Introduction

The European bison or wisent (Bos bonasus (Syn.: Bison bonasus) Linnaeus, 1758) represents a textbook example of successful ex situ population management and reintroduction following severe bottlenecks and extinction in the wild in 1927. Ex situ and in situ population management is based on the world’s first studbook for a threatened species (European Bison Pedigree Book; EBPB) established for conservation purposes1. Today’s global population size of > 9,500 is the result of this successful population management during the last almost 100 years1,2,3. Despite this success, the species is still threatened by genetic erosion due to a small gene pool resulting from a total of only 12 founders with uneven founder representations4,5,6. Besides this massive bottleneck, the population went through several other contractions in population size before and after7,8 the establishment of the breeding programme in 1923, with the latest happening during World War II9. Additional bottlenecks still happen through initial founder effects when reintroducing a limited number of animals from captivity into the wild in the framework of reintroduction programmes. While it is presently not fully understood to which degree reduced genetic diversity hampers population fitness and adaptability to changing environmental conditions, an increased susceptibility to diseases, such as posthitis or balanoposthitis is commonly suspected to be a likely consequence of low genetic diversity and high inbreeding coefficients1,10.

The current B. bonasus population is managed separately in two breeding lines: the lowland line (LL), representing the natural subspecies Bos bonasus bonasus Linnaeus, 1758, originated from seven founders. The lowland-Caucasian line (LC) was founded by 11 founders of B. b. bonasus, including the seven founders of LL, and a single male of Bos bonasus caucasicus Turkin & Satunin, 1904. The LC is factually managed as an open population, whereas gene flow from LC into LL is undesired and its prevention is considered a priority in European bison conservation management5.

Because of the genetic issues mentioned above it is pivotal to track genetic diversity and relatedness in both ex situ as well as reintroduced populations of the European bison. However, due to genetic homogeneity of the species, standard approaches of using microsatellite markers for genetic monitoring as well as individual identification are not applicable in European bison conservation management. Tokarska et al.11 showed that single-nucleotide polymorphisms (SNPs) are more suitable to assess identity and paternity compared with microsatellites. Another important issue is DNA sampling: in contrast to often impractical and undesired invasive sampling, the ability to use non-invasive samples to assess viable genetic population data from appropriate numbers of individuals could be a valuable tool for e.g. monitoring wild species or for the use in behavioural studies12,13,14,15,16.

Consequently, a comprehensive genetic assessment with a reliable molecular method accompanying the existing conservation management in the wisent is needed to enable further preservation of genetic depletion of the already low intraspecific diversity in the long-term. Here, we present a novel reduced 96 SNP panel applicable for non-invasive samples of the European bison. The new modular marker panel tackles several conservation-relevant issues: (i) individual discrimination, (ii) parental assignment, (iii) sex determination, (iv) assessment of genetic diversity within the population, (v) breeding line discrimination and (vi) cross-species detection for European bison. Molecular resolution of parental assignment and genetic diversity in the wisent measures were evaluated with genealogical studbook data. Additionally, we evaluated the applicability of the SNP panel for further Bovini (Gray, 1821) with potential conservation relevance in basic applications.

Results

General assay performance and selection of the final 96 SNP marker panel

Invasive DNA samples, such as muscle tissue and blood, were used as reference samples to validate the informative value of each SNP marker in European bison for individual discrimination, breeding line assignment as well as cross-species detection and sex determination. Furthermore, samples with degraded DNA with the focus on dung were utilised to test for the suitability of the SNP markers for the non-invasive approach. A detailed description of the selection of the final SNP set as well as the different sample types, sampled species, sampling method, sample storage and DNA extraction can be found in the “Methods” section.

In total, we found 226 candidate SNP loci designated for the abovementioned applications from literature (see “Methods” section), and additionally designed SNPtype assays for five gonosomal SNP loci from public sequences as described in more detail in the “Methods” section. From those initially tested 231 SNP markers, 111 markers failed to amplify, showed no interpretable clusters or SNP polymorphism in the European bison and were thus excluded after the first round of wet laboratory tests. From the remaining 120 SNP markers a final set of 96 SNPs was selected based on best performance with non-invasively collected samples to accommodate for the applicability with the 96.96 microfluidic chip format. This final 96 SNP marker panel with overlapping subsets consisted of 90 autosomal markers for individual discrimination, 63 markers for parental assignment and the assessment of genetic diversity as well as 18 markers for breeding line discrimination between LL and LC. Six candidate SNP assays in the gonosomal amelogenin (AmelY1, AmelY2, AmelY3, AmelX1, AmelX2) and the zinc finger gene (ZFXY), respectively, were validated for sex determination in European bison and other bovines. Five assays showed consistent amplification for invasive samples, whereof four were excluded in later testing phase due to failing with non-invasive samples. Though no template controls (NTCs) were amplified within the X-chromosomal cluster, the locus AmelY1 was still found to be suitable due to the distinct Y-chromosomal-associated allele cluster and was finally included in the 96 SNP panel.

Subsets for parental assignment and genetic diversity assessment were tested for Hardy–Weinberg equilibrium (HWE) and linkage disequilibrium (LD) across 58 non-first-degree relatives, resulting in a selection of 63 unlinked markers in HWE. The R2-based LD calculations estimated high linkage especially for posthitis-associated loci of the panel (Supplementary Fig. S3).

The mean call rate for non-invasive samples was 92.4% and the mean genotyping error (GE) was 1.9%, with allelic dropouts (ADOs) = 1.6%, and false alleles (FAs) = 0.3%. AmelY1 showed a GE rate of 0.044 across non-invasive samples. The mean call rate from invasive samples was 98%, while the mean GE rate over all marker was close to 0 (Supplementary File SNP_marker_list_details.xlsx).

Modular subsets of the 96 SNP panel

Individual discrimination

We found 90 polymorphic SNPs that allow for secure individual discrimination in European bison (see Fig. 1), which is essential for genetic monitoring of populations and a prerequisite for further analyses. The microsatellite panel with 11 loci used in the pilot study did not reach sufficient resolution for the probability of identity (PID) and the probability of identity among siblings (PIDsib), which is considered to be a sufficiently low threshold for most applications involving natural populations17. In contrast, the SNP subset of 90 polymorphic markers reached a PID ≤ 0.0001 with ≤ 10 markers and PIDsib ≤ 0.0001 with ≤ 18 markers for B. bonasus (Fig. 1).

Figure 1
figure 1

Probability of identity (PID) and probability of identity among siblings (PIDsib) of genotyped microsatellites (n = 11) and autosomal SNPs (n = 95) for European bison. Horizontal dashed red line: PID threshold for natural populations by Waits et al.17 is not overcome by the microsatellite panel. SNP-based PID reaches threshold at approximately 10, PIDsib at approx. 18 loci for the European bison. Approximations of PID and PIDsib close to zero are reached approx. with 13 and 24 loci, respectively. The x-axis was cut at locus combination of 30 loci for more conciseness whereby the approximation of the SNP-based PIDs does not change after 30 loci. PIDsib estimations of the microsatellite panel are outside of the scale. PID and PIDsib for all other Bos species for which individualisation was possible based on 95 autosomal SNPs are provided in Supplementary Fig. S4.

The mean number of allele mismatches found between pairs of genotypes within the total wisent population were 28.2 (LC: 29.5; LL: 26.5), for American bison 11.2, for gaur 6, for banteng 4.1 and highest for domestic cattle with 34.9. The lowest value for European bison (= 17) was found between two first-degree relatives. The lowest number of allele mismatches in the American bison was 7, for domestic cattle 23, for gaur and banteng 4, also all between two first-degree relatives each (Fig. 2).

Figure 2
figure 2

Detected number of mean allele mismatches between individual genotypes (genotypes consisting of 95 loci) of European bison (both breeding lines separately) as well as American bison, domestic cattle, gaur and banteng. Lowest allele mismatches are highlighted in red. Individual sample size per group is noted (n). Allele mismatches between genotypes of five unrelated cattle individuals are > 40 loci.

Sex determination

In genetic monitoring of populations sex determination provides often crucial information for sex-biased behaviour or analysis of relatedness. One out of six gonosomal SNPs (GTA0242198) was found to be suitable for sex determination in European bison. Correct sex determination failed for six European bison cows out of a total of 137 individuals (4.4%). These six individual samples showed three to four FAs in the Y-chromosomal cluster within six replicates. Sex determination was also possible with American bison, yak, domestic cattle, gaur, banteng, water buffalo, lowland anoa, mountain anoa, Cape buffalo and forest buffalo. Over all 11 species (235 individuals) 92.9% were correctly determined, 4.4% were false positive and 2.8% not determinable.

Parental assignment

Parental assignment based on molecular data is beneficial to resolve genealogy if pedigree data is incomplete or not available. To test for applicability of the final SNP panel, parental assignment for comparison with the pedigree data was conducted with a subset of 63 SNPs for 137 individual genotypes (see exemplary family network with 23 relatives in Fig. 3). Of those, 128 were individually assigned during sampling in the field, while nine individual genotypes originate from not individually assigned samples. According to the studbook, 48 parental assignments were expected to be detected between the available genotypes. From these, 41 maternal and paternal relationships were correctly identified. In eight cases, the parent-offspring (PO) relationship was detected but the offspring was assumed to be the parent or vice versa. In all of those latter cases the genotype of the second parent was unknown. In seven cases the expected PO relationship was not identified. In eight cases, PO relationships were estimated false-positively compared to pedigree data. Five of these false positives were assigned to second-degree relatives, one to a third-degree and one to a fourth-degree relative with recent inbreeding involved. Despite one case of a second-degree relative all false-positive parental assignments between individuals were obtained if no true parental genotypes were available in the molecular sample set. No false-positive parental assignments between individuals of the two breeding lines were estimated.

Figure 3
figure 3

An exemplary family network to document the integration of molecular kinship analysis into the present pedigree data from the European Bison Pedigree Book (EBPB). Three generations of 23 individuals assigned to LL were sampled and genotyped from three holders in the Netherlands and Germany (Lelystad (Natuurpark), Duisburg (Zoo), and Springe (Wisentgehege)). Circles represent female individuals and squares male individuals (filled symbols: genotyped). Green edges around the individuals represent successful molecular sex verification, whereas solid red edges represent falsely positive sex assignments and dashed red edges, where no molecular sex assignment was possible. All genotypes are based on a single sample per individual. Triple edges: sample was not individually assignable in the field but was assigned to an individual with the genotype based on sex determination and parental assignment. Different colours of the genealogical lineages represent different verification states: green: genetically verified kinships from the EBPB; blue: genetically assigned kinships with lacking data in the EBPB; red: kinship from the EBPB not genetically verified; black: kinships genetically not verifiable due to missing genotypes. 10 parental assignments (sired by ‘EBPB#7591’ and ‘EBPB#10081’) with unknown maternities from the EBPB were included to visualise the high degree of at least half-sibling relationships of the females/potential mothers in Lelystad; grey dashed: presumed kinships not verifiable due to missing genotypes and missing data in the EBPB. Asterisk: case of inbreeding. All breeding line assignments of the displayed individuals were genetically verified (not noted here).

Two out of twelve originally individually unassignable field samples were assigned to known individuals documented in the EBPB through their as well genotyped parents: ‘Durana’ (EBPB#11813) and ‘Odila’ (EBPB#13951).

Genetic diversity

All 63 non-linked markers in HWE (Supplementary File SNP_marker_list_details.xlsx) were used for the assessment of genetic diversity in the European bison in comparison to pedigree-derived values. Generally, gene diversity (GD) and heterozygosity values (HS/uHE) were stable within but not consistent between molecular and pedigree data, whereas the F-statistics showed comparable values between both data sets (Table 1). The F-statistics tend to be variable even based on same molecular or pedigree data depending on the utilised software and its calculation method. Notably lower genotype samples sizes negatively affected mostly heterozgosities and F-statistics and caused erroneous calculations most prominently in the FIS (Table 1). If calculated per breeding line, LC showed a consistently higher genetic diversity than LL (Supplementary Table S4).

Table 1 Genetic diversity measures based on SNP genotypes and pedigree data for different sets of European bison individuals.

Breeding line discrimination

A subset of 18 SNP markers provided the lowest false-positive rate in breeding line assignments according to current requirements in wisent conservation5. This marker subset with the highest resolution was identified when the FST threshold per locus was set to a minimum of 0.075. It includes two out of six loci with private alleles found in LC among 137 individuals in this study (Supplementary File SNP_marker_list_details.xlsx).

Seven individuals (5.1%) with the Bayesian genetic clustering (STRUCTURE) and five individuals (3.6%) with the maximum likelihood genetic clustering (adegenet) were false-positively assigned to a breeding line (Bayesian: total: n = 5, LC: n = 4, LL: n = 1; Maximum Likelihood: total: n = 4, LC: n = 4, LL: n = 0) or were not clearly assignable (Bayesian: total: n = 2, LC: n = 1, LL: n = 1; Maximum Likelihood: total: n = 1, LC: n = 0, LL: n = 1; Fig. 4). Four samples from Russia were constantly false-positively assigned to LL based on the given breeding line assignment.

Figure 4
figure 4

Assignment probabilities [%] based on 18 loci selected for breeding line discrimination between LC (n = 76) and LL (n = 61) in the European bison: (a) Bayesian genetic clustering computed with STRUCTURE; (b) Maximum-likelihood genetic clustering computed with adegenet. The black line shows the previously assigned lineage distinction (LC: blue; LL: orange). Dashed red lines indicate assignment thresholds. Bars tarnished red mark individuals with unexpected lineage assignment; bars tarnished grey mark individuals not assignable with genotypic data according to the assignment threshold. Brown arrows: F1 breeding line hybrids. White asterisks: LC individuals with at least one of the six private alleles found in LC. See Supplementary Table S5 for the order of individuals shown here.

Cross-species detection

Since most of the utilised SNPs were detected to be polymorphic in non-target species and considering the applied approach for in situ monitoring, we tested SNP subsets to discriminate, assign and exclude samples from other evolutionarily significant units (ESUs). All non-target taxa with SNP call rates > 80% (16 ESUs in 10 Bovini species; Fig. 5) could be distinguished from B. bonasus in a Principal Coordinates Analysis (PCoA) based on 95 or 31 (for domestic cattle) loci (Fig. 6). Samples from more distantly related taxa showed generally much lower call rates and less SNP polymorphism (Fig. 5). See Supplementary File SNP_marker_list_details.xlsx for SNP subsets suited for cross-species identification between several other ESUs within Bovini along with provided reference genotypes from a broad phylogenetic diversity of this tribe (Supplementary File Genotype_lists.xlsx).

Figure 5
figure 5

SNP call rate (%) for 95 autosomal SNPs in the European bison and 15 non-target species with corresponding numbers of individuals (n). The length of a solid bar indicates the mean SNP call rate for each analysed species. Blue bars reflect all groups classified to the genus Bos, blue-grey bars groups classified to the subtribe Bubalina and grey bars species outside of Bovini. A SNP call rate of at least 80% call rate (red dashed line) is the threshold for inclusion into further analysis. The orange-hatched bars show the percentage of found polymorphism over 95 loci within the groups. The cladogram reflects known evolutionary relationships between the species18. The asterisk points out the tribe of Bovini.

Figure 6
figure 6

(a) PCoA of 137 European bison (both breeding lines) and 116 individuals of 10 non-target Bovini species (16 ESUs) with a SNP call rate over 80% utilising all 95 autosomal SNP loci. (b) PCoA of 137 European bison and 15 domestic cattle (four major lineages) utilising a subset of selected 31 SNP loci. Clusters containing higher taxa like the subgenus Bibos (Hodgson, 1837) and the subtribe Bubalina (Rütimeyer, 1865) are marked in grey circles. Eigenvalues (a): axis 1: 233.68; axis 2: 158.19; eigenvalues (b): axis 1: 33.46; axis 2: 26.85.

Discussion

Resolution of the new SNP panel

The genetic assessment of wildlife populations via non-invasive samples reduces undesired anthropogenic interference as much as possible and consequently became common practice in wildlife genetic studies16. Once developed, such reduced marker panels for genotyping of non-invasive samples with genome-wide SNPs provide a standardised, fast-applicable, and low-cost genomic approach for conservation19. Previously, it has been shown that genotype recovery for non-invasive samples is overall higher using microfluid SNP panels compared with frequently utilised microsatellites20. In line with von Thaden et al.20 we also found high informative content, reliability and reproducibility of genotypes of our microfluidic SNP panel, with a high average genotyping quality across samples (average call rate = 92.4%, GE rate = 1.9%).

To gain for maximum resolution of the panel, we decided to accept increased amplification rates in NTCs for some selected loci. Occasional fluorescence of NTCs are known in SNP genotyping and is considered to be no major concern due to marker-specificity and inconsistency in genotype yields from NTCs19. With the marker GTA0242130 all NTCs showed fluorescence and solely clustered with the homozygous YY cluster. Nevertheless, this marker was kept because of the overall good clustering. If the downstream analysis was not negatively impacted lower call rates were also tolerated: a single autosomal marker (GTA0250956) showed a drastically lower call rate of 76.2%. Since this marker is highly informative for breeding line discrimination (FST = 0.112 in a set of 58 individuals not in a first-degree relationship) with GE rate of 2.2%, it was kept. Invasive samples generally showed complete call rates and minor GE rates and thus no need to be replicated with the current SNP panel.

The European bison is the only recent wild cattle in its current distribution21. However, within all native regions of the European bison, domestic cattle and partly domestic water buffalos occur as livestock22,23 and their faeces could thus be confused during field sampling (see Supplementary Discussion for a more extensive discussion on bovid dung as a considerable genetic sample type). Therefore, it is important that obtained genotypes can be reliably assigned to the correct species to avoid biased results in a genetic monitoring. With the SNP panel presented here all genotyped Bovini could be distinguished from the European bison and furthermore, clustered according to their ESU. The proximity of the cattle cluster to the European bison cluster can be attributed to the fact that all autosomal SNPs in this study were originally detected in B. primigenius (Fig. 6a). This also causes the strikingly high degree of SNP polymorphism in this species (Fig. 5). With a subset of 31 selected SNPs from the novel marker panel it is possible to genetically distinguish B. primigenius from B. bonasus (Fig. 6b; Supplementary File SNP_marker_list_details.xlsx).

The new SNP panel allowed for safe individual discrimination, with considerable allele differences between most individuals. The lowest number of allele mismatches between individuals of European bison was 17 loci between first-degree relatives. This is two to three times higher than allele mismatch thresholds allowing individual discrimination known from similar SNP panels for other species24,25, resulting in a high degree of confidence. This is roughly consistent if considering the commonly used probability threshold for natural populations by Waits et al.17: approximately 18 SNPs would be sufficient for reliable individual discrimination (Fig. 1).

The GE rate of 0.04 for the sex marker led to six failed individual sex determinations (three false positives and three not determinable) out of a total of 137 European bison. Despite occasional misidentifications, which typically occur in genetic information derived from non-invasively collected samples26, this marker set will be helpful in assessing sex ratios and sex-related behaviour in free roaming European bison populations.

Reliable individual genotypes can be used for parentage analysis, which is highly susceptible towards genotyping errors27. Previous studies conclude that 50–60 SNPs selected for high heterozygosity would be sufficient to resolve paternity in the European bison11,28. The number of required loci depends on the breeding line and the grade of information regarding the parents11,28,29. A 100 SNP panel has been published for parental assignment for LL exclusively29, of which a portion of markers were included in the current panel. Wojciechowska et al.3 developed a subset of 50 SNPs for parental assignment applied for both breeding lines. With the reduced 63 SNP subset in our study, parental assignment was also successful for LC and additionally proved effective for non-invasively collected samples. In difficult cases as shown in the exemplary family network (Fig. 3), where recent inbreeding meets low genetic diversity in LL, which is expected to require more loci to resolve PO relationships28 the panel resolution reaches its limits and partially fails to disentangle kinship. Such cases show that only the combination of genetic assessment with available studbook and other metadata will allow to resolve patterns of relatedness with high certainty30,31. This combined approach is state of the art in other comprehensive genetic population monitoring assessments32 and in line with the conclusion of other studies that parental assignment is strongly facilitated in case of one known parent11,28,29.

Despite the high genetic similarity due to recent origin from an overlapping subset of founders and ongoing one-directional gene flow from LL to LC5, the presented SNP panel allowed for reliable discrimination of the two breeding lines as an overarching requirement for conservation actions5. While breeding line discrimination has previously been achieved with sets of 1,53628 and 30 selected SNPs3, our subset of just 18 markers achieved a comparable resolution including F1 breeding line hybrids, which are formally assigned to LC following the official management definition5 (Fig. 4). Among the tested samples only four individuals from ‘Russia’ documented as LC individuals clearly clustered in LL regardless of the utilised clustering method. While wild herds founded only by LL individuals in Russia are known33, we have no detailed information regarding those particular samples, and thus the reason for the apparent incongruency cannot be deduced here. It is recommended to apply both genetic clustering methods complementarily to identify marginal assignments based on the minimum breeding line discrimination threshold. We suggest assigning an individual only if both methods verify at least 60% probability based on the SNP genotype.

The finding of six private alleles within LC is not surprising since this breeding line carries genetic material of five additional founders including one bull from a separate subspecies5. The absence of any of six private alleles in 16 LC individuals (Fig. 4) shows the low information content just relying on those markers and the need for a more discriminative markers if aiming for a robust breeding line separation as the one presented here. The discriminative value for SNP alleles published by Kamiński et al.34, which were described to be private for one breeding line, could not be confirmed in our study. This can be explained by the small and presumably not representative sample size of only ten individuals genotyped in the aforementioned study.

Neither the private alleles nor the other discriminative markers have assignable genetic origins from one of the two subspecies, B. b. bonasus or B. b. caucasicus and could be a consequence of distinct breeding management during past decades. The marker subset for breeding line discrimination presented here is thus not suitable for a validation of both the breeding lines as ESUs. Solely designed to assign individuals to the currently predefined anthropogenic breeding lines it cannot be applied to argue for or against the separate management of the two breeding lines within the European bison.

Comparing genetic diversity estimates between studbook and molecular data

Given its history of consecutive bottlenecks and genetic depletion, an appropriate genetic marker system for B. bonasus should, besides individual discrimination and parentage analysis, allow for measures of genetic diversity in order to aid population management35. For this we selected 63 autosomal unlinked loci in HWE found to be polymorphic in the European bison. The SNPs utilised in this panel were originally detected in domestic cattle3,11,28,29,34,36,37,38,39. Though common practice40,41,42, it is obvious that such a reduced number of SNPs found in a related species as well as an ascertainment bias from selecting for high polymorphism in our target species will not allow for unbiased estimates of genetic diversity43,44. Thus, any results regarding genetic diversity using this SNP panel need to be interpreted with caution.

Different aspects were considered to reduce an ascertainment bias in the current SNP panel. Studies assessing genetic diversity often face the problem of incomplete population sampling35. In this study, the pedigree-based founder representation of the genotype set (n = 99) was compared with a larger pedigree data set of in total 1,296 individuals including all genotyped individuals up to all known founders to validate its population representativity beforehand (Table 1; Supplementary Fig. S1). An overall ascertainment bias can be reduced when ancestral populations are used to develop SNP panels applied on derived populations45. Until today, reintroductions of European bison are largely sourced from the captive population, which therefore resembles an ancestral population from which the majority of individuals for the SNP selection process originated. Overall, our genotyped individuals represent approx. 1.5% of the current generally highly admixed46 global ancestral population (status 20202).

Not surprisingly, estimates of relatedness or inbreeding based on sufficient pedigree data are generally more accurate than marker-based estimates47. However, often no pedigree data is available for conservation-related population studies. Even for the otherwise well documented European bison, this is the case for reintroduced free-roaming herds. Additionally, pedigree-based estimations may suffer from underestimated inbreeding in the founder population48 as well as uncertainties towards the correctness of parental assignments, which can result in an accumulation of errors over time. This concern has been raised as well for the EBPB49,50. It is also known that genetic diversity estimates, whether based on pedigree or molecular data, suffer from small sample sets especially with small gene pools caused by inbred populations and/or sample sets with high portions of closely related individuals51,52. Thus, estimation accuracy will be increased by larger sample sizes and decreasing sampling variance of reference genotypes47, particularly within the breeding lines.

Since the 63 SNPs utilised for genetic diversity estimations were specifically selected for high polymorphism, it is obviously not appropriate to directly compare pedigree-based GD values with molecular-based heterozygosities. Still, it is interesting to note that SNP-based fixation indices resemble the pedigree-based values (Table 1). The relatively low FIS is caused by high intermixture within the breeding lines, whereas rare gene flow between LC and LL is manifested in the second highest fixation estimated in the FST. Overall, we found a high degree of admixture over the population, despite of the species’ strongly reduced gene pool. This finding, which is consistent with a recent study utilising 22,602 SNPs46 is a consequence of the successful population management during the last decades. The highest fixation seen in the FIT is caused by different allele frequencies within the breeding lines compared to the total population and is known as the Wahlund effect53. Changes in fixation indices among populations can be caused by dynamic processes such as genetic drift, gene flow, migration or bottleneck events54. Since one of the biggest threats for the European bison is genetic erosion, the new SNP panel can be used to effectively track such trends and changes in genetic diversity and aid conservation efforts aiming at the establishment of stable populations in the wild. Thus, long-term monitoring of genetic diversity will also enable an evaluation of laborious and costly reintroduction efforts for decision makers.

Potential application on other Bovini species

The IUCN red list contains 12 Bovini species (Syncerus spp. included in this study are recognised as conspecific) of which 9 species are listed as threatened (VU: n = 2; EN: n = 4; CR: n = 3)55,56,57,58,59,60,61,62,63,64,65,66. A genetic assessment of those wild cattle, similar to the European bison is therefore of considerable interest. The SNP marker panel presented here was solely developed for B. bonasus. However, as all autosomal SNPs were originally discovered in B. primigenius but are still polymorphic in the European bison, those to some degree evolutionary conserved orthologous SNPs may allow for utilisation in closely related species. Demonstrably, this SNP panel can be utilised for sex determination in all Bovini species (success rate of 92.9%) as well as for individualisation in American bison (both subspecies), domestic cattle (with all four major lineages), gaur (including gayal) and banteng from non-invasive samples. Thus, the new SNP panel developed for the European bison has instant potential for basic population genetics or conservation applications in other threatened wild cattle and may serve as basis for further optimised panels.

Implementation in conservation and research of European bison

The SNP panel presented here has been specifically developed for current questions and needs in ex and in situ conservation of the European bison. Free-roaming European bison are not listed individually in the EBPB and therefore lack genealogical documentation2. The new SNP panel provided here allows the assessment of relationships between wild individuals without the need to catch or harm the animals, and allows for continuous, systematic genetic monitoring, which is recommended to improve in situ conservation efforts67. Genetic population monitoring generates important information for decision makers and can also help raise public awareness68,69. The panel may be as well used to generate sound data in human-wildlife conflicts, which may arise due to damages in forestry or agriculture70. To allow for an effective long-term monitoring of wild populations, it is strongly recommended to genotype all reintroduced founder individuals. Complementing this approach with a subsequent continuous non-invasive genetic monitoring will allow to track population developments over time and help disentangle the effects of e.g. genetic drift, population isolation, migration, and/or changes in (effective) population sizes, home ranges and social structure following reintroductions71.

Even more than 50 years since the first reintroductions, the captive wisent population is still the source for current rewilding efforts. Therefore, an assessment of the ex situ population must go hand in hand with the in situ conservation actions re-establishing Europe’s last species of wild bovines. Ex situ breeding strategies based on pedigrees are tested to be efficient if sufficient genealogical data is available for a species48,72. Until today, this pedigree data is utilised for breeding, culling and reintroduction recommendations5,73. However, due to the above-mentioned weaknesses of pedigree-based estimations on genetic diversity an independent assessment is needed. Further unintended documentation errors in the EBPB are still possible due to certain husbandry conditions, unknown paternity in herds with several mature bulls or natural behaviours like alloparental care, especially non-maternal suckling, known in European bison1. Formally unknown maternal relationships, genetically identified with the new SNP panel presented here, already have found their way into the EBPB. SNP-based marker-assisted breeding strategies in addition to the traditional practice based on the EBPB have been recommended before36,74. This might be especially true for populations with high inbreeding, where it is presumably more important to practice population management based on genetic diversity instead of management purely based on heredity.

Besides its obvious application in population monitoring, the SNP panel may as well serve in research projects aiming at studying various aspects of conservation-relevant European bison biology, e.g. to investigate the influence of dominant male mating behaviour on the genetic structure and effective population size of the species. Furthermore, due to its robustness towards low quality samples, the analysis of collection specimens75 and historical hunting trophies76,77 could provide interesting insights into the development of genetic diversity over time. Recently, the focus on posthitis-associated SNPs39,78 paves the way for an utilisation of genetic assessments of this disease important for wisent conservation management. In prospect, twelve posthitis-associated markers were included into the current SNP panel (Supplementary File SNP_marker_list_details.xlsx). Due to the lack of presence-absence information of posthitis in the genotyped individuals of this study, further investigation is needed.

Despite of the moderate marker number our SNP panel provides a viable tool to monitor reintroductions, validate, revise and construct pedigrees, and assess population structures where no pedigree data is available. Thus, the new SNP panel represents an optimised compromise between the needed non-invasive sampling method, cost-efficiency needed for the application in conservation and the resulting informative accuracy, which is demonstrably and reasonably sufficient for the purpose it was developed for. While other recently presented SNP panels lack implementation in appropriate assays3,11,28,29,34,79 the presented marker panel is non-invasive genotyping approach for the European bison ready to be used in conservation and monitoring studies. Ongoing real-world application comprises dung-based genetic monitoring of the reintroduced European bison in the Țarcu Mountains, Romania (LIFE RE-Bison; LIFE14 NAT/NL/000987). We propose the wider use of this panel both for ex situ population management as well as genetic monitoring of reintroduced European bison.

Methods

All statistical analyses and most graphical visualisations were conducted using R v3.6.080 within RStudio v1.0.4381.

Pedigree data

All EBPB editions from 1947 to 2018 were reviewed to assess genealogical data and to create a total pedigree data set of all European bison sampled in this study (n = 337) up to the founders. The software mPed82 was used to convert the pedigree data into a readable format for PMx v1.5.2018042983.

Sampling and sample storage

This study focused mainly on the collection of faecal samples, however, hair, urine, saliva and nasal secretion as valuable non-invasive sample types were also collected. No animals were trapped, harmed or killed for this study. Invasive sample types like muscle tissue were used as reference samples and originated from study-unrelated samplings. Non-invasive samples were collected in compliance with the respective local and national laws. Within this study 253 individual genotypes from European bison (n = 137; LC: n = 76; LL: n = 61) and additional 15 species were analysed: ten Bovini species in 16 ESUs: American bison [Bos bison (Linnaeus, 1758): n = 35]; plains bison [B. b. bison (Linnaeus, 1758): n = 22]; wood bison [B. b. athabascae (Rhoads, 1897): n = 13], domestic yak [Bos mutus grunniens (Linnaeus, 1766): n = 9], domestic cattle in four ESUs [Bos primigenius (Bojanus, 1827): n = 15]; taurine cattle [B. p. taurus (Linnaeus, 1758) in eight breeds: n = 10; African humpless shorthorn cattle: n = 1; sanga: n = 1; indicine cattle/zebu B. p. indicus (Linnaeus, 1758) in three breeds: n = 384,85], gaur [Bos gaurus (Smith, 1827): n = 10]; Indian gaur [B. g. gaurus (Smith, 1827): n = 6]; gayal [B. g. frontalis (Lambert, 1804): n = 4], Javan banteng [Bos javanicus javanicus (d'Alton, 1823): n = 8], water buffalo [Bubalus arnee bubalis (Linnaeus, 1758): n = 5; river-type: n = 4; swamp-type: n = 186,87], lowland anoa [Bubalus depressicornis (Smith, 1827): n = 7], mountain anoa [Bubalus quarlesi (Ouwens, 1910): n = 1], Cape buffalo [Syncerus caffer (Sparrman, 1779): n = 14] and forest buffalo [Syncerus nanus, (Boddaert 1785): n = 12]. For cross-species tests five further species with each one individual were included: Eurasian elk (Alces alces alces (Linnaeus, 1758)), common red deer (Cervus elaphus elaphus Linnaeus, 1758), Central European wild boar (Sus scrofa scrofa Linnaeus, 1758), European red fox (Vulpes vulpes crucigera (Bechstein, 1789)) and human (Homo sapiens Linnaeus, 1758) (Supplementary File Sample_list.xlsx).

Captive sampling was done in 37 institutions from eight European countries. Samples from free-roaming LL individuals originate from the Białowieża and Knyszyńska forests in Poland and a single bull shot near Lebus in Germany in 2017. Samples from free-roaming LC individuals were collected in Russia and the Rothaar mountains in Germany between 1990 and 2017. Samples from non-Bovini species were taken from our internal collection of wildlife samples.

For sampling of faeces, hair, body liquids like urine, saliva, nasal secretion or blood from environmental surfaces sterile gloves and cotton swabs were used. Beside storage of faecal swab samples in InhibitEX buffer (Qiagen, Germany) all swabs and hair samples were stored in a filter paper and pressure lock bags including a silica gel sachet. Most pure urine samples were collected from urine-soaked snow in winter88. In order to test optimised faecal sampling for genetic analysis, several sampling and preservation methods were previously validated in a pilot study (Supplementary Information), resulting in two equally-suited approaches: (i) collection of 10–15 g of interior faecal matrix with a one-way forceps and storage in 33 ml of 96% EtOH, (ii) swabbing the interior part of faeces and storage in InhibitEX buffer. For this study no tissue samples were invasively collected, unless as by-product from occasionally conducted mandatory earmarking by zoo personnel.

All samples were stored at room temperature (RT; 20–21 °C), except blood samples in Ethylenediaminetetraacetic acid (EDTA), which were stored at −20 °C. Beside from dead individuals some fresh blood samples independently originate from veterinarian procedures occurring alongside this study. Some beforehand stored blood samples were also provided by some holders (collected between 2014 and 2019).

DNA extraction

DNA extraction of non-invasive or minimally invasive samples (hairs, scats, saliva swabs) was conducted in a laboratory dedicated to processing of non-invasively collected sample material12. The QIAamp Fast DNA Stool Mini Kit (Qiagen) for faecal samples and the QIAamp DNA Investigator Kit (Qiagen) for all other non-invasive sample types, respectively, were used to extract DNA on the QIAcube system (Qiagen) generally following manufacturer’s instructions with some adjustments (Supplementary Tables S8S10). DNA from invasive samples was extracted with the Blood&Tissue Kit (Qiagen) according to the manufacturer’s protocol. Nucleic acid concentrations of DNA extracts from invasive samples were measured with a Nanodrop spectrophotometer. Isolated DNA was stored at 4 °C until use.

Pilot study: faecal sampling, preservation and sample storage methodology

To account for the aforementioned methodological challenges, we tested for best practice in faecal sampling, sample preservation and DNA extraction from wisent dung. Mainly faeces, but other invasive and non-invasive sample types of the European bison were analysed with a set of 14 polymorphic out of 21 microsatellite markers from non-coding regions originally developed for different even-toed ungulate species and a sex determination marker89 to evaluate the applicability of the different sampling and storage methods. In the present study, 16 of these markers were applied for the first time to European bison. Using Generalised Linear Mixed Models (GLMMs), we statistically evaluated sampling, sample preservation and DNA extraction of wisent dung and used these results to extrapolate the finally used best practice (Supplementary Information).

Selection of SNP loci and SNPtype assay design

All autosomal SNP loci tested in this study originate from the BovineSNP50 Genotyping BeadChip and BovineHD Genotyping BeadChip (Illumina). A set of 231 informative SNP loci for the European bison was selected from available publications for initial testing (Supplementary File SNP_marker_list_details.xlsx): 14 SNPs with the strongest association to posthitis78, 43 most polymorphic SNPs from Kamiński et al.34, respective 43 loci from Oleński et al.29 filtered by PID, additionally 44 SNP loci from unpublished data by high polymorphic information content (PIC) and 81 SNPS for breeding line discrimination using loci with highest contrary allele frequencies between LL and LC. It is noted that further promising SNP loci from the study Wojciechowska et al.28 for more accurate breeding line discrimination were not available due to missing indication of used loci. For sex determination, a SNP (ZFXY) found in the homologous zinc finger gene distinguishing between the gonosomal ZFX and ZFY with a C/T transition90 was included. Five gonosomal SNPs were identified in the amelogenin gene of European bison, plains bison, taurine cattle and zebu, yak, banteng and gayal using sequence information from GenBank® (http://www.ncbi.nlm.nih.gov/genbank; Supplementary File SNP_marker_list_details.xlsx). Subsequently, SNPtype assays were designed based on sequence information of approx. 300 bp for each SNP locus using the web-based D3 assay design tool (Fluidigm corp.). SNPs were rejected from the initial selection if not traceable at the European Bioinformatics Institute (EMBL-EBI; http://www.ebi.ac.uk) to avoid SNP duplicates or if primer design by Fluidigm corp. failed.

SNP panel development and genotyping

We followed the development guidelines for genotyping degraded samples with reduced SNP panels provided in von Thaden et al.25 to obtain a final 96 SNP panel for implementation into a microfluidic chip system. The following sample set was used during the entire testing phase: 46 invasive reference samples (LL: n = 17; LC: n = 21; taurine cattle: n = 6; plains bison: n = 2) and 90 non-invasively collected samples. For initial wet laboratory tests, we used 150 in silico SNPtype assays in two partitioned genotyping runs to filter for markers with (i) proper amplification and (ii) high informative value. Assays showing failed amplification or indistinct clustering were excluded for final panel selection. All reference samples were normalised before genotyping towards the recommended concentration of 60 ng/µl (Fluidigm). Those samples did not undergo a STA (specific target amplification) pre-amplification step to enrich the target regions for SNP genotyping.

In the next step, serial dilutions of the reference sample set were prepared to concentrations of 5 ng/µl, 1 ng/µl and 0.2 ng/µl and genotyped with the remaining pool of SNPs after filtering to test the markers’ applicability on low template concentrations and subsequent pre-amplification.

Specific target amplification and SNP genotyping

The SNP genotyping procedure using 96.96 Dynamic Arrays™ with integrated fluidic circuits (IFCs)91 was conducted according to the manufacturer’s protocol for genotyping with SNPtype™ Assays (Advanced Development Protocol 34, Fluidigm corp.). Low DNA samples were pre-amplified in a modified STA for enrichment of the target loci before the SNP genotyping PCR. The pre-amplification of the target regions was conducted using 14 cycles for invasive samples and 28 cycles with extracts from non-invasive samples according to von Thaden et al.25.

All experiments and sample setups included NTCs (no template controls) and STA NTCs. In all experiments NTCs and samples were replicated.

Validation of SNP markers and scoring procedure

Raw data analyses of all runs were conducted with Fluidigm SNP Genotyping Analysis v4.1.2 software (Fluidigm) after 38 thermal cycles. Automated clustering and allele scoring of every SNP marker was manually checked and corrected if needed according to the guidelines suggested by von Thaden et al.20. During the development phase every SNP cluster was compared to its profile in former chip runs to keep uniformity in allele scoring. If the clustering pattern of SNP markers diverged to the pattern in former runs the complete marker was disregarded and scored as ‘No Call’ for all samples. Alleles appearing too far from the centre of a cluster were ranked as FAs and were also scored as ‘No Call’.

Validation of genotyping errors

Genotyping errors (GE) of each single replicate were calculated based on a consensus multilocus genotype (subsequently called reference genotype) which was built using all replicates of a sample (for consensus genotypes see Supplementary File Genotype_lists.xlsx). Accordingly, the following rules were applied: in general, the majority rule was applied across replicates. Loci equally scored as homo- and heterozygous were considered heterozygous. For all autosomal loci: if a locus was scored partly to be heterozygous and both opposite homozygous genotypes were found at least twice in other replicates, the genotype was defined as heterozygous. If every possible zygosity was shown in triplicates, the locus was considered to be heterozygous as well. If both homozygous genotypes were scored the more frequent zygosity was assigned. If both homozygosities were scored with 50%, no zygosity was assigned in the consensus. Sex information for the tested individual was used as reference for calculation of the sex markers` GE. Loci with GE rates < 0.05% were excluded from the final panel.

Characteristics of the final 96 SNP panel

The 96 SNPs of the final panel are distributed throughout all B. primigenius chromosomes except autosome 25, which was not represented in the initially tested 231 SNPs as well (Supplementary File SNP_marker_list_details.xlsx). With 2n = 60, the European bison carries the same number of chromosomes92, which suggest a similar distribution of the used SNPs in both species.

Several applications of GenAlEx v6.593 were used for evaluation and assessment of the molecular data as explicitly noted below. A test for LD of the 90 autosomal markers polymorphic in the European bison was conducted using squared allelic correlation (R2) utilising the R package LDheatmap94.

Cross-species detection

Five cross-species markers (GTA0250958, GTA0250953, GTA0250963, GTA0250909, GTA0250962) were selected to be monomorphic in the European bison and polymorphic in the most common sympatric bovine species (domestic cattle) or sister species (American bison), respectively. Those five markers were utilised for cross-species detection only.

In total, 24 taxa/ESUs were selected for the cross-species test on the basis of the following criteria: potentially sympatric with the European bison95,96 and represent candidates for potential confusion in environmental traces such as faeces and stripping damage or sample contamination due to faecal wallowing. All further allopatric Bovini, representing the closest living relatives up to the tribe level collectable in Europe were also included for cross-species detection and were tested for the applicability of the new panel. Human was included to test for methodological contamination. All samples with a SNP call rate over 80% were analysed with a PCoA using all 95 autosomal loci executed in GenAlEx.

Individualisation

The discriminative power of the polymorphic autosomal SNP set (90 loci) and of the microsatellite panel (11 loci, data from pilot study) was assessed by estimating PID and PIDsib in GenAlEx. The loci were sorted according to the highest expected heterozygosity (HE).

The number of allele mismatches between individual genotypes were compared: the lowest number of allowed allele mismatches were expected between close relatives and were used as a guidance threshold for individual discrimination. Except for the sole mountain anoa all genotype sets per species contained first-degree relatives. Only those Bovini species were considered with an allele mismatches ≥ 1.

Parental assignment

The software Colony v2.0.6.697, using the Full-likelihood analysis method was utilised to estimate Parent-Offspring (PO) relationships between all 137 individuals with a subset of 63 SNPs in HWE and without loci in LD. The Full-likelihood method was chosen because it was shown to be the most accurate method of Colony98. The estimations were computed with default assumptions except the following settings: male and female polygamy and inbreeding were assumed since both cases were present in the data set. Very high likelihood precision with allele frequency updates in a very long run was executed. All 137 individuals were put in as offspring and assigned to their sex with the probability of a sire or a dam in the data set = 0.5. No parental sibling inclusion or exclusion were added. It was only excluded for every individual to be its own parent. These settings were chosen to simulate a blind genetic monitoring study where only information is available from the genotypes and the sex determination marker. Genotyping error rates were assumed to be 0.0001 per locus because the used consensus genotypes were generated from at least triplicates and assumed to be reliable.

For validation, an exemplary family network of 23 individuals was chosen, whereof relationships of a bigger part were known. This showcase included three generations from different parks (different sample types from different collectors), many possible parents in siblinghoods, a case of inbreeding, individually assigned and not assigned samples as well as individuals with undocumented maternities and thus, visualise the full range of applications for parental assignment (Fig. 3).

Breeding line discrimination

Based on 58 individual genotypes without first-degree relatives (LC: n = 35; LL: n = 23) GenAlEx was used to identify markers with highest FST in each of the breeding lines to minimise an allele frequency bias by relatedness. If both parents were genotyped, the offspring were removed to obtain the highest allele variation possible. Two methods for genetic clustering were applied to the descriptive markers to test the robustness of the breeding line marker subset across different statistical approaches. Thus, the subsequent analysis was conducted assuming K = 2. A minimum breeding line discrimination threshold of 60% probability was set for both genetic clustering methods. Finally, the selected 18 SNP markers were applied to all 137 individuals including first-degree relatives for breeding line discrimination.

Bayesian genetic clustering

To infer the presence of a distinct breeding line structure the systematic Bayesian clustering approach of STRUCTURE v2.3.499,100,101 was used for microsatellite (Supplementary Fig. S6) and SNP genotypes (Fig. 4) with burn-in periods of 250,000 repetitions and 500,000 MCMC (Markov Chain Monte Carlo) repeats. The simulations were set with K = 1–10 each with ten iterations. STRUCTURE HARVESTER102 was used to select the most likely K value. CLUMPP v1.1.2 was used to combine the iterations of the most likely K value with the FullSearch algorithm among 10 K103.

Maximum-likelihood genetic clustering

The function snapclust104 implemented in the R package adegenet v2.1.1105,106 was used to infer the presence of distinct genetic structures between the two breeding lines. The Bayesian information criterion (BIC) among K = 1–10 was used to estimate the most likely K value (Supplementary Fig. S5).

Assessment of molecular genetic diversity

To select a marker subset for the assessment of genetic diversity in the European bison all markers deviating from HWE within 58 non-first-degree-relatives were discarded utilising χ2 test in GenAlEx and Arlequin visualised in ternary plots (Supplementary Fig. S2) performed with the R package HardyWeinberg v1.6.3107,108. Allelic richness, expected (HE), unbiased expected (uHE) and observed heterozygosity (HO) as well as the F-statistics were measured for all European bison individuals and for each breeding line. Molecular based heterozygosities and F-statistics (FIT, FST, FIS) were calculated in GenAlex and FSTAT v2.9.4109.

PMx110 was used to generate genetic values from pedigree data. PMx provides two methods to calculate pedigree-based gene diversity (GD): from kinship matrix as well as gene drop method111. For the latter method genetic default assumptions (1000 gene drop iterations, autosomal mendelian inheritance mode) were used. GD is equivalent to HE111,112 and was therefore used for pedigree versus molecular data comparisons. For clarification and as it is output by each software, GD will always refer to the pedigree-based values within this study, whereas HE is referring to molecular-based values. Additionally, pedigree-based FST, FIS and FIT were generated in ENDOG v4.8113.

The pedigree-based and SNP-based F-statistics were also compared. In order to do this, two pedigree data sets were used for PMx: for a direct comparison the pedigree-based genetic values were computed including only the successfully SNP-genotyped individuals with known genealogy (n = 99) and their assigned ancestors (n = 982) up to the founders. To evaluate the representativeness of those pedigree-based genetic values, the same calculations were conducted with all sampled individuals with known genealogy in this study (n = 227) and their assigned ancestors up to the founders (n = 1296).

Visualisation and data set conversion

Boxplots were generated with the R packages ggplot2 v3.2.0114 and gridExtra v2.3115. The cladogram of the Bovini and other non-target species was conducted in Mesquite v3.61 (build 927)116. CONVERT v1.31117 was used to adjust data sets for implementation in several analysis programs. The R package genetics v1.3.8.1.2118 was used to transform data sets into partly required genotype data sets.