Natal habitat preference induction in large mammals—Like mother, like child?

Abstract Habitat selection has received considerable attention from ecologists during the last decades, yet the underlying forces shaping individual differences in habitat selection are poorly documented. Some of these differences could be explained by the early experience of individuals in their natal habitat. By selecting habitat attributes like those encountered early in life, individuals could improve resource acquisition, survival, and ultimately fitness. This behavior, known as natal habitat preference induction (NHPI), could be particularly common in large mammals, because offspring generally stay with their mother for an extended period. We used three complementary approaches to assess NHPI in a marked population of woodland caribou (Rangifer tarandus caribou): (a) population‐based resource selection functions (RSFs), (b) individual‐based RSFs, and (c) behavioral repeatability analyses. All approaches compared the behavior of calves in their natal range to their behavior as independent subadults during the snow‐covered (Dec–Apr) and snow‐free (May–Nov) seasons. Using RSFs, we found that the magnitude of habitat selection between calf and subadult stages differed for most covariates, yet the signs of statistically significant effects (selection vs. avoidance) were generally the same. We also found that some habitat selection tactics were highly repeatable across life stages. Notably, caribou responses to habitat disturbances were highly repeatable year‐round, meaning that different individuals reacted differently, but consistently, to disturbances. This study highlights the potential role of natal habitat preference induction in shaping individual differences in habitat selection in large mammals and provides valuable knowledge for the management and conservation of a threatened species.


| INTRODUC TI ON
Early life experience could shape individual differences in adult behavior and thus have major ecological and evolutionary implications (Immelmann, 1975). Notably, experience with filial or environmental stimuli may induce subsequent preference for these stimuli (Dethier, 1982). Well-known examples of induced preference are the reactions of newborns to their presumed parents (Lorenz, 1935) or the search image acquired by predators after successfully capturing a prey (Ishii & Shimada, 2010). Induced preference for natal habitat, hereafter referred to as natal habitat preference induction (NHPI), occurs when habitat attributes encountered by an individual in its natal habitat increase the likelihood that it will select similar attributes as an adult (Davis & Stamps, 2004). NHPI could thus shape individual differences in habitat selection, or habitat selection "personalities" (Leclerc et al., 2016;Stamps & Groothuis, 2010).
Individual variability is increasingly considered in animal behavior and life-history studies (Hamel et al., 2018;Réale, Reader, Sol, McDougall, & Dingemanse, 2007;Stamps, Briffa, & Biro, 2012). For example, studies have measured the temporal stability of individual differences in habitat selection patterns (Leclerc et al., 2016), or have linked individual differences in habitat selection (Leclerc, Dussault, & St-Laurent, 2014;McLoughlin, Boyce, Coulson, & Clutton-Brock, 2006) and space-use patterns (Lafontaine, Drapeau, Fortin, & St-Laurent, 2017) to individual differences in life-history traits. Other studies have exposed the pitfalls of traditional population-based habitat selection analyses that do not account for individual variability (Lesmerises & St-Laurent, 2017). Although the importance of individual variability in habitat selection and its effects on life history are increasingly acknowledged, the potential forces that shape this variability, such as NHPI, remain poorly understood (Stamps & Groothuis, 2010).
Empirical support for NHPI originates from laboratory experiments on insects, and more recently from studies on wild birds and small mammals (reviewed by Davis & Stamps, 2004). The growing theoretical and empirical evidence for NHPI suggests that it could be an important source of individual variability in habitat selection.
NHPI could be particularly common in large mammals, for which the natal period, that is, the period between birth and independence from the mother, is typically long (Ralls, Kranz, & Lundrigan, 1986).
This long natal period could favor the evolution of NHPI if individuals were able to adapt their phenotype to natal-like habitat attributes, a phenomenon referred to as adaptive phenotypic plasticity (Stamps & Davis, 2006;Stamps, Krishnan, & Willits, 2009;Via et al., 1995).
Alternatively, experience learned from the reactions of the mother to various habitat attributes could trigger NHPI in large mammals, especially if this experience improved decision-making during the establishment in a new home range (Hoppitt et al., 2008;Stamps & Davis, 2006;Stamps et al., 2009).
The main objective of this study was to test for a potential role of NHPI in shaping individual differences in habitat selection in a sedentary population of boreal woodland caribou (Rangifer tarandus caribou) in Charlevoix, Québec, Canada ( Figure 1). Boreal caribou usually occur in old-growth forests providing abundant lichens, grasses, forbs, and deciduous shrubs, away from disturbed areas carrying high predation risk (such as harvested cutblocks; Rettie, Sheard, & Messier, 1997;Leblond, Dussault, Ouellet, & St-Laurent, 2016 We compared the habitat selection of caribou calves in their natal range to their selection as independent subadults using GPS telemetry. Because calves could not be equipped with GPS collars (Section 2.2 below), we used habitat selection of their mother as a proxy of their own habitat selection during their first year of life. This approach was appropriate because caribou calves are "followers" (Espmark, 1971), that is, they stay close to their mother until ~1 year of age. We hypothesized that NHPI influenced caribou habitat selection and predicted that the selection by individuals before (as calves in their natal range) versus after the separation from their mother (as independent subadults in their postdispersal range) would not differ. We also hypothesized that habitat selection would be more repeatable among life stages of a given individual than among individuals. Following this hypothesis, we predicted that variance in habitat selection coefficients among life stages would be low relative to variance among individuals. We accounted for the potential effects of seasonality by assessing habitat selection separately during the snow-covered and snow-free seasons, and we interpreted our results in light of varying degrees of range fidelity displayed by individuals in our study population (Lafontaine et al., 2017).

| Study area
The study area (7,250 km 2 ) was at the southern fringe of the boreal forest, in the Charlevoix region of Québec, Canada ( Figure 2). It included Grands-Jardins National Park, as well as portions of Hautes-

| Caribou capture and telemetry
Between April 2004 and March 2008, we captured 27 adult females using a net-gun fired from a helicopter (Potvin & Breton, 1988) and equipped them with GPS telemetry collars (models TGW-3600 or TGW-4600, Telonics Inc., Mesa, AZ, USA). Some of these individuals had been equipped with VHF collars during a previous study (Sebbane, Courtois, & Jolicoeur, 2008). Depending on model and year, we programmed GPS collars to record a location every 2, 3, 5, or 7 hr. Every 1 or 2 years, we recaptured individuals to download telemetry data and replace batteries. Individuals were monitored until March 2012, when collars were programmed to drop using an automated release mechanism.
Between spring 2004 and spring 2007, we located pregnant females by helicopter every 1-3 days during the calving period, looking for the presence of a calf (details in . These regular surveys allowed us to identify and capture 55 calves soon after birth (see Appendix S1: Table   A1). GPS collars were too heavy to be placed onto newborn calves (1,300 g). Instead, we fitted calves with a 15-g ear-tag VHF transmitter (Holohil Al-2C, Carp, Ontario, Canada) or an expandable 400-g VHF collar (model M2510B; Advanced Telemetry Systems, Isanti, MN, USA) equipped with mortality sensors, with the intent of recapturing them at the subadult stage. Of the 55 calves captured, seven died from unknown causes, one from drowning, and 19 were killed by predators during the first 5 weeks after birth .
We lost track of eight additional calves due to transmitter/collar defects. We recaptured the remaining 20 individuals at 2.1 ± 1.6 years (mean ± SD) and equipped 15 of them with a GPS collar.

| Life stages and seasons
In all analyses, we used the locations of a mother accompanied by her calf as a proxy for the location of her calf during its first year of life. We then used the locations from the same offspring during its first complete year of GPS monitoring to evaluate its behavior as a subadult. During the calf stage, individuals followed their mother F I G U R E 2 Map of the study area, showing the natal ranges of calves and the postdispersal ranges of subadults during the snow-covered (Dec-Apr) and snow-free seasons (May-Nov) used to study natal habitat preference induction in a boreal population of woodland caribou in Charlevoix, Québec, Canada, 2004Canada, -2011 everywhere and occupied their natal range. During the subadult stage, individuals were independent from their mother and occupied their postdispersal range. Calf and subadult life stages were separated by a period of 460 ± 198 days (min = 239; max = 670) during which we did not know the location of the individual. We considered two seasons based on snow-cover data from the Forêt Montmorency weather station in our study area (www.mddelcc. gouv.qc.ca/climat/donnees/). The snow-covered season went from 1 December to 30 April, and the snow-free season went from 1 May to 30 November.

| Range fidelity
We evaluated the range fidelity exhibited by caribou in our study area, that is, the tendency for individuals to reuse the natal habitat after the natal period. This step was necessary because our habitat selection analyses could not distinguish between animals

| Population-and individual-based resource selection functions
To assess habitat selection, we imported all GPS locations into ArcGIS 10.3.1 and associated each location to habitat attributes obtained from digital maps (see Table 1). We included all land cover types as binary variables (using 50-90-year-old conifer-dominated forests as the reference category), as well as topography and distances to both road types as continuous variables in resource selection functions (RSF; Manly, McDonald, Thomas, McDonald, & Erickson, 2002). We used a binary-dependent variable (1 = GPS location, 0 = random location) within a use-availability design (Johnson, Nielsen, Merrill, McDonald, & Boyce, 2006). RSFs allowed us to contrast used habitat characteristics with those of an equivalent number of randomly generated locations in natal and postdispersal ranges during both seasons. Contrary to our analyses on range fidelity, we estimated ranges using 100% minimum convex polygons (MCP) with the genmcp tool in the Geospatial Modeling Environment (Beyer, 2012) for our RSF analyses (see Figure 2). We used MCPs rather than BBMMs to insure that ranges were sufficiently large to adequately represent availability (Leclerc, Dussault, & St-Laurent, 2012).
We estimated seasonal population-based RSFs by fitting generalized linear mixed models using the glmer function in the lme4 package (Bates, Mächler, Bolker, & Walker, 2015) in R. We used individual identity as a random intercept in all population-based models to account for differences in sample size among individuals (Gillies et al., 2006). To compare the habitat selection of calves to the selection of subadults, we added a life stage variable to all models and used it in interaction with all other fixed variables in the models. For both seasons, we built a set of seven a priori candidate models (Table 2)  To further evaluate the occurrence of NHPI in the population, we applied the most parsimonious seasonal population-based habitat selection models to each individual (while also eliminating the random effect) using the glm function in R. To achieve this, we used individual RSFs (Leblond et al., 2016) which we compiled to highlight individual differences in habitat selection. Our goal was to better distinguish population responses from "filial" responses to habitat attributes and to assess whether individual responses in habitat selection were masked by populational responses (Lesmerises & St-Laurent, 2017). To avoid overestimating among-individual variance in repeatability models (which would have overestimated repeatability), we accounted for functional responses in habitat selection (i.e., variations in habitat selection in response to changes in resource availability; Leclerc et al., 2016). We did this by adding the proportion of random locations that fell within each land cover type as a fixed effect in repeatability models. We used this proportion as a proxy of the availability of each land cover type in each range.

| RE SULTS
Of the 15 calves that survived their first year of life and that we were able to equip with a GPS collar, respectively, 9 and 8 could be paired to a GPS-monitored mother during the snow-covered and snow-free seasons. We restricted our final sample to these individuals because TA B L E 1 Variables used in resource selection functions used to assess NHPI in a boreal population of woodland caribou in Charlevoix, Québec, Canada, 2004Canada, -2011 Natal and postdispersal ranges during the snow-covered season had respective sizes of 50 ± 29 km 2 (mean ± SD) and 30 ± 21 km 2 on average (estimated using 100% MCPs; n = 9). The mean percentage of overlap was 32% ± 23% and ranged between 0% and 65%. During the snow-free season, natal and postdispersal ranges had respective sizes of 104 ± 97 km 2 and 96 ± 37 km 2 on average (n = 8). The mean percentage of overlap was 45% ± 23% (range = 26%-81%). Seasonal home ranges were thus on average more than twice as large during the snow-free compared to the snow-covered season, irrespective of life stage. The percentage of overlap was also larger during the snow-free than the snow-covered season.

| Population-and individual-based resource selection functions
For both seasons, the model that best explained habitat selection by caribou was the global model including land cover types, topography, and distances to active and derelict roads (

| Behavioral repeatability
By directly quantifying the repeatability of habitat selection between life stages while accounting for functional responses to habitat availability, we found several habitat attributes for which behavioral responses were highly repeatable across life stages (Table 5). Notably, we found high repeatability in the response of caribou to mature conifer and deciduous stands, ≤5 and 6-20-year-old disturbances, and the "other" category during the snow-free season. During the snowcovered season, only ≤5-year-old disturbances, active roads, and the "other" category had significant repeatability values (Table 5).
Interestingly, responses toward habitat disturbances were generally more repeatable than responses toward natural habitat covariates, irrespective of season. Indeed, the mean annual (i.e., both seasons combined) repeatability for ≤5-and 6-20-year-old disturbances, the "other" category, and active roads combined was 0. Notes. Separate models were used for the snow-covered (Dec-Apr) and snow-free (May-Nov) seasons. All models included interactions between all fixed effects and Life stage, as well as CaribouID as a random effect. See Table 1 for a list of the variables included in each group of variables (i.e., Land cover, Topography, and Roads). AIC c : Akaike's information criterion corrected for small sample sizes; ΔAIC c : difference in AIC c compared to the most parsimonious model (ΔAIC c = AIC ci − AIC cmin ); k: number of parameters.
TA B L E 2 Ranking of population-based resource selection functions comparing seasonal habitat selection by caribou calves in their natal range to selection by the same individuals as subadults in their postdispersal range, used to assess NHPI in a boreal population of woodland caribou in Charlevoix, Québec, Canada, 2004Canada, -2011 individuals avoided and five individuals selected for 6-20-year-old disturbances during the snow-free season (i.e., high among-individual variance), selection coefficients remained relatively constant from calf to subadult stages for all eight individuals (i.e., low withinindividual variance; Figure 3).

| D ISCUSS I ON
We used long-term monitoring of individuals and habitat selection analyses to assess the occurrence of NHPI in a wild population of woodland caribou. Contrary to our prediction, the habitat selection of subadults differed statistically from the selection of calves for most habitat covariates at the population and individual levels.

| Linking RSFs and behavioral repeatability analyses to study NHPI
Numerous studies have determined the repeatability of behavioral traits in wild animals (see Bell et al., 2009), yet the repeatability of habitat selection has seldom been assessed (but see Leclerc et al., 2016). Individual-based RSFs allowed us to compare habitat selection among life stages and highlighted potential individual effects masked at the populational level (Lesmerises & St-Laurent, 2017). Yet, only when we combined telemetry-based resource selection functions with repeatability analyses were we really able to determine the proportion of variance explained by individual differences in habitat selection (Leclerc et al., 2016;Niemelä & Dingemanse, 2014 Notes. Individual models were composed of the same habitat covariates as the most parsimonious population-based resource selection functions without random effects (see Table 3). For each individual Life stage × habitat covariate, a blue overlay was used to indicate that selection statistically differed between life stages (i.e., significant interaction), and an orange overlay was used to indicate that selection did not statistically differ between life stages. Letters S, A, and P were, respectively, used to represent instances when calves selected, avoided, or used habitat in proportion to their availability. In cases when selection was similar among life stages, this letter also indicated selection by subadults. to lead to increased predation rates in declining caribou populations (Leech, Jelinski, DeGroot, & Kuzyk, 2017), including our study population Leblond et al., 2016). Our study provides insights as to why caribou populations may be susceptible to human development; that is, behavioral adaptation to anthropogenic disturbances is unlikely to occur quickly in caribou, if at all, because individuals tend to repeat the same habitat selection tactics across matrilines. That is, of course, only true if NHPI is maintained until females rear their own calves, an assumption that  the average population response for given habitat characteristics (Nakagawa & Schielzeth, 2010 (Johnson, 1980). The argument could be made, however, that habitat selection tactics are more likely to be similar in overlapping ranges than in totally separate areas. The fact that postdispersal ranges were predominantly composed of new areas relative to natal ranges (55%-68% of ranges did not overlap, depending on seasons) supports the hypothesis that caribou were displaying NHPI and that range fidelity was unlikely to be the sole factor explaining repeatability of habitat selection in our study. Moreover, considering the large area of postdispersal ranges and the high heterogeneity of habitat attributes found in these ranges, individuals had the opportunity to display a different habitat selection tactic across life stages even when parts of their natal and postdispersal ranges overlapped.

| NHPI as a mechanism explaining habitat selection
The potential role of NHPI in shaping habitat selection is reminiscent of the age-old "nature versus nurture" debate (Plomin, 1994).
Selection for a stimulus experienced in the natal habitat could be "innate," that is, individuals could have a genetic predisposition to select habitat features that improve their fitness based on the genotypic legacy of their ancestors. Under this hypothesis, individual differences in habitat selection would originate from different genotypes in the population, and theory predicts that the frequency of various habitat selection tactics should be proportional to the fitness benefits provided by these differential strategies. On the other hand, habitat selection could be learned through the observation of parents, or derived from experience of the environment during early life. To determine whether preference for natal-like habitat is inherited or induced through experience would require additional information (e.g., paternal identity and behavior; Morehouse, Graves, Mikle, & Boyce, 2016). Even then, the task of identifying all processes intervening in habitat selection would be complex, as both innate and acquired processes could act simultaneously (Pigliucci, 2001).
Nevertheless, the high repeatability of habitat selection across life stages in our study population suggests that NHPI could partly explain the habitat selection tactics of boreal caribou.

| NHPI in wildlife management and conservation
NHPI and range fidelity may benefit individuals by improving their knowledge of the habitat, allowing them to avoid predators more efficiently, or to access better food, shelter, or mates (Berteaux & Boutin, 2000;Davis & Stamps, 2004). However, in some circumstances, these behaviors may be maladaptive, partly explaining the difficulties of some species to adapt to recent environmental changes Lamb, Mowat, McLellan, Nielsen, & Boutin, 2017). In intensively managed areas where the rate of habitat alteration is high, animals exhibiting NHPI could select habitat in ways that are suited to past rather than current conditions, causing individuals to settle in poor or sink habitat (Piper, Palmer, Banfield, & Meyer, 2013). Such a response has been proposed to explain the seemingly maladaptive habitat selection tactics and subsequent poor recruitment rates of female caribou in our study population Leblond et al., 2016). In regions subjected to climate change, caribou could lag behind their optimal climatic envelope (Schloss, Nuñez, & Lawler, 2012) because of their fidelity to familiar space (Lafontaine et al., 2017). Moreover, the naivety of individuals regarding local mortality risks could impede the success of caribou translocations in areas where conditions differ from those found in the natal habitat of translocated individuals (Le Gouar, Mihoub, & Sarrazin, 2012;. We argue that the implications of early-life habitat selection and the factors that influence this behavior, such as NHPI, should be considered in the application of wildlife management, especially for species of high conservation concern such as caribou.

ACK N OWLED G M ENTS
We would like to thank B. Baillargeon, L. Breton, P. Dubois, J.-G.

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N
BL was responsible for analyzing data and writing the manuscript.
SDC and MHSL contributed to writing the manuscript. CD conceived the project and contributed to writing the manuscript. ML contributed to conceiving the project, oversaw analyses, and contributed to writing the manuscript. All authors approved the final manuscript.

DATA ACCE SS I B I LIT Y
Data used to perform repeatability analyses are available from the