Identification of copy number variants polymorphism associated with productive traits in Holstein. Analyzing their predictive ability

Copy number variants (CNV) are structural variants consisting in duplications or deletions of genomic fragments longer than lkb of size that are variable across individuals and heritable. Several of these CNV have already been associated with phenotypic variants in human as well as in livestock species. The objective of this study was to identify copy number variants associated to seven economical traits: fat yield, protein yield, somatic cell score, days open, stature, foot angle and udder depth in 1-lolste.in. CNVs were detected by exploiting hybridization signal intensities from the BovineSNP50BeadChip and using the PennCNV software. More than 2000 CNVRs were identified, 315 of which appeared segregating (MAF>0.005) in the bul.is with available phenotypic data. Frequentist and bayesian approaches were used in the association analyses. Both fat and protein yield shared associated CNVRs. Among the genes contained within these regions highlight GHI, SREB, STAT5, TNRF and ACLY which have been previously associated to fat yield in different studies. Besides, growth hormone gene was also associated to animal stature, which suggests a pleiotropic effect, in agreement with previous studies. Further, five different genes (DGATI, IGFAL, GFER, PPARA and CPT1B) included within CNVRs associated with both protein and fat yield have also been previously reported as related to both traits in Holstein. Additionally, we report that six different CNVRs are significantly associated with somatic cell count. Among the genes contained within these regions, the LRPIB has been already related with the trait. Genomic prediction was conducted on those traits using SNP and CNV.R information. 1198 bulls divided according to its birth year in training set (1021 bulls) and testing set (183 bulls) were used in this analysis. The conducted genomic predictions using SNP information were evaluated using a semi¬parametric G-BLUP. Secondly, CNVR information was added to the model assuming a double exponential prior distribution for their effects. Only the 52 most polymorphic CNVRs (MAF less than 0.005) were used for genomic prediction models. Three metrics were calculated to evaluate the predictive ability of the different methods: the correlation coefficient between phenotypic data and predicted EBV was used as a measure of accuracy, minimum square error and regression coefficient that predicts the degree of bias of the EBV prediction. Results showed that including CNVR information within the model provides in average, an increase of 1 per cent for Fat Yield, Protein Yield, Somatic Cell Score, Foot Angle and Udder Depth accuracy. In addition, the regression slope for days open and foot angle were almost equal to the unity when including CNVRs within the model, which proves non biased estimates. In general, genomic prediction accuracy did not change in a relevant manner at including information from the exact number of copies of CNVs, regarding the traditional G-BLUP evaluation. The present study reported associations of CNV regions with important economical traits, among them several match with previously reported QTLs and candidate genes, supporting the obtained results. Additionally, CNVRs have been incorporated into genomic predictions models and the results seem promising. However it should be taken into account the limitations of the study, regarding the relatively small sample size and the limitations in the methodology to detect CNVs. Therefore, some complementary analysis using quantitative PCR or genome sequencing would probably allow us to validate these results. In any case, the use of higher density SNP chip (Bovine 600K SNP chip) would probably facilitate a more precise CNV identification to better understand the genetic architecture of economic important traits and perform better genomic predictions and selection.