The influence of structural variants from 2445 pigs on gene expression and complex traits Authors: Zhongzi Wu, Lu Gui, Jianchao Hu, Xiaoyun Chen, Zhiyan Zhang, Jun Gao, Weiwei Liu, Lusheng Huang Published: 2026 ABSTRACT The impact of structural variants on gene regulation and complex traits in pigs remains poorly understood. Here, we present a resource comprising 60 long-read sequencing and 2445 high-quality short-read sequencing samples from 120 pig breeds/populations worldwide. These samples include 293 high-depth short-read sequencing samples and three long-read samples generated in this study. Integrating two graph-based genotyping approaches, we construct a structural variant map encompassing 158,288 deletions, 137,493 insertions, 660 inversions, and 10,063 duplications. We perform eQTL mapping in liver, skeletal muscle, backfat, and intramuscular fat tissues from an F6 hybrid pig family, finding that 49.01% of expressed genes are significantly associated with cis-region structural variants. We further explore the contribution of structural variants to domestication and population differentiation. We find a 53 bp insertion on chromosome 11 that significantly down-regulates GPC6 expression in liver tissue, potentially contributing to adaptive selection in northern Chinese pig populations. Additionally, FST analyses between miniature and commercial pig breeds identify candidate structural variant loci associated with body size. We next integrate results from structural variant-eQTL and GWAS to find a 50 bp deletion at chr4:75,613,295 that significantly reduces PLAG1 expression in skeletal muscle tissue and is associated with body weight at 240 days of age. Introduction Pigs (Sus scrofa) have a rich evolutionary history shaped by complex natural and anthropogenic processes. Following their divergence from ancient ancestors, pigs underwent independent domestication events in different regions, notably in East Asia and Europe. Subsequent intense artificial selection has driven the development of numerous breeds with highly distinct phenotypic traits1–3. For example, Tibetan pigs have adapted to the hypoxic conditions of high-altitude environments, whereas Min pigs in China and Russian wild boars exhibit strong cold resistance traits suited to extremely frigid climates. Notably, European commercial breeds, such as the Yorkshire and Landrace, have been optimized for rapid growth rate and large body size, underpinning modern industrial pork production. In contrast, many miniature pig breeds, including the Göttingen minipig and the Bama Xiang pig, have gained prominence for their small body size, making them as ideal companion animals and as valuable models for biomedical research4. Advances in pig genomics have substantially expanded our understanding of the genetic basis underlying these phenotypic diversities. High-quality reference genome assemblies and extensive population genomics analyses have facilitated the identification of numerous single nucleotide polymorphisms (SNPs) associated with economically and biologically important traits5,6. In addition to SNPs, structural variants (SVs), including deletions (DEL), insertions (INS), duplications (DUP), and inversions (INV), represent a major source of genomic variation that can reshape genome architecture and phenotypic traits in pigs7–10. For example, our previous study showed that a 2.3 kb deletion in the ABO gene reduces the concentration of N-acetyl-galactosamine in the intestinal tract and subsequently decreases the abundance of Erysipelotrichaceae11. In addition, a recent SV study of 1060 pigs identified a SINE insertion in the MYO5A gene that may alter splicing patterns and cause coat color variation12. These findings highlight the functional importance of SVs in shaping phenotypic diversity. Therefore, systematic analysis of porcine SVs is essential to reveal genetic mechanisms driving trait diversity, advance complex trait understanding, and enable precision breeding. Recent technological advances, particularly in third-generation sequencing (TGS), have markedly improved our ability to detect and characterize SVs. Unlike next-generation sequencing (NGS) technologies, which rely on short reads and are often limited in resolving large and complex genomic rearrangements, TGS platforms such as PacBio and Oxford Nanopore provide much longer reads that enable more accurate and comprehensive SV discovery13. Furthermore, graph-based approaches for genome representation and analysis have emerged as powerful tools for detecting and genotyping SVs across populations, offering a more complete view of genomic diversity beyond linear reference biases. These methods have proven effective in capturing previously undetectable variants and understanding their functional impacts14,15. Graph-based SV analysis is emerging as a powerful approach in porcine genetics, offering important insights into the genetic architecture underlying complex traits and adaptive evolution9,16–18. However, current studies are limited by insufficient sample sizes, inadequate breed diversity, and, in particular, a scarcity of high-quality third-generation sequencing data, which severely restricts the comprehensive discovery of SVs. In addition, the lack of integrated multi-omics datasets, including genomics, transcriptomics, and phenomics, further hampers in-depth functional characterization of SVs. In this study, we leveraged a large collection of pig short-read sequencing (SRS) datasets, combined with long-read sequencing data and graph-based structural variant genotyping strategies, to construct a comprehensive atlas of SVs across diverse pig breeds. We further integrated expression quantitative trait locus (eQTL) analyses across multiple tissues to explore the regulatory roles of SVs, and investigated their contributions to domestication, environmental adaptation, and complex trait formation. Our work serves as a valuable resource for understanding the structural basis of genetic and phenotypic diversity in pigs and highlights the underappreciated functional importance of SVs in the pig genome. Results Sample characteristics and SV reference panel in the Thousand Pig Genome Project Building upon the first phase of the Thousand Pig Genome Project (TPGP), which focused on key SNP and InDel variants across 1103 pigs19, we expanded the dataset by incorporating 1322 publicly available and 293 newly generated high-depth NGS samples. After stringent quality control and deduplication, the final dataset comprises 2486 individuals, including 2445 Sus scrofa representing 120 breeds or populations and 41 outgroup individuals from the Suidae family representing nine populations. Specifically, the panel includes 733 (29.49%) Asian domestic pigs, 111 (4.58%) Asian wild boars, 983 (39.54%) European domestic pigs, 39 (1.45%) European wild boars, 579 (23.29%) Eurasian hybrid pigs, and 41 (1.65%) outgroup samples (Fig. 1A; and Supplementary Data 1). Both neighbor-joining (NJ) tree and principal components analysis (PCA) based on genome-wide variants recapitulate known breed structures and geographic stratification patterns (Fig. 1A, B). To further enhance genome representation, we also integrated 57 publicly available and 3 newly generated long-read sequencing (LRS) datasets from 39 breeds/populations (Fig. 1A; and Supplementary Data 2). Compared to existing pig SV related studies, our resource includes the largest and most comprehensive sample set to date (Fig. 1C).Fig. 1Geographic distribution and sample summary of 2445 pigs and 41 outgroup individuals.A World map illustrating the geographic origins of 129 breeds or populations, comprising 120 pig breeds and 9 outgroups. Bar plots indicate sample sizes for each breed. Red stars denote breeds with at least one available LRS data. Left panel shows a neighbor-joining tree constructed using genome-wide SNP and InDel variants from all 2486 samples. Geographic visualizations were generated in R (v4.2.0) using the ggplot2 and maps packages. The basemap was derived from the “world” dataset provided by the maps package in R. B Principal component analysis based on genome-wide SNP and InDel variants across all samples. C Comparison of sample sizes between this study, the Pig1K Phase I project19, and other representative pig structural variation studies12. ASW, Asian wild boar; ASD, Asian domedtic pig; CEC, China-Europe crossbred; EUD, European domestic pig; EUW, European wild boar; OUT, Outgroups. Geographic distribution and sample summary of 2445 pigs and 41 outgroup individuals. A World map illustrating the geographic origins of 129 breeds or populations, comprising 120 pig breeds and 9 outgroups. Bar plots indicate sample sizes for each breed. Red stars denote breeds with at least one available LRS data. Left panel shows a neighbor-joining tree constructed using genome-wide SNP and InDel variants from all 2486 samples. Geographic visualizations were generated in R (v4.2.0) using the ggplot2 and maps packages. The basemap was derived from the “world” dataset provided by the maps package in R. B Principal component analysis based on genome-wide SNP and InDel variants across all samples. C Comparison of sample sizes between this study, the Pig1K Phase I project19, and other representative pig structural variation studies12. ASW, Asian wild boar; ASD, Asian domedtic pig; CEC, China-Europe crossbred; EUD, European domestic pig; EUW, European wild boar; OUT, Outgroups. To comprehensively identify structural variants, we employed two complementary graph-based genotyping pipelines. First, SVs were identified from 60 LRS genomes to generate a high-confidence callset, which was subsequently genotyped across all SRS samples using Paragraph20. Second, we implemented a short-read-based SV detection pipeline (Manta21 + Graphtyper222) independent of LRS data. The union of these SVs provides a robust foundation for downstream regulatory and population genomic analyses (Fig. 2A). From the LRS data, we detected an average of 32,533 deletions, 35,152 insertions, 29 duplications, and 260 inversions per sample, with Asian pigs exhibiting a higher SV burden than European pigs, likely due to reference bias and elevated genetic diversity in Asian populations (Fig. 2B). To benchmark detection strategies, we compared SVs identified from NGS, LRS, and assembly-based methods in a Meishan pig23, finding that LRS captured ~97.79% more SVs than SRS, while assembly-based calls were limited due to haplotype incompleteness23 (Figure S1A, S1B). The merged high-confidence SV library from 60 LRS samples contains 300,934 non-redundant SVs, including 114,053 deletions, 184,293 insertions, 670 duplications, and 1918 inversions (Figure S1C). Benchmark analysis revealed that the precision and recall of the SV library reached 0.946 and 0.943, respectively (Fig. 2C). Length distribution analyses revealed enrichment peaks at ~300 bp and ~7 kb for deletions and insertions, respectively, corresponding to SINE and LINE elements, supporting the biological credibility of the callset (Fig. 2D). In addition, 908 super large structural variants (>10 kb) were also identified in this SV library, some of which overlap with protein-coding genes (Figure S1D).Fig. 2Workflow for graph-based SV detection and the description of SV reference panels.A Overview of the SV detection and analysis pipeline employed in this study. Silhouettes of the world map were generated in R (v4.2.0) using the ggplot2 and maps packages. The basemap was derived from the “world” dataset provided by the maps package in R. B Number of SVs identified from 60 LRS samples, including 36 from Asian pig populations, 14 from European populations, and 10 from Eurasian mixed pigs. The figure above illustrates the cumulative distribution curve, while the figure below presents individual statistics. C Concordance between the SV reference panel derived from this study (PB, PacBio sequencing) and the SV calls generated by another sequencing platform (ONT, Nanopore sequencing). Analyses were performed on three independently sequenced pig genomes (n = 3). Each point represents one genome analyzed across the two platforms. Boxplots indicate the 25th, 50th (median), and 75th percentiles, and whiskers extend to a maximum of 1.5 times the interquartile range. D Length distribution of SVs in the high-confidence reference panel. Workflow for graph-based SV detection and the description of SV reference panels. A Overview of the SV detection and analysis pipeline employed in this study. Silhouettes of the world map were generated in R (v4.2.0) using the ggplot2 and maps packages. The basemap was derived from the “world” dataset provided by the maps package in R. B Number of SVs identified from 60 LRS samples, including 36 from Asian pig populations, 14 from European populations, and 10 from Eurasian mixed pigs. The figure above illustrates the cumulative distribution curve, while the figure below presents individual statistics. C Concordance between the SV reference panel derived from this study (PB, PacBio sequencing) and the SV calls generated by another sequencing platform (ONT, Nanopore sequencing). Analyses were performed on three independently sequenced pig genomes (n = 3). Each point represents one genome analyzed across the two platforms. Boxplots indicate the 25th, 50th (median), and 75th percentiles, and whiskers extend to a maximum of 1.5 times the interquartile range. D Length distribution of SVs in the high-confidence reference panel. Comprehensive structural variation atlas derived from 2445 pig genomes Across the 2445 Sus scrofa genomes, we identified 40,539,082 SNPs, 4,491,905 InDels, 158,288 deletions, 137,493 insertions, 10,063 duplications, and 660 inversions. Mendelian error rates estimated from 47 trios were low for SNPs/InDels (0.004–0.005%) and acceptable for SVs (0.52–3.86%), with duplications exhibiting the highest error rate but still below 5% (Fig. 3A), demonstrating reliable SV genotyping. Further analysis revealed that this SV atlas achieved a precision of 0.84 and a recall of 0.68, resulting in an F1 score of 0.75, which was substantially higher than that obtained using NGS data alone for SV detection (F1 score = 0.53) (Figure S2A, B). Moreover, the results of the NJ tree and PCA based on SV are similar to those of SNP (Figure S3). The reliability of SV genotyping was further supported by randomly selected PCR validations (Figure S4, and Supplementary Data 3). On average, each genome carried 22,667 deletions, 17,706 insertions, 531 duplications, and 75 inversions. Cumulative count curves showed that SV discovery has reached saturation, underscoring the value of large-scale WGS (Fig. 3B, and Figure S5A). We found that 38.39% SVs are common variants (MAF > 0.05, Figure S4C) and 42.46% of SVs overlapped with genes, among which 1.21% were related to the coding sequence (CDS) region (Figure S5B, S6A). Meanwhile, we also observed a number of SVs that overlapped with the genomic positions of various epigenetic status (Figure S6B). These results suggest that the SVs may have significant functional impacts.Fig. 3Comprehensive SV landscape identified from 2445 pigs.A Left panel: The number of different types of variations. Right panel: Mendelian error rates evaluated using 47 trios. Boxplots indicate the 25th, 50th (median), and 75th percentiles, and whiskers extend to a maximum of 1.5 times the interquartile range. B Distribution of SV counts across different pig populations. Asian pigs, European pigs, and their Eurasian hybrid populations. C Circos plot illustrating genome-wide SV density. From outer to inner tracks: chromosomes, inversions, duplications, insertions, and deletions. D Concordance between SVs identified in this study and those from public databases, including SVs aggregated from four major datasets12,16,28,36. E Total heritability estimation of 14,835 genes in 266 liver samples from the F6 population. Three groups (SNPs/InDels only, SVs only, SVs and SNPs/InDels) were compared for this study. Comprehensive SV landscape identified from 2445 pigs. A Left panel: The number of different types of variations. Right panel: Mendelian error rates evaluated using 47 trios. Boxplots indicate the 25th, 50th (median), and 75th percentiles, and whiskers extend to a maximum of 1.5 times the interquartile range. B Distribution of SV counts across different pig populations. Asian pigs, European pigs, and their Eurasian hybrid populations. C Circos plot illustrating genome-wide SV density. From outer to inner tracks: chromosomes, inversions, duplications, insertions, and deletions. D Concordance between SVs identified in this study and those from public databases, including SVs aggregated from four major datasets12,16,28,36. E Total heritability estimation of 14,835 genes in 266 liver samples from the F6 population. Three groups (SNPs/InDels only, SVs only, SVs and SNPs/InDels) were compared for this study. Genome-wide mapping revealed a non-uniform distribution of SVs, with a significant hotspot identified on chr5:39.75-40.50 Mb (n = 312, Fold Change=3.39, Pchi-square test = 2.22e-16). Deletions and insertions were particularly enriched toward chromosomal ends (Fig. 3C). Comparison with existing SV datasets revealed that 45.11% of the SVs identified here were absent from previously reported SV catalogs, with the greatest contribution from Asian wild pigs (Fig. 3D, and Figure S6C). To evaluate linkage disequilibrium (LD) between SVs and nearby SNPs, we calculated pairwise LD (r²) within ±100 kb windows. Notably, 43.38% of SVs were incompletely tagged by surrounding SNPs (LD, r² <0.8), whereas only 7.93% of SNP–SNP pairs showed such incomplete tagging, indicating that a substantial fraction of SVs captures genetic variation that is less well represented by nearby SNPs24 (Figure S7). Further analysis demonstrated that incorporating SVs increased the estimated global heritability of expression for 14,835 genes in 266 liver tissues25 from the F6 population compared with models including SNPs/InDels alone (mean heritability: 0.583 vs. 0.514; Wilcoxon rank sum test, P = 2.2e-16). These results suggest that SVs captured in this study help explain a substantial proportion of the “missing heritability” not accounted for by conventional variant types (Fig. 3E). Impact of structural variations on gene expression across liver, muscle, backfat, and intramuscular fat tissues in pigs To investigate the regulatory impact of SVs on gene expression, we conducted cis-eQTL analyses using WGS and RNA-seq data from 292 individuals in a hybrid family (Supplementary Data 1 and Supplementary Data 4). This study included four metabolically and economically relevant tissues: liver, longissimus dorsi muscle (LD), backfat, and intramuscular fat (IMF) (Fig. 4A; Supplementary Data 5-8). After stringent quality control, a total of 101,828 high quality SVs and 17,736 expressed gene were used to perform eQTL mapping. We identified 68,309, 42,657, 5985 and 8480 significant SV-eGene associations (P < 1e-6) in liver, LD, IMF, and backfat, respectively. On average, ~15,000 genes were expressed per tissue, with 43.01% exhibiting cis-regulatory signals, including 23.51% significantly associated with cis-acting SVs. Overall, ~49.01% (8693 vs 17736) of expressed genes were significantly associated with nearby SVs (Fig. 4B).Fig. 4Impact of SVs on gene expression across multiple pig tissues.A Overview of the SV-eQTL analysis strategy in four tissues(muscle, liver, backfat, and intramuscular fat) using a hybrid family design. B Summary of the number of expressed genes, eGenes, and SV-associated eGenes (SV-eGenes) identified in each tissue. C–F Comparison of P value between top associated SNPs (top eSNPs) and top associated SVs (top eSVs) in liver (C), muscle (D), intramuscular fat (E), and backfat (F). Points above the diagonal in red indicate that the lead eSV has a more significant P-value than the corresponding lead eSNP, whereas points in blue indicate the opposite. G Linkage disequilibrium structure surrounding a 16-kb DUP on chromosome 14 and its association with BMPR1A expression in muscle tissue. Boxplots indicate the 25th, 50th (median), and 75th percentiles, and whiskers extend to a maximum of 1.5 times the interquartile range. H Linkage disequilibrium structure surrounding a 560-kb DUP on chromosome 8 and its association with KIT gene expression in backfat tissue. Boxplots indicate the 25th, 50th (median), and 75th percentiles, and whiskers extend to a maximum of 1.5 times the interquartile range. Impact of SVs on gene expression across multiple pig tissues. A Overview of the SV-eQTL analysis strategy in four tissues(muscle, liver, backfat, and intramuscular fat) using a hybrid family design. B Summary of the number of expressed genes, eGenes, and SV-associated eGenes (SV-eGenes) identified in each tissue. C–F Comparison of P value between top associated SNPs (top eSNPs) and top associated SVs (top eSVs) in liver (C), muscle (D), intramuscular fat (E), and backfat (F). Points above the diagonal in red indicate that the lead eSV has a more significant P-value than the corresponding lead eSNP, whereas points in blue indicate the opposite. G Linkage disequilibrium structure surrounding a 16-kb DUP on chromosome 14 and its association with BMPR1A expression in muscle tissue. Boxplots indicate the 25th, 50th (median), and 75th percentiles, and whiskers extend to a maximum of 1.5 times the interquartile range. H Linkage disequilibrium structure surrounding a 560-kb DUP on chromosome 8 and its association with KIT gene expression in backfat tissue. Boxplots indicate the 25th, 50th (median), and 75th percentiles, and whiskers extend to a maximum of 1.5 times the interquartile range. Notably, fine-mapping analyses identified hundreds of eSVs (62−205 per tissue) with posterior probabilities exceeding 0.5 (Supplementary Data 5-8). And we also found dozens of lead SVs (58−93 per tissue) exhibited stronger association signals than the corresponding lead SNPs across tissues (Fig. 4C-F). For instance, a 16 kb duplication on chromosome 14 was significantly associated with BMPR1A expression in muscle (P = 1.51e-14; PPH4 = 0.99), showing a stronger signal than the lead SNP at the locus (Fig. 4G). BMPR1A has been linked to growth and developmental traits26. Similarly, a 560-kb duplication on chromosome 8 significantly influenced KIT expression in backfat tissue (P = 7.54e-26; PPH4 = 0.5; Fig. 4H). KIT has been previously implicated in coat color variation27. Importantly, comparison analysis of pure black and pure white pigs in F6 populations supports the role of KIT gene copy number as a genetic basis for coat color (Figure S8A). Meanwhile, three publicly available Hi-C datasets representing three different pig breeds8 also support the above-mentioned connection between 560-kb duplication and coat color (Figure S8B–D). We next characterized general properties of eSVs. eSVs were enriched near eGene bodies, with frequency declining with increasing distance (Figure S9A). Duplications were overrepresented among eSVs, highlighting their potential as regulatory elements. Additionally, 36.14% of eSVs modulated two or more eGene, and 82.99% of genes were regulated by multiple SVs. Finally, we assessed tissue specificity, finding that 70.01% (n = 37,921) of eSVs and 81.23% (n = 81,245) of eSV–eGene pairs were tissue-specific, while only 3.75% (n = 2,029) eSVs and 1.81% (n = 1806) of eSV–eGene pairs were shared across all four tissues (Figure S9B-D). Population-differentiated SVs reveal signatures of domestication and environmental adaptation in pigs Owing to the breadth and depth of our genomic dataset, we systematically investigated allele frequency shifts of structural variants associated with domestication and environmental adaptation (Fig. 5A, B). In the comparison between Asian wild boars (n = 109) and Asian domestic pigs (n = 535), we identified 181 significantly differentiated SVs (dSVs, top 0.1% FST; Supplementary Data 9) and 17,622 significantly differentiated SNPs (dSNPs), which associated with 1294 nearby protein-coding genes. Functional enrichment pointed to processes including immune regulation and protein tyrosine kinase activity (Figure S10A). Top-ranking dSVs included a 287 bp deletion at chr4:82,722,164 (SFT2D2, FST = 0.929), a 1163 bp deletion at chr15:94,385,174 (PMS1, FST = 0.883), and a 319 bp insertion at chr9:91,997,173 (NUP42, FST = 0.849) (Fig. 5A). Previous work from our group has suggested that NUP42-associated SVs may play a role in organogenesis during pig development28.Fig. 5Role of SVs in pig domestication and environmental adaptation.A Genomic distribution of highly differentiated SVs between Asian domestic pigs and Asian wild boars. The green and yellow dots represent SVs, and the gray dots represent SNPs/InDels. B Genomic distribution of highly differentiated SVs between cold-adapted large-bodied pigs from northern China (CL) and heat-adapted large-bodied pigs from southern China (HL). C A 300 bp INS at chr4:82,709,026 is significantly associated with SFT2D2 expression in liver tissue (n = 266). Pie charts show the genotype frequencies of this INS in Asian domestic pigs and wild boars. D A 53 bp INS at chr11:61,847,761 is significantly associated with GPC6 expression in liver tissue (n = 266). Pie charts indicate genotype frequencies of this INS in the northern Chinese pigs and other Chinese pig populations. E A 3577 bp DEL at chr14:12,5621,895 is significantly associated with TRUB1 expression in muscle tissue (n = 289). Pie charts display genotype frequencies of this DEL in the northern Chinese pigs and other Chinese pig populations. Boxplots indicate the 25th, 50th (median), and 75th percentiles, and whiskers extend to a maximum of 1.5 times the interquartile range. Statistical significance was assessed using a Wald test. Role of SVs in pig domestication and environmental adaptation. A Genomic distribution of highly differentiated SVs between Asian domestic pigs and Asian wild boars. The green and yellow dots represent SVs, and the gray dots represent SNPs/InDels. B Genomic distribution of highly differentiated SVs between cold-adapted large-bodied pigs from northern China (CL) and heat-adapted large-bodied pigs from southern China (HL). C A 300 bp INS at chr4:82,709,026 is significantly associated with SFT2D2 expression in liver tissue (n = 266). Pie charts show the genotype frequencies of this INS in Asian domestic pigs and wild boars. D A 53 bp INS at chr11:61,847,761 is significantly associated with GPC6 expression in liver tissue (n = 266). Pie charts indicate genotype frequencies of this INS in the northern Chinese pigs and other Chinese pig populations. E A 3577 bp DEL at chr14:12,5621,895 is significantly associated with TRUB1 expression in muscle tissue (n = 289). Pie charts display genotype frequencies of this DEL in the northern Chinese pigs and other Chinese pig populations. Boxplots indicate the 25th, 50th (median), and 75th percentiles, and whiskers extend to a maximum of 1.5 times the interquartile range. Statistical significance was assessed using a Wald test. Environmental adaptation in latitudinal differentiation in northern Chinese pigs reveal distinct genetic and phenotypic signatures shaped by regional selective pressures. A total of 174 dSVs and 21,522 SNPs (top 0.1% FST; Supplementary Data 10) were identified between cold-adapted large-bodied pigs from northern China (CL, n = 124) and heat-adapted large-bodied pigs from southern China (HL, n = 262), with 1057 protein-coding genes located within their flanking regions. Enrichment analyses suggested involvement in regulation of hormone levels and energy metabolism (Figure S10B). The most differentiated SV was a 98 bp insertion within the first intron of GPC6 on chromosome 11 (FST = 0.81). Notably, we detected a cluster of highly differentiated SVs surrounding GPC family on chromosome 11, a gene family previously associated with fat deposition in multiple species29–31, suggesting a potential role in porcine thermoregulation (Fig. 5B). To further elucidate the potential functional relevance of domestication- or environment-associated dSVs, we integrated them with cis-eQTL data derived from four tissues (Supplementary Data 9-10). Notably, 16.57%–41.76% of dSVs were significantly associated with the expression of at least one gene in at least one tissue, suggesting a widespread regulatory impact. For instance, a 300 bp insertion at chr4:82,709,026, highly differentiated between Asian domestic pigs and Asian wild boars, was significantly associated with the expression of SFT2D2 in liver, IMF and muscle tissue (FST = 0.909; PLiver = 2.91e-31; PIMF = 1.07e-17; PLD = 1.55e-27). SFT2D2 encodes a membrane-associated protein implicated in vesicular transport and lipid trafficking, processes essential for intracellular signaling and metabolic regulation (Fig. 5C). Another example is a 53 bp insertion at chr11:61,847,761, which shows strong allele frequency differentiation between northern and southern Chinese pig populations and is significantly associated with liver expression of GPC6 (FST = 0.647; PLiver = 3.74e-23) (Fig. 5D). Further analysis revealed that this INS is located in the intergenic region upstream of GPC6 and overlaps with the enhancer sequence in liver tissue32, suggesting that it may influence enhancer activity and thereby regulate GPC6 expression (Figure S11). GPC6, a member of the glypican family, modulates key signaling pathways involved in cell proliferation and metabolism, and the PheWAS results from PigBiobank33 suggest that GPC6 is significantly correlated with backfat thickness (P = 1.33e-5, Supplementary Data 11). Therefore, its expression modulation may underlie adaptive responses to environmental stressors such as temperature variation. In addition, we also identified a 3.5-kb length upstream deletion of TRUB1 that is highly differentiated between northern and southern pig populations and significantly modulates TRUB1 expression in muscle (Fig. 5E). PheWAS analyses further implicate TRUB1 in the regulation of palmitoleic acid content (P = 1.63e-7, Supplementary Data 11), suggesting a potential role of this structural variant in lipid metabolism–related traits. Structural variants contribute to body size determination in domestic pigs In contrast to the large body size of commercial pig breeds, numerous miniature pig breeds worldwide exhibit markedly reduced adult body weights ranging from 25 to 75 kg, making them well-suited as companion animals or biomedical models. To explore whether SVs contribute to the genetic architecture underlying body size variation in pigs, we performed a population differentiation analysis between 305 miniature pigs and 913 large-bodied commercial pigs using FST. A total of 225 SVs and 18,762 SNPs showed strong population stratification (top 0.1% FST; Supplementary Data 12), with putative functional relevance. GO and KEGG enrichment analyses based on 800 nearby genes revealed a significant overrepresentation of terms related to regulation of protein kinase activity, and neuro-regulatory processes (Figure S12). Notably, the most highly differentiated SV was a 316 bp deletion located in the first intron of the PLAG1 gene on chromosome 4. This deletion was associated with both PLAG1 expression in skeletal muscle tissue (PeQTL = NA; Pwilcox.test(0/0 vs 1/1) = 3.40e-08) and 240-day body weight (Pwilcox.test(0/0 vs 1/1) = 0.0029) in the F6 population (Fig. 6A-D). Consistent with our findings, previous studies have implicated this deletion as a likely contributor to body size variation in pigs16. The traceability analysis revealed that this 316 bp SINE/tRNA insertion originating in Asian northern wild pig genomes subsequently introgressed into European commercial pig populations, likely selected for improvement of body size traits. Among the top ten SVs ranked by FST, the flanking regions of several SVs harbored candidate genes previously linked to body size regulation, including PLAG1, LCORL, and CDK1034,35 (Fig. 6A), further supporting their functional relevance in growth trait evolution.Fig. 6Role of SVs in determining body size traits in pigs.A Genome-wide SV-based FST analysis between miniature pigs and commercial pig populations to identify candidate SVs potentially associated with body size. The green and yellow dots represent SVs, and the gray dots represent SNPs/InDels. B Local linkage disequilibrium pattern surrounding a 316 bp DEL locus on chromosome 4. C A 316-bp DEL at chr4:75,648,976 is significantly associated with PLAG1 expression in muscle tissue (n = 289) of the F6 chimeric population. D The same 316-bp DEL is significantly associated with body weight at 240 days of age in the F6 chimeric population (n = 886). E Genome-wide SV-based GWAS analysis of body weight trait in F6 population. F Co-localization analysis of SV-GWAS and muscle tissue SV-eQTL in F6 population. G A 50 bp DEL at chr4:75,613,295 is significantly associated with PLAG1 expression in muscle tissue (n = 289) of the F6 hybrid population. H The same 50 bp DEL is significantly associated with body weight at 240 days of age in the F6 hybrid population (n = 886). Boxplots indicate the 25th, 50th (median), and 75th percentiles, and whiskers extend to a maximum of 1.5 times the interquartile range. Statistical significance was assessed using a two-tailed Wilcoxon test. Role of SVs in determining body size traits in pigs. A Genome-wide SV-based FST analysis between miniature pigs and commercial pig populations to identify candidate SVs potentially associated with body size. The green and yellow dots represent SVs, and the gray dots represent SNPs/InDels. B Local linkage disequilibrium pattern surrounding a 316 bp DEL locus on chromosome 4. C A 316-bp DEL at chr4:75,648,976 is significantly associated with PLAG1 expression in muscle tissue (n = 289) of the F6 chimeric population. D The same 316-bp DEL is significantly associated with body weight at 240 days of age in the F6 chimeric population (n = 886). E Genome-wide SV-based GWAS analysis of body weight trait in F6 population. F Co-localization analysis of SV-GWAS and muscle tissue SV-eQTL in F6 population. G A 50 bp DEL at chr4:75,613,295 is significantly associated with PLAG1 expression in muscle tissue (n = 289) of the F6 hybrid population. H The same 50 bp DEL is significantly associated with body weight at 240 days of age in the F6 hybrid population (n = 886). Boxplots indicate the 25th, 50th (median), and 75th percentiles, and whiskers extend to a maximum of 1.5 times the interquartile range. Statistical significance was assessed using a two-tailed Wilcoxon test. To prioritize variants with regulatory potential, we integrated these high-FST SVs with the cis-eQTL datasets from four tissues. This analysis identified 101 dSVs that were significantly associated with the expression of at least one gene in one or more tissues. (Supplementary Data 12). Of particular interest, we revealed that a 65 bp insertion located in chr6:333,242 is on the first intron of the CDK10 gene, significantly affecting the expression of the CDK10 gene in liver tissue (Fst = 0.961; Pliver = 1.01e-17) (Figure S13A). Another case is that one 220 bp insertion (chr18:50,807,395) affects the expression of the NUDCD3 gene in muscle, liver, and IMF, thereby influencing body size differentiation (Fst = 0.953; PLD = 4.82e-17, PIMF = 6.43e-16, PLiver = 5.38e-19, Figure S13, Supplementary Data 12). PheWAS analyses further implicate NUDCD3 in the regulation of loin muscle area (P = 9.67e-9, Supplementary Data 11)33. Finally, genome wide SV-GWAS analysis based on the adult weight of 886 samples also identified the significant QTL region on chromosome 4, overlapped with the above-mentioned significant differentiation region (Fig. 6E). We discovered the highest significantly associated SV is a 50 bp deletion located 33.29 kb upstream of PLAG1 that was in strong linkage disequilibrium with the aforementioned 316 bp deletion (LD > 0.8; Fig. 6B). Moreover, the posterior probability of colocalization analysis at this site is 0.86 (Fig. 6F). And this 50 bp DEL exhibited even stronger associations with PLAG1 expression in muscle (PLD = 3.14e-10) and the strongest associations with 240-day body weight (PGWAS = 9.02e-08; Fst = 0.956) in the F6 population (Fig. 6G-H). Further analysis revealed that this SV can explain 15.19% of the phenotypic variations of the body weight trait in F6 population. Therefore, these results nominate the 50 bp deletion as a promising candidate SV influencing body size in pigs and highlight the regulatory complexity of the PLAG1 locus. Discussion Structural variants represent a substantial and underexplored source of genomic diversity, particularly in species subject to strong artificial selection, such as pigs. In this study, we constructed a comprehensive high-confidence SV resource by integrating 60 LRS samples and 2445 SRS individuals across wild and domestic pig populations, enabling population-scale characterization of SV distribution, differentiation, and potential functional impacts. By further combining this resource with multi-tissue SV-eQTL analyses in an F6 hybrid pig family, we show that ~49.01% of expressed genes are regulated by cis-acting SVs. Compared with previous pig genomics studies12,28,36, our work is distinguished by its larger sample size, broader breed representation, improved accuracy of SV detection, and the integration of multi-tissue transcriptomic resources. Collectively, our findings underscore the critical roles of SVs in shaping genomic architecture and complex trait evolution during pig domestication and breed diversification. At the population level, we show that SVs contribute to domestication and environmental adaptation through differentiated allele frequencies coupled with regulatory effects. A representative example is a 53 bp insertion upstream of GPC6 on chromosome 11, which exhibits strong allele frequency differentiation between northern and southern Chinese pig populations and significantly downregulates GPC6 expression in liver tissue. This insertion overlaps a liver enhancer element, suggesting a plausible regulatory mechanism. GPC6 encodes a glypican family proteoglycan involved in modulating Wnt, Hedgehog, and BMP signaling pathways, which play key roles in metabolism and energy homeostasis37,38. Consistent with this functional context, previously studies associate GPC6 with backfat thickness and lipid metabolism30,33, suggesting that SV-mediated modulation of hepatic GPC6 expression may contribute to metabolic adaptation to environmental factors such as temperature variation. A similar pattern is observed for a 3.5 kb upstream deletion affecting TRUB1, which shows strong population differentiation between northern and southern pigs and significantly alters TRUB1 expression in skeletal muscle. TRUB1 encodes a tRNA pseudouridine synthase involved in RNA modification, a regulatory layer increasingly linked to metabolic control39,40. Its association with palmitoleic acid content further suggests that SV-mediated regulation of TRUB1 may influence lipid metabolism and regional adaptive traits. Beyond environmental adaptation, SVs identified in this study play prominent roles in body size divergence between miniature pig breeds and commercial pigs. By integrating genome-wide SV differentiation with SV-eQTL and SV-GWAS analyses, we identified 101 differentiated SVs significantly associated with gene expression in one or more tissues. Among these, a 50 bp deletion near PLAG1 emerges as a particularly strong candidate. PLAG1 is a well-established regulator of growth and body size in livestock species41,42. In our analyses, this deletion exhibits extreme population differentiation, shows stronger associations with PLAG1 expression in skeletal muscle and body weight at 240 days of age. Additional SVs illustrate the diverse regulatory mechanisms through which structural variation influences growth. A 65 bp intronic insertion within CDK10 shows strong population differentiation and significantly affects CDK10 expression in liver tissue. CDK10 encodes a cyclin-dependent kinase involved in cell cycle regulation and cellular proliferation. Notably, loss-of-function mutations in CDK10 have been shown to cause severe growth retardation and developmental defects in both humans and mice, highlighting its essential role in organismal growth35. Nevertheless, the molecular mechanisms through which structural variation in CDK10 affects body size variation in pigs remain to be clarified. Similarly, a 220 bp insertion associated with NUDCD3 expression across muscle, liver, and intramuscular fat tissues represents a multi-tissue regulatory SV. NUDCD3 is involved in cytoskeletal organization and intracellular transport, and its association with loin muscle area33 suggests that SV-mediated regulation of this gene may contribute to coordinated changes in growth and body composition. However, several limitations of this study should be acknowledged. Although newly generated data were incorporated, much of the sequencing data were obtained from public resources, introducing heterogeneity in sequencing depth and platforms. In addition, the hybrid F6 family design and limited RNA-seq resource used for eQTL mapping may not fully capture the allelic diversity present in natural populations. Finally, despite the advantages of graph-based genotyping, SV discovery remains partly constrained by short-read sequencing, leading to underrepresentation of complex rearrangements and reference-absent sequences. Future studies incorporating deeper long-read sequencing, more balanced population sampling, and integrative multi-omics resources will further refine SV catalogs and strengthen functional interpretation. In summary, this study establishes structural variants as a key contributor to regulatory variation, domestication, environmental adaptation, and complex trait divergence in pigs. By integrating a population-scale, graph-based SV reference with multi-tissue transcriptomic and phenotypic data, we move beyond SV cataloging to directly link structural variation with molecular and phenotypic consequences. These findings provide a robust framework for incorporating SVs into future pig genomics research and precision breeding strategies. Methods Ethics statement All of the procedures involving animals are in compliance with the care and use guidelines of experimental animals established by the Ministry of Science and Technology of China. The study protocol was approved by the Animal Welfare and Ethics Committee of Jiangxi Agricultural University (Approval No. JXAU2011-006 for F6 population and JXAULL-2023-11-15 for indigenous breeds). Sample collection and quality control As an extension of the Thousand Pig Genomes Project, we expanded our dataset by incorporating 293 new re-sequencing samples and 1322 additional samples from public repositories2,43–63, based on the initial 1103 samples collected in the first phase19. After removing duplicate entries and low-quality sequences, we obtained a final dataset comprising 2445 whole-genome sequencing (WGS, paired-end read length ≥ 100 bp; sequencing depth ≥ 10×; genome coverage ≥ 90%) samples from 120 pig breeds/populations for subsequent analyses. To minimize biases caused by sequencing depth and improve computational efficiency, we down-sampled 361 deeply sequenced samples (> 40×) into 35× coverage using Picard v2.24.0 DownsampleSam function with parameters STRATEGY = ConstantMemory and RANDOM_SEED = 999. The average sequencing coverage for the 2,445 samples used in the subsequent analyses was 22.15×, with a range of 10–40×. This dataset includes 733 Asian domestic pigs, 111 Asian wild boars, 983 European domestic pigs, 39 European wild boars, and 579 Eurasian hybrid pigs. Of these, 293 samples were generated by our laboratory in this study. These animals were maintained under standard husbandry conditions and slaughtered at adult age (240 ± 10 days) for tissue collection. In addition, we also download 41 evolutionarily related outgroup individuals for comparative analysis. All WGS data are publicly accessible via NCBI and GNBC. Detailed summary statistics are provided in Supplementary Data 1. Among the Eurasian hybrid pigs, 292 new re-sequencing samples were derived from the F6 generation of an eight-way composite pedigree established by our laboratory11. This pedigree was created through rotational crosses among eight breeds: Bama Xiang, Laiwu, Erhualian, Tibetan, Duroc, Yorkshire, Landrace, and Hampshire pigs, with each ancestral breed contributing 12.5% of the genetic background from the F3 generation onward. The 292 F6 individuals, forming the core measured cohort, were sequenced at high coverage (depth > 30×), and were additionally profiled by RNA sequencing of liver (n = 266), skeletal muscle (n = 289), intramuscular fat (n = 99), or backfat (n = 124) tissues. Each RNA-seq sample yielded at least 10 Gb of data. Moreover, body weight at 240 days of age was recorded for all 292 pigs. SNPs and InDels calling and filtering We aligned the WGS data to the Sscrofa11.1 reference genome using BWA-MEM v0.7.17-r118864. PCR duplicates were marked and removed using Samblaster v0.1.2665, followed by sorting and indexing of the alignment files with Samtools v1.1066. Sequencing coverage and depth were assessed using Mosdepth v0.3.667. Variant calling of SNPs and InDels was performed jointly across all samples using Graphtyper v2.7.5 with default parameters68. Quality control of the resulting variants was conducted using PLINK2 v2.0.0-a.5.22LM69 with the following criteria: (1) retain only biallelic variants (--max-alleles 2); (2) remove variants with a missing rate > 5% (--geno 0.05); (3) remove variants with minor allele frequency <0.001 (--maf 0.001); (4) set genotypes to missing if read depth <5 or genotype quality (GQ) < 20. Further filtering was applied using Bcftools v1.966 to retain variants with a FILTER status of PASS and an AAScore ≥ 0.5. The resulting high-quality variant set was used for downstream analyses. Structural variants genotyping using Graphtyper2 pipeline22 We aligned WGS data to the Sscrofa11.1 reference genome using BWA-MEM v0.7.17-r1188, generating corresponding BAM files for each sample. Structural variants were initially identified separately in each sample using Manta v1.6, and only variants with a FILTER status of “PASS” and a quality score (QUAL) ≥ 30 were retained for downstream analysis. To refine inversion (INV) calls, we reannotated INV-type SVs using the convertInversion.py script. For each sample, we required that the SVs identified by Manta should be supported by at least one of the two additional SV detection pipelines, Smoove v0.2.6 (https://github.com/brentp/smoove) or INSurVeyor v1.1.270. Subsequently, we merged SV calls across all samples using Svimmer v0.1, yielding a non-redundant set of 679,801 candidate SVs. Population-scale joint genotyping of these SVs was performed using Graphtyper2 v2.7.5. To obtain a high-confidence SV set, we applied stringent quality control filters to the joint-genotyped variants, retaining only those that met the following criteria: (i) for all SV types except inversions, mode must be “AGGREGATED”; (ii) SV length between 50 bp and 1,000,000 bp; (iii) FILTER status of “PASS”; (iv) PASS_ratio greater than 0.5; (v) missing genotype rate ≤ 50%; and (vi) minor allele frequency (MAF) ≥ 0.001. For each genotype, we additionally assigned a missing value when the (vii) genotype depth (DP) < 8, or (viii) genotype quality (GQ) < 20; Following these filtering steps, a total of 205,379 high-quality SVs were identified across the 2,486 samples using Graphtyper2 pipeline, comprising 107,364 deletions (DEL), 87,021 insertions (INS), 10,063 duplications (DUP), and 931 inversions (INV). Structural variants genotyping using paragraph20 To complement SV discovery, particularly for large insertions, we constructed an SV reference panel based on the 60 long-read sequencing data. We obtained 17 ONT datasets and 34 PacBio (Pb) datasets from public repositories, supplemented by six ONT and three Pb datasets from our laboratory, representing major global pig breeds (details in Supplementary Data 2.7,8,17,18,48,58,71–84). SVs were initially called using Sniffles v2.5.385 with default parameters. To further refine this SV set, SVs were independently detected from the 60 LRS datasets using cuteSV v2.1.286 and SVIM v2.0.087 with default parameters, respectively. For each sample, SVs detected by Sniffles, cuteSV86, and SVIM were merged using JasmineSV v1.1.588 with the parameters --min_overlap 0.5 and --max_dist 500. Only Sniffles calls supported by at least one of the other two callers (cuteSV or SVIM) were retained. All samples were merged using Jasmine v1.1.5 to produce a final, high-confidence SV true set. Insertions and deletions larger than 50 bp and smaller than 1 Mb were extracted from the reference panel and used as reference SVs. Genotyping of these SVs across the 2486 samples was performed using Paragraph v2.3 with default parameters. The individual VCF files were merged using Bcftools merge. Post-genotyping filtering included: (1) removal of genotypes with DP < 5; (2) removal of sites with missing genotype rate > 50%; (3) removal of variants with MAF < 0.001; and (4) retention of genotypes with FT = “PASS”. Following these filtering steps, a total of 205,886 high-quality SVs were identified across the 2486 samples using Paragraph pipeline, comprising 86,715 deletions (DEL), 119,171 insertions (INS). Merge the paragraph and graphtyper2 callsets For deletions and insertion, we merged the SV callsets generated by Graphtyper2 and Paragraph, assigning higher priority to variants identified by Paragraph. We used Jasmine v1.1.588 with the parameters --min_overlap 0.5 and --max_dist 500 to determine whether SVs of the same type from the two callsets overlapped substantially, and manually incorporated Graphtyper2-specific SVs into the Paragraph callset to generate an updated, non-redundant SV set. Among these, 76,474 SVs (34,950 DEL and 41,524 INS) were detected by both pipelines, representing the overlapping set. Subsequently, inversions and duplications genotyped by Graphtyper2 were incorporated into the updated SV collection. Finally, SVs specific to 41 outgroup genomes were removed, resulting in the final comprehensive SV callset used for downstream analyses. All benchmark analysis was performed using truvari v5.4.0 software89. Functional annotation We obtained the Sscrofa11.1 v110 gene annotation file from Ensembl (https://asia.ensembl.org/index.html) and parsed the genomic coordinates of 5’ untranslated regions (5’ UTRs), exons, introns, 3’ untranslated regions (3’ UTRs), and intergenic regions. To characterize the regulatory landscape, we additionally downloaded chromatin state segmentation data, encompassing 15 distinct functional annotations across the pig genome (http://genome.ucsc.edu/s/zhypan/susScr11_15_state_14_tissues_new)90. Repetitive elements within the Sscrofa11.1 assembly were annotated using RepeatMasker v4.1.091, and the corresponding repeat coordinates were manually extracted. Using Bedtools v2.31.0, we intersected the structural variant dataset with the above genomic features to calculate counts and enrichment statistics. For downstream functional interpretation, we conducted Gene Ontology enrichment analyses with the R package clusterProfiler v4.6.092, following default settings. SV-eQTL analysis in liver, muscle, backfat and intramuscular fat tissues RNA sequencing reads from liver, muscle, backfat, and intramuscular fat tissues were aligned to the Sscrofa11.1 reference genome using STAR v2.7.11b93, followed by sorting and indexing with samtools v1.9. Gene-level quantification was performed with featureCounts v1.6.394 using annotations from Sscrofa11.1.v110. And we defined expressed genes as those with transcripts per million (TPM) > 0.5 in at least three samples. All SVs with MAF > 0.05 and call rate >0.8 of each tissue/population were used to perform eQTL mapping. SV-eQTL analyses were conducted using the fastGWA framework of GCTA v1.94.0beta95, with corrections for genetic kinship structure, sex, age, the top three principal components (PC) derived from genotype data, and PEER factors inferred from gene expression data. Specifically, 20 PEER factors were applied for tissues with large sample sizes (liver and muscle; n > 200), whereas 10 PEER factors were used for tissues with smaller sample sizes (backfat and intramuscular fat). Cis-eQTLs were defined as variants located within ±1 Mb of each gene, and associations reaching an empirical significance threshold of P < 1e-6 were considered statistically significant. FINEMAP v1.496 was run with the --sss and --n-causal-snps 5 options, restricting the analysis to configurations with at most five causal variants. LDAK v6.197 was used to estimate the total heritability of gene expression for 14,835 genes measured in 266 liver samples25 from the F6 population. Separate genetic relationship matrices were constructed using SNPs/InDels only, SVs only, and all variants jointly. Prior to heritability estimation, genetic variants were pruned to remove nearby SNPs in near-perfect linkage disequilibrium using the LDAK-thin procedure with parameters --window-prune 0.98 and --window-kb 100. For all analyses performed with LDAK-thin, the parameter --constrain YES was applied to prevent negative heritability estimates. SV-FST analysis between different population We evaluated the genetic differentiation of SVs between different populations using FST analysis from VCFtools v0.1.16. Three pairwise comparisons were performed: (a) Asian wild boars (n = 109) versus Asian domestic pigs (n = 535), excluding pigs that have belong to high-altitude regions; and (b) 14 miniature pig breeds (n = 305) versus commercial pig breeds (n = 913). The miniature pig group comprised breeds with adult body weights below 100 kg (Bamaxiang, Luchuan, Wuzhishan, Diannanxiaoer, Xishuangbanna, Tibetan, Yucatan, Goettingen, Ossabaw, and MiniLewe), whereas the commercial pig group included breeds with adult body weights exceeding 200 kg (Duroc, Landrace, Yorkshire, and Pietrain). SVs ranking within the top 0.1% of the genome-wide FST distribution were considered significantly differentiated. (c) To investigate climatic adaptation, two groups were defined based on both environmental adaptation and body size: cold-adapted large-bodied pigs (CL, n = 124; Bamei, Hetao, Min, Mashen, Laiwu, and Nanyang) and heat-adapted large-bodied pigs (HL, n = 262; excluding Bamaxiang, Luchuan, Wuzhishan, Diannanxiaoer, and Xishuangbanna). As both CL and HL consist of large-bodied pigs, this design effectively minimizes confounding effects of body size, thereby enabling a more robust contrast between cold and heat adaptation. Candidate genes located within 10 kb upstream or downstream of significant variants were identified. Functional enrichment analyses were performed using the R package clusterProfiler92. We also downloaded raw Hi-C data for three samples from NCBI with accession number of PRJNA7542508. The Hi-C reads were aligned to the pig reference genome to generate interaction matrices using the Juicer (v1.8.8) pipeline, which were then visualized in Juicebox (v1.11.08)98. SV-GWAS analysis in the F6 population In the F6 population, body weight at 240 days of age was recorded for a total of 886 individuals, comprising 292 high-depth whole-genome sequenced samples (>30×) used in this study and 594 previously published low-depth samples (~8×)11. SV genotyping was performed across all samples using the Graphtyper2 pipeline, with subsequent filtering based on a call rate > 0.8 and minor allele frequency (MAF) > 0.05. SV-based genome-wide association studies (SV-GWAS) were conducted using the linear mixed model implemented in GEMMA v0.9899, with sex and age at measurement included as covariates. A genome-wide significance threshold was set at 1e-06. The coloc package100 is used to perform colocalization analysis between SV-GWAS and SV-eQTL. Statistical analysis Graphs and associated statistical tests were performed using R software (version 4.2.0). Unless otherwise specified, statistical tests were two-tailed Wilcoxon tests. The exact sample sizes for each group are provided in the boxplot legends to ensure transparency and reproducibility. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary information Description of Additional Supplementary Files Supplementary Data 1 Supplementary Data 2 Supplementary Data 3 Supplementary Data 4 Supplementary Data 5 Supplementary Data 6 Supplementary Data 7 Supplementary Data 8 Supplementary Data 9 Supplementary Data 10 Supplementary Data 11 Supplementary Data 12 Reporting Summary Transparent Peer Review file Supplementary information Description of Additional Supplementary Files Supplementary Data 1 Supplementary Data 2 Supplementary Data 3 Supplementary Data 4 Supplementary Data 5 Supplementary Data 6 Supplementary Data 7 Supplementary Data 8 Supplementary Data 9 Supplementary Data 10 Supplementary Data 11 Supplementary Data 12 Reporting Summary Transparent Peer Review file Source data Source data Source data