Telemetry and genetics reveal asymmetric dispersal of a lake‐feeding salmonid between inflow and outflow spawning streams at a microgeographic scale

Abstract The degree of natal philopatry relative to natal dispersal in animal populations has important demographic and genetic consequences and often varies substantially within species. In salmonid fishes, lakes have been shown to have a strong influence on dispersal and gene flow within catchments; for example, populations spawning in inflow streams are often reproductively isolated and genetically distinct from those spawning in relatively distant outflow streams. Less is known, however, regarding the level of philopatry and genetic differentiation occurring at microgeographic scales, for example, where inflow and outflow streams are separated by very small expanses of lake habitat. Here, we investigated the interplay between genetic differentiation and fine‐scale spawning movements of brown trout between their lake‐feeding habitat and two spawning streams (one inflow, one outflow, separated by <100 m of lake habitat). Most (69.2%) of the lake‐tagged trout subsequently detected during the spawning period were recorded in just one of the two streams, consistent with natal fidelity, while the remainder were detected in both streams, creating an opportunity for these individuals to spawn in both natal and non‐natal streams. The latter behavior was supported by genetic sibship analysis, which revealed several half‐sibling dyads containing one individual that was sampled as a fry in the outflow and another that was sampled as fry in the inflow. Genetic clustering analyses in conjunction with telemetry data suggested that asymmetrical dispersal patterns were occurring, with natal fidelity being more common among individuals originating from the outflow than the inflow stream. This was corroborated by Bayesian analysis of gene flow, which indicated significantly higher rates of gene flow from the inflow into the outflow than vice versa. Collectively, these results reveal how a combination of telemetry and genetics can identify distinct reproductive behaviors and associated asymmetries in natal dispersal that produce subtle, but nonetheless biologically relevant, population structuring at microgeographic scales.

In this study, we use a combination of telemetry and genetics to investigate the interplay between putative homing/straying behaviors and genetic differentiation among spawning streams at a microgeographic scale (sensu Richardson, Urban, Bolnick, & Skelly, 2014) in nonanadromous brown trout. The species typically exhibits hierarchical population genetic structure across a range of spatial scales (Lobón-Cerviá & Sanz, 2017), sometimes down to scales of <1 km (Carlsson, Olsén, Nilsson, Øverli, & Stabell, 1999), implying either low straying rates at these microgeographic scales or fine-scale local adaptation that constrains gene flow if straying does occur. Brown trout often exploit lakes for growing and rearing, which can involve short-or long-distance migrations between natal streams and lacustrine habitat. A particularly interesting scenario arises where brown trout spawn in both inflowing and outflowing streams, but co-occur in a more productive lake habitat for much of their lives.
Juveniles spawned in lake-outflows must conduct upstream feeding migrations to reach lake habitat, actively swimming against the flow of the river. Such upstream feeding migrations would presumably be maladaptive in lake-inflow streams, as this behavior would move juveniles away from, rather than toward, the lake, where growth opportunities are higher. Thus, inflowing and outflowing streams may exert differing selective pressures by virtue of their flow direction, which in turn could promote genome-wide genetic divergence via the above mechanisms. Indeed, Jonsson, Jonsson, Skurdal, and Hansen (1994) demonstrated that the offspring of inflow and outflow spawning brown trout displayed different directional migratory responses to water current, a population-specific juvenile rheotactic response pattern that has been identified in various salmonid species (Bowler, 1975;Brannon, 1972;Kaya, 1991;Kelso & Northcote, 1981;Raleigh, 1967Raleigh, , 1971Raleigh & Chapman, 1971). Additionally, brown trout populations that utilize common lake-feeding habitat but are genetically, behaviorally and morphologically distinct have been found to display reproductive isolation by homing back to separate inflow or outflow rivers for spawning (Ferguson & Mason, 1981;Ferguson & Taggart, 1991;Jacobs et al., 2018). Reproductive isolation appears to promote similar differentiation among sympatric lake-dwelling populations of rainbow smelt (Osmerus mordax) (Taylors & Bentzent, 1993), Arctic charr (Salvelinus alpinus L.) (Jonsson & Jonsson, 2001), Dolly Varden (Salvelinus malma) (Markevich, Esin, & Anisimova, 2018), and sockeye salmon (Oncorhynchus Nerka) (Moreira & Taylor, 2015). It remains unknown, however, whether consistent, accurate homing behavior and associated genetic divergence occurs between lake-inflow and lake-outflow streams at microgeographic scales of less than 100 m.
Here, we investigated these issues in a small lake in the west of Ireland that is fed and drained respectively by a single inflowing and a single outflowing stream, separated by less than a hundred meters of lake habitat. We hypothesized that a combination of preand postzygotic isolating mechanisms should have produced weak to moderate neutral genetic differentiation between brown trout originating in the inflow and outflow streams, and that gene flow patterns between these groups may not be symmetrical. As our first aim, we used PIT-tag telemetry to monitor lake-to-stream movements of spawning-sized trout to determine whether some fish exhibited behavior consistent with philopatry (only detected in one of the streams) while the behavior of others was consistent with straying (detected in both streams). A fish detected during the spawning season in only one of the streams may, of course, have been born in the other and thus have been exhibiting straying behavior. In the absence of more direct methods for detecting homing versus straying, genetic techniques can be used to assign fish sampled as adults in the lake to population genetic clusters that may correspond to inflow versus outflow spawning streams if gene flow is restricted. Similarly, parental movements may be inferred indirectly using genetic sibship analysis: if fry sampled in both streams assign to the same half-sib group, for example, this suggests that one of their parents spawned in both streams. Our second aim was therefore to use a range of genetic analyses, including clustering approaches, to identify any such half-sibships, test for fine-scale population structure, characterize patterns of gene flow between the streams (symmetric vs. asymmetric), and interpret these patterns in light of the behavioral data and vice versa.

| Study area
The Burrishoole catchment in the northwest of Ireland is a complex freshwater system comprised of three main lakes linked by a network of rivers and streams that drain an area of approximately 83km 2 . Bunaveela Lough (54º01′18″ N 9º32′43″ W), the most northerly and the most elevated of the three lakes, has a surface area of 46 ha, a maximum depth of 23 m, and supports populations of brown trout (Salmo trutta L.), Atlantic salmon (Salmo salar L.), Arctic char (Salvelinus alpinus L.), and European eel (Anguilla anguilla L.). Seine netting surveys indicate that trout are relatively abundant in the lake, outnumbering salmon and char within the littoral zone by more than five to one. The lake is fed by a single inflowing stream, the Fiddaunveela, and drained by a single outflowing stream, the Goulaun. The straight line distance between the point at which the Fiddaunveela flows into the lake and the point at which the Goulaun flows out of the lake is 98 m. The inflow is a shallow and flood-prone stream of approximately 2,010 m in length, draining a steep valley to the southeast of the lake. The outflow flows southwest from Bunaveela Lough for 10,345 m before entering Lough Feeagh (410 ha), increasing significantly in size as it approaches Feeagh. Due to the regulating effect of Bunaveela, the upper stretches of the outflow are less prone to rapid fluctuations in flow rates than the inflow. Although pH during baseflow is circumneutral in the two streams, they are both small, poorly buffered and oligotrophic, and therefore offer limited feeding and growth opportunities to resident trout (hydrological conditions described in Appendix Table A1). In contrast, much of Bunaveela Lough is comparatively well buffered and productive due to limestone and sandstone deposits (Whelan, Poole, McGinnity, Rogan, & Cotter, 1998) and consequently the lake represents an alternative feeding habitat where growth may be less constrained (trout size distributions in streams and lake shown in Appendix Figure A1).

| Sampling
Trout fry and parr were captured during the summers of 2017 and 2018 by electrofishing (electrofisher model: Hans Grassl IG600) in the upper and lower sections of the outflow (n = 181) and the upper, middle, and lower sections of the inflow (n = 208) (Figure 1). A total of 500 additional trout (fry, parr, and adults) were captured from the southeastern shore of Bunaveela Lough (henceforth "Bunaveela") by seine netting (9-mm mesh) across six dates between October 2016 and October 2018 (details in Appendix Table A2). All trout were anaesthetized in pH-buffered tricaine methane sulfonate (80 mg/L), measured (fork length, FL, mm), and weighed (to 0.1 g), and a small clip (c. 2 mm 2 ) was taken from the caudal fin and stored in 95% ethanol for genetic analysis. Each trout of >70 mm (n = 605) was implanted with a uniquely coded 12 mm half-duplex (HDX) passive integrated transponder (PIT) tag (Biomark 134.2 kHz ISO HDX). PIT tags were implanted into the peritoneal body cavity through a needle inserted just posterior to the tip of the pectoral fin and to one side of the midventral line at the tips of the pleural ribs. After sampling, anaesthetized fish were moved to a tank of aerated fresh river or lake water and monitored until their equilibrium was fully regained and active swimming recommenced. Once recovered, all fish were released back into the site from which they were originally captured. PIT tagging and fin clipping were carried out in accordance with S. observed during subsequent analysis, a combination of electrofishing and seine netting was used to capture trout of a range of sizes in Lough Feeagh and three of its tributaries (the Rough River, the Lodge River, and the Glenamong River) (Figure 1) over 32 dates between October 2016 and October 2018 (Appendix Table A2).
Additionally, between December 2016 and March 2019 one fish trap was operated continuously in the Rough River (RR), a river that is utilized by lake-feeding trout for spawning, and two other traps were operated at tidal limit of the river system. Trout that had been actively captured or trapped in either of the above cases were measured, visually assessed for maturity status, scanned for PIT tags, and all untagged individuals >70 mm (n = 3,617) were PIT tagged. A length distribution for trout confirmed as mature (i.e., displaying physical characteristics indicative of imminent or recent mating) was generated and used to designate a minimum threshold "mature" length (mean FL minus 1 SD of mature fish) for use in behavioral analyses. A conservative threshold of one standard deviation below the mean was used here in order to maximize the chances of only including true mature fish in the telemetry analyses (as we were interested in spawning movement behaviors specifically), that is, to exclude larger immature fish that may have been similar in size to relatively small mature fish.
Most of the trout that were PIT-tagged in Bunaveela (87.6%) were only caught once (up to 811 days prior to the spawning period during which they were detected on an antenna), and therefore, their actual size at spawning time was unknown. To estimate this, and thereby exclude any individuals that were below our threshold mature length from subsequent analyses, we used FL data from 87 PIT-tagged lake-feeding trout that were recaptured during the study period (from 15 to 505 days after tagging) to calibrate a linear model to describe growth per degree day in the statistical program R v3.5.2 (R Core Team, 2018). This then allowed us to infer the likely growth (from tagging to spawning period) of fish that were only measured at tagging, giving an estimate of their FL during the spawning period. Previous studies have shown that individual growth rate within fish populations is largely a function of temperature and individual size (Boltaña et al., 2017;Handeland, Imsland, & Stefansson, 2008;Neuheimer & Taggart, 2007). Throughout the study period, high frequency lake surface temperature data were recorded in Lough Feeagh (~7 km SW of Bunaveela) (de Eyto et al., 2019) and, due to the physical similarities and geographical proximity of the lakes, these temperatures were used as a proxy for Bunaveela surface water temperatures. By including a base temperature (T 0 ), calculations were limited to temperatures relevant to growth, or "growing degree days" (GDD) (Chezik, Lester, & Venturelli, 2014

| Monitoring behavior-PIT telemetry
In order to investigate the movement of lake-feeding trout during the spawning season, as well as the degree of movement between the upper and lower Burrishoole catchment, a network of five swimthrough, cross-channel HDX PIT antennae powered by Oregon RFID multiplexing readers was installed between August and September 2017 and maintained for the duration of the study period. HDX PIT antennae generate an electromagnetic field that wirelessly powers any nearby HDX PIT tag, causing the tag to transmit a unique 12-digit identification number that is subsequently received and recorded by the antenna reader along with the date and time. The scan rates of all readers were set to ten transmit-receive cycles/s. Two antennae were A combination of daily data on the maturity status of trout moving upstream and downstream through the RR fish traps and PIT-derived behavioral data for Bunaveela-tagged trout deemed mature was used to designate a spawning migration period for Burrishoole trout, running from the 1st of November to the 28th of February. Annual movements of mature-sized lake-tagged trout past the fluvial antennae occurred almost exclusively during this period (see results and Appendix Figure   A2), suggesting that movements recorded at this time of the year were primarily motivated by reproduction rather than exploratory foraging.
For the purpose of characterizing discrete behavioral tactics, mature trout detected only on the inflow antennae were categorized simply as "inflow-only" (IO) fish, and mature fish detected only on the upper outflow antennae as "outflow-only" (OO) fish. Fish from each of these categories could be exhibiting homing behavior to their natal streams, or straying behavior if born in the other stream-the telemetry data alone cannot distinguish these, but genetic inference may facilitate such distinctions (see below). Mature trout detected on antennas in both streams during the spawning window were categorized simply as B fish (standing for "both"), which could be exhibiting either homing or straying, or both; clearly they moved from the lake to both streams, but they may not have spawned in them.

| DNA extraction
Genomic DNA was extracted from caudal tissue from 853 samples using the Promega Wizard® SV 96 Genomic DNA Purification System. DNA quality and quantity were assessed on agarose gels by comparison with a Quick-Load® Purple 100 bp DNA Ladder (New England Biolabs). Concentration of DNA for PCR was adjusted to 2-10 ng/µl.

| Parentage and sibship analysis
In order to control for the influence of full-sibling groups on genetic structure analyses and identify mixed-site half-sibling groups (i.e., half-sibs sampled in different places) that could be indicative of nonphilopatric parental breeding behavior, the maximum likelihood method implemented in COLONY v2.0.6.5 (Jones & Wang, 2010) was used to infer parent-offspring relationships as well as half-and fullsibling groupings among all sampled individuals. COLONY input settings were female and male polygamy with inbreeding; dioecious and diploid; medium length run; full likelihood method; updating of allele frequencies; sibship scaling; and no sibship priors. Three replicate runs were conducted with differing seeds, and only assignments that were identified with >90% probability in at least two runs were accepted.
Salmonids are not known to display long-term monogamous breeding behavior over extended spatial scales (Taggart et al., 2001)  places. Thus, if scenario (c) and (d) occur regularly, the probability that half siblings hatch in different places should be higher than that for full siblings. The proportion of groups that contained both inflow-and outflow-sampled juvenile members was therefore compared between full and half siblings using the prop.test function in R (with "sites" here corresponding to geographic groups, see below). Groups containing Bunaveela-sampled members were excluded from this analysis because our sampling indicates the lake is primarily a feeding habitat for older trout that have already moved to the lake from their natal stream. The null hypothesis of equal proportions corresponds to a situation where there are no prehatching differences in spatial distribution of full-versus half-sibling families, or initial differences do exist but are erased by extensive and random posthatching movements. A significant difference in proportions (with inflow-outflow mixed-site groups more common among half siblings) therefore provides indirect evidence for lack of fidelity to a single spawning stream by polygamous parents, coupled with limited movement of fry or parr between inflow and outflow sites.
In order to avoid bias in population structure analyses that were excluded from subsequent analyses. PIT-tagged individuals that had been detected on antennae were given preference for retention, followed by individuals with the highest number of scored loci, followed by individuals that had been assigned the highest sample identification number.

| Calculating population genetic summary statistics
Prior to conducting population structure analyses, temporally distinct samples were merged based on sampling site where genetic differentiation between sampling years was low. To test for temporal genetic structure within our sampled sites, we calculated within-site between-time pairwise linearized F ST values (Reynolds, Weir, & Cockerham, 1983;Slatkin, 1995) and their significance in Arlequin v3.5.2.2 (Excoffier & Lischer, 2010) using 10,000 dememorization steps and a Markov chain length of 100,000. Without any multiple test correction of p-values, no within-site between-time comparisons were significant, and we thus merged samples across years for all sites, resulting in six geographically defined groups (GGs) ( Figure 1;  (Do et al., 2014). This single-sample method has been shown to provide similar or higher precision in estimates of N e than other available methods when applied to highly polymorphic microsatellite data with limited temporal variation such as those used in this study (Waples & Do, 2010).

| Population structure and gene flow
To test for population structure within and among GGs, we used STRUCTURE v.2.3.4 (Falush, Stephens, & Pritchard, 2003Hubisz, Falush, Stephens, & Pritchard, 2009 In order to identify the likely natal stream (i.e., inflow or upper outflow) of mature lake-feeding trout identified behaviorally as having moved into the inflow only (IO), the outflow only (OO), or both (B), the three behavioral classifications were treated as sampling populations in a STRUCTURE analysis that also included one group comprised of all juveniles sampled in the upper outflow site (GG2) and a second group comprised of all juveniles sampled in the inflow sampling sites (GG4, GG5, and GG6). Location priors were applied to the stream-sampled groups but not to the IO, OO, or B groups. This analysis was conducted after purely GG-based analyses had been used to investigate geographic patterns of genetic structure, allowing K to be fixed at an appropriate value.
Rates of recent immigration (a proxy for gene flow) between the inflow and the outflow were estimated using BayesAss v3.0.4 (Wilson & Rannala, 2003). BayesAss implements a Bayesian inference framework to estimate the fraction of individuals in a sampled population that are migrants derived from a second (or multiple) sampled population(s) per generation. All samples from the three inflow GGs (GG4, GG5, and GG6) were pooled to form a single "inflow" group (n = 146), while the upper outflow samples (GG2) formed a second "upper outflow" group (n = 124). All samples used were from trout that were below the threshold maturity FL at time of sampling, with the majority being young-of-the-year. A burn-in period of 500,000 was used followed by 5,000,000 MCMC iterations with a sampling interval of 1,000 steps. Adjustable acceptance rates fell within the optimal range of 20%-60% suggested by Wilson and Rannala without the need to alter mixing parameters. All analyses were conducted 10 times, each with different initial seed values, and the stationary distributions of the associated chains were compared in order to test for convergence between runs.
Broad spatial patterns of genetic differentiation were assessed within the greater Burrishoole catchment in order to test for evidence of isolation by distance (IBD) and explore broad patterns of gene flow between the upper and lower catchment. Pairwise linearized F ST values (Reynolds et al., 1983;Slatkin, 1995) and associated p-values were calculated among all GGs in Arlequin and the measuring tool in QGIS (v3.8.0-Zanzibar) was used to measure the minimum traversable waterway distance between all sampling sites. The resulting geographic and genetic distance matrices were assessed for evidence of IBD using a Mantel test (Mantel, 1967) of matrix correspondence as implemented in the "ape" package v5.2 (Paradis & Schliep, 2018) in R.

| Defining maturity status and spawning period
The mean FL of trout that were visually identified as mature during the study period (i.e., ripe, gravid, or spent) was 201.2 mm (SD = 36.5 mm, n = 322). We thus set 165 mm as the threshold for designating maturity status (~mean -1 SD). At the fish traps, 87.3% of visibly mature trout sampled were above this threshold size. A combination of historical data and recent sampling indicates that outside of the spawning period only 0.46% of trout sampled in the inflow and 0.18% of trout in the upper outflow exceed 165mm, while 26.1% of trout sampled in Bunaveela exceed this length (Appendix Figure A1).
Growth per GDD of recaptured lake-feeding trout was found to be influenced by both initial FL and T 0 . A base temperature of 5°C the FL of Bunaveela-resident trout that were recaptured more than once during the project and whose intervening growth between recapture events had been excluded from the model calibration process (n = 10, R 2 = .91, p < .001, RSE = 6.674).
The vast majority (99.0%) of visibly mature trout captured at the Rough River fish trap during the study period were caught between the dates we designated as the spawning period (1 November to 28 February). Similarly, 96.8% of all detections from mature-sized trout that had been tagged in Bunaveela were recorded on our fluvial antennae between these dates (Appendix Figure A2).   Table A2) were detected moving downstream past the lower outflow antenna into the lower catchment nor were any of these fish captured as smolts in the fish traps at the tidal limit.

| Parentage and sibship
COLONY identified 52 full-sibling groups with p ≥ .9 within the total successfully genotyped dataset (n = 844) in at least two out of three runs. Full-sibling groups ranged in size from 2 to 11 individuals.
Eighteen groups contained only members sampled in a single stream (inflow or outflow), four groups were dyads containing one inflow and one Bunaveela-sampled member, one group was a dyad containing one upper outflow and one Bunaveela-sampled member, and 29 groups were dyads containing two Bunaveela-sampled members.
No full-sibling groups contained both inflow-and outflow-sampled members. More than 90% of Bunaveela-sampled dyad members were >85 mm in length (mean FL = 144.3 mm), implying that these Bunaveela-sampled fish were not young-of-the-year and had most likely moved from their natal stream to the lake, unless some lakespawning had occurred (in contrast, mean FL in stream-sampled groups was 60.1 mm).
Mixed-site groups were relatively frequent among half-sibling groups compared with full-sibling groups, with 74 out of 163 half-sibling groups (45.4%) composed of two individuals sampled in separate locations (Table 3). While 61 of these mixed-site half-sibling dyads contained one member sampled in Bunaveela, 11 dyads contained one member sampled in the inflow and a second member sampled in the outflow. In eight of these 11 dyads, both siblings were sampled as young-of-the-year (<70 mm). When groups containing Bunaveela-sampled members were excluded from analyses (leaving 44 half-sibling groups and 18 full-sibling groups), a significantly higher proportion of half-sibling groups (25%) contained both an inflow-and an outflow-sampled member compared with full-sibling groups (0%) (p = .049).
Twenty-four individuals in the total dataset were assigned as parents of other individuals within the dataset (Appendix Table   A3). All of these parents were captured within Bunaveela while 12 offspring were sampled in the lake, seven in inflow and five in the upper outflow. Based on the COLONY results, 82 full-sibling samples, representing ~10% of genotyped individuals, were removed from the dataset prior to performing genetic structure analyses in order to prevent full-sibling groups from biasing population genetics analyses.

| Population genetic summary statistics
After pooling temporal samples within sites, and after sequential Bonferroni correction, locus One-102-b showed a significant departure from HWE in the lower outflow group (GG1). No other locus departed significantly from HWE in any other GG, and no locus pairs showed significant LD in any GG. Genetic structure analyses conducted both with and without the One-102-b locus were highly congruent, indicating that this single departure from HWE had no significant impact on observed patterns of genetic structure between GGs. Consequently, results described hereafter refer to analyses utilizing 14 loci, include locus One-102-b.

| Population structure and gene flow
Level one of hierarchical STRUCTURE analysis identified evidence of two genetic clusters (K = 2) (Appendix Figures A3 and A4).
These two clusters could be broadly described as "upper catch- In order to investigate lower hierarchical levels of genetic structure within the upper catchment cluster, a STRUCTURE analysis was performed on the five GGs associated with the upper catchment cluster (GG2-GG6). Although the mean ln(K) values for this analysis were marginally higher for K = 1 than K = 2, this difference was minor when compared with all other modeled values of K, suggesting that ln(K) plateaus at K = 2 (Appendix Figure A5).
Furthermore, at K = 2, a large proportion of individuals had strong assignment to one cluster or the other, and with spatial nonhomogeneity in the distribution of assignment scores (Figure 2b), a pattern consistent with the presence of genuine population structure (Pritchard, Wen, & Falush, 2007). At K = 2, most individuals in the three inflow groups (GG4, GG5, and GG6) assign strongly to a sin- Level 1 (a) assigned among four predefined groups matched in size to the four donor GGs (Appendix Figure A7b). These findings indicate that the patterns shown in Figure 2 represent genuine geographically based structure rather than artifacts from Bayesian priors.
No evidence of lower hierarchical levels of genetic substructuring was detected within the inflow or outflow GGs during subsequent analyses of these groups.
When the OO, IO, and B groups of mature PIT-tagged lake-sampled fish were included in a STRUCTURE analysis (fixed at K = 2) that also included an upper outflow and a merged inflow group (with the latter two groups composed offry and parr sampled in the streams), each of the three behaviorally defined groups exhibited distinct assignment patterns that resemble the characteristic assignment patterns of the inflow and upper outflow groups. In particular, the IO group, the B group, and the merged inflow group were characterized by consistently high individual assignment to a single "inflow" cluster BayesAss results indicate that recent migration rates between the inflow and upper outflow streams have been strongly asymmetrical and are characterized by a predominantly downstream direction of migration from the inflow to the outflow. Less than 1% (0.95%) of the trout in the inflow are estimated to be migrants derived from the outflow group (per generation), while 32.02% of trout in the upper outflow are estimated to be migrants derived from the inflow group (per generation). The associated 95% credible intervals, a Bayesian analogue to confidence intervals, are 0% to 2.4% and 30% to 34.01%, respectively. There was a high degree of convergence between multiple runs (Appendix Table A4), indicating that that the asymmetric pattern detected by BayesAss reflects genuine differences in migration rates between the inflow and outflow. Genetic diversity generally increased from upstream to downstream, with the minimum A R recorded in the uppermost group (6.01 in the upper inflow, GG6), and the maximum A R recorded in the lowest group (7.02 in the lower outflow, GG1).
Isolation-by-distance analyses revealed that genetic distance between GG pairs, expressed as Slatkin's pairwise linearized F ST [F ST / (1 − F ST )], ranged from <0.001 to 0.028 while geographic distance between sampling sites ranged from 89 to 5,645 m (Table 4)

| D ISCUSS I ON
Previous studies of brown trout have shown that populations spawning in inflowing versus outflowing streams of the same lake are often genetically and phenotypically differentiated, implying limited gene flow among them and potential local adaptation (Ferguson & Taggart, 1991;Jonsson et al., 1994;Linløkken, Johansen, & Wilson, 2014;Massa-Gallucci et al., 2010;Palmé et al., 2013). In cases where no obvious physical barriers to dispersal exist, some combination of prezygotic behavioral isolating mechanisms (that reduce the likelihood of infloworigin fish straying into outflow streams and successfully mating there, or vice versa) and postzygotic ecological barriers (reduced fitness of hybrid offspring) may constrain effective dispersal and thus gene flow.
Here, we explored these issues at a microgeographic scale, that is, a fine spatial scale that is within the typical "dispersal neighborhood" of the species (Richardson et al., 2014). Our telemetry results showed that the potential for dispersal between inflow and outflow streams is high in the Bunaveela system, given that ~30% of lake-tagged trout that were detected by our antennae were detected moving into both streams. Genetic sibship analysis indicated that some fry sampled in TA B L E 4 Pairwise genetic distances and geographic distances between all geographic groups (GGs). Genetic distances [F ST /(1 − F ST )] are located in the bottom left diagonal. Geographic distances (m) are located in the top right diagonal. Genetic distances with significant associated p-values (after applying as sequential Bonferroni correction) are indicated by bold text for gene flow was consistent with the rather weak genetic differentiation we documented between the streams. Importantly, however, the genetic data also revealed higher rates of recent migration from the inflow into the outflow than vice versa. Given the short dispersal distances involved and the lack of any obvious physical barriers, our findings point collectively toward interesting asymmetries in pre-or postzygotic isolating mechanisms, which we discuss further below.

| Interpreting the telemetry and population structure results in light of each other
A key goal of this study was to couple PIT telemetry with microsatellite-based population genetic inference to obtain complementary insights into homing/straying behaviors and the implications for population structuring. A priori, we expected to find some genetic differentiation between the inflow and outflow streams, given that (a) brown trout can exhibit genetic structuring at fine spatial scales (Carlsson et al., 1999;Lobón-Cerviá & Sanz, 2017), (b) previous studies (see above) have found genetic differences between inflow and outflow streams, and (c) isolation-by-adaptation mechanisms can be reasonably hypothesized (see below). We indeed documented subtle population structuring within the upper Burrishoole catchment, with two putative clusters that corresponded to some degree, but not perfectly, with inflow versus outflow spawning streams.
The weak nature of this structure precluded us from being able to pinpoint cases of "pure straying" behavior, which would require assigning lake-tagged individuals with high confidence to a given natal stream (inflow or outflow) and subsequently demonstrating with PIT telemetry that they were only detected in the opposite stream during the spawning season. However, the existence of "pure homing" and mixed homing/straying tactics could be tentatively inferred.
Lake-tagged fish detected during the spawning period in the inflow stream only (IO group) exhibited consistently high individual assignment (mean Q i = 0.97, SD = 0.02) to a single putative "inflow cluster"; thus, the IO group, which comprised ~39% of all fish detected on the inflow antennas, likely represented mostly "inflow-origin homers." A second genetic cluster corresponded most closely with the upper outflow (GG2), as indicated by the fact that the mean cluster 2 Q i for this group was 0.55 (SD = 0.17) in the STRUCTURE analysis that excluded the behavioral groups (Figure 3b). The lower and more variable assignment here compared to the very high and consistent assignment of inflow groups (GG4-GG6) to the putative inflow cluster (cluster 1) likely reflects substantial net immigration from inflow to outflow; thus, there is weak genetic differentiation between them and high levels of admixture in the outflow.
Additional sampling further downstream in the outflow would perhaps have revealed increasing assignment to cluster 2, making the overall distinction with respect to the inflow cluster clearer. When the behavioral groups were included in the analysis, OO fish importantly exhibited higher, although again more variable, assignments to the putative outflow cluster (mean cluster 2 Q i = 0.17, SD = 0.11) compared with IO fish (mean cluster 2 Q i = 0.03, SD = 0.02) or B fish (mean cluster 2 Q i = 0.10, SD = 0.06). This again indicates admixture in the OO group and suggests that these fish were of mixed ancestry, consistent with gene flow from the inflow to the outflow.
Thus, OO fish may represent "outflow-origin homers" (born in outflow, returned to outflow), "inflow-origin strayers" (born in inflow, moved to outflow), or some mix of both. The genetic assignment patterns point more toward OO fish being predominantly outflow-origin homers, but we have less confidence in this inference relative to our inference regarding the IO fish. Interestingly, the B group were characterized by individual assignment scores to the putative inflow cluster (mean Q i = 0.90, SD = 0.06) that were lower and more variable relative to IO fish assignments to the same cluster, but higher and less variable relative to OO fish. This suggests that many of the B group fish originated in the inflow stream-having a genetic signature that was more "inflow in nature"-but may have actually spawned in both streams, again consistent with net gene flow from inflow to outflow. Thus, B fish may be both homers and strayers, although clearly the definition of these terms is contingent on the criteria one uses to define "distinct" genetic populations, which may not map cleanly onto population units defined using demographic criteria (Waples & Gaggiotti, 2006).
The identification of juvenile half-sibling dyads (i.e., sharing a single parent) containing both inflow-and outflow-sampled fry indicates that some individuals indeed spawn in both streams. An alternative scenario whereby half siblings were all born in the same stream, but some then moved as fry to the other stream seems highly unlikely, as these fry were sampled early in life on the nursery grounds prior to when extensive dispersion is believed to occur.
Additionally, the proportion of dyads containing both inflow-and outflow-sampled members relative to same site dyads was significantly higher among half siblings than among full siblings, indicating that juvenile movement between sites (which is likely to be equally common within half-and full-sibling groups) does not account for the prevalence of these inflow-outflow half-sibling dyads.

| Asymmetric dispersal and the maintenance of genetic structure between inflow and outflow
Bayesian analyses indicated that contemporary migration rates between the two streams are strongly asymmetrical in a predominantly downstream direction, implying that inflow-to-outflow dispersal significantly exceeds outflow-to-inflow dispersal. The program BayesAss estimated that in recent years, ~31% of trout in the upper outflow are migrants derived from the inflow (strayers), which accorded well with our telemetry observation that ~37% of ostensible outflow spawners belonged to our B group, that is, they were detected during the spawning period in both streams. These findings were also consistent with the Bayesian clustering (STRUCTURE) results, whereby juvenile trout sampled in the inflow were characterized by consistently high individual assignment to a single cluster, indicating that effective migrants into this stream are relatively rare.
In contrast, fry sampled in the outflow exhibited higher and more variable assignment to a second cluster, indicating that comparatively high rates of immigration result in greater admixture at this location. Taking all genetics and telemetry results together, it appears that trout originating in the inflow are more prone to betweenstream dispersal than trout originating in the outflow.
Flow direction effects may combine with behavioral and olfactory mechanisms to bias dispersal in favor of inflow to outflow. For example, inflow-born fry may be easily washed downstream into the lake before olfactory imprinting on inflow water, while outflow fry cannot be washed into the lake-they must actively swim against the flow to reach the lake. Additionally, as spawning time approaches, lake-dwelling outflow-origin fish that approach the mouth of the inflow would receive odor cues that "smell wrong," assuming they first imprinted on outflow water before moving to the lake. In contrast, lake-dwelling inflow-origin fish would presumably detect no odor cues from the outflow stream until they enter it, so such exploratory movements may be more common among inflow-origin trout.
Moreover, for spawning-age fish of either origin, moving from the lake into the outflow can be completely passive whereas moving into the inflow requires active locomotion.
More generally, asymmetry in gene flow tends to be associated with systems driven by directional physical processes such as flowing water (Sundqvist, Keenan, Zackrisson, Prodöhl, & Kleinhans, 2016), wind (Cook & Crisp, 2005;Sanmartı et al., 2007), and pelagic currents (Pringle et al., 2011;Storch & Pringle, 2018). In such scenarios, prevailing air or water currents tend to act as prezygotic mechanisms that promote dispersal from "upstream" populations to "downstream" populations while restricting dispersal in the opposite direction. This asymmetric dispersal coupled with differences in selective pressures (and possible variation in habitat quality) can lead to interesting source-sink population dynamics with complex evolutionary outcomes (Kawecki & Holt, 2002). Additionally, net dispersal from upstream to downstream should lead to higher genetic diversity in downstream populations relative to upstream populations, all else being equal. Indeed, the most downstream group within our set of samples (lower outflow, GG1) exhibited the highest genetic diversity (Table 1)  An alternative explanation for the maintenance of inflow-outflow genetic differentiation is that reduced fitness of hybrids resulting from matings between inflow-and outflow-origin fish acts as a postzygotic isolating barrier to effective dispersal. Inflow-born fry must move downstream to reach productive lake-rearing habitat, whereas outflow-born fry must move upstream to reach the lake. A genetic basis for fry movement direction has previously been shown in inflow versus outflow systems (Jonsson et al., 1994), and thus, alleles for downstream fry movement may be selected against in outflow-born hybrids and vice versa in inflow-born hybrids. F1 hybrids may therefore die at higher rates or grow less well due to not reaching lake habitat, in each stream, effectively reducing rates of gene flow between the streams. Such isolation-by-adaptation processes can also contribute to genetic differentiation at neutral markers, for which divergence is promoted by genetic drift and constrained by gene flow (Nosil et al., 2008;Orsini et al., 2013). The low to moderate N e values (Table 1) estimated for these populations imply that nontrivial genetic drift may indeed be in operation. The existence of postzygotic barriers could also foster the evolution of prezygotic isolating mechanisms, that is, reinforcement, where, for example, inflow-origin fish "prefer" to mate with other inflow-origin fish and vice versa in order to avoid producing less-fit hybrids. It is also worth noting that while the isolation-by-distance pattern we document ( Figure 3) at a broader catchment scale is consistent with simple isolation-by-dispersal limitation and asymmetric gene flow, it could also be produced by isolation-by-adaptation process if ecological dissimilarity-and therefore the extent of local adaptation-also increases with distance (Orsini et al., 2013).
While our research strongly suggests that mature trout that move from Bunaveela to both the inflow and the outflow often breed successfully in both streams, the nocturnal nature of these activities prevented us from directly observing individuals during spawning.
Consequently, we cannot unequivocally confirm that the putative straying behavior detected during this study resulted in effective dispersal. On the other hand, it is possible for female salmonids to shed PIT tags during spawning (Bateman, Gresswell, & Berger, 2009;Foldvik & Kvingedal, 2018;Taylor et al., 2011). Indeed, such tag loss by females may account for the male bias observed among putative strayers. If so, the actual proportion of strayers among the trout tagged during this study may exceed the proportion reported here. Finally, although brown trout primarily spawn in fluvial habitats, they are capable of successfully spawning in lakes if suitable hydrological conditions are present (Arostegui & Quinn, 2019). It is possible that some of the mature-sized lake-tagged trout that were not detected in either the inflow or outflow were in fact lacustrine spawners. Predation by large brown trout, otters (Lutra lutra), eels, cormorants (Phalacrocorax carbo), and herons (Ardea cinerea) may also account for some of the discrepancy between the number of trout that we tagged and the number we subsequently detected. Due to the difficulty of operating efficient PIT antennae in lakes, and the difficulty of distinguishing lake feeding from lake-spawning related detections, it was not possible to assess the local prevalence of lake-spawning during this project, nor the influence of such behavior on local population structure.

| Concluding remarks
The geographic scale at which natal philopatry and natal dispersal operate plays an important role in regulating gene flow and determining population structure within the landscape. Our research here indicates that a small expanse of intervening lake habitat can have a significant influence on rates of dispersal and philopatry among trout populations that spawn in inflowing and outflowing streams. Furthermore, it appears that in such scenarios inflow-to-outflow dispersal may significantly exceed outflow-to-inflow dispersal. Analogous asymmetric dispersal patterns are found in various species, with important demographic (Storch & Pringle, 2018) and evolutionary (Kawecki & Holt, 2002) consequences. Isolation-by-adaptation type mechanisms may promote the maintenance of genetic differentiation at neutral, in addition to non-neutral, loci between inflow and outflow populations, while asymmetric dispersal will tend to increase genetic diversity of outflow populations. Through these processes, the presence of lake habitat between inflow and outflow streams, despite providing no physical barrier to movement or dispersal, may facilitate the evolution and persistence of local adaptations in salmonid populations at finer geographic scales than has been traditionally suggested in the litera-  Mobley et al., 2019).

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

AUTH O R CO NTR I B UTI O N S
RF, TR, PMcG, and RP conceived and designed the experiment. RF and RP collected the data. RF, TR, JC, KP, and PP analyzed the data.
RF led the writing of the manuscript. DC ensured compliance with scientific animal protection. All authors contributed to drafts of the manuscript and gave final approval for publication.

F I G U R E A 5
Mean Ln probability (STRUCTURE method) of K = 1-8. GG2-GG6 F I G U R E A 6 Delta K (Evanno method) of K = 2-7. GG2-GG6 F I G U R E A 7 STRUCTURE plots of inflow (GG4-GG6) and upper outflow (GG2) geographic groups (K = 2), in which: (a) GG2 was divided into three population groups (STR4, STR5, and STR6) equal in size to GG3, GG4, and GG5 prior to analysis. (b) All samples from the inflow and upper outflow groups (GG2, GG4, GG5, and GG6) were randomly assigned among four randomized groups (RGs) equal in size to GG2, GG4, GG5, and GG6 prior to analysis