Local Ancestry and Selection in the Genomes of Russian Black Pied Cattle

Abstract

The Russian Black Pied (RBP) is one of Russia’s most popular dairy cattle breeds. It was developed in the USSR during the 1930s by crossing Russian native cattle with Dutch cattle. Since the mid-1970s, the RBP has been cross-bred with Holstein cattle to enhance milk production. To trace haplotypes and selection signatures in the RBP genomes formed during cross-breeding and selection, we conducted local ancestry inference and scans for selection signatures in 12 resequenced RBP animals. Scans for selection using hapFLK and window-based F_ST identified 65 putatively selected regions across the genome, with 4 identified by both methods. Candidate missense variants were detected within these regions, including KIAA1755, CNBD1, and MPZL3, which may be related to milk traits, pathogen resistance, and climate adaptation. Local ancestry inference and functional annotation analyses suggest that regions with a higher fraction of native haplotypes are enriched for genes associated with keratin filaments, antimicrobial defence, the immune system, protein digestion, and amino acid transporters. The results of our study could be utilised for conservation purposes and the improvement of this and other breeds.

1. Introduction

Cattle are a vital milk, meat, and draft power source worldwide. In modern Russian animal husbandry, cattle farming primarily focuses on milk production, with dairy and dual-purpose pedigree herds numbering 2.6 million head (compared to 0.3 million for beef herds) in 2023 [1,2]. One of the most popular dairy breeds is the Russian Black Pied (RBP). This breed was developed in the USSR during the 1930s by crossing local Russian cattle from various climate zones with imported breeds of Dutch origin, primarily the Dutch Black Pied [3,4]. Since the mid-1970s, the cross-breeding of the RBP with the Holstein breed, commonly referred to as “Holsteinisation”, has occurred with the intention of increasing milk productivity [3,4]. Consequently, the RBP now possesses a significant proportion of the Holstein ancestry [5,6]. Recently, many highly Holsteinised RBP populations have been officially recognised as Holstein. As a result, the proportion of the RBP in the national pedigree dairy herd decreased from 55.6% in 2015 to 17.6% in 2023 [1]. Concerns exist regarding the Holsteinisation of the RBP leading to the loss of unique, potentially adaptive alleles passed down from local cattle populations [5].

Despite extensive cross-breeding with Holstein, the RBP retains adaptive features inherent to Russian native cattle. For instance, in winter, the coat of the RBP provides superior thermal insulation properties compared to pure Holsteins [7,8]. Also, the comparative genetic data suggest that the RBP harbours alleles conferring genetic resistance to leucosis development at higher frequencies than the Holstein breed [9]. Given the complex origin of the RBP, uncovering the changes in their genome resulting from cross-breeding and follow-up selection could be essential for conservation purposes and genetic improvement of this breed.

One approach to detecting such genetic changes is scanning for signatures of selection. Selection signatures are genomic footprints left by artificial or natural selection. Examples include shifts in allele or haplotype frequencies in extreme cases, resulting in runs of homozygosity and longer haplotypes at the selected sites [10]. The analyses for signatures of selection in cattle breeds that are exceptionally adapted to their environments or chosen for productive characteristics offer insights into the genetic foundation of their distinctive traits. For instance, a study on Yakut, the northernmost cattle breed, revealed a unique breed-specific amino acid change in the highly conserved NRAP gene. This mutation was absent from other cattle breeds but present in every Yakut cattle individual. Moreover, this mutation was also observed in 16 species of hibernating or cold-adapted mammals across six distinct phylogenetic orders, marking the only documented case of convergent evolution at the amino acid level shared between a single breed and multiple wild animals [11]. Signatures of selection scans in the Wagyu breed, famous for its top-quality marbled meat, pointed to several candidate genes for marbling (DGAT1, IQGAP2, RSRC1, DIP2B [12], ZPLD1, KLF2, GPCPD1, SLC25A10, GAA, KLF13, and LRP2BP [13]). It should be noted, however, that methods for detecting selection signatures may demonstrate reduced statistical power when applied to admixed populations [14,15]. An alternative approach widely employed for analysing the genomes of admixed populations is local ancestry inference (LAI). This type of analysis facilitates the genome-wide identification of the genetic origins of chromosomal segments in admixed individuals. At the population level, genomic regions exhibiting deviating ancestry proportions may indicate selection [16].

The roots of the LAI analysis can be traced back to early attempts to ascertain the ancestry of hybrid populations, which initially relied solely on morphology. For instance, in a 1937 paper, Bisschop analysed various morphological traits of African Sanga cattle (hump, thighs, hindquarters, rump, skeleton, etc.) and argued that the Sanga originated from cross-breeding between long-horned Zebu and humpless Egyptian long-horned cattle [17]. Later, genetic markers such as blood groups and protein polymorphisms allowed for a more detailed study of the ancestry of populations. Stormont and colleagues (1986), for example, analysed blood groups (B system phenogroups) in Beefalo cattle, a hybrid of bison (Bison bison) and domestic cattle, concluding that Beefalo was primarily derived from Charolais, Angus, and Hereford breeds. Simultaneously, gel electrophoresis of blood proteins (haemoglobin, transferrin, and carbonic anhydrase) revealed very few bison-specific markers in the Beefalo population. This contradicts the widely accepted assertion that Beefalo cattle have three-eighths bison ancestry [18]. Interestingly, a recent genomic study confirmed the latter result, revealing that most Beefalo animals exhibit no detectable bison ancestry. Their phenotypic resemblance to bison is likely due to an indicine component [19].

The introduction of new genetic markers, including microsatellites, restriction fragment length polymorphisms, and single-nucleotide polymorphisms (SNPs), along with significant advancements in high-throughput SNP genotyping technologies, has facilitated the development of software tools capable of estimating the ancestry proportions for specific individuals. Among these tools, STRUCTURE and ADMIXTURE have likely emerged as the most widely utilised [20]. Their invention provided insights into the genetic histories of various human, animal, and plant populations. In cattle studies, their utilisation helped clarify the origin and genetic composition of European, African, Russian, Iranian, Chinese, Indian, and numerous other native breeds [6,21,22,23,24,25,26]. The aforementioned tools were designed to estimate global ancestry. Meanwhile, the development of tools for local ancestry inference has lagged slightly behind, as this type of analysis heavily depends on high SNP density [27]. Consequently, the increase in density of SNP arrays in the late 2000s and the growing availability of whole-genome sequencing in the early 2010s accelerated the creation of several software tools for LAI analysis, with RFMix [28] probably becoming the most popular.

In studies of hybrid cattle populations, the LAI analysis allowed for identifying genomic regions where ancestry fractions significantly deviated from their genome-wide averages, presumably due to artificial selection or genetic adaptation (e.g., [29,30,31,32]). The most significant finding in cattle discovered through the LAI approach is the evidence for mitonuclear coadaptation in African hybrid indicine × taurine populations. Despite extensive admixture, all African hybrid cattle populations possess a taurine mitochondrial genome. By employing LAI analysis, Ward and colleagues (2022) demonstrated that the nuclear genes encoding mitochondrial proteins have a notably higher proportion of taurine ancestry than the rest of the genome in these populations. The authors concluded that this phenomenon supports the hypothesis of incompatibility between the indicine nuclear genome and the taurine mitochondrial genome [33]. Overall, the LAI approach is applicable in various research fields, ranging from genome-wide association studies to investigations of epistatic interactions and gene regulation [34,35,36].

It can be concluded that contemporary bioinformatics methods now facilitate uncovering genomic changes in admixed populations resulting from cross-breeding and subsequent selection. With this in mind, we conducted local ancestry inference and scans for signatures of selection in the Russian Black Pied cattle using whole-genome resequencing data.

2. Materials and Methods

2.1. Data Preparation

The key steps of our analytical workflow are illustrated in Figure 1. Twenty-four Russian Black Pied (RBP) cattle individuals were genotyped on the GGP HD150K array (~139,000 SNPs) at Neogen Europe Ltd. (Ayr, UK) in our earlier work [6]. Among them, eleven animals were excluded from the analysis as they had second-degree relatives in the sample, as indicated by PI_HAT statistics (>0.25) produced by PLINK v.1.9 software (using the “--genome” option) [37]. This step aims to ensure the representativeness of the samples for further resequencing, which is compromised in the presence of close relatives (as the same chromosomal segments are sequenced multiple times). The remaining thirteen animals were resequenced using Hiseq4000 technology at Novogene Co., Ltd. (Hong Kong, China) for approximately 50 Gbp each. The raw resequencing data were processed following the “Fastq to GVCF guidelines” of the 1000 Bull Genomes Project (1KBGP) [38].

To discover local ancestry and signatures of selection, we examined breeds genetically closest to the RBP. We utilised the 1KBGP set of genomes, which contains whole-genome resequencing data from over 5000 animals of taurine cattle breeds worldwide [39]. The 1KBGP and our RBP data were combined using the following procedure: The gVCF files of RBP individuals and gVCFs of over 500 additional animals from 29 breeds mapped against the same reference genome (ARS-UCD1.2) were merged using the GenotypeGVCFs module of the GATK v.3.8-1-0-gf15c1c3ef package [40]. These samples from diverse taurine breeds were used to reduce potential ascertainment bias when identifying polymorphic sites, which could influence our selection of related breeds [41]. In the resulting VCF file, only SNPs were retained (GATK options “-T SelectVariants -selectType SNP”), which were subsequently filtered by the VariantFiltration GATK module using the hard-filtering parameters “QD < 2.0||FS > 60.000|| MQ < 40.00||MQRankSum < −12.5||ReadPosRankSum < −8.0” adopted from GATK Best Practices [42]. Another commonly used filtering option for GATK is the variant quality score recalibration (VQSR) method, which relies on a machine learning algorithm [43]. Simulations show that VQSR outperforms hard filtering when retaining true variants but has a similar or even higher fraction of false positives [44]. The VQSR requires well-curated databases of known genetic variants currently lacking for cattle. The multiallelic SNPs were split into biallelic variants using BCFtools version 1.11 (“norm -m -any” option) [45]. The dataset was LD-pruned using PLINK v.1.9 software with the “--indep-pairwise 100 kb 100 0.2” option. The thirteen RBP animals were extracted using the “--keep” option, and the resulting file, containing approximately 62,000 SNPs, was merged with the 1KBGP set using the “--merge-list --keep-allele-order” option. This combined dataset was employed to analyse population structure using ADMIXTURE software v.1.3.0, with a cluster range (K) of 2–12 [46] and measuring population differentiation (F_ST [47]) between RBP and 1KBGP breeds using the EIGENSTRAT package [48]. Based on the results of these analyses (Figure S1, Table S1), the five breeds closest to the RBP were selected: Holstein–Friesian (HOL), German Black Pied (Deutsche Schwarzbunte Niederungsrind, DSN), Meuse–Rhine–Yssel (MRY), Kholmogory (KHO), and Yaroslavl (YAR).

Resequencing data for the breeds HOL, DSN, MRY, KHO, YAR, and YKT (in which Yakut cattle were used as a genetically distant outgroup in the hapFLK analysis) were downloaded from the Sequence Read Archive database [49], with a focus on samples with the highest genome coverage. As a quality control step, we processed the raw resequencing data to generate autosomal VCF files containing high-quality biallelic SNPs, as previously described. Following LD pruning, ADMIXTURE analysis was conducted to exclude putatively admixed or misidentified individuals. As a result, for each of the breeds mentioned above, we selected the twenty most purebred samples with the highest genome coverage (see Table S2 for IDs). One admixed animal was excluded from subsequent analysis in the Russian Black Pied breed. The chosen gVCF files of the RBP (n = 12) and six additional breeds (n = 20 × 6 = 120) were merged using the GenotypeGVCFs module of GATK. The resulting VCF file comprised 31,737,129 polymorphic sites (SNPs and indels). After removing indels, applying a hard-filtering procedure, and splitting multiallelic variants, 25,844,010 SNPs were kept.

2.2. Scans for Selection Signatures

We conducted scans for signatures of selection in the genomes of the RBP using hapFLK and F_ST methods. For the hapFLK analysis, the dataset was filtered for minor allele frequency and call rate in PLINK v.1.9 (--maf 0.1 --geno 0.1). The number of haplotype clusters (K) was estimated using fastPhase software v.1.2 [50]. HapFLK v.1.4 software [51] was run with the following parameters: “--K 30 --nfit=30 --kfrq”. This analysis was conducted for the set involving the RBP and closely related breeds, with YKT serving as the outgroup. P-values were calculated using the “rlm” function from the “MASS” R package [52], assuming a normal distribution for the hapFLK statistic values. The correction for multiple testing was performed using the “qvalue” R package [53]. In our analysis, we considered the putatively selected regions as those containing at least one SNP with a q-value < 0.01. The upstream and downstream boundaries were defined by the first SNP with a q-value > 0.2. To identify breeds under selection, haplotype diversity plots and local trees were visualised using scripts provided with hapFLK (hapflk-clusterplot.R, local_reynolds.py, and local_trees.R).

The F_ST between the RBP and the combined set of five closely related breeds was calculated for sliding windows and individual SNPs using VCFtools software v.0.1.13 [54]. The window-based analysis was performed using the options “--fst-window-size 50000 --fst-window-step 25000 --max-missing 0.9”. The single-point analysis was carried out without the options mentioned above. It was restricted to SNPs with an overall call rate of at least 90% and at least 80%, specifically in the RBP. Windows containing fewer than 100 SNPs were excluded, and the top 0.1% of weighted F_ST windows were chosen as candidates. For individual SNPs, those with F_ST > 0.4 and annotated by the SNPeff tool [55] as missense and nonsense substitutions, start-and-stop codon losses, and splice acceptor and donor variants were considered candidates. Where needed, the linkage disequilibrium between variants was estimated with PLINK v.1.9 (“--ld” option).

Duplicated regions in a genome may result in the detection of false-positive SNP sites and create an apparent excess of heterozygotes [56,57]. Therefore, to account for the potential influence of duplications on selection scans, we also searched for copy number variants (CNVs) in the genomes of the RBP using the “cn.MOPS” R package [58] in the windows of 700 SNPs. The identified CNVs were merged into CNV regions (CNVRs) with the “bedmap” function of the BEDOPS toolkit v.2.4.41 [58] using at least 50% reciprocal overlap in at least two animals. The putatively selected regions, with over 10% of their length covered by duplications, were excluded from the analysis. Additionally, we excluded candidate SNPs identified through single-point F_ST analysis that were found within duplications.

2.3. Local Ancestry Inference (LAI)

RFMix [28] has probably been the most used local ancestry inference (LAI) analysis method for many years. The Gnomix algorithm was recently developed and claims to outperform RFMix in accuracy [59]. We performed population genetics simulations with AdmixSim software v.2 [60] to choose between these algorithms. In brief, we simulated a hybrid population (n = 1000) with ancestry fractions of 80%, 10%, and 10% based on three different breeds from the 1KBGP, allowing it to “evolve” over twenty generations. For one of the two breeds with a minor contribution, we additionally simulated admixture (50%) with various 1KBGP breeds to emulate an impure reference. Next, we ran RFMix and Gnomix on the hybrid population using one admixed and two pure reference populations. RFMix was executed with the “--reanalyze-reference” option, which is suitable for admixed reference (https://github.com/slowkoni/rfmix/blob/master/MANUAL.md (accessed on 27 May 2024)). Gnomix was employed with the default parameters. We calculated the correlations between the simulated and detected ancestry fractions in the genomes of the hybrid population for both methods. As a result, Gnomix outperformed RFMix in revealing components with admixed (r = 0.66 vs. 0.10) and pure (0.87 vs. 0.24, 0.84 vs. 0.31) references. Following this, we explored different options in Gnomix to maximise the correlation. The best result was obtained with the “calibrate: True” option, surpassing the default parameters (for admixed reference: 0.76 vs. 0.66; for pure references: 0.90 vs. 0.87, 0.89 vs. 0.84).

The history of the RBP breed indicates that its genome is expected to include three major components inherited from Russian local cattle, Dutch local breeds, and Holsteins. Therefore, we employed three reference panels for analysis. The panel for the Russian native component (n = 28) comprised the most purebred animals from YAR (n = 13) and KHO (n = 15). The DSN (n = 20) was the reference panel for Dutch native cattle ancestry. Holstein samples (n = 20) constituted the panel for Holstein ancestry. It is worth noting that Holsteins are related to the Dutch local breed known as Dutch Friesian (Dutch Black Pied) in the Netherlands or German Black Pied (DSN) in Germany [25,61]. However, after many generations of intensive selection, the Holstein breed has significantly diverged from its ancestral population [25], as our analysis shows (Figure S1). We also conducted principal component analysis (PCA) using PLINK v.1.9 software (“--pca” option) on an LD-filtered dataset (“--indep-pairwise 50_5_0.4” option resulting in approximately 824K SNPs) comprising close breeds (utilised for hapFLK analysis). Our PCA results confirm that DSN and Holstein individuals form two distinct, non-overlapping clusters (Figure S2). In light of this, we refer to the component in the RBP genome inherited directly from Dutch local breeds as “Dutch native” to differentiate it from the modern Holstein component. The VCF file was initially phased with Beagle v.5.4 [62] using a likelihood-based version of the genetic map from Qanbari and Wittenburg [63]. The Gnomix analysis was conducted using the “calibrate: True” option and the aforementioned genetic map.

2.4. Functional Annotation of Genome Regions and Gene Sets

We performed functional enrichment analyses and functional annotation clustering using the DAVID tool [64]. The annotation of selected genome regions with genes was conducted using the ARS-UCD1.2 Ensembl annotation file (https://ftp.ensembl.org/pub/release-110/gtf/bos_taurus/Bos_taurus.ARS-UCD1.2.110.gtf.gz (accessed on 5 September 2024)). As a result, all genes from this file were utilised as the background for the DAVID analyses. Functional enrichment analyses were carried out for three gene sets: (1) genes with >50% intersection with regions whose ancestries deviate from the genome-wide average; (2) genes with >50% intersection with putatively selected regions identified by window-based F_ST and/or hapFLK analyses; and (3) genes containing candidate variants identified by single-point F_ST analysis. We used q-value (FDR) = 0.05 in the functional enrichment analysis as the significance threshold. An enrichment score (ES) of >1.3 was applied as the cut-off threshold in our clustering analysis.

We searched for overlaps with known quantitative trait loci (QTLs) from CattleQTLdb [65] for the regions identified in the selection scans. To simplify the count of overlaps, if two or more QTLs for the same trait fell into a particular region, they were counted as a single QTL.

3. Results

3.1. Scans for Selection Signatures

A comprehensive list of regions identified in the RBP by hapFLK and window-based F_ST is presented in Table S3. The hapFLK analysis uncovered 20 regions (Figure 2A) that are putatively selected in the RBP breed, of which two overlapped with duplications and were thus excluded from further analysis. Of the remaining 18 regions, 9 are organised into clusters of 2 or 3 closely located regions on BTA2, BTA13, BTA14, and BTA15. Six regions exhibited selection signatures shared with another breed: three with MRY, two with HOL, and one with YAR (Table S3). Following the window-based F_ST analysis (Figure 2A), the top 99 intervals were combined into 53 non-overlapping regions, of which two were excluded due to excessive overlap with duplications. Four regions contained the intersection of the hapFLK and window-based F_ST results (Table S3). The single-point F_ST analysis (Figure 2A) identified 96 missense and 3 nonsense variants with F_ST > 0.4 (Table S4). One missense variant was discovered in a duplicated region and was subsequently excluded. Twelve missense mutations from six genes were identified in six putatively selected intervals, two overlapping between the hapFLK and window-based F_ST analyses (

Table 1. Candidate missense variants from regions putatively selected in the Russian Black Pied (RBP) breed, their alternative allele frequencies in the RBP and closely related breeds, and the 1000 Bull Genomes Project (1KBGP).

Region (BTA)	Method	Position (BTA)	Ref/Alt Allele	Gene	Amino Acid Change	FST	Alternative Allele Frequency
							RBP	Close Breeds	1KBGP
2:131125001-131200000	FST	2:131164309	C/T	RAP1GAP	Ala704Val	0.445	0.417	0.06	0.059
		2:131164365	C/T	RAP1GAP	Pro723Ser	0.445	0.417	0.06	0.054
		2:131165149	T/C	RAP1GAP	Leu751Pro	0.445	0.417	0.06	0.067
		2:131165157	C/G	RAP1GAP	Pro754Ala	0.445	0.417	0.06	0.066
		2:131164249	T/C	RAP1GAP	Leu684Pro	0.426	0.417	0.065	0.136
		2:131165166	G/C	RAP1GAP	Asp757His	0.425	0.417	0.065	0.065
2:133015733-133049432	hapFLK	2:133035362	C/A	HTR6	Gly6Val	0.426	0.542	0.113	0.145
13:65900001-66125000	FST	13:65971107	G/A	SAMHD1	His89Tyr	0.436	0.292	0.025	0.047
13:67075001-67125000	FST, hapFLK	13:67108721	T/C	KIAA1755	Lys432Glu	0.548	0.792	0.18	0.209
14:75434681-75456117	FST, hapFLK	14:75438515	T/G	CNBD1	Asn275His	0.466	0.708	0.18	0.297
15:28900001-28975000	FST	15:28919395	G/A	MPZL3	Leu218Phe	0.561	0.375	0.025	0.064
		15:28940819	C/A	MPZL3	Ala13Ser	0.561	0.375	0.025	0.063

and Table S5).

The missense variants identified at the intersection of window-based F_ST and hapFLK methods are located in the KIAA1755 and CNBD1 genes. The KIAA1755 gene encodes an uncharacterised protein (https://www.genecards.org/cgi-bin/carddisp.pl?gene=KIAA1755 (accessed on 14 October 2024)). According to a genome-wide association study (GWAS) in Holstein cows, an SNP near the KIAA1755 gene is associated with a somatic cell score, an essential indicator of mastitis resistance [66]. Moreover, a transcriptome study revealed that KIAA1755 significantly upregulates its expression in the mammary glands during pregnancy in pigs and, consequently, may play a role in developing the swine mammary gland [67]. The missense variant from the KIAA1755 gene (BTA13:67108721, Lys432Glu, F_ST = 0.548) has an alternative allele frequency of 0.79 in the RBP, with all the animals being carriers. In five closely related breeds, this frequency does not exceed 0.33.

The CNBD1 gene encodes cyclic nucleotide-binding domain-containing protein 1, which has a poorly characterised function (https://www.genecards.org/cgi-bin/carddisp.pl?gene=CNBD1 (accessed on 14 October 2024)). There is evidence that the function of this gene in cattle is linked to pathogen resistance. Thus, according to a GWAS in German Black Pied cattle, CNBD1 is associated with faecal egg counts of gastrointestinal nematodes [68]. In another GWAS involving Holsteins, an SNP close to the CNBD1 gene was associated with susceptibility to Mycobacterium avium subsp. paratuberculosis infection [69]. In the RBP, the missense variant of the CNBD1 gene (BTA14:75438515, Asn275His, F_ST = 0.466) has an alternative allele frequency of 0.71 and a carrier rate of 0.92. In the other five breeds, its frequency does not exceed 0.33.

Other mutations are found in genes identified by window-based F_ST or hapFLK methods. Among these, the most prominent are two missense variants in the MPZL3 gene (BTA15:28919395, Leu218Phe, F_ST = 0.561; BTA15:28940819, Ala13Ser, F_ST = 0.561), located in the region with the highest weighted F_ST among those containing genes. The MPZL3 gene encodes myelin protein zero-like protein 3 (https://www.genecards.org/cgi-bin/carddisp.pl?gene=MPZL3 (accessed on 14 October 2024)), which is localised in the mitochondria. MPZL3 plays a role in various biological processes, including regulating the hair follicle cycle. According to Cao and colleagues (2007), a missense mutation in MPZL3 led to severe skin and hair abnormalities in mice [70]. Another study shows that the MPZL3 knockout mice developed sebaceous hypertrophy and hair loss [71]. The results relevant to cattle come from a study of the Tianzhu white yak (Bos grunniens), adapted to cold climates. The authors discovered that the MPZL3 gene regulates the hair follicle cycle in this species by performing a scan for RNA editing sites [72]. The two missense mutations in the MPZL3 gene identified in our study are in strong linkage disequilibrium (r² = 1, D’ = 1) and exhibit the same frequencies in the RBP and closely related breeds. In the RBP, the alternative allele frequency for these variants is 0.38, with a carrier rate of 0.67. In other closely related breeds, the alternative alleles are either absent or present at low frequencies (0.03-0.05).

3.2. Local Ancestry Inference (LAI)

According to the LAI analysis (Figure 3, Table S6), the genome-wide average fractions of Russian native, Dutch native, and Holstein ancestries in the RBP genome were estimated to be 10.13%, 11.84%, and 78.03%, respectively. The ancestry fractions in certain regions largely deviated from these values. Thus, in the genomes of the RBP, 22.30 Mb (0.9% of the autosomal genome) and 9.65 Mb (0.39%) exhibited ≥50% of Russian and Dutch native ancestries, respectively. When combining these components (i.e., non-Holstein ancestry), 114.92 Mb (4.62%) contained ≥50%. Furthermore, 2.31 Mb (0.09%) had >90% non-Holstein ancestry. None of the regions were inherited 100% from Russian or Dutch native cattle. Considering local estimates for Holstein ancestry, 506.56 Mb (20.35%) had a fraction of >90%. Among this segment, 76.62 Mb (3.08%) were of 100% Holstein origin.

Some regions in the RBP genome displayed a high proportion of segments from Russian native cattle, either alone or when combined with segments from Dutch native cattle. The only region exhibiting over 90% Russian native ancestry was identified on chromosome 10, covering over two megabases (23.18–25.25 Mb) (Figure 3). This interval broadly corresponds to the bovine TRA/TRD locus (22.2 Mb–25.9 Mb) [73]. In cattle, the TRA/TRD locus contains multiple T cell receptor alpha and delta genes playing a role in adaptive immunity [73,74]. Another region of interest is located on chromosome 23 and spans ~125 Kb (25.69–25.82 Mb) (Figure 3). More than 80% of haplotypes in this genomic interval match Russian or Dutch native cattle segments. This region contains the BOLA-DRB3 gene coding for bovine major histocompatibility complex class II. The BOLA-DRB3 polymorphisms are a known genetic factor for BLV infectivity, proviral load [75], and resistance/susceptibility to BLV-induced leucosis in cattle [76,77,78].

Based on the LAI results, we calculated ancestry fractions for putatively selected regions (Table S7). Of the 65 putatively selected areas, 16 exhibited over 90% Holstein ancestry. Four genes (H4C14, OR2D3G, HECA, and TXLNB) were identified within these intervals. One region detected by hapFLK (BTA1:47764970-47814477) displayed 100% Holstein ancestry with no annotated genes. None of the putatively selected regions had more than 50% Russian native or Dutch native origins, even when these components were combined.

3.3. Functional Annotation of Gene Sets

The functional enrichment in regions exhibiting ≥50% ancestry from Russian native cattle resulted in 117 statistically significant terms. The most prevalent among these were associated with keratin (“Keratin_II”, “keratinisation”, “keratin filament”, etc.), nucleases (“Nuclease”, “RNAse_Pc”, “RNA nuclease activity”, etc.), amino acid transporters (“AA_transporter”, “L-ornithine transmembrane transporter activity”, “AA/rel_permease1”, etc.), antimicrobial defence (“defence response to Gram-positive bacterium”, “Beta/alpha-defensin_C”, “Antibiotic”, etc.), the immune system (“MHC II”, “inhibitory MHC class I receptor activity”, “Intestinal immune network for IgA production”, etc.), protein digestion (“Aspartic_peptidase_A1”, “PEPTIDASE_A1”, “Protein digestion and absorption”, etc.), and sensory transduction (“olfactory receptor activity”, “sensory perception of smell”, “Sensory transduction”, etc.) (Table S8). The functional annotation clustering for the same regions produced 14 statistically significant clusters. The top cluster (ES = 20.6) included terms related to keratin and filaments. The second, third, and fourth top clusters primarily pertained to amino acid transporters (ES = 10.2), nucleases (ES = 8.6), and antimicrobial defence (ES = 6.9), respectively. Among the remaining clusters were those linked to the immune system (ES = 3.7 and 3.1), sensory transduction (ES = 3.1), and protein digestion (ES = 3) (Table S9). The functional enrichment analysis for regions showing ≥50% Dutch native ancestry resulted in 25 statistically significant terms. Most were linked to CRAL-TRIO and GOLD domains (“CRAL_TRIO_N”, “CRAL-TRIO_dom_sf”, “GOLD_dom”, “DOMAIN:GOLD”, etc.) and metabolism (“Metabolism of xenobiotics by cytochrome P450”, “Glutathione metabolism”, “Drug metabolism-other enzymes”, etc.) (Table S8). The functional annotation clustering for these regions resulted in two significant clusters also comprising terms related to CRAL-TRIO and GOLD domains (ES = 6.2) and metabolism (ES = 2.3) (Table S9).

The functional enrichment analysis for regions exhibiting >90% Holstein ancestry produced one significant term (“S_100”) (Table S8). The functional annotation clustering yielded eight clusters (Table S9). The top cluster primarily included terms associated with the small calcium-binding protein S100 (ES = 2.3). The second and third top clusters were related to chitinase activity (ES = 2) and motor proteins (ES = 1.7). The functional annotation for regions with 100% Holstein ancestry resulted in 14 significant terms related to the homeobox domain (“Homeobox_metazoa”, “DNA_BIND:Homeobox”, “Homeobox_CS”, etc.) and neurotransmission (“Neurotrans-gated_channel_TM”, “Neurotransmitter_ion_chnl_CS”, “Neur_channel”, etc.) (Table S8). The functional annotation clustering yielded two significant clusters: one comprised terms associated with the homeobox domain and RNA polymerase (ES = 2.7), and the other included terms related to neurotransmission (ES = 2.1) (Table S9). In addition, we conducted functional annotation for regions with <10% of the Holstein component (i.e., >90% of Russian native and Dutch native ancestries combined). The functional enrichment analysis resulted in twelve significant terms, related to the transmembrane transport of amino acids (“DOMAIN:Cationic amino acid transporter C-terminal”, “L-ornithine transmembrane transport”, “amino acid transmembrane transporter activity”, etc.) (Table S8). The functional annotation clustering produced no clusters with ES > 1.3.

The functional annotation analyses for window-based selection scan regions did not reveal enriched terms or clusters. The functional enrichment analysis based on genes identified in the single-point analysis yielded no significant terms. However, the functional annotation clustering based on this gene set revealed three significant clusters related to the cyclic nucleotide-binding domain (ES = 2), peptidases (ES = 1.7), and leucine-rich repeats (ES = 1.4) (Table S10).

The search for intersections with QTLs identified 36 out of 65 putatively selected consensus regions containing at least one QTL (Table S11). In CattleQTLdb, traits are categorised into higher-order groups. The most common were QTLs that fell into the categories of “Milk Association” (47 QTLs), “Reproduction Association” (14 QTLs), “Production Association” (9 QTLs), and “Meat and Carcass Association” (7 QTLs). Among the traits, the most prevalent were QTLs linked to “Milk protein yield” (6 QTLs), “Milk fat yield” (5 QTLs), “Conception rate” (4 QTLs), “Body weight” (4 QTLs), “Milk fat percentage” (4 QTLs), “Milk kappa-casein percentage” (4 QTLs), “Milk caproic acid content” (4 QTLs), and “Milk caprylic acid content” (4 QTLs).

4. Discussion

This study focused on determining the local ancestry inference in Russian Black Pied cattle genomes. We also conducted a scan for genome signatures of selection in this breed using whole-genome resequencing data. The primary conclusion from our study is that despite strong Holsteinisation, the genomes of the Russian Black Pied breed still retain high-frequency haplotypes likely inherited from Russian (and Dutch) native cattle. Genes in these regions are involved in the immune system and resistance to pathogens. Furthermore, we identified promising candidate genetic variants in putatively selected genomic regions that could be utilised in future studies examining the association between genotype, economically important traits, and adaptive traits in cattle, with potential applications in marker-assisted selection and gene editing.

Similar to our study, the research on composite Xinjiang Brown cattle, developed through crossing the commercial Swiss Brown dairy breed with local Kazakh cattle, demonstrates an excess of Kazakh haplotypes enriched in pathways related to bacterial and protozoan infections [29]. This suggests that links between native haplotypes and immune response are not exclusive to the Black Pied breed. Our observation that the immune response BOLA-DRB3 gene from the BoLA region originates from Russian or Dutch local cattle haplotypes aligns with a study of a heat-resistant composite Brangus (Angus/Brahman) breed, where the BoLA region was shown to be primarily inherited from tropical Brahman cattle [30]. One of the main reasons for creating composite breeds is to merge commercial breeds’ productivity with local cattle’s adaptive traits. The observed phenomenon confirms that this approach is effective at the genome-wide level.

Furthermore, our BOLA-DRB3 finding aligns well with comparative genetic data. In a study by Lazebnaya (2022), the overall frequency of BOLA-DRB3 alleles conferring resistance to bovine leucosis was found to be significantly higher in the RBP (0.29) compared to Holsteins (0.18), as summarised across several populations [9]. According to veterinary regulations in Russia, all animals displaying pathomorphological or haematological signs of bovine leucosis must be culled [79]. Therefore, it can be assumed that the BOLA-DRB3 locus in the RBP is experiencing purifying selection against alleles that predispose to leucosis, which have been introduced by the Holstein breed, in which they are more prevalent. The same could be speculated about the TRA/TRD locus for >90% originating from Russian native cattle and involved in adaptive immunity.

Apart from the immunity and pathogen resistance genes, the regions with an excess of Russian native haplotypes in our study demonstrate functional enrichments likely related to the climate adaptations of the Black Pied breed. The highest enrichment score was observed for the keratin filament category, with keratins being the primary component of cattle hair, affecting coat and horn characteristics and playing an essential role in adaptation to different climates [80]. This is supported by studies from Picard and colleagues (1999), who showed that the keratin sheath of horns in bovid species from lower temperate zones has lower thermal conductance than that of horns from the tropics, thus restricting heat loss in colder environments [81]. Other genomic studies imply the importance of keratin filament functional categories in the cold adaptation of cattle. This is supported by the study of CNVRs in Yakut and Kholmogory cattle breeds, both known for their cold adaptation, indicating that CNVRs shared between these two breeds involve genes enriched in terms related to keratin filaments [82]. A genome scan in Ladakhi cattle reared at high altitudes and exposed to extreme cold has revealed signatures of selection in genes coding for keratin proteins. Cattle adaptation to low temperatures relies on multiple features, including the insulative properties of skin and coat, thermoregulatory (shivering and non-shivering thermogenesis, piloerection, vasoconstriction, regulation of respiratory rate, etc.) and behavioural (huddling and changes in body posture) responses to cold [83]. Therefore, the enrichment of Russian native cattle haplotypes for other terms or clusters, such as those related to metabolism (amino acid transporters and protein digestion) or sensory transduction, may also be linked to adaptations to the harsh climate in the Russian Black Pied. On the other hand, the abovementioned metabolic terms/clusters might reflect the selection for productive traits Russian native cattle populations have experienced in the past. For example, amino acid transporters are necessary for cattle, other ruminants, and non-ruminant species to uptake amino acids from the gastrointestinal tract into the blood [84]. Similarly, protein digestibility in the bovine gastrointestinal tract is an economically important trait that differs significantly for various feedstuffs [85].

The regions in the RBP genome with the highest fractions of Holstein ancestry are enriched for terms and clusters that may be linked to economically important traits. Thus, the regions with >90% of Holstein haplotypes demonstrate enrichments related to the S100 proteins. These calcium-binding molecules are involved in various physiological processes [86]. In cattle, the S100 proteins are among the main components of the teat canal lining and are presumably involved in the resistance of cows to mastitis [87]. The regions with 100% Holstein ancestry are enriched for terms related to neurotransmission (synaptic transmission), which may be explained by selection for behavioural traits [88].

In contrast to the ancestral inference results showing that native haplotypes are likely under purifying selection, the genes detected in selection scans (mainly positive selection) did not show significant functional enrichments except those identified in single-point analysis. This could be due to low statistical power because of a relatively small sample size (n = 12), which is a limitation of our study. On the other hand, the search in CattleQTLdb revealed multiple overlaps of putatively selected regions with bovine QTLs, which are most commonly associated with milk traits. This finding is unsurprising, as the RBP is a dairy breed, with selection efforts primarily focused on improving milk productivity (milk yield, fat yield, fat percentage, etc. [89]).

There are also substantial findings related to individual genes. For example, GRIK2 is encoding for a glutamate receptor involved in cold sensing and, therefore, can participate in adaptation to low temperatures [90]. The SLC6A19 gene encodes a neutral amino acid transporter whose inhibition reduces protein absorption and increases urinary loss of amino acids [91,92]. Selection for changes in these genes could influence the RBP’s productive traits.

Our limited sample size and the differences in the selection scan approach could account for the discrepancies with the previous study on the RBP [93]. We found an overlap for two genes (LRRIQ3 and BNIP1), with an earlier study suggesting that these genes have strong signals detectable on smaller and larger datasets and by different methods. The difference in gene sets could also be attributed to distinct genotyping technologies (sequencing and DNA arrays) and the genetic makeup of the studied populations. The genome resequencing approach is not biased toward preselected SNP sets, while the genotyping arrays are.

Our study centred on whole-genome resequenced datasets, facilitating the search for candidate causative variants in the putatively selected regions. One such mutation we identified is the missense mutation in the MPZL3 gene. This gene is associated with the thermal insulation properties of hair coats and thermoregulatory phenotypes. Unlike earlier Russian Yakut and Yaroslavl cattle studies [11,94], we did not uncover breed-specific variants under selection in the RBP cattle. This outcome is expected for composite breeds with a relatively lower proportion of unique genetics than breeds predominantly formed locally. Indeed, our recent scan for signatures of selection in the local Buryat cattle, which have not undergone cross-breeding with other (including commercial) breeds, has revealed numerous unique variants absent from the 1KBGP [12]. However, we identified some alleles (e.g., in the BNIP1) that can be regarded as nearly breed-specific in the RBP due to their low frequency in other population breeds.

All putatively selected alleles in our study also segregate in other breeds, suggesting the parental population origins of these variants. Therefore, our data are in agreement with observations from multiple species, indicating that selection typically acts on standing variations (which endure in a population across many generations) rather than on “de novo” mutations [95,96,97]. Such variants may be maintained in populations by chance, or alternatively, their persistence may be prolonged due to linkage disequilibrium with sites under balancing selection [98]. According to simulation studies, phenotypic changes in response to selection initially rely almost entirely on standing variations. However, the contribution of “de novo” mutations increases over time [99]. Therefore, the absence of recent mutations among the putatively selected variants in our RBP sample might indicate that this breed has a high potential for further improvement through selective breeding.

It is also notable that variants from eleven genes (BNIP1, ADAMTS5, ENSBTAG00000018549 (SPAG16), ADAMTS1, TMEM67, NCOA6, RBM12B, OR11L1, EEF2K, KCNH6, and SOGA1) pinpointed by single-point analysis had a minor allele frequency of <1% in the 1KBGP dataset. According to commonly used classification, such variants are considered “rare”. The rare variants often have large phenotypic effects and are associated with disease [100]. In cattle, a whole-genome imputation study has shown that rare variants contribute, at least in part, to “missing heritability” [101]. The minor allele frequency of the aforementioned variants in the 1KBGP dataset ranges from 0.1% to 0.96%, whereas in the RBP, it is between 17% and 29%. Considering carrier rates, the total fraction of animals either homozygous or heterozygous for the alternative allele ranges from 33% to 55%. Given this, it may be speculated that these rare variants play an important role in shaping the RBP phenotype.

In the context of breed-specific artificial selection and adaptation, it is noteworthy that many alleles putatively selected in the RBP likely existed at lower frequencies in the parental populations but seem not to have experienced selective pressure in those populations. For example, the variant Lys432Glu of the KIAA1755 gene selected in the RBP was likely present in all three parental populations, as supported by its current frequencies in HOL, DSN, and YAR/KHO. However, haplotype diversity plots (Figure 2B) indicate that this variant did not undergo selection in these populations. The same applies to the Ala13Ser and Leu218Phe variants from the MPZL3 gene, which were putatively selected in the RBP but not in its Holstein and Russian native ancestors (Figure 2B). This phenomenon may be attributed to variations in breed management practices between the RBP and its parental populations. Alternatively, it might reflect differences in the genetic backgrounds of the breeds, which can influence the phenotypic effects of alleles and, consequently, their response to selection [102,103]. We may hypothesise that the variants revealed in KIAA1755, MPZL3, and other genes are particularly important to the productivity and adaptation of the RBP and could be utilised to enhance this breed.

The variants identified in this study should be further investigated for their associations with adaptive and/or economically important traits, with the potential use in marker-assisted selection in the Russian Black Pied breed. Furthermore, as these variants are likely enriched for functional polymorphisms, they may also have applications in genomic selection, which is known to benefit from including causative SNPs in the breeding value estimation [104].

In conclusion, our study reveals haplotypes, genes, and candidate variants subject to purifying and positive selection in the Russian Black Pied breed. Our results may be used to conserve the breed and enhance its genetics.