Assessing consequences of anthropogenic hybridization: Great Lakes mallard movement and resource selection

Abstract

Background

The release of domesticated animals is resulting in anthropogenic hybridization with their wild congeners and there is concern that these interactions are a risk to conservation. In North America, large-scale releases of domesticated mallards have continued for over a century, yet the potential ecological consequences are not well understood. Towards this, we assessed whether hybridization between game-farm and wild mallards was associated with variation in movement and resource selection behavior, which in turn can affect survival and reproduction.

Methods

We coupled genomic, telemetry, and remotely sensed spatial data for 592 female mallards sampled from 2021 to 2023 in the Great Lakes region of North America and tested for association between an individual’s level of domestic ancestry, their local and migratory movement, and resource selection during the non-breeding period. We used linear mixed models and general linear models to examine factors affecting daily movement distance and autumn migration probability, respectively. We implemented a step-selection function and conditional logistic regression to estimate relative selection strength of land cover types relevant to mallard resource use.

Results

We recovered 44% of GPS-marked birds as genetically wild, with the remaining 56% being feral game-farm or wild × game-farm mallard hybrids. Increased domestic ancestry was associated with shorter daily movement distances, decreased likelihood of migration, and higher selection for urban developed land cover.

Conclusions

Lower mobility and selection of urban areas by individuals with greater domestic ancestry warrant further investigation as factors potentially affecting survival and reproduction.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s40462-026-00657-7.

Background

The distribution and abundance of wildlife populations is directly linked to movement ecology through movement and resource selection behaviors, and indirectly via effects of these movement processes on demographic parameters such as survival and fecundity [1]. Individuals face tradeoffs when selecting resources, including balancing energy expenditure and resource acquisition with predation risk [2–5], resulting in demographic costs and benefits [6]. In addition to tradeoffs associated with movement decisions at the scale of an individual’s home range, migratory decisions can also lead to demographic consequences [7–10]. Knowledge of movement ecology is thus valuable for understanding population ecology and developing management strategies for wildlife conservation [11].

Land use change, pollution, climate change, and many invasive species are well-recognized anthropogenic factors threatening wildlife conservation [12–14]. Human driven changes to phenology and landscapes can alter animal movement distances, migration timing, and resource use [15–17]. Conversely, the consequences of introgressive hybridization between domesticated and wild congeners is a less studied conservation risk [18]. Contact between domestic and wild animals can occur incidentally through expansion of human development or habitat loss [19], and directly via release of domesticated organisms for species conservation or to augment harvested populations [20, 21]. Research has identified adverse behavioral and demographic responses resulting from hybridization with captive-reared fish [22–26] and wildlife [20, 27–29]. Further, technological advancements in genomics have revealed examples where domestic hybridization is widespread [21, 30] to the level of threatening the genetic identity of wild populations [31, 32].

Although comprehensive reviews of wildlife stocking are infrequent, particularly with regard to relating genetics to traits potentially affecting fitness [32], the capacity for the release of artificially selected animals to cause changes in morphology, movement behavior, and demography within wild populations is documented [33–37]. Long-term releases of domestic animals could negatively impact wild populations if gene flow transfers heritable domestic traits that are maladaptive for individuals within free-ranging populations. This relationship, though, between natural selection and population dynamics can be complex, as the type of natural selection and ecological context influence how phenotypic composition affects population change [38]. Quantifying the impact of hybridization between domestic and wild individuals can also be difficult as the additive effects of introgression on fitness may be slow to develop [39]. Thus, wild populations undergoing extensive domestic stocking and hybridization over many generations may be at greater risk.

The mallard (Anas platyrhynchos) is one of the most highly supplemented terrestrial vertebrates, with an estimated 200,000 to 500,000 domestically derived game-farm mallards released annually in eastern North America since the 1920s [40–42] and > 2.5 million annually released in Europe [43, 44]. Game-farm mallards were domesticated from wild mallards in Eurasia [45] and are now primarily released at private game-farms and shooting preserves for the purpose of sport harvest [42]. In North America, free-ranging mallards are delineated into three populations for harvest management, defined by breeding geography and administrative flyway (i.e., eastern, midcontinent, and western mallards; [46]). Game-farm mallard releases primarily occur(ed) in the Atlantic Flyway [42], where recent research revealed mallards in the U.S. portion are now a hybrid swarm (introgressive hybridization has progressed until nearly all individuals are hybrids) where about 90% of individuals are wild game-farm mallard hybrids [21, 45]. Annual surveys indicate eastern mallard breeding abundance declined from 1.4 million in 1998 to 1.2 million in 2023 [47], primarily driven by population decline in the northern U.S. portion of the Atlantic Flyway [48]. Similarly, the Great Lakes component of the midcontinent mallard population (Michigan, Minnesota, and Wisconsin survey areas) declined from an estimated 1.2 to 0.5 million during 2000–2023 [47]. Hybridization with game-farm mallards could be one factor contributing to the eastern and Great Lakes mallard population declines [45, 49, 50], and quantifying effects of hybridization on individual movement behaviors could reveal mechanisms by which hybridization affects population ecology.

Using early banding data, Lincoln [51] posited that captive-reared (i.e., raised in captivity and likely of domestic ancestry) mallards released into the wild had shorter dispersal distances and greater non-hunting mortality compared to wild North American mallards. More recent studies found that domestic mallards were typically harvested within 3 km of release locations [52] and had significantly shorter migration distances [33, 53] and smaller home range areas [54, 55] than wild mallards [56]. Mallards are facultative migrants and migratory decisions have been linked to proximate weather cues that affect energy expenditure and food availability [57, 58]. Therefore, migratory decisions likely evolved to enable selection of areas that ultimately enhance survival [7]. Correspondingly, if hybridization with game-farm mallards diminishes mobility, migration incidence, and/or the ability to select habitat types providing necessary resources, survival and/or fecundity could be reduced [59]. Genetics could therefore be a source of individual heterogeneity if variation in movement behavior associated with genotype leads to differences in individual survival or reproduction [60]. Thus, quantifying the effects of hybridization on movement, migration, and resource selection could aid understanding of factors affecting mallard population ecology.

Our objective was to examine non-migratory and migratory movement and non-breeding period resource selection of hen mallards in relation to their genotype as one component of a broader study to examine Great Lakes mallard movements and population dynamics. Although movement and habitat selection during the breeding period can affect demographic rates [61–63], cross-seasonal effects occur when processes or events in one time period influence the success of an animal in a subsequent time period. Cross-seasonal effects can drive heterogeneity in individual fitness [64] and are documented in waterfowl [65–68]. In addition to their importance in linking movement ecology to population ecology, non-breeding period movement, migration, and habitat selection information could have implications for mallard harvest and habitat management.

We predicted that feral game-farm and hybrid (wild × game-farm) mallards would move shorter distances and would be less likely to migrate than wild mallards, but that selection of land cover types would be similar for wild and feral/hybrid mallards. Whereas previous studies suggested captive-reared and released mallards were less mobile [33, 34, 52, 54, 55], recent genetic sampling of hunter-harvested mallards in Ohio found 65% of harvested birds had > 10% domestic ancestry [69], suggesting hybrid mallards use areas open to hunting. Evaluating whether movement and resource selection differs between wild and feral/hybrid mallards could aid understanding of potential ecological consequences of hybridization. Finally, the extensive stocking history and monitoring of mallards in North America present an opportunity to use this system as a case study of anthropogenic hybridization.

Methods

Study area

We captured and marked female mallards with GPS-GSM transmitters in Bird Conservation Regions (BCRs; [70]) 12 (Boreal Hardwood Transition), 22 (Eastern Tallgrass Prairie), and 23 (Prairie Hardwood Transition) in the Great Lakes states of Michigan, Wisconsin, Ohio, Indiana, and Illinois, USA (hereafter, Great Lakes region; Fig. 1). Bird Conservation Regions are landscape planning units comprised of similar ecosystems and bird communities and are relevant for landscape conservation planning in the Upper Mississippi/Great Lakes (UMGL) Joint Venture (JV) region [71]. Mallards were the most abundant nesting duck in the Great Lakes region [46] and predicted nesting densities were greatest in the Prairie Hardwood Transition (BCR 23) [72]. We monitored movements of GPS-marked mallards within North America, which occurred in an area bounded by approximately 51.5⁰N and 32⁰N, and 75.2⁰W, and 99.8⁰W.

Fig. 1.

Capture and GPS-GSM transmitter deployment locations for female mallards captured in Michigan, Wisconsin, Illinois, Indiana, and Ohio, 2021–2023 in relation to banding site type (urban or rural) and Bird Conservation Region (BCR) 12 (Boreal Hardwood Transition), 22 (Eastern Tallgrass Prairie), and 23 (Prairie Hardwood Transition)

Mallard capture and data collection

We captured ducks using baited traps, rocket- and spring-propelled nets, handheld nets, and via night-lighting from 4 March – 4 October annually during 2021, 2022, and 2023. We aged ducks as after hatch year (adult), hatch year (juvenile capable of flight) or local (juvenile incapable of flight) and sexed birds via plumage characteristics [73]. We banded mallards with a size 7 United States Geological Survey (USGS) aluminum leg band. We attached 20 g Ornitela GPS-GSM (global system for mobile communications) transmitters (OrniTrack-E20 4GCT C48; Ornitela, Vilnius, Lithuania) dorsally to female mallards via 4.5 mm-wide elastic straps. Transmitters were attached with two separate elastic loops in 2021 and via an X-shaped design consisting of a single piece of elastic in 2022 and 2023. We weighed each hen mallard to the nearest 10 g and attached transmitters only to individuals which weighed greater than 700 g so transmitters comprised < 3% of an individual’s body mass (mean = 1.9%). Transmitters were programmed to record a location every 30 min, 2 h, or 4 h when battery charge was > 50%, < 50% and > 25%, or < 25%, respectively, and transmitted data once every 24 h when connected to cellular networks.

The number of transmitters attached to birds in each BCR was chosen to be roughly proportional to the total estimated breeding mallard abundance within each BCR [72]. We classified the capture sites where transmitters were deployed as urban if the proportion of developed land cover (low, medium, or high intensity developed, or developed open space as classified by the 2021 National Land Cover Database [NLCD; [74]) within a 7 km (mean of mallard maximum daily net displacement [49]) radius of the capture site was > 0.5, or rural if ≤ 0.5. Whereas transmitter deployment locations classified as urban or rural could contain cover classes associated with typical mallard habitat (e.g., emergent herbaceous wetlands, forested wetlands, open water, agriculture) in the 7 km radius, the proportion of developed land cover differed.

Finally, we drew ~ 0.1 ml of blood from the medial metatarsal vein of each GPS-marked mallard. Ducks were released immediately after processing at capture locations. Approval to capture, band, and attach transmitters was provided by the Michigan State University institutional animal care and use committee (IACUC) permit PROTO202100046 and USGS Bird Banding Laboratory permit 03110.

DNA extraction, sequencing, and genetic ancestry analyses

We extracted genomic DNA from blood samples using a DNeasy Blood & Tissue kit following the manufacturer’s protocols (Qiagen, Valencia, CA, USA). DNA quality was visually assessed on a 1% agarose gel to ensure high molecular weight bands. We followed ddRAD-seq (double-digest restriction site associated DNA sequencing) library protocols outlined in DaCosta and Sorenson [75] and Lavretsky et al. [76]. In short, genomic DNA was enzymatically fragmented using SbfI and EcoRI restriction enzymes, and Illumina TruSeq compatible barcodes ligated for future de-multiplexing. The barcode-ligated fragments were then size-selected using optimized double-sided bead selection protocols [77]. Libraries were then quantified with a Qubit 3 Flourometer (Invitrogen, Carlsbad, CA, USA) and pooled in equimolar amounts and sent to Novogenetics LTD (Sacramento, California, USA) for 150 base-pair, single-end chemistry sequencing on an Illumina HiSeq X. Raw Illumina reads were de-multiplexed using the ddRADparser.py script of the BU ddRAD-seq pipeline [75] based on perfect barcode/index matches. Comparable sequences from previously published wild and domestic mallards (i.e., known game-farm and Khaki Campbell’s) were included and served as respective references [21, 78–80]. For each sample, we first trimmed or discarded sequences of poor quality using TRIMMOMATIC [81], and then the remaining quality reads aligned to a chromosomal-level reference wild mallard genome [82] using the BURROWS WHEELER ALIGNER v. 07.15 [83]. Samples were then sorted and indexed in SAMTOOLS v. 1.7 [81] and combined using the mpileup function with the following parameters “-c –A -Q 30 -q 30.” All steps through mpileup were automated using a custom Python script [21]. Next, we used VCFTOOLS v. 0.1.15 [84] to filter variant call format (VCF) files for any base-pair missing > 10% of samples that also included a minimum base-pair depth of 5X (i.e., 10X per genotype) and quality per base PHRED scores of ≥ 30.

All nuclear population structure was based on independent bi-allelic ddRAD-seq autosomal single nucleotide polymorphisms (SNPs), and without using a priori assignment of individuals to populations or species. The final dataset was obtained by using VCFTOOLS [84] to first extract bi-allelic SNPs, and then PLINK v.1.9 [85] to filter for singletons (minimum allele frequency: 0.0014), any SNP missing ≥ 10% of data across samples, as well as any SNPs found to be in linkage disequilibrium (LD). We randomly excluded all but one SNP for any positions found to be in significant LD (r² > 0.5). Population structure was first visualized using a principal components analysis (PCA) as implemented in PLINK v.1.9 [85]. Next, the program ADMIXTURE 1.3 [86] was used to attain per sample maximum likelihood estimates of population assignments for each individual, with datasets formatted for the ADMIXTURE analyses using PLINK v.1.9 [85], and following steps outlined in Alexander and Lange [87]. Given previous work consistently recovering optimum K population models of 3 when assessing mallards, game-farm mallards, and Khaki Campbells [21, 45, 69], ADMIXTURE analyses were run for population models of K of 2 and 3 with a 10-fold cross validation, incorporating a quasi-Newton algorithm to accelerate convergence [88]. Each analysis used a block relaxation algorithm for point estimation and terminated once the change in the log-likelihood of the point estimations increased by < 0.0001. Finally, standard deviations around each point estimate were calculated based on 1,000 bootstrap replicates. Ancestry assignments and their standard errors were used to recategorize samples, with individuals having ancestry assignment plus standard-error ≥ 0.92 as parental (i.e., wild mallard, game-farm mallard, or Khaki Cambpell) and all others as hybrids. For hybrids, we attempted to assign generational class [see 69]. Although we acknowledge that demarcated generational class may not reflect an individual’s true generational age as there are multiple combinations that result in the same general genomic proportion (e.g., 60:40 mates with a 40:60), generational classes still provide appropriate categorization of individuals given expected ancestry proportions (i.e., F1 = 50:50, F2 = 75:25). The expected generational backcross categories are provided for conceptual reference. Assignment probabilities are also interpretable as an estimate of the proportion of an individual’s genes that are of wild ancestry, and thus we considered proportion wild ancestry as a continuous covariate in movement analyses as the primary covariate to represent an individual’s genotype.

Movement

We performed data preparation and analyses in Program R [89] and used the move package (version 4.2.6) [90] to retrieve GPS data from Movebank. We defined the non-breeding period as 16 August to 29 February based on Coluccy et al. [91] and the earliest date of nest incubation observed in this study (13 March; [49]), and included data collected during this timeframe from 2021 to 2024 in movement and resource selection analyses. We censored GPS locations where a satellite fix was not obtained or only one satellite was successfully contacted, and/or locations with a horizontal dilution of precision ≥ 5 [92]. We monitored individuals from marking to reported harvest by a waterfowl hunter, mortality or transmitter loss indicated by transmitter temperature and accelerometer data and verified by field observation when possible, or transmitter failure (transmitter stopped sending data with no indication of mortality). We excluded location data from individuals that provided GPS data for < 7 days after release [5, 93].

We modeled the mean sum of step lengths (i.e., movement distance) in each 24-hour period during non-migratory movements in relation to the estimated proportion of wild mallard ancestry within individuals using linear mixed models in the lme4 package (version 1.1–38) [94]. First, we inspected the sampling rate of GPS fixes and excluded six individuals with a median fix rate > 30 min (highest programmed location frequency) because estimation of step lengths assumes the length of time between pairs of consecutive locations is equal. We resampled the remaining data to a location every 30 min to ensure time between each location was consistent and only used successive locations to calculate step lengths. We calculated step lengths as the linear distance between consecutive GPS locations using the package amt (version 0.3.0.0) [95]. Step lengths provide an estimate of movement distance over a given time period and do not require the assumption that individuals are range resident (plots of variance in position over time reach an asymptote, indicating the amount of space used eventually becomes constant), an assumption needed to estimate home range area [96]. We excluded step length sums collected during autumn or spring migration that were > 30 km [58]. Thus, the daily sum of step lengths provided an index of total non-migratory distance moved in each 24-hour period and facilitated comparison of local daily movement distances. Inspection of the distribution of step length sums and QQ (quantile-quantile) plots indicated a log transformation of daily movement distance was appropriate because residuals were not normally distributed. We modeled log-transformed daily step length sum (daily movement distance) as a function of the individual proportions of wild ancestry and capture site type (urban or rural) with a random intercept of mallard ID to account for random individual variation. We included effects of site type and proportion wild ancestry as covariates on daily movement distance, as well as a null model, resulting in three candidate models. We ranked models using Akaike’s Information Criterion [97] adjusted for small sample size () using the package MuMIn (version 1.48.11) [98]. We calculated predicted daily movement distances from the top-ranked model treating proportion wild ancestry as a continuous variable. To predict how backcrossing of hybrid with wild mallards influences daily movement distance, we also fit a model treating genotype as a discrete predictor where the generational categories of feral, first generation (F1), < F2, F2, < F3, F3, and wild corresponded to 0 to < 0.43, 0.43 to < 0.63, 0.63 to < 0.72, 0.72 to < 0.82, 0.82 to < 0.86, 0.86 to < 0.92, and 0.92 to 1 proportion wild ancestry, respectively (also see [69]). Note that the < F2 and < F3 categories included hybrids that did not fit into an ordered generational category (i.e., unknown generational backcrosses). We used the emmeans package (version 2.0.1) [99] to calculate predicted daily movement distance for each generational category.

We defined autumn migration as a relocation with a total displacement ≥ 30 km [58] during 16 August to 29 February. We estimated the probability of autumn migration, conditional on an individual being alive on 1 January (to ensure mortality did not preclude the ability to observe migratory behavior), in relation to proportion wild ancestry and capture site land cover (urban or rural) using general linear models with a binomial response (migrated, did not migrate) and logit link [94]. We excluded 11 of 385 individuals which provided migration data for 2 or more years from this analysis to avoid repeated measures on individuals. Candidate models included all combinations of covariates with additive effects of proportion wild ancestry, capture site land cover, capture latitude, and capture state (MI, WI, OH, IN, or IL). We examined covariates for multicollinearity prior to modeling. We ranked models using via the package MuMIn [98].

Resource selection

To compare resource selection across mallard genotypes, we implemented a step-selection function analysis to account for resource availability given mallard movement characteristics [100]. Modeling movement metrics such as step length and turning angle (angle measured between the previous and current direction of movement) within the time series framework of GPS data provides a method to account for spatial and temporal autocorrelation [101] of animal locations while identifying which resources are biologically available to individuals [100, 102]. This approach provides more realistic estimates of availability than using randomly sampled locations because movement patterns affect which resources are available and availability affects how animals move [103]. We merged land cover classes in the 2021 USGS NLCD [74] and 2020 Land Cover of Canada (LCC) database [104], clipped to the extent of mallard GPS locations, and reclassified the raster into eight cover classes relevant to mallard resource selection. Land cover classes included emergent herbaceous wetland, forested/woody wetland, open water, developed (low, medium, and high intensity developed and developed open space), grassland (grassland/herbaceous and hay/pasture), forest (upland deciduous, evergreen, needleleaf, and mixed forest), agriculture (cultivated cropland), and other (barren, scrub shrub, and shrubland).

We first used Hidden Markov Models (HMMs) to assign unobserved behavioral states [105–107] using observed step lengths and turning angles. We fit HMMs in the moveHMM package (version 1.11)) [108] using a gamma distribution for step length (meters) and a Von Mises distribution for turning angle (radians). We specified HMMs to include three behavioral states because we expected step length and turning angle distributions could be used to identify resting or within-resource patch movement, local movement between resource patches, and migratory movements. HMM covariates included all combinations of additive effects of hour of day, proportion wild ancestry, and age, which were hypothesized to affect movement characteristics among individuals and over time [109]. We ranked HMMs using AIC [97] and used the top model to assign the most probable behavioral state to each location. We used the distribution of step lengths and turning angles associated with each behavioral state to randomly sample six available locations for every observed (used) location [109, 110]. Then we extracted the land cover type at each used and available point via the raster package [111].

We estimated the relative selection strength [112] of land cover types using two-step conditional logistic regression in the TwoStepCLogit package (version 1.2.6) [113]. Relative selection strength (RSS) is analogous to a risk ratio, and in the case where resource units such as potential habitat types are categorical, RSS is the probability of selecting a given habitat type divided by the probability of selecting a reference cover type [112]. Probability of selection is the probability an individual selects a resource unit given the resource unit is encountered [114], with selection quantified as an observed GPS location within a given resource unit. In wildlife telemetry, researchers typically only have data to estimate which resource units were used and which units were available, rather than which units were used and unused. When data are in the used versus available format and resource covariates are discrete, the exponential resource selection probability function is inestimable, thus only permitting use of a resource selection function to estimate relative selection and not probability of selection [115]. We used models with fixed effects of land cover type and random effects of mallard ID to account for differences in habitat selection between individuals [109, 110]. We set open water as the reference category and calculated odds ratios of selection, or RSS. Relative Selection Strength estimates indicate whether each land cover type was less selected (odds ratio < 1 with confidence intervals non-overlapping 1), similarly selected (confidence intervals overlap 1), or more selected (odds ratio > 1 with confidence intervals non-overlapping 1) than open water. Importantly, odds ratios of selection allow for comparison of the strength of land cover type selection between mallards of different genotypes, informing if and how selection of land cover types differs by genotype. We considered three genotypic categories in this analysis which corresponded to observed differences in daily movement distance and migration incidence: early generational hybrids (Feral and F1 hybrids; <63% wild), late generational hybrids (F2 and F3 hybrids; 63% to 92% wild), and wild (> 92% wild).

Results

We attached GPS transmitters to 592 hen mallards in 2021 (n = 194), 2022 (n = 241), and 2023 (n = 157). Mallards were captured using baited traps (n = 331), rocket- and spring-propelled nets (n = 166), hand-held nets (n = 55), and night-lighting (n = 40). Transmitters were deployed across all study area states, BCRs, and mallard ages (Table 1). Individual mallards were monitored for an average of 189 days (range: 2–1,037 days) during the study. We excluded 24 individuals from analyses for which there were < 7 days of GPS monitoring data available following release. Locations collected from 16 August through 28 February in 2021–2022 (n = 744,364), 2022–2023 (n = 1,181,940), and through 29 February in 2023–2024 (n = 1,035,939) were extracted for analysis. After censoring 553 locations (227 in 2021–2022, 147 in 2022–2023, and 179 locations in 2023–2024), there were 2,968,883 locations used for non-breeding period analyses.

Table 1.

Number of hen mallards marked with GPS-GSM transmitters by bird conservation region (BCR), state, and age (AHY = after hatch year; HY = hatch year; L = local) in the Great Lakes region from 2021–2023. BCR 23 is the Prairie Hardwood Transition, BCR 22 is the Eastern Tallgrass Prairie, and BCR 12 is the Boreal Hardwood Transition

	Michigan			Wisconsin			Ohio			Indiana			Illinois
	AHY	HY	L	AHY	HY	L	AHY	HY	L	AHY	HY	L	AHY	HY	L	Total
BCR 23	127	49	3	80	51	10	61	30	0	12	11	2	0	2	0	438
BCR 22	-	-	-	-	-	-	2	2	0	28	9	0	26	10	0	77
BCR 12	12	36	1	7	18	3	-	-	-	-	-	-	-	-	-	77
Total	139	85	4	87	69	13	63	32	0	40	20	2	26	12	0	592

Genetic ancestry

Five hundred and ninety (of 592) samples provided usable genomic material. There were 33,216 (of 34,230) independent bi-allelic ddRAD-seq autosomal SNPs retained for nuclear population structure analyses. The first two principal components (Fig. 2) explained 26.8% of the variance, and clearly separated reference wild mallards, game-farm mallards, and Khaki Campbells, with samples from this study distributed between wild reference mallard set and game-farm mallards (Fig. 2). Given no evident clustering with Khaki Campbells, we attained assignment probabilities by analyzing a dataset excluding them and plotted ADMIXTURE assignment probabilities under a K population of two. In total, we recovered 44% of GPS-marked hen mallards as genetically wild, whereas 56% were classified as feral or wild x game-farm mallard hybrids. Total sample sizes for feral, F1, < F2, F2, < F3, F3, and wild individuals were 22, 37, 63, 61, 60, 92, and 255, respectively. The percentage of wild mallards captured in each state ranged from 8.1% in Indiana to 63.9% in Wisconsin. Generally, fewer hybrids were captured in the northern and western portions of the study area (Table 2). Of mallards captured at rural sites (n = 353), 58.4% were wild and 41.6% were feral or hybrids. Mallards captured at urban sites (n = 237) were comprised of 21.9% wild and 78.1% feral or hybrid individuals.

Fig. 2.

Principal component plot (A) of the first 2 principal components explaining variation in nuclear ddRAD sequencing results for reference wild (WMA), domestic game-farm mallard (GFM), and Khaki Campbell samples and genetic samples collected from female mallards marked with GPS-GSM transmitters in Michigan, Wisconsin, Illinois, Indiana, and Ohio, 2021–2023 (Study Samples; grey circles). ADMIXTURE assignment probabilities (B) for individual Great Lakes mallard study samples by capture state, and wild and game-farm mallard reference samples

Table 2.

Sample size (n) and the proportion of wild (≥ 0.92 wild ancestry) and hybrid (< 0.92 wild ancestry) mallards captured by banding site type (urban or rural) within Illinois (IL), Indiana (IN), Michigan (MI), Ohio (OH), and Wisconsin (WI) from 2021–2023

		Proportion		n
State	Site Type	Wild	Hybrid	Wild	Hybrid
IL	Rural	0.684	0.316	13	6
	Urban	0.111	0.889	2	16
IN	Rural	0.143	0.857	3	18
	Urban	0.049	0.951	2	39
MI	Rural	0.492	0.508	60	62
	Urban	0.305	0.695	32	73
OH	Rural	0.446	0.554	33	41
	Urban	0.238	0.762	5	16
WI	Rural	0.829	0.171	97	20
	Urban	0.212	0.788	11	41

Movement

The top-ranked model predicting log-daily movement distance indicated a positive linear effect of proportion wild ancestry () and received all the model weight (Table 3). The proportion of the variance explained by the fixed and random effects (conditional R²) was 0.29. We report estimates from the top-ranked model. Proportion wild ancestry was positively associated with average daily movement distance (Fig. 3). The model treating genotype as discrete categories corresponding to estimated generational assignments also predicted an increase in daily movement distance as generational backcrosses approached wild (Fig. 4). Feral, F1, and < F2 hybrids were predicted to move significantly shorter daily distances than were wild mallards (non-overlapping Cis). The 95% Cis for F2, < F3, and F3 hybrid groups overlapped the 95% Cis for wild mallards. Wild mallards were predicted to have 2.52, 1.72, 1.32, 1.20, 1.23, and 1.02 times the daily movement distance compared to feral, F1, < F2, F2, < F3, and F3 hybrids, respectively.

Table 3.

Model selection table using Akaike’s Information Criterion adjusted for small sample size () for linear mixed models predicting log-transformed daily movement distance (sum of step lengths) as a function of proportion wild ancestry (pwild) and capture site type (site; urban or rural) for female Great Lakes mallards marked with GPS-GSM transmitters in Michigan, Wisconsin, Illinois, Indiana, and Ohio from 2021–2023. Table includes model loglikelihood (logLik), difference in from the top-ranked model (Δ), model weight (ω), and number of parameters (K)

Model	logLik	AICc	ΔAICc	ω	K
pwild	-92897.0	185802.1	0.0	1.00	4
site	-92925.5	185859.0	57.0	0.00	4
Null	-92938.7	185883.3	81.2	0.00	3

Fig. 3.

Predicted average daily movement distance (solid line) and 95% confidence intervals (dashed lines), and average individual daily movement distance (points) for female mallards marked with GPS-GSM transmitters in Michigan, Wisconsin, Illinois, Indiana, and Ohio, 2021–2023. Point color corresponds to estimated ancestry assignment where feral, first generation (F1), < F2, F2, < F3, F3, and wild corresponded to 0 to < 0.43, 0.43 to < 0.63, 0.63 to < 0.72, 0.72 to < 0.82, 0.82 to < 0.86, 0.86 to < 0.92, and 0.92 to 1 proportion wild ancestry, respectively

Fig. 4.

Predicted mean daily movement distance (points) and 95% confidence intervals (error bars) in relation to mallard ancestry assignment for mallards marked with GPS-GSM transmitters in Michigan, Wisconsin, Illinois, Indiana, and Ohio, 2021–2023. Color corresponds to estimated ancestry assignment where feral, first generation (F1), < F2, F2, < F3, F3, and wild corresponded to 0 to < 0.43, 0.43 to < 0.63, 0.63 to < 0.72, 0.72 to < 0.82, 0.82 to < 0.86, 0.86 to < 0.92, and 0.92 to 1 proportion wild ancestry, respectively

The top model predicting autumn migration probability included effects of proportion wild ancestry, banding site type (urban or rural), and capture latitude (Table 4). Migration probability was positively associated (Fig. 5) with proportion wild ancestry () and with capture latitude (). Mallards marked at urban sites were less likely to migrate than those marked in rural areas (). For rural capture sites at the mean latitude (42.74⁰N), wild mallards were 1.97, 1.37, and 1.16 times more likely to migrate than F1, F2, and F3 hybrids, respectively. Mallards that migrated from mid-August through February primarily went to the southern Great Lakes states or mid-latitude states in the Mississippi and Atlantic flyways, with few individuals moving farther south than Tennessee (Supplemental File 1).

Table 4.

Model selection table using Akaike’s Information Criterion adjusted for small sample size () for general linear models (binomial response and logit-link function) predicting autumn migration probability in relation to proportion wild ancestry (pwild), capture site type (site; urban or rural), capture latitude (lat), and capture state for female mallards marked with GPS-GSM transmitters in Michigan, Wisconsin, Illinois, Indiana, and Ohio, 2021–2023. Table includes model loglikelihood (logLik), difference infrom the top-ranked model (Δ), model weight (ω), and number of parameters (K)

Model	logLik	AICc	ΔAICc	ω	K
site + pwild + lat	-192.2	392.48	0.00	0.70	4
lat + site	-194.2	394.51	2.03	0.25	3
site + pwild	-196.2	398.53	6.05	0.03	3
state + site	-194.6	401.45	8.97	0.01	6
site	-200.4	404.78	12.30	0.00	2
lat + pwild	-201.0	408.15	15.67	0.00	3
pwild	-204.9	413.79	21.31	0.00	2
state + pwild	-200.9	413.95	21.47	0.00	6
lat	-209.7	423.48	31.00	0.00	2
state	-209.3	428.81	36.33	0.00	5
age	-218.9	441.85	49.38	0.00	2

Fig. 5.

Predicted autumn (16 August – 29 February) migration probability (lines) and 95% confidence intervals (shaded areas) as a function of capture latitude (panels), capture site land cover (urban or rural), and proportion wild ancestry for female mallards marked with GPS-GSM transmitters in Michigan, Wisconsin, Illinois, Indiana, and Ohio, 2021–2023. Vertical lines on x-axis (rug) denote proportion wild ancestry and capture site types of individual GPS-marked female mallards

Resource selection

The top-ranked HMM describing movement characteristics included three behavioral states and effects of hour of day, proportion wild ancestry, and age (Table 5). We used this model to randomly sample available land cover types from the distribution of step lengths and turning angles associated with each individual’s movement state at each observed GPS location (Supplemental Files 2 and 3). Open water and emergent herbaceous wetland were the most-selected land cover types across all mallard genotypes (Table 6; Fig. 6). The aquatic cover types of open water (used as the reference category), emergent herbaceous wetland, and forested wetland were generally more highly selected than were terrestrial cover types (all other land cover types). The ‘Other’ category (including barren land, scrub-shrub, and shrubland) was highly selected by early generational hybrids. Early generational hybrids had significantly greater selection for urban developed land cover than did late generational hybrids and wild mallards. Early generational hybrids were 1.9 times more likely and late generational hybrids were 1.2 times more likely to select developed land cover than were wild mallards. Early generational hybrids were significantly less likely to select agriculture than were late generational hybrids and wild mallards. Trends in RSS estimates for urban land cover followed trends in the proportion of observed (used) locations in urban land cover (Supplemental File 4), with early generational hybrids having the highest proportion of locations in urban land cover and the lowest use of emergent and forested wetlands. Therefore, as the proportion of game-farm ancestry increased, individuals were more likely to select and spend a greater proportion of time in developed areas. Further, feral/hybrid mallards were less likely to use emergent herbaceous and forested wetlands than were wild mallards (Supplemental File 4).

Table 5.

Hidden Markov Model (HMM) selection table using Akaike’s Information Criterion (AIC) for models describing step lengths and turning angles associated with unobserved movement states and hour of day, proportion wild ancestry (pwild), and age (juvenile, adult) effects for female mallards marked with GPS-GSM transmitters in Michigan, Wisconsin, Illinois, Indiana, and Ohio, 2021–2023. Table includes model loglikelihood (logLik), difference in AIC from the top-ranked model (ΔAIC), and model weight (ω)

Model	logLik	AIC	ΔAIC	ω
Hour + pwild + Age	11440733.1	22825345.0	0.0	1.0
Hour + pwild	11417106.5	22826619.5	1274.5	0.0
Hour + Age	11436896.4	22831759.0	6414.0	0.0
Hour	11439479.8	22834264.9	8919.9	0.0
pwild + Age	11413277.8	22872599.4	47254.3	0.0
pwild	11415847.5	22873844.8	48499.8	0.0
Age	11436267.7	22879011.6	53666.6	0.0
Null	11412634.5	22881506.2	56161.1	0.0

Table 6.

Resource selection coefficients and 95% confidence intervals (CI) for female mallards marked with GPS-GSM transmitters in Michigan, Wisconsin, Illinois, Indiana, and Ohio, 2021–2023. Genotype categories include the range of proportion wild ancestry in parentheses

Genotype Category	Land Cover Type	Selection Coefficient (95% CI)
Wild (> 92%)	Herbaceous Wetland	0.05 (-0.01,0.11)
Wild (> 92%)	Forested Wetland	-0.36 (-0.44,-0.27)
Wild (> 92%)	Agriculture	-1 (-1.11,-0.88)
Wild (> 92%)	Grassland	-0.65 (-0.77,-0.53)
Wild (> 92%)	Forest	-1.34 (-1.48,-1.2)
Wild (> 92%)	Urban	-1.09 (-1.21,-0.97)
Wild (> 92%)	Other	-0.39 (-0.52,-0.25)
F2, F3 Hybrid (63–92%)	Herbaceous Wetland	0.13 (0.06,0.2)
F2, F3 Hybrid (63–92%)	Forested Wetland	-0.26 (-0.36,-0.16)
F2, F3 Hybrid (63–92%)	Agriculture	-1.05 (-1.18,-0.92)
F2, F3 Hybrid (63–92%)	Grassland	-0.68 (-0.79,-0.56)
F2, F3 Hybrid (63–92%)	Forest	-1 (-1.12,-0.88)
F2, F3 Hybrid (63–92%)	Urban	-0.83 (-0.92,-0.73)
F2, F3 Hybrid (63–92%)	Other	-0.44 (-0.58,-0.3)
Feral, F1 Hybrid (< 63%)	Herbaceous Wetland	0.15 (0,0.29)
Feral, F1 Hybrid (< 63%)	Forested Wetland	-0.41 (-0.68,-0.13)
Feral, F1 Hybrid (< 63%)	Agriculture	-1.66 (-2.02,-1.29)
Feral, F1 Hybrid (< 63%)	Grassland	-0.76 (-1.04,-0.48)
Feral, F1 Hybrid (< 63%)	Forest	-1.51 (-1.81,-1.21)
Feral, F1 Hybrid (< 63%)	Urban	-0.4 (-0.58,-0.22)
Feral, F1 Hybrid (< 63%)	Other	-0.07 (-0.32,0.18)

Fig. 6.

Estimated odds ratios representing relative selection strength (probability of selecting the given land cover type divided by the probability of selecting open water) and 95% confidence intervals (CI) for female mallards marked with GPS-GSM transmitters in Michigan, Wisconsin, Illinois, Indiana, and Ohio, 2021–2023. Open water was the reference category, designated by the solid horizontal line. An odds ratio less < 1 with CIs excluding 1 indicate that land cover type was less selected than open water, CIs overlapping 1 indicate that land cover type was selected similarly to open water, and estimates > 1 with CIs excluding 1 indicate the land cover type was more selected than open water. Genotype categories correspond to the percentage wild ancestry

Discussion

As predicted, we found that mallards with greater amounts of game-farm ancestry moved shorter daily distances (Fig. 4) and were less likely to migrate than wild mallards (Fig. 5). While we predicted that selection of land cover types would be similar between wild and hybrid mallards, we found that individuals with greater game-farm ancestry had higher use and stronger selection of urban developed land cover than did wild mallards (Fig. 6; Supplemental File 4). The inverse relationship between proportion wild ancestry and relative selection strength and use of developed land cover in this study indicates that hybridization of wild with game-farm mallards likely influences resource use and selection. Mallards with a higher proportion of domestic ancestry may have morphological, physiological, or behavioral traits that make them better suited to urban landscapes. Artificial selection in captivity may intentionally or unintentionally select for traits that are appropriate for semi-domestic settings or that are maladaptive in wild settings [45]. Based on previous studies, released captive-reared or domestic mallards are likely to be more susceptible to predation [116], more susceptible to hunter harvest [54, 117, 118], and have lower annual survival [52, 119]. A recent analysis found that game-farm mallard hybrids made shorter and later spring migrations but did not detect individuals settling near urban locations [53], possibly because the mallards monitored in that study did not include as broad a range of domestic ancestry as individuals marked in this study. Selection and use of urban areas should generally reduce exposure to predation and harvest pressure and increase access to supplemental feeding [120]. In addition, we found ancestry cut-offs were associated with shifts from game-farm to wild mallard-like movement behavior. Use of emergent herbaceous and forested wetlands increased, and use of urban cover types decreased, with increasing wild mallard ancestry.

Habitat conservation efforts aimed at restoring, enhancing, and protecting rural wetlands may benefit wild mallards, but likely provide less benefit to early generational hybrid mallards. The continued growth of metropolitan areas [121] could expand a niche occupied by feral and hybrid mallards, while regional losses of wetlands and grasslands [72] threaten wild mallards. Thus, conserving rural wetlands and grasslands could be an important factor in sustaining wild mallard populations. Opportunity for public viewing could be a potential social benefit arising from mallard use of urban areas [122].

Increases in the proportion of domestic ancestry were associated with shorter daily movements and individuals that were less likely to migrate during autumn and winter. Lower mobility by individuals with greater domestic ancestry is consistent with numerous studies examining dispersal or movement distances of captive-reared mallards released into the wild [33, 34, 51, 52, 54]. The average daily movement distance for wild mallards in this study of 5.0 km (95% CI: 4.7, 5.4) was similar to mallards marked in California where daily movement distance estimates ranged from 4.9 km to 5.9 km [123] and wild mallards marked in Sweden that moved 5.1 km per day [34]. Our daily movement distance estimate for feral game-farm mallards of 1.98 km (1.58, 2.47) was similar to mean distance moved among game-farm mallards released in Sweden (i.e., ~ 1 km per day; [34]). A remaining question is whether lack of migratory behavior in early generational hybrids is due to morphological and/or physiological limitations, lack of response to migratory cues, or a combination of factors. Wild mallards captured at urban sites were less likely to migrate than wild mallards captured at rural sites, and access to open water, anthropogenic food sources, and thermal refugia [124, 125] in urban areas could contribute to reduced migration in both wild and hybrid mallards. Lack of migratory behavior could be maladaptive for individuals using rural areas as freezing conditions and snow cover limit access to open water and forage, which are proximate factors associated with mallard migration [57, 58]. Failure of early generational hybrids to migrate could explain why fewer were found in northern latitudes in the Great Lakes region, where migration probability increases in relation to more extreme winter weather and a non-migratory strategy may increase mortality risk. Climate or environmental changes may contribute to individual heterogeneity in migratory behavior in wild mallards. The primary strategies employed by wild mallards in this study were to make short (i.e., within the Great Lakes states) migrations or local movements to areas with open water during the winter. These movements correspond to contemporary northward shifts in dabbling duck band recoveries during January and February [126], and to the distribution of band recoveries from mallards marked in the Great Lakes states [49].

Genetic sampling of mallards in the Great Lakes region provided further evidence that large-scale releases of game-farm mallards [40, 42] in eastern North America (i.e., the Atlantic and Mississippi flyways) during the past century has resulted in flow of domestic genes into wild mallard populations [127]. The proportion of hybrids in our study was intermediate between the U. S. portion of the Atlantic Flyway where previous research suggested about 90% of mallards were wild game-farm hybrids [127] and the Central Flyway where sampled mallards were almost entirely comprised of wild individuals [45]. Previous sampling indicated most releases of game-farm mallards likely occurred in eastern North America [40, 42]. Our findings support this assumption at a finer spatial scale, with generally lower proportions of hybrid mallards occurring westward across states in the Great Lakes region.

Hybrids were more prevalent than wild mallards at urban sites in all study states, and only five hybrids with < 50% wild mallard ancestry were encountered at rural banding locations. We acknowledge that feral and hybrid mallards could be more susceptible to capture than wild mallards when using bait due to their domestic history lending to decreased human avoidance, as well as foraging strategy [128]. In addition to being exposed to supplemental feeding [34, 54], captive-bred mallards are known to settle in urban settings post-release [52]. Although the behavioral change among captive-bred or hybrid individuals may positively bias their capture over wild mallards, any increases in hybrids demonstrating such behavior would still reflect prevalence across locations. Consequently, our results support that increases in game-farm mallard ancestry results in significantly greater selection of urban habitat (Fig. 6).

Collectively, immobility and selection of urban areas warrants further examination of the ecological fitness of wild domestic game-farm mallard hybrids. Hybridization with game-farm mallards apparently contributes to shorter movement distances, reduced migration, and selection and use of developed land cover. Variation in movement behaviors related to genetics thus has the potential to result in individual heterogeneity in demographic parameters if mobility and resource selection influence survival or fecundity. Our findings indicate introgressive hybridization of game-farm with wild mallards affects movement ecology, and these findings could inform both future research and policies to conserve wild mallards [129].

Conclusions

Results of this study quantify the effects of hybridization between wild and game-farm mallards on movement ecology in the Great Lakes region and provide considerations relevant to mallard habitat and harvest management. Hybridization and increased proportions of domestic genetic ancestry were associated with lower mobility and selection and use of urban areas relative to local movement, migration, and resource selection by wild mallards. Conserving rural wetland habitats likely benefits wild mallards more than those with greater domestic ancestry, while urban areas could serve as a niche for mallards with more domestic ancestry. Moreover, stronger selection for urban areas and reduced migration incidence by mallards with greater domestic ancestry likely reduces their contribution to hunter harvest at a local scale and in areas south of the Great Lakes.

Our results indicate that third generation (F3) hybrids generally have similar daily movement distances and similar tendencies to migrate as observed in wild mallards, and that resource selection of late generational hybrids (F2 and F3; 62–92% wild ancestry) approaches that of wild mallards. Therefore, backcrossing of hybrid mallards with wild individuals relatively quickly (i.e., within approximately three mallard generations) results in behaviors similar to those observed in wild mallards. Given concern with declining Great Lakes and eastern mallard populations [46], further research to determine demographic rates across the spectrum of hybrid ancestry, while considering associated variation in movement behavior and resource selection, is warranted.

Electronic Supplementary Material

Below is the link to the electronic supplementary material.

Abbreviations

AHY

After Hatch Year

AIC

Akaike’s Information Criterion

BCR

Bird Conservation Region

ddRAD

Double-digest Restriction Site Associated DNA Sequencing

DNA

Deoxyribonucleic Acid

GPS

Global Positioning System

GSM

Global System for Mobile Communications

HMM

Hidden Markov Model

HY

Hatch Year

IACUC

Institutional Animal Care and Use Committee

LCC

Land Cover of Canada

LD

Linkage Disequilibrium

NLCD

National Land Cover Database

PCA

Principal Components Analysis

SNP

Single Nucleotide Polymorphism

UMGL JV

Upper Mississippi / Great Lakes Joint Venture

USGS

United States Geological Survey

RSS

Relative Selection Strength

VCF

Variant Call Format

Author contributions

Benjamin Z. Luukkonen: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Writing – original draft. Scott R. Winterstein: Conceptualization, Formal analysis, Funding acquisition, Methodology, Project administration, Resources, Supervision, Writing – review and editing. Daniel B. Hayes: Conceptualization, Formal analysis, Funding acquisition, Methodology, Project administration, Resources, Supervision, Writing – review and editing. Drew N. Fowler: Conceptualization, Formal analysis, Investigation, Funding acquisition, Methodology, Resources, Supervision, Writing – review and editing. Gregory J. Soulliere: Conceptualization, Funding acquisition, Resources, Writing – review and editing. John M. Coluccy: Conceptualization, Funding acquisition, Methodology, Resources, Writing – review and editing. Brendan Shirkey: Conceptualization, Funding acquisition, Resources, Writing – review and editing. John Simpson: Conceptualization, Funding acquisition, Resources, Writing – review and editing. Amy A. Shipley: Conceptualization, Funding acquisition, Resources, Writing – review and editing. Jason M. Winiarski: Resources, Writing – review and editing. Benjamin J. O’Neal: Conceptualization, Funding acquisition, Resources, Writing – review and editing. Joshua M. Osborn: Resources, Writing – review and editing. Barbara A. Avers: Conceptualization, Funding acquisition, Resources, Supervision, Writing – review and editing. Vergie M. Musni: Data curation, Formal analysis. Philip Lavretsky: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Resources, Writing – review and editing.

Funding

This study was support by grant GG 751210000001883 from the Great Lakes Fish and Wildlife Restoration Act, grant WLD2101 from the Michigan Department of Natural Resources Wildlife Division, grant ID 2010704 from the National Science Foundation, and through contribution from Ducks Unlimited, Dave and Denise Bunning through the Sunshine Charitable Foundation, Michigan State University, Winous Point Marsh Conservancy, Franklin College, Indiana Department of Natural Resources, Upper Mississippi/Great Lakes Joint Venture, and by the Joeseph Laurence Maison scholarship and Joseph G. Schotthoefer Memorial Student Award by Safari Club International, Michigan Involvement Committee.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

A Michigan State University institutional animal care and use committee (IACUC) approved this study (permit PROTO202100046). Approval to capture and mark birds was provided by U. S. Geological Survey Bird Banding Laboratory permit 03110.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.