Analysis of a large dataset reveals haplotypes carrying putatively recessive lethal and semi-lethal alleles with pleiotropic effects on economically important traits in beef cattle

Background In livestock, deleterious recessive alleles can result in reduced economic performance of homozygous individuals in multiple ways, e.g. early embryonic death, death soon after birth, or semi-lethality with incomplete penetrance causing reduced viability. While death is an easy phenotype to score, reduced viability is not as easy to identify. However, it can sometimes be observed as reduced conception rates, longer calving intervals, or lower survival for live born animals. Methods In this paper, we searched for haplotypes that carry putatively recessive lethal or semi-lethal alleles in 132,725 genotyped Irish beef cattle from five breeds: Aberdeen Angus, Charolais, Hereford, Limousin, and Simmental. We phased the genotypes in sliding windows along the genome and used five tests to identify haplotypes with absence of or reduced homozygosity. Then, we associated the identified haplotypes with 44,351 insemination records that indicated early embryonic death, and postnatal survival records. Finally, we assessed haplotype pleiotropy by estimating substitution effects on estimates of breeding value for 15 economically important traits in beef production. Results We found support for one haplotype that carries a putatively recessive lethal (chromosome 16 in Simmental) and two haplotypes that carry semi-lethal alleles (chromosome 14 in Aberdeen Angus and chromosome 19 in Charolais), with population frequencies of 8.8, 15.2, and 14.4%, respectively. These three haplotypes showed pleiotropic effects on economically important traits for beef production. Their allele substitution effects are €2.30, €3.42, and €1.47 for the terminal index and €1.03, − €3.11, and − €0.88 for the replacement index, where the standard deviations for the terminal index are €22.52, €18.65, and €22.70 and for the replacement index they are €31.35, €29.82, and €35.79. We identified ZFAT as the candidate gene for semi-lethality in Aberdeen Angus, several candidate genes for the lethal Simmental haplotype, and no candidate genes for the semi-lethal Charolais haplotype. Conclusions We analysed genotype, reproduction, survival, and production data to detect haplotypes that carry putatively recessive lethal or semi-lethal alleles in Irish beef cattle and identified one lethal and two semi-lethal haplotypes, which have pleiotropic effects on economically important traits in beef production. Electronic supplementary material The online version of this article (10.1186/s12711-019-0452-z) contains supplementary material, which is available to authorized users.

Background Reproduction inefficiency has a major negative effect on profitability in cattle farming [1][2][3][4]. Over the last few decades, reproduction has deteriorated in both dairy and beef cattle in many countries, due to strong selection on production traits and their antagonistic relationship with reproduction [5,6], and probably also to inbreeding. In recent years, this trend has begun to be readdressed in many countries by emphasising reproduction traits in animal recording and breeding goals [7,8]. Furthermore, the advent of affordable high-density genotyping has enabled genomic selection jointly for production and reproduction traits [9,10] and screening for recessive lethal alleles [11][12][13][14][15].
Recessive lethal alleles are one of the genetic causes of reproduction inefficiency, because they result in embryonic death when an embryo is homozygous. They can also reduce viability of embryos and liveborn animals when their effect is not fully expressed (incomplete penetrance). Death is an easy phenotype to score but reduced viability is not as easy to identify. However, in some cases, it is observed as reduced conception rates, longer calving intervals, or lower survival for live born animals. Such alleles tend to be rare, but if their frequency increases, the number of affected (homozygous) embryos increases and so does the economic loss. Recessive lethal alleles may increase in frequency in a population because of genetic drift, linkage with favourable alleles, and heterozygote advantage. For example, a recessive lethal allele shows heterozygote advantage if it has a positive pleiotropic effect on production traits thereby conferring an advantage to heterozygous carriers, but not to affected (homozygous) embryos [16]. This would lead to the selection of heterozygous carriers as parents of the next generation, propagate the recessive lethal allele across the population, and in turn significantly affect reproduction [17].
The availability of high-density genotype data has enabled screening of livestock populations for recessive lethal alleles. There are reported successes in dairy cattle [11][12][13][14], beef cattle [15], and pigs [18,19]. The screening is based on the detection of the absence of homozygotes, either in whole populations or within carrier families. Several recessive haplotypes that show pleiotropic effects on yield, longevity and fertility in dairy cattle have been reported [20].
Recently, the Irish Cattle Breeding Federation (ICBF) has initiated large-scale genotyping of commercial beef cattle in Ireland [21]. To date, more than 1.3 million purebred and crossbred animals have been genotyped. These data will enable genomic selection to increase the productivity and reduce the environmental footprint of Irish cattle [22]. It can also be a powerful resource for biological discovery, including detection of recessive lethal or semi-lethal alleles, should they exist.
The objectives of our study were to discover haplotypes that carry putatively recessive lethal or semi-lethal alleles in the most numerous Irish beef cattle breeds, and to quantify their pleiotropic effects on traits under selection. We phased the genotypes in sliding windows along the genome for 132,725 purebred animals from five breeds and used five tests to identify haplotypes with absence or reduced homozygosity. We corroborated the identified haplotypes with reproduction and postnatal survival records. We found support for one haplotype that carries a putatively recessive lethal allele and two haplotypes that carry putatively recessive semi-lethal alleles. Substitution analysis showed that these haplotypes had pleiotropic effects on economically important traits in Irish beef cattle.

Methods
In this paper, we analysed genotype, reproduction, survival, and production data to identify and validate haplotypes that carry putatively recessive lethal or semi-lethal alleles in Irish beef cattle. Briefly, our approach consisted of: (1) Genotype filtering, imputation, and phasing, (2) Identification of haplotypes that carry putatively recessive lethal or semi-lethal alleles, (3) Reproduction and survival analysis to corroborate the identified haplotypes, (4) Pleiotropy analysis to estimate haplotype effects on production traits, (5) Identification of candidate genes.

Genotype filtering, imputation, and phasing
We used single nucleotide polymorphism (SNP) array data from the ICBF for purebred Aberdeen Angus, Charolais, Hereford, Limousin, and Simmental animals. The animals were genotyped with four different, but overlapping, Illumina SNP arrays (HD, IDBv1, IDBv2, and IDBv3; see Table 1), with most of the animals genotyped on the IDBv3 array (> 64%). We filtered the genotype data on autosomal chromosomes, missing information, and heterozygosity. We excluded SNPs with a low genotype call rate (< 0.90), and animals with a low genotype call rate (< 0.90) or high heterozygosity (> 6 standard deviations from mean heterozygosity). After filtering, 132,725 animals were available: 22 [23], and then they were phased with AlphaPhase version 1.9.1 [23] (using a core and tail length of 320 SNPs) into 20 SNP haplotypes in sliding windows. This means that the genotypes at each locus were phased 20 times in sliding windows, with the start of each window being moved along by one SNP at a time.

Identification of haplotypes that carry putatively recessive lethal or semi-lethal alleles
To identify haplotypes that carry putatively recessive lethal or semi-lethal alleles, the phased haplotypes from 132,725 animals were analysed for complete absence or a reduced level of homozygosity within the analysed breed (see Table 1). We used five different tests that were grouped into three categories: (i) population, (ii) carrier mating, and (iii) semi-lethality. The first and second categories are based on the probability of observing complete absence of recessive homozygous animals in the whole population or in the progeny from carrier matings. Thus, they assume that recessive lethal alleles are completely penetrant. The third category tests the null hypothesis that observed and expected numbers of recessive homozygous animals are equal. Thus, it is able to detect recessive semi-lethal alleles, with some homozygotes not suffering from deleterious effects or some haplotypes that imperfectly tag a recessive lethal allele. From this point on, we will use HH to denote wild type homozygotes, Hh to denote heterozygotes, and hh to denote homozygotes for a haplotype that carries putatively recessive lethal alleles.

Population test
The first test identifies haplotypes that carry putatively recessive lethal alleles based on their population frequency. The probability of observing no hh animals (P hh ) depends on the Hh frequency (c) and the number of genotyped animals (N): P hh = (1 − c 2 /4) N . The number of expected hh animals (E hh ) under these conditions is equal to: E hh = (N/4)·c 2 . This test is the same as that used by VanRaden et al. [12].

Carrier mating tests
The second category consists of two tests that identify haplotypes that carry putatively recessive lethal alleles based on the expected number of hh progeny from Hh × Hh matings. We analysed both sire × dam and sire × maternal grand-sire Hh × Hh matings. The probability of observing no hh animals follows a Bernoulli process and can be calculated as P hh = 0.75 C for C sire × dam Hh × Hh matings and P hh = 0.875 C for C sire × maternal grand-sire Hh × Hh matings. The expected number of hh progeny was calculated as C/4 for sire × dam Hh × Hh matings and C/8 for sire × maternal grand-sire Hh × Hh matings. These tests are partially the same as that used by VanRaden et al. [12].

Semi-lethality tests
The third category tests the null hypothesis that the observed and expected number of hh animals do not differ based on sire × dam Hh × Hh or sire × maternal grandsire Hh × Hh matings. From the number of observed and expected hh (O hh and E hh ), Hh (O Hh and E Hh ), and HH (O HH and E HH ) progeny, we calculated a one-way Chi square test statistic: The test statistic has one degree of freedom.

Identification of haplotypes that carry putative recessive lethal or semi-lethal alleles
Haplotypes that show absence or a reduced level of homozygosity at p < 5 × 10 −8 (−log 10 p > 7.30) were considered as candidates for carrying putatively recessive lethal or semi-lethal alleles that cause early embryonic death or reduced postnatal survival. In order to decrease the false discovery rate, only 1 Mb long regions with at least nine haplotype tests with p < 5 × 10 −8 were considered. From each of these regions, only the most significant haplotype test from each of the three test categories was selected. We analysed the effect of each of these haplotypes on cow reproduction and progeny survival to corroborate lethality and on production traits to evaluate pleiotropy.

Reproduction and survival analysis to corroborate the identified haplotypes
We used routine cow reproduction and progeny survival records from the ICBF database. We tracked 44,351 insemination records that pertain to purebred Aberdeen Angus, Charolais, Hereford, Limousin, or Simmental cattle, for which both the cow and the AI sire that it is mated to, were genotyped. There were 7540 records for Aberdeen Angus, 11,520 for Charolais, 5566 for Hereford, 15,404 for Limousin, and 4321 for Simmental. Altogether these inseminations resulted in 21,196 progeny. Each resulting calf had records for its date of birth and, if deceased, its date of death and the cause of death (natural death or slaughtered). All the new born calves were followed from birth until November 2017. If they were still alive at the end of data collection their records were treated as censored.
To quantify the effect of identified haplotypes on reproduction, we determined the insemination success rate based on the number of days between a recorded insemination and the subsequent calving date. Insemination success rate was calculated as a ratio between the number of born progeny and the number of inseminations. These statistics were reported for the three mating types: (1) Hh × Hh, (2) Hh × HH, and (3) HH × HH. The interval between insemination and calving was calculated from 40,769 records for which the date of the corresponding calving was known. In order to accommodate shorter or longer than average gestation length, a window of 7, 14, or 21 days was allowed on either side of the breed average gestation length, calculated as the average gestation length for the set of individuals recorded for that breed. If a calving occurred outside of the window, the insemination was deemed to be unsuccessful. If a haplotype harbours recessive lethal alleles, Hh × Hh matings are expected to have a 25% lower insemination success since there will be 25% hh progeny and an interval between insemination and calving longer than 21 days, which is the average time between each standing oestrus in cattle.
To quantify the effect of identified haplotypes on postnatal survival, we used survival analysis with the Cox proportional-hazard model. We compared hazard ratios between the progeny from: (1) Hh × Hh, (2) Hh × HH, and (3) HH × HH matings. If a haplotype affects postnatal survival, we would expect hh progeny from Hh × Hh matings to have a higher hazard ratio. Since two causes of death were reported (natural death and slaughtered), we analysed data in two ways: (1) we treated records from slaughtered animals as censored, and (2) we treated their records as uncensored.

Economic impact
For the estimation of economic impact of putative lethal or semi-lethal haplotypes on the whole Irish beef population, we built a simple economic model. This model was based on the number of purebred sire × purebred dam, purebred sire × crossbred dam, and crossbred sire × crossbred dam of Aberdeen Angus, Charolais, and Simmental animals, the expected number of progeny genotypes, haplotype frequencies, insemination success rate, number of days between the subsequent inseminations, daily costs for maintaining a cow, and substitution effects. We obtained these inputs from this study, the ICBF database, and the E-profit monitor-an online financial analysis tool [25].

Candidate genes
We used Bioconductor version 3.7 with the BioMart database version Ensembl Genes 92 using the btaurus_ gene_ensembl to find candidate genes within the regions of the identified haplotypes [26]. We used the UMD3.1 cattle reference genome [27]. We extracted the protein coding genes and searched for lethal effects observed with homologous genes with the same function in the Mouse Genome Informatics database [28].

Results
Our results identified one haplotype that carries a putatively recessive lethal allele and two haplotypes that carry putatively recessive semi-lethal alleles, which were located on chromosome 16 in Simmental, chromosome 14 in Aberdeen Angus, and chromosome 19 in Charolais, respectively. These haplotypes showed pleiotropic effects on economically important beef production traits. We identified the zinc finger and AT-hook domain containing (ZFAT) gene as the potential causative gene for semilethality in Aberdeen Angus.

Identification of haplotypes that carry putative recessive lethal or semi-lethal alleles Haplotype tests
In total, there were between 36 and 109 haplotypes with significant absence or reduced levels of homozygosity at p < 5 × 10 −8 in the five breeds. Table 2 shows the number of significant haplotypes by test and breed. The largest number of significant haplotypes was observed for Limousin (109) followed by Charolais (86), Aberdeen Angus (69), Hereford (57), and Simmental (36). Among the different tests, the semi-lethality test on sire × dam matings had the largest number of significant haplotypes. The carrier mating test on sire × dam matings had the second largest number of significant haplotypes for Charolais, Hereford, and Simmental.

Identification of haplotypes that carry putative recessive lethal or semi-lethal alleles
Across all five breeds, there were 19 one-Mb long regions with at least nine haplotypes that showed significant absence or a reduced level of homozygosity. These were considered as candidate regions with haplotypes that carry putatively recessive lethal or semi-lethal alleles. Figure 1 shows the negative logarithm of all test p-values across the whole genome and the 19 candidate regions are listed in Additional file 1: Table S1, with five regions for Aberdeen Angus, two for Charolais, three for Hereford and Limousin, and six for Simmental. Some of the regions were in close proximity and appeared as a single region (see Fig. 1). This was the case for three regions on chromosome 14 in Aberdeen Angus, two regions on chromosome 6 in Hereford, two regions on chromosome 23 in Limousin, and three regions on chromosome 13 and two regions on chromosome 16 in Simmental. The region between 48 and 49 Mb on chromosome 19 was detected in all the breeds.
From the 19 candidate regions, we selected 22 haplotypes with the highest significance within each of the three category tests. Table 3 lists the 22 haplotypes and their characteristics. We named the haplotypes by using a code including the breed (AA for Aberdeen Angus, CH for Charolais, HE for Hereford, LI for Limousin and SI for Simmental), the chromosome, and the haplotype number within the breed (H for haplotype). Except for HE19H3 and SI19H6, which were identified with tests from two categories (carrier mating test with sire × dam matings from the carrier mating tests and semi-lethality test with sire × dam matings from the semi-lethality tests), all haplotypes were identified by tests from a single category.

Reproduction and survival analysis to corroborate the identified haplotypes Reproduction analysis
Two out of the 22 identified haplotypes, AA14H3 and SI16H5, were associated with decreased insemination success rate and a longer interval between insemination and calving. Table 4 shows the insemination success rate and the length of the interval between   20.2% lower (AA14H3), and 48.6, 52.5, and 52.7% lower (SI16H5) compared to the HH × HH matings for windows of 7, 14, and 21 days on either side of the average gestation length for each breed, respectively. The insemination success rates were also lower in HH × Hh matings than in HH × HH matings, but the differences were smaller. For both haplotypes (AA14H3 and SI16H5) the interval between insemination and calving was longer for Hh × Hh matings than for HH × Hh matings, which was in turn longer than for HH × HH matings. The interval between insemination and calving for Hh × Hh matings was 30 days longer than for HH × HH matings for both AA14H3 and SI16H5 haplotypes, and that for HH × Hh matings was longer than for HH × HH matings by 4 days for AA14H3 and by 9 days for SI16H5.

Survival analysis
One of the identified haplotypes, CH19H2 was associated with a significantly (p < 0.05) increased hazard ratio for postnatal survival (i.e. decreased survival) of progeny from Hh × Hh matings. CH19H2 is located on chromosome 19 (48,874,999-48,902,319 bp) in Charolais. Additional file 2: Table S2 presents the hazard ratios for progeny from HH × Hh and Hh × Hh matings compared to those for progeny from HH × HH matings, for which records for culled progeny were treated as censored or complete. Progeny of individuals that carried the CH19H2 haplotype had a 36% higher probability of dying or being slaughtered during their life compared to the progeny of HH × HH parents.

Frequencies for haplotypes that carry putatively recessive lethal or semi-lethal alleles
All three haplotypes that carry putatively recessive lethal or semi-lethal alleles associated with decreased insemination success or reduced postnatal survival had high frequencies. Table 5 summarizes the statistics for the AA14A3, CH19A2, and SI16A5 haplotypes. Haplotype frequencies were 15.2% for AA14A3, 14.4% for CH19A2, and 8.8% for SI16A5. Ninety-five recessive homozygous individuals were observed for AA14A3, 83 for CH19A2, and none for SI16A5. This suggests that the SI16A5 haplotype carries a putatively recessive lethal allele, whereas AA14A3 and CH19A2 haplotypes carry putatively recessive semi-lethal alleles. Table 5 shows that the number of observed recessive homozygotes was always smaller than the expected number for different types of comparisons. SNP alleles present on the AA14A3, CH19A2, and SI16A5 haplotypes are in Additional file 3: Table S3.

Pleiotropy analysis
To investigate potential pleiotropic effects on traits under selection, we estimated the substitution effects of the AA14H3, CH19H2, and SI16H5 haplotypes. Figure 2 shows standardized substitution effects, presented in the units of a trait standard deviation, for 15 traits that are part of the terminal or replacement index. A significant substitution effect for the AA14H3 haplotype was Probability of not observing recessive homozygotes with the T3 test a 6.7 × 10 −9 9.0 × 10 −9 3.

Economic impact
The expected economic impact estimated with a simple economic model is presented in Additional file 4: Table S4. The annual losses due to decreased viability and decreased survival for each haplotype were estimated to reach ~ €570,000 for AA14H3, ~ €930,000 for CH19H2, and ~ €210,000 for SI16H5. The annual gains due to positive associations with economically important traits were estimated to reach ~ €110,000 for AA14H3, ~ €70,000 for CH19H2, and ~ €50,000 for SI16H5. Under the current population structure in Ireland, the estimated national annual net effects were negative: − €450,000 for AA14H3 − €860,000 for CH19H2, and − €160,000 for SI16H5. The large negative net effect of the CH19H2 haplotype is due to the larger purebred population in Charolais than in Aberdeen Angus and Simmental. The economic model shows that, under the current population structure, the maximum annual net effects would be achieved with haplotype frequencies of 0.02% for AA14H3 (+€7,000), 0.01% for CH19H2 (+€2,000), and SI16H5 (+€4,000) [see Additional file 5: Figure S1].

Candidate genes
We identified one protein coding gene in the region of AA14H3, none in the region of CH19H2, and several in the region of SI16H5. Zinc finger and AT-hook domain containing (ZFAT) is the only protein coding gene located between 8,064,004 and 8,927,881 bp on chromosome 14 of the AA14A3 haplotype. ZFAT is associated with prenatal or perinatal lethality in the Mouse Informatics Database. Additional file 6: Table S5 lists the 64 protein coding genes located between 51,811,400 and 53,434,159 bp on chromosome 16 of the SI16H5 haplotype, among which five, i.e. SKI proto-oncogene (SKI), G protein subunit beta 1 (GNB1), ATPase family AAA domain-containing protein 3 (ATAD3A), zinc finger and BTB domain containing 17 (ZBTB17), and caspase-9 (CASP9)) are associated with prenatal or perinatal lethality in mice.

Discussion
Our results reveal one haplotype that carries a putatively recessive lethal allele (SI16A5) and two haplotypes that carry putatively recessive semi-lethal alleles (AA14A3 and CH19A2) that are candidates for lethality or semilethality in homozygous state. These three haplotypes were identified by the haplotype analysis of 132,725 purebred animals of five beef cattle breeds from the ICBF database. The results also suggest pleiotropy, since these haplotypes include or are in strong linkage with variants that have substantial effects on economically important traits. In the light of these results, we discuss (1) the identification of haplotypes that carry putative recessive lethal and semi-lethal alleles, (2) the frequency of the identified haplotypes, (3) their economic impact, and (4) their potential genetic causes.

Identification of haplotypes that carry putative recessive lethal or semi-lethal alleles
We used five tests on sliding haplotypes and phenotypic information to search for haplotypes that carry recessive lethal or semi-lethal alleles. We used haplotypes in a sliding window of 20 SNPs, which was previously shown to be a good compromise between haplotype diversity and detection of putative recessive lethal alleles with a moderate SNP density array [15]. The tests were based on the number of expected recessive homozygous individuals either in the whole population or from matings between carriers. Along with the tests for complete absence of recessive homozygotes, we also used a set of tests that allowed for some recessive homozygotes but tested whether their number was significantly smaller than expected. This allowed us to take possible incomplete penetrance, incomplete linkage between haplotype and putative recessive lethal alleles, structural variation, or genotyping, phasing, and imputation errors into account. In the following, we consider that all are semi-lethal.
The top 22 candidates from the haplotype analysis were further corroborated with cow insemination and progeny survival analysis, and three haplotypes showed phenotypic evidence of lethality. Recessive lethal alleles are expected to decrease insemination success and extend the interval between insemination and calving. In the case of semi-lethality, some embryos are expected to survive, but these may have a decreased viability that will in turn affect their survival.
Although there were many haplotypes with a significant absence or reduced level of homozygosity, which indicated that they carried putatively recessive lethal or semi-lethal alleles, only two haplotypes were supported by the reproduction analysis. In particular, there was a strong signal observed at the end of chromosome 19 for all five cattle breeds, which was supported by phenotypic data only in the Charolais breed. A highly significant departure from Hardy-Weinberg equilibrium was observed for SNPs that were located at the end of chromosome 19 with an excessive heterozygosity level (results not shown). This may indicate genotyping errors or structural variation in this region [29] that could explain the strong signal observed.

Frequency of haplotypes that carry putatively recessive lethal or semi-lethal alleles
For our methods to detect a haplotype that carries putatively recessive lethal or semi-lethal alleles, we needed a relatively high haplotype frequency or many Hh × Hh matings. Because most haplotypes are rare, this means that the methods we used had sufficient power for a very small proportion of the haplotypes. Following the probability calculations of the different tests and fixing the number of genotyped animals to the number available for this study, we were able to assess the minimum required haplotype frequency with a significance of p < 5 × 10 −8 , i.e. 5.42% for Aberdeen Angus, 4.14% for Charolais, 7.29% for Hereford, 3.83% for Limousin, and 7.56% for Simmental. The percentages of haplotypes with a frequency higher than these thresholds were very low: 0.98% for Aberdeen Angus, 0.68% for Charolais, 1.01% for Hereford, 0.56% for Limousin, and 0.52% for Simmental. Similarly, reaching such significance with the carrier mating test would require at least 59 sire × dam carrier (Hh × Hh) matings and no homozygous progeny. The percentages of haplotypes reaching this threshold were even smaller than for the population test: 0.44% for Aberdeen Angus, 0.23% for Charolais, 0.37% for Hereford, 0.20% for Limousin, and 0.16% for Simmental. Furthermore, reaching significance with the carrier mating test would require 126 sire × maternal grand-sire carrier (Hh × Hh) matings and no homozygous progeny. The percentage of haplotypes reaching this threshold were even smaller than for the carrier mating test with sire × dam matings, i.e. 0.06% for Aberdeen Angus, 0.04% for Charolais and Hereford, 0.03% for Limousin, and 0.02% for Simmental. Finally, reaching significance with the semi-lethality test required that the Chi square test statistic be higher than 29.72. The percentages of haplotypes that reached this threshold in the current study was less than 0.01% for all the breeds.
Since recessive lethal and semi-lethal alleles are usually rare and breed-specific, they are difficult to detect in a population, in particular for recent mutations since modern breeding programs try to avoid inbreeding to preserve genetic variation and maintain fitness. Thus, breeders are likely to indirectly avoid carrier matings because they avoid matings between close relatives. Still, there are some alleles that are propagated through the population via popular sires, have pleiotropic effects, or are in incomplete linkage with an allele that has a positive effect on a trait that was selected for in the past. A classic example of a recessive lethal allele propagated with a popular sire is the complex vertebral malformation (CVM) of calves in the Holstein breed. The causal mutation was propagated by the bull Carlin-M Ivanhoe Bell that was incidentally also a carrier of the bovine leukocyte adhesion deficiency (BLAD) [30]. An example of a pleiotropic recessive lethal allele is a 660-kb deletion that causes embryonic death in Nordic Red cattle [14]. Since this deletion is positively associated with milk yield, it is quite common (13-32%) in the Nordic Red cattle population. Yet another example is the Weaver syndrome which is expressed postnatally between the age of 5 and 8 months [31], with Weaver-carriers also producing more milk and fat, which have been strongly selected for in the past [31,32].
High haplotype frequencies and substitution effects on traits under selection suggest that the frequency of the detected haplotypes in this study may have increased because of pleiotropy or linked selection. The observed frequencies are high compared to expected frequencies with mutation-selection equilibrium. We can use Nei's [33] formula to approximate the expectation and variance of the frequency of a recessive lethal allele in a finite population, where µ is the mutation rate, N e the effective population size, and s the selection coefficient against the mutant lethal: With a high mutation rate of µ = 10 −5 and an N e similar to that of some beef cattle populations of 300, e.g. [34,35], the expected equilibrium frequency is very low, regardless of whether the lethal allele is fully (s = 1) or partially penetrant (s = 0.75): q ≈ 0.0004 and σ ≈ 0.0031 for a fully penetrant lethal allele, and q ≈ 0.0005 and σ ≈ 0.0036 for a partially penetrant lethal allele. This analysis clearly shows that there are other forces, i.e., selection, that maintain the high frequency of the three haplotypes.

Economic impact
Analysis of the economic impact of the three haplotypes on the whole Irish beef population supports the role of pleiotropy in maintaining high haplotype frequencies, which should be managed with appropriately optimised breeding programs. The economic impact could be increased if haplotype carriers are identified and mating between carriers is avoided. This is relatively easy to achieve when there is only one or a few haplotypes for which carrier matings are to be avoided. However, if there are many lethal haplotypes to select against, it will be much more difficult to achieve within pure breeds. Crossbreeding could overcome this issue for breed-specific haplotypes.

Genetic causes of lethality and pleiotropy
We found one candidate gene for one haplotype that carries a putatively recessive lethal allele, no candidate genes for another, and several for the third one. ZFAT was the only protein coding gene that overlapped with the AA14H3 haplotype. ZFAT is a transcription factor involved in immune-regulation and apoptosis [36,37] and plays a role in the development of the hematopoietic system [38]. The Mouse Genome Informatics database reported complete early embryonic lethality for a knock-out allele of ZFAT. ZFAT is also known to be associated with human height [39][40][41] and body size in horses [42]. This is consistent with our results, since we observed an insemination success rate that was 20.2 to 24.2% lower (based on different gestation window size) for the AA14H3 Hh × Hh matings than for the HH × HH matings, and the haplotype was positively associated with weight-related traits and feed intake. Sixty-four protein coding genes overlapped with the SI16H5 haplotype, of which five SKI, GNB1, ATAD3A ZBTB17, and CASP9 were associated with prenatal or perinatal lethality in the Mouse Genome Informatics database [43][44][45][46][47]. This haplotype displays a positive association with feed conversion ratio, which may explain its high frequency in the population.
There were no homozygous individuals for the SI16H5 haplotype, which suggests that the identified haplotype tags the putative recessive lethal allele(s). The existence of homozygous individuals for the AA14H3 and CH19H2 haplotypes suggests that there may be multiple underlying haplotypes with and without lethal allele(s), but we are unable to distinguish them with the current genotyping data. Another possible reason is incomplete penetrance, resulting in some homozygotes not suffering from the deleterious effects of the allele.
Fine-mapping these haplotypes with sequence data could potentially resolve the fine haplotype structure, and identify causal variants for lethality and pleiotropic effects on the selected traits. Efforts to identify such causative variants are ongoing.

Conclusions
We identified three haplotypes, one that carries a putatively recessive lethal allele in Aberdeen Angus and two that carry semi-lethal alleles, one in Charolais and one in Simmental. Carrier matings in Aberdeen Angus and Simmental had a lower success of artificial insemination and a longer interval between insemination and calving. Progeny from carrier matings in Charolais had a higher probability of dying or culling. The haplotype in Aberdeen Angus was in a region that contains the ZFAT gene, which was previously shown to be lethal in mice. The haplotype in Simmental was in a region that contains several candidate genes that potentially cause prenatal or perinatal lethality. Substitution analysis indicated that the haplotypes had pleiotropic effects on the traits under selection. The substitution effects were positive for the terminal index and positive to negative for the replacement index. Further work is needed to identify the underlying causative variants that impact the traits under selection and fitness. Identification of haplotype carriers will improve the breeding success and fitness of beef cattle populations.