Preferred habitat and effective population size drive landscape genetic patterns in an endangered species

1. Introduction

Maintaining wildlife habitat connectivity amidst the mosaic of human impacted landscapes has become a global conservation priority [1]. A lack of connectivity, particularly in the small isolated populations that typify many endangered species, leads to a multitude of demographic and genetic consequences. These include inbreeding depression [2], compromised immune response [3], loss of adaptive potential [4] and heightened susceptibility to demographic and environmental stochasticity [5]. Landscape genetics methodologies address the interactions between environmental features and the evolutionary processes such as gene flow, genetic drift and selection, and thus the mechanisms by which negative genetic impacts can be manifested. These methodologies are increasingly coupled with landscape resistance models to guide management decisions in identifying where best to set aside corridors, construct habitat linkages and otherwise promote connectivity [1,6,7].

Landscape resistance is a hypothesized measure of a landscape feature's impediment to gene flow. Studies using landscape resistance models often lack empirical data and instead rely heavily upon expert opinion to identify habitat variables important to resistance [6,7]. However, this qualitative approach has been shown to suffer from lack of repeatability and poor performance in describing actual landscape costs [8]. Some studies have begun to incorporate radio-telemetry data to identify key environmental metrics for constructing more objective landscape resistance surfaces [7,9]. A common method in studies of terrestrial animal ecology is to build models of resource selection functions (RSFs; e.g. [10]) to infer a species’ preferred habitat. Shafer et al. [9] have recently demonstrated the superiority of RSFs over null isolation-by-distance (IBD) and isolation-by-barrier models in predicting genetic structure. Prior limitations in landscape genetics research may be circumvented by using advances in empirically derived landscape resistance models.

While landscape genetics can enhance conservation planning, it cannot ignore the fundamental evolutionary processes that underlie metapopulation dynamics and their spatial and temporal scales. For example, few studies have taken into account the potential impact of population size on patterns of genetic diversity and the potential for genetic drift to obstruct resistance models from identifying important landscape genetic relationships, particularly when dealing with endangered species. Genetic drift is an evolutionary process with important implications in conservation biology due to its sensitivity to population fluctuations and temporal and geographical isolation [11–13]. The leading parameter that reflects evolutionary changes in population dynamics, such as drift, is the effective population size (N_e), defined as the size of an idealized population exhibiting the same rate of random genetic drift as the population under consideration [14], and which can roughly be seen as the number of breeders that contribute genes across generations. The loss of genetic diversity caused by genetic drift is inversely proportional to N_e, following approximately H_t/H₀ = [1–(1/2N_e)]^t, where H_t/H₀ is the reduction in heterozygosity after t generations. Consequently, knowledge of N_e can be a powerful tool in conservation as a predictor of genetic diversity loss, inbreeding and, perhaps most important to landscape genetics studies, population differentiation.

In this study, we conducted a detailed landscape genetics analysis that incorporates the effect of N_e among the factors determining patterns of genetic diversity. Specifically, we analyzed the primary environmental and demographic variables that drive ungulate population substructure in endangered woodland caribou (Rangifer tarandus). Our a priori landscape models were chosen to test specific hypotheses regarding the factors thought to have the greatest conservation impact on threatened species, including caribou. We test four hypotheses that population genetic structure is influenced by: (i) preferred habitat availability, (ii) anthropogenic barriers, (iii) predation risk, or (iv) reduced N_e. If small populations are experiencing rapid genetic drift owing to small N_e, this may explain a large variance component in population-pairwise genetic distances and obscure the signal from landscape variables. Woodland caribou offer an ideal species to explore these hypotheses as their ecological and conservation challenges are well documented in the literature [15–17] and genetic and spatial telemetry data are available for many populations [10,18].

2. Methods

3. Results

4. Discussion

Our study is one of the first to assess the impact of N_e in a landscape genetics framework. We found that after accounting for geographical distance, preferred habitat availability (figure 3, RSF) and N_e were the most significant explanatory variables in determining genetic distances between herds (table 2). This pattern was further supported when we controlled for RSF-based resistance and the partial Mantel r-values approached zero for most landscape variables (table 2), indicating, with one exception (N_e), that no other variables explained genetic variability after accounting for preferred habitat availability. RSF models have only recently begun to be applied in landscape genetics, but are already demonstrating superior results to traditional landscape models [9].

After accounting for the effects of geographical distance, only the RSF and N_e were significant in explaining the variability among genetic relationships in both F_ST (RSF was best followed by N_e) and D_c (N_e was best, followed by RSF; table 2). Similarly, in tests controlling for the variability explained by the RSF, only N_e (as measured by D_c) was a significant explanatory variable. Interestingly, after controlling for N_e, the genetic distance correlations with the anthropogenic variables increased for D_c (table 2). This improvement of correlation might simply be a statistical artifact owing to non-independence between some covariates (e.g. a suppressor effect [37]). However, this trend effectively reveals how, once the variance accounted for by N_e is partialled out, the residual variance is even more associated with all the gene-flow-related variables. Similarly, when the variance associated with RSF is removed, there remains a significant proportion explained by N_e. Given the strong association between random drift and N_e, our analysis starkly partitions genetic differentiation into a component caused by gene flow, and another, often overlooked, component that is due to random genetic drift; though predicted by theory [14], this pattern is very difficult to demonstrate in nature [1]. The interactive effects of drift and habitat fragmentation likely result in even stronger landscape fragmentation effects in small populations where drift is strongest; a detail that cannot be ignored in the precariously small populations that typify many endangered species [1].

The use of parameters related to population size have largely been absent from landscape genetics projects. However, ecological research on caribou found that censused population size was one of the key variables explaining genetic relationships [38,39], but did not explicitly test for landscape resistance at the same time. In Arctic char, Wollebæk et al. [40] found genetically assessed N_e to be the major cause of contemporary population differentiation. Thus, the effects of genetic drift due to small populations, rather than by geographical isolation due to distance, can be a major explanatory variable of population genetic relationships, and in some cases this may obfuscate the strength of relationships between genetic structure and landscape variables. This emphasizes that landscape genetics studies conducted without accounting for the effects of N_e are likely flawed and could result in misleading conservation recommendations.

The 50/500 rule often cited by conservation practitioners [41] postulates that a minimum of N_e = 50 is necessary to prevent a damaging rate of inbreeding in the short term, but that an N_e of 500–1000 is required for long-term genetic integrity [42]. Our figures of N_e were therefore concerning. Results from assignment tests [43] provide evidence of recent historical metapopulations in these caribou, a pattern that is also supported at the continental scale [18]. It is likely these same metapopulation dynamics allowed for the maintenance of genetic diversity through gene flow among demographically viable populations across unfragmented habitats. Given the persistently small populations sizes for these caribou, management strategies to protect them should emphasize fostering connectivity among caribou herds [18,43] and rebuilding past metapopulation relationships.

Despite the documented decline in population sizes for all caribou populations [19], tests for excess heterozygosity, as an indicator of recent bottlenecks, was detected consistently only for the Banff (BNP) population, and is corroborated by a negative F_IS value that is also indicative of heterozygosity excess (table 1; [2]). The Brazeau (BRZ) and Maligne (MAL) populations were of similar size, but there is some evidence that these two herds have recently exchanged migrants, which would effectively buffer them from the potential impacts of a population reduction [13], and explain non-significant heterozygosity excess. The Banff population suffered local extinction from an avalanche in 2009 [5]. Barring an increase in population size, both BRZ and MAL are at immediate risk of similar stochastic extinction [44].

Predation by wolves has been noted as the most important proximate threat to the persistence of threatened caribou populations [45]. Caribou natural history characteristics describe anti-predation behaviour through geographical spatial separation [27,46]. As such, a historic pattern of avoiding specific regions that provide good habitat for wolves could provide a natural barrier to gene flow. However, in our study, the correlation of predation risk to genetic distances was highly autocorrelated with geographical distance (table 2). Thus, predation by wolves may be too ephemeral to become a permanent landscape barrier that would influence gene flow. Alternatively, our model of predation risk may reflect wolf occurrence on a human manipulated landscape that is too recent to show up in genetic signatures. Anthropogenic barriers are frequently cited as major concerns for connectivity of fragmented populations [1,6]. With caribou, human-mediated landscape changes are predicted to be a major influence on population structure [45], particularly in Alberta [19]. However, similar to wolf predation in our study, after accounting for geographical distance or RSF, the relationship between anthropogenic features and genetic distance mostly disappeared (partial Mantel tests, table 2). The lack of a strong relationship independent of geographical distance may be due to the time lag of a genetic response to the anthropogenic features [47,48]. In caribou, a detectable numerical response to human land use changes have been documented to take several decades [49] and the potential negative impacts of anthropogenic features, even at the low density revealed here, cannot be dismissed.

The implications for endangered species such as woodland caribou are twofold: (i) in the threatened populations analysed here, modern anthropogenic features do not appear to yet have significant impacts on gene flow by themselves, but have been shown to reduce population size [38], thus leading to increased drift; (ii) conservation efforts should focus on preserving preferred caribou habitat to maintain the natural pattern of landscape resistance in caribou metapopulation dynamics.

5. Conclusion

Promoting connectivity among populations of threatened species in heterogeneous landscapes impacted by human disturbance is further complicated by the fact that most species already exist within discontinuous mosaics of preferred habitat [1,9,38]. Here, we demonstrated that the greatest predictor of genetic connectivity in caribou of west-central Alberta is preferred habitat availability. The distribution of preferred habitat demonstrated in figure 2a, and the associated resistance surface of that habitat (figure 3) exhibit the classic matrix of suitable habitat interwoven within a matrix of unsuitable space on which we would expect metapopulation dynamics to operate [50].

Metapopulation theory dictates that throughout the metapopulation, localized extinctions take place at the population level, only to be recolonized in the future. A reduction in connectivity (e.g. by habitat destruction or landscape barriers) lowers per patch immigration rate, thus inhibiting the rescue effect [50] and resulting in declines in abundance and occupancy of remaining patches [51,52]. For caribou, telemetry data (for females) indicated little movement between populations [38,43], suggesting a breakdown in such metapopulation dynamics. The Banff population illustrates the danger of decoupling metapopulation dynamics, as exemplified by its persistent isolation for decades with no new migrants, which ultimately resulted in stochastic localized extinction [5] and loss of that patch's genetic contribution to the metapopulation. Our results therefore emphasize the importance of habitat within and between population ranges for the viability of the metapopulation and its discrete elements.

The correlation that we found between genetic differentiation and low population numbers provides an empirical link between habitat loss and fragmentation [1,6]. The failure of demographic rescue in local populations points to lack of preferred habitat between populations and to a paucity of effective migrants, which, in turn, may be due to a synergistic relationship between declining caribou numbers and population density-dependent dispersal behaviour [38]. Little is known about the dispersal patterns of male caribou, but the lack of female dispersal in caribou is directly correlated with small isolated subpopulations throughout the Canadian Rockies [38,39,43]. In our study, after preferred habitat, the second variable that best explained levels of genetic structure was N_e (table 2), and if drift continues in these declining and isolated populations, barring any demographic rescue, we would predict that N_e will tend to explain an increasingly larger proportion of the variability.

Population size, and in particular N_e, is an important variable that tends to be neglected in most landscape genetic studies [53]. Genetic drift due to small, isolated, populations can lead to spatial structuring in markers such as microsatellites that are frequently used in landscape genetic research. A failure to specifically account for the N_e component of genetic drift may lead to erroneous inferences of population structure strictly based on landscape features and, as a result, will fail to pinpoint crucial demographic processes important to conservation.

Funding statement

Support was provided by the Alberta Department of Sustainable Resource Development, British Columbia Ministry of the Environment, BC Ministry of Forests, Canadian Association of Petroleum Producers, Conoco-Philips, NSERC, Parks Canada, Petroleum Technology Alliance of Canada, Royal Dutch Shell, UCD SEED funding, Weyerhaueser Company, Alberta Innovates, Alberta Conservation Association, the Y2Y Conservation Initiative, and the National Aeronautic and Space Agency (NASA) under award no. NNX11AO47G.