Multi-breed genetic parameters and genome-wide association studies for mortality rate at birth in pigs

Background Piglet mortality is an economically important complex trait that impacts sow prolificacy in the 6 pig industry. The genetic parameters estimations and genome-wide association studies will 7 help us to better understand the genetic fundamentals of piglet mortality. However, compared 8 with other economically important traits, a little breakthrough in the genetic analyses of the 9 trait has been achieved. 10 In this study, we used multi-breed data sets from Yorkshire, Landrace, and Duroc sows and 12 characterized the genetic and genomic properties of mortality rate at birth by treating each 13 parity as a different trait. The heritability of mortality rate from parity I to III were estimated 14 to be 0.0630, 0.1031, and 0.1140, respectively. The phenotypic and genetic correlations with 15 its component traits were all positive with ranges from 0.0897 to 0.9054, and 0.2388 to 0.9999, 16 respectively. Integrating the results, we identified 21 loci that were detected at least by two 17 tools from standard MLM, FarmCPU, BLINK and mrMLM, and these loci were annotated to 18 22 genes. The annotations revealed that the gene expressions were associated with the 19 reproductive system, nervous system, digestive system, and embryonic development, which 20 are reasonably related to the piglet mortality. properties birth were findings provide much information for understanding the genetic and genomic fundamentals 24 of farrowing mortality and also identify candidate molecular markers for breeding practice.


28
Piglet mortality-related traits are a category of economically important traits that provide direct 29 or indirect metrics of piglet deaths and produce heavy economic losses and welfare concerns 30 to the pig industry (1, 2). The majority of piglet mortality traits have been documented to be 31 complex traits with low heritability ranging from 0.03 to 0.17 (3). In theory, the outcome of 32 mortality is the tri-interactions between piglet, sow and environment, and the phenotypic 33 variation could be caused by diverse systematic and non-systematic factors, including genetic 34 background (breed), parity, season, disease, management, piglet vitality, and sow's behavior 35 such as crushing and starvation (4-6). For the high economic merit, there has been a growing 36 emphasis on reducing piglet losses in pig breeding programs of some countries. 37 Past experiences from breeders revealed that selection on litter size increases piglet mortality 38 and the intensive couplings with litter size implied that there might exist strong negative 39 linkage disequilibrium (LD) or opposite pleiotropy in the cross-trait genetic architectures (7). 40 However, to date, there have been no more than 10 reports publicly available to describe the 41 genomic fundamentals of piglet mortality-related traits (8,9). Compared with other 42 economically important traits, little breakthrough in the genetic dissections of piglet mortality-43 related traits has been achieved. The limited progress cannot underpin a pinpoint understanding 44 of genetic properties of piglet mortality-related traits, and further efforts are needed. 45 There are usually two time points to measure mortality, including at birth and at weaning (10) 46 . No matter at weaning or at birth the mortality is measured, the metrics are derived from its 47 component traits, i.e. litter size-related traits. When dealing with piglet mortality as well as its 48 components, there remains an important concern for parity. It is still unable to reach a 49 consensus about how to treat this type of data sets in practice. In theory, during the first parities, 50 the reproductive organs of gilt are still undergoing developmental changes, while for higher 51 parity sows the risk of death increases due to oxytocin insufficiency and ruptured umbilical 52 cord (11). Given these, many researchers treated each parity as a different trait in genetic 53 analyses. For example, Roehe & Kennedy (1995) reported that the genetic parameters of litter 54 size were estimated with each parity treated as a different trait (12). More studies revealed that 55 the estimations of genetic parameters varied between different parities in different pig cohorts 56 (13,14). In addition, researchers also found that the reproductive traits in different parities had 57 a different genetic architecture (15). So, it's quite sound to treat each parity as a different trait.

58
There were growing studies that used multi-breed data sets for genetic parameter estimation 59 and GWAS. In general, the multi-breed approach has potential advantages, such as enlarging 60 the sample size by putting the multi-breed individuals together, capturing the genetic variants 61 both within and across breeds, and improving the accuracy of genetic evaluation. For example, 62 a simulation study has evaluated the efficiency of the multi-breed approach, and reported that 63 the multi-breed approach could improve the accuracy of genomic estimated breeding values 64 (GEBVs) for the second breed with fewer sizes (16). It was also found that the multi-breed 65 models had a positive effect on the genetic parameter estimations (17). Raven et al. (2014) 66 declared that the multi-breed approach could accurately locate the highly conserved functional 67 mutations because the mixed population had lower levels of long-range LD (18). The multi-68 breed approach has been widely proven to be feasible in genetic analyses.

69
Knowledge of genetic property for a trait is involved in many aspects, in which genetic 70 parameters and genomic architecture are two important ones. The aim of this study was to 71 characterize the genetic property of mortality rate at birth using the mixture data sets from 72 Yorkshire, Landrace, and Duroc sows. In this study, the genetic parameters including breeding 73 value, heritability, and genetic correlation between piglet mortality and its component traits 74 from parity I to III were estimated, and GWAS on piglet mortality was performed to identify 75 the genome-wide variants and putative genes underlying the variability of piglet mortality. This 76 study would accelerate our understanding of the molecular fundamentals of piglet mortality 77 and provides potential markers for pig breeding programs.  (23) and the imputed genotype data were also filtered, using the same conditions as the former.

133
To increase the detection credibility, we used a combined approach for GWAS. Concretely  figures from D to F. In this study, the piglet mortality was defined as a ratio trait that is 166 calculated as the ratio of total number dead (TND) over total number born (TNB). In usual, a 167 ratio trait is departure from the normal distribution, and it was found that the phenotypes of 168 piglet mortality from parity I to III followed a heavy skewed distribution. Compared with the 169 heavy skewed distribution of phenotypes, all distribution curves of the EBVs from parity I to 170 III had two tails with a positively skewed distribution, which were relatively closer to the 171 normal distribution. For more detailed information, the descriptive statistics of raw data sets 172 was given in Additional Table 4.

180
The results of heritability estimation for piglet mortality were presented in  FarmCPU, BLINK, and mrMLM were used to run the GWAS analyses. In the GWAS analyses, 206 the target trait from parity I to III have different sample sizes (parity I, n=1331; parity II, 207 n=1220; parity III, n=980). The top ten SNPs identified by each tool for each parity were listed 208 in Additional Table 5. After extracting and summarizing the results of GWAS, Figure 2 showed 209 the circular-Manhattan plots of piglet mortality traits from MLM, FarmCPU, BLINK, 210 mrMLM. In addition, the Venn diagrams were drawn to identify the intersections of the top ten 211 SNPs from four tools (Figure 3).

212
In total, 21 SNPs were identified, of which, 6 belonged to parity I, 5 belonged to parity II, and 213 10 belonged to parity III. For the identified SNPs from different parities, no overlapping was 214 observed. All identified SNPs passed the permutation test, and were statistically confirmed.

215
The SNP symbols, the smallest p-values from GWAS, and permutated p-values of the 216 identified SNPs were listed in the Additional Table 3.

Gene annotations 231
All SNPs that passed the permutation test were further used for gene annotations. In total, we 232 obtained 22 candidate genes that harbor or near the 21 identified SNPs. The position 233 information of 21 SNPs and corresponding 22 genes were shown in Table 3. Among them, the 234 positions of MARC0113660 and DRGA0008818 are located within 5.8kb, and there is only 235 one gene (STPG2) in this region. According to the annotation criterion, there were two genes 236 CDK8 and WASF3 that were both close to ALGA0060358. Three SNPs, including 237 ALGA0036320, H3GA0018655 and ASGA0029165, were annotated to be close to four genes 238 that included CABYR, OSBPL1A, IMPACT, and HRH4. It can be found that, among these 239 SNPs, there were totally eight SNPs clustered on SSC 8. Furthermore, we used the information 240 extracted from TISSUES database (TISSUES 2.0) to visualize the digital tissue expression 241 profiles for target genes. When drawing the heat map, three genes were dropped because there 242 was no expression information for them. At last, the heat map of tissue expressions for 19 243 annotated genes from different tissues was presented in Figure 3. The heat map revealed that 244 most of these genes have been expressed in reproductive and urinary system, nervous system, 245 and digestive system, and many expressions were detected in fetus. The tissue expression 246 profiles revealed that the identified genes are intuitively related to the physiological processes 247 contributing to piglet mortality, such as embryo development.

254
In the pig industry, piglet mortality is intensively related to sow prolificacy, which is generally 255 defined as the number of piglets weaned per sow per year (PSY). It is of high importance to 256 characterize the genetic properties of piglet mortality. In this study, the piglet mortality was 257 defined as a ratio trait that was re-constructed by its component traits, which needed

289
In this work, we proposed a combined approach to increase the detection credibility in GWAS.

290
In the pipeline, the standard MLM, FarmCPU, BLINK, mrMLM were simultaneously utilized 291 to identify the putative SNPs, and the permutation test was followed to statistically confirm the and then contribute to piglet mortality (42).

328
In brief, piglet mortality at birth was found a low heritability trait. All phenotypic and genetic 329 correlations between piglet mortality and its component traits were estimated to be positive.

330
Integrating the results from standard MLM, FarmCPU, BLINK, and mrMLM, we identified 21 331 loci and 22 genes associated with piglet mortality. Most of these genes were annotated to be 332 expressed in the reproductive system, nervous system, digestive system, and embryonic 333 development, which are reasonably related to piglet losses. This study advances our 334 understanding of the genetic and genomic fundamentals of piglet mortality and also provides 335 candidate genes that could be potentially used for pig breeding programs, genomic selection, 336 and further investigations.