Genome-wide association studies on body-weight and body-size traits among Zhedong white geese based on a modified genotyping-by-sequencing method Authors: Yunzhou Yang, Shaojia Zhai, Haodong Liu, Huiying Wang, Daqian He Published: 2025 ABSTRACT BackgroundLCORL is a crucial gene that regulates body weight and size in several organisms, including the graylag geese (Anser anser). However, the effect of this gene in swan geese (Anser cygnoides) has been deemed negligible. As various indigenous goose breeds (A. cygnoides) are systemically selected for their growth performance, a comprehensive understanding of the genetic basis of growth performance-related traits is imperative for goose breeding.ResultsA goose double-digest genotyping-by-sequencing (ddGBS) protocol was established by implementing dual restriction enzyme (EcoRI and BfaI) digestion of genomic DNA, followed by the ligation of adapters with paired-end index sequences. The ddGBS method was used to perform genome-wide association studies on body weight in Zhedong white goose, with the aim to identify candidate genes. The novel genome-wide genotyping method detected > 100 K bi-allelic markers per sample and exhibited a high genotyping accuracy demonstrated by the Sanger sequencing method. Moreover, this strategy was used to conduct a longitudinal genome-wide association study (GWAS) on body-weight traits (0–10 weeks), which identified 19 significant loci. These loci exerted significant effects on body weight, but could only explain a small proportion of the phenotypic variation (≤ 0.51%). Nine candidate genes, including growth-related genes (e.g., THADA), were identified in the proximity of significant GWAS markers. The GBS-based GWAS on body size detected candidate variants associated with shank circumference. Each of the four significant single-nucleotide polymorphisms explained 0.57%–0.58% of the phenotypic variance. Furthermore, screening analyses of genomic selection signals between high- and low-body-weight groups at 10 weeks of age detected 754 potential loci under selection. However, no extremely divergent regions were identified in the genome.ConclusionsThe modified goose ddGBS approach provided an alternative genotyping workflow for geese, facilitating GWAS on quantitative traits. The preliminary findings of this study revealed that several genetic loci with minor effects, rather than single major genes, regulated body-weight and body-size traits in Zhedong white goose. Furthermore, short-term artificial selection pressure on body-weight traits could improve phenotypes. However, it mildly reshaped the genome on a large scale.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12864-025-12288-0. Background Although pork is the most commonly consumed meat in China, the consumption of poultry meat has been increasing in recent decades, and goose meat substantially contributes to the overall supply of poultry meat. At present, China has the largest population of meat-type geese, producing the highest amount of goose meat globally. The goose meat industry aims to enhance production efficiency by developing successful commercial lines similar to those used in broiler chickens [1]. Recently, to improve reproduction and growth performance, systematic breeding practices have been implemented among various indigenous goose breeds. Body-weight and body-size traits are the main candidates for research and selection among farm animals; selection based on the phenotypic values of these traits has been effective. Recent developments in molecular selection (e.g. genomic selection) has provided extremely useful tools for artificial selection, resulting in improvement of candidate traits [2]. The essential procedure in molecular selection was the genotyping of genetic variants that are highly related to phenotypes. Specifically, molecular selection was performed to establish a molecular dictionary that records the effects of genome-wide variants on phenotypes (the reference population). It is axiomatic that increasing prediction accuracy becomes the first task in molecular breeding practices. By achieving this goal, one approach is to incorporate variants identified through genome-wide association studies (GWAS) into the prediction marker set and assign them a higher weighting. This method has demonstrated its ability to enhance the predictive power of molecular selection [3]. Although molecular selection for body weight and size in geese has not yet been initiated, pilot GWA studies on candidate traits are expected firstly, then providing valuable variants that can be used in future genomic selection applications. Zhedong white goose, a meat-type breed in China, is popular among farmers and consumers owing to its early sexual maturity, high-quality meat, and rapid and early growth rate. However, the major genes or genomic regions regulating variations in body weight and size in Zhedong white goose have not been identified. This limits further genetic resource conservation and molecular selection. A previous study detected genes or gene networks that regulate variations in body weight in Zhedong white geese via comparative transcriptome sequencing, including IGF1, GHRHR, and NCAPG-LCORL [4]. These genes have also been found in other species due to their essential roles in regulating body weight [5–8]. Among them, LCORL was believed to be the major gene modulating the body weight of Zhedong white goose. Specifically, variants in LCORL were responsible for >26% body weight variance [8]. However, the LCORL gene only had minor effects on the body weight in the Zhedong white goose [9, 10]. Thus, to detect more candidate genes and genomic fragments, a comprehensive assessment of the genetic basis of body-weight and body-size traits in Zhedong white geese is required. Performing such an investigation can help the development of a high-yielding, high-quality domestic goose breed. These efforts can promote the development of the goose industry, contributing to an efficient meat supply. Elucidating whole genome-wide distributed variants is essential for screening and identifying candidate genes or genomic regions that regulate quantitative traits [11, 12]. Recently, there have been advancements in technologies, such as whole-genome resequencing (in-depth or low-coverage) and genotyping chips (solid or liquid). However, the high cost associated with large-scale sequencing and genotyping for breeding populations is substantial, as numerous individuals are required to obtain precise breeding value estimation or to narrow down the regions of candidate genes [13]. The reduced representation genome sequencing technology has been proven to be a highly efficient and cost-effective genotyping method [14, 15]. In addition to providing sufficient genetic markers for GWAS and genomic selection, reduced representation genome sequencing can also decrease sequencing costs and the complexity of data analysis. Furthermore, several methodologies based on reduced representation genome sequencing principles have been developed [16, 17] and successfully applied to the study of several organisms—including pigs, ducks, and various crops [18–20]. It is hypothesized that the implementation of reduced representation genome sequencing in the context of domestic goose breeding will decrease genotyping cost and facilitate comprehensive analyses of the genetic mechanisms underpinning body-weight and body-size traits. This study aimed to develop a modified double restriction enzyme digest genotyping-by-sequencing (ddGBS) method and investigate the molecular basis of body-weight and body-size traits in Zhedong white geese using ddGBS-GWA. Methods Animals and samples This study used 200 Zhedong white goose geese (half male and half female) obtained from a breeding farm in Zhejiang Province. The gender of Zhedong white goose geese was determined via anal genital examination. The geese were all hatched on the same day and raised under the same rearing conditions. All geese have free access to feed and water under the natural light condition through the whole experiment period. Professionally trained personnel recorded the body measurements. At the end of the experimental period, blood samples (1 mL) were collected from the wing vein of each goose in a disposable blood collection tube and stored at 4 °C until DNA extraction. Moreover, body weight was recorded at 1 (BW1), 2 (BW2), 4 (BW4), 7 (BW7), 8 (BW8), and 10 (BW10) weeks of age. All geese were allowed to fast for 24 h. However, they had free access to water before measurement. At 10 week of age, the following eight phenotypes associated with goose body size were recorded: semi-submerged length (SSL), defined as the distance from the tip of the mouth to the midpoint of the hip connection; neck length, defined as the distance from the anterior edge of the first cervical vertebra to the base of the neck; body oblique length, defined as the distance between the shoulder joint and the ischial tuberosity on the body surface; keel length, defined as the distance from the front end of the keel process to the end of the keel on the body surface; shank circumference (SC), defined as the circumference of the thinnest part of the shank; chest width, defined as the distance between the shoulder joints measured using calipers; chest depth, defined as the distance from the first thoracic vertebra to the anterior edge of the keel on the body surface measured using calipers [21]; and middle toe length, defined as the length between the acrocheir and the end of the middle toe. Modified ddGBS Modifications in the Goose DdGBS library structure Based on previously reported methods [14, 24], a next-generation GBS framework was engineered via library indexing and adapter architecture optimization (Supplementary Figure S1). Major changes in library structures were as follows: (1) fragmentation of the genomic DNA occurred using double-restriction endonuclease (one enzyme producing longer adhesive ends and the second enzyme producing shorter ends); (2) specific barcodes were generated using GBSX [22], and barcodes were tagged in sequences with EcoRI overhangs; and (3) pair-ended adapters from the TruSeq DNA library construction technology were adopted for fragment enrichments and final library construction. Compared with the traditional GBS protocols, ddGBS was expected to exhibit enhanced multiplexing capabilities and similar genotyping quality; supplementary Table S1 provides the primer sequences used. Library Preparation and quality control Single-strand adapters tagged with barcodes and bases complementary to the overhangs of EcoRI and CviAII were synthesized by Sangon (Shanghai). These oligos were annealed to double-strand sequences in annealing buffer (Beyotime, D0251). Then, they were ligated to the digested fragments, and after restriction digestion, fragmented DNA was subjected to adapter ligation using T4 DNA ligase (New England Biolabs, M0202), with a precisely formulated mixture containing paired-end adapters. The library was amplified with TruSeq-compatible polymerase chain reaction (PCR) primers under optimized thermocycling conditions (13 cycles). Post-amplification purification and size selection were performed using SPRIselect bead-based fractionation (Beckman Coulter, B23319), selectively retaining DNA fragments within a range of 500–800 bp. The library quality was evaluated based on its concentration (determined using quantitative real-time PCR) and fragment length distribution (Agilent 2100-H, USA). Moreover, libraries with a concentration of > 2 nmol/µL and one main peak were subjected to paired-end sequencing (PE150) on the NovaSeq X Plus platform (Illumina). Variant calling pipeline and genotype validation Variant calling pipeline Read quality parameters, such as read length distribution, base quality, and GC content, were assessed using FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Furthermore, the function module of process_radtags from Stacks was implemented to split sequencing reads based on barcodes for each sample [25]. The Trimmomatic software was used to remove adapters and low-quality sequences (ILLUMINACLIP: adapters.fa:2:30:10 LEADING:9 TRAILING:9 SLIDINGWINDOW:4:15 MINLEN:100) [26]. To obtain the sorted binary alignment/map format (bam) files, the split sequencing data were aligned with the goose reference genome (GCF_040182565.1) [27, 28]. Next, the gstacks and populations modules of the Stacks software were used to call single-nucleotide polymorphisms (SNPs) based on the aligned reads. Genotype validations The reproducibility and accuracy of ddGBS sequencing was validated by sequencing multiple batches of the same individuals. Specific primers were designed to amplify target sequences around six polymorphic sites on five chromosomes (Chr) (Supplementary Table S2). The amplicons were subjected to Sanger sequencing (Sangon, Shanghai). Finally, the genotyping accuracy of ddGBS was determined based on its genotype consistency with Sanger sequencing experiments. Imputation for missing genotypes Before subsequent analyses, variants called from the ddGBS workflow should be filtered. Imputation was applied to increase the genotyping rate and decrease the missing rate. The raw vcf file was first fed to VCFtools to filter out genotypes with a missing rate of >0.6 and a minor allele frequency (MAF) of < 0.05 [29]. Then, variants with a low genotyping quality (GQ < 20) and read depth (DP < 10) were filtered out using BCFtools [30]. Thereafter, the BEAGLE (v4.0) software was used to impute missing genotypes with default values (java -jar beagle gt = original_vcf.gz out = imputed_vcf) [31]. Imputed SNPs were also subjected to second-round quality control. The quality of genotype imputation was mainly determined by the parameter of genotyping probability (GP) produced by BEAGLE (v4.0). If the fraction of genotypes for each variant had a GP of < 0.8 among >90% of the samples, the variant was considered as low quality and excluded. A home-in python script was written to finish the filtration steps [8]. Genome-wide association studies and estimation of genotype effects on phenotypes The EMMAX software was used to perform GWAS on body weight [32]. Weekly body weight and size were all analyzed using the following model: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y\;=\;X\beta\;+\;S_y\;+\;g\;+\;\varepsilon,$$\end{document} where y is the vector of the observed weekly body weight and size values. X is the n×1 confounder matrix (e.g., gender in this study). β and S are the coefficients of fixed effects and genotypes (only bi-alleles used in this study), and γ is the effect size for all SNPs. g is the n×1 random vector caused by genetic relatedness (g ~ N(0, σ2*K)), and K is the genetic relatedness matrix. The emmax-kin module in the EMMAX package was used to calculate the Balding–Nichols (BN) kinship matrix to produce K. SNPs located on sex chromosomes (Z and W) were not included. Suggested significance threshold was 1E-5 while Bonferroni significance threshold (0.05/142057) was also applied to GWA studies on body size traits due to the small sample size, aiming to mitigate risks of false positive signals. Phenotypic values due to different genotypes were also calculated after removing the gender effects by considering sex as an important covariate in a linear regression model. Specifically, the original phenotypes were first fit to sex in a general linear model (Yori~Sex). The model returned the residuals (Yres) that could not be captured by the gender effect. The residuals were generally considered as the real phenotypes mainly controlled by genotypes. The genotypes and returned residuals were fit into another model (Yres~SNP). The average phenotypic values of different genotypes explained variance for each locus was calculated using RStudio [33]. Genomic differentiation between sub-populations Geese with the highest and lowest BW10 values were selected and assigned to two groups. Each group comprised 10 male and 10 female geese. The two groups were used for comparative genomic analyses. For FST and allele frequency difference calculations, a certain degree of missing values in genotypes is permitted. To preserve sufficient markers as many as possible, the differentiation analyses did not employ imputation and second-round filtration as in GWA studies. As long as they passed the first-round quality control, they will be used for differentiation analyses. SNPs located on sex chromosomes (Z and W) were not included. The pairwise population differentiation levels (--weir-fst-pop) and allele frequency (--freq2) were calculated using VCFtools [29]. For the FST calculation, the values were 20-kb window and 10-kb step size between windows. Enzyme selection and restriction site prediction for GBS To account for sequencing costs and marker density, different combinations of high-frequency enzymes (such as MseI, BfaI, ApekI, and CviAII) and low-frequency enzymes (such as EcoRI and BglII) were selected to simulate the effects of enzyme digestion on the goose genome. GBSX software was used to simulate fragmentation for different restriction enzymes (perl GBSX.pl -d enzyme cut sequences -f reference genome) [22]. The GBSX software searches the cut sites along the reference genome based on the input sequences. The reference genome established from a swan goose breed was selected because it was a telomer-to-telomer (T2T) chromosomal genome assembly with fewer gaps [23]. Outputs from the above-mentioned commands contained the cut sites along the genome and the number of genomic fragments with different sizes. Based on the distribution of read numbers and lengths, candidate enzymes (New England Biolabs) were selected for subsequent verification experiments. Results Optimization and validation of DdGBS Goose genomic DNA digestion with different restriction enzyme combinations Based on the in silico simulation performed using the GBSX software, EcoRI and four other enzymes were selected as candidates based on their digested fragment distributions (Fig. 1A). Single or double enzymes can generate thousands of fragments. The combination of EcoRI and BfaI was used in this study (Fig. 1B). The fragments generated by this combination were mainly distributed within a range of 100–900 bp, with 60% distributed within a range of 500–800 bp. This fragment size range was suitable for library construction via the Illumina sequencing platform. The total number of fragments was moderate considering the coverage and sequencing cost. Furthermore, CviAII and MseI produced abundant fragments. However, CviAII is expensive and cannot be easily obtained from suppliers. MseI could not completely digest the genomic DNA (Supplementary Figure S2). Fig. 1Simulation and experimental validation of enzymatic digestion. (A) Digestion fragments statistics based on an in silico simulation. (B) The digestion of 200 ng of DNA by EcoRI and BfaI yields smeared small fragments. (C) EcoRI and CviAII combination completely digests the genomic DNA to small fragments Simulation and experimental validation of enzymatic digestion. (A) Digestion fragments statistics based on an in silico simulation. (B) The digestion of 200 ng of DNA by EcoRI and BfaI yields smeared small fragments. (C) EcoRI and CviAII combination completely digests the genomic DNA to small fragments Genotyping performance The fragment distribution exhibited a bell-shaped curve. The length of the main fragments was within a range of 564–645 bp (Supplementary Figure S3). After trimming the adapters and low-quality reads, approximately 1.38 GB (0.99–1.63 GB) of clean data was obtained in each sample, indicating an averaged sequencing depth of 1.14× on the genome scale. Moreover, the numbers of genotyped sites in four samples were 257,427, 122,895, 193,903, and 192,420 (DP = 10, Fig. 2A). Meanwhile, the number of common genotyped SNPs between two samples was within a range of 262,412–288,100 (87.67%–96.13%, Supplementary Table S3). The genomic interval between two neighboring SNPs was 3.72 kb on the genome scale (Supplementary Figure S4). Samples that were sequenced in different sequencing projects exhibited a high genotype concordance when the sequencing depth (DP) was 10 (> 98%, Fig. 2B and Supplementary Table S4). Genotyping results with different library PCR primers exhibited a high concordance for the same samples (> 91% at DP = 10, Fig. 2C). This concordance can be improved using a stringent DP threshold (e.g., > 95% at DP = 20). Fig. 2General statistics of genotype-by-sequencing (GBS) in four samples. (A) Single-nucleotide polymorphism (SNP) number based on different allele depths. (B) Genotype concordance of the same individuals from two sequencing trials. (C) Genotype concordance of same samples using different sequencing polymerase chain reaction (PCR) primers. DP: depth of the loci General statistics of genotype-by-sequencing (GBS) in four samples. (A) Single-nucleotide polymorphism (SNP) number based on different allele depths. (B) Genotype concordance of the same individuals from two sequencing trials. (C) Genotype concordance of same samples using different sequencing polymerase chain reaction (PCR) primers. DP: depth of the loci Genotyping quality and validation of random loci using Sanger sequencing The genotypes of the detected variants were verified using Sanger sequencing. Six pairs of designed primers had an optimal performance. The products of PCR with these primers were sequenced. The heterozygous and homozygous genotypes were selected. All sites that were sequenced had the same genotypes as those from GBS (Table 1). Table 1Genotypes of the loci from genotyping-by-sequencing (GBS) were validated using Sanger sequencingChromosomeChrPosRefAltSamplesGeno-GBSGeno-SangerNC_089873.1Chr169,130,392GAID1A/AA/AID2G/AG/AID3G/GG/GID4G/GG/GNC_089887.1Chr1213,118,660CTID1C/CC/CID2C/TC/TID3T/TC/CID4C/CNANC_089882.1Chr1014,533,258CTID2C/TC/TID1C/TC/TID3C/CC/TID4C/TC/CNC_089885.1Chr137,651,611TCID1T/TT/TID2T/TT/TID3C/CT/TID4T/TNANC_089876.1Chr47,854,120CTID1C/CC/CID2C/CC/CID3T/TC/CID4C/CNA6,322,926GAID1G/GG/GID2A/AA/AID3A/GA/GID4A/GA/G Table 1Genotypes of the loci from genotyping-by-sequencing (GBS) were validated using Sanger sequencingChromosomeChrPosRefAltSamplesGeno-GBSGeno-SangerNC_089873.1Chr169,130,392GAID1A/AA/AID2G/AG/AID3G/GG/GID4G/GG/GNC_089887.1Chr1213,118,660CTID1C/CC/CID2C/TC/TID3T/TC/CID4C/CNANC_089882.1Chr1014,533,258CTID2C/TC/TID1C/TC/TID3C/CC/TID4C/TC/CNC_089885.1Chr137,651,611TCID1T/TT/TID2T/TT/TID3C/CT/TID4T/TNANC_089876.1Chr47,854,120CTID1C/CC/CID2C/CC/CID3T/TC/CID4C/CNA6,322,926GAID1G/GG/GID2A/AA/AID3A/GA/GID4A/GA/G Genotypes of the loci from genotyping-by-sequencing (GBS) were validated using Sanger sequencing GWAS on weekly body weight Statistics of DdGBS sequencing and genotyping among Zhedong white Goose This study examined 174 samples (male: 104, female: 70) with BW1, BW2, BW4, BW7, BW8, and BW10 data to GBS. The body-size traits of 103 samples (male: 59, female: 44) were examined and recorded. Average number of bases for each sample was 1,996,371,465, indicating a sequencing depth of 1.62× compared with the reference genome. The gstacks software detected polymorphisms at 1,961,411 loci, with a high missing rate among samples. The first-round filtration kept 198,526 biallelic SNPs which were subjected to subsequent imputation and second-round filtration. After removing low-quality imputation variants, 142,504 were retained and they were all genotyped among all samples. Among them, 3,818 variants were on Z chromosome and were not kept for subsequent GWAS. The same samples were sequenced twice using different index primers, and genotypes were also verified using Sanger sequencing. Results revealed high consistency and reliability (Supplementary Table S5–S6). GWAS on weekly body weight of Zhedong white Goose A longitudinal GWAS of Zhedong white goose genotype using the ddGBS approach revealed a weight-associated genomic architecture. For BW1, BW2, BW4, BW7, BW8, and BW10 (Supplementary Table S7), 19 genome-wide loci significantly associated with body weight were identified on Chr1, Chr2, Chr3, Chr8, Chr11, and Chr12 (Fig. 3A and F). Chr 1, 2, 8, and 11 were detected twice for body-weight traits. Meanwhile, other Chr (such as Chr3, Chr5, and Chr12) were only detected once for BW4 and BW8. The number of SNPs significantly associated with BW7 was 7. This value was higher than that associated with other traits (Fig. 3D; Table 2). Fig. 3Genome-wide association studies on body weight in Zhedong white geese. (A–F) Manhattan plots (left panels) and quantile-quantile (Q-Q) plots (right panels) for body weight at 1, 2, 4, 7, 8, and 10 weeks of age. Chromosomal positions are displayed on the X-axis, while the P values are shown on the Y-axis. The red dashed line was the suggested significance threshold (-log10(1E-5)) Genome-wide association studies on body weight in Zhedong white geese. (A–F) Manhattan plots (left panels) and quantile-quantile (Q-Q) plots (right panels) for body weight at 1, 2, 4, 7, 8, and 10 weeks of age. Chromosomal positions are displayed on the X-axis, while the P values are shown on the Y-axis. The red dashed line was the suggested significance threshold (-log10(1E-5)) Phenotypic variance explained by candidate variants A simple general linear model was used to fit genotypes to the weekly body weight. Table 2 depicts the phenotypic variation explained by the model for each locus (See phenotypic values due to genotypes in Table S8). All 19 SNPs explained 0.11%–0.51% variations on weekly body weight phenotypes, with most explaining 0.11%–0.2% of variations. The SNPs on Chr1 explained the most phenotypic variations of BW7 (all > 0.24%). Neighboring SNPs on the same Chr explained almost the same phenotypic variations due to a high linkage disequilibrium (LD) (Supplementary Table S8). Beta column was the SNP effects estimated using the EMMAX software. Table 2Genome-wide significant single-nucleotide polymorphisms (SNPs) associated with age-stratified body weightAge (Weeks)ChrPosition (bp)SNP Ann.♀P-valueNearest GeneBetaEx. Var. (%)#BW11204,428,021rs_1_2044280213.46E-0611.920.1143227,883,399rs_2_278833993.12E-0611.190.1192227,889,404rs_2_278894047.02E-0610.280.1097BW21114,162,361rs_11_141623611.30E-06 UBE2Q2 −39.250.1209BW4328,243,354rs_3_282433548.61E-06 THADA 87.750.1022BW7124,948,378rs_1_249483781.77E-06 PTPRZ1 −123.48640.5111129,853,356rs_1_298533567.81E-06 DOCK4 −75.89530.3051129,856,618rs_1_298566187.50E-06 DOCK4 −82.66220.30651187,811,641rs_1_1878116415.21E-06−38.70110.24091187,811,658rs_1_1878116585.21E-06−38.70110.2409210,356,247rs_2_103562471.43E-06 PTPRN2 −171.98760.1942252,475,310rs_2_524753108.36E-06−48.37460.1531BW8548,980,334rs_5_489803349.49E-06 TDP1 341.55790.147810,287,984rs_8_102879847.70E-07 TADA1 234.53920.1321825,626,201rs_8_256262016.93E-06 AGBL4 −156.88110.1212118,530,326rs_11_85303269.35E-06 TICRR 282.420.1251BW1025,559,444rs_2_55594442.82E-06199.73470.109525,559,529rs_2_55595292.69E-06198.7320.108125,623,937rs_12_56239375.85E-06−417.10.1059♀ The SNP nomenclature comprises Chr number and positions, starting with the letters rs. For example, rs_1_204428021 indicates an SNP at the position of 204,428,201 bp on Chr1. # The phenotypic variance explained by the SNP Table 2Genome-wide significant single-nucleotide polymorphisms (SNPs) associated with age-stratified body weightAge (Weeks)ChrPosition (bp)SNP Ann.♀P-valueNearest GeneBetaEx. Var. (%)#BW11204,428,021rs_1_2044280213.46E-0611.920.1143227,883,399rs_2_278833993.12E-0611.190.1192227,889,404rs_2_278894047.02E-0610.280.1097BW21114,162,361rs_11_141623611.30E-06 UBE2Q2 −39.250.1209BW4328,243,354rs_3_282433548.61E-06 THADA 87.750.1022BW7124,948,378rs_1_249483781.77E-06 PTPRZ1 −123.48640.5111129,853,356rs_1_298533567.81E-06 DOCK4 −75.89530.3051129,856,618rs_1_298566187.50E-06 DOCK4 −82.66220.30651187,811,641rs_1_1878116415.21E-06−38.70110.24091187,811,658rs_1_1878116585.21E-06−38.70110.2409210,356,247rs_2_103562471.43E-06 PTPRN2 −171.98760.1942252,475,310rs_2_524753108.36E-06−48.37460.1531BW8548,980,334rs_5_489803349.49E-06 TDP1 341.55790.147810,287,984rs_8_102879847.70E-07 TADA1 234.53920.1321825,626,201rs_8_256262016.93E-06 AGBL4 −156.88110.1212118,530,326rs_11_85303269.35E-06 TICRR 282.420.1251BW1025,559,444rs_2_55594442.82E-06199.73470.109525,559,529rs_2_55595292.69E-06198.7320.108125,623,937rs_12_56239375.85E-06−417.10.1059♀ The SNP nomenclature comprises Chr number and positions, starting with the letters rs. For example, rs_1_204428021 indicates an SNP at the position of 204,428,201 bp on Chr1. # The phenotypic variance explained by the SNP Genome-wide significant single-nucleotide polymorphisms (SNPs) associated with age-stratified body weight ♀ The SNP nomenclature comprises Chr number and positions, starting with the letters rs. For example, rs_1_204428021 indicates an SNP at the position of 204,428,201 bp on Chr1. # The phenotypic variance explained by the SNP Effects of genotypes on body weight This study investigated the effects of different genotypes in all 19 loci detected using GWAS on body weight. Original body weight can be significantly biased by genotype effects. Therefore, the residuals of body weight from the regression model were used as adjusted phenotypic values. This will also generate negative values for body weight. For BW1, one locus at 204,428,021 on Chr1 and two loci at 27,883,399 and 27,889,404 on Chr2 were identified. The two SNPs on Chr2 were tightly linked to each other. However, only one was displayed here. The SNPs rs_1_204428021 and rs_2_27883399 exerted significant effects on body weight. The reference alleles of these two SNPs (T and A) increased the body weight by 23.37 g. Meanwhile, the alternative alleles (C and G) decreased the body weight by 23.78 g (Fig. 4A and B). Fig. 4Genotypic effects of partial loci on weekly body weight. The y-axis values were calculated following the general regression model (See materials and methods). Genotypes were encoded as 0, 1, and 2. Three numbers referred to homozygotes of the reference allele, heterozygotes, and homozygotes of alternative alleles. One genotype for the other 10 loci was harbored by one or two samples. They are shown in Supplementary Table S7 Genotypic effects of partial loci on weekly body weight. The y-axis values were calculated following the general regression model (See materials and methods). Genotypes were encoded as 0, 1, and 2. Three numbers referred to homozygotes of the reference allele, heterozygotes, and homozygotes of alternative alleles. One genotype for the other 10 loci was harbored by one or two samples. They are shown in Supplementary Table S7 For BW2, BW4, and BW10, the number of loci detected was lower. Nevertheless, they exerted large effects. The animals with the TT genotype at rs_11_14162361 were 79.33 g (BW2) heavier than those with the CC genotype (Fig. 4C). At the loci rs_3_28243354 and rs_2_5559444, the homozygotes of alternative alleles (TT and GG) increased BW4 and BW10 by 172.33 and 387.72 g, respectively (Fig. 4D and I). BW7 and BW8 were controlled by 7 and 4 loci, respectively (Table 2). Due to a low MAF, only two SNPs for each trait are shown in Fig. 5. Both SNPs controlling BW7 were located on Chr2. The homozygotes of alternative alleles (2: GG in Figs. 4E and 42: AA in Fig. 4F) increased the body weight by 1119.94 and 856.01 g, respectively. For the loci rs_8_25626201 and rs_8_10287984, the homozygotes of alternative alleles (2: GG in Figs. 2 and 4G: AA in Fig. 4H) decreased BW8 by 273.72 and 520.06 g, respectively. Fig. 5Genome-wide association (GWA) studies on body size phenotypes among Zhedong white geese. The significance threshold in the multiple comparisons was determined using the Bonferroni correction method owing to the small sample size. Stringent criteria can mitigate false positive signals. Red and blue dashed lines suggest the significance threshold ((-log10(1E-5)) and the Bonferroni correction significance threshold ((-log10(0.05/142057)), respectively Genome-wide association (GWA) studies on body size phenotypes among Zhedong white geese. The significance threshold in the multiple comparisons was determined using the Bonferroni correction method owing to the small sample size. Stringent criteria can mitigate false positive signals. Red and blue dashed lines suggest the significance threshold ((-log10(1E-5)) and the Bonferroni correction significance threshold ((-log10(0.05/142057)), respectively GWAS of body size among Zhedong white Goose Detection of genomic regions associated with eight body-size traits Among the 103 samples, 142,057 SNPs were used for GWAS on eight body-size traits and the mapping of significant variants for the SC trait. Six variants on Chr1, Chr2, Chr12, and Chr18 were found to be significantly related to SC variations (Fig. 5C). These variants were located at 200,582,099 on chr1 (rs_1_200582099), 3,106,288 on chr12 (rs_2_3106288), 2,703,645 on chr18 (rs_18_2703645), and 55,980,310, 56,036,725, and 56,077,718 on chr2 (rs_2_55980310/56036725/56077718). Of these, three SNPs (rs_2_55980310/56036725/56077718) were in one LD block together with 16 other SNPs. Most LD values (R2) between markers in this block were > 0.92 (Supplementary Table S9). Effects of significant markers on the SC trait Six SNPs were significantly related to SC. The effects of four variants (rs_1_200582099, rs_12_3106288, rs_18_2703645, and rs_2_55980310) on the SC trait were calculated. These four markers explained > 0.5% of SC variations. For the markers of rs_1_200582099, rs_18_2703645, and rs_2_55980310, the reference alleles significantly decreased (Fig. 6A and D) or increased (Fig. 6B) the SC values (P < 0.01). Fig. 6Genotypic effects of partial loci on the shank circumference (SC). The y-axis values were calculated using formula 1 (See the materials and methods section). Two other markers on chromosome 2 were tightly linked to rs_2_55980310 and exerted similar effects Genotypic effects of partial loci on the shank circumference (SC). The y-axis values were calculated using formula 1 (See the materials and methods section). Two other markers on chromosome 2 were tightly linked to rs_2_55980310 and exerted similar effects Potential genomic regions under artificial selection Differential regions between the high-BW10 (4764.73 ± 332.46 g) and low-BW10 (3457.79 ± 350.26 g) were investigated based on allele frequency difference values and population differentiation index (FST). Of 198,526 SNPs, 198,207 were located on autosomes and they were used for differentiation analyses. The limited numbers of high signals were detected on some Chr (e.g., Chr5). However, the allele frequency difference values between the high-BW10 and low-BW10 groups exerted rapid effects of artificial selection on alleles (Fig. 7A). On 12 Chr, 23 SNPs with an allele frequency difference of > 0.5 were detected (Supplementary Table S10). An analysis of population differentiation showed that the effects of artificial selection on body weight altered the differentiation on a genome-wide scale, as 754 SNPs with high FST values, accounting for 1% of total SNPs, were distributed on 28 Chr (Supplementary Table S11). None of the Chr exhibited significant differentiations between the high-BW10 and low-BW10 groups. However, the genomic region at 168.43–177.68 Mb on Chr1 exhibited consecutive FST value changes (pink shadowed region in Fig. 7B). Fig. 7Effects of artificial selection on the allele frequency and population differentiation. A: Differential allele frequencies between the high-BW10 and low-BW10 groups. B: Differential population differentiation levels between the high-BW10 and low-BW10 groups. Pink rectangle in Fig. 7B refers to the region between 168,430,001 and 177,680,001 bp on chromosome 1. The dashed line refers to a FST value of 0.149 which detected 754 SNPs Effects of artificial selection on the allele frequency and population differentiation. A: Differential allele frequencies between the high-BW10 and low-BW10 groups. B: Differential population differentiation levels between the high-BW10 and low-BW10 groups. Pink rectangle in Fig. 7B refers to the region between 168,430,001 and 177,680,001 bp on chromosome 1. The dashed line refers to a FST value of 0.149 which detected 754 SNPs Discussion Herein, the general GBS library architecture was updated, and a genotyping pipeline was constructed for geese using two restriction enzymes (EcoRI and BfaI). Consequently, GWA studies on body-weight traits were successfully conducted. Identification of genome-wide variants with high-accuracy using Goose DdGBS Compared with previously reported reduced representation genome sequencing methods, this newly modified goose ddGBS method offers typical advantages. First, the cost-effectiveness of the reagents used is notable. Compared with the whole-genome sequencing (WGS)-based genotyping method, ddGBS requires significantly fewer sequencing data for reliable variant calling, which indicates a significantly lower cost. Further, ddGBS experiments are performed without the need for specialized supplies, a feature which is important for large-scale genotyping in farm animal genomics and genomic selection studies. Second, the distribution of markers across the genome is characterized by a high density. The utilization of EcoRI/BfaI combinations facilitated the identification of over 100,000 SNPs. Surprisingly, it ensured a high locus reproducibility (up to 96% common loci across samples; Supplementary Table S3) and achieved small variant intervals (~ 3.75 kilobases). Third, the high reliability and accuracy of ddGBS are supported by consistent genotyping across sequencing batches and validation via Sanger sequencing in domestic geese. Compared with chicken ddGBS that was previously published, the adoptions of widely used paired-end TruSeq primers and complementary bridge primers, along with goose ddGBS barcodes, provided more convenience in terms of library construction and higher multiplexity potentials. This enhanced ddGBS technique is consequently a dependable, precise, and economic solution for capturing genomic representativeness, thereby enabling genetics and genomics studies in geese. Identification of significant but different genomic regions via a ddGBS-based GWAS on goose weekly body weight and size traits Based on the variants detected using the modified GBS method, a GWAS on body weight and body size was conducted. This study identified 19 loci significantly correlated with body weight and five loci with body size. The loci varied in terms of different traits. Among the seven most significant loci for BW7, five were on Chr1. The growth rate of geese was highest at 7 and 8 weeks of age (Table 2 and Supplementary Table S6). Thus, more genes may be involved in supporting rapid growth and development. Based on previous studies, Chr1 harbored some genomic fragments that regulated body weight. The region between 43.07 and 194.63 Mb on chicken Chr1 harbored body weight-related QTLs. In this region, SLC25A30 [34], NEK3 [34], REELT [35], FAM168A [35], PPFIA2 [36], and DUSP6 [36] were related to life-span, body weight, feed efficiency, carcass performance, and body size. Moreover, advanced intercross lineages of chickens with different founders have both genomic regions on Chr1, exerting remarkable effects on the body weight of chicken [37–39]. Further studies with a large sample size must be performed to determine the exact positions regulating goose body weight on Chr1. Previous studies have reported candidate genes regulating goose body weight and size. The GH, LDLRAD4, and GPR180 variants were related to body weight in graylag and swan goose breeds [21, 40, 41]. This study detected THADA, which has been reported to be associated with goose SSL. However, the correlation of THADA with body weight was not previously investigated [21]. In this study, GWAS on SSL did not capture THADA, which may be attributed to small sample size and breed differences. However, THADA may be an important gene for regulating goose growth and development. This is because an SNP (rs_3_28243354), which was in the 28th intron of THADA, could increase BW4 by 172.33 g. The SNP of rs_18_2703645 was very close to variants between 2.86 and 6.65 Mb on Chr18 which have been demonstrated as significant ones regulating body weight and body size traits among 120-days-old lion-head geese [42]. To some extent, similar results obtained from different researches verified the reliability of our study based on the ddGBS-GWAS. Minor effects of loci on body weight and size Body-weight and body-size traits are quantitative traits (widely also known as polygenic traits). However, the determination of major gene effects can help identify QTL even in small populations [43]. According to a previous study, LCORL exerted large effects on the body weight of the Landes goose. An SNP upstream of LCORL explained 26% of body weight variations at 70 days of age [44]. Conversely, this study did not detect any gene with large effects on goose body weight or body size. Each of the 19 loci could only explain < 1% of phenotypic variations. This trend was also confirmed by genetic effect sizes estimated using the EMMAX software. The values were distributed mainly around zero, detecting no evident outliers (Supplementary Figure S5). Currently, association studies on goose body weight hardly detected the presence of major genes among swan geese [21, 45], except the one in lion-head geese [42]. This raises the fundamental question of why few major effect genes can be detected in swan geese. Several reasons should be considered thoroughly. In addition to body weight and size, which are polygenic traits in Zhedong white geese controlled by micro-effect loci, allele frequency should also be considered. If alleles with large effects show extremely low frequencies in Zhedong white geese, they cannot be easily genotyped in small-size populations. Our previous study found an extremely low frequency of the large-effect allele located in the LCORL upstream [9]. ddGBS only sequences a small part of the genome (2.3% in maize [14] and 13% in duck [20]). This further reduces the chance to genotype the rare alleles even if they are present in the population. Second, an “omnigenic” model of complex traits was proposed several years ago [46]. This theory claimed that there is a cellular regulatory network that modulates complex traits. This network comprised core and peripheral genes. Peripheral genes affect phenotypes via the modification of core gene expression patterns through the highly interconnected networks. These peripheral genes spread along the whole genome with small effects. However, they outnumbered the core genes by hundred times, showing extremely small effects but contributing most part of the heritability. Based on this theory, genes with large effects cannot be detected unless core genes mutate and lead to dysfunction. Similar to other studies on growth performance, small-size population prevented us from finding expected genes. In the future, large-size populations plus different mapping methods (linkage analysis and haplotype-based GWAS) are still required as ddGBS can substantially reduce genotyping cost. Notably, the phenotypic variance estimated using our linear model should be cautiously considered because the omission of relatedness in our model might biased the explained variance estimation. However, the effects of specific markers estimated using the EMMAX software was convincing as main confounders were all adjusted. Luckily, primary conclusions of SNP effect estimation were consistent with between different methods (Table 2). Although few phenotypic variations could be explained, their effects should not be ignored. For example, the effects of homozygotes of reference alleles increased BW1 by 23.78 g, which was 12.78% of the mean BW1 (183.01 g in Supplementary Table S6). Thus, these can be considered as large-effect loci. Due to the small sample size, the exact average weekly body weight for each genotype at the same loci was not accurate. For the markers associated with SC, only three samples carrying certain homozygotes were identified. This led to a fact that the exact phenotype gains caused by advantageous alleles cannot be precisely estimated. Mild effects of intensive phenotypic selection on body-weight traits on the genome This study did not involve geese selection. However, sampling geese with the lowest and highest body weight (BW10) simulated a 10% selection intensity in male or female geese. Theoretically, genomic regions underlying growth and development should also be under selection [47, 48]. Via the investigation of genome-wide allele frequency changes or population differentiation levels, these body weight-regulating regions can be detected. Comparative analysis of the genomic changes between geese with high and low BW10 revealed several SNPs along the whole genome with large allele frequency differences or FST values. However, extremely significant genomic regions associated with body weight were not detected. This finding indicated that there were no major genes or variants that regulated body weight variations in Zhedong white geese [9]. Thus, artificial selection for increasing body weight did not significantly modify the genome architecture in specific regions within a short duration (e.g., one generation in this study) even in cases of high selection intensity (10%). Significant changes can be detected only in populations subjected to selection for more than a few decades [6, 38]. For the first few generations of selection, artificial selection exerted mild effects on the whole genome instead of a few local regions. The genomic region (168–171 Mb) on Chr1 of chickens harbor the most significant loci contributing to chicken body weight, and this finding was found in different populations [6, 38, 39]. Considering that QLTs were detected frequently in other species, it deserves our attention that there perhaps existed significant genetic loci controlling goose body weight. Compared with previous studies, the current study analyzed a limited number of samples, which decreased the power of GWAS on quantitative traits. However, this research provided some genomic coordinates of the causal genetic variants. The regions mapped using ddGBS-GWAS, allele frequency difference, and FST values must be validated in large and selected populations. Conclusions The modified GBS is a reliable and relatively cost-effective approach for developing genetic markers among geese. The high genotyping accuracy and repeatability and the selection of restriction enzymes in this approach can increase the probability of detecting novel polymorphic loci that could not be identified using detailed WGS. In addition, the modified GBS approach can be applied in goose genetics research and molecular breeding. In this study, the modified GBS approach was used to detect genetic loci associated with body-weight and body-size traits in Zhedong white goose. The key loci influencing body weight varied at each stage of growth. Therefore, the findings of this study must be validated using a large sample size. Supplementary Information Supplementary Material 1. Supplementary Material 1. Supplementary Material 2. Supplementary Material 2.