狼-犬杂交：种群结构混淆遗传指派与杂交分类。 机翻

Summary

Accurate hybrid detection is increasingly important for conservation, yet genetic assignment can be strongly biased when structured populations are analyzed as if they were homogeneous. This problem is especially relevant in recolonizing systems, where individuals from differentiated source populations meet and admix. We present and evaluate a monitoring-compatible framework that accounts for population structure before testing for hybridization, without relying on reference populations from outside the study area. Using four years of routine wolf monitoring in Czechia, a Central European contact zone of differentiated lineages, we show that ignoring population structure can inflate admixture and generate false signals of hybridization. Population-specific calibration combined with independent validation resolved ambiguous assignments and provided no evidence for ongoing wolf-dog introgression within the monitored wolf population. Our results demonstrate that population structure must be explicitly considered when interpreting hybridization signals and provide a practical framework for conservation monitoring in other structured wildlife populations.

Introduction

Hybridization has long been recognized as an important novelty-generating evolutionary process.1 In a conservation context, however, it may also represent a substantial threat to the genetic integrity and long-term persistence of wild populations.2 This risk is particularly pronounced when wild species occur in contact with closely related domestic forms whose population sizes are substantially larger, allowing for asymmetric introgression and potential genetic swamping.3 Considerable research effort has therefore been devoted to monitoring hybridization in numerous taxa.4,5,6,7,8 Despite this endeavor, detection of hybridization remains one of the most challenging analytical problems in conservation genetics.9

Domestic animals and their wild ancestors diverged relatively recently, usually within the last thousands or tens of thousands of years,10 which does not provide sufficient time for the development of reciprocal monophyly across most of the genome. As a result, shared ancestral polymorphism may persist and mask signals of recent admixture.11 Moreover, easy-to-detect first-generation (F1) hybrids are typically rare and most often only detected under conditions of extensive or ongoing hybridization.3,12 In most natural systems, introgression proceeds through repeated backcrossing, progressively diluting domestic ancestry and further complicating the identification of admixed individuals.12 These challenges in detection of admixed individuals are also amplified by the fact that routine monitoring is commonly based on non-invasive or environmental samples, such as scats, hair, or saliva,13 which are often characterized by low quantity and quality of DNA. Consequently, despite substantial advances in high-throughput sequencing methodologies, classical monitoring approaches still predominantly rely on short tandem repeat (STR) loci.7 Together, these factors make detection of hybridization particularly sensitive to analytical assumptions and marker choice.

These challenges are especially evident in systems involving domestic-wild hybrids. Hybridization between gray wolf (Canis lupus) and domestic dog (C. l. familiaris) is one of the most extensively studied examples of anthropogenic hybridization in large carnivores and has been increasingly reported across Europe.14 It appears to be most prominent in southern European regions characterized by large populations of free-ranging dogs.15,16,17 However, the risk of hybridization is increasing even in areas with limited numbers of free-ranging dogs due to the ongoing demographic and spatial expansion of the wolf populations18,19 and the consequently growing opportunities for wolf-dog encounters. Wolf-dog hybridization has attracted significant attention across all sectors, from research through policymaking20,21 to media, which creates pressure for robust scientific data involving reliable hybrid determination approaches.22,23 In Europe, wolf-dog hybrids are not only a biological concern but also a governance issue, as their classification can affect legal status, reporting under conservation frameworks, and management decisions, which can sometimes have strong political implications. For example, in Slovakia, the alleged presence of hybrids was one of the main reasons for changing the protection status of wolves and re-establishment of their hunting.24 Recent policy-oriented syntheses emphasize that the lack of a shared, operational definition of wolf-dog hybrid and the lack of standardized diagnostic criteria hinder coordinated monitoring and policy implementation.23,25 As morphological determination of hybrids has proven to be insufficient,26 molecular methods are required to reliably distinguish wolves, dogs, and their hybrids.

Numerous approaches have been applied to detection of wolf-dog hybridization, using a wide range of genetic markers, most commonly STRs,27 segments of mitochondrial DNA (mtDNA),28 and single nucleotide polymorphisms (SNPs),29 in combination with diverse analytical frameworks and software (see Dziech7 for a review of used methods). While these studies have substantially advanced our understanding of hybridization dynamics, most have been performed on genetically rather homogeneous populations or have paid limited attention to the genetic structuring within the analyzed datasets. In contrast, research from areas where distinct wolf populations came into secondary contact is scarce, and evaluating hybridization in these regions may be challenging as population structure can impact hybridization testing and lead to erroneous results.30 Specifically, admixture among wild populations may be misinterpreted as domestic introgression, leading to inflated false-positive hybrid assignments or obscuring true hybrid individuals.29,31,32

One such area is Czechia, which lies at the crossroads of three wolf populations: the Central European lowland population, the Carpathian population, and the Alpine population (sensu Hindrikson et al.33)—creating a mosaic of individuals with varying degrees of population admixture.34,35,36 This complex population structure represents a particularly demanding scenario for hybridization testing. Hybridization risk can be expected to be meager due to the low number of free-ranging dogs in Czechia, and a single hybrid case has been reported so far.14 However, as the wolf population is rapidly expanding,19,37 the threat hybridization poses to its genetic integrity33,38 should not be underestimated. This situation therefore provides an opportunity to assess the reliability of commonly used hybridization detection approaches in the context of mixing among three genetically differentiated wolf lineages.

In this study, we aimed to (1) establish a conservative, monitoring-compatible workflow for wolf-dog hybrid detection that accounts for intraspecific wolf structure and does not require external reference wolf populations from outside the study area and (2) apply this workflow to the Czech wolf population to evaluate how population structure affects hybrid inference and to estimate the current frequency of recent wolf-dog hybridization. To address these objectives, we used STR markers commonly applied in genetic monitoring as the primary screening tool, followed by targeted confirmation with an SNP panel designed for wolf-dog admixture detection in individuals with ambiguous STR-based assignments.

Results

Amplification success, species determination, and final dataset

Over the course of four wolf years, a total of 2,413 samples were collected. Out of these samples, 1,245 (51.6%) met the quality threshold of at least 13 successfully amplified STR loci needed for an acceptable consensus genotype. The 1,168 unsuccessful samples did not reach the quality threshold or contained genetic information from multiple individuals. The proportion of successfully genotyped samples ranged across sample types, from kill swabs (39.8%) through hair samples (43.8%), urine samples (47.9%), scats (54.7%), other samples (57.6%), and blood samples (80%) to tissues (88.5%). Out of the 1,245 successfully amplified samples, 60 (4.8%) and 41 (3.3%) were identified in the clustering analysis as originating from foxes and dogs, respectively, and were excluded. A total of 380 individuals were retained after performing identity analysis and constituted the final dataset D1.

Sensitivity analysis

The analyses performed without prior accounting for population structure produced inconsistent assignment outcomes across methods. The STRUCTURE analysis of the dataset D1 and comparative dogs revealed the highest support for the presence of two clusters (K = 2), where a large proportion of the wolves clustered with dogs or showed varying degrees of wolf-dog admixture. Similarly, some of the results for higher numbers of clusters failed to reliably differentiate dogs and wolves (K = 2 and K = 3 in Figure 1A). NewHybrids also misclassified some of the wolves as pure dogs or assigned them to various hybrid categories (Figure 1B). On the contrary, dogs and wolves were clearly distinguished in the principal component analysis (PCA) plot (Figure 1C).

[图：Results of the analyses performed with the entire dataset D1 without addressing population structure (A) STRUCTURE resul]

Population structure

STRUCTURE analysis of the D1 dataset showed the strongest support for two population clusters (K2). Based on the results for K2, applying a 0.8 threshold allocated 275 individuals to one cluster (hereafter referred to as cluster C1), 84 to the other (hereafter cluster C2), and the remaining 21 showed mixed ancestry (Figure 2).

[图：Population structure within dataset D1 (A) STRUCTURE results for K = 2. Each column represents one of the 380 individual]

Threshold determination

The run of STRUCTURE with simulated genotypes showed the highest support for K2 in both population clusters. The highest qw score of the simulated BC1W was 0.915 for cluster C1 (Figure 3A) and 0.9108 for cluster C2 (Figure 3D). Using these thresholds of BC1W, 13 and 10 out of the 100 simulated wolf genotypes, respectively, were incorrectly identified as potential hybrids. None of the simulated hybrid categories were identified as pure wolves, indicating a 0% type II error rate.

[图：Results of the analyses performed with a dataset of simulated genotypes of parental (wolf and dog) and hybrid (F1, F2, B]

When the simulated genotypes were analyzed in NewHybrids, the highest pw score of the simulated BC1W was 0.9599 for cluster C1 (Figure 3B) and 0.9484 for cluster C2 (Figure 3E). Using these thresholds, 17 and 15 out of the 100 simulated wolf genotypes, respectively, were incorrectly identified as potential hybrids. None of the simulated hybrid categories were identified as pure wolves, reaching the desired 0% type II error rate.

Application of thresholds to the empirical data

The STRUCTURE analyses for our empirical data conducted for each population cluster separately alongside comparative dogs, showed the strongest support for the presence of two clusters corresponding to comparative dogs and tested individuals. The application of the 0.9150 threshold to the results for cluster C1 revealed 5 individuals among the pure cluster C1 members and 16 among the individuals of mixed population ancestry with assignment probabilities below the threshold (Figure 4A). When the 0.9108 threshold was applied to cluster C2, all pure members of cluster C2 met the threshold, while it was lower than the threshold in 10 individuals of mixed population ancestry (all ten also failed to meet the threshold when analyzed with cluster C1; Figure 4D). Combining the results for both population clusters, 15 individuals failed to fulfill the conditions to be considered pure wolves—10 from the group of individuals of mixed population ancestry and 5 from the pure cluster C1 members.

[图：Results of the analyses performed with a dataset of cluster C1 individuals (A–C) or cluster C2 individuals (D–F), admixe]

The application of the 0.9599 threshold to the NewHybrids results for cluster C1 revealed 8 of the pure cluster C1 members and 16 of the individuals of mixed population ancestry with posterior probabilities to belong to the wolf category below the threshold (Figure 4B). When the 0.9484 threshold was applied to the results for cluster C2, pw of all pure cluster C2 members met the threshold, while it was lower than the threshold in 12 individuals of mixed population ancestry (all 12 also failed to meet the threshold when analyzed with cluster C1; Figure 4E). Combining the results for both population clusters, 20 individuals could not be classified as pure wolves based on our criteria—12 from the individuals of mixed population ancestry group and 8 from the pure cluster C1 members.

All fifteen individuals that failed to meet the criteria to be classified as pure wolves based on the STRUCTURE analyses also showed posterior probabilities to belong to the wolf category below the threshold in NewHybrids. The remaining five animals that failed to be classified as pure wolves in NewHybrids met the threshold to be classified as pure wolves in STRUCTURE (Table S2).

Verification of hybrid status

The 20 individuals that could not be assigned to pure wolves were further investigated. Amplification of mtDNA was successful in 19 of the 20 animals, revealing six haplotypes previously described in Pilot et al.44 The most frequent haplotype was w1 (n = 10), followed by w6 (n = 5), w14 (n = 1), w19 (n = 1), w22 (n = 1), and w23 (n = 1).

STRUCTURE analysis of the SNP genotypes of these 20 individuals showed the highest support for the presence of two clusters, clearly distinguishing the dogs from wolves. Out of the 20 individuals, 19 clustered with the wolves with an assignment probability (qw) of 0.945 or higher (Figure 5A). Similarly, NewHybrids clearly distinguished dogs and wolves, and the same 19 individuals showed a 1.0 posterior probability of belonging to the wolf category (pw; Figure 5B). Consequently, these 19 animals were confirmed to be pure wolves, while the remaining individual (CW24_212) showed intermediate values between wolves and dogs and was considered a hybrid.

[图：Results of the SNP analyses performed with individuals that did not meet the thresholds to be classified as pure wolves ]

The detected hybrid (CW24_212) was only sampled once in March 2024 in the Krkonoše National Park in northern Czechia, and the scat sample yielded a genotype of 16 successfully amplified loci out of the 20 loci used (80%). Amplification of amelogenin was unsuccessful, and the individual’s sex could therefore not be determined. The individual displayed ambiguous results in the different analyses. It was determined to have mixed ancestry based on the initial STRUCTURE analysis assessing population structure within the dataset D1, with an assignment probability of 0.6988 to cluster C1 and 0.3012 to cluster C2. When analyzed with cluster C1, it showed an assignment probability of 0.5622 to the dog cluster in STRUCTURE, and NewHybrids indicated the highest posterior probabilities that the individual belongs to the F2 (0.42803) and to the BC1W (0.37498) hybrid categories. The individual was positioned on the periphery of the wolf cluster in the PCA. When analyzed with cluster C2, it was assigned with an assignment probability of 0.9774 to the dog cluster in STRUCTURE, and NewHybrids showed the highest posterior probability (0.9445) that the individual belongs to the pure dog category. The individual was positioned closer to the dog cluster in the PCA (Figure 4). The SNP analysis showed an assignment probability of 0.799 to the dog cluster in STRUCTURE and the posterior probability of 0.99141 to belong to the BC1D hybrid category in NewHybrids. The individual was outside the wolf and dog clusters in the PCA but was located closer to the dog cluster (Figure 5). The w19 D-loop haplotype was detected in this individual.

Discussion

In this study, we propose and evaluate a monitoring-oriented, conservative framework for wolf-dog hybrid detection in a genetically structured population (Figure 6). Czechia represents such a scenario, as it lies at the intersection of differentiated wolf populations,34,35 producing admixed wolf genotypes that may be misinterpreted as wolf-dog introgression if population structure is not accounted for.30 We therefore applied a sequential multi-step workflow that incorporates intraspecific structure, applies conservative simulation-calibrated thresholds, and verifies ambiguous cases using independent marker systems. This framework was designed to be compatible with routine monitoring and to avoid dependence on external “pure wolf” reference populations from outside the study area, while still enabling robust hybrid inference through conservative screening and targeted verification. By contrasting analyses that do not account for population structure with structure-specific analyses, our study demonstrates how population structure affects hybrid inference in contact zones and provides a practical workflow for estimating recent wolf-dog hybridization in monitoring datasets.

[图：Scheme of the workflow The figure was created in Inkscape.]

Our threshold strategy was intentionally conservative and designed to minimize type II error, i.e., the misclassification of admixed individuals as pure wolves. This objective is appropriate for early-warning monitoring where missing a true hybrid could have substantial conservation or regulatory consequences. Using HYBRIDLAB simulations, thresholds were derived from the most wolf-like hybrid class (BC1W), which is expected to be the most difficult to detect.27,45 As a consequence, the screening step is expected to flag some pure wolves as potentially admixed. This trade-off was evident in our simulated datasets, where 10%–20% of simulated pure wolves fell below the derived thresholds. On the other hand, none of the simulated F1, F2, or BC1W individuals were misclassified as pure wolves, achieving the desired 0% type II error. In applied monitoring, such a design inevitably increases the number of suspect individuals and may overestimate hybridization if screening outputs are interpreted without verification.

Although STRUCTURE and NewHybrids produced broadly concordant screening outcomes, NewHybrids identified a slightly larger number of individuals failing the criteria to be considered pure wolves than STRUCTURE. STRUCTURE and NewHybrids are both Bayesian, MCMC-based programs, but they evaluate hybridization in different ways, which can lead to varying numbers of individuals identified as potentially admixed. For each individual, STRUCTURE estimates the proportion of its genome best explained by each genetic cluster.46,47 Hybrid status is then inferred indirectly by applying an assignment threshold to membership in the wolf cluster. In contrast, NewHybrids is explicitly designed for hybrid classification. It evaluates how well an individual’s multilocus genotype matches the expected patterns of predefined categories and returns the posterior probability of belonging to each class.48 Importantly, thresholds are not directly transferable between studies and are not directly comparable between the two approaches. They depend strongly on the number and informativeness of markers and should ideally be calibrated using simulations (e.g., HYBRIDLAB) to control misclassification.27,45,49 In wolf-dog systems, simulation-based evaluations further show that increasing marker resolution typically increases assignment confidence and allows more stringent thresholds, while also highlighting that backcrossed and introgressed individuals remain among the most challenging classes to identify reliably.27,45 Consequently, NewHybrids may indicate a larger set of individuals as non-pure under conservative settings, but differences between the programs can also reflect residual structure within the reference parental groups.27

To verify individuals flagged as admixed by STR screening, we applied the wolf-dog ancestry-informative SNP panel developed by Harmoinen et al.,29 which was specifically designed to bridge the gap between genome-wide inference and routine monitoring based on non-invasive samples. Selected markers show strong allele-frequency differentiation between wolves and dogs and provide reliable discrimination of parental individuals and early-generation hybrids. In our workflow, this SNP confirmation step therefore served as a targeted, higher-resolution test applied only to the subset of individuals below STR-based thresholds, allowing us to distinguish true dog introgression from conservative STR “false positives” expected under a screening design that prioritizes not missing any hybrids. Importantly, this does not diminish the value of STRs in this context. Unlike the SNP panel developed explicitly for hybrid detection, STR genotyping in routine monitoring has to serve multiple purposes, from species verification, individual identification, and parentage testing to demographic and population structure analyses.50,51,52 STR panels are also cost-effective when time-sensitive management cases are needed (often within hours to a few days) and are robust to the low DNA quantity and fragmentation typical of non-invasive material.

Mitochondrial haplotypes provided an additional, independent line of evidence to contextualize the individuals flagged by STR-based screening. Because mtDNA reflects maternal inheritance, it cannot by itself confirm or exclude hybridization, but it is useful for assessing whether the maternal lineage of a suspect individual is consistent with the haplotype pool expected in the study area. In our dataset, all but one of the successfully amplified individuals carried typical wolf haplotypes previously reported. The most frequently detected haplotype in our study, w1, is found in the north-eastern European wolves44 and is the most common haplotype in the Central European lowland wolf population,35,53 whose distribution area extends to Czechia.34,35 The w6 and w14 haplotypes are typical for the Carpathian34,54 and Dinaric-Balkan55,56 populations, the former of which is represented in the eastern part of Czechia.34 The w22 haplotype is characteristic of the Alpine population,34,53 whose members have dispersed and admixed into the Central European lowland population in south-western Czechia.35 The w23 haplotype has been reported in both the Dinaric-Balkan55 and Alpine34 populations. These detected haplotypes imply the local origin of the individuals and do not suggest long-distance dispersal. Therefore, mtDNA results support the interpretation that most STR-flagged individuals represent wolves embedded within the regional gene pool rather than recent immigrants with atypical maternal ancestry. Although the w19 haplotype identified in the remaining individual is reported in wolves in eastern Europe,44 it is also commonly detected in domestic dogs.28

Out of the 380 individuals analyzed within this study, one was confirmed as a recent hybrid after SNP verification, indicating a low prevalence of recent wolf-dog hybridization in Czechia. This finding is in concordance with the low number of free-ranging dogs in the country, as high hybridization rates reported across Europe are commonly linked to large populations of such dogs15,16,17 and low hybridization rates are associated with their absence.57,58 Moreover, wolf hunting leads to the disruption of wolf population dynamics and a lack of conspecific mating partners, which can result in wolf-dog hybridization.59,60,61 The fact that the gray wolf is listed as a strictly protected species and its hunting is prohibited by the Act of the Czech National Council on Nature and Landscape Protection (No. 114/1992 Coll.)62 and the Law on Hunting (No. 449/2001 Coll.)63 could be reflected in the low hybridization rates in Czechia.

The single individual identified as a hybrid showed varying degrees of wolf-dog admixture across analyses but was generally more similar to the comparative dog cluster than to either of the wolf clusters. Similar cases of backcrossing into dogs have been reported from Mediterranean regions with high densities of free-ranging dogs,61,64,65,66 whereas such conditions are not typical for Czechia, where free-ranging dogs are controlled. The individual also carried the mtDNA haplotype w19, which has been reported in wolves but is also commonly detected in domestic dogs, a finding consistent with a dog-associated maternal lineage. Importantly, this individual was detected only once in a heavily visited tourist area of the Krkonoše region and was not redetected despite continued intensive monitoring in the broader area in the following seasons (until fall 2025). Such dog-associated individuals may arise through several pathways, including recent backcrossing into dogs, or, in some cases, human-mediated breeding; however, the available evidence does not allow us to distinguish among these possibilities in the present case. Taken together, these observations are consistent with an isolated, non-recurring event involving a dog-associated individual with wolf ancestry, and do not indicate ongoing hybridization or the establishment of a free-ranging admixed lineage in the region.

Conclusions

Our results demonstrate that wolf-dog hybrid inference based on routine STR monitoring can be strongly confounded by intraspecific population structure in contact-zone systems. Analyses that ignore population structure can misassign wolves as dogs or hybrids and inflate apparent admixture, whereas structure-specific calibration using simulations provides a transparent way to control error rates and to interpret “suspect” assignments conservatively. By combining cluster-specific thresholding (STRUCTURE and NewHybrids) with verification using independent marker systems (mtDNA and SNPs), our workflow offers a monitoring-compatible and cost-effective framework for hybrid assessment that does not rely on external reference wolf populations from outside the study area and explicitly incorporates independent validation of putative hybrids. Applying this framework, we found no evidence of ongoing wolf-dog introgression within the monitored wolf population; the only confirmed non-wolf assignment involved a dog-associated backcross individual detected once during the study period.

Our study supports a tiered approach in which STR genotyping retains high operational value for large-scale monitoring and rapid screening, while standardized SNP assays provide robust confirmation for the limited subset of ambiguous cases. Such reproducible pipelines are essential where hybrid classification has management consequences and where population structure and admixture among wolf lineages are expected to increase as recolonization proceeds.

Limitations of the study

Several limitations should be considered when interpreting hybridization estimates from monitoring datasets. Our workflow is optimized to detect recent wolf-dog hybridization and early-generation backcrosses under routine monitoring conditions, but low levels of introgression, especially late-generation backcrossing, remain difficult to distinguish from background variation without genome-wide data.9,45 In addition, although the SNP panel provides a standardized confirmation step for ambiguous cases, its performance still depends on DNA quality and completeness of genotypes obtained from non-invasive samples.13,29 Finally, because assignment thresholds are marker- and dataset-specific, they should be recalibrated if the marker set or the composition of comparative dogs changes substantially.49 Similarly, our analyses demonstrated that assignment patterns in domestic dogs also depended on the composition of the analyzed dataset and the particular wolf population used for comparison. This observation likely reflects the extensive shared genetic variation between wolves and dogs, as well as the heterogeneous demographic histories of domestic dog breeds, including founder effects, inbreeding, genetic drift, and varying levels of historical wolf ancestry. Because our dog dataset was designed to capture broad domestic dog diversity rather than to represent specific dog populations or breeds, we do not consider these patterns informative for assessing introgression into dogs.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Barbora Černá Bolfíková (bolfikova@ftz.czu.cz).

Materials availability

This study did not generate new unique reagents.

Data and code availability

Acknowledgments

We thank the many dedicated volunteers and professional staff of the Friends of the Earth Czech Republic Carnivore Conservation Program, the Nature Conservation Agency of the Czech Republic, Czech University of Life Sciences Prague, Mendel University in Brno, and the Krkonoše and Šumava National Park administrations for their efforts in collecting samples in the field and processing them in the laboratory. We thank Berardino Cocchiararo, Gregor Rolshausen, and the lab team of the Senckenberg Center for Wildlife Genetics for conduction of SNP analyses. The study was supported by the 10.13039/100014809Technology Agency of the Czech Republic under the Environment for Life Programme through the Genetic Monitoring of Wolves project (SS07010447), the Internal Grant Agency of the Czech University of Life Sciences Prague (IGA 20253125), Interreg Germany/Bavaria-Czechia 2021–2027 (BYCZ01-001), Interreg Germany/Saxony-Czechia 2014–2020 (100686869) and 2021–2027 (100400831), 10.13039/501100013855Ministry of the Environment of the Czech Republic (CZ.05.4.27/0.0/0.0/20_139/0013781), and Interreg Central Europe 2021–2027 (LECA CE0100170). Computational resources were provided by the e-INFRA CZ project (ID: 90254), supported by the Ministry of Education, Youth, and Sports of the Czech Republic.

Author contributions

Conceptualization, B.Č.B., P.H., E.C., and K.E.; methodology, B.Č.B., P.H., E.C., and K.E.; investigation, E.C., P.M., N.T., and J.Š.; data curation, E.C., P.M., N.T., and J.Š.; formal analysis, E.C., B.Č.B., K.E., and P.H.; resources, J.J., M.K., A.V., C.N., P.H., and B.Č.B.; funding acquisition, B.Č.B., P.H., J.J., M.K., A.V., and C.N.; supervision, B.Č.B. and P.H.; writing – original draft, B.Č.B., E.C., and K.E.; writing – review and editing, all authors.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

Method details

Study design and analytical framework

The analytical framework of this study reflects the conditions of routine genetic monitoring, in which genetic structure and admixture composition of the focal wolf population are typically unknown at the time of analysis. Although hybridization studies commonly rely on predefined reference wolf populations representing putatively “pure” wolves,15,27,66 such approaches assume that the ancestry of the monitored wolf population is known in advance. This assumption may not hold in recolonizing populations and secondary contact zones, where individuals can originate from multiple genetically differentiated wolf populations. Accordingly, no external reference datasets representing putatively “pure” wolf populations were used. All wolf samples were collected within Czechia and western Slovakia, and genetic data were generated as part of the national wolf monitoring program and Czech-Slovak transboundary research.73 Domestic dogs were included as a comparative group to enable hybridization inference. These comprised individuals whose samples were collected within the wolf monitoring and identified as dogs, as well as a number of dog breeds including pastoral and hunting breeds, which are most likely to be present in the area of wolf-dog co-occurrence. Czechoslovakian wolfdogs were also included as a breed with documented wolf introgression74,75,76 to control for potential intermediate genotypes in wolf–dog assignment analyses. This design allowed us to evaluate hybridization under realistic monitoring conditions without assuming prior knowledge of wolf population origin or relying on predefined reference wolf populations from outside the study area.

STR markers were used as a primary screening tool, reflecting their routine application in genetic monitoring, to assess wolf–dog admixture categories using analytical approaches described below. For individuals exhibiting ambiguous assignments, we applied an SNP hybridization panel specifically developed for wolf–dog hybrid detection as a targeted confirmation step (see Figure 6 for an overview of the workflow).

Sample collection

Samples used in this study were collected throughout four consecutive wolf years (2020/2021, 2021/2022, 2022/2023, 2023/2024), with each wolf year starting on May 1st of one calendar year and ending on April 30th of the following calendar year. Sample collection was carried out by trained volunteers, rangers, and zoologists from the Nature Conservation Agency of the Czech Republic, Friends of the Earth Carnivore Conservation Program, Faculty of Forestry and Wood Technology of the Mendel University in Brno, Faculty of Environmental Sciences of the Czech University of Life Sciences Prague, Šumava National Park, and Krkonoše National Park within the routine wolf monitoring. Sampling was predominantly non-invasive and included scats, hair, urine, estrus blood and saliva traces. Tissue samples originated from deceased animals and were provided by the Nature Conservation Agency of the Czech Republic, which is authorized to handle such material. Sampling in protected areas was conducted by authorized staff or in cooperation with the relevant protected-area administrations, and all sampling activities complied with applicable national regulations. No animals were captured, handled, or disturbed for the purpose of this study.

Samples were stored in 96% ethanol at −20°C and processed in the molecular genetics laboratory of the Faculty of Tropical AgriSciences, Czech University of Life Sciences Prague, which holds approval (No. CZ 11712934) for the storage and use of animal material in accordance with Section 48(1)(i) of Act No. 166/1999 Coll., on veterinary care and amending related laws, and relevant European legislation (Regulation (EC) No. 1069/2009 and Commission Regulation (EU) No. 142/2011).

A total of 2,413 samples were collected, comprising 1,496 scats, 430 kill-swabs, 267 urine samples, 121 hair samples, 40 estrus blood samples, 26 tissues, and 33 other samples (hair and tissue from prey, saliva, vomit, ticks). The comparative dog dataset consisted of previously published reference genotypes from known domestic breeds, used in earlier studies.34,35 These included pastoral and hunting breeds relevant to the study area, as well as Czechoslovakian wolfdogs, a breed with documented wolf ancestry.27

Sample processing

DNA was extracted from the samples using commercial kits and following the manufacturers’ protocols. In case of scats, extraction was performed using 180–220 mg of the top layer of the sample and the QIAamp Fast DNA Stool Mini Kit (QIAGEN) or NucleoSpin DNA Stool Kit (Macherey-Nagel). Urine samples were processed using the QIAamp Fast DNA Stool Mini Kit (QIAGEN) after being diluted in a 30:66:3 ratio with 96% ethanol and 3M sodium acetate and centrifuged at 5,500 × g and 4°C for 60 min. The Genomic DNA Mini Kit (Blood/Cultured Cell) (Geneaid) was used to extract DNA from venous blood, while the DNeasy Blood and Tissue Kit (QIAGEN) was used for coagulated blood. The DNeasy Blood and Tissue Kit (QIAGEN) or Genomic DNA Mini Kit (Tissue) (Geneaid) were used for all other sample types. Killswabs were detached from the shafts and shredded. Tissue samples were cut up into pieces of up to 25 mg, hair samples were trimmed to only include fragments with follicles, and dried blood from ticks was fragmented into small pieces. Saliva and vomit samples were centrifuged at 5,500 × g for 30 min, and DNA was extracted from the dried pellet. The amount of the lysis buffer was modified according to the sample volume and solubility. For elution, 30–100 μL of the respective elution buffer was used.

In all samples, a panel of 20 STR loci and the sex-determining gene amelogenin (Table S1) was amplified by polymerase chain reaction (PCR) using a multiple-tube approach77 and the Multiplex PCR Plus Kit (QIAGEN). The methodology was identical to that described by Hulva et al.,35 the only change being the removal of the FH2017 locus due to its low amplification success. Each sample was amplified in 2–5 replicates until a consensus genotype could be reached. Following the approach of Tumendemberel et al.,78 we required heterozygous and homozygous alleles to be present in at least two and three independent PCR replicates, respectively, to be included in the consensus genotype. Individuals that could not be confidently assigned as pure wolves based on STR genotypes (see details below) were subjected to additional analyses to assess their potential hybrid ancestry. In these samples, the left domain of the hypervariable region of the mitochondrial control region (D-loop) was amplified using the Thr-L15926 (5′-CAATCCCCCGGTCTTGTAACCC-3′) and DL-H16340 (5′-CCTGAAGTAGAAACCAGATG-3′) primers,67 following the methodology of Hulva et al.34 These individuals were also genotyped on a 96 SNP panel, specifically designed to detect wolf-dog hybridization.29 Amplification using a microfluidic chip approach was performed on a Biomark HD system (Standard BioTools, formerly Fluidigm).

Genotypes and removal of unsuccessfully amplified samples

The STR data were edited and binned in Geneious Prime 2025.2.2 (www.geneious.com) using the Microsatellite Plugin.68 Allele calls were inspected manually, and ambiguous peaks, peak imbalance, stutter artifacts, and loci with inconsistent amplification were treated as missing data. Homozygous genotypes were accepted only after repeated amplification of the same allele, whereas heterozygous genotypes required repeated observation of both alleles. Loci that could not be scored consistently were excluded from the consensus genotype. Samples were retained for individual identification and downstream analyses only when reliable consensus genotypes were obtained for at least 13 of the 20 STR loci, corresponding to ≥65% of the marker set.

Species determination

Species determination has been routinely conducted as part of national wolf monitoring since 2014, with analyses performed on an annual basis using Bayesian clustering implemented in STRUCTURE 2.3.4.46 This routine screening step aims to distinguish wolves from other canids and to exclude clear non-wolf samples prior to downstream analyses. At the end of each wolf year, all consensus genotypes of sufficient quality were analyzed alongside reference genotypes of foxes (N = 5), jackals (N = 2), dogs of various breeds (N = 36) and wolves representing the Central European lowland (N = 36), Carpathian (N = 36), and Alpine (N = 25) wolf populations. These three wolf populations represent the major source populations currently contributing to wolf occurrence and admixture dynamics in Czechia.34,35 In the context of the present study, these comparative genotypes served exclusively for routine species determination and quality control and were not used as reference populations in the subsequent hybridization analyses.

All comparative genotypes were constructed using the 20 STR markers utilized in this study, except for the Alpine wolves, whose genotypes only included twelve of the loci (see Table S1). The STRUCTURE analysis was run using the No admixture and Correlated allele frequencies models. The No admixture model was selected because the objective of this preliminary analysis was to assign samples to major canid groups rather than to infer admixture proportions, while the Correlated allele frequencies model was used because it has been shown to improve assignment performance among closely related populations and species.47 Identically to all STRUCTURE analyses performed in this study, five replicates for K 1–5 were run with a burn-in of 200,000 iterations followed by 800,000 iterations, the results of the five replicates were summarized in STRUCTURE Selector,69 and the most supported number of clusters was determined using the ΔK method.70 Individuals assigned to the dog or fox cluster were further excluded to focus downstream analyses on potentially admixed wolf individuals only.

Identification of unique individuals

Wolf multilocus genotypes were screened for matching individuals using identity analysis in Cervus 3.0.771 with a minimum of eight matching loci and a maximum of seven mismatches allowed. The allowed number of mismatches was set based on the optimal alleleMismatch value computed in the allelematch package (version 2.6.0)72 in R.41 All cases of matching genotypes were verified using available metadata (raw data, date and location of sample collection), and those confirmed to be identical with the final number of mismatches not exceeding four were used to construct consensus genotypes. For downstream hybridization analyses, one consensus genotype per individual was retained, corresponding to the first detection of that individual within one of the four analyzed wolf years (2020/2021–2023/2024). These genotypes constituted the dataset used for hybridization analyses (hereafter referred to as dataset D1).

Sensitivity analysis

Because hybridization testing in monitoring practice is frequently performed on pooled datasets without prior assessment of intraspecific structure, we first analyzed the entire dataset D1 as a baseline (“unstructured”) scenario. This analysis was not intended to identify hybrids per se, but rather to evaluate how strongly the omission of population structure affects wolf–dog assignment and hybrid classification outcomes and to provide a direct comparison with the structured workflow applied subsequently. The entire dataset D1 (N = 380) was analyzed in STRUCTURE alongside comparative dogs (N = 67) using the Admixture and Correlated Allele Frequencies models. These models were selected because contemporary European wolf populations experience secondary contact and admixture, and correlated allele frequencies improve the resolution of subtle population structure among closely related populations. The same data were analyzed in NewHybrids 1.148 using uniform priors and 1,000,000 iterations following a burn-in of 200,000 (identical settings to other NewHybrids analyses in this study). The results were additionally visualized using PCA implemented in the adegenet package (version 2.1.11)40 in R.41 This approach allowed us to quantify the extent to which population structure alone can generate apparent signals of admixture.

Population structure analysis

Prior to hybridization analysis, we characterized the population structure of the empirical wolf dataset D1 because admixture among differentiated wolf populations may generate assignment patterns resembling wolf–dog hybridization, as demonstrated in the sensitivity analysis described above. The purpose of this step was therefore not to classify individuals as pure or admixed, but to identify the major wolf genetic backgrounds present in the dataset and to define population-specific calibration groups for subsequent threshold calibration.

Population structure was inferred using STRUCTURE with the Admixture and Correlated Allele Frequencies models. The highest supported number of clusters (hereafter referred to as population clusters) was identified using ΔK. Individuals with an assignment probability (q) ≥0.8 to a single population cluster were considered representative of that population cluster. We emphasize that this threshold was not used to classify hybrid status or to identify genetically “pure” wolves. Instead, it served solely to identify individuals with relatively homogeneous population ancestry that could be used for simulation-based calibration of population-specific hybridization thresholds and to reduce the effect of admixture among wolf populations during subsequent hybridization analyses.

Individuals with (q) < 0.8 were considered to show mixed assignment, which could reflect admixture among wolf populations, wolf–dog hybridization, or both. Distinguishing among these possibilities was not the objective of this analysis and was only addressed in the subsequent hybridization analyses. Importantly, these individuals were not excluded from subsequent analyses, but analyzed independently within each inferred wolf population cluster. The q = 0.8 was applied as a pragmatic cutoff for pre-defining cluster membership in the population-structure step, in line with common practice in STRUCTURE-based assignment analyses, where individuals below this threshold are often interpreted as admixed.79,80,81

Identification of STRUCTURE and NewHybrids thresholds

In order to distinguish hybrids from pure wolves, we first defined assignment thresholds for classifying admixed individuals using simulated genotypes generated in HYBRIDLAB 1.1.49 HYBRIDLAB uses allele frequencies from empirically defined parental populations to generate multilocus genotypes representing pure and hybrid classes, allowing the expected genetic distributions of different hybrid categories to be evaluated. Because population structure was detected within the wolf dataset, simulations were performed separately for each population cluster. For each population cluster, a representative panel of dogs (N = 67) and wolves (N = 67) was selected, using only individuals classified as non-admixed, and maintaining equal sample sizes as recommended.49 The selection of wolves was fully randomized in small and geographically restricted population clusters. Larger population clusters covering broader geographic areas were divided into regions, and it was ensured that the number of individuals selected from each region reflected the proportional representation of the region in the population cluster. For each population cluster, we simulated 100 genotypes for both parental populations (wolves and dogs) and for each hybrid category: first-generation hybrids (F1), second-generation hybrids (F2), first-generation backcrosses to wolves (BC1W) and first-generation backcrosses to dogs (BC1D). These simulated datasets were subsequently used to define assignment thresholds based on the expected genetic overlap among pure and hybrid classes.

The simulated dataset for each population cluster was subsequently analyzed using STRUCTURE and NewHybrids. In STRUCTURE, the Correlated Allele Frequencies model was retained because wolves and domestic dogs represent recently diverged and historically interconnected gene pools with substantial shared genetic variation. This model has been shown to improve the inference of subtle population structure and admixture among closely related populations47 and is routinely applied in studies of wolf–dog hybridization and introgression.3,17,29 To define conservative thresholds for hybrid detection, we focused on minimizing type II error (misclassification of hybrid individuals as pure wolves). Because first-generation backcrosses to wolves (BC1W) represent the hybrid class most genetically similar to pure wolves and therefore the most difficult to detect, threshold values were derived from this category. Specifically, for STRUCTURE, the maximum assignment probability to the wolf cluster (qw) observed among simulated BC1W individuals was used as the threshold. For NewHybrids, the maximum posterior probability of assignment to the pure wolf category (pw) among simulated BC1W individuals was selected as the threshold. The derived thresholds were subsequently applied to the empirical dataset to identify individuals consistent with pure wolf ancestry and to flag individuals with potential hybrid origin. Additionally, the simulated individuals for each of the population clusters were visualized using a PCA plot.

Application of STRUCTURE and NewHybrids thresholds to empirical data

STRUCTURE and NewHybrids analyses were then performed on the defined empirical population clusters combined with comparative dogs (N = 67), using the settings described above. Individuals with a qw exceeding the cluster-specific threshold in the STRUCTURE analysis and whose pw simultaneously exceeded the cluster-specific threshold in the NewHybrids analysis were classified as pure wolves. Because individuals of mixed population ancestry were included in all population clusters, they were regarded as pure wolves, if their qw and pw were higher than the respective STRUCTURE and NewHybrids thresholds in at least one of the population clusters. The data for each population cluster, together with comparative dogs, were additionally visualized using a PCA plot.

Verification of individuals below thresholds

Individuals that failed to pass the thresholds set in STRUCTURE or NewHybrids to be considered pure wolves were subjected to additional analyses to confirm their status using mitochondrial D-loop and autosomal SNP markers. In the case of D-loop, the obtained sequences were edited in Geneious Prime, trimmed to 230 base pairs, and aligned with the reference sequences from Pilot et al.44 using the Geneious alignment algorithm. Haplotypes were determined and labeled according to the codes used in Pilot et al.44

The SNP alleles were scored using Standard BioTools SNP Genotyping Analysis Software 1.0.2. The tested individuals were combined with comparative dogs of diverse breeds (N = 15) and selected comparative pure wolves (N = 52) with varying degrees of admixture among population clusters. They were analyzed in STRUCTURE using the Admixture and Correlated allele frequencies models and in NewHybrids, and the data were also visualized in a PCA plot.