Two different population samples from the Irish Kerry cattle breed were used for this study (Figure 1). The first population sample consisted of peripheral blood and semen straw genomic DNA collected and purified from 36 male and female Kerry cattle in 1991/92, which are a subset of the Kerry cattle population sample (n = 40) we have previously described and used for microsatellite-based population genetics analyses (MacHugh et al., 1997, 1998). Pedigree records and owners were consulted to ensure that a representative sample of animals was obtained. This Kerry population sample group is coded as KY92.

The second Kerry cattle population sample was collected in 2011/12 from 19 different herds located across southern and western Ireland. Performagene (PG-100) nasal swab DNA collection kits were used for biological sample collection (DNA Genotek Inc., Ottawa, Canada). Nasal swab DNA samples were collected from a total of 75 male and female Kerry cattle and owners were consulted to ensure that a representative sample of animals was obtained. This Kerry population sample group is coded as KY12. Genomic DNA was purified from 0.5 ml of each PG-100 nasal swab sample using the laboratory protocol recommended by the manufacturer (DNA Genotek Inc.).

Illumina® Bovine SNP50 BeadChip (Matukumalli et al., 2009) genotyping on all 111 Kerry genomic DNA samples (KY92 and KY12 sample panels plus nine blinded sample duplicates for quality control purposes) was performed by Weatherbys Scientific (Co. Kildare, Ireland).

For comparative population genomics analyses, equivalent SNP data for a range of other breeds were obtained from previously published work (Decker et al., 2009; Flori et al., 2009; Gibbs et al., 2009; Matukumalli et al., 2009; Gautier et al., 2010; Park et al., 2015). The breed SNP data were split into two discrete composite data sets: a European breed SNP data set (EU) and a SNP data set for a subset of European breeds originating from Britain and Ireland (BI). A population sample of West African N'Dama B. taurus cattle from Guinea (NDAM) was also used as an outgroup for the phylogenetic analyses. Table 1 provides detailed biogeographical information on the cattle breed samples used for the present study.

Genomic non-exclusion—in other words, genome-wide SNP profiles completely compatible with parent-offspring relationship (with allowance for very low-level genotyping error)—were used to identify animals from the KY92 and KY12 population samples that were parent-offspring pairs. One of the two animals in each pair was then randomly removed to generate the working SNP data set. Following this procedure, quality control and filtering based on recorded SNP genotypes was performed as detailed below for the EU and BI data sets.

Prior to quality control and filtering there were 54,057 SNPs in the EU data set (608 animals, including KY92 and KY12) and in the BI data set (354 animals, including KY92 and KY12). SNP quality filtering was performed using PLINK version 1.07 (Purcell et al., 2007), such that individual SNPs with more than 10% missing data and a minor allele frequency (MAF) of ≤0.01 (1%) were removed from both data sets; however, for analyses of genomic inbreeding and runs of homozygosity (ROH) the MAF filtering threshold was not imposed. Only autosomal SNPs were retained and individual animal samples with a SNP call rate less than 90% were also removed from each of the two data sets.

SNP quality control and filtering were performed across breeds/populations (by data set) for construction of phylogenies and ancestor graphs, multivariate analysis, investigation of population structure and detection of signatures of selection. For intrapopulation analyses of effective population size (N_e) and genomic inbreeding, all SNPs genotyped (54,057) were filtered within breeds/populations as detailed above. However, an additional filtering procedure was used to remove SNPs deviating from Hardy-Weinberg equilibrium (HWE) with a P-value threshold of < 0.0001. Also, for the N_e analysis, a more stringent MAF threshold of 0.05 was used.

Using the filtered genome-wide SNP data, PLINK v1.07 was also used to generate identity-by-state (IBS) values for all pairs of Kerry cattle (KY92 and KY12), including the nine blinded sample duplicates for sample verification and tracking purposes.

Maximum likelihood (ML) phylogenetic trees with ancestry graphs were generated for the EU and BI data sets using the TreeMix (version 1.12) software package (Pickrell and Pritchard, 2012). The West African B. taurus NDAM breed sample (n = 22) was used as an outgroup. TreeMix was run without using SNP blocks (as described in the TreeMix software documentation) and ML phylogenetic trees were generated with no migration edges (m = 0) up to ten migration edges (m = 10).

To visualise the main axes of genomic variation among cattle breeds and individual animals, multivariate principal component analysis (PCA) was performed for the composite EU and BI SNP data sets using SMARTPCA from the EIGENSOFT package (version 4.2) with default settings (Patterson et al., 2006).

To further investigate genetic structure and admixture history for Kerry cattle and other breeds the fastSTRUCTURE software package (Raj et al., 2014) was used to analyse the EU and BI data sets for a range of K possible ancestral populations (K = 2–15). For the present study, the simple prior approach described by Raj et al. (2014) was used, which is sufficient for modelling population/breed divergence. To identify the “true” K-value for the number of ancestral populations, a series of fastSTRUCTURE runs with pre-defined K-values were examined using the chooseK.py script (Raj et al., 2014). Outputs from the fastSTRUCTURE analyses were visualised using the DISTRUCT software program (Rosenberg et al., 2002) using standard parameters.

Current and historical N_e trends were modelled with genome-wide SNP linkage disequilibrium (LD) data for the KY92 and KY12 populations plus a selection of BI and EU breeds using the SNeP software tool as described by Barbato et al. (2015). This method facilitates estimation of historical N_e values from SNP linkage disequilibrium (LD) data using the following equation (Corbin et al., 2012):

where N_T is the effective population size t generations ago calculated as t=(2f(ct))-1 (Hayes et al., 2003), c_t is the recombination rate defined for a specific physical distance between SNP markers, radj2 is the LD value adjusted for sample size and the recommended default α value = 1 was used to correct for the occurrence of mutation (Barbato et al., 2015). In addition, the SNeP program option for unphased SNP data was used for the analyses described here.

Individual animal genomic inbreeding was evaluated as genome-wide autozygosity estimated from SNP data using ROH and the F_ROH statistic introduced by McQuillan et al. (2008). The F_ROH statistic was calculated as the ratio of the total length of defined ROH (L_ROH) to the total length of the autosomal genome covered by SNPs:

PLINK v1.07 was used to define ROH using a sliding window approach and procedures modified from previous recommendations for Illumina® Bovine SNP50 BeadChip and similar SNP data sets (Purfield et al., 2012, 2017). The criteria for defining individual ROH were set such that the ROH was required to be at least 500 kb in length, with a minimum density of one SNP per 120 kb and that there was a gap of at least 1,000 kb between each ROH. A sliding window of 50 SNPs was incrementally advanced one SNP at a time along the genome; each discrete window could contain a maximum of one heterozygous SNP and no more than two SNPs with missing genotypes. Following Purfield et al. (2012) all filtered genomic SNPs (without a MAF threshold), including those located in centromeric regions, were used to estimate F_ROH values for individual animals.

In the absence of hard selective sweeps, single selection tests using high-density SNP data do not perform well in detecting signatures of selection from individual livestock breeds (Kemper et al., 2014). Therefore, for the present study, genomic signatures of selection were identified using the composite selection signal (CSS) method introduced by Randhawa et al. (2014). The CSS method has been shown to be a robust and sensitive approach for detecting genomic signatures of selection underlying microevolution of complex traits in livestock (Randhawa et al., 2015). The CSS is a weighted index of signatures of selection from multiple estimates; it is a nonparametric procedure that uses fractional ranks of constituent tests and does not depend on assumptions about the distributions of individual test results.

As described in detail by Randhawa et al. (2014), the CSS method can be used to combine the fixation index (F_ST), the change in selected allele frequency (ΔSAF) and the cross-population extended haplotype homozygosity (XP-EHH) tests into one composite statistic for each SNP in a population genomics data set. For the present study, we used 36,621 genome-wide SNPs genotyped in 98 individual Kerry cattle samples (from both the KY92 and KY12 populations) and a sample of 102 randomly selected cattle (six random cattle from each breed of the EU data set). To mitigate against false positives, genomic selection signatures were only considered significant if at least one SNP from the set of the top 0.1% genome-wide CSS scores was flanked by at least five SNPs from the set of the top 1% CSS scores.

The Ensembl BioMart data mining resource (Smedley et al., 2015) was used to identify genes within ± 1.0 Mb of each selection peak (Ensembl release 90, August 2017). Following this, Ingenuity® Pathway Analysis (IPA®: Qiagen, Redwood City, CA, USA; release date June 2017) was used to perform an overrepresentation enrichment analysis (ORA) with this gene set to identify canonical pathways and functional processes of biological importance. The total gene content of Ensembl release 90 version of the UMD3.1 bovine genome assembly (Zimin et al., 2009) was used as the most appropriate reference gene set for these analyses (Timmons et al., 2015).

Genomic non-exclusion identified 20 parent-offspring pairs from the KY12 population and one sample from each pair was randomly removed (sample codes: KY12_01, KY12_05, KY12_13, KY12_14, KY12_17, KY12_18, KY12_19, KY12_46, KY12_55, KY12_67). Thereafter, general SNP quality control and filtering led to additional samples being excluded (KY12_26, KY12_28 and KY12_54), giving a total filtered KY12 population sample of 62 animals for downstream population genomics analyses.

After SNP quality control and filtering across the two composite data sets (EU and BI), there were 36,621 autosomal SNPs from 605 individual animals in the EU data set and there were 37,395 autosomal SNPs from 351 animals in the BI data set. When the West African NDAM breed sample (n = 22) was included for the ML phylogenetic tree and ancestry graph analyses, the number of SNPs used was 36,000 from 627 animals for the EU data set and 37,490 from 373 animals for the BI data set. The final numbers of SNPs used for individual breed/population analyses of N_e and genomic inbreeding after all quality control and filtering (including additional filtering for deviations from HWE) are shown in Table 2.

All data sets, including EU and BI composite data sets and individual breed/population data sets had total SNP call rates of > 99%. The IBS values estimated for Kerry cattle (KY92 and KY12) from filtered genome-wide SNP data are reported in Supplementary Table 1 and described further in section Genomic Relationship and Analysis of Inbreeding.

Table 2 provides genome-wide H_o values for each of the breeds/populations used for the present study. The lowest genome-wide H_o value was observed for the West Africa NDAM B. taurus breed, which is likely a consequence of ascertainment bias introduced by a focus on polymorphic SNPs in European B. taurus during design of the Illumina® Bovine SNP50 BeadChip (Matukumalli et al., 2009).

Generally, as shown in Table 2 for the EU and BI breeds and populations, local landrace or heritage breeds display lower H_o values compared to more widespread production breeds such as the Simmental (SIMM), Holstein (HOLS) or Charolais (CHAR) breeds. In addition, as might be expected, production breeds originally derived from minor island populations (Jersey [JRSY] and Guernsey [GNSY]) also exhibit relatively low H_o values. In the context of genetic conservation it is therefore encouraging that the KY92 and KY12 population samples display intermediate H_o values that are at the upper end of the range observed for the heritage breeds.

To examine microevolutionary patterns of genetic differentiation and gene flow among cattle breeds and populations, ML phylogenetic ancestry graphs were generated using TreeMix. For the EU data set, the ML tree topology was consistent for all values of m, with the exception of m = 2 migration edges, where the Hereford breed (HRFD) was observed to group with the HOLS breed. The ML tree generated with m = 5 is shown in Figure 2, which highlights the genetic similarity of the Northern European breeds (British, Irish and Scandinavian). As expected the two Kerry population samples (KY92 and KY12) are genetically very similar and emerge on the same branch as the HRFD breed. It is also noteworthy that there is a high-weight migration edge between the British Shorthorn breed (BSHN) and the root of the two Kerry population samples, supporting the hypothesis of historical gene flow from the British Shorthorn breed into the ancestral population of modern Kerry cattle (Curran, 1990).

For the ML trees generated using the BI data set, breed/population differentiation was less apparent, possibly due to similar biogeographical origins for these breeds and/or smaller sample sizes for some of the populations sampled. Figure 3 shows the ML tree generated with m = 5 for the BI data set. For m = 5, all migration edges stem from the BSHN/Lincoln Red (LNCR) branch, including a medium-weight migration edge connecting to the Kerry cattle branch. These results support the hypothesis that there was significant gene flow during the eighteenth and nineteenth centuries from British Shorthorn cattle into the ancestral populations for a range of modern British and Irish cattle breeds (Grobet et al., 1998; Felius et al., 2011, 2015).

To investigate inter- and intra-population genomic diversity and genetic relationship among individual animals from multiple cattle breeds and populations, PCA was performed using genome-wide SNP data. Principal component plots of the first (PC1) and second (PC2) principal components are shown in Figures 4, 5 for the EU and BI data sets, respectively.

In Figure 4, for the EU data set, PC1 and PC2 account for 18.2% and 16.8% of the total variation for PC1–10, respectively. The PC1 plot axis differentiates the British Angus (ANGU), Red Angus (RANG) and BSHN and Irish KY92 and KY12 populations from the rest of the European breeds, including the British HRFD and GNSY and JRSY Channel Islands breeds. In addition, the ANGU and RANG and the Kerry (KY92 and KY12) emerge at the opposite extremes of the PC2 plot axis. In Figure 5, for the BI data set, PC1 and PC2 account for 23.7% and 22.7% of the total variation for PC1–10, respectively. The PC1 plot axis recapitulates PC2 in Figure 4 and differentiates the Kerry (KY92 and KY12) from the ANGU and RANG breeds with the other British breeds emerging between these two extremes. These results highlight the genetic distinctiveness of the Kerry cattle breed in comparison to a wide range of British production and heritage landrace cattle breeds and support their status as an important cattle genetic resource that should be prioritised for conservation.

The PC2 plot axis in Figure 5 differentiates the HRFD breed from the other British and Irish breeds and reveals substantial genetic diversity among individual HRFD animals. However, in this context, it is important to note that the pattern of genetic diversity revealed here for the HRFD population sample may be due to ascertainment bias as a consequence of the strategy used to design the Illumina® Bovine SNP50 BeadChip. In this regard, many of the SNPs that constitute this first-generation SNP array were identified from heterozygous positions in the inbred Hereford female (L1 Dominette 01449) bovine genome assembly or through comparisons of random shotgun reads from six diverse cattle breeds that were aligned directly to the same Hereford genome assembly (Matukumalli et al., 2009). This approach to SNP array design will inevitably lead to elevated intrabreed genomic variation using the Illumina® Bovine SNP50 BeadChip with Hereford cattle (Meuwissen, 2009) and accounts for the dispersed pattern of individual HRFD samples in Figure 5.

Examination of Figures 4, 5 indicates that two of the KY12 animals sampled may exhibit a genetic signature of ancestral crossbreeding with another cattle population, which, anecdotally, is likely to have been due to crossbreeding with Angus cattle. Therefore, another PCA plot was generated (Supplementary Figure 1) that shows PC1 and PC2 for individual animals from the KY92, KY12, ANGU and RANG population samples. The two animals exhibiting a genetic signature of possible crossbreeding (KY12_06 and KY12_58) are indicated on Supplementary Figure 1. Notwithstanding the KY12_06 and KY12_58 data points, the genetic similarity among all Kerry cattle sampled is evident by comparison of the tight KY92 and KY12 sample cluster to the dispersion of the ANGU and RANG samples on the PCA plot in Supplementary Figure 1.

The values for the variation accounted for by PC3, PC4 and PC5 in Figure 4 (EU data set) are relatively high (15.7, 13.8, and 10.4%, respectively). For Figure 5 (BI data set), the variation accounted for by PC3 is also relatively high (19.1%). Therefore, we generated additional PCA plots of PC1 for each of the two data sets vs. these additional principal components (see Supplementary Figures 2–5).

The results of the fastSTRUCTURE analyses using the EU and BI data sets are shown in Figures 6, 7, respectively. For both analyses, the Kerry cattle (KY92 and KY12) cluster as a single group at K = 2 and are differentiated from all other European or British and Irish cattle breeds. The other breed group that is clearly differentiated at K = 2 in Figure 7 is the cluster composed of the ANGU and RANG breeds. These results mirror the pattern shown for PC1 in Figure 5, and again emphasise the genetic distinctiveness of Kerry cattle compared to other European production and landrace heritage breeds. Using the chooseK.py script the “true” number of clusters corresponding to the likely number of ancestral populations was estimated to be between 12 and 14 for the EU data set and either 7 or 8 for the BI data set.

For both data sets, animals from the KY12 population sample appear to be more genetically homogenous compared to the KY92 population sample. This observation may be a consequence of increasing use, since the early 1990s, of small numbers of artificial insemination (AI) Kerry sires. It is also noteworthy that the two individual animals detected with a substantial signature of putative historical crossbreeding (KY12_06 and KY12_58) show marked patterns of population admixture in the fastSTRUCTURE results, which are indicated by red arrows in Figures 6, 7.

The results from modelling historical N_e in a selection of production and heritage cattle breeds and populations (KY92, KY12, DXTR, BSHN, BGAL, LNCR, ANGU, JRSY, and HOLS) are provided in Supplementary Table 2 and visualised in Figure 8. The “demographic fingerprints” (Barbato et al., 2015) of the two Kerry populations shown in Figure 8 and tabulated in Supplementary Table 2 are more similar to those of the production breeds with large census populations (BSHN, ANGU, JRSY, HOLS) than the other heritage breeds with relatively small census population sizes (DXTR, BGAL, LNCR). The KY92, KY12, BSHN, ANGU, JRSY, and HOLS populations show a declining trend from historical N_e peaks between 1,500 and 2,000 more than 900 generations ago to N_e values estimated to be less than 200 within the last 20 generations. One the other hand, the DXTR, BSHN, BGAL and LNCR populations display a more severe decline from historical N_e peaks between 2,500 and 4,000 more than 900 generations ago to N_e values estimated to be less than 150 within the last 20 generations.

It is important to keep in mind these N_e trends may be partly a consequence of the relatively small sample sizes for the DXTR, BGAL, and LNCR breeds (see Table 2), coupled with different histories of migration, gene flow and, in particular, strong artificial selection in the production cattle populations. Notwithstanding these caveats, the most recent modelled N_e values for the KY92 and KY12 population samples are 89 and 88, respectively. These values are N_e estimates for 12 generations in the past and assuming a generation interval of between 4 and 6 years, which is based on a pedigree estimate from a similar heritage cattle population of 5.66 (Mészáros et al., 2015), this corresponds to between 48 and 72 years before 2012 (for the KY12 population). This is approximately the period between 1940 and 1965, which is during the time that the Kerry breed started to decline precipitously in census population size and also N_e estimated from herd book data (O'hUigín and Cunningham, 1990; Food and Agriculture Organization, 2007).

From a conservation perspective, livestock populations generally exhibit N_e values relative to total census population sizes (N_c) that are substantially lower than seen in comparable wild mammal populations (Hall, 2016). Also, estimation of N_e using methods such as SNeP that leverage genome-wide SNP linkage disequilibrium (LD) data will tend to underestimate N_e because of physical linkage between many of the SNPs in the data set (Waples et al., 2016). Nevertheless, taking this into account, there is still cause for concern that the most recent N_e values modelled for the KY92 and KY12 population samples are below the critical N_e threshold of 100 recommended by Meuwissen (2009) for long-term viability of discrete livestock breeds and populations.

Supplementary Table 1 shows a genomic relationship matrix in terms of genotype IBS for the genome-wide SNP data generated for individual animals in the KY92 and KY12 population samples. Close genomic relationship between individual animals sampled from the same herd is evident in the SNP genotype IBS values between samples. In addition, the relatively low genomic relationship between the KY12_06 and KY12_58 outlier samples (Figures 4–7) and the rest of the Kerry cattle sampled is also evident in Supplementary Table 1. These data emphasise the value of intrapopulation genomic relationship values for identifying animals (in this case, KY12_06 and KY12_58) that should not be used in breeding programmes. They also highlight the potential of genome-wide SNP data for providing a systematic approach to prioritising males and females with minimum genomic relationship for breeding to minimize loss of genetic diversity and maintain or increase N_e (Gandini et al., 2004; Meuwissen, 2009; de Cara et al., 2011, 2013).

Genome-wide autozygosity estimated from SNP data using F_ROH and the F_ROH statistic are visualised in Figure 9 for individual animals from the KY92 and KY12 population samples and a range of European comparator breeds. Additional summary ROH data is provided in Supplementary Table 3 and also Supplementary Figure 6, which reveals marked inter-population differences in ROH length and demonstrates that the SNP density of the Illumina® Bovine SNP50 BeadChip is too low to reliably capture ROH below 5 Mb in length, an observation previously reported by Purfield et al. (2012).

There is significant variation in F_ROH values among individual animals and between breeds and populations. The non-parametric Wilcoxon rank sum test was performed on F_ROH distributions for all pairwise population/breed comparisons with application of the Bonferroni correction P-value adjustment for multiple statistical tests (Supplementary Table 4). This analysis demonstrated that the KY12 population sample exhibited a significantly higher mean F_ROH value than the KY92 population sample (0.098 vs. 0.079; P_adjust = 0.0081). This is important from a conservation genetics perspective, indicating that genome-wide autozygosity, which is highly correlated with conventional pedigree-based estimates of inbreeding (F_PED) for cattle (Purfield et al., 2012; Ferencaković et al., 2013; Martikainen et al., 2017), has increased for the Kerry cattle population in the 20 years between sampling of the KY92 and KY12 populations.

The importance of understanding and quantifying genome-wide autozygosity for genetic conservation purposes has recently been highlighted through correlation of F_ROH with inbreeding depression for a range of production traits in domestic cattle (Bjelland et al., 2013; Pryce et al., 2014; Kim et al., 2015). Importantly, F_ROH has also been shown to correlate with inbreeding depression for bovine fertility traits in both males (Ferencaković et al., 2017) and females (Kim et al., 2015; Martikainen et al., 2017). Finally, according to basic population genetic principles, recent inbreeding captured by F_ROH will lead to recessive deleterious genomic variants emerging at a population level—a phenomenon that has been studied in both humans and cattle (Szpiech et al., 2013; Zhang et al., 2015).

The results of the genome-wide scan for signatures of selection using the CSS method in the Kerry cattle breed are shown in Figure 10. Six distinct selection signatures were detected on BTA9, BTA12, BTA16, BTA17, BTA19, and BTA28. A total of 178 genes were located within the genomic ranges ± 1.0 Mb of selection peaks and 32 of these genes were located within the boundaries of a selection peak. Supplementary Table 5 provides detailed information for these 178 genes.

A single gene was located within the BTA9 selection peak—the phosphodiesterase 7B gene (PDE7B), which has been associated with neurobiological processes (de Gortari and Mengod, 2010) and has been previously linked to genetic changes associated with dog (Canis lupus familiaris) domestication and behaviour (Freedman et al., 2016). A single gene was also located within the BTA16 selection peak—the dorsal inhibitory axon guidance protein gene (DRAXIN), which encodes a protein that regulates axon guidance, neural circuit formation and vertebrate brain development (Islam et al., 2009; Shinmyo et al., 2015). Twenty-four genes were located within the BTA17 selection peak, including BICDL1, RAB35, and RNF10, which have been associated with neurobiology and brain development (Hoshikawa et al., 2008; Schlager et al., 2010; Villarroel-Campos et al., 2016) and SIRT4 and COQ5 that function in cellular metabolism (Kawamukai, 2015; Elkhwanky and Hakkola, 2017). Six genes were located within the BTA28 selection peak, including, most notably, the Rho GTPase activating protein 22 gene (ARHGAP22), which has recently been associated with bovine fertility as an mRNA expression biomarker for oocyte competence in cumulus cells (Melo et al., 2017).

To obtain a broader perspective on natural and artificial selection acting at a population level on the Kerry cattle genome, a functional gene set enrichment approach (GSEA) was taken using IPA with the 178 genes located within ± 1.0 Mb of each selection peak (Supplementary Table 5). Of these 178 genes, 141 could be mapped to the IPA knowledgebase and the summary results for the IPA Physiological System Development and Function category are shown in Supplementary Table 6, revealing an enrichment of biological processes associated with nervous system development and behaviour. This functional enrichment coupled with the neurobiologically relevant single-gene selection peaks on BTA9 (PDE7B) and BTA16 (DRAXIN) suggests that natural and/or artificial selection related to brain development and behaviour has been important in the microevolution of the Kerry cattle breed. In this regard, it is therefore noteworthy that Kerry cattle, including bulls, are recognised as being comparatively docile and easy to manage (Curran, 1990).

The genome-wide phylogenetic and population genetic analyses detailed here demonstrate that Kerry cattle represent an important farm animal genetic resource, befitting the breed's status as a livestock population with a unique history of adaptation to the climate and physical geography of southwest Ireland at the edge of Western Europe. Notably, from a genetic conservation and breed management perspective, high-resolution comparative PCA (Figures 4, 5) and genetic clustering results (Figures 6, 7) demonstrate that Kerry cattle are markedly distinct from other British and European cattle populations. This observation may also be placed in the context of recent paleogenomic studies that have detected ancient gene flow from wild British aurochs (B. primigenius) into the ancestors of present-day Kerry cattle (Orlando, 2015; Park et al., 2015; Upadhyay et al., 2017).

The current genetic status of the Kerry cattle population is underlined by analyses of genetic effective population size (N_e) and inbreeding using genome-wide SNP data. As shown in Table 2, genome-wide observed heterozygosity (H_o) is relatively high in the KY92 and K12 population samples, particularly for endangered heritage cattle breeds. However, it has been long recognised that monitoring N_e is a more important tool for rational breed management and long-term conservation of endangered livestock populations (Notter, 1999; Gandini et al., 2004; Biscarini et al., 2015). As shown in Figure 8 and Supplementary Table 2, the Kerry cattle population has a recent demographic trend of N_e decline, to the point where the most recent modelled N_e values are below the recommended threshold for sustainable breed management and conservation (Meuwissen, 2009). There is also cause for concern that genomic inbreeding estimated using genome-wide autozygosity (F_ROH) and visualised in Figure 9 has increased significantly in the 20-year period between the sampling of the KY92 and KY12 Kerry cattle populations.

In a more positive light, as shown in the present study, detection of discrete signatures of selection using the relatively low-density Illumina® Bovine SNP50 BeadChip is encouraging for wider studies of genome-wide microevolution in endangered heritage livestock populations such as Kerry cattle. Future surveys of heritage livestock populations that use higher-density SNP array platforms and ultimately whole-genome sequence data could provide exquisitely detailed information on the genomic regions and associated polygenic production, health, fertility and behavioural traits shaped, over many centuries, by the agro-ecology and pre-industrial farming systems of southwest Ireland.

In conclusion, the results presented here for the Kerry cattle population demonstrate that population genomics analyses of large SNP data sets can provide useful information concerning the microevolution and recent genetic history of heritage livestock breeds. In particular, we would recommend that comparable surveys in other populations consider the use of genome-wide scans for signatures of selection, which can provide a functional genomics perspective on evolutionary adaptations to particular agricultural environments and production systems.

The authors IWR and SP are employed by IdentiGEN, Ltd. The other authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.