Causal variants in animal genomes: approaches to identification, phenotypic impact, and implications for selective breeding

Abstract

This review provides a comprehensive analysis of causal variants as the most informative genetic markers in livestock breeding, highlighting their potential to enhance the precision and efficiency of animal selection strategies. It discusses the types and classifications of genetic variants arising in animal genomes, addresses complex issues related to genetic variant terminology, and evaluates their prevalence and significance in phenotypic variability and evolution. The review explores the molecular mechanisms underlying genetic causality, detailing how variants in both regulatory and coding regions can influence gene expression regulation and the structural and functional properties of transcripts and proteins. It also outlines the diversity of methodological approaches used to identify causal variants and assess their functional impact, incorporating both experimental techniques and bioinformatic tools. Prominent examples of causal variants identified in the genomes of mammals and birds are presented, demonstrating their influence on traits such as growth rate, body conformation, milk and meat production and quality, reproductive performance, coat color, and disease resistance. In addition, the review describes current strategies and future prospects for the use of causal variants as genetic markers to improve animal productivity traits within marker-assisted and genomic selection frameworks. Finally, it addresses the conceptual and practical challenges in distinguishing truly causal variants from those merely statistically associated with traits, emphasizing the need for more rigorous criteria and methodological standards in this area of research.

Introduction

The breeding process in livestock can be regarded as a specialized form of evolution, wherein targeted artificial selection serves as the primary driving force [1–4]. Evolutionary events within this context are observable across various levels of biological organization — from molecular to population level — encompassing shifts in morphological, physiological, and biochemical structures and functions. The genome, and its alterations, constitutes the fundamental object of study in this scenario. Genome variability, sustained by the continual emergence of mutations and recombination events, provides new genetic information for artificial selection aimed at enhancing productivity traits [5, 6]. As a consequence, selection induces shifts in allelic frequencies, leading to the fixation of genetic variants in populations that exhibit causal properties concerning these traits. “Causal” genetic variants are typically defined as those for which cause-and-effect relationships with specific phenotypes have been established [7–10]. The identification of causal genes and their variants is one of the important objectives in applied animal genetics. Using these variants as polymorphic genetic markers for productivity traits significantly enhances the efficacy of contemporary breeding methodologies, including marker-assisted selection and genomic selection [7, 8, 11, 12]. These polymorphic markers most accurately reflect genotype–phenotype relationships, thereby enabling a more precise assessment of animal genotypes in breeding strategies.

Most productivity traits exhibit quantitative inheritance, being influenced by numerous genes that collectively contribute to the phenotype. These genes can interact in various ways, with each gene exerting different levels of influence on the expression of specific traits [13, 14]. The identification of causal variants in the so-called major genes of quantitative trait loci (QTLs) is of particular interest, as their using as genetic markers in both marker-assisted selection and genomic selection is expected to have a significant impact [15–21]. At the same time, the search for truly causal variants can be methodologically challenging, since causal variants are often in linkage disequilibrium (LD) with nearby variants that are not themselves responsible for the trait. As a consequence, associations attributed to these linked, non-causal variants may weaken or disappear in subsequent generations or may be breed-specific [22]. Moreover, it seems appropriate to identify causal variants in other genes and loci outside well-characterized QTLs, whose influence on a specific trait may be less pronounced; in addition, the same polymorphic variants may have a minor effect on some traits but significantly affect the formation of others [9].

It should also be noted that the impact of causal variants manifests through distinct molecular pathways. Some causal variants act through nucleotide substitutions in protein-coding regions of genes, leading to alterations in amino acid sequences and, consequently, changes in protein structure and function [7, 23]. Other causal variants exert their influence by modifying gene expression regulation, transcript processing, or the dynamics of protein synthesis [7]. Together, these mechanisms illustrate the molecular heterogeneity of causal effects and underscore the need for integrated approaches to variant identification and validation.

This review examines the variability of animal genomes as a source of polymorphisms of various types, which serve as the foundation for genetic markers. It explores terminology and classification issues related to genetic variants, outlines the molecular mechanisms underlying their causal effects and functional consequences, and reviews methodological approaches for identifying causal genes and variants. In addition, the review summarizes representative examples of direct genetic markers successfully identified and applied in animal breeding and discusses existing strategies and prospects for improving the use of causal variants within marker-assisted and genomic selection frameworks.

Genetic variation in animals

In this review, genetic variation within the genome is understood as alterations in nucleotide sequence that, at the population level, manifest as genetic polymorphism among individuals of the same species. Genetic variation contributes to intraspecific diversity and enables adaptation to changing selective pressures, whether natural or artificial, particularly in the context of livestock productivity traits.

Basic terms and classification of genetic variants

Several terms describe nucleotide sequence alterations, including “polymorphism,” “mutation,” and “sequence variant,” but their usage varies across studies, complicating interpretation and comparison. “Mutation” is often defined as any rare sequence change, whereas “polymorphism” refers to sequence alteration occurring at a population frequency of at least 1% [24–26] or 5% [27]. The distinction is also sometimes based on pathogenicity, with mutations considered disease-causing and polymorphisms non-pathogenic [24]. These classifications are limited, as allele frequencies may change with expanding datasets, and the same variant can have beneficial, neutral, or deleterious effects depending on environmental context [28]. In addition, the terms may differ in contextual usage: “mutation” often refers to the process or event of nucleotide sequence alteration (e.g. germline or somatic mutations), whereas “polymorphism” refers to a stable condition or quality resulting from such changes (e.g. multi-allelic state at the population level) [26, 28]. To reduce ambiguity, the neutral term “sequence variant” (or “variant”) is preferred [24], as it is consistent with current annotation practices, such as those employed by the Ensembl genome browser [29] or Sequence Ontology (SO) [30], and avoids implicit assumptions regarding allele frequency or phenotypic effect.

Genetic variants are commonly classified according to the nature of DNA sequence changes that generate them. A widely used scheme, adopted in the scientific literature and databases such as Ensembl Variation, distinguishes between short variants (including substitutions and small insertion-deletion variants) and structural variants, which typically span from more than 50 base pairs to over a megabase [31–33]. Earlier classification further subdivided structural variants into submicroscopic (1 kb—3 Mb) and microscopic (> 3 Mb) categories [34].

Figure 1 illustrates the number of variants in these two categories across domestic animal species represented in the Ensembl Variation 115 database [29, 35–37]. Notably, datasets containing structural variants are available for fewer species than datasets containing short variants. Additionally, the number of known short variants significantly exceeds that of structural variants. At present, it remains difficult to precisely determine the ratio between short and structural variants, as the effective identification of structural variants has only become possible with advances in sequencing technologies that enable long-read genome sequencing [31].

Fig. 1.

Comparison of the number of known short (sequence) and structural variants across domestic animal species and the human genome. Variant counts were obtained using Ensembl BioMart [36] from Ensembl Variation release 115. Short variant counts were derived from the complete Ensembl Variation datasets [29], whereas structural variant counts were limited to variants archived in DGVa [37]. The following reference genome assemblies were used: human (GRCh38.p14), chicken (bGalGal1.mat.broiler.GRCg7b), horse (EquCab3.0), cattle (ARS-UCD2.0), pig (Sscrofa11.1), sheep (ARS-UI_Ramb_v3.0), goat (ARS1), dog (ROS_Cfam_1.0), and domestic cat (F.catus_Fca126_mat1.0)

Structural variants are also less well studied in terms of their functional impact, particularly concerning gene expression, genome structure and organization, and epigenetic modifications [31]. Consequently, their potential causal roles in phenotypic variation remain insufficiently understood. Moreover, a general disparity exists in the study of both short and structural variants across different species. Species that have been more extensively studied using modern sequencing approaches exhibit a greater number of identified variants than those that have received less attention. For example, the number of known genetic variants in humans is more than an order of magnitude greater than in cattle or pigs, two common agricultural species (Fig. 1), largely due to differences in sequencing effort, quality of genome assemblies, and database completeness rather than inherent biological differences.

A more detailed and arguably the most comprehensive classification of genetic variants to date is that provided by Sequence Ontology [30]. The term “sequence alteration” (SO:0001059) encompasses all other terms used to describe various types of genetic variants. Sequence Ontology definitions are integrated into Ensembl, and are also employed in this review to categorize different genetic variants. Below, the major variant types are described to outline their “mechanistic” classification based on the nature of their occurrence in the genome. This overview also addresses ambiguities in definitions and distinctions between such variant types that often complicate their interpretation and understanding of their functional consequences, including whether the variants are neutral or causal.

Substitutions

In the Sequence Ontology framework, substitutions (SO:1000002) are defined as sequence alterations in which the length of the variant is identical to that of the reference. This class comprises single nucleotide variants (SNVs; SO:0001483) and multiple nucleotide variants (MNVs; SO:0002007), also known as single- and multi-nucleotide polymorphisms (SNPs and MNPs), respectively [38, 39].

SNVs represent one of the most prevalent variant types, occurring approximately every 200—300 base pairs in mammalian genomes [40, 41], and hundreds of millions of SNVs have been cataloged across animal species in public repositories [42–44]. SNVs arise primarily from point mutations caused by replication errors, mutagen exposure, or impaired DNA repair, and a subset of these variants directly alters protein sequence, gene regulation, or splicing. Although many SNVs are neutral, their high genomic frequency and their role as primary targets for genome-wide association studies (GWAS) and genomic selection, due to their density and ease of genotyping [9, 14], mean that a substantial proportion of well-characterized causal variants belong to this variant type, and that SNVs are frequently associated with economically important traits in breeding populations [7, 45, 46].

Although MNVs in humans and animals have attracted increasing attention [47], their interpretation remains challenging [47, 48] because automated variant-calling pipelines frequently misannotate them as adjacent independent SNVs, leading to inaccurate functional predictions by tools such as VEP (Variant Effect Predictor) [35], ANNOVAR [49], and SnpEff [50]. A well-documented example illustrating the impact of such misannotation on the analysis of MNV causality is the AA to GC substitution in the bovine DGAT1 gene, which leads to a causal amino acid change (K232A) affecting phenotype, but is annotated in Ensembl as two independent SNVs (rs109234250 and rs109326954) [51]. Consequently, misannotation of MNVs may obscure true causal variants and reduce the accuracy of downstream analyses and functional variant prioritization.

Insertion-deletion variants

Insertions (SO:0000667) and deletions (SO:0000159) are variants that add or remove consecutive nucleotides, respectively. Indels (SO:1000032) in formal Sequence Ontology and Ensembl usage refer to sequence alterations that include both an insertion and a deletion at the same site, although in broader evolutionary and phylogenetic context [52–54] the term indel is often used to refer to either an insertion or a deletion.

Insertion-deletion variants are classified as either short or structural variants based on the length of the altered sequence. In animals, insertion-deletion variants are the second most common type of variation, occurring about tenfold less frequently than SNVs [55, 56], although they affect a comparable total number of nucleotides [57, 58].

A specific type of insertion is duplication (SO:1000035), in which a segment of the nucleotide sequence is repeated. Duplications serve as important driving force of evolutionary processes by providing genetic material for genetic novelty [59]. Gene duplications, in particular, lead to paralogy [60] and contribute to the emergence of new genes and gene functions [61, 62]. Consequently, duplications contribute not only to phenotypic variability but also to long-term evolutionary processes, including lineage diversification and speciation [61, 62], sometimes placing their impact beyond individual causal variants and at the level of genome evolution.

Although mobile genetic elements such as transposable elements (TEs) are formally annotated as insertion-deletion variants (SO:0001837, SO:0002066), their autonomous mobility and evolutionary dynamics set them apart from conventional insertion-deletion variants [63–65]. In animal genomes, particularly in mammals, TEs constitute at least half of the genomic DNA and are a major source of genetic diversity [31, 66–69]. Their impact reflects both the scale and diversity of mutations they generate, ranging from local sequence changes to large genomic rearrangements. While insertions in heterochromatic regions are often neutral, integration into euchromatic regions, especially within or near genes, can alter gene expression and protein structure [70, 71]. Through these processes, TEs act not only as generators of genetic variability but also as potent causal variants, capable of inducing phenotypic changes and driving long-term genome evolution [67, 72].

Sequence length alterations

Sequence length alterations comprise structural variants associated with changes in the copy number of certain nucleotide sequences [73]. In the Sequence Ontology, these include copy number variation (SO:0001019), short tandem repeat variation (SO:0002096), and simple sequence length variation (SO:0000207), with the first two used in Ensembl annotations.

Copy number variations (CNVs) involve deletions or duplications of genomic segments of approximately 1 kb or larger [74, 75], and are important contributors to genotypic and phenotypic variability in animals, covering up to 13% of the genome [76–78]. CNVs play a role in generating genetic diversity within populations, while individual CNVs serve as the basis for genetic markers, which may be associated with economically important traits in livestock animals [79–81].

Tandem repeats represent a significant source of polymorphism in both human and animal genomes [82, 83]. Repeats with short units (typically 2—4 or 2—6 bp) are commonly termed microsatellites or short tandem repeats (STRs), whereas longer units (10—100 bp) are referred to as minisatellites [82, 84, 85]; the term VNTR is used inconsistently, either for minisatellites alone [86–88] or as a collective label for both classes [89]. Tandem repeats distributed across intergenic regions, introns, and exons, with a predominant presence in non-coding regions [90, 91], with minisatellites tending to accumulate in (sub)telomeric regions [92]. Persistent inconsistencies in terminology, size thresholds, and localization criteria regarding tandem repeats complicate cross-study comparisons and interpretation [92].

The formation of tandem repeats (both micro- and minisatellites) as structural variants [93, 94] is driven by recombination errors, unequal crossing-over, polymerase slippage during DNA replication and repair [85, 95, 96], and activity of mobile genetic elements [92], while minisatellites may also arise through the progressive expansion of microsatellites [85]. From a causal perspective, microsatellite instability serves as an indicator of DNA mismatch repair deficiency and is associated with oncological diseases in humans and animals [97], as well as with specific inherited disorders [85]. At the same time, genome stability relies on the preservation of tandem repeats in centromeres and telomeres [98].

Both micro- and minisatellites are used in DNA fingerprinting [85, 99]. Owing to their high polymorphism, microsatellites form the basis of STR profiling, which is extensively applied in evolutionary studies, pedigree verification, breed differentiation, and forensic identification in animals and humans [100–102]. Species-specific STR panels have been developed for numerous taxa [103–107], alongside ongoing efforts to standardize STR systems across species [83, 108] and the establishment of STR locus databases [109–112]. Beyond their use in STR profiling, microsatellites are increasingly recognized for their roles in gene regulation and phenotypic variation [93, 113, 114], supporting their relevance as potentially causal variants in animal genomes.

Inversions

Inversions (SO:1000036) are structural variants in which a continuous nucleotide sequence is inverted in the same position. In animal genomes, they range from 130 kb to 100 Mb, with an average length of 8.4 Mb [115], and they occur as paracentric inversions (excluding the centromere) or pericentric inversions (including the centromeric region).

Phenotypically, inversions have been directly linked to complex trait variation across diverse animal taxa. In fish, inversions are associated with changes in migratory behavior and physiological adaptations, while in birds they influence mating behavior, body size, plumage, aggression, testis size, and steroid metabolism [115, 116]. Furthermore, inversions can modulate gene expression, as exemplified by their influence on the expression of the KIT gene in horses, which determines coat coloration [117]. Additionally, an inversion has been identified within a QTL in rainbow trout that is associated with disease resistance [118].

Inversions restrict the occurrence of crossing-over during meiosis in heterozygous individuals, which contributes to reproductive isolation. This suppression of recombination, combined with the accumulation of additional mutations within the inverted region, facilitates genetic divergence and ultimately promotes speciation [115, 119]. Accordingly, inversions are recognized as drivers of evolutionary change, with documented roles in radiation of birds [120] and in genomic restructuring events preceding the emergence of the Homo genus [115, 119]. In this context, inversions may exert an evolutionary impact that exceeds that of individual causal variants, shaping not only traits but also species boundaries.

Translocations

Translocations (SO:0000199) are chromosomal aberrations in which a segment of the genome is relocated to a new position due to chromosome breakage, most commonly between non-homologous chromosomes [121], although some definitions also include intra-chromosomal translocations [122]. They are typically classified as reciprocal translocations, involving exchange of segments between two non-homologous chromosomes, or Robertsonian translocations, which result from centric fusion of (usually acrocentric) chromosomes into a single chromosome composed of their long arms [123].

From a causal perspective, the biological relevance of translocations lies in the possibility of genetic material loss (in unbalanced translocations) and meiotic behavior. While carriers of balanced translocations may not exhibit phenotypic abnormalities, during meiosis, the formation of unbalanced gametes can lead to reduced fertility and an increased risk of producing offspring with partial (in reciprocal translocations) or complete (in Robertsonian translocations) trisomy or monosomy [124]. In animals, translocations are among important structural variants affecting breeding efficiency as numerous studies have demonstrated strong associations between chromosomal translocations and decreased fertility, increased rates of early embryonic mortality, and, consequently, hypoprolificacy [125–132].

Molecular mechanisms of causality of genomic variants

The causal influence of a genetic variant on phenotype arises from molecular mechanisms that affect gene expression regulation or alter the structural and functional properties of gene products, including proteins and RNAs. This section describes the principal molecular pathways through which genomic variants can exert causal effects, with particular emphasis on how variant type is linked to functional impact. Within this framework, functional annotation serves as a systematic means of relating sequence variation to genomic features and molecular entities. A central concept in variant annotation is the “consequence,” defined as the predicted effect of a variant on a specific gene, transcript, or protein. The Ensembl Variation 115 database [29] defines 41 distinct consequence categories, each corresponding to standardized terms in the Sequence Ontology [30]. Figure 2 summarizes the relationships among selected consequences associated with variants located in gene, regulatory, and intergenic regions.

Fig. 2.

Relationships among different variant consequence terms based on their genomic localization relative to transcribed genes. Variants are categorized into gene, regulatory region, and intergenic categories, which are further subdivided into specific consequence types, shown from left to right in the figure. High-impact variant consequences are marked in red, moderate-impact consequences in yellow, low-impact consequences in green, and modifier or undefined-impact consequences in grey, according to impact categories defined by the SnpEff tool [50]. The figure primarily includes consequence terms used in Ensembl Variation, together with their corresponding Sequence Ontology (SO) accessions, while the complete hierarchy of consequence terms is defined in the Sequence Ontology [30]

Variant consequence annotation offers an initial, descriptive assessment of the potential impact of a variant in relation to the protein-coding or regulatory regions and therefore serves as a preliminary indicator of functional relevance and potential involvement in phenotypic variation. According to the SnpEff annotation tool [50], frameshift, stop-gained, start-lost, stop-lost, splice-acceptor, and splice-donor variants are classified as high-impact (disruptive) changes, likely causing protein truncation, loss of function, or triggering nonsense-mediated decay. Inframe insertions and deletions, as well as missense variants, are considered moderate-impact (non-disruptive) changes that may alter protein function. Synonymous variants, including start- and stop-retained variants, along with some splice variants, are generally classified as low-impact, as they are unlikely to affect protein behavior. Other variant consequences, particularly those affecting regulatory regions, are categorized by SnpEff as “modifiers,” reflecting the limited predictive power of sequence-based annotation for their functional effects [50] (Fig. 2). Consequently, consequence-based predictions are most informative for variants within transcripts, whereas inferring functional and causal effects for regulatory or intergenic variants remains substantially more challenging.

Another important consideration is the relationship between variant type and variant consequence. Variants of the same type can have different consequences; for example, SNVs may have missense, synonymous, initiator codon, terminator codon, untranslated region, intronic, splicing, regulatory, or intergenic consequences. Conversely, identical consequence can arise from different variant types. For instance, missense consequences can result from both SNVs and MNVs, while splice-site and initiator (or terminator) codon variants may be caused by either SNVs or insertion-deletion variants. Additionally, intergenic and regulatory consequences can arise from a range of variant types, including structural variants.

In the following subsections, the main types of genomic variants are considered in terms of the molecular mechanisms through which they can influence gene products or gene regulation, thereby providing a mechanistic bridge between functional annotation, empirical identification strategies, and evidence for phenotypic causality.

Substitutions, their effects and consequences

Substitutions, including SNVs and MNVs, are widely distributed across the genome and can have nearly all the consequences outlined in Fig. 2, except those directly related to sequence length changes. They may occur in gene, regulatory or intergenic regions, and their phenotypic effects arise through changes in protein structure and stability, transcript processing, or gene regulatory activity. Particularly significant are substitutions that affect gene products — nucleic acids and proteins. Of special interest are substitutions in exonic regions that alter protein function.

Missense variants, which change a codon to encode a different amino acid, modify the protein’s amino acid sequence and can affect its structural and functional properties. In particular, missense variants can influence protein folding and binding free energies (ΔG) as well as reorganize networks of intramolecular chemical interactions. These changes may lead to altered protein stability at higher levels of structural organization, including effects on folding, oligomerization of protein monomers, and interactions with ligands [23]. While these variants do not typically cause complete loss of function, they can have significant effects, especially when they occur in protein domains responsible for receptor function, interactions with bioregulatory molecules, or enzymatic activity [133]. Many well-characterized variants with strong evidence for causality in cattle and pigs are missense variants [7].

Other substitutions affecting the initiator and terminator codons have more serious effects. In the case of substitutions that destroy the initiator codon (start-lost variant), protein synthesis does not occur. In the case where the variant leads to the destruction of the terminator codon (stop-lost variant), or the appearance of a premature terminator codon (stop-gained variant), this leads to the synthesis of an elongated or short product. As a consequence, the mRNA may be degraded by nonsense-mediated decay or non-stop decay, or a protein product may be synthesized that is either non-functional or has significantly altered functions [134, 135, 136]. SNVs and MNVs resulting in start-lost, stop-lost and stop-gained consequences are frequently found in mammalian genomes and may be strong candidate or putative causal variants affecting phenotype [46, 137–141].

In contrast to protein-altering variants, synonymous substitutions change the transcript sequence without affecting the amino acid sequence of the encoded protein. However, their potential role in phenotypic variability, including productivity traits in farm animals, remains an interesting topic for discussion and experimental study. While few synonymous variants have been experimentally validated as causal, nucleotide substitutions that do not alter codon can still modulate protein synthesis rates, co-mRNA folding and stability, as well as other regulatory processes [142]. In a broader sense, they may affect the molecular mechanisms of gene expression. A detailed analysis of the impact of synonymous mutations at various cellular levels is presented in the review [143], which also provides evidence of their involvement in certain diseases.

Equally important are substitutions located in non-coding regions, including intronic and UTR regions, splice sites, and other regulatory elements. Their functional significance lies in their potential to influence splicing and isoform balance, as well as to affect transcription factor binding sites, enhancers, DNA methylation regions, and non-coding RNA motifs [144]. Consequently, substitutions in non-coding regions can modulate gene expression. Since regulatory effects on gene expression are not exclusive to substitutions (SNVs and MNVs) but also occur with other variant types, such as insertion-deletion variants, the potential causal roles of non-coding variants with regulatory functions is addressed in the separate following subsections.

Effects and consequences of insertion-deletion variants

Insertion-deletion variants represent a functionally relevant type of genomic variation that can contribute to phenotypic variation and may represent candidate or putative causal variants in animals. Like substitutions, insertion-deletion variants can occur in gene, intergenic, and regulatory regions. When located in exonic regions, their impact is often most pronounced. Insertion-deletion variants may either preserve the reading frame or disrupt it, leading to different functional consequences. Inframe variants alter the amino acid sequence by adding or removing residues, with effects that depend on the position and structural context of the change within a protein domain [145]. In contrast, frameshift-inducing variants disrupt the triplet reading frame and lead to proteins with aberrant C-terminal sequences, stop-gained or stop-lost consequences, resulting in non-functional proteins or mRNA decay [134]. Insertion-deletion variants in regulatory regions may impact the regulation of gene expression or splicing processes, which are discussed in following subsection.

Sequencing technologies enable precise localization of insertion-deletion variants and their assessment in association studies and prioritization as candidate variants. Given their nature, insertions and deletions tend to be a more radical event for a DNA region compared to substitutions, making them more likely to disrupt regulatory or coding sequences. Animal genomic studies increasingly document how insertion-deletion variants intersect with traits of economic importance [7, 146–148].

Mobile genetic elements, inherently classified as insertion-deletion variants, can play a significant role in influencing animal traits and may underlie putative causal variants. The insertion of DNA fragments associated with mobile element activity can alter the expression of TE-translocated genes. These genes are influenced by chromatin conformation at the integration site (position effect), such as its open or closed state, changes in their relative position to enhancers or silencers, or interactions with newly introduced transcriptional regulatory elements. Conversely, integration events can also affect the activity of neighboring genes, leading to regulatory changes of the other nature [70].

When mobile elements integrate into coding regions, they can disrupt gene expression, introduce frameshifts, or completely block transcription. Insertions into regulatory regions, whether within or outside genes, may alter gene expression in specific tissues due to interactions with tissue-specific enhancers. If mobile elements insert into introns without affecting expression, their impact is generally limited, as introns are removed during splicing. However, this does not apply when such insertions interfere with splicing signals, potentially altering transcript processing and gene function [70]. For example, the insertion of short interspersed nuclear elements (SINEs) into an intron of the porcine growth hormone receptor (GHR) gene has been shown to reduce its expression by acting as a repressor [66, 149]. These structural rearrangements, as a source of polymorphism, are considered valuable genetic markers for animal breeding and selection [67, 149]. By mediating the expression of genes harboring candidate or putative causal variants linked to productive phenotypes [150], they may contribute directly to trait variation. In the study [67] the creation of comprehensive TE maps for domesticated species was proposed to support genetic breeding studies, particularly in assessing the potential influence of TE dose effects on specific traits.

Effects of copy number variations

The productivity phenotype is influenced not only by the expression of specific allelic variants and the activity of regulatory DNA regions but also by the number of copies of these genes and regions. CNVs may involve coding sequences, regulatory regions, or both, and can therefore influence phenotypes through multiple, partially independent molecular mechanisms [151]. One of the principal molecular mechanisms underlying CNV effects is the alteration of gene dosage resulting from changes in copy number, which leads to corresponding changes in transcript and protein abundance. Such dosage effects can directly affect quantitative traits that are sensitive to the concentration of gene products or to the stoichiometric balance of proteins within molecular complexes [152]. In addition, CNVs can expose recessive alleles through hemizygous deletions that would otherwise remain phenotypically silent [151].

CNVs can also disrupt gene architecture, as breakpoint junctions may interrupt coding sequences, resulting in truncated or nonfunctional gene products, or generate gene fusion events that produce novel transcripts with altered or aberrant functions [79, 151, 153]. When CNVs involve regulatory elements such as promoters, enhancers, or silencers, changes in copy number may alter the accessibility, dosage, or spatial organization of these elements [151]. Furthermore, duplicated gene copies may lack essential regulatory sequences or become embedded in a different chromatin environment, leading to altered expression levels compared with the original locus [80, 151].

Thus, CNV as structural variations represent an independent factor of phenotypic variability and may represent putative causal variants influencing specific phenotypic traits [154, 155]. For example, CNVs can affect coat color, as demonstrated by duplication of the KIT gene in pigs [156] and duplication of the ASIP gene in sheep [157] and goats [158]. CNVs have also been implicated as causal factors in developmental and inherited disorders, including osteopetrosis, anhidrotic ectodermal dysplasia, copper toxicosis, intersexuality, cone degeneration, periodic fever, and dermoid sinus across various animal species [80]. Additionally, numerous CNVs have been found to be strongly associated with productivity traits in various cattle breeds [79, 81, 153, 155, 159–162] and other livestock species [76, 163–165].

Short tandem repeats’ effects

STRs constitute a highly polymorphic component of animal genomes and can influence phenotypic variation through their effects on gene regulation. When STRs are located within introns and 3'- or 5'-UTR regions, variation in repeat length does not modify the amino acid sequence of the encoded protein but may affect molecular processes underlying differential expression and splicing of multiple genes [93].

STRs associated with gene expression variation (eSTRs) are considered functionally relevant because changes in repeat length can influence nucleosome positioning, the formation of non-canonical DNA or RNA secondary structures [166], and transcription factor binding [167], thereby potentially affecting transcriptional output. In this context, STRs represent plausible regulatory variants that may contribute to complex productivity traits in animals by quantitatively modifying gene expression [91, 114].

Establishing a direct causal role for individual STRs remains challenging, while transcriptomic analyses combined with association studies provide an important framework for identifying putative target genes of eSTRs and for assessing their regulatory impact. Several well-characterized eSTRs have been identified near known signals from GWAS [91, 114], some of which overlap with epigenomic markers of promoters, enhancers, or other open chromatin regions. Such colocalization supports a regulatory mechanism but does not, by itself, constitute definitive proof of causality.

Nevertheless, multiple well-documented examples illustrate how STR variation can plausibly contribute to phenotypic differences and disease susceptibility in animals. For instance, the study [168] demonstrated an association between STR polymorphism in the 3'-UTR of the SLC11A1 gene and resistance to brucellosis in cattle. Similarly, it was identified [169] an association between the microsatellite locus (GAA)_n in the intron of the ITPR1 gene and progressive gait abnormality in dogs. Moreover, (AC)_n repeats in the promoter of the porcine IGF1 gene are associated with variation in its expression [170], while similar repeats in the promoter of the SIX1 gene are significantly associated with carcass weight and backfat thickness in pigs [171].

Effects of regulatory variants on transcription and splicing

The influence of a significant proportion of experimentally validated and putative causal variants is associated with alterations in gene expression regulation, highlighting the relevance of separately considering variants located in non-coding regulatory regions. SNVs and insertion-deletion variants represent the primary variant types that can have regulatory consequence. Their relevance is supported by the observation that many QTLs associated with productivity traits in livestock are located in non-coding regions, and that GWAS signals are enriched in regulatory regions active in trait-relevant tissues, as shown, for example, in cattle [172].

Regulatory variants can influence phenotype through at least two closely related molecular mechanisms: changes in expression levels and alterations in RNA splicing. Variants affecting expression (eVariants) may modify transcription factor binding sites, chromatin accessibility, or enhancer-promoter interactions, leading to quantitative or context-dependent changes in gene expression [173–175]. Variants affecting splicing (sVariants) may disrupt canonical splice sites, create cryptic splice sites, or alter splicing regulatory motifs, thereby changing exon inclusion, transcript structure, or isoform balance [176, 177]. Because gene expression level and isoform composition directly shape cellular pathways and physiological processes, variants associated with expression QTLs (eQTLs) or splicing QTLs (sQTLs) represent biologically plausible candidates for causal variation underlying complex productivity traits [172, 178].

While the functional consequences of variants in coding sequences can often be predicted with a reasonable degree of confidence, the effects of variants in regulatory loci remain challenging to determine. Consequently, identifying the regulatory elements and specific variants responsible for altered gene expression remains challenging and may require integration of genomic, transcriptomic, and epigenomic data [172, 179, 180].

Numerous examples exist of putative causal variants associated with animal productivity traits and diseases that are localized in non-coding regions of the genome involved in gene expression regulation. The identification and validation of such variants remain key objectives in elucidating causal genetic mechanisms [181, 182]. In one study [178], SNV data and RNA sequencing analysis of gene expression were used to identify regulatory regions associated with beef quality in bulls. Major eQTLs and sQTLs were identified and classified as membrane-associated or cytoskeletal proteins, transcription factors, or DNA methylases, with polymorphisms in these genes being candidates for causality. A study partially similar to the previous one was conducted on sheep [183]. In the study [184], numerous bovine sQTLs were identified and found to be widely distributed across different tissue types. The significant overlap between bovine sQTLs and complex trait QTLs underscores the contribution of regulatory mutations to phenotypic variation. A comparison of eQTL and sQTL identification results with GWAS findings for four traits in dairy cattle revealed the regulatory polymorphism rs109421300 in the DGAT1 gene, which may serve as a marker for milk yield, fat yield, protein yield, and somatic cell count in milk [185]. An example of a study investigating the effect of a splice variant on important disease-related traits in cattle is the study [186], in which authors identified a c.124-2A > G splice variant in the RNF11 gene of Belgian Blue cattle, disrupting inflammatory response regulation. Homozygosity for this mutation leads to severe inflammatory lesions and premature death in one-third of affected individuals.

Regulatory impact of variants in non-coding RNA genes

Certain genes are actively transcribed in the cell, yet their transcripts are not translated into proteins. It has been established that such genes play a specific role in the functioning of the genetic apparatus of cells and contribute to the formation of phenotypic variability in the organism [187]. Structurally and functionally heterogeneous RNAs transcribed from such genes are united under the general name of non-coding RNAs (ncRNAs). NcRNAs contribute to the regulation of productivity traits and may harbor candidate or putative causal variants. Moreover, one of their most important functions is regulatory, which is realized at the transcriptional and post-transcriptional levels. Among ncRNAs, the most intensively studied classes include short non-coding RNAs and long non-coding RNAs (lncRNAs).

MiRNAs represent a broad group of short non-coding RNAs. MiRNAs are single-stranded short non-coding RNA molecules of 18—25 nucleotides in length, originating from intragenic regions of the genome (introns and partly exons), as well as intergenic regions, sometimes organized in clusters, in various genomic regions [188, 189]. MiRNAs can promote the degradation or inhibit the translation of target mRNAs by binding directly (typically with imperfect complementarity) to target sites, most commonly within the 3'-UTR of the mRNA [190–192]. One miRNA can target multiple mRNAs, while a single mRNA can be regulated by several different miRNAs [193, 194]. MiRNAs serve as key modulators of gene expression at the post-transcriptional level, play an important role in immune responses in farm animals, and are associated with variation in productive phenotypes [189]. MiRNA expression profiles in livestock are primarily associated with traits such as productivity, fertility, embryonic development, and disease resistance. MiRNAs are also known to play a role as biomarkers in the diagnosis of certain animal diseases [195].

Mutations occurring in miRNA genes can change the structure of the regulatory molecule and, consequently, affect its binding to the target mRNA. The result of this event may be an alteration in the expression of the target gene, leading to an induced phenotypic response. However, polymorphisms in target gene regions, identified experimentally or through bioinformatic analysis, may also affect their affinity for the corresponding miRNAs and can be considered as potential markers for selection in livestock species. Identified variants in miRNA and in the interacting regions of target genes, when associated with livestock productivity traits, may thus be used as genetic markers [196].

Some variants in miRNAs represent candidate or putative causal variants in relation to productivity traits. There are a number of examples indicating a candidate role for such variations in cattle, which are given in the study [197]. Sequence variants of miRNAs ssc-miR-206, ssc-miR-133b, ssc-miR-1 in pig and of gga-miR-1666 in chicken affect the processing and levels of mature miRNAs and are associated with variation in muscle fiber composition, meat quality and lean meat production in pigs [198, 199], as well as carcass traits and growth in chicken [200], respectively. Similar results for other miRNAs in chickens are presented in the study [201].

LncRNAs are RNA molecules longer than 200 nucleotides. They perform various functions at both the transcriptional and post-transcriptional levels, acting in both cis- and trans-regulatory positions [202, 203]. The involvement and regulatory role of such RNA molecules have been demonstrated primarily through lncRNA profiling of human diseases, including tumorigenesis, cardiac function, aging, and immune system development. For livestock species, however, relevant information remained limited for a long time [204]. It is currently known that lncRNAs are able to regulate gene expression through participation in epigenetic mechanisms, structural changes in chromatin, changes in promoter activity, thereby potentially influencing the phenotypic variability of complex traits in domestic animals [205, 206]. Transcriptomic studies in different animal species have shown that transcription of lncRNAs spans approximately 75% of the genome [207], and that some lncRNAs, despite containing only short open reading frames, can nevertheless encode multiple peptides [207, 208]. LcRNAs and their characterization in the genomes of livestock and other animal species represent an important component in bridging the genotype–phenotype gap and may harbor candidate or putative causal variants for productivity traits. For example, in the skeletal muscle of mice and sheep, long non-coding RNAs such as Linc-RAM, Lnc-mg, Linc-MD1, and others have been shown to participate in myogenesis [207, 209]. Additionally, LncRNA-Six1 has been associated with increased cell proliferation and muscle growth in chicken [210].

NcRNAs also include small interfering RNAs (siRNAs), which are involved in RNA interference — a process that modulates gene expression by inhibiting translation or preventing the transcription of target genes. RNA interference technology has been applied in genetic engineering of animals to achieve targeted suppression of gene expression without deleting the gene itself [211]. Possible regulation of retrotransposon transcripts and pseudogene-complementary transcripts by endo-siRNAs is suggested [212].

PIWI RNAs (piRNAs) are also non-coding regulatory RNAs. As epigenetic factors, they aggregate PIWI proteins to form a repression complex. piRNA can recognize target nascent transposable elements, initiate regulatory transcriptional gene silencing, and ensure genome integrity [213, 214]. In mammals, piRNAs are predominantly found during testicular development and spermatogenesis. PiRNAs are also able to inhibit abnormal expression of transposons and regulate gene expression to ensure oocyte genomic integrity and normal physiological activity for oogenesis. PiRNAs mediate the degradation of a wide variety of mRNAs and lncRNAs in spermatogenic cells [214].

Thus, progress in the study of regulatory non-coding RNAs provides an additional source for searching for genetic variants. The action of such variants is realized through different molecular mechanisms that ensure their regulatory effect on target mRNAs. Variants arising in regulatory non-coding RNAs may hold potential causality with respect to productivity traits. Overall, ncRNAs expand our understanding of the genetic diversity that underlies broad phenotypic variability.

Methodological approaches to identifying causal genes and variants

The search for causal variants generally follows a trajectory from mapping QTLs to identifying individual candidate genes, known as quantitative trait genes (QTGs), within these loci. This process is then refined further to pinpoint specific causal or “functional” nucleotides, referred to as quantitative trait nucleotides (QTNs) [17, 215]. A wide range of methodological approaches has been developed, each appropriate to different tasks, including QTL mapping, causal gene identification, and causal variant characterization. The hierarchical organization of these approaches is illustrated in Fig. 3. In identifying causal variants, it is essential to consider genetic variants with different effects. These include not only mutations in protein-coding regions, which may alter protein structure and function, but also regulatory variants that affect gene expression, splicing, and other molecular processes. Recognizing the diverse nature and mechanisms of genetic variants is important for understanding their roles in phenotypic diversity and disease susceptibility.

Fig. 3.

Task-oriented hierarchy of methodological approaches to QTL mapping and identification of causal genes and variants

In some cases, a gene can be classified as a candidate not solely based on its localization within a specific QTL, but also based on the known biological function of the protein it encodes, on the “literature annotation of the gene function” [216]. Such an annotation is a justification for its possible causality and a basis for conducting association studies with the aim of establishing causal relationships between the variants found in it and the corresponding productivity traits [217]. However, this approach may exclude from analysis a whole range of genes and genomic loci that contribute to the manifestation of a particular trait but whose involvement is not obvious. Some genes, even if they are not part of known physiological pathways, may still influence the development of the trait under study [13].

Given the polygenic nature of quantitative traits, a methodical approach to identifying causal variants — starting with the detection of QTLs — seems more justified and comprehensive. Two complementary approaches are used for QTL mapping: linkage mapping and association mapping. Linkage mapping is based on tracking the co-segregation of a trait and genetic markers in breeding or experimental crosses, in which animals that differ in the trait of interest are crossed and followed across several generations. In contrast, association mapping searches for statistical associations between genetic markers and phenotypic variation in largely unrelated individuals sampled directly from a population [218–220]. When association mapping is performed on a genome-wide scale to identify markers associated with productivity traits, this approach is commonly referred to as a genome-wide association study (GWAS) [221].

Numerous studies have been conducted on mapping QTLs that affect various parameters of animal productivity, and their results have been compiled into QTL databases for different biological species [222, 223]. Many QTLs in these databases, identified through linkage analysis, span dozens of map units and contain hundreds of genes, forming broad genomic regions [215, 217]. This complicates the further identification of QTGs and QTNs. The application of GWAS has improved the resolution of QTL detection, allowing for more precise delineation of QTL boundaries and facilitating the refinement of candidate gene lists within the identified loci [224–227].

In the early stages of QTL mapping method development, amplified fragment length polymorphism (AFLP) markers were used [228]. Later, microsatellite loci come into use [229]. Currently, SNV markers [230] and an expanding array of DNA sequencing data play a key role in GWAS [231, 232]. GWAS enables the identification of genomic loci involved in the regulation of a productivity trait, within which causal genes and variants can subsequently be sought [233, 234]. GWAS technology, especially with the advent of high-throughput next-generation sequencing (NGS) techniques [235], has accelerated the search for causal variants and enabled the inclusion of a greater number of relevant genetic markers for productivity traits in DNA typing panels [216, 236].

However, GWAS faces certain methodological challenges. Thus, given the involvement of a large number of genes in the formation of a quantitative productivity trait, it is not always possible to detect a statistically significant impact of individual genes with a small contribution [237]. This limitation can be overcome by using large populations in association analysis [232, 238]. However, when working with farm animals, this approach may entail significant labor and time costs. Similarly, GWAS requires large groups of evaluated animals to assess the influence of rare gene alleles on the formation of quantitative productivity traits. As a possible option, however, in the latter case, associations of rare variants with traits can be identified if the impact of this variant is highly significant [231]. Notably, increasing the sample size of animals can be facilitated by utilizing data from commercial genomic selection and accelerated through international collaboration to create larger and more diverse datasets. A meta-GWAS approach is being developed to combine summary statistics from multiple individual GWAS studies [8].

Another solution to the problem of analyzing rare alleles may be conducting GWAS in specially prepared inbred lines, where the frequency of some minor alleles is higher than in the original populations [233, 239, 240]. However, in this case, since inbreeding leads to a loss of genetic diversity, information on the role of other gene variants will not be obtained. Moreover, the implementation of experimental inbreeding in farm animal populations with relatively long generational intervals and potential misalignment with commercial breeding goals appears challenging.

An alternative and conceptually complementary approach is multi-population (or multi-breed) GWAS, which can, in contrast, provide increased genetic diversity and distinct LD patterns across animal populations. By jointly analyzing data from multiple populations, this approach exploits differences in LD structure to improve mapping resolution, facilitates the detection of shared causal variants across populations, and increases power for loci that segregate at low frequency within individual populations but are more common across the combined dataset [241–243].

In addition, statistical significant variants identified through SNV marker-based GWAS may account for only a small portion of genetic variation in the phenotype, highlighting the need for increased attention to other forms of genomic variation [244]. In general, the statistical power to detect associations between DNA variants and traits depends on the size of the experimental sample, the distribution of effect sizes among causal genetic variants segregating in the population, the frequency of these variants, and the LD between observed genotyped DNA variants and unknown causal variants [231]. According to [232], increasing sample size and diversity is more beneficial than increasing DNA variant density.

As an alternative to the relatively time-intensive sequencing of the entire genomic DNA through whole genome sequencing (WGS), sequencing of the exome and other transcribed regions of the genome is commonly employed [245, 246]. This approach enables faster GWAS; however, it evaluates only variants localized in specific regions of the genome. Although variants in transcribed regions are widely believed to be most frequently associated with phenotypic changes, mutations located outside genes and coding sequences that may potentially affect the regulation of gene expression may be excluded from the analysis [14, 247]. Targeted sequencing and the resequencing of specific loci or selected candidate genes identified through GWAS are also commonly used [248].

It is well established that, along with causal mutations, non-causal genetic variants that are in LD with them also segregate, forming a common haplotype block [249]. In livestock species, historical demographic events such as domestication, breed formation, artificial selection, and small effective population sizes have produced extended LD relative to many outbred human populations, with haplotype blocks spanning larger chromosomal intervals and elevated pairwise correlation between variants at greater physical distances [250]. Like the targeted causal variants, these non-causal variants are statistically associated with the manifestation of the trait [233]. On the one hand, this phenomenon enables the implementation of GWAS technology; on the other hand, it complicates the identification of causal variants, particularly under conditions of low genetic marker density. Moreover, the same LD-associated variants may be linked to different causal variants [8, 251]. However, such associations may be unstable due to genetic drift, recombination processes, and selective pressure, potentially disintegrating after several generations. Moreover, they are often characterized by population- and breed-specificity [14]. Accordingly, to address the problem of variants in LD and to identify truly causal variants, it may be necessary to apply multi-breed GWAS and meta-GWAS approaches. In this context, non-causal polymorphisms that show statistical associations with productivity traits due to LD are referred to as LD markers [22, 252, 253].

Because the lead variant identified in GWAS (defined as the variant with the smallest p-value) is not necessarily the causal variant and may instead represent an LD marker [254], recent methodological developments emphasize statistical fine-mapping as an essential analytical step following GWAS, aimed at distinguishing truly causal variants from correlated, non-causal markers within associated loci. A variety of approaches can be applied to fine-mapping, including heuristic methods, penalized regression models, and Bayesian methods [254]. Among these approaches, Bayesian methods have become widely used in fine-mapping analyses, as they estimate the posterior inclusion probability for each variant, providing a probabilistic measure of its likelihood of exerting a causal effect [254]. Fine-mapping has been successfully applied to identify causal variants in multiple animal species, including cattle, sheep, and pigs [9, 255–262]. One of the key priorities in the development of fine-mapping approaches has been their adaptation for variant prioritization in populations characterized by high levels of relatedness, which is typical of agricultural animal populations [262]. In addition, an important factor in improving fine-mapping performance is the integration of genomic annotations that assign biological functions to both coding and regulatory variants, as well as the integration of GWAS results with gene expression data [254].

Several methodological approaches, either independent of GWAS or complementary to GWAS, are also employed to identify causal genes and variants. For example, in certain cases, gene expression analysis is performed in relevant organs and tissues of animals for genes mapped within a single QTL. It is generally assumed that the expression level of a causal gene should correlate with the synthesis of the protein associated with the manifestation of the trait of interest [263–266]. To identify and confirm the causality of mutations, the phenotypic effect of expressing different allelic variants of a gene can be analyzed ex vivo in cell culture [267]. Moreover, experimental allelic variants that do not occur in nature can be engineered by introducing specific mutations and subsequently evaluated in vitro. One of the approaches to identifying and confirming gene causality involves gene knockout or targeted gene modification using genome editing tools such as CRISPR-Cas9 technology [268, 269]. Similar objectives can be achieved by suppressing gene expression using microRNA [7, 270] or by knockdown mediated by siRNA [271]. If the disruption or inhibition of a particular gene leads to the loss or alteration of the corresponding phenotypic trait, this provides strong evidence supporting the causal role of that gene. Comparable insights into the causal roles of individual genes can also be obtained using opposite gene-engineering strategies, such as knock-in approaches [272] or targeted gene overexpression [273, 274], which assess whether the introduction or increased expression of specific genes or allelic variants induces or enhances the trait of interest.

Gene set enrichment analysis (GSEA) and pathway analysis approaches can also be used to interpret GWAS results from the perspective of biological function and underlying mechanisms at a systems level [275]. These approaches assess whether collections of genes defined by shared participation in molecular pathways or annotated functional gene sets are statistically overrepresented among loci or genes associated with a given trait [275, 276]. Curated resources such as Gene Ontology [277, 278] and the Kyoto Encyclopedia of Genes and Genomes (KEGG) [279–281] are commonly used to test for enrichment of gene sets associated with specific biological functions or pathways among GWAS signals [282, 283]. This strategy contributes to a systems-level understanding of the biological processes underlying complex quantitative traits.

Expression and splicing quantitative trait loci (eQTLs and sQTLs) are part of a broader class of molecular QTLs, which represent genomic loci associated with variation in intermediate molecular phenotypes, including gene expression levels, alternative splicing, DNA methylation, chromatin accessibility, and protein abundance [284]. These loci are distributed across both intergenic and intragenic regions of the genome and provide a mechanistic link between DNA sequence variation and complex phenotypic traits. Elucidating the regulatory basis of trait variation therefore requires the integration of GWAS genotypic data with molecular QTL analyses derived from high-throughput profiling approaches, such as RNA sequencing. Loci supported by both GWAS and molecular QTL evidence can be interpreted as regulatory with respect to specific productivity traits. Accordingly, the joint analysis of molecular QTLs and GWAS data enables more accurate prioritization of causal mutations and their functional roles, as GWAS alone identifies variants associated with traits but does not directly reveal the underlying biological mechanisms [285].

Beyond simple overlap analyses, several integrative statistical frameworks have been developed to formally connect trait-associated loci with molecular QTLs. Colocalization analyses test whether GWAS and molecular QTL signals at a given locus are consistent with a shared causal variant, thereby strengthening causal inference and reducing the likelihood that observed associations are driven by LD alone [286–288]. Analysis based on Mendelian randomization (including summary-data-based Mendelian randomization, SMR), further extend this framework by evaluating whether the effect of a genetic variant on a complex trait is mediated through an intermediate molecular phenotype, such as gene expression or splicing [285, 289]. In addition, transcriptome-wide association studies (TWAS) integrate GWAS summary statistics with expression reference panels to identify genes for which predicted expression levels are associated with the trait of interest [290–294]. Collectively, these approaches provide complementary strategies for prioritizing candidate genes and regulatory mechanisms underlying GWAS loci [293].

To identify alternatively spliced genes and corresponding sQTLs, it is useful to compare the nucleotide sequence of the target gene with the mRNA transcribed from it. The assumption of alternative splicing is based on the absence of a deletion in the genomic DNA locus containing the analyzed gene and the presence of multiple mature mRNA isoforms detected in the cytoplasm. Causal mutations in the gene (mainly substitutions and insertion-deletion variants) that disrupt splicing signals, leading to the loss or addition of exons, are responsible for the formation of new alternative mRNAs or the absence of expected splice variants. In this case, splice variants can be the cause of changes in the translation products of alternatively spliced mRNA of the analyzed gene, potentially leading to the formation of abnormal proteins [177]. However, it should be remembered that alternative splicing plays a important role in molecular processes, contributing to the diversity of pathways that lead to different tissue phenotypes [295, 296]. This process can result in the formation of multiple transcripts and protein isoforms, which may play similar or entirely different roles across various mammalian tissues [297]. Thus, the presence of short transcripts that differ from the genomic sequence of the gene under study should be interpreted as the result of a splice variant only when considering the potential tissue specificity of alternative splicing of the mRNA precursor. Additionally, whole-transcriptome RNA sequencing technology enables the identification of all transcripts synthesized in specific organs and tissues at a given time. This approach enables the search for alternative transcript variants and, consequently, potential splice mutations not just for a single gene but for many expressed genes simultaneously. However, targeted transcript analysis and sequencing approaches have also proven effective for identifying variants affecting canonical splice sites located at exon–intron boundaries [186, 298]. In addition, such analyses may help identify the activation of cryptic splice sites caused by mutations [299].

Several methodological approaches are employed to detect CNVs in genomes, enabling their characterization and association with phenotypic traits. Genome-wide CNV detection is based primarily on hybridization-based platforms, including SNV microarrays and array comparative genomic hybridization (aCGH) [76, 80, 165, 300–302], as well as high-throughput sequencing approaches [303]. In array-based methods, CNVs are inferred from differences in signal intensity relative to a reference genome, whereas sequencing-based approaches detect CNVs through changes in read depth, discordant paired-end mapping, or split-read signals. High-coverage WGS provides improved resolution compared with array-based methods, particularly for detecting small or complex CNVs [80]. Quantitative PCR (qPCR) is also used for validation of candidate CNVs [304]. GWAS of CNV associations with productivity traits and animal diseases is a key approach for identifying copy number variation regions (CNVRs), followed by their functional annotation and validation of causality. Several reviews [80, 154] provide updated information on identified CNVRs across various animal genomes and their established phenotypic associations. When discussing STRs, which involve repeated sequences smaller than CNVs, it is noteworthy that STR mapping relies heavily on either WGS or targeted sequencing of specific regions, whereas STR typing is based on locus-specific polymerase chain reaction (PCR) amplification followed by electrophoretic separation of DNA fragments [83, 91, 100, 305].

The application of NGS techniques has enabled the WGS of multiple animals, providing extensive data on variants within genomes. Consequently, the computational processing of these data has become increasingly important. In this context, the functional annotation of variants and assessment of their potential impact using bioinformatics methods represents a relevant research direction. The Ensembl VEP enables the functional annotation of variants, including their genomic localization, associations with genes and transcripts, consequences on protein sequences, SIFT scores, and, in some cases, their population frequencies [35]. A key challenge in this field is identifying approaches to assess molecular effects of variants, thereby enabling the prediction of their causal impact on phenotype. One in silico approach focuses on the analysis of nucleotide and amino acid sequences, incorporating factors such as evolutionary conservation at specific sites and the physicochemical properties of substituted nucleotides or amino acids. Another approach examines the structural organization of nucleic acids and proteins, predicting the effects of nucleotide and amino acid variants on biopolymer structure and function. This may involve free energy calculations, the study of different conformational states, and ligand-receptor interactions, which can be carried out using computational techniques such as molecular docking and molecular dynamics simulations [23].

Given that in silico methods enable predictions of variant effects at the molecular level, they may provide valuable insights into the identification of causal variants. One study [306] investigated the impact of missense variants using in silico predictions followed by association analysis, demonstrating that impact of missense mutations on phenotypes is hard to predict in silico. In a separate study [307], researchers attempted to identify potentially causal missense variants in the pig TERT gene using solely in silico approaches. Another study [23] demonstrated the effectiveness of a multi-stage bioinformatic approach in identifying causal missense variants in pig and cattle genes. When analyzing large genomic datasets, bioinformatic methods can play a crucial role in prioritizing candidate causal variants, thereby enhancing the efficiency of association studies and optimizing the selection of animals based on valuable genetic markers. Given the conceptual parallels between this approach and rational drug design — where candidate compounds are systematically selected based on predicted biological activity — the term “rational selective breeding design” has been previously proposed [23].

The aforementioned examples illustrate the diversity of methodological approaches used to identify causal genes and variants, reflecting the complexity of the molecular processes underlying gene expression and its regulatory mechanisms, the multi-level metabolic pathways involved in gene function, and the quantitative nature of animal productivity traits. Consequently, no universal methodological strategy currently exists for establishing the causality of specific genes and variants associated with phenotypic traits [7].

Causal genes and variants as markers of productivity traits

This section presents several causal and candidate genes, as well as corresponding variants in the genomes of various animal species, that have been successfully used in breeding programs as markers of productivity traits. The list of genetic variants employed in marker-assisted and genomic selection continues to expand across different species. However, the causality of specific variants often remains a subject of debate for extended periods, until sufficient scientific evidence is accumulated to confirm consistent associations with particular traits or to elucidate the molecular mechanisms underlying their phenotypic effects.

Distinguishing between causal variants and causal genes is frequently challenging. This difficulty arises, for example, when a single gene harbors multiple variants associated with different phenotypes or with varying magnitudes of the same trait, or when robust associations have been established at the level of a gene or QTL but not for a specific variant. In such cases, the apparent association may reflect a single causal variant in strong LD with nearby neutral markers, or, alternatively, the presence of multiple causal variants within the same gene, as has been reported for several coat color-related loci. Several papers have synthesized the existing data on causal genes and variants linked to specific traits in different livestock species [7, 45, 308–310].

In this review, we highlight the most prominent examples of causal variants associated with a range of economically important animal characteristics, including growth rate, body conformation, milk and meat production and quality, reproductive performance, coat or plumage characteristics, and disease resistance. For clarity, the data are organized by individual genes for which robust associations with productivity traits have been established, with explicit indication of the variant or variants considered causal or candidate in each case. In addition, we outline effective methodological approaches that have been employed to establish the causality of specific genetic variants.

Myostatin gene (MSTN)

Myostatin, a member of the transforming growth factor β (TGF-β) superfamily, functions as a negative regulator of skeletal muscle mass in animals. It is a structurally and functionally conserved protein, and its deficiency in several vertebrate species determines the “double-muscling” phenotype [311]. Evidence for the causality of the gene in the double muscling phenotype comes from its inactivation in knockout experiments conducted in several animal species [7].

In cattle, the MSTN gene is located on bovine chromosome 2 and defines a QTL that regulates linear type traits [312, 313]. Several dozen different alleles, including SNV variants, insertions and deletions, have been detected in bovine MSTN [311, 314, 315]. Some of these genetic variants are significantly associated with the growth rate and carcass parameters of animals, while others are considered the primary cause of muscle hypertrophy [316]. The importance of MSTN as a genetic marker in commercial production has been demonstrated in various beef cattle breeds [317–319].

In various sheep breeds exhibiting the double-muscling phenotype, more than 80 SNVs in the MSTN have been identified to date [320–322]. Allelic variations in the MSTN gene have also been identified in goats; however, they are not associated with the double-muscling phenotype [323]. In horses, MSTN variants are associated with muscle fiber composition [324]. One study [325] demonstrated that MSTN may be a candidate gene for muscle hypertrophy in pigs, such as the Pietrain breed.

Thus, the MSTN gene and some its variants are putative causal in several domestic animal species. However, in not all species the causality of MSTN is linked to the double-muscling phenotype. In some species, variants are associated with other traits related to body conformation and growth performance, which do not lead to the expression of the double-muscling phenotype [311].

Insulin-like growth factor 2 gene (IGF2)

One of the most important traits in pig breeding is muscle growth and fat deposition [326, 327]. It has been established that a QTL located at the distal end of porcine chromosome 2 significantly influences these traits [328–330]. The QTL includes the IGF2 gene, which encodes the IGF2 hormone that regulates cell proliferation, growth, migration, and differentiation. The hormone is predominantly synthesized during the early stages of embryonic and fetal development in various tissues of the animal’s body. Gene expression is characterized by imprinting, with only the paternal allele being transcribed. A regulatory variant g.3072G > A in the intron 3, which has a major imprinting effect and is localized in an evolutionarily conserved CpG island (chromosome 2:1,483,817, genome assembly Sscrofa11.1), is described as a causal [326]. The mechanism of its effect is based on the disruption of a binding site for a nuclear transcriptional repressor, leading to a threefold increase in IGF2 mRNA expression in the skeletal muscles of pigs that inherited this mutation through the paternal lineage.

The QTL containing the causal g.3072G > A variant was identified through traditional mapping, and its boundaries were narrowed using the haplotype-sharing approach to a 500 kb region that included the IGF2 gene and the causal mutation, which could be identified by sequencing. The causal role of this variant was demonstrated through transcriptional analysis following the transfection of mouse myoblasts with gene constructs containing either the normal or the mutant allele. The construct containing the normal gene variant functioned as an active repressive element, whereas the fragment harboring the mutation exhibited significantly reduced repressive activity [326]. The causal g.3072G > A variant was absent in a small populations of European and Asian wild boars, as well as in pig breeds that had not been subjected to intensive selection for increased muscle growth. In contrast, this mutation has been found at high frequency in several breeds that have been subject to strong selection for lean growth [326], indicating that the mutation is susceptible to selective pressure.

Growth hormone receptor gene (GHR)

The growth hormone receptor, encoded by the GHR gene, mediates the physiological effects of growth hormone and, has been extensively investigated for its association with productivity traits in farm animals. In various species, significant associations between GHR and economically important traits have been established [331–334]. However, the causality of specific GHR variants is currently well-supported primarily in cattle, where GHR is located within a QTL on bovine chromosome 20 [335, 336]. In cows, genetic variations in the GHR have been associated with phenotypic variability in milk production and fertility [337, 338]. In proving the causality of GHR in dairy cattle, the primary role was played by widely accepted approaches, including the choice of a functionally relevant candidate gene for the expression of the aforementioned traits [339, 340], as well as GWAS [341]. Several variants have been identified within the GHR gene, among which the missense variant rs385640152 (F279Y), located in exon 8 (chromosome 20:31,888,449, genome assembly ARS-UCD2.0), is considered causal. This variant exhibits a strong association with both the quality and quantity of milk production [335, 336, 342]. In addition, rs109014416 variant (chromosome 20:32,125,768) in 5'-UTR is considered a candidate variant for traits such as the mean calving interval and age at first calving [343].

Nuclear receptor subfamily 6 group A member 1 gene (NR6A1)

The NR6A1 receptor is involved in regulating pluripotency genes and may serve as a molecular target for promoting evolutionary changes in animal body structure [344]. A QTL containing the NR6A1, which identified as a candidate associated with the number of vertebrae in pigs, is localized on chromosome 1 [345]. The number of vertebrae is an important trait in pig production, as it directly influences the size of valuable meat cuts and the overall carcass structure [346]. The most likely causal variant in the NR6A1 gene is the missense substitution rs326780270 (chromosome 1:265,347,265, genome assembly Sscrofa11.1), also known as c.748C > T, which leads to the replacement of proline with leucine at position 192 of the protein sequence (P192L) [345, 347]. The NR6A1 c.748T allele, which is associated with a greater number of vertebrae, is fixed in most Western pig breeds, while in most local Chinese short carcass breeds, it is a minor allele [345]. Thus, the causality of the NR6A1 gene in relation to the number of vertebrae in pigs is supported by the functional significance of the encoded receptor and was demonstrated through experiments mapping it to the corresponding region of porcine chromosome 1. Evidence for the causality of the rs326780270 variant lies in its localization within the domain responsible for the receptor’s interaction with corepressors, which is essential for the NR6A1 function in regulating gene expression [7]. Additionally, its strong association with the number of vertebrae in pigs and the fixation of the c.748T allele in pig breeds selected for an increased number of vertebrae further support its causality.

Melanocortin 4 receptor gene (MC4R)

The MC4R gene encodes a transmembrane G protein-coupled receptor that plays a key role in regulating energy homeostasis in animals [348]. In pigs, the MC4R gene is located in a genomic region on porcine chromosome 1 at approximately 160 Mb [349, 350], where the rs81219178 variant (chromosome 1:160,773,437, genome assembly Sscrofa11.1), also known as c.892G > A, is significantly associated with increased back fat thickness, carcass weight, moisture content, and saturated fatty acids levels in meat [351–354]. This variant results in an D298N substitution in the receptor protein, altering its structural and functional properties [351]. Evidence for the causality of rs81219178 is supported by unambiguous association study results obtained in pigs of different breeds and in divergently selected populations within a single breed [355]. Molecular evidence for causality was obtained through functional analysis of MC4R allelic variants expression in a human cell line [356]. It was demonstrated that, unlike the D298 variant, the N298 MC4R variant does not stimulate cAMP production in response to NDP-α-MSH stimulation. Whereas, the D298 variant is essential for normal MC4R signaling to adenylate cyclase. Thus, the N298 allele leads to a loss of normal receptor function and a decrease in melanocortin signaling in pigs.

Protein kinase AMP-activated non-catalytic subunit gamma 3 gene (PRKAG3)

PRKAG3 gene is associated with meat quality in pigs [357]. A missense rs1109104772 variant (chromosome 15:120,863,533, genome assembly Sscrofa11.1) in this gene (also known as c.749G > A), which results in a non-conservative R250Q substitution, leads to increased glycogen levels in meat [358, 359], a low pH, and a deterioration in meat product quality [360]. Evidence for the causality of the c.749G > A substitution is based on the fact that the PRKAG3 gene is part of a QTL associated with pork meat quality traits [357] and was initially considered functionally significant for the formation of its key characteristics. Several original studies reviewed in [7] demonstrated a strong association between the c.749G > A variant and pork meat quality.

ATP-binding cassette transporter G2 gene (ABCG2)

The membrane-associated protein ABCG2, with broad substrate specificity, can expel various physiological compounds, toxins, and xenobiotics from cells. In cows, the ABCG2 gene is part of a QTL influencing the milk protein spectrum, as well as the synthesis and composition of milk fats [7, 12, 361]. In one study [263] the missense variant rs43702337 (chromosome 6:36,599,640, genome assembly ARS-UCD2.0) was described, which corresponds to the amino acid substitution Y581S and is associated with enhanced transport function in both in vitro and in vivo experiments. The Y581S variant of the bovine ABCG2 gene is associated with the efficiency of drug secretion into milk, including antibiotics such as danofloxacin and enrofloxacin, which are widely used in veterinary medicine [362]. This genetic variation has also been reported to influence milk yield and composition [263, 363]. The evidence supporting the causality of the ABCG2 gene is based on the physiological relevance of the encoded protein’s function, as well as the analysis of gene expression across different physiological stages in cows, compared to the expression of two other genes (SPP1 and PKD2) within the candidate region [7, 364]. The Y581S mutation was the only variant found to be associated with milk fat and protein composition, providing evidence of its causality [263].

Diacylglycerol acyltransferase 1 gene (DGAT1)

Several studies in cattle have identified QTLs for milk production traits on chromosome 14 [365, 366]. Linkage mapping studies led to the identification of a strong candidate gene in this region encoding DGAT1 [364, 367, 368], an enzyme that regulates the rate of triglyceride synthesis in adipocytes. A number of studies reviewed in [364], demonstrated a polymorphism in exon 8 of the DGAT1 gene in Bos taurus, characterized by a double AA to GC substitution (chromosome 14:611,019-611,020, genome assembly ARS-UCD2.0; annotated as two SNVs rs109234250 and rs109326954 instead of MNV in Ensembl). This substitution resulted in an amino acid K232A change in the encoded protein, and was associated with milk protein and fat content, as well as milk yield. One review paper [369] presents findings indicating that an increase in milk fat and protein percentages is associated with the K232 variant of DGAT1, while an increase in milk yield is linked to the A232 variant in dairy cattle. The determination of the causal relationship of the DGAT1 gene with milk productivity traits in cattle is based on its functional significance and its belonging to orthologous genes, as DGAT1 is associated with milk productivity traits in different species. In addition, its functionality is confirmed by experiments in which both copies of the gene were knocked out in mice, leading to the absence of lactation in knockout animals [370]. Evidence for the causality of the K232A mutation includes its established association with milk yield, protein and fat content, and fatty acid composition. In vitro functional analysis demonstrated that the gene variant encoding K232 of the DGAT1 enzyme led to the synthesize of 1.5 times more triglycerides than the variant with A232 [7].

Prolactin receptor gene (PRLR)

The prolactin receptor belongs to the class I cytokine receptor superfamily of transmembrane receptors. In cattle, the prolactin receptor gene (PRLR), due to its functional relevance, has consistently been considered a potential genetic marker for a number of productivity traits. Indeed, several mutations in the PRLR gene have been shown to be associated with shorter hair and lower follicle density — traits characteristic of the “slick” hair phenotype in the Senepol cattle breed. The corresponding locus, which includes the PRLR, was mapped through linkage analysis [7]. In the study [371] it was proposed a causal variant rs517047387 in the PRLR involving a one-nucleotide frameshift deletion (chromosome 20:39,099,214, genome assembly ARS-UCD2.0) which introduces a premature stop-codon and loss of 120 C-terminal amino acids in PRLR of Senepol cattle. Several other causal variants, also encoding truncated proteins in different cattle breeds, have been reported [7, 372]. In addition, the study [373] identified multiple associations of polymorphisms in PRLR with milk production, reproductive traits, and heat stress response characteristics in a Holstein population using SNV-based association analysis.

In birds, PRLR also exhibits a causal role. In commercial poultry production, autosexing is employed — an early sex differentiation of chicks based on indirect phenotypic traits. One such sexually dimorphic trait is the rate of leg feathering at hatching. In chickens and turkeys, the PRLR gene, which is involved in reproductive and other essential physiological processes, also influences feathering morphology. In turkeys, a causal polymorphism associated with this trait has been identified — a 5-base pair deletion (chromosome Z:9,426,019-9,426,023, genome assembly Turkey_5.1) in the last exon of the PRLR gene (resulting in a frameshift), which leads to the loss of a portion of the receptor protein at its C-terminus, specifically 98 amino acids [374]. The identification of the causal deletion in turkeys is the result of a GWAS study, which localized a genomic region on the Z chromosome that was statistically associated with chick feathering rate at hatching, followed by the discovery of a 5-bp deletion within that region. In chickens, the molecular basis for the effect on feathering rate is also the loss of a portion of the PRLR protein (149 amino acids from the C-terminus of the receptor), but this is caused by a causal tandem duplication of 176 kb that encompasses part of the PRLR gene [375, 376]. To detect this duplication in the chicken genome, quantitative PCR was performed to determine CNV, along with sequencing of the sex-linked locus responsible for feathering. These two examples illustrate similar selective pressures on the same trait in different domestic bird species, resulting in distinct mutations, each of which leads to a truncation of the same protein.

Ryanodine receptor 1 gene (RYR1)

Ryanodine receptors (RYR1, RYR2, RYR3) are membrane-associated calcium-release channels [377, 378]. In both humans and various animal species, variants in the genes encoding these receptors have been investigated for their associations with a range of diseases [377–379]. Among these, RYR1 variants are particularly notable, having been identified as candidate or causal for malignant hyperthermia across multiple species, thereby providing strong evidence for the causality of RYR1 gene [380].

In pigs, the RYR1 gene is located within the halothane locus, mapped to chromosome 6. This gene plays a critical role in the regulation of porcine stress syndrome, a condition characterized by susceptibility to malignant hyperthermia [381, 382]. The economic value of stress-sensitive animals is diminished due to reduced meat quality and hypersensitivity to environmental stressors. However, these animals are also characterized by a higher carcass meat content [383]. The RYR1 variant rs344435545 (chromosome 6:47,357,966, genome assembly Sscrofa11.1) has been identified, with the T allele being associated with stress sensitivity [383]. This variant is also known as g.1843C > T and causes R615C substitution in the protein. Genotyping of pigs for this variant has become common in breeding and crossbreeding programs, both to improve herd health and to produce fattening pigs heterozygous for the RYR1 gene. Heterozygous individuals are characterized by high meat yield, stress resistance, and an absence of the meat quality defects typically observed in stress-sensitive animals. The causality of the rs344435545 polymorphism in pigs is confirmed by its strong association with manifestations of stress syndrome across breeds and the concordance between RYR1 rs344435545 genotypes and results from the classical halothane stress sensitivity test [7].

In horses, the missense variant in RYR1 originally referred as c.7360C > G and resulting in the R2454G amino acid substitution [384] may be considered as putative causal variant. In the reference genome assembly EquCab3.0 this variant is localized at chromosome 10 in position 9,678,680 (no rs ID available) and contributes to R2455G substitution in protein. Among horses heterozygous for this variant, approximately 51% of deaths are attributed to anesthesia-induced malignant hyperthermia or severe acute rhabdomyolysis with hyperthermia. Notably, no homozygous individuals have been identified, which may suggest that homozygosity for this variant is incompatible with life [385].

In dogs, malignant hyperthermia has been associated with the c.1643T > C candidate missense variant in RYR1, resulting in the V548A substitution (chromosome NC_006583.3:g.114562165, genome assembly CanFam3.1) [386], originally reported as V547A [387]. One of the most recent studies also identified RYR1 candidate variants leading to P152L and G2375R substitutions in dogs with malignant hyperthermia [388]. These findings underscore the need for further investigation into additional causal variants within RYR1.

Synaptogyrin-2 gene (SYNGR2)

Synaptogyrin-2 is an integral membrane protein, whose functions, as well as those of the genes encoding the entire synaptogyrin family, remain largely unknown but are at least partially related to transport in neuronal cells [389]. A direct interaction between SYNGR2 and the non-structural proteins of several viruses has been established, which plays a role in the formation of intraplasmic inclusions during the infectious process [390]. A GWAS of viral load in crossbred pigs infected with PCV2 identified two loci in chromosome 12 [270], one of which was the precisely mapped SYNGR2 gene. A putative causal missense variant rs3473454700 (chromosome 12:3,797,515, genome assembly Sscrofa11.1) was detected in the gene itself, corresponding to the amino acid substitution R63C, located in a conserved domain of the protein.

According to [7], evidence for gene causality is based on the analysis of SYNGR2 expression following circovirus exposure [270], the identification of SYNGR2’s role in viral replication, and the observed reduction in PCV2 replication after gene knockout using siRNA. Evidence for the causality of the polymorphism was provided by association study, which demonstrated that the strongest association with viral load was observed for the missense variant R63C, rather than for another locus with an indel in LD with R63C and located near the BIRC5 gene. The exclusion of this indel as a causal variant was based on a CRISPR-Cas9 genome editing experiment and the lack of specific expression of the BIRC5 gene.

Genes associated with hindlimb feathering in birds

Leg feathering is a trait under strong selection in ornamental chicken breeds. The two main loci controlling this trait are located on chromosomes 13 and 15. Within these loci, one of candidate causal variants is an SNV located 25 kb upstream of the TBX5 gene (T-box transcription factor 5) on chromosome 15. Another is a 17.7 kb deletion located approximately 200 kb upstream of the PITX1 (paired-like homeodomain 1) transcription factor gene on chromosome 13 [391]. These mutations activate TBX5 expression and suppress PITX1 expression, respectively. Remarkably, in domestic pigeons, a similar phenotype is also caused by non-coding mutations located upstream of the same two genes, TBX5 and PITX1. Genetic evidence identifying the candidate causal variants at these loci was obtained through a combination of classical linkage mapping and high-resolution mapping based on WGS data from chickens with and without feathered legs [391].

Genes regulating prolificacy in sheep

Three genes that influence prolificacy in sheep, including ovulation rate and litter size, are commonly referred to as fecundity (Fec) genes: the bone morphogenetic protein receptor type 1B (BMPR1B, also known as FecB) gene located on chromosome 6; the bone morphogenetic protein 15 (BMP15, or FecX) gene located on the X chromosome; and the growth differentiation factor 9 (GDF9, or FecG) gene located on chromosome 5 [392]. These three genes encode proteins that belong to the TGF-β superfamily [393]. BMP15 and GDF9 are secreted from oocytes and bind to the BMPR1B receptor on granulosa cells, participating in the regulation of cellular differentiation, follicular atresia, and oocyte maturation. This interaction suggests the potential for a synergistic effect of polymorphisms among these genes [394].

In 2001, three independent research groups [395–397], employing conceptually similar methodologies based on QTL mapping followed by sequencing, identified that the known FecB^B mutation responsible for the Booroola phenotype in sheep is SNV (rs418841713; chromosome 6:30,050,621; genome assembly ARS-UI_Ramb_v2.0) in the BMPR1B gene. This mutation results in a Q249R amino acid substitution and leads to a 1.5- to 3.0-fold increase in ovulation rate, corresponding to approximately 1.0 and 1.5 additional lambs per lambing in heterozygous and homozygous ewes, respectively [398].

Several mutations in the BMP15 gene have been identified with putative causal effects on prolificacy or as candidate variants for such a role [398–400]. These include: FecX^H (rs413916687; chromosome X:54,284,142), a stop-gained mutation (Q23Stop) found in the Hanna breed; FecX^I (rs398521635; chromosome X:54,284,117), encoding a V31D substitution in the mature protein in the Inverdale breed; FecX^L (rs3508196091; chromosome X:54,284,051), encoding a C53Y substitution found in Lacaune sheep; FecX^B (rs3508195618; chromosome X:54,283,913), encoding an S99I substitution found in the Belclare breed; as well as two variants affecting the unprocessed protein: FecX^G (rs425019156; chromosome X:54,284,295), introducing a Q240Stop (also known as Q239Stop), and FecX^R (rs421419167; chromosome X:54,284,537-54,284,553), encoding a 17-amino-acid frameshift deletion, found in the Cambridge and Rasa Aragonesa breeds, respectively. Genomic localization of these polymorphisms has been facilitated through QTL mapping, sequencing, and other molecular techniques such as single-strand conformation polymorphism (SSCP) detection, and duplex PCR [401–404]. Additionally, novel variants in BMP15 continue to be reported [399, 405].

Numerous allelic variants with missense or synonymous consequences have also been identified in the GDF9 gene through sequencing, with mutations localized in the signal peptide, propeptide, and mature peptide regions. These include G0 [406], G1-G8 (with G7 also referred to as FecG^F [407, 408] and G8 as FecG^H) [403], FecG^E (also FecG^SI) [409, 410], FecG^V [411], FecTT [412], FecG^A [413], S2 [414], and others [406, 415]. Among these, the following mutations are frequently highlighted for their effects on fertility and ovulation rate: G1 (rs410123449; chromosome 5:42,116,345), encoding an R87H substitution found in Moghani, Ghezel, and Garole breeds; FecG^F (rs403536877; chromosome 5:42,114,376), encoding V371M in Norwegian White sheep (Finnsheep); FecG^H (rs3487550980; chromosome 5:42,114,303), encoding S395F in the Belclare and Cambridge breeds; FecG^E (rs1092755620, position 5:42,114,453), encoding F345C in the Santa Ines breed; FecG^V (rs3508195812; chromosome 5:42,114,544), encoding R315C in Brazilian sheep (Vacaria allele); and FecTT (rs3508195635; chromosome 5:42,114,208), encoding S427R in the Thoka breed [415, 416].

It is important to note that when studying variants in BMPR1B, BMP15, and GDF9, the effects of combinations of allelic variants across different polymorphisms are of great significance. Moreover, some variants may cause infertility in the homozygous state [398].

Genes regulating coat color

A number of genes are involved in the regulation of melanogenesis and the determination of coat color, including melanocortin 1 receptor (MC1R), agouti signaling protein (ASIP), tyrosinase (TYR), tyrosinase-related protein 1 (TYRP1), tyrosinase-related protein 2 (TYRP2), KIT proto-oncogene receptor tyrosine kinase (KIT), melanocyte-inducing transcription factor (MITF), and endothelin receptor type B (EDNRB), among others [417, 418]. These genes define several allelic series that collectively determine the final pigmentation phenotype, including the Extension (E), Agouti (A), Albino (C), and Dilute (D) loci [419].

The MC1R and ASIP genes, associated with the extension and agouti loci respectively, exhibit epistatic interaction. In the presence of the dominant extension allele (E^D) at the MC1R locus, binding of α-melanocyte-stimulating hormone (α-MSH) to MC1R is activated, leading to the production of eumelanin (black or brown coat color). Conversely, the homozygous presence of the recessive e allele results in the synthesis of pheomelanin (red or yellow coat color). When the wild-type extension (E⁺) allele is present, coat color is modulated by the action of ASIP. The dominant A allele of ASIP encodes a protein that competes with α-MSH for MC1R binding, promoting pheomelanin synthesis, while the recessive a allele permits continued eumelanin production [417–419].

In sheep, the missense variant M73K (rs3508196008; chromosome 14:14,251,947, genome assembly ARS-UI_Ramb_v2.0) and D121N (rs409651063; chromosome 14:14,252,090) in MC1R are causal for the dominant E^D phenotype [420], while the R67C mutation (rs3508195989; chromosome 14:14,251,928) is associated with the recessive e phenotype [421]. In pigs, the L102P substitution, originally annotated as L99P (rs45434630; chromosome 6:181,883 in genome assembly Sscrofa11.1), and D124N, originally annotated as D121N (rs326921593; chromosome 6:181,818), in MC1R are responsible for the dominant E^D1 (Meishan, Large Black) and E^D2 (Hampshire) phenotypes, respectively [422–424]. The A243T variant, originally known as A240T (rs321432333, chromosome 6:181,461), is associated with the recessive e allele in Duroc pigs [422–424]. Additionally, a 2-bp frameshift insertion (rs5522776030, chromosome 6:182,120-182,121), which may represent a case of somatic reversion, in combination with D121N, contributes to the black-spotted phenotype on a red or white background (denoted as E^p) as observed in the Pietrain breed [422, 424, 425]. Recent studies utilizing CRISPR/Cas9 gene editing in the Duroc breed have significantly advanced the understanding of the functional consequences of MC1R mutations [426].

In cattle, the L99P substitution (rs109688013; chromosome 18:14,705,671, genome assembly ARS-UCD2.0) in MC1R is associated with the dominant black (E^D) phenotype, while a 1-bp frameshift deletion (rs110710422; chromosome 18:14,705,684-14,705,685) results in the recessive red (e) phenotype [427, 428]. Several additional variants in cattle are compatible with the wild-type E⁺ phenotype [427, 429] which is responsible for combinations of red or reddish brown and black coat coloration. In horses, the S83F substitution (rs68458866; chromosome 3:36,979,560, genome assembly EquCab3.0) is responsible for the chestnut coloration and represents a recessive mutation (e) at the MC1R locus [430].

The ASIP gene also is a key regulator of melanogenesis, playing an important role in determining coat color in animals [431], while also contributing significantly to the modulation of lipid metabolism [432]. The inheritance of coat color and the function of the Agouti locus have been extensively investigated through classical genetic approaches [433, 434]. Molecular genetic studies, including GWAS, have further validated the central role of the Agouti locus in coat color determination and have identified various gene variants encoding this protein [435].

In sheep, the dominant white allele (A^Wt) is the result of a structural rearrangement involving the duplication of the ASIP and ACHY coding regions along with the promoter region of ITCH [157, 436]. Conversely, loss-of-function variants such as the deletion g.100_105del (D5) and the g.5172T > A variant in ASIP have been associated with the formation of a recessive allele that results in a black coat phenotype [157, 418, 437]. One of the most compelling demonstrations of ASIP’s causal role in coat color variation in sheep comes from gene editing experiments using CRISPR/Cas9 [436].

In cattle, WGS has revealed an upstream structural variant, ASIP-SV1, comprising a 1155-bp deletion followed by the insertion of a transposable element of more than 150 bp, which is associated with a dark coat phenotype [438]. Additionally, an in-frame deletion located in the last exon (rs519457228; chromosome 13:63,667,797-63,667,802, genome assembly ARS-UCD2.0) is also considered a candidate variant influencing pigmentation [438]. In horses, a recessive 11-bp deletion in the ASIP gene (rs3091770233; positions 22:26,067,463-26,067,473, genome assembly EquCab3.0) has been identified using sequencing and cross-species PCR amplification; this deletion leads to loss of gene function and results in a black coat color phenotype [417, 439, 440].

The KIT gene regulates melanocyte migration and has impact on coat color. Variants in KIT can result in dominant white phenotypes in several species. In pigs, for instance, dominant KIT alleles can mask MC1R variation due to epistatic effect. The recessive i allele represents the wild-type form of KIT, while the dominant white phenotype arises from a complex structural mutation involving a 450-kb CNV and a splice mutation in intron 17. This splice mutation mutation occurs in at least one copy of the gene (alleles I¹, I², or I³) and is characteristic of breeds such as Large White and Landrace [422, 424]. When two normal copies of KIT are present, the phenotype with black patches on a white coat (I^P) is observed [441, 442]. If other regulatory variant affects a single copy of the gene in the absence of the splice mutation, a belted phenotype (I^Be), as seen in Hampshire pigs, can occur [443]. SSCP, quantitative PCR and sequencing methods played a major role in identifying the causality of KYT gene variants in pigs [441–443].

In horses, numerous KIT variants have been identified, conventionally designated W1-W35. These include insertions, deletions, and variants with missense, stop-gain, and splice-site consequences [417, 444]. Multiple KIT variants may co-occur within individuals as combinations of up to three have been documented. Many of KIT variants are not observed in the homozygous state, likely due to their lethality [417, 444]. Other KIT mutations in horses are associated with distinct pigmentation patterns, such as Tobiano, Sabino, and Roan [417, 444]. In cattle, KIT is considered a candidate gene for spotted coat phenotypes, though the causality of specific variants remains to be conclusively demonstrated [428].

Additionally, other genes such as TYR and TYRP1 have been implicated in coat color determination. For example, in Asinara white donkeys, the H202D substitution in TYR results in albinism [417]. In Chinese pig populations, a 6-bp deletion in TYRP1 has been associated with the brown coat color phenotype [422]. In cattle, a frameshift insertion in the TYR gene results in a premature stop codon at residue 316 in the homozygous state, causing albinism in Brownveih and Brown Swiss breeds. Furthermore, in Dexter cattle, the TYRP1 H434Y (rs3423268465; chromosome 8:31,633,328, genome assembly ARS-UCD2.0) substitution is responsible for the brown coat color [428], while in sheep, the C290F (rs402624085; chromosome 2:81,188,093, genome assembly ARS-UI_Ramb_v2.0) substitution in TYRP1, when homozygous, causes a shift from dark to light coat coloration [445].

Descriptions of the genetics underlying coat color formation and the associated causal variants in these and other pigmentation-related genes are available in several reviews focused on species such as cattle [428], sheep [418], pigs [422], horses, and donkeys [417, 419, 444].

Application of causal variants in animal selective breeding

In this review, we presented multiple examples of genetic variants associated with animal productivity traits. Collectively, these cases illustrate the methodological heterogeneity involved both in variant discovery and in the subsequent establishment of causality. Numerous additional examples of causal and candidate genetic variants already applied, or proposed for application, in breeding programs and genotype-based animal evaluation are documented in a wide range of original and review articles [7, 45, 308–310].

The primary motivation for identifying causal variants in animal populations lies in their potential use as direct genetic markers of economically important traits. Direct markers are generally more effective than markers based on LD, although they are also more difficult to identify [446]. The ultimate objective is the targeted artificial selection in animal populations, aimed either at increasing the frequency of favorable genotypes associated with enhanced productivity and disease resistance, or at eliminating genotypes linked to heritable diseases or low production performance. Such directed modification of the genetic structure of a population can be implemented through marker-assisted selection or genomic selection frameworks [446–449].

Marker-assisted selection is based on the use of stable, heritable genetic markers to guide animal selection [446, 447]. Conceptually, selection is performed on a marker that is statistically associated with the target trait under improvement. The effectiveness of marker-assisted selection depends strongly on the nature of the marker itself and on the genetic architecture of the trait, including the number of loci contributing to phenotypic variation [446, 447]. Consequently, the efficiency of marker-assisted selection may be reduced or even absent when LD-based markers are used, as their linkage to true causal variants may decay over generations or fail to reproduce in genetically distinct populations [450]. Similar limitations arise when the selected marker corresponds to a variant that explains only a small proportion of the total quantitative variance of the trait. Therefore, a challenge for marker-assisted selection is the polygenic nature of most complex traits [446, 451], as well as the possible presence of multiple independent causal variants within a single gene. For example, in the case of MSTN gene, several SNVs and insertion-deletion polymorphisms have been shown to exert putative causal effects on production-related traits in cattle and sheep [7, 311].

In this context, an important aspect concerns what should be regarded as a truly causal variant. In most cases, a causal variant is defined as one that has a direct effect on the phenotype and on one or more traits [446, 452]. At present, the most high-throughput approach for identifying such variants (and potentially the largest source of candidates) is GWAS combined with fine-mapping, which can statistically support a direct effect of a variant on a trait [256]. However, from a molecular biology perspective, a direct effect on the phenotype implies the existence of an underlying biological mechanism explaining how the variant influences the trait [453].

Such molecular mechanisms may involve changes in the structure or function of gene products, including proteins or various classes of RNA, or may operate at the regulatory level, for example through effects on transcriptional activity, alternative splicing, DNA methylation, or other epigenetic or regulatory processes. Thus, the presence of a specific molecular mechanism of action may be considered a necessary criterion for defining a variant as causal, although establishing such a mechanism is not always feasible. Importantly, molecular effects based on specific biological processes are expected to be relatively stable across individuals, unless compensated for or neutralized by other genetic or regulatory mechanisms. Consequently, a truly causal marker is more likely to retain its effect across generations and is less likely to be specific to a particular breed or line within a species.

In contrast, the inability to identify or validate the molecular mechanism underlying a variant’s effect on a phenotype may result in the introduction of markers into marker-assisted selection schemes that are supported solely by statistical association. Such markers may yield irreproducible or population-specific results [446]. A well-known example is SNV in the PRLR gene of pigs, which has been described in the literature as a candidate marker for reproductive performance for more than two decades. Depending on the population studied, either the reference allele, the alternative allele, or no allele has been reported to be associated with improved reproductive traits [454–459]. However, other studies failed to identify reproducible QTL related to reproductive function in this genomic region, casting doubt on the causal role of this variant [460, 461].

A significant step forward in breeding methodologies is genomic selection, which was initially described as marker-assisted selection on a genome-wide scale [462, 463]. Unlike classical marker-assisted selection, which is typically restricted to a limited panel of variants in individual genes, genomic selection allows the effects of genetic variation across the entire genome on target productivity traits to be taken into account [449, 451, 464]. Theoretically, genomic selection can provide a substantial advantage over marker-assisted selection due to the much larger number of markers that can be incorporated into the model [451]. In practice, genomic selection also encounters a number of implementation difficulties [464], moreover its performance depends critically on the underlying LD structure of the population. In most cases, the majority of markers included in genomic prediction models are not themselves causal but are in LD with causal variants. Consequently, the accuracy of genomic prediction depends on the extent to which LD between markers and causal variants is preserved [451].

Over successive generations, recombination reduces the strength of LD between markers and causal loci, particularly when selection candidates become genetically distant from the reference population used to train the model. This decay of LD leads to a reduction in the accuracy of estimated breeding values [451]. A related limitation arises in the context of cross-population or cross-breed prediction, where differences in LD structure and allele frequencies between populations can substantially reduce the transferability of genomic prediction models [451, 465–467].

Several strategies can mitigate these limitations. Increasing the size of the reference population and the density of genotyped markers may improve the likelihood that markers are in strong LD with causal variants. Incorporating multiple breeds or populations in training datasets can also enhance the robustness of predictions across diverse genetic backgrounds. Ultimately, these approaches aim to improve the tagging of causal variants, either by identifying them directly or by using markers that are in very close LD with them [451, 465–469].

Furthermore, in genomic selection, the effects of all analyzed variants are aggregated to estimate the breeding value of an animal. The commonly used genomic best linear unbiased prediction (GBLUP) model incorporates marker information through a genomic relationship matrix and assumes that all marker effects are normally distributed with equal variance, implying that all markers are expected to contribute equally to the trait [11, 451]. Several approaches have been proposed to relax this assumption. Bayesian genomic selection methods, for example, typically assume that only a small proportion of variants have non-zero effects and assign higher weights to variants that are more likely to be in LD with causal loci. An alternative strategy is a two-stage approach in which a GWAS is first used to identify variants with the strongest effects on the target trait, and these variants are subsequently incorporated into the genomic prediction model. In addition, QTL information can be integrated into machine learning frameworks to improve the prediction of breeding values [11, 451].

One particularly promising direction is the use of functional annotations of genetic variants, such as their localization within coding sequences, promoters, enhancers, or other regulatory elements, and the incorporation of this information into Bayesian models, such as BayesRC [470], to prioritize variants with a higher likelihood of being causal [471]. This concept is actively promoted within the Functional Annotation of Animal Genomes (FAANG) initiative [471].

The incorporation of variant weights derived from bioinformatic predictions may further enhance genomic selection models. Previous study has demonstrated that several known causal missense variants can exert effects on protein structure and function when evaluated using combinations of predictive bioinformatic tools [23]. In this context, modern deep learning-based methods, such as AlphaMissense [472] and AlphaGenome [473], which aim to predict the functional consequences of missense and non-coding variants, respectively, offer the potential to systematically prioritize candidate causal variants across animal genomes.

Taken together, these developments suggest that continued refinement of genomic selection models toward weighted frameworks that integrate genome annotation data, bioinformatics-based effect predictions, and fine-mapping GWAS results is both feasible and promising. Under such a paradigm, genomic selection informed by causal variant data across the entire genome is expected to achieve maximal efficiency [469].

Conclusions

The preferential use of causal variants as genetic markers for productivity traits offers a more precise evaluation of the genetic potential of animals and facilitates faster progress toward desired breeding outcomes. Nevertheless, identifying such variants remains a complex task, particularly given the polygenic inheritance patterns that underlie most productivity traits. A key challenge lies in the absence of a standardized methodological framework to determine the causality of variants in any given context. This complexity arises from the diverse physical nature of mutations and their corresponding polymorphic variants (sequence and structural variant types) which exert their influence through varied mechanisms on gene expression and the structure of gene products (variant consequences). These mechanisms are further complicated by the specific pathways through which phenotypes are formed.

Conceptually, the search for causal variants follows a general trajectory: identifying trait-associated regulatory regions, pinpointing candidate genes and functional sequences, and ultimately detecting specific causal nucleotides. However, methodological approaches must be adapted to the variant type, its consequence regarding transcript or protein product, and the specific parameters of the study design. Furthermore, the lack of clear criteria distinguishing truly causal variants from those merely associated with phenotypes presents a significant barrier to the development of a cohesive theory of genetic causality. This is compounded by limited understanding of the molecular mechanisms underlying many associations, insufficient replication of results across diverse populations, and the difficulty in distinguishing direct causal variants from those in LD with causal loci.

We propose that several key features — direct phenotypic influence, a clearly defined molecular mechanism, and predominantly population-independent effects — can serve as foundational criteria for establishing causality. These principles may facilitate the development of more consistent methodological standards for the identification and validation of causal variants, ultimately enhancing the reliability of genetic markers derived from them.

Abbreviations

α-MSH

α-Melanocyte-stimulating hormone

aCGH

Array comparative genomic hybridization

ABCG2

ATP-binding cassette transporter G2

AFLP

Amplified fragment length polymorphism

ASIP

Agouti signaling protein

BMPR1B

Bone morphogenetic protein receptor type 1B

BMP15

Bone morphogenetic protein 15

CNV

Copy number variation

CNVR

Copy number variation region

DGAT1

Diacylglycerol acyltransferase 1

EDNRB

Endothelin receptor type B

EVA

European Variation Archive

FAANG

Functional Annotation of Animal Genomes

GBLUP

Genomic best linear unbiased prediction

GDF9

Growth differentiation factor 9

GHR

Growth hormone receptor

GSEA

Gene set enrichment analysis

GWAS

Genome-wide association study

IGF2

Insulin-like growth factor 2

KEGG

Kyoto Encyclopedia of Genes and Genomes

KIT

KIT proto-oncogene receptor tyrosine kinase

LD

Linkage disequilibrium

MC1R

Melanocortin 1 receptor

MC4R

Melanocortin 4 receptor

MITF

Melanocyte-inducing transcription factor

MNP

Multi-nucleotide polymorphism

MNV

Multiple nucleotide variant

MSTN

Myostatin

NGS

Next-generation sequencing

NR6A1

Nuclear receptor subfamily 6 group A member 1

PCR

Polymerase chain reaction

PITX1

Paired-like homeodomain 1

qPCR

Quantitative polymerase chain reaction

QTG

Quantitative trait gene

QTL

Quantitative trait loci

QTN

Quantitative trait nucleotide

PRKAG3

Protein kinase AMP-activated non-catalytic subunit gamma 3

PRLR

Prolactin receptor

RYR1

Ryanodine receptor 1

SINE

Short interspersed nuclear element

SMR

Summary-data based Mendelian randomization

SNP

Single nucleotide polymorphism

SNV

Single nucleotide variant

SO

Sequence Ontology

SSCP

Single-strand conformation polymorphism

STR

Short tandem repeat

SYNGR2

Synaptogyrin-2

TBX5

T-box transcription factor 5

TE

Transposable element

TWAS

Transcriptome-wide association study

TYR

Tyrosinase

TYRP1

Tyrosinase-related protein 1

TYRP2

Tyrosinase-related protein 2

UTR

Untranslated region

VEP

Variant Effect Predictor

VNTR

Variable number of tandem repeat

WGS

Whole genome sequencing

Authors’ contributions

V.B. and M.P. were responsible for preparation of sections “Genetic variation in animals,” “Molecular mechanisms of causality of genomic variants,” and “Methodological approaches to identifying causal genes and variants.” V.B., A.S., and M.P. were responsible for preparation of section “Causal genes and variants as markers of productivity traits.” M.P. was responsible for preparation of section “Application of causal variants in animal selective breeding.” V.B. prepared Introduction and Conclusions. All authors read and approved the manuscript.

Funding

The authors did not receive any specific funding for either the study or its publication.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.