A systematic survey to identify lethal recessive variation in highly managed pig populations

Small effective population size can lead to inbreeding depression. The cause of inbreeding depression is the accumulation of (recessive) deleterious alleles increasing in frequency and becoming expressed in homozygous state due to drift in small populations [1]. In domesticated populations, despite strong artificial selection for desired traits, selection on relatively rare variation (allele frequency < 10%) is usually very inefficient [2]. The eradication of deleterious variation is challenging by applying either traditional or even more recent ‘genomic’ breeding strategies. The inefficiency of purging deleterious variation, even from highly managed populations, is particularly apparent if there is an unpredictable, or poorly characterized relationship between genotype and phenotype. For instance, when a homozygous deleterious phenotype leads to very early death of the developing embryo, the only observed consequence is a (somewhat) lower fertility of the parents. It is estimated that livestock species harbour 2–4 fold higher variation in the genome compared to humans [3, 4]. Accordingly, the number of non-synonymous mutations has been shown to be greater in livestock animals, suggesting a large reservoir of potentially deleterious variation in livestock [3].

Domestication in general, and modern breeding industry in particular, constitutes the largest and longest lasting genetic experiment ever conducted. In many cases, especially in pig and poultry breeding, commercial breeding organizations apply their genetic improvement efforts at a small number of elite breeding lines [5]. These breeding lines are usually fairly closed, i.e. exchange between breeding lines is infrequent. Moreover, depending on species and breeding purpose, the effective populations sizes vary from small (several hundred at most) to very small (dozens of animals). With the adoption of genomic selection, a large proportion of the animals in the pure bred elite lines, i.e. the selection candidates, are genotyped using high to medium density SNP assays to estimate their genomic estimated breeding value (GEBV). Breeding values itself are not very efficient in eliminating rare deleterious variation. However, the availability of a large number of genotyped, pedigreed individuals enables the unravelling of the genetic basis of rare disorders in the population. Recessive deleterious variants can be identified by testing for statistical depletion, or even the absence, of specific haplotypes in homozygous state. This haplotype approach is a powerful tool [6–10] originally developed for cattle by vanRaden et al. (2011). The power of this method heavily depends on the number of genotyped individuals. If applied to tens of thousands of genotyped animals, a number prohibitively large usually for academic budget, but routinely attained in many commercial breeding lines nowadays, even very rare deleterious haplotypes can be detected (frequency < 2%). Genotyping large numbers of domestic animals currently still relies on the use of dedicated SNP assays (‘SNP chips’). Since these assays are designed with a bias towards high minor allele frequency (MAF), causal variants of serious syndromes are unlikely to be present on the assay. The haplotype based approach is therefore more efficient in capturing deleterious variation compared to individual SNPs, which are expected to be often in low linkage disequilibrium (LD) with the causal variant. Significant depletion of haplotype homozygosity is an indication of decreased viability, and these haplotypes are likely to harbour deleterious mutations causing embryonic lethality (EL) in homozygous state. Studies in mouse showed a lethal knockout phenotype for about 30% of mouse genes in homozygous state, suggesting a large proportion of potential embryonic lethal genes in Mammalia [11].

Fertility traits, such as total number born (TNB), are among the relevant phenotypes systematically recorded in pig breeding that can provide phenotypic support for (early) embryonic lethality. In addition, recorded phenotypes, such as number of stillborn (NSB) and the number of mummified pigs (MUM), can provide additional information on genetic defects that are lethal, or seriously compromising the survival probability later in foetal development. In theory, assuming prenatal death of homozygotes, a loss of 25% of each litter is expected if two heterozygous carriers of the lethal variant mate (C x C mating). In mammals, death of an embryo or foetus usually does not result in spontaneous termination of the pregnancy, when also carrying living young. Instead, foetuses go through a process of desiccation and encapsulation, known as mummification [12]. Foetal mummification can occur from day 35 of gestation until parturition (when the skeletal system is developing), and has been associated with several risk factors like large litter size and infectious disease [13]. Also, several studies have already identified QTLs associated with pig reproductive traits, including mummified piglets [14–17]. Parameters, such as time of death and any morphological abnormalities found in mummified piglets [12], can provide insights in the developmental consequences of a specific genetic defect without any further welfare concerns for the mother or live siblings. A systematic analysis to identify specific genetic defects as risk factors, however, has not been conducted.

In this study, we aim to identify novel genetic loci associated with both early and late termination of development during gestation in pigs. Moreover, we examine the occurrence and frequency of lethal haplotypes in the studied breeds, as well as their impact on fertility related traits. Finally, we show that missing homozygosity in highly managed livestock populations can be a result of early lethality caused by low frequency recessive lethal haplotypes.

Highly managed, domesticated populations are expected to be under selection against inbreeding depression. Indeed, our results show that high frequency occurrence of potentially monogenic lethal or debilitating alleles is rare in commercial populations, despite relatively low effective population sizes. This is in contrast to other domesticated populations that are far less well managed, such as dogs and horses, and that can carry high frequencies of deleterious alleles [23, 24]. The few examples that exist for commercial breeding populations, mostly from Holstein cattle, indicate that the effects of deleterious traits are often masked, because they involve early embryonic lethality, which reveals itself only indirectly as depressed parent fertility. Moreover, some of these lethal alleles are maintained in the population as a result of balancing selection, where heterozygotes show an advantageous phenotype [25]. However, even in those cases, these alleles are usually kept at low frequency.

It is unlikely that purging can remove all or even most of the detrimental variation because even modern genomic breeding programs are inefficient in capturing genotype-phenotype relations of low frequency alleles. Our study reveals that the frequency of the haplotypes exhibiting missing homozygosity ranges from 0.5–11%, showing that we have the statistical power to detect very rare deleterious haplotypes in our populations, but also confirming that, as expected, truly lethal recessive variants are invariably infrequent.

The approach chosen for this study relies on the premise that unexpected absence of homozygotes results from unviability of the homozygous deleterious allelic state. Ideally, if we would have sequence data for many thousands of animals, we would be able to directly infer the absence of specific homozygous allele states, e.g. alleles that impair the required protein function, or, alternatively, alleles that are in complete LD with such variants. However, we used low to medium density genotype data, and the SNPs on the chips have been primarily chosen based on their relatively high MAF in most breeds, unlikely to include deleterious variants. To overcome this problem, the haplotype based analysis chosen in this study applied a sliding window approach from 0.5 to 4 Mb. Selecting optimum window sizes is not trivial and depends on population structure, SNP density, recombination rate, and haplotype frequencies within the examined genomic region. For example, by selecting large haplotypes, we increase the risk of analysing recombinant haplotypes. Moreover, by selecting very small haplotypes, non-unique haplotypes might be selected (overlap between distinct haplotypes). We solely used information from complete trios (both parents and offspring genotyped) to calculate the expected number of homozygotes. This number, however, is likely an underestimation, because not all genotyped animals are in genotyped trios. Also, the haplotype approach is unlikely to capture all deleterious variants, as any rare variant that resides only on a common haplotype will be missed. However, rare variants that coincide with rare haplotypes can be robustly detected, with the number of genotyped offspring being the limiting factor for statistical power.

In total, 145 haplotypes showed a significant deficit of homozygotes. Of these, 35 haplotypes showed a negative effect on at least one of the three fertility traits examined, indicating that indeed these 145 haplotypes are highly enriched for variation that can lead to embryonic lethality or prenatal death. The overwhelming majority of these haplotypes are located at chromosomal regions not previously linked to fertility [19]. Only four genomic loci are shared between the three breeding populations, and all four of these were previously identified as copy number variable (CNV) regions [18]. We hypothesize that these four haplotypes are likely false positives, as these CNV events can cause duplication of genetic markers potentially introducing polymorphisms. This could lead to inter-locus cross-hybridization of oligo’s on the chip, causing all individuals to become heterozygous for a particular marker, or a set of markers, generating haplotypes with missing homozygotes.

Several of the identified haplotypes did not show a significant effect on fertility. In some cases, the number of C x C matings was too low to obtain significant statistical support. There are three additional explanations for the absence of an effect on fertility. First, recombination hotspots can potentially result in an excess of heterozygotes that carry recombinant haplotypes. This seems especially apparent at the chromosome ends. Second, since we are examining commercial breeding lines, there is selection on the animals that are genotyped. Piglets are selected for genotyping based on their performance on numerous traits depending on their particular breeding purpose, e.g. growth rate, back fat, fertility, number of teats, and leg quality. Piglets that exhibit unfavourable phenotypes early in life are likely not genotyped and could end up as “missing homozygotes” in our analysis. One example of such a phenotype in pig breeding is the number of teats. Piglets with less than 12 teats are often immediately removed from the population. Third, embryos dying very early (e.g. before time of implantation) are likely replaced by other embryos, as breeding sows are very likely to produce a far higher number of ova than can be accommodated in the uterus.

For one specific haplotype (LW19), identified only in the Large White line, we found evidence from 88 carrier matings that homozygous animals die mid-gestation, and become mummified (fivefold increase in the number of mummified piglets). The association of mummified piglets and haplotype LW19 is very likely an underestimation, since the number of mummified piglets is not always recorded equally strict at all breeding farms, especially for the embryos that died early in gestation for which mummies are small. The haplotype showed a 19% reduction in TNB, less than the 25% loss expected under HWE. One explanation for this discrepancy is that the whole litter might potentially be aborted if a large proportion of the litter dies during gestation, and will therefore not be recorded. We identified the BMPER gene as a likely candidate gene causing the defect. In human, the perinatal lethal skeletal disorder diaphanospondylodysostosis (OMIM: 608,022) is associated with homozygous or compound heterozygous mutations in the BMPER gene [26]. Characteristics include a small chest, abnormal vertebral segmentation, and posterior rib gaps [26]. Homozygous knockout mice exhibit neonatal lethality associated with abnormal lung and skeleton development [27, 28]. Moreover, heterozygotes for a null allele exhibit abnormal lung development [28]. BMPER is involved in the negative regulation of bone morphogenetic proteins (BMPs), a group of growth factors involved in the formation of bone and cartilage [29]. Variation in this gene has been associated with increased body size and rump length in cattle [30], and higher intramuscular fat content in pig [31]. Evidence for similar early termination of development comes from human and mouse studies [26, 28]. We observed multiple haplotypes with a deficit of homozygotes surrounding the LW19 haplotype associated with similar phenotypic effects (increase in number of mummified piglets, decrease in TNB). It is likely that these haplotypes are not in complete LD with the causal variant. Interestingly, these haplotypes are surrounding a region under selection, despite low LD with the selected haplotype, we hypothesize that this could be a remnant of genetic hitchhiking in the past as this locus has previously been associated with increased body weight and ovulation rate [17, 32]. Therefore, LW19 might have been subjected to genetic hitchhiking, although we did not find direct evidence supporting this (LW19 is not in LD with the neighbouring haplotypes under selection). More recent recombination might have lowered the LD, but as a result of previous hitchhiking the haplotype still segregates in the population.

Despite the limited impact on crossbred products, given that most haplotypes are population specific, eradication of these haplotypes is still desired. Especially because embryonic lethality leading to mummification does not only have negative economic consequences for the pig breeder, but also results in reduced animal welfare such as health risk for the sow [12]. Our study can directly impact positively on current breeding programmes, by avoiding C x C matings to lower the frequency of the lethal recessive haplotypes in the elite breeding lines. Furthermore, if a causal variant is found, avoiding homozygotes could be combined with a low-level selection to eradicate the variant after a number of generations. Many risk factors have been associated with an increase in mummified piglets [33], most of them, however, are independent of the foetus’s genetic material. In our study, we found that about 2% of all recorded mummies in the Large White breed can be attributed to C x C matings for the LW19 haplotype. We believe that the majority of the total recorded mummified piglets are not a direct effect of a genetic defect carried by the unborn foetus, because several other factors, especially many pathogens can cause mummified or stillborn piglets. Therefore, this proportion of 2% likely represents a much larger fraction of the total number of mummified piglets directly caused by a genetic defect carried by the unborn foetus.

The use of molecular tools, in particular in genomic selection, has considerably increased breeding progress over the past years. Despite this, identification of low-frequency deleterious recessive alleles, present in livestock populations, remains a challenging task. The expected low frequency of this type of variation, and therefore marginal effects on fertility traits at the population, contribute to this challenge. However, the routine large scale genotyping of domesticated animals has opened up new possibilities to detect these low frequency deleterious alleles. A systematic genomic survey for missing homozygosity, as applied in this study, is especially promising when thousands of individuals are genotyped. Using this method, novel genetic defects can be identified and fully characterised. Moreover, the increased use of whole genome sequence (WGS) data for breeding purposes opens new opportunities to directly infer deleterious variants from the sequence itself [3].

Scanning for depletion of homozygous haplotypes provides a powerful tool to identify deleterious recessive alleles, especially for large genotyped livestock populations. Our results confirm the existence, relative rarity, and severe effects on fertility and welfare of lethal haplotypes in commercial livestock populations. We show that these haplotypes, apart from reduced fertility in the parent animal, also cause large numbers (several hundred at population level) of stillbirths (‘mummified piglets’), even within highly managed populations. Moreover, the method applied in this study is increasing in potential, as growing numbers of genotyped animals are becoming available for breeding purposes. Finally, this study will facilitate at least partial purging of lethal variation by avoiding matings producing affected or non-viable progeny, demonstrating its value for current breeding programs and animal welfare.