisms in cattle. BMC Genomics 2010, 11:232. Taylor JF, et al: Selection of single-nucleotide polymorphisms and quality 32. Liu GE, Ventura M, Cellamare A, Chen L, Cheng Z, Zhu B, et al: Analysis of of genotypes used in genomic evaluation of dairy cattle in the United recent segmental duplications in the bovine genome. BMC Genomics States and Canada. J Dairy Sci 2009, 92:3431-3436. 2009, 10:571. 9. Cole JB, VanRaden PM, O’Connell JR, Van Tassell CP, Sonstegard TS, 33. Marioni JC, White M, Tavare S, Lynch AG: Hidden copy number variation Schnabel RD, et al: Distribution and location of genetic effects for dairy in the HapMap population. Proc Natl Acad Sci USA 2008, 105:10067-10072. traits. J Dairy Sci 2009, 92:2931-2946. 34. Sonstegard TS, Barendse W, Bennett GL, Brockmann GA, Davis S, 10. Flori L, Fritz S, Jaffrezic F, Boussaha M, Gut I, Heath S, et al: The genome Droegemuller C, et al: Consensus and comprehensive linkage maps of response to artificial selection: a case study in dairy cattle. PLoS ONE the bovine sex chromosomes. Anim Genet 2001, 32:115-117. 2009, 4:e6595. 35. Rhead B, Karolchik D, Kuhn RM, Hinrichs AS, Zweig AS, Fujita PA, et al: The 11. Scherer SW, Lee C, Birney E, Altshuler DM, Eichler EE, Carter NP, et al: UCSC Genome Browser database: update 2010. Nucleic Acids Res 2010, 38: Challenges and standards in integrating surveys of structural variation. D613-D619. Nat Genet 2007, 39:S7-15. 36. Troy CS, Machugh DE, Bailey JF, Magee DA, Loftus RT, Cunningham P, et al: 12. Conrad DF, Pinto D, Redon R, Feuk L, Gokcumen O, Zhang Y, et al: Origins Genetic evidence for Near-Eastern origins of European cattle. Nature and functional impact of copy number variation in the human genome. 2001, 410:1088-1091. Nature 2009, 464:704-712. 37. Caramelli D: The Origins of Domesticated Cattle. Human Evolution 2006, 13. Redon R, Ishikawa S, Fitch KR, Feuk L, Perry GH, Andrews TD, et al: Global 21:107-122. variation in copy number in the human genome. Nature 2006, 38. Eichler EE: Widening the spectrum of human genetic variation. Nat Genet 444:444-454. 2006, 38:9-11. 14. Feuk L, Carson AR, Scherer SW: Structural variation in the human 39. Matsuzaki H, Wang PH, Hu J, Rava R, Fu GK: High resolution discovery and genome. Nat Rev Genet 2006, 7:85-97. confirmation of copy number variants in 90 Yoruba Nigerians. Genome 15. McCarroll SA, Altshuler DM: Copy-number variation and association Biol 2009, 10:R125. studies of human disease. Nat Genet 2007, 39:S37-S42. 40. Larkin DM, Pape G, Donthu R, Auvil L, Welge M, Lewin HA: Breakpoint 16. Marques-Bonet T, Girirajan S, Eichler EE: The origins and impact of primate regions and homologous synteny blocks in chromosomes have different segmental duplications. Trends Genet 2009, 25:443-454. evolutionary histories. Genome Res 2009, 19:770-777. 17. Zhang F, Gu W, Hurles ME, Lupski JR: Copy number variation in human 41. Nei M, Rooney AP: Concerted and birth-and-death evolution of health, disease, and evolution. Annu Rev Genomics Hum Genet 2009, multigene families. Annu Rev Genet 2005, 39:121-152. 10:451-481. 42. Liu GE, Li RW, Sonstegard TS, Matukumalli LK, Silva MV, Van Tassell CP: 18. Peiffer DA, Le JM, Steemers FJ, Chang W, Jenniges T, Garcia F, et al: High- Characterization of a novel microdeletion polymorphism on BTA5 in resolution genomic profiling of chromosomal aberrations using Infinium cattle. Anim Genet 2008, 39:655-658. whole-genome genotyping. Genome Res 2006, 16:1136-1148. Hou et al. BMC Genomics 2011, 12:127 Page 11 of 11 http://www.biomedcentral.com/1471-2164/12/127 43. Seroussi E, Glick G, Shirak A, Yakobson E, Weller JI, Ezra E, et al: Analysis of copy loss and gain variations in Holstein cattle autosomes using BeadChip SNPs. BMC Genomics 2010, 11:673. 44. Fontanesi L, Beretti F, Martelli PL, Colombo M, Dall’olio S, Occidente M, et al: A first comparative map of copy number variations in the sheep genome. Genomics 2010. 45. Fontanesi L, Martelli PL, Beretti F, Riggio V, Dall’olio S, Colombo M, et al: An initial comparative map of copy number variations in the goat (Capra hircus) genome. BMC Genomics 2010, 11:639. 46. Ramayo-Caldas Y, Castello A, Pena RN, Alves E, Mercade A, Souza CA, et al: Copy number variation in the porcine genome inferred from a 60 k SNP BeadChip. BMC Genomics 2010, 11:593. 47. Decker JE, Pires JC, Conant GC, McKay SD, Heaton MP, Chen K, et al: Resolving the evolution of extant and extinct ruminants with high- throughput phylogenomics. Proc Natl Acad Sci USA 2009, 106:18644-18649. 48. Livak KJ, Schmittgen TD: Analysis of relative gene expression data using real-time quantitative PCR and the 2(-Delta Delta C(T)) Method. Methods 2001, 25:402-408. 49. Graubert TA, Cahan P, Edwin D, Selzer RR, Richmond TA, Eis PS, et al: A high-resolution map of segmental DNA copy number variation in the mouse genome. PLoS Genet 2007, 3:e3. doi:10.1186/1471-2164-12-127 Cite this article as: Hou et al.: Genomic characteristics of cattle copy number variations. BMC Genomics 2011 12:127. Submit your next manuscript to BioMed Central and take full advantage of: • Convenient online submission • Thorough peer review • No space constraints or color figure charges • Immediate publication on acceptance • Inclusion in PubMed, CAS, Scopus and Google Scholar • Research which is freely available for redistribution Submit your manuscript at www.biomedcentral.com/submit

Journal

BMC Genomics – Springer Journals

Published: Feb 23, 2011

There are no references for this article.