Background

Ducks are frequently raised as poultry for the consumption of their meat and eggs by humans. Duck meat is commonly acknowledged for its rich flavor, high amino acid and polyunsaturated fatty acid content, and comparatively low-fat levels [1, 2]. Breeders consider traits related to growth and body weight composition to be the most critical in broiler ducks [3]. Meat production and growth traits are influenced by numerous polygenic QTLs, with each QTL making a small contribution to the trait [4, 5]. To overcome the challenge, genome-wide associated studies (GWAS) have been widely used to identify genetic loci affecting complex traits such as growth rate and meat production. Currently, several candidate loci associated with economic traits have been identified [6, 7]. For example, the variants of IGF2BP1 determined the variation in body size and feed efficiency [8]. However, how to mechanistically decipher these candidate loci contributing to agronomic traits in ducks is a challenge. Nowadays, with the development of molecular phenotypic measurements, we can elucidate the genetic architecture of complex traits using these molecular phenotypes, such as gene expression information [9, 10].

Analyzing expression quantitative trait loci (eQTLs) is the most effective approach for evaluating how sequence variants influence gene expression within their native genomic and cellular environments [11]. This approach has been extensively documented in the series of Genotype-Tissue Expression (GTEx) project [12,13,14] in human. Currently, several eQTL studies have been conducted on farm animals, including cattle [15], pigs [16], sheep [17], and chicken [18]. For example, they observed that the production traits related to eQTLs also affect the expression of MGST1 [19] and SLC37A1 [20] in dairy cattle. However, eQTL analysis has not been conducted in ducks. Therefore, it is desirable to know whether eQTLs within GWAS signals contributed to duck economic traits.

In this study, we hypothesized that the colocalization of eQTL mapping and GWAS strategy could help identify essential candidate genes and causal DNA variants vital to growth and carcass traits. We integrated GWAS results of duck carcass and growth traits with the transcriptomes from seven tissues to prioritize genes and variants that influence duck economic traits through transcriptome. This work provided a duck cis-eQTL catalog consisting of 820 samples from seven tissues. The study results provide valuable resources for understanding the genetic effects on the transcriptome and also suggest the underlying molecular mechanisms of potentially causal functional variants in duck economic traits.

Methods

RNA-seq analysis

We downloaded 820 RNA-Seq datasets from Sequence Read Archive (SRA; https://www.ncbi.nlm.nih.gov/sra) and BIGD (https://bigd.big.ac.cn/bioproject/; Table S1). The adaptor and low-quality reads were trimmed and adaptor by Trimmomatic v0.39 [21]. The clean reads were mapped to the duck reference genome (ZJU1.0, https://www.ncbi.nlm.nih.gov/datasets/genome/GCF_015476345.1/) using STAR aligner [22]. To quantify gene expression, we used the mapped reads to calculate the gene expression levels using transcript per million (TPM) by StringTie [23] based on the annotation of RefSeq (GCF_015476345.1). A gene was considered expressed if it had a TPM threshold of ≥ 0.1 in at least 20% of the samples [15].

Genotype imputation

PCR duplicates of STAR alignments were marked using MarkDuplicates in Picard (http://broadinstitute.github.io/picard/). We splited reads into exon segments and trimmed any sequences overhanging into the intronic regions using SplitNCigarReads modules of GATK v4.2.6.1 [24]. Then, we recalibrated base quality scores based on known genomic variants using BaseRecalibrator and ApplyBQSR modules of GATK. Then, we carried out joint-calling of all GVCF samples using the GenotypeGVCFs module of the GATK tool. The low-quality variants were filtered out using –filter-expression “FS > 30.0 & QD < 2.0”. We also check the variants consistency of different tissues from the same individual in PRJNA419583 study. The average of consistency was 99.6% (Table S2). The variants were imputed to DNA sequence variants level based on a multiple-breed reference panel consisting of 2215 Pekin ducks and 289 public individuals from Shaoxing ducks (n = 166) and wide mallards (n = 123) by Beagle 5.4 [25]. We removed variants with dosage R-squared (DR2) of less than 0.8, genotype call rates of less than 90%, minor allele frequency (MAF) of less than 0.05 or variants located outside of autosomes using PLINK v1.90 [26]. On average, we retained 1,112,867 autosomal variants with average DR2 of 0.92 for eQTL mapping.

Cis-eQTL mapping

Gene expression values were normalized across samples using the inverse normal transformation. To account for population effects, we incorporated principal components (PCs) of genotype into the eQTL analyses. The number of PCs selected for the analysis was determined by testing their significance using EIGENSTRAT v6.1.4 (https://alkesgroup.broadinstitute.org/EIGENSOFT/). We ultimately decided to use the first three PCs for the eQTL analysis, as the test was not significant (P-value = 0.061) when more than three PCs were included (Table S3). Additionally, in order to address hidden batch effects and other sources of technical or biological variation, we estimated latent covariates for gene expression levels in each tissue using the Probabilistic Estimation of Expression Residuals (PEER) [27]. We retained 15 PEER confounding variables because the posterior variances of factor weights were nearly at their minimum values (Fig. S1). For blood tissue, we also conducted cis-eQTLs using 20 PEER confounding variables (https://github.com/WentaoCai/Duck_eQTL_results), we identified 1029 eGenes with 20 peer variables compared to 1040 eGenes using 15 peer variables. The number of overlapped eGenes was 952 (Fig. S2). We defined potential cis-eQTLs as variants located within a 1 Mb proximity ups or downstream of the transcription start sites (TSSs) of genes [28]. The cis-eQTLs mapping was conducted using a nominal P-value threshold that corresponded to FDR ≤ 0.05 for each gene by fastQTL [29]. The scripts of data analysis can be accessed on https://wentaocai.github.io/eQTL-analysis/. The gene lists were subjected to GO and KEGG analyses using DAVID with a significance threshold set at a P-value ≤ 0.05 [30]. We estimated the replication rate of cis-eQTL across different tissues using π1 statistic of the qvalue R package [31].

GWAS and Colocalization

The phenotype of 50 growth and carcass traits were collected from 941 Pekin XMallard F2 individuals. The summary of phenotype records can be found in Table S4. The genotypes were conducted using resequencing technology on Illumina HiSeq X Ten, which was described in our previous study [8]. After mapping, call SNPs, and quality control, 13,064,619 variants for 941 individuals were obtained for GWAS. The phenotype was adjusted for covariates, including sex, feed room, and the first three PCs of genotypes. We used a linear mixed model for the association test of each SNP using the GCTA -mlma [32].

$${\varvec{y}}={\varvec{X}}{\varvec{\beta}}+{\varvec{u}}+{\varvec{\varepsilon}},$$

where y is adjusted phenotype; X is a vector of genotypes of a variant at the locus tested; β is the effect size of the variant; u is a vector of random polygenic effects ~ N (0, \({{\varvec{G}}{\varvec{\sigma}}}_{{\varvec{g}}}^{2}\)), where G is genomic relationship matrix constructed from all variants; ε is a vector of residual errors.

We conducted the colocalization analysis of GWAS variants and eQTLs using Coloc R package [33]. The variants with P-value < 5 × 10−5 and their neighboring ± 50 Kb extracted from GWAS summaries of 50 growth and carcass traits were obtained for colocalization analysis [34]. GWAS signals and eQTLs were considered colocalizing if a posterior probability (H4) of the shared signal was > 0.4 [35].

Results

Transcriptome profile of duck primary tissues

We obtained 22.5 billion clean reads from a total of 820 publicly available RNA-seq samples, including seven tissues (251 liver, 184 muscle, 120 blood, 71 adipose, 87 ovary, 66 spleen, and 41 lung) from 17 breeds (Table S1). Pekin ducks accounted for the largest proportion of samples (36.2%), indicative of their extensive economic significance worldwide. Approximately 88.9% of the total reads were successfully mapped to the reference genome (Table S1). Under the expression threshold of TPM ≥ 0.1 in at least 20% of the samples, 16,485 (65.7% of total genes) genes were expressed across tissues. Using PCA analysis, we can successfully distinguish samples according to different tissues and reconstruct the relationships between these tissues based on their expression levels (Fig. 1A). The expression profiles accurately reflected tissue types (Fig. 1B), which reaffirmed the high quality of these profiles and underscores their suitability for our subsequent analysis. We called variants from RNA-Seq samples and imputed each genotype using a reference population of 2,504 ducks. When applying PCA clustering to imputed genotypes, as anticipated, the samples exhibited clustering patterns corresponding to their respective breeds (Fig. 1C).

Fig. 1
figure 1

Principal component analysis (PCA) and hierarchical clustering of samples. A Sample clustering (n = 820) using PCA based on gene expression levels. B Hierarchical clustering of 820 samples using gene expression levels, sample clustering is affected by both tissues and breeds. C PCA of samples (n = 820) based on imputed genotypes

Full size image

Genetic effects on gene expression

Considering all tissues, our analysis detected a total of 113,374 cis-eQTLs associated with 12,266 genes. These genes represent approximately 54.8% of all autosomally expressed genes. The numbers of eGene (gene with significant cis-eQTLs) for liver, muscle, blood, ovary, adipose, spleen, and lung were 7,301, 5,706, 1,040, 2,390, 989, 368, and 775, respectively (Table 1). The most notable cis-eQTLs detected in the liver were found to influence the expression of TIMMDC1, a gene crucial for the assembly of the membrane arm of mitochondrial Complex I (Fig. 2A). The most significant cis-eQTLs of muscle affected expression of PHOSPHO2, involved in dephosphorylation (Fig. 2B). The Manhattan plots of cis-eQTLs for other tissues are shown in Figs. S3, S4, S5, S6 and S7. The cis-eQTL results of seven tissues are accessed on https://github.com/WentaoCai/Duck_eQTL_results.

Table 1 The summary of cis-eQTL results in seven tissues. The genes, with an expression threshold of TPM (transcripts per million) ≥ 0.1 in at least 20% of samples on autosomes, account for the expressed genes. The eGene is a gene with significant cis-eQTLs. The eVariant is a genetic variant regulated at least one gene
Full size table
Fig. 2
figure 2

Manhattan plot of cis-eQTLs. A The Manhattan plot illustrates the nominal P-value (y-axis) for all cis-eQTLs in liver (left). Points colored dark red or pink represents significant cis-eQTLs, whereas those in black or gray were SNPs that did not reach a significant threshold. The expression of TIMMDC1 with three genotypes in liver (right). B The Manhattan plot shows the nominal P-value (y-axis) for all cis-eQTLs in muscle (left). The expression of PHOSPHO2 with three genotypes in muscle (right)

Full size image

Character of cis-eQTLs

Consistent with previous work in humans [36], most of the significant duck cis-eQTLs clustered around the TSS of target genes (Fig. 3A). An average of 45.3% of significant cis-eQTLs fell within 100 kb around the TSS. We found an enrichment of low P values closer to TSSs, showing that cis-eQTLs are more likely to be located within this distance (Fig. 3B and Fig. S8). Cis-eGenes of most tissues exhibited significantly higher expression levels compared to non-eGenes (Wilcoxon test, Bonferroni adjusted P-value < 0.05), with the exception of lung and spleen tissue (Fig. 3C). To assess the sharing patterns of cis-eQTLs between tissues, we computed the π1 statistics for each pair of tissues. We observed that cis-eQTLs in blood exhibited lower conservation across different tissues., while cis-eQTLs from any tissue exhibited a strong sharing pattern to liver tissue (Fig. 3D). We detected that any two tissues shared 39.1% of eGenes (Fig. 3E), and 10.7% of eVariants (Fig. S9). The 244 eGenes shared by at least four tissues were involved in basic metabolic pathways, such as nucleoplasm, mitochondrion, cytosol, and ATP binding (Table S5). Interestingly, we detected that the expression of THNSL1, SF3B6, PPIL3, PHOSPHO2, LOC119713243, and LOC119717646 were affected by cis-eQTLs in almost all tissues except for spleen. On average, 10.7% of cis-eQTLs were involved in at least two eGenes.

Fig. 3
figure 3

The characters of cis-eQTLs. A The density of top significant cis-eQTLs for each tested gene around TSS. B The P-value distribution of top significant cis-eQTLs for each tested gene in liver. C The expression of eGenes of cis-eQTL compared to non-eGenes. The significant P-value were calculated using Wilcoxon test. D Pairwise sharing patterns of cis-eQTL (π1 value) across tissues. E The overlaps of eGenes across tissues

Full size image

Colocalization of cis-eQTLs and GWAS loci

We conducted GWAS between genomic variants and 50 growth and carcass traits in F2 ducks (Table S4). We identified a total of 2,136 unique variants that showed significant association (P < 5 × 10–8) with 47 agronomic traits, of which, 1,039 variants were associated with at least two traits (Table S6). We obtained the variants with P-value < 5 × 10−5 and their neighboring ± 50 Kb for colocalization analysis. We observed that the eQTLs for 49 eGenes in seven tissues were colocalized with 41 traits, resulting in a total of 94 tissue-gene-trait pairs (Fig. 4, Table 2, Fig. S10, and Table S5). In liver, we detected that eQTLs of SRSF4 (serine and arginine rich splicing factor 4) were colocalized with GWAS signals of jejunum length and total intestine length (Fig. 5A). The top GWAS signal of jejunum length significantly affecting the expression of SRSF4. The eQTLs of GSS (glutathione synthetase) were colocalized with GWAS signals of shank length (Fig. 5B). There was a strong LD (r2 = 1) between the top eQTLs signal and the top GWAS signal. The top GWAS signal of shank length is also significantly associated with the expression of GSS. Previously, we detected the variants located IGF2BP1 gene affecting the carcass traits and body size [8]. Here, we confirmed these GWAS signals can affect the phenotype of eviscerated weight, heart weight, neck length, and skeleton weight through the change of IGF2BP1 expression (Fig. 5C and Table S7).

Fig. 4
figure 4

The colocalized results between cis-eQTLs and GWAS signals. A Manhattan plot illustrates the colocalization results (H4 > 0.4) between liver eQTLs and GWAS signals. The x-axis represents the P-value of lead eQTLs (points) across traits in liver. The right labels are colocalized SNP-gene-trait pairs. The different categories of tissues are distinguished by colors (B) Manhattan plot illustrates the colocalization results (H4 > 0.4) between muscle cis-eQTLs and GWAS signals

Full size image
Fig. 5
figure 5

Examples of colocalized results between eQTLs and GWAS signals. A The colocalization between jejunum length GWAS and SRSF4 eQTLs of liver. The colors of the variants are determined based on their linkage disequilibrium values with the most significant variant. B The colocalization between shank length GWAS and GSS eQTLs of liver. C The colocalization between eviscerated weight GWAS and IGFBP1 eQTLs of liver. D The colocalization between BW28 (body weight at 28 days) GWAS and NDUFC2 eQTLs of muscle

Full size image
Table 2 The colocalized results of eGene-trait pairs. Colocalization analyses were conducted between cis-eQTLs and GWAS traits
Full size table

In muscle tissue, the eQTLs of NDUFC2 (NADH: ubiquinone oxidoreductase subunit C2) were colocalized with BW28 (body weight at 28 days; Fig. 5D). The top significant GWAS signals for BW28 and the top significant eQTLs of NDUFC2 were the same. It appeared that there were two independent loci in eQTL signals of liver GSS and muscle NDUFC2 (Fig. 5B and D). To confirm whether the eQTL signals were derived from the same causative variant, we repeated cis-eQTL mapping using the top eQTL signal of GSS or NDUFC2 as a covariate. No significant signals were detected using this conditional analysis (Fig. S11), indicating the eQTL signals of liver GSS and muscle NDUFC2 arise from one causative variant. The eQTLs of LOC113840004 were colocalized with 20 traits, implying this uncharacterized gene would be an important functional gene in ducks. In other tissues, we observed the eQTLs of EIF3 (E74 like ETS transcription factor 3) in adipose were associated with the GWAS of BW56 body weight at 56 days (Table S7). The eQTLs of RUNDC1 (RUN domain containing) in blood were detected colocalized with the GWAS of ileum length. The eQTLs of IPPK (inositol-pentakisphosphate 2-kinase) in ovary were detected colocalized with the GWAS of neck weight percentage. Overall, the colocalization analysis enhanced our ability to identify potentially causal genes and gain a deeper understanding of the genetic underpinnings of complex traits in ducks.

Discussion

In this study, we created a comprehensive catalog of eQTLs across multiple tissues, further expanding the list of candidate genes and potential variants affecting important agronomic traits in ducks.

Consistent with other species [36, 37], a high density of signals was observed near the TSSs of their respective genes. The majority of eQTLs appear to be tissue-specific, suggesting intricate and distinct genetic mechanisms governing gene expression across different tissues. The functional annotation of tissue-shared eGenes revealed that genetic variants more prominently and frequently impact genes related to immune and metabolic functions, which is consistent with previous findings in human [38] and cattle [37]. Similar to cattle [37], these cis-eGenes in duck were found to exhibit higher expression levels than non-eGenes in most tissues.

The colocalization analysis between GWAS signals and cis-eQTLs in the seven tissues facilitates the identification of causal genes for association signals that were previously unresolved. Several GWAS signals were observed to colocalize with cis-eQTLs. SRSF4 belongs to the family of Arginine-Serine-rich (SR) proteins, which play a crucial role in constitutive splicing and also regulate alternative splicing [39]. Like other SRSFs, SRSF4 shuttles between the nucleus and cytoplasm, and mediate mRNA regulation, including export, stability, and translation [40, 41]. SRSF4 is associated with human colon adenocarcinomas [42]. SRSF4 has a relatively high expression in the liver, stomach, and intestine [43]. The upstream variants of SRSF4 have significantly regulated the expression of SRSF4 in the liver for both cattle [15] and pigs [16]. GSS is responsible for catalyzing the condensation of gamma-glutamylcysteine and glycine to produce glutathione (GSH) [44]. GSH is essential for a multitude of processes, such as safeguarding cells from oxidative damage, facilitating amino acid transport, detoxifying foreign compounds, preserving protein sulfhydryl groups in a reduced state, and serving as a cofactor for several enzymes [45]. The mutations of GSS are associated with body height in humans [46, 47]. The length of the shank directly determines a poultry's height. The top GWAS signal of shank length is also significantly associated with the expression of GSS, which implies that GSS could be a candidate gene for the shank length and body height of the duck. IGF2BP1 belongs to the insulin-like growth factor 2 mRNA-binding protein family and plays a role in RNA transport, cell proliferation, differentiation, and metabolism by regulating the mRNA localization, stability, and translation of specific target genes. [48, 49]. IGF2BP1 is a well-known candidate gene for body size in ducks [8], chickens [50], and humans [51]. Here, we confirmed the variants of IGF2BP1 can regulate the duck body size by changing the IGF2BP1 expression in liver.

NDUFC2 is a subunit of the mitochondrial membrane respiratory chain NADH dehydrogenase (Complex I) [52]. A deficiency in NDUFC2 results in cellular-level mitochondrial dysfunction and increased oxidative stress [53]. NDUFC2 appears to be downregulated in the skeletal muscle cells of individuals with insulin resistance and is linked to insulin secretion in vivo [54, 55]. The mutations of NDUFC2 are associated with body mass index [56]. Both BW28 GWAS and NDUFC2 eQTLs shared the same of top significant variant, implying NDUFC2 may act as a candidate gene for duck body weight. ELF3, an important member of the E74-like transcription factor family, involves in inflammatory response [57] and adipogenic differentiation [58]. Leptin is a potent pro-inflammatory and pro-catabolic factor, and its downstream actions are mediated by ELF3 [59]. The mutations near ELF3 were associated with body mass index in humans [60]. The eQTLs of ELF3 in adipose were associated with BW56. RUNDC1 is an inhibitor of the tumor suppressor p53 [61], which can negatively modulate autophagy by blocking fusion between autophagosomes and lysosomes [62]. The eQTLs of RUNDC1 in blood were associated with ileum length. We can hypothesize that RUNDC1 is a candidate gene for ileum length. These results are crucial for gaining a deeper understanding of the molecular mechanisms underlying specific traits by considering gene expression and the functional characteristics of genes associated with these traits.

Although our study released thousands of eGenes by analyzing public RNA-seq data, there are still some limitations. For example, some tissues have limited sample sizes, resulting in the detection of a small number of eGenes in several tissues, such as spleen, lung, and adipose. The potential biases in variant detection, population stratification resulting from the inclusion of data from multiple breeds, and confounding factors stemming from diverse experimental designs may not have been completely resolved when using publicly available RNA-seq data. In addition, the variety and sample size of our multiple-breed reference panel for genotype imputation remain limited, which may introduce biases in the genotype results. This study serves as a pioneering investigation into the field of duck eQTLs. The summary statistics of duck eQTLs are still unavailable, which impedes a more detailed investigation of colocalized signals and thus an evaluation of possible pleiotropic effects. Our analysis of GWAS signals within the context of duck eQTLs serves as a preliminary foundation for future investigations in this field.

Conclusions

The interpretation of genetic mechanisms underlying complex traits based on the molecular phenotype of primary tissues in ducks was rare. Through the integration of eQTL and GWAS data, we have constructed a molecular QTL map in ducks, which aids in unraveling genetic association signals by identifying candidate genes, such as SRSF4, GSS, IGF2BP1, NDUFC2, ELF3, and RUNDC1. Moreover, these newly identified cis-eQTLs and candidate genes will enhance the accuracy of genomic prediction and contribute to the benefit of genetic improvement programs in duck breeding.