Figures
Abstract
We conducted a selection signature analysis using the chicken 60k SNP chip in two chicken lines that had been divergently selected for abdominal fat content (AFC) for 11 generations. The selection signature analysis used multiple signals of selection, including long-range allele frequency differences between the lean and fat lines, long-range heterozygosity changes, linkage disequilibrium, haplotype frequencies, and extended haplotype homozygosity. Multiple signals of selection identified ten signatures on chromosomes 1, 2, 4, 5, 11, 15, 20, 26 and Z. The 0.73 Mb PC1/PCSK1 region of the Z chromosome at 55.43-56.16 Mb was the most heavily selected region. This region had 26 SNP markers and seven genes, Mar-03, SLC12A2, FBN2, ERAP1, CAST, PC1/PCSK1 and ELL2, where PC1/PCSK1 are the chicken/human names for the same gene. The lean and fat lines had two main haplotypes with completely opposite SNP alleles for the 26 SNP markers and were virtually line-specific, and had a recombinant haplotype with nearly equal frequency (0.193 and 0.196) in both lines. Other haplotypes in this region had negligible frequencies. Nine other regions with selection signatures were PAH-IGF1, TRPC4, GJD4-CCNY, NDST4, NOVA1, GALNT9, the ESRP2-GALR1 region with five genes, the SYCP2-CADH4 with six genes, and the TULP1-KIF21B with 14 genes. Genome-wide association analysis showed that nearly all regions with evidence of selection signature had SNP effects with genome-wide significance (P<10–6) on abdominal fat weight and percentage. The results of this study provide specific gene targets for the control of chicken AFC and a potential model of AFC in human obesity.
Citation: Zhang H, Hu X, Wang Z, Zhang Y, Wang S, Wang N, et al. (2012) Selection Signature Analysis Implicates the PC1/PCSK1 Region for Chicken Abdominal Fat Content. PLoS ONE 7(7): e40736. https://doi.org/10.1371/journal.pone.0040736
Editor: Ioannis P. Androulakis, Rutgers University, United States of America
Received: December 21, 2011; Accepted: June 12, 2012; Published: July 11, 2012
Copyright: © 2012 Zhang et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This research was supported by China Agriculture Research System (No. CARS-42), National 863 project of China (No. 2010AA10A102) and National 973 Project of China (No.2009CB941604). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
The chicken (Gallus gallus) is an important model organism that bridges the evolutionary gap between mammals and other vertebrates [1]. Research on human obesity typically uses body mass index (BMI) as the phenotypic measure of obesity [2]–[12]. However, BMI is affected by variations in the entire body, including bones, muscles and body fat, and is not specific for abdominal fat, a major problem in obese people. Although indirect measures of human abdominal fat are available [13], [14], direct measures are unavailable. In chickens, abdominal fat weight (AFW) can be measured directly, and experiments could be designed to identify genetic variants associated with abdominal fat content (AFC). The results from this type of experiment may provide useful comparative information for human obesity research and lead to genetic improvement for reduced abdominal fat in chickens.
Selection for rapid growth in chickens has always been accompanied by increased fat deposition [15], [16]. Excessive fat deposition can decrease feed efficiency and cause consumer rejection of the meat [17], and cause difficulties in meat processing [18]. The measurements of fatness are often laborious and expensive by slaughtering birds, which prevents genetic selection on the basis of an individual's measures of fatness. Knowledge of the genetic factors associated with fatness will facilitate genetic selection using genetic markers without the necessity to assess the phenotype of the selected individuals. Genome-wide association studies (GWAS) provide a powerful approach to the identification of the genetic factors associated with phenotypes. However, GWAS is affected by variations in phenotypic measures and by genetic drift and hitchhiking of the genome. In contrast, selection signature analysis [19]–[21] does not rely on phenotypic measurements and could be a promising approach to address the statistical noise from drift and hitchhiking. An integrated analysis of selection signature and GWAS provides a new approach that has the strengths of both methods. A joint analysis of selection signature and GWAS using two divergent chicken lines has been reported [22].
The goal of this study was to identify genome changes and genes associated with selection for high and low AFC using multiple signals of the selection signature in an experimental chicken population that had undergone 11 generations of divergent selection for high and low AFC. The GWAS analysis was used to assess the phenotypic effects of the selection signatures identified in this study and was also used to identify likely causal locations of the selection signatures with dense gene coverage.
Materials and Methods
Ethics statement
All animal work was conducted according to the guidelines for the care and use of experimental animals established by the Ministry of Science and Technology of the People's Republic of China (Approval number: 2006-398), and was approved by the Laboratory Animal Management Committee of Northeast Agricultural University.
Animals
The broilers used in this study were from two Northeast Agricultural University (NEAU) broiler lines divergently selected for AFC (NEAUHLF). The NEAUHLF lines have been selected since 1996 using abdominal fat percentage (AFP = AFW/body weight) and plasma very low-density lipoprotein (VLDL) concentration as selection criteria. The G0 generation of NEAUHLF came from the same grandsire line originating from the Arbor Acres broiler, which was then divided into two lines according to their plasma VLDL concentration at 7 weeks of age. The G0 birds were mated (one sire: four dams) to produce 25 half-sib families for each line, with an average of 70 G1 offspring per family in two hatches. From G1 to G11, the birds of each line were raised in two hatches and housed in pens with five birds per cage. Plasma VLDL concentrations were measured for all male birds, which had free access to feed and water at 7 weeks, and the AFP of the male birds in the first hatch was measured after slaughter at 7 weeks. Sib birds from the families with lower (lean line) or higher (fat line) AFP than the average value for the population were selected as candidates for breeding, considering the plasma VLDL concentration and the body weights of male birds in the second hatch and the egg production of female birds in both hatches. These birds were kept under the same environmental conditions and had free access to feed and water. Commercial corn-soybean-based diets that met all National Research Council (NRC) requirements were provided. From hatch to 3 weeks of age, the birds received a starter feed (3,000 kal ME = kg and 210 g = kg CP) and from 4 weeks of age to slaughter the birds were fed a grower diet (3,100 kal ME = kg and 190 g = kg CP). A total of 475 individuals (203 from the fat line and 272 from the lean line) from generation 11 of NEAUHLF were used in this study. The AFP of the fat line was 3.75 times that of the lean line at generation 11 (Table 1), and the phenotypic (AFP) changes over the 11 generations are shown in Figure 1.
SNP selection and genotyping
The 60k chip had a total of 57,636 SNPs, and 45,578 SNPs with a minor allele frequency (MAF) of 5% or greater and a call rate of 95% in the combined sample of the lean and fat lines were selected for use in this study. Individuals with pedigree error or 5% or more missing SNP genotypes were removed. Of the 45,578 SNPs, 45,005 had known chromosome locations and were distributed across 28 autosomes, the Z chromosome and two linkage groups (LGE22C19W28_E50C23 and LGE64). The number of SNPs per chromosome ranged from 2 to 7,135 with a mean distance of 22.51 kb between adjacent SNPs (Table 2). Genomic DNA was isolated from venous blood samples using a phenol-chloroform method from 20 μL of venous blood collected in EDTANa2-coated tubes and stored at −20°C. Genotyping of the chicken 60k SNP chips from Illumina Inc. was performed by DNA LandMarks Inc., Quebec, Canada, using 75 μL of approximately 50 ng/μL genomic DNA.
Statistical analysis
Selection signature analysis.
The selection signature analysis used a combination of various signals of selection, including long-range heterozygosity changes [19], long-range allele frequency differences (AFD) and standardized AFD between the lean and fat lines, linkage disequilibrium (LD) and haplotype analyses [23], and extended haplotype homozygosity (EHH) analysis [24]. Heterozygosity and AFD measures were used for the first screening of selection signatures, and LD, haplotype frequencies and EHH were analyzed as additional evidence of selection signatures and as indications of whether selection had occurred in one line or both lines. The LD and haplotype analyses used Haploview [23], and the extended haplotype homozygosity (EHH) analysis was carried out using Sweep-1.1 [24]. Phased genotypic data as input files for Sweep 1.1 were produced using FASTPHASE [25].
For long-range AFD and heterozygosity measures, we used 0.5 Mb sliding windows of SNP markers as the genome length. Two long-range heterozygosity measures were calculated following the method in [19]: standardized heterozygosity in the lean line (Z_lean) and standardized heterozygosity in the fat line (Z_fat). For each sliding window, we also calculated the AFD and standardized AFD (Z_AFD) between the lean and fat lines. For each chromosome, the Z_AFD between the two lines used the chromosome mean and standard deviation of the AFD values, while Z_lean and Z_fat each used within-line mean and standard deviation of the heterozygosity values of the entire chromosome. This type of within-line standardization is more conservative than across-line standardization using the pooled mean and standard deviation of the chromosome over the two lines. The criterion for declaring selection was the use of extreme AFD and extreme standardized AFD and heterozygosity values, following the approach in [19]. Threshold values of the above measures for declaring significance were AFD ≥0.44, Z_AFD = 4.0, and Z_lean = Z_fat = ±5.0, and the percentages of markers above these threshold values were 0.27%, 0.09%, 0.11%, and 0.02%, respectively.
The two AFD measures (AFD and Z_AFD) were used to compensate for the weakness of the three measures of heterozygosity (Z_lean and Z_fat) in cases of ‘p-q sweep’, where heterozygosity measures are expected to fail to detect genome changes due to selection. Let p0 and q0 represent the allele frequencies of alleles 1 and 2 in the unselected population, and let pt and qt represent the allele frequencies for the same alleles after t generations of selection. Then, heterozygosity has no change at generation t if pt = q0 and qt = p0 (p-q sweep) assuming Hardy-Weinberg equilibrium. Therefore, the use of heterozygosity could miss significant allele frequency changes that result in a p-q sweep. For this reason, increases in heterozygosity (rather than decreases only) were also considered (in addition to AFD and Z_AFD) when identifying chromosome regions subjected to selection, because increases in heterozygosity could also be a result of substantial allele frequency changes when the initial population had extreme frequencies.
SNP association analysis.
The significance testing of SNP effects on AFW and AFP used four methods, the GLS-LS version of EPISNP [26], [27] was the main method for reporting GWAS results, and PLINK [28] was used as a secondary method for reporting GWAS results. The GLS-LS version uses generalized least squares (GLS) to estimate fixed non-genetic effects and uses least squares (LS) for testing SNP effects after removing the GLS estimates of fixed non-genetic effects from the phenotypic values. The statistical model was Y = SNP + f + e, where Y was the phenotype value, SNP was the SNP marker effect, f was a random family effect to account for sib correlation, and e was the residual effect. Each SNP was tested for two effects, additive and dominance effects. The genome-wide 5% type-I error with the Bonferroni correction was considered to indicate genome-wide significance. For two traits (AFW and AFP), two effects per test (additive and dominance effects) and 45,578 SNP markers, the threshold P-value for declaring genome-wide significance was (0.05)/[(2)(2)(45,578)] = 2.74×10−7 = 10−6.56. The statistical model for PLINK was the same as that for the GLS-LS version of EPISNP, except that the family effect was treated as a fixed effect. Gene locations were based on Ensembl [29] and NCBI [12]. Full gene names from Ensembl and NCBI for the candidate genes identified in Table 3 and Figure S1 are given in Table S1.
Results and Discussion
Multiple signals of selection identified ten selection signatures on chromosomes 1, 2, 4, 5, 11, 15, 20, 26 and Z as signatures of selection for high and low AFC in the 11 generations. The candidate genes and regions with multiple genes in the selection signatures were PC1/PCSK1, PAH-IGF1, TRPC4, GJD4-CCNY, NDST4, NOVA1, GALNT9, the ESRP2-GALR1 region with five genes, the SYCP2-CADH4 region with six genes, and the TULP1-KIF21B region with 14 genes. Table 3 describes general characteristics of these selection signatures. A summary of the selection signatures and chromosome regions with highly significant SNP effects is given in Figure 2, which also displays the locations of human obesity genes [2–12; Table S2] on the chicken genome. Six selection signatures on chromosomes 1, 2, 4, 5 and 15 each involved one or two genes while the remaining four signatures on chromosomes 11, 20, 26 and Z each involved 6–14 genes.
Black: selection signature and gene in the selection signature (nearly all these regions had SNP effects with genome-wide significance); Blue: chromosome region with highly significant SNP effects but not declared as a selection signature. Purple: human obesity gene nearest to a selection signature or a region with highly significant SNP effects for abdominal fat weight (AFW) and abdominal fat percentage (AFP); Red: human obesity gene inside selection signature; Green box: not studied.
Long-range AFD between the lean and fat lines, Z_AFD, Z_lean, and Z_fat in 0.5 Mb sliding windows of SNP markers were used for initial genome-wide screening of selection signatures (Figure S2). Linkage disequilibrium (LD) patterns [23] (Figure S3), extended haplotype heterozygosity (EHH) analysis [24] (Figure S4) and haplotype analyses [23] (Table S3) further confirmed the initial evidence of selection signatures, provided evidence of whether selection had occurred in one line or both lines, and modified the sizes of some selection signatures. In addition, the EHH analysis added information about a specific gene or genes that may have been subject to selection.
The PC1/PCSK1 region of chromosome Z: the most heavily selected chromosome region
The most significant region in the 11 generations of divergent selection for high and low AFC was the 0.73 Mb region of 55.43–56.16 Mb of chromosome Z, with 26 intronic SNP markers. This region had the largest long-range AFD and the largest Z_AFD between the lean and fat lines for all chromosomes, with an average AFD greater than 0.70 and peak Z_AFD of 4.98 in 0.5 Mb sliding windows of SNP markers (Figure 3, Table 3). No other selection signature had such large average AFD, particularly for a 0.73 Mb distance with 26 SNP markers. Single locus AFD between the lean and fat lines for the 26 SNP markers in this region were in the range of 0.68–0.78, and dropped to 0.11–0.46 outside this 0.73 Mb region, although large AFD values existed further away from this region. This 0.73 Mb region had seven genes, Mar-03, SLC12A2, FBN2, ERAP1, CAST, PC1, and ELL2, noting that only a downstream portion of Mar-03 and an upstream portion of ELL2 were in this region. Of these seven genes, only PC1 is known to have biological functions highly relevant to AFC. In humans, PC1 is also known as PCSK1 [12]. This gene plays a key role in regulating insulin biosynthesis [12], and is on the leptin/melanocortin pathway that is related to obesity [30]. Variants in PCSK1 have been found to be associated with human obesity [8]–[12]. The complete chicken PC1 DNA sequence is 56,014,540–56,043,792 bp in length. This sequence matches 16 segments of the human PCSK1 DNA sequence [29] and has 79.1% identity with the human PCSK1 DNA sequence [31]. No match or identity between the chicken PC1 and any other human DNA sequence was found. For this reason and considering the known relevance of PC1/PCSK1 to obesity, we named this 0.73 Mb seven-gene region as the PC1/PCSK1 region.
The haplotype and LD analyses showed that the two lines had two main haplotypes and shared a common recombinant haplotype in two haplotype blocks separated by the recombination point at 55,736,592–55,808,443 bp, shown between markers 1,141 and 1,142 in Figure 4. The main haplotype in the lean line (Haplotype 1 in Table 4) had a frequency of 0.683 and did not exist in the fat line. The main haplotype in the fat line (Haplotype 2 in Table 4) had a high frequency of 0.75 and a low frequency of 0.017 in the lean line, and these two main haplotypes had opposite alleles at each of the 26 SNP markers (Table 4). The recombinant haplotype (Haplotype 3 in Table 4) in each line was due to recombination between Haplotypes 1 and 2 (between markers 1,141 and 1,142 in Figure 4) and had similar frequencies in the two lines (0.193 in the lean line and 0.196 in the fat line). This indicates that the recombinant type is not of great importance to either line and that both sides of the recombination point had causal effects. The lean line had three other haplotypes with a combined frequency of 0.12, and the fat line also had three other haplotypes with a combined frequency of 0.0524 (Table 4).
WPR: a recessive white line of White Plymouth Rock chicken from France (n = 94). AK: Anak, a commercial broiler chicken introduced from Israel (n = 51). CAU-F2: an F2 resource population produced from reciprocal crosses of Silky Fowl and White Plymouth Rock at China Agricultural University (n = 112).
The LD and EHH results also showed that this 0.73 Mb region was potentially subject to selection, and that the fat line had stronger selection than the lean line. The fat line had strong LD values, r2 = 1 in both blocks for adjacent SNPs, except for one SNP pair in the first block and two SNP pairs in the second block, and had three fixed SNP markers in the first block (Figure 4). The fat line had higher r2 values and more fixed SNP markers than the lean line, which did not have fixed SNP markers in both haplotype blocks, indicating that the 0.73 Mb region had greater selection pressure in the fat line than in the lean line. Note that the default algorithm for defining haplotype blocks [32] implemented by Haploview [23] defined different haplotype blocks (Figure S3A) from those shown in Figure 4. The results of EHH analysis (Figure 5) were in agreement with the LD analysis. For the lean line, EHH analysis defined essentially two haplotype blocks separated by the recombination point between the lean and fat haplotypes (Haplotype 1 and 2 in Table 4). The third haplotype block covered only about 30 kb of the selection signature. For the fat line, two haplotype blocks were defined for a total distance that exceeded the selection signature distance by about 910.309 kb. Therefore the two haplotype blocks in the fat line covered nearly 1 Mb more than the first two haplotype blocks in the lean line. The EHH values of both haplotype blocks in the fat line were also higher and were at 100% for longer distances than those of the first two haplotype blocks in the lean line. These EHH results indicate that the fat line was subjected to stronger selection pressure in this 0.73 Mb region. The LD values of the lean and fat lines were obviously stronger than those of three chicken lines without selection for abdominal fat (White Plymouth Rock (WPR), Anak (AK) and an F2 population constructed by China Agricultural University (CAU-F2), shown on the right of Figure 4), which indicates that this chromosome region had been subjected to strong selection in the lean and fat lines.
The chromosome 1 selection signature at the PAH-IGF1 region
This selection signature was the only selection signature identified by decreased heterozygosity in the fat line (Z_fat = −5.99, Table 3) in the 57.05–57.16 Mb region that included PAH and was about 150 kb upstream of IGF1, a gene known to be associated with fatness in chicken [33], [34]. The 60k chicken SNP chip had two SNP markers in IGF1 but neither met the minor allele frequency (MAF) requirement (MAF >5%). Consequently, no SNP marker inside IGF1 was present in this study and it is unknown whether IGF1 was part of the PAH selection signature. However, four SNP markers that were immediately upstream of IGF1 were fixed, and another three SNPs were either fixed or nearly fixed with allele frequencies of 0.015, 1.0 and 0.983 in the fat line, while the AFD between the two lines for these seven markers were 0.14–0.53, indicating that IGF1 could have been selected for high AFC. The LD signals and EHH values in the region spanning IGF1 were considerably stronger in the fat line than in the lean line (Figure S3B, Figure S4A), further indicating potential involvement of IGF1 in chicken AFC. The Haploview [23] defined two haplotype blocks in the region of 56,922,109–57,998,003 bp with 17 genes, and Sweep 1.1 defined one haplotype block in the same region, which showed strong LD signals and EHH value downstream of IGF1. This could be due to causal effects downstream of IGF1 or hitchhiking effects due to selection in the PAH-IGF1 region.
The chromosome 1 selection signature at the TRPC4 region
This selection signature was at 176.08–176.29 Mb and was identified by a large standardized AFD value (Z_AFD = 4.23, Table 3). The EHH analysis identified one haplotype block approximately in the same region in the lean and fat lines, with considerably stronger EHH values in the fat line than in the lean line (Figure S4B). TRPC4 was at 176,097,670–176,235,464 bp, and therefore this selection signature essentially was due to selection on TRPC4 in the fat line. This indicates that TRPC4 is associated with increased AFC. The LD signals were stronger in the fat line than in the lean line (Figure S3C).
The chromosome 2 selection signature at the GJD4-CCNY region
This selection signature was identified by a large AFD value (0.47) and a large Z_AFD value (4.15) in the 12.48–12.80 Mb region with two genes, GJD4 at 12,756,816–12,760,762 and CCNY at 12,774,998–12,818,966 bp (Table 3). The EHH analysis defined a haplotype block at 12,750,042–12,858,095 in the lean line, covering both GJD4 and CCNY (Figure S4C). This region had elevated LD values in the lean line (Figure S3D). These results suggest that both GJD4 and CCNY are associated with selection for low AFC in the lean line.
The chromosome 4 selection signature at the NDST4 region
The chromosome 4 selection signature at 57.43–57.79 Mb that includes NDST4 was identified by two measures, the second largest AFD value of 0.50 (along with the chromosome 26 signature) and a large Z_AFD value (4.26). The LD and EEH values strongly suggested that selection in this region was in the lean line only (Figure 6). The NDST4 gene was at 57,673,334–57,758,645 bp. This gene region had strong LD signals and the highest EHH values in the lean line for over 1 Mb distance (Figure 6), which indicates a strong association of NDST4 with low AFC. The EHH analysis defined a 0.58 Mb core region at 57,192,825–57,775,670 bp, and EHH values for the main haplotype.
The chromosome 5 selection signature at the NOVA1 region
The chromosome 5 selection signature at 35.02–35.65 Mb region that contains NOVA1 was identified by a Z_AFD value of 4.4. The Z_lean value of −4.07 indicates that selection in this region was mainly in the lean line. The EHH values provided the strongest evidence that selection in this region occurred in the lean line (Figure 7). The two haplotype blocks defined by EHH analysis in the lean line spanned the region of 34,363,844–35,929,248 bp, with the main haploype in each block having a high frequency of 0.68. NOVA1 at 34,863,690–35,327,411 nearly covered the total length of the two haplotype blocks. This strongly suggests that NOVA1 was involved in low AFC. The lean line also had more fixed loci and stronger LD than the fat line (Figure 7).
The chromosome 11 selection signature at the ESRP2-GALR1 region
The 3.20–3.40 Mb region of chromosome 11 was the only selection signature identified by decreased heterozygosity in the lean line (Z_lean = −5.80, Table 3). The Z_lean value of −5.80 indicated that selection in this region was in the lean line, and this was confirmed further by the LD and EHH results (Figure S3E, Figure S4D). Five genes were located in this selection signature but none of these genes was known to have a role in fat content. Two genes reported to be associated with human obesity are in the vicinity of this region, MMP2 [6], [7] which is 0.28 Mb downstream and FTO [2]–[5] which is 1.45 Mb downstream of this region.
The chromosome 15 selection signature at the GALNT9 region
This selection signature was identified by a high AFD value of 0.44 (Table 3) at the 2.15–2.50 Mb region containing GALNT9, which is at 2,393,778–2,485,829 bp. The EHH analysis defined one halplotype block in both lines at 2,263,761–2,470,336 bp, covering 87 kb of the 92 kb GALNT9 gene (Figure S4E). The main haplotype in the lean line had higher EHH values than the main haplotype in the fat line, and the third most frequent haplotype in the fat line had high EHH values and was likely to have been subjected to selection for high fat content. The lean line had five more fixed loci and stronger LD than the fat line (Figure S3F).
The chromosome 20 selection signature at the SYCP2-CADH4 region
The chromosome 20 selection signature at 6.83–7.29 Mb had a peak AFD value of 0.46. This 0.46 Mb region contained six genes. This is a complex region but the long-range AFD and heterozygosity analysis had good agreement with the EHH analysis. Five haplotype blocks in the lean line and four haplotype blocks in the fat line were defined in the EHH analysis. In the lean line, haplotype blocks 1 and 2 covered five genes, SYCP2, PPP1R3D, ENSGALG00000024473, C20orf177, and ENSGALG00000023777, block 3 covered a blank space, and blocks 4 and 5 covered 109.85 kb of the 414.10 kb CADH4 at 7,213,069–7,627,166. In the fat line, the first two blocks covered approximately the same region of six genes that was covered by the first two blocks in the lean line, block 3 covered the same blank area covered by block 3 in the lean line, and block 4 covered the same area of CADH4 covered by the last two blocks in the lean line (Figure S4F). The EHH values in the fat line generally were higher than those in the lean line. The EHH patterns suggest that this region was subjected to selection in both lines. The LD signals in this region were the weakest among the selection signatures identified in this study (Figure S3G). These relatively weak LD signals could be due to the fact that chromosome 20 had considerably more markers per Mb than other chromosomes, e.g., 104.89 SNPs/Mb for this chromosome and 35.51 SNPs/Mb for chromosome 1 (Table 2). Chromosome 20, along with the Z chromosome, had the largest average AFD (0.24) among all chromosomes, excluding the linkage group LGE64 that only had two markers (Table 2).
The chromosome 26 selection signature at the upstream telomere region
The 0.06–0.29 Mb region of chromosome 26 with 14 genes was the only selection signature identified by increased heterozygosity in the fat line (Z_fat = 5.02, Table 3), and was the only selection signature identified by another two measures of selection signature, AFD and Z_AFD (Table 3). The peak AFD value (0.50), along with the peak AFD value of the chromosome 4 selection signature, was the second largest of the entire genome, next to the AFD in the PC1/PCSK1 region. This region had strong LD in both lines. The fat line had stronger LD values but the lean line had a larger number of fixed SNP markers (Figure S3H). One haplotype block was defined in each line by EHH analysis. The block size was 55,909–302,113 bp for the lean line and 55,909–334,821 for the fat line (Figure S4G). The EHH values indicated selection in both lines but the fat line had stronger selection than the lean line in this region.
Haplotype frequencies in selection signatures of the lean and fat lines
Most of the selection signatures had a line-specific main haplotype that either did not exist or had a very low frequency in the other line. In addition to the PC1/PCSK1 region discussed above, selection signatures at TRPC4, GJD4-CCNY, NOVA1, GALNT9, SYCP2-CADH4 and TULP1-KIF21B had such line-specific haplotypes (Table S3). For the TRPC4 region, the main lean line haplotype with a frequency of 0.481 did not exist in the fat line, while the main fat line haplotype with a frequency of 0.462 had a very low frequency of 0.032 in the lean line. This trend was the same for the GJD4-CCNY, NOVA1, GALNT9, and SYCP2-CADH4 regions. The selection signature at TULP1-KIF21B of chromosome 26 had a line-specific main haplotype with a frequency of 0.539 in the fat line, while the main lean line haplotype with a high frequency of 0.808 had a low frequency of 0.253 in the fat line.
For the PAH-IGF1 region, the main haplotype of the fat line had a high frequency of 0.984 and the same haplotype had a relatively low frequency of 0.392 in the lean line, while the second most frequent (0.329) haplotype in the lean line did not exist in the fat line (Table S3). For the NDST4 region, the main haplotype in the lean line had a high frequency of 0.742 and a low frequency of 0.16 in the fat line, which did not have a dominant main haplotype (Table S3). For the chromosome 11 signature in ESRP2-GALR1, the second most frequent fat line haplotype with a frequency of 0.359 did not exist in the lean line, and the main haplotypes in both lines had a substantial frequency difference, 0.795 in the lean line and 0.529 in the fat line (Table S3). Overall, the data on haplotype frequencies provided strong additional evidence that the regions with selection signatures identified by AFD, heterozygosity change, LD and EHH indeed were subjected to selection.
Phenotypic effects of selection signatures
Genome-wide association analysis detected a total of 569 SNP markers with phenotypic effects on AFW and AFP and with genome-wide significance (P<10−6.56, Figure S1). Of the 569 SNPs, 342 (60%) were significantly associated with both AFW and AFP, and the other 40% were associated with either AFW or AFP. In the literature, 216 quantitative trait loci (QTL) for traits related to abdominal fat in chickens have been reported [35]. Approximately 46% of the 569 SNPs were located in 39% of the 216 reported QTL regions. Nearly all the selection signatures detected in this study had significant SNP effects on AFW and AFP with genome-wide significance, or were in the close proximity to significant SNP effects (Table 3), although most SNP effects in or near the selection signatures were not ranked among the most significant.
The 55.43–56.16 Mb PC1/PCSK1 region of chromosome Z, which was identified as the most highly selected region of the entire chicken genome had SNP effects on AFW and AFP with genome-wide significance. The rankings of the SNP effects were not among the highest. However, with the knowledge that both haplotype blocks shown in Figure 4 were required to have the highest or lowest AFW and AFP values, the relatively low ranking of the SNP effects could be due to the single-locus analysis, which could detect only half of the phenotypic effects in one haplotype block of the selection signature, thus yielding lower statistical significance. The AFD values overlapped with the SNP effects almost exactly (Figure 8A).
Each selection signature on chromosome 1 (PAH-IGF1 and TRPC4 regions) was close to a group of SNP effects, and a human obesity gene, MTIF3 [2]–[3], was close to the TRPC4 region (Figure 8B). Selection signatures at the GJD4-CCNY region of chromosome 2 (Figure 8C) and at the GALNT9 region of chromosome 15 (Figure S1) had SNP effects that were ranked high among all chromosomes and were the highest among all selection signatures. The chromosome 5 signature at NOVA1 had a significant SNP for AFW and AFP, the chromosome 11 selection signature nearly overlapped with SNP effects that were close to MMP2, a human obesity gene [6], [7], and the chromosome 26 signature had a cluster of SNP effects with similarly low significance (Figure S1). The chromosome 20 signature had a large number of significant SNP effects for AFW and AFP, with CADH4 having one of the most significant SNP for AFW and AFP and five other significant SNPs for AFW. The SNP effects upstream of CADH4 were in an intergenic region. The peak AFD values and significant SNP effects overlapped well, similar to the overlap between AFD and SNP effects in the PC1/PCSK1 region (Figure S1).
The chromosome 4 signature at NDST4 had one significant SNP for AFW and AFP that was ranked low, and was between two large groups of SNP effects (Figure 8D). On the left side of the NDST4 region was a large group of SNP effects in an 8.68 Mb region at 42.95–51.63 Mb. Near the downstream end of this large 8.68 Mb region was LEPROTL1 (a leptin receptor) at 50,714,780–50,717,850 bp. This molecule is on the leptin/melanocortin pathway, which is related to obesity [25]. This large (8.68 Mb) region had three locations with large AFD values, the left end, the right end and the IGFBP7-LEPROTL1 region. The SGCZ region downstream of the NDST4 selection signature had the most significant SNP effect and substantial AFD (0.35).
Gene name in black: gene in the selection signature; Gene name in blue: gene in the chromosome region with highly significant SNP effects but not declared as a selection signature; Gene name in purple: gene associated with human obesity [2]–[12]. Blue circles and red plus signs: SNP effects from EPISNP [26], [27]; Green triangles: SNP effects from PLINK [28].
Candidate genes for chicken AFC
The most apparent candidate gene was PC1/PCSK1 in the most significant 0.73 Mb region at 55.43–56.16 Mb of chromosome Z. This region clearly stood out as the region with the strongest evidence of selection in the entire chicken genome, in both the lean and fat lines in this study. In addition, the known function of PCSK1 in regulating insulin biosynthesis and its known association with human obesity [8]–[12] make PC1/PCSK1 the most likely candidate gene to have a causal effect on AFC. However, haplotype analysis showed that PC1/PCSK1 should not be the only gene with causal effect in this region, and that the evidence in this region was consistent with a two-locus model with two causal loci flanking the recombination point at 55,736,592–55,808,443 bp inside FBN2.
Upstream of the recombination point were part of Mar-03, SLC12A2, and the upstream two-thirds of FBN2. None of these three genes has known biological functions specifically related to fat metabolism. However, based on chromosome positions, Mar-03 was on the left end of this region and should be least likely to have causal effects, because a substantial hitchhiking effect upstream of Mar-03 should have been observed if Mar-03 had been one of the most significant genes. This analysis leaves SLC12A2 and the upstream two-thirds of FBN2 to be the candidate genes for causal effects. Downstream of the recombination point were the downstream one-third of FBN, ERAP1, CAST, PC1/PCSK1 and part of ELL2. In this block, PC1/PCSK1 is the most apparent candidate gene but the downstream one-third of FBN, ERAP1, CAST and ELL2 could not be excluded from having causal effects. However, ELL2 is unlikely to be the target of selection given that AFD values downstream of ELL2 dropped rather sharply (Figure 3).
Selection signatures that occur in single genes should make those genes apparent targets for candidate genes, such as PAH-IGF1, TRPC4, GJD4, CCNY, NDST, NOVA1 and GALNT9. The chromosome 20 signature at 6.83–7.29 Mb involved five genes with CADH4 being the largest gene of that region. The chromosome 11 signature of the ESRP2-GALR1 region at 3.20–3.40 Mb had five genes without known biological functions relevant to AFC. However, a human obesity gene (MMP2) was only 0.28 Mb downstream at 3.68–3.72 Mb, and a second human obesity gene (FTO) was less than 1.45 Mb downstream at 4.85–4.89 Mb, making this region an interesting region for candidate genes. Within this region, the most significant SNP was at 4.40 Mb for AFW and AFP (Figure S1), 0.68 Mb downstream of MMP2 and 0.45 Mb upstream of FTO, indicating that the region of MMP2-FTO could contain a causal effect. The chromosome 26 selection signature at 0.06–0.29 Mb involved about 14 genes. In this region, the 0.14–0.27 region had five significant SNPs for AFW and AFP, with one SNP in TBC1D22B and two SNPs in CAC1S. The upstream telomere region of chromosome 26 is a gene-dense region, and current evidence would not pinpoint with good accuracy to a causal location in this region. The general conclusion for this region is that a causal effect should exist in the 0–0.5 Mb region of chromosome 26.
In summary, ten selection signatures on chromosomes 1, 2, 4, 5, 11, 15, 20, 26 and Z were identified in the current study. The 0.73 Mb PC1/PCSK1 region of the Z chromosome at 55.43–56.16 Mb was the most significant selected region. The PC1/PCSK1 gene in this region might be important for chicken abdominal fat content.
Supporting Information
Figure S1.
SNP effects on abdominal fat weight (AFW) and abdominal fat percentage (AFP). Blue circles and red plus signs: SNP effects from EPISNP [26], [27]; Green triangles: SNP effects from PLINK [28].
https://doi.org/10.1371/journal.pone.0040736.s001
(PDF)
Figure S2.
AFD and Z values in 0.5 Mb sliding windows of SNP markers.
https://doi.org/10.1371/journal.pone.0040736.s002
(PDF)
Figure S3.
Linkage disequilibrium patterns of the selection signatures.
https://doi.org/10.1371/journal.pone.0040736.s003
(PDF)
Figure S4.
Extended haplotype homozygosity in selection signatures.
https://doi.org/10.1371/journal.pone.0040736.s004
(PDF)
Table S1.
Full gene names of candidate genes for chicken abdominal fat content. The genes in Table 3 and genes in black and blue colors in Figure S1 are included. Human obesity genes are not included, except PC1/PCSK1.
https://doi.org/10.1371/journal.pone.0040736.s005
(DOC)
Table S2.
Human obesity genes on the chicken genome.
https://doi.org/10.1371/journal.pone.0040736.s006
(DOC)
Table S3.
Haplotype frequencies in selection signatures.
https://doi.org/10.1371/journal.pone.0040736.s007
(DOC)
Acknowledgments
The authors gratefully acknowledge members of the Poultry Breeding group of the College of Animal Science and Technology in the Northeast Agricultural University for help in managing the chicken lines and data collection.
Author Contributions
Conceived and designed the experiments: HL YZ NL. Performed the experiments: HZ ZW Shouzhi Wang NW QW HL. Analyzed the data: HZ XH LM Shengwen Wang YD. Contributed reagents/materials/analysis tools: YW LL XH ZT. Wrote the paper: HZ ZW YD HL.
References
- 1. Hillier LW, Miller W, Birney E, Warren W, Hardison RC, et al. (2004) Sequence and comparative analysis of the chicken genome provide unique perspectives on vertebrate evolution. Nature 432: 695–716.
- 2.
Zhao J, Bradfield JP, Zhang H, Sleiman PM, Kim CE, et al. (2011) Role of BMI-Associated Loci Identified in GWAS Meta-Analyses in the Context of Common Childhood Obesity in European Americans. Obesity (Silver Spring). https://doi.org/10.1038/oby.2011.237
- 3. Speliotes EK, Willer CJ, Berndt SI, Monda KL, Thorleifsson G, et al. (2010) Association analyses of 249,796 individuals reveal 18 new loci associated with body mass index. Nat Genet 42: 937–948.
- 4. Thorleifsson G, Walters GB, Gudbjartsson DF, Steinthorsdottir V, Sulem P, et al. (2009) Genome-wide association yields new sequence variants at seven loci that associate with measures of obesity. Nat Genet 41: 18–24.
- 5. Willer CJ, Speliotes EK, Loos RJ, Li S, Lindgren CM, et al. (2009) Six new loci associated with body mass index highlight a neuronal influence on body weight regulation. Nat Genet 41: 25–34.
- 6. Morgan AR, Han DY, Thompson JM, Mitchell EA, Ferguson LR (2011) Analysis of MMP2 promoter polymorphisms in childhood obesity. BMC Res Notes 4: 253.
- 7. Han DH, Kim SK, Kang S, Choe BK, Kim KS, et al. (2008) Matrix Metallopeptidase 2 Gene Polymorphism is Associated with Obesity in Korean Population. Korean J Physiol Pharmacol 12: 125–129.
- 8. Benzinou M, Creemers JW, Choquet H, Lobbens S, Dina C, et al. (2008) Common nonsynonymous variants in PCSK1 confer risk of obesity. Nat Genet 40: 943–945.
- 9. Choquet H, Stijnen P, Creemers JW (2011) Genetic and Functional Characterization of PCSK1. Methods Mol Biol 768: 247–253.
- 10. Kilpeläinen TO, Bingham SA, Khaw KT, Wareham NJ, Loos RJ (2009) Association of variants in the PCSK1 gene with obesity in the EPIC-Norfolk study. Hum Mol Genet 18: 3496–3501.
- 11. Chang YC, Chiu YF, Shih KC, Lin MW, Sheu WH, et al. (2010) Common PCSK1 haplotypes are associated with obesity in the Chinese population. Obesity (Silver Spring) 18: 1404–1409.
- 12.
NCBI website. Accessed 2011 Aug 22.
- 13. Parikh RM, Joshi SR, Menon PS, Shah NS (2007) Index of central obesity – A novel parameter. Med Hypotheses 68: 1272–1275.
- 14. Méthot J, Houle J, Poirier P (2010) Obesity: how to define central adiposity? Expert Rev Cardiovasc Ther 8: 639–644.
- 15. Nones K, Ledur MC, Ruy DC, Baron EE, Melo CM, et al. (2006) Mapping QTLs on chicken chromosome 1 for performance and carcass traits in a broiler x layer cross. Animal Genetics 37: 95–100.
- 16. Havenstein GB, Ferket PR, Scheideler SE, Rives DV (1994) Carcass composition and yield of 1991 vs 1957 broilers when fed “typical” 1957 and 1991 broiler diets. Poult Sci 73: 1795–1804.
- 17.
Kessler AM, Snizek PN Jr, Brugalli I (2000) Manipulação da quantidade de gordura na carcaça de frangos In: Anais da Conferência APINCO de Ciência e Tecnologia Avícolas. APINCO, Campinas, SP, Brazil Press.
- 18.
Chambers JR (1990) Genetics of growth and meat production in chickens. In Poultry breeding and genetics Edited by: Crawford RD. Amsterdam: Elsevier Press.
- 19. Rubin CJ, Zody MC, Eriksson J, Meadows JRS, Sherwood E, et al. (2010) Whole-genome resequencing reveals loci under selection during chicken domestication. Nature 464: 587–591.
- 20. Sabeti PC, Varilly P, Fry B, Lohmueller J, Elizabeth Hostetter E, et al. (2007) The International HapMap Consortium: Genome-wide detection and characterization of positive selection in human populations. Nature 449: 913–918.
- 21. Flori L, Fritz S, Jaffrézic F, Boussaha M, Gut I, et al. (2009) The genome response to artificial selection: a case study in dairy cattle. PLoS One 4: e6595.
- 22. Johansson AM, Pettersson ME, Siegel PB, Carlborg O¨ (2010) Genome-Wide Effects of Long-Term Divergent Selection. PLoS Genet 6(11): e1001188.
- 23. Barrett JC, Fry B, Maller J, Daly MJ (2005) Haploview: analysis and visualization of LD and haplotype maps. Bioinformatics 21: 263–265.
- 24. Sabeti PC, Reich DE, Higgins JM, Levine HZ, Richter DJ, et al. (2002) Detecting recent positive selection in the human genome from haplotype structure. Nature 419: 832–837.
- 25. Scheet P, Stephens M (2006) A fast and flexible statistical model for large-scale population genotype data: applications to inferring missing genotypes and haplotypic phase. Am J Hum Genet 78: 629–644.
- 26. Ma L, Runesha HB, Dvorkin D, Garbe JR, Da Y (2008) Parallel and serial computing tools for testing single-locus and epistatic SNP effects of quantitative traits in genome-wide association studies. BMC Bioinformatics 9: 315.
- 27. Mao Y, London NR, Ma L, Dvorkin D, Da Y (2006) Detection of SNP epistasis effects of quantitative traits using an extended Kempthorne model. Physiol Genomics 28: 46–52.
- 28. Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MAR, et al. (2007) PLINK: a toolset for whole-genome association and population-based linkage analysis. Am J Hum Genet 81: 559–575.
- 29.
Ensembl website. Accessed 2011 Aug 24.
- 30. Mutch DM, Clément K (2006) Unraveling the genetics of human obesity. PLoS Genet 2: e188.
- 31.
UCSC website. Accessed 2011 Aug 18.
- 32. Gabriel SB, Schaffner SF, Nguyen H, Moore JM, Roy J, et al. (2002) The structure of haplotype blocks in the human genome. Science 296: 2225–2229.
- 33. Ikeobi CO, Woolliams JA, Morrice DR, Law A, Windsor D, et al. (2002) Quantitative trait loci affecting fatness in the chicken. Anim Genet 33: 428–435.
- 34. Zhou H, Mitchell AD, McMurtry JP, Ashwell CM, Lamont SJ (2005) Insulin-like growth factor-I gene polymorphism associations with growth, body composition, skeleton integrity, and metabolic traits in chickens. Poult Sci 84: 212–219.
- 35.
Animal QTLdb website. Accessed 2011 Aug 29.