Dispersal is a complex process involving the interaction between animal behavior and landscape structure (Baguette and Dyck, 2007). This is because landscapes in which organisms move are heterogeneous and composed of elements with dynamically changing costs and benefits for dispersing organisms (Belisle, 2005). Spatial models of animal distribution, commonly assume that dispersal strategies are fixed and evolutionarily stable. That is, individuals move according to species-specific decision rules, for example to avoid inbreeding or low suitability habitat (Lima and Zollner, 1996; Bonte et al., 2012). But dispersal is a complex and multi-phase life-history process that is under selection at each of its stages (Clobert et al., 2009) and, therefore, should rarely be invariant.

Sex-biased dispersal, for example, is a well-documented form of intraspecific behavioral variation (Pusey, 1987) and is commonly assumed to be a near-universal avian trait. Though, it is rarely considered in conservation planning, empirical study designs or in models of landscape connectivity. A recent study tested for sex-biased movement in an understory forest bird species in fragmented landscapes finding that males indeed had a reduced inter-patch movement success when compared to females (Awade et al., 2017). This finding confirms that sex-biased movements occur in fragmented landscapes, and should be more widely considered in connectivity-conservation scenarios; but also, that inter-individual variation in general should be considered. Given especially that dispersal behavior is heritable and under selection, differences among traits related to the different stages of dispersal (i.e., emigration, transfer in the matrix and immigration) should be expected among populations that persist in different landscape contexts (e.g., continuous or fragmented forests).

Indeed, organisms that persist in fragmented landscapes do exhibit inter-population variation in dispersal traits related to landscape structure. Butterfly populations from fragmented or disturbed habitats differ in dispersal related traits when compared to populations from continuous or undisturbed habitats (Merckx and Van Dyck, 2006; Öckinger and Van Dyck, 2012). In a long-term capture–recapture study of forest birds in Amazonia, conspecific individuals exhibited differences in maximum dispersed distances depending on whether they occurred in forest fragments or continuous forest (Van Houtan et al., 2007). Studies in temperate forests demonstrate that birds hatched in isolated territories have greater natal dispersal distances (Pasinelli et al., 2004) and a delay of dispersal events (Lens and Dhondt, 1994).

Fahrig (2007) proposed a conceptual model for dispersal evolution in different landscape contexts. The model predicts that individuals evolving in continuous habitats, with little or no experience with habitat boundaries, would exhibit low resistance to cross boundaries that they do encounter. In fragmented landscapes, those individuals experience high mortality in the matrix due to the usually high-risk matrix encountered after crossing a boundary, and therefore have generally low dispersal success, which makes them maladapted in human modified landscapes. On the other hand, in young birds from fragmented landscapes that developed exposed to a risky matrix and its boundary conditions, there should be strong selection for appropriate risk-averse movement decisions including behavioral avoidance of crossing dangerous habitat boundaries. This model assumes selection and genetic adaptation over longer time scales as the main driver of boundary responses but behavioral variation in the short term can also emerge by behavioral plasticity (Stamps and Biro, 2016).

Animals have the ability to change their behavior to reduce risk (Lima and Dill, 1990). During exploration and dispersal movements, risks to birds abound but the collection and use of information about those risks can make dispersal safer for individuals (Delgado et al., 2014). However, dispersing animals can only use information if they are able to correctly interpret available cues, for example those related to different kinds of predation threats (Huang et al., 2012). The capacity of using this information can be either inherited or obtained from direct experience with their environment (Spiegel and Crofoot, 2016). That is, movement behavior in fragmented landscapes might also emerge as some individuals adjust their movement behavior based on state-dependent conditions. For example, wild great tits (Parus major) became fast movers or fast explorers (i.e., exhibit increased risk taking behavior) when their probability of survival was experimentally decreased (Nicolaus et al., 2012).

Within birds and other vertebrates, exploratory behavior is a key trait underlying individual variation in use of space and resources because it provides the means to learn about and utilize the environment (Mettke-Hofmann et al., 2006). Thus, individuals that spend their juvenal exploratory phase in fragmented landscapes may be exposed to many different cues, information types and sources, and develop different decision-making strategies when compared to individuals that grow up in a continuous habitat without complex boundary conditions (Spiegel and Crofoot, 2016; Reader, 2015). Additionally, juvenal individuals that make poor decisions in gap and forest boundary crossing may get weeded out, leaving adults in fragmented habitat with better spatial decision-making regarding edges and their dangers (Cosentino and Droney, 2016).

We conducted a translocation experiment with radio-tracking to assess whether inter-patch dispersal success is different among individuals of populations belonging to contrasting landscapes (i.e. fragmented vs. continuous forest) in a Neotropical rainforest bird species, the White-shouldered Fire-eye (Pyriglena leucoptera, Thamnophillidae). Our main hypothesis, derived from Fahrig's (2007) framework and other empirical evidence reviewed above, is that lack of previous experience with a fragmented landscape should hinder successful inter-patch movement in White-shouldered Fire-eyes raised in continuous unbroken forest (and vice versa; birds living in fragmented habitat would disperse better in a fragmented landscape). Furthermore, we propose that this pattern is a consequence of higher emigration propensity in edge-naïve birds from continuous forest (i.e. low resistance to cross boundaries) with higher mortality in the matrix, while individuals from the fragmented landscape should show lower emigration propensity (i.e. high resistance to cross boundaries) but an overall higher dispersal success because once they leave a fragment they are more successful crossing the matrix. To help reveal differences in emigration propensity, we subjected birds to a short behavioral trial in a cage to determine their location on the fast-slow exploratory continuum, because there seems to be a correlation between ‘slow’ exploring in birds and higher survival in the face of risky decision-making (Hall et al., 2015).

Therefore, our specific predictions are as follows: individuals from fragmented landscapes (i) should take longer to leave a small forest patch, (ii) spend less time crossing matrix habitat when moving (thereby minimizing exposure to predators) and, consequently, (iii) exhibit greater success in arriving to a neighboring forest patch and, finally, in the behavioral trials, (iv) birds from the fragmented landscape should be slower explorers than birds raised in continuous forest. Our study design cannot disentangle selective genetic change from behavioral adjustments, but rather aims to test whether inter-population variation in dispersal traits can arise in response to human-driven landscape change. We also seek to provide information for better management practices to enhance connectivity in fragmented landscapes.

Utilizing the rationale set out in Fahrig's (2007) conceptual model for non-optimal animal movement in human-modified landscapes we investigated whether individuals of P. leucoptera taken from different populations of origin (i.e. with or without previous experience with fragmented landscapes) would exhibit (a) different behavioral patterns when moving freely through a fragmented landscape, (b) different inter-patch dispersal success and (c) different exploratory behavior in a novel captive environment. In agreement with model predictions we found that individuals taken from continuous habitat, presumably with no previous experience of fragmented boundaries or non-forest habitats, had a higher emigration propensity (i.e. reduced boundary response), lower success when crossing the matrix, and consequently, exhibited an overall reduced dispersal success in human-modified landscapes.

The central implication of this study (and central prediction of Fahrig's model) is that sudden landscape changes may result in non-optimal spatial behaviors that can lead directly to reduced population persistence in fragmented landscapes. On the other hand, our findings (as predicted) revealed that P. leucoptera individuals living in older fragmented landscapes had an overall higher dispersal success than individuals from the continuous forest population, suggesting that species can adapt to life in fragmented landscapes with time (Cheptou et al., 2017).

Larger inter-patch distances result into reduced dispersal success as Awade et al. (2017) demonstrated in the same system. But individuals from the fragmented population crossed longer distances more successfully than individuals from the continuous forest. Interestingly, time spent in the matrix was similar among populations, contrary to expectations, suggesting that once past the boundary individuals from both populations moved quickly to minimize exposure to risk (Lima and Dill, 1990). However, individuals from the fragmented population, moving equivalent inter-patch distances as individuals from the continuous forest, experienced reduced probability of mortality in the matrix. Specifically, the distance at which individuals from the fragmented landscape had a 50% chance of dispersing successfully (ca. 250m) is about 2.5 times larger than the distance for individuals from the continuous forest (ca. 100m). This suggests that their previous exposure to the dangers of fragmented landscapes helped them move more safely (Reader, 2015).

The differences in emigration propensity, or boundary response, observed in individuals from the different populations may be explained by the differences in their exploratory behavior that we detected (i.e., reduced exploratory score in individuals from the fragmented population). Fast versus slow exploration, assessed via activity levels in novel environment tests, is one of several traditional axes of personality syndrome (Carter et al., 2013) that drive inter-individual variation in many types of behaviors, including dispersal propensity (Dingemanse et al., 2003; Debeffe et al., 2013). Individuals (and species) that move more frequently, visit a larger volume of the cage and visually scan their environment more, fall within a “fast explorer” behavioral profile (Huang et al., 2015). But these fast explorers explore their environment more superficially when compared to slow explorers (Van Oers et al., 2004). Fast exploration is therefore commonly used as a proxy for risk-taking behavior (faster explorers take more survival and reproductive risks; Nicolaus et al., 2012).

On the other hand, slow exploration can predict better performance at some problem-solving tasks, including discrimination of vocal cues (Guillette et al., 2011), or a higher perception of incident predation risk (Huang et al., 2012) and presumably lower chance of attack by avian predators. Many such slow versus fast exploration functional differences could be related to the differences we observed in dispersal success and movement patterns in P. leucoptera. Individuals from the fragmented landscape were overall slower explorers when compared to the continuous forest individuals. If birds from fragmented landscapes are risk-averse and more accurate in interpretation of information cues, then these behavioral patterns could indeed enable more successful dispersal.

Individuals from the fragmented landscape vocalized more also, raising several possible factors. The birds could have been seeking conspecific social information (King and McGregor, 2016) because birds raised in riskier fragmented habitat may rely more heavily on social cues about risks and rewards (Griffin, 2004). For example, individuals seeking suitable habitat, under a model of conspecific attraction, will use vocalizations of conspecifics to orient toward suitable, and minimize time in unsuitable habitat (Fletcher, 2006). Alternatively, higher calling rates may simply signify an increased level of anxiety and perception of greater predation risk that birds from continuous forest are too naïve to sense (Sieving et al., 2010).

Our study was not designed to disentangle the relative importance of genetic adaptation versus behavioral plasticity (or individual behavioral adjustments) in determining the observed inter-population behavioral variation in dispersal patterns and success. We argue, however, that this variation is likely a consequence of individual behavioral adjustments that result from state-dependent decisions in different landscape contexts. In fragmented landscapes P. leucoptera individuals may be adjusting their behavior as they gain experience and encounter edges and the risky matrix. As juvenile birds they may learn to be slow, and careful explorers with greater risk-aversion (Reader, 2015). This is more likely than a landscape-level difference in genetic frequencies of slow-exploring individuals in fragmented areas for two reasons. First, fragmentation in this area is only ca. 150 years old (Ribeiro et al., 2009) and P. leucoptera has too a long generation time to have changed that much. Second, this species can probably disperse far enough to insure at least some gene flow among continuous and fragmented populations within our study region, as evidenced by high haplotype sharing among southern populations (Maldonado-Coelho, 2012); even a few dispersal events among populations may counteract gene frequency changes due to selection (Lenormand, 2002). It certainly is likely that given enough isolation, an evolutionary signature could emerge quickly because our data suggest that the mortality rate of non-optimal dispersers could represent hard selection (Wade, 1985). Future studies on similar systems should not discount the simultaneous occurrence of both mechanisms in attempts to discern underlying mechanisms.

Our findings have important implications for connectivity management and conservation of populations in fragmented landscapes. Observed behavioral adjustment as reported for P. leucoptera may explain in part why fewer species have gone extinct than expected in the Atlantic forest (Brooks et al., 1999). This, however, should not be interpreted as a high resilience of forest bird species to landscape change because behavioral adjustments may occur only under a small range of conditions. For P. leucoptera, distances above 350m result in very low probability of dispersal success regardless of their population of origin. At such distances corridors may be a more effective measure to increase connectivity. In fact, probability of occurrence of P. leucoptera and other forest-dependent species is very low in landscapes with less than 30% of forest cover (Banks-Leite et al., 2014; Boscolo and Metzger, 2011). Therefore, our results should be rather interpreted as a call for caution for sudden and abrupt landscape changes that reduce the chances of adaptive selection of species traits, or of individuals to adjust their behavior to the new environment. But when landscape change is unavoidable, a gradual transformation should be preferred coupled with movement-friendly matrix management (see Biz et al., 2017) and considering habitat configuration when forest cover is below 50% (Villard and Metzger, 2014) to increase connectivity in fragmented landscapes.

The authors declare no conflicts of interest.