Whole-genome resequencing reveals candidate mutations for pig prolificacy

Abstract

Changes in pig fertility have occurred as a result of domestication, but are not understood at the level of genetic variation. To identify variations potentially responsible for prolificacy, we sequenced the genomes of the highly prolific Taihu pig breed and four control breeds. Genes involved in embryogenesis and morphogenesis were targeted in the Taihu pig, consistent with the morphological differences observed between the Taihu pig and others during pregnancy. Additionally, excessive functional non-coding mutations have been specifically fixed or nearly fixed in the Taihu pig. We focused attention on an oestrogen response element (ERE) within the first intron of the bone morphogenetic protein receptor type-1B gene (BMPR1B) that overlaps with a known quantitative trait locus (QTL) for pig fecundity. Using 242 pigs from 30 different breeds, we confirmed that the genotype of the ERE was nearly fixed in the Taihu pig. ERE function was assessed by luciferase assays, examination of histological sections, chromatin immunoprecipitation, quantitative polymerase chain reactions, and western blots. The results suggest that the ERE may control pig prolificacy via the cis-regulation of BMPR1B expression. This study provides new insight into changes in reproductive performance and highlights the role of non-coding mutations in generating phenotypic diversity between breeds.

1. Background

Breed diversity has been shaped by a long history of domestication and selection [1]. The germplasm characteristics of individual breeds, including obvious phenotypic and reproductive performance differences, were established during this process. The pig, an important source of animal protein, was domesticated approximately 10 000 years ago [2–4]. Significant differences among pig breeds have been demonstrated in behaviour, fertility, growth and local adaptation [1,5]. While this genetic diversity makes it possible to uncover the causes of phenotypic variation, only a small number of loci, genes, and mutations responsible for important traits, such as lean growth [6–8], vertebrae numbers [9,10], high-altitude adaptation [11], and temperature adaptation[12], have been identified.

Due to the low heritability of pig reproductive traits [13], no causal genes or mutations affecting litter size are known, although many studies have been conducted using quantitative genetic analysis in cross gilts [14–16] or by polymorphism scans of candidate genes in diverse pig populations [17–22]. In sheep, an autosomal mutation in the Booroola fecundity gene FecB increases ovulation rate and litter size [23,24]. The association between prolificacy and the mutation, which occurs within the gene encoding the bone morphogenetic protein (BMP) receptor 1B (BMPR1B), has been confirmed in a transgenic model [25]. The first quantitative trait locus (QTL) associated with litter size in pigs was the oestrogen receptor (ESR) locus [21]. The follicle-stimulating hormone beta subunit gene has also been considered a major candidate gene controlling litter size [26]. Studies focused on reproductive traits have identified additional QTLs and candidate genes in different pig breeds [14,15,27,28].

Many studies have focused on mutations in coding regions that underlie Mendelian and common complex traits [29]. However, even a single nucleotide mutation within a noncoding region, such as a regulatory element, can result in a modified phenotype [30], and the majority of transcription factors regulate the expression of target genes by binding to a specific conserved motif [31]. In order to identify possible causal mutations for pig prolificacy, including those in non-coding regions, we performed a genome-wide and single-base comparison between highly and relatively less prolific breeds. Our strategy leveraged the annotation resources of the human genome and incorporated functional experiments. The results suggest that changes in regulation play an essential role in the prolific phenotype, and demonstrate that an integrated genomics approach improves the accuracy of the genetic dissection of complex traits.

2. Results

(a). Identification of genetic variation among pig populations

The Chinese Taihu pig group, which includes the Meishan, Erhualian, Mi, Shawutou, Jiaxing Black and Fengjing breeds [32], comprises the most prolific pigs in the world, producing approximately four more piglets per litter than American or European commercial breeds [33]. Taihu pigs and four other breeds from Asia or Europe can be clustered into four groups by neighbour joining according to prolificacy and geographical differences (electronic supplementary material, table S1 and figure S1). Equal amounts of genomic DNA from each group were pooled and sequenced using high throughput technology, providing about 10X coverage of the genome. For each population, more than 240 million raw reads were obtained and the average read depth was more than 10.15X. After filtering, 236 million clean reads remained. Of these, 93.96–95.98% reads were mapped to the reference genome (Sscrofa10.2). Approximately 13 million SNPs were identified.

(b). Selective sweeps specific to Taihu pigs

Under selection, beneficial variants and surrounding neutral variants will increase in frequency in a population (selective sweeps), resulting in low heterozygosity in the population and increased differentiation around the beneficial sites, as compared to non-selected populations. For each SNP, we calculated the pooled heterozygosity of the Taihu group (Het), and three fixation index (F_ST) values between Taihu and the other three pig groups across the genome using 100 kb sliding windows with 50 kb overlap. Because the main goal was to identify the unique genetic background of Taihu pigs relative to the other three groups, the minimum value of the three F_ST values (minF_ST) was used to represent genetic differentiation between the Taihu pigs and the others. Of the three pairwise comparisons, Taihu versus Asian-control (a sample combining the Diannan small ear and Tibetan breeds), accounted for 94.78% of the lowest Fst values, as these pigs are the most related to each other and are therefore less differentiated. The analysis yielded 43 Het - minF_ST outliers with minF_ST > 0.35 and Het < 0.02 [12,34,35]. After merging sweeps separated by less than 1 Mb, 39 genomic regions were identified as candidate selective sweeps specific to Taihu pigs (electronic supplementary material, table S2 and figure 1).

Figure 1.

Selective sweeps in the Taihu pig. (a) Het-minF_ST outliers are within the red box. (b) Manhattan plot for minF_ST values obtained from a sliding window analysis of the pig genome. (c) Selective sweeps that overlap BMPR1B (left) and BMP15 (right). Het: heterozygosity, minF_ST: minimum F_ST values calculated by comparing the Taihu pig to each of three control groups.

After examining genes that overlapped with the 39 putative selective sweeps, we identified genes potentially associated with reproductive traits (electronic supplementary material, table S2). A comparison between the selective sweeps and the Animal QTL Database [36] revealed 32 selective sweeps that overlapped with one or more QTLs related to reproductive traits. The selective sweeps identified in Taihu pig had strong bootstrap support and was overlapped with a list of QTLs for reproductive traits. Twenty-seven sweeps overlapped with QTLs related to teat number, and 11 sweeps included QTLs that were related to total number born (three sweeps on chromosome 6 and eight sweeps on chromosome 8). Among these 11, one sweep (chr8:134.05–134.15 Mb) harboured a QTL peak (123.3–147.4 Mb) with effects on several pig reproductive traits (total born alive, litter size and prenatal survival), located approximately 1.28 Mb from the QTL peak marker (SW1551) [14,15]. This sweep also overlapped with bone morphogenetic protein receptor type-1B (BMPR1B), a major fecundity gene first identified in prolific Booroola Merino sheep [23,37,38]. Another fecundity gene (FecX/bone morphogenetic protein 15 (BMP15)) identified in sheep [39–41] also overlapped with a selective sweep (chrX:49–51.55 Mb). The two sweeps containing BMPR1B and BMP15 showed high allele frequency differences and low heterozygosity in Taihu pigs (figure 1).

Besides coding genes, we identified several non-coding genes that overlapped with selective sweeps. One sweep (chr3:86.4–87.9 Mb) occurred at an extremely conserved syntenic locus across vertebrates (electronic supplementary material, figure S2); this locus contains lincRNAs (LINC01122 and LOC101927285 in human), and is predominantly expressed in brain and reproductive tissues [42]. We conjecture that this region may harbour lincRNAs that affect reproductive traits across vertebrates. Another example is selective sweep chrX:34.6–35 Mb which contains 7SK, a non-coding snRNA that regulates gene transcription in the murine Oct4-GiP embryonic stem cell line [43].

(c). Positive selection signal at non-coding sites

Since highly differentiated SNPs are likely to be the direct targets of selection, or to occur near loci under selection [35], we searched for Taihu-specific SNPs across the entire genome. To reduce biased estimates of allele differences caused by sequencing coverage variations, SNPs with low sequencing depths in any one population were set aside. For the common (or major) allele of each SNP, the frequency difference was computed between Taihu pigs and each of other three groups (ΔAF). As described above for minF_ST, the minimum value (minΔAF) of three ΔAF values was used to represent the genetic difference between Taihu pigs and the other breeds; 93.62% of minΔAF was contributed by the comparison between Taihu and the Asian-control. Approximately 132 000 Taihu-specific SNPs with minΔAF ≥ 0.75 were obtained (approx. 1% of the total) (table 1); these common alleles were considered as fixed or nearly fixed in Taihu pigs. To refine the analysis, SNPs with minF_ST > 0.35 were selected, yielding 810 sites. We compared these to the 810 sites with the highest minΔAF value; 83.06% of the members of these two sets overlapped.

Table 1.

Fisher's exact test for regions under positive selection during breeding of the Taihu pig.

regions	total (SNPs)	Taihu-specific SNPs	specific fraction	Fisher estimate	p-value
genome	13 000 917	132 078	1.02%	—	—
genic	3 564 062	38 034	1.07%	1.072	1.92 × 10−29
intergenic	9 436 855	94 044	1.00%	0.933	1.92 × 10−29
exonic	82 844	686	0.82%	0.813	2.63 × 10−8
intronic	3 181 483	34 427	1.08%	1.089	2.94 × 10−41
conservative	494 033	5539	1.12%	1.109	1.29 × 10−13
cons-coding	47 840	432	0.90%	0.887	1.28 × 10−2
cons-non-coding	446 193	5107	1.14%	1.133	1.18 × 10−17
cons-intronic	103 109	1244	1.21%	1.192	2.54 × 10−9
cons-intergenic	322 863	3661	1.13%	1.121	2.79 × 10−11

Genomic location enrichment analysis indicated that a significant excess of Taihu-specific SNPs is in or near genes (including exons, introns, 5′ UTRs, 3′ UTRs, 1 kb upstream from the transcription start site, and 1 kb downstream from the transcription termination site) compared to intergenic regions (Fisher's exact test, p = 1.92 × 10⁻²⁹) (table 1). The data were then examined more closely to determine which sub-regions were targeted preferentially during the breeding process. A significant excess of Taihu-specific SNPs was found in introns (Fisher's exact test, p = 2.94 × 10⁻⁴¹), and a corresponding lack was noted in exons (Fisher's exact test, p = 2.63 × 10⁻⁸). This result could be due to strong purifying selection in exonic regions, or indicate strong positive selection in intronic regions. By aligning the pig and human genomes, we identified approximately 0.5 million conserved SNPs and found that Taihu-specific SNPs are enriched at these conserved sites (Fisher's exact test, p = 1.29 × 10⁻¹³). Again, a significant lack of SNPs (Fisher's exact test, p = 1.28 × 10⁻²) was observed at coding sites, and significant excess at conserved non-coding, intronic and intergenic sites (Fisher's exact test, all the p-values < 2.54 × 10⁻⁹). The excess of Taihu-specific SNPs in non-coding regions, particularly at conserved non-coding regions, suggests that non-coding mutations have played an important role in the breeding of the Taihu pig.

(d). Taihu-specific selection targets morphogenesis

To examine functional targets relevant to the breeding of the hyper-prolific Taihu pig, the 1000 genes having ≥10 Taihu-specific SNPs in gene regions (electronic supplementary material, table S3) were subjected to gene ontology (GO) overrepresentation analysis. GO terms associated with morphogenesis, such as developmental process, anatomical structure development, and anatomical structure morphogenesis, were significantly enriched (table 2 and electronic supplementary material, table S4; all p-values obtained using Bonferroni tests were less than 0.002). Major genes involved in developmental pathways were identified; these included the TGF-β/smad signalling pathway (e.g. BMPR1B, BMP6, BMP7, SMAD9 and BMPR2) and the Wnt/β-catenin signalling pathway (e.g. WNT7B, CDH12 and CSNK1E). The results indicate that positive selection in Taihu pigs acted primarily on genes affecting development, consistent with molecular and morphological studies that revealed differences in ovarian and uterine development in Taihu pigs early in pregnancy [44]. To assess whether the GO overrepresentation analysis might be biased due to differences in gene lengths, we divided the number of Taihu-specific SNPs in each gene region by its corresponding gene length (SNP number/gene length ≥ 10 SNPs/kb), and ranked them to obtain a new list of 1000 genes. Again, the GO analysis performed using these genes was enriched for GO terms associated with embryonic and fetal development (data not shown).

Table 2.

Selected overrepresented gene ontology (GO) terms. Complete list is in electronic supplementary material, table S4.

GO biological process	observed	expected	Bonferroni p-value
single-organism developmental process (GO:0044767)	210	141.6	1.46 × 10−5
developmental process (GO:0032502)	211	142.85	1.84 × 10−5
anatomical structure development (GO:0048856)	196	132.36	6.70 × 10−5
multicellular organismal development (GO:0007275)	176	118.7	4.26 × 10−4
anatomical structure morphogenesis (GO:0009653)	108	65.35	1.83 × 10−3
system development (GO:0048731)	160	108.52	2.72 × 10−3
cell development (GO:0048468)	83	47.63	6.54 × 10−3
neurogenesis (GO:0022008)	71	39.28	1.18 × 10−2
cell adhesion (GO:0007155)	56	28.48	1.35 × 10−2

(e). An intronic mutation of BMPR1B is a causal candidate for pig prolificacy

Eleven of the 39 candidate selective sweeps overlapped QTLs that affect the number of piglets born. Sorting by minF_st values from large to small, the sweep that included BMPR1B (chr8:134.05–134.15 Mb) ranked second. The ranked first sweep (chr8:70.15–70.55 Mb) was overlapped with the same QTL to BMPR1B, which was associated with reproductive traits. However, the sweep with the second largest value (chr8:134.05–134.15 Mb) not only included the first intron of BMPR1B, but it was also very close to the QTL peak for litter size on the long arm of chr8 [14,15]. We therefore focused our attention on this sweep.

BMPR1B has been confirmed as a major fecundity gene in sheep [23,37,38]. No coding mutations within BMPR1B exhibited differences in Taihu-specific allele frequencies (data not shown). However, co-located with the minF_ST peak (figure 1c) is a Taihu-specific haplotype (chr8:134 092 836–134 094 077; approximately 1.2 kb in length), where 18 of 19 SNPs show a marked difference in allele frequencies between Taihu and other breeds (figure 2). To confirm that the 1.2 kb region is unusual in Taihu pigs, we genotyped 10 SNPs by direct sequencing or restriction fragment length polymorphism (RFLP) analysis in 176 pigs from six different breeds (electronic supplementary material, figure S3). With the exception of a rare mutation (variant allele frequency < 0.05), all 10 SNPs showed distinct allele frequencies in the Taihu pig (figure 2 and electronic supplementary material, table S5). Published datasets from 29 pig breeds (data library ‘Genome Diversity’, uploaded by rico@bx.psu.edu and the European Nucleotide Archive (ENA); accession PRJEB9922) were used to confirm the 1.2 kb Taihu-specific haplotype [2] (electronic supplementary material, table S6). The haplotype completely resolves into Taihu and other breed types (electronic supplementary material, figure S4). A previous study showed that Jiangquhai pigs have the closest phylogenetic relationship to Taihu pigs (Erhualian and Meishan) [12]; Nevertheless, the Taihu-specific haplotype was only observed in one of three Jiangquhai pigs. When additional Jiangquhai pigs were examined, the dominant Jiangquhai haplotype was clearly not the Taihu-specific haplotype (data not shown). Two Meishan pigs presented a heterozygous genotype, suggesting that gene flow from foreign breeds (such as landrace or Duroc) might exist in this breed.

Figure 2.

Strong selection on BMPR1B. (a) A selective sweep (black bar) of pig chr8 falls within a BMPR1B intron. The peak minF_ST signal region is shown in red. (b) Alignment of peak region to its corresponding (homologous) sequences in the human genome. In the pig SNPs track, Taihu-specific SNPs are shown in red. An ESR1 binding site is located within the peak sweep region (grey bar). The binding site was predicted by ENCODE/HudsonAlpha/Analysis (http://genome.ucsc.edu/ENCODE/). The lower panel shows that the BMPR1B intronic region is conserved in mammals but absent in oviparous animals.

Genome alignment shows that the 1.2 kb sequence is conserved in mammals but absent in oviparous animals such as birds, amphibians and fishes (figure 2). Importantly, we found a conserved oestrogen response element (ERE) in the 1.2 kb conserved region, and four Taihu-specific SNPs are located in this ERE (designated hereafter as Taihu-ERE) (figure 2).

Since it is likely to regulate transcription of BMPR1B, we investigated whether the Taihu-ERE exhibits altered activation by the type I receptor of oestrogen (ESR1). Luciferase assays were performed to compare the Taihu-ERE with the version of the element obtained from Duroc pigs (Duroc-ERE). The Taihu-ERE generated significantly higher luciferase activity (figure 3a), suggesting that it may bind ESR1 more efficiently and elevate the expression of BMPR1B. We then examined BMPR1B expression (mRNA and protein) in the endometrium of Taihu and Duroc pigs 72 days post-pregnancy. Expression is significantly higher in Taihu than in Duroc pigs (figure 3b,c). The number of endometrial glands in the Taihu pig is also greater than in the Duroc pig (p = 9.18 × 10⁻⁸) (figure 3d–f), consistent with a finding that BMPR1B-deficient mice are defective in endometrial gland formation [25]. Finally, the in vitro binding efficiency of Taihu-ERE and Duroc-ERE were compared in a chromatin immunoprecipitation assay. Taihu-ERE showed higher binding activity with ESR1 than Duroc-ERE (figure 3g,h). Taken together, our results suggest that the Taihu-ERE mutation contributes to pig fecundity by elevating BMPR1B expression in endometrial cells to promote uterine development during pregnancy.

Figure 3.

Functional assays for BMPR1B. (a) Analysis of ESR1 binding to two ERE haplotypes. The y-axis shows the tests and controls. The x-axis shows the luciferase activity (Firefly) normalized to reference luciferase activity (Renilla). (b,c) Relative quantification of BMPR1B mRNA (b) and protein (c) expression in the endometrium of DR and TH pigs 72 days post-pregnancy. (d,e) Tissue section analysis of DR (d) and TH (e) endometrium 72 days post-pregnancy. Endometrial glands are denoted with black arrows. (f) Analysis of the number of endometrial glands. (g) ChIP analysis of ESR1 binding to ERE sequences from DR and TH pigs. (h) Band intensity analysis of the data shown in (g). DR and TH values were normalized to their respective inputs. Data are expressed as the mean ± s.e. of three replicates. DR: Duroc, TH: Taihu, ESR1: the type I receptor of oestrogen, ERE: ESR1 response element. Statistical analysis was conducted by Student's t-test with Microsoft Excel. * and **: 0.05 and 0.01 significance levels, respectively.

To further explore the effect of Taihu-ERE, 380 reproductive records (including litter size and number born alive) were analysed for possible associations between reproductive traits and the Taihu-ERE haplotype (defined as H2). The association analysis showed that number born alive was significantly higher in both H2H2 and H1H2 types than in H1H1 type pigs. H2H2 type litter size was approximately 0.2 higher than in H1H1 pigs, but the difference was not significant. No significant difference was observed between H1H2 and H2H2 types in either litter size or number born alive (table 3).

Table 3.

Association analysis between the Taihu-specific haplotype and reproductive traits. Notes: values are expressed as mean ± standard error. H1, TATCCTAAGACCGC; H2, CGCTCCGCACGTAG. Different superscript characters (a and b) indicate significant differences between haplotypes.

	H1H1 (n = 176)	H1H2 (n = 120)	H2H2 (n = 39)
litter size	10.56 ± 0.35a	10.74 ± 0.38a	10.77 ± 0.51a
number born alive	8.92 ± 0.36b	9.31 ± 0.38a	9.26 ± 0.52a

3. Discussion

The extent to which non-coding variations contribute to breed-specific phenotypes is largely unknown. Only a few non-coding mutations have been confirmed to cause phenotypic variations in animal domestication studies, such as an intronic mutation of IGF2 influencing muscle development [6] and non-coding mutations for tame behaviour in rabbits [35]. A recent report describes non-coding cis-regulatory changes in GDF6 that influence skeletal evolution [45]. These examples demonstrate that non-coding variations can play an important role in specific phenotypes. In this study, Taihu pigs exhibited significantly excessive genetic differences in comparison with other breeds at non-coding sites, especially in functional or conserved elements, suggesting that non-coding variations played a crucial role in the development of the Taihu breed. Furthermore, we identified a variation within a cis-regulatory element in the BMPR1B intron, distinct from the missense mutation identified earlier in BMPR1B in sheep [23].

Apart from the sweeps related to fertility, we also identified sweeps related to other distinctive characteristics of the Taihu breed. For example, three sweeps appear to be related to adiposity metabolism (chr1:248–249.5 Mb; chr8:70.25–70.55 Mb; chrX:49.05–51.55 Mb) [46–48]. Eleven other sweeps overlap with published QTLs related to ear area, ear weight and ear erectness (Taihu pigs have exceptionally large and floppy ears) [49]. One sweep overlaps with a coat colour QTL [50]. Two sweeps overlap with QTLs associated with scapula length, and another six sweeps are associated with vertebra number, lumbar vertebra number, and thoracic vertebra number [51–53]. Additionally, the sweep located at chr1:245.15–245.3 Mb overlaps with a QTL related to susceptibility to umbilical hernia in the pig [54]. These results strengthen our conclusion that the candidate sweeps are associated with Taihu-specific traits. The remaining selective sweeps, which have not yet been analysed, may be related to other characteristics of the Taihu breed.

In our study, Taihu-specific SNPs were obtained by detecting genetic differentiation between Taihu pigs and other breeds. However, genetic differentiation not only occurs as a result of selection, but also through other factors, such as introgression and random genetic drift. In particular, because BMP15 harbours an introgressed fragment from a possibly extinct Sus species, and is associated with local adaptation in Chinese pigs [12], it might be an example of introgression. Random genetic drift can also lead to genetic differentiation between isolated lines. Changes in gene frequency resulting from drift are cumulative and have no direction [55]. Theoretically, the effects of drift on gene frequency dominate in the absence of other factors, such as selection, when the idealized population size is small. However, Taihu pigs are the result of long-term artificial selection, since high prolificacy has always been a goal during domestication due to its significant benefits. Because the Taihu-specific haplotype in Taihu pigs is nearly fixed, and the selective sweeps we identified were significantly associated with high prolificacy, we conclude that drift is not the primary contributor to genetic differentiation in this case. Nonetheless, we do not discount the importance of introgression or random genetic drift. We will examine their effects on the frequency of genes related to prolificacy in an idealized natural population in the future.

A recent report suggested that Asian pig populations might share common ancestries within certain geographical regions [56]. However, the Taihu, Tibetan and DNSE pigs used in this study are from three distinct areas, and therefore could have been domesticated from different wild boar populations. Although modern Chinese domesticated pigs are thought to be direct descendants of pigs originally domesticated along the Yellow River, we cannot exclude the possibility that pigs were domesticated independently in other parts of China [57,58]. Thus, some Taihu-specific SNPs could be remnants from an ancient lineage of wild boars. So, it is also likely that Taihu-specific SNPs were derived from selection during or after domestication.

In contrast with human research, a frustrating barrier in domestic animal studies is the lack of functionally annotated genomic data, such as epigenetic datasets. This shortcoming severely limits the ability to interpret the importance of noncoding mutations and other features. Here, we integrated population genetics, comparative genomics and molecular experiments to identify convincing candidate mutations responsible for pig prolificacy, a complex trait that is difficult to decipher through traditional recombination segregation analyses due to low heritability [13]. The results demonstrate that our integrated approach can efficiently identify functional elements underlying important phenotypes in domesticated animals, and also provides targets for subsequent investigations to improve pig fecundity.

4. Material and methods

Detailed information was described in electronic supplementary material.

Data accessibility

Raw reads have been deposited in the NCBI Sequence Read Archive under accession PRJNA358108.

Authors' contributions

K.L.W. and C.X.W. conceived and designed the experiments. W.T.L., K.J.W., M.M.Z., M.Z.Z., Y.F.L., Q.Y.L. and H.G.B. performed the experiments. Q.G.L. and W.T.L. designed and performed bioinformatics analysis. H.T. and L.F.Z. provided and prepared the samples and materials. W.T.L., K.J.W. and Y.M.Z. analysed the data, wrote the manuscript and prepared the figures. All authors reviewed the manuscript.

Competing interests

The authors declare no competing financial interests.

Funding

This study was supported by the National Basic Research Program of China (973 Program, Grant 2014CB138501), the National Transgenic Animal Breeding Grand Project (2014ZX08006-005), the programme for Changjiang Scholars and Innovation Research Teams in the University (IRT_15R62), Beijing Innovation Consortium of Swine Research System (BAIC02-2017), and the Huazhong Agricultural University Scientific and Technological Self-Innovation Foundation (2014RC020).