Using multiple sources during reintroduction of a locally extinct population benefits survival and reproduction of an endangered freshwater fish

Abstract Through using different sources, population reintroductions can create genetically diverse populations at low risk of harmful inbreeding and well equipped for adaptation to future environments. Genetic variation from one source can mask locally nonoptimal alleles from another, thereby enhancing adaptive potential and population persistence. We assessed the outcomes in survival, growth and reproduction of using two differentiated sources (genetically diverse Yarra and moderately diverse Dartmouth) for translocations and stocking to reintroduce the endangered Australian freshwater Macquarie perch Macquaria australasica into the Ovens River. For stocking, same‐ and different‐population parents (“cross‐types”) were used during hatchery production. Genetic samples and data on individual fish were collected over three years of monitoring the Ovens. We genetically assigned Ovens fish to their broodstock parents and tested whether cross‐type and genetic dissimilarity between parents are associated with offspring survival, and whether cross‐type and parental dissimilarity or individual genetic diversity are associated with somatic growth rates of stocked fish. We genetically identified translocated fish and assessed local recruit ancestry. Of 296 Ovens fish, 31.1% were inferred to be stocked, 1.3% translocated and 67.6% locally born. Cross‐type strongly predicted survival of stocked offspring: those with two Yarra parents had the highest survival, followed by offspring with two‐population, then Dartmouth, ancestry. Of the Ovens recruits, 59.5% had Yarra, 33.5% two‐population and 7.0% Dartmouth ancestry, despite 67% of stocked and 98% of translocated fish originating from Dartmouth. Offspring with two Yarra parents grew faster than offspring of Dartmouth or two‐population ancestry. Although Dartmouth fish appear to be less fit in the Ovens compared to Yarra fish, possibly due to deleterious variation or genetic or plastic maladaptation, they contribute to the reintroduced population through local interbreeding with Yarra fish and relatively high survival of stocked offspring of two‐population ancestry. Thus, combining compatible stocks is likely to benefit restoration of other wildlife populations.


| INTRODUC TI ON
Human-driven environmental change has resulted in unprecedented biodiversity loss (IPBES, 2019). Counteracting this through restoration of declining or previously extinct populations requires a multi-faceted approach, including mitigating the original drivers of decline, considering population sizes, genetic diversity and fit between the current and future environment and genetic composition, morphology, physiology and behaviour of introduced individuals (IUCN/SSC 2013). Small populations experience increased extinction risk due environmental and demographic stochasticity, and loss of genetic diversity, reduced ability to adapt to new conditions, and reduced fitness due to inbreeding (inbreeding depression) (Frankham et al., 2017). Rehabilitated habitat might be suboptimal for individuals translocated from any single source, either because they are adapted to different environments through genetic changes or phenotypic plasticity (e.g. epigenetic modifications experienced during development; Duncan et al., 2014), or because the source population accumulated deleterious variation through genetic drift or relaxed selection (Ferchaud et al., 2018;Huff et al., 2010). Populations established from a single source will usually capture only a subset of the source's genetic diversity (Furlan et al., 2020).
Using multiple populations as sources for reintroduction can increase genetic diversity and the ability to adapt to changing environments and decrease the risk of inbreeding depression in the translocated population, as well as increase the likelihood that some translocated individuals or their descendants will thrive in unpredictable or changing environments (Barrett & Schluter, 2008;Binks et al., 2007). Outcrossing of inbred populations that do not exhibit known predictors of outbreeding depression has overwhelmingly positive fitness effects that extend beyond the F 3 generation (Frankham, 2015(Frankham, , 2016. Even with noninbred sources, fitness and growth of translocated populations can be elevated by genetic admixture through heterosis, masking deleterious/maladaptive alleles, and rapid adaptation (Binks et al., 2007). Using multiple sources can be appropriate if outbreeding depression (fitness loss due to interbreeding between genetically incompatible or differently adapted populations) and behavioural differences between populations are unlikely, and in the case of uncertainties, adaptive management could alleviate concerns (IUCN/SSC 2013). Gene flow between recently isolated populations that have been evolving in similar environments should have low risk of harming fitness and undermining local adaptation (Frankham et al., 2017). Even with unexpected mating asymmetries or dominance of one stock, mixing distinct populations can still be overall beneficial (Huff et al., 2010;Thavornkanlapachai et al., 2019), although unexpected lack of interbreeding (e.g. due to behavioural differences) will result in lower effective population size than anticipated.
Illustrative cases from a range of wildlife and management scenarios could accelerate the uptake of evolutionary thinking by conservation managers (Cook & Sgrò, 2017). The fitness benefits of using multiple sources for founding populations have been demonstrated in few wildlife species, and monitoring outcomes remains rare (Attard et al., 2016;Thavornkanlapachai et al., 2019;White et al., 2018). Here, we used genetic monitoring to test for the effect of using multiple source populations on survival and recruitment of an endangered freshwater fish, Macquarie perch (Macquaria australasica Cuvier 1830) during a reintroduction programme in a south-eastern Australian river where it had become extinct through human impacts.
The Macquarie perch was once common and widespread through the Murray-Darling Basin (MDB) of Eastern Australia, but experienced severe population declines and isolation through human impacts (Faulks et al., 2011;Ingram et al., 2000). The attendant population differentiation and very low genetic diversity in most remaining populations makes them vulnerable to inbreeding depression and decreased adaptive potential (Pavlova et al., 2017).
The risk of outbreeding depression on mixing populations was assessed to be low, so augmented gene flow was recommended as an urgent management action, adopted by the species-wide recovery plan (Pavlova et al., 2017;Commonwealth of Australia, 2018).
Macquarie perch became extinct in the Ovens River in the southern MDB in the 1980s, due to excessive recreational fishing, increased sedimentation, habitat degradation and introduction of predatory and competitor fish (Cadwallader, 1981;Trueman, 2011).
Over the past two decades, strict fishing regulations have been applied, and major habitat rehabilitation of the Ovens River included livestock removal, restoring riparian revegetation and in-stream woody habitat, pest control and reintroduction of another extirpated threatened fish species (Department of Environment, Land, Water and Planning, 2017;Lyon et al., 2012).
Macquarie perch reintroduction into the Ovens River began in 2011 with annual stocking of fingerlings produced at the Snobs Creek Hatchery by pairing broodstock males and females from Lake Dartmouth or Yarra River populations. Lake Dartmouth, formed by damming the Mitta Mitta River in 1979, contains the largest remaining MDB population of the species, with fish residing in the lake and extending into the upper Mitta Mitta River and adjacent streams (Tonkin et al., 2014). Although a naturally riverine species, Macquarie perch can occupy reservoirs, given access to suitable survival of stocked offspring of two-population ancestry. Thus, combining compatible stocks is likely to benefit restoration of other wildlife populations.

K E Y W O R D S
genetic diversity, genetic management, genetic monitoring, multiple source populations, offspring survival, population reintroduction, recruitment, stocking, translocation riverine habitat for spawning (Cadwallader, 1981). The Yarra River population was established between 1907 and 1943 from multiple southern MDB sources (Cadwallader, 1981;Trueman, 2011) and is the most genetically diverse MDB population, containing unique genetic variation now extinct at the sources (Pavlova et al., 2017).
Although Dartmouth and Yarra populations have high genetic diversity and effective population sizes compared to those of other Macquarie perch populations, both can be expected to have compromised ability to adapt to environmental changes, which could be improved through increased genetic variation (Pavlova et al., 2017).
Single-source crosses were used to produce fingerlings for stocking until breeding seasons 2014-2017, when interpopulation (Dartmouth × Yarra) crosses replaced Yarra × Yarra crosses in the conservation breeding programme (Table 1, Figure 1a,b). From 2014 to 2017, stocking was combined with annual translocations of juvenile fish from Lake Dartmouth, which at the time had abundant young fish following the breaking of the "Millennium drought" (Tonkin et al., 2014). Small numbers of adults from Lake Dartmouth, and retired broodstock from both populations, were also released into the Ovens (Table 1, Figure 1c, Figure A2 in Appendix S1).
Several factors might potentially drive different levels of survival and recruitment in the Ovens River for fish of different ancestry.
First, Yarra and Dartmouth Macquarie perch could have phenotypic differences inhibiting free interbreeding (genomic incompatibilities are unlikely, Pavlova et al., 2017). Second, genetically diverse fish descended from multiple stocks may have higher fitness compared to fish of single-population ancestry due to heterosis, balancing selection or masking of deleterious alleles accumulated in isolated populations (Charlesworth, 2015;Ferchaud et al., 2018). Accordingly, higher fitness could be expected for the offspring of Dartmouth × Yarra pairs, and for offspring of two Yarra parents, themselves descended from multiple diverse stocks. Third, fish from one source might be better suited to conditions in the Ovens (Huff et al., 2010).
If this were true, it is more plausible that Macquarie perch from the Yarra River population were better suited to the Ovens than those from Dartmouth, because over the last 40 years (five generations, assuming a generation time of 7 years; Pavlova et al., 2017) Dartmouth fish have spent most of every year in lacustrine environments. Here, they could have experienced plastic changes, relaxed selection for riverine conditions and/or positive selection for lake conditions, potentially making them less suited to the Ovens River in only a few generations (Bell et al., 2004;Duncan et al., 2014;Lahti et al., 2009).
Adaptations to stream environments have not been assessed for Macquarie perch, but are known in fishes (Bunt et al., 2004).
Here, we use genotypic data from reduced-representation genome-wide markers in genetic analyses including identity, parentage, kinship and genetic cluster assignment and hybrid analysis.
These analyses were applied to identify, among individuals captured during annual monitoring in the Ovens River (2016-2018), hatchery-produced offspring, translocated fish and offspring of translocated fish, and to assign genetic ancestry to local recruits. We use these data to answer the following questions: (a) Do offspring of different genetic ancestry (i.e. of cross-types involving Dartmouthonly, Yarra-only or two-population parents) and more genetically dissimilar parents have higher survival following stocking? (b) Are genetic ancestry, and individual genetic diversity or parental dissimilarity, associated with individual growth rates? (c) Do fish of different genetic ancestry have different levels of recruitment in the Ovens? (d) Do Yarra and Dartmouth fish freely interbreed in the wild? By using a nonlethal genetic method of monitoring stocking and translocation success in a wild population, our study provides rare information on growth and survival of stocked fish, local recruitment, and fitness outcomes when multiple source populations are used to reintroduce an extinct population.

| Study system: artificial breeding, stocking and translocations to the Ovens River
Artificial breeding of Macquarie perch was performed at the Snobs Creek Hatchery using established methods (Ho & Ingram, 2013).
Broodstock were collected as "running ripe" fish from the Yarra River, and from Lake Dartmouth (captured during their annual temporary spawning migration into the Mitta Mitta River), in November-December of 2010-2017 (austral spring-summer). Eggs from each female were mixed with milt of one male and incubated separately; the number of hatched larvae was estimated before pooling (Table A1 in Appendix S1). Pooled larvae were grown to fingerlings (0.5-1 g) in fertilized earthen ponds. Fingerlings harvested from ponds were released the following January-February across 10 sites along a ~50 km stretch of the Ovens (>94,000 in total; Figure 1b, Table A2 in Appendix S1; Figure A1 in Appendix S1).
A total of 1801 fish were also translocated to the Ovens from 2014 to 2017 (Table A3 in Appendix S1), including wild-caught juveniles (<3-year old) and adults from Dartmouth (89.1% and 19.9%, respectively; N = 1739; Figure A2 in Appendix S1) and retired broodstock from Dartmouth (N = 32), Yarra (N = 27) and of unknown origin (N = 3). Fin-clip tissue samples for DNA analyses were collected from 98/108 broodstock (Table A1 in Appendix S1) and 1770/1801 translocated fish, and stored in absolute ethanol at −20°C.  (Table A4, Figure A1 in Appendix S1). These sites encompassed all stocking and translocation release sites, plus upstream and downstream of them. Fish were caught using boat-mounted electrofishing equipment (Smith-Root® Model 5 GPP), measured for total length (nearest mm) and weight (nearest g), fin-clipped and released at their capture site. TA B L E 1 Summary of management actions for each breeding season/year and inferences of origin (stocked, translocated and locally born) and genetic ancestry (DxD-Dartmouth; DxY-two-population, YxY-Yarra ancestry) for the Ovens fish captured during monitoring (data in Data S1) Ltd., using reduced-representation sequencing DartSeq™ protocols (Kilian et al., 2012), following Nguyen et al. (2018). SNP data were filtered using the dartR package (Gruber et al., 2018) in R (R Core Team, 2019). Unless otherwise stated, a single biallelic SNP per locus with reproducibility = 1 (estimated based on re-processing of ~25% samples) was retained, loci missing scores in >10% of individuals and 41 individuals missing >25% of genotypes were removed (Data S1;

| Fish monitoring
F I G U R E 1 History of stocking and translocation across seven breeding seasons and their outcomes inferred from genetic monitoring for Macquarie perch in the Ovens River: (a) number of broodstock pairs that produced larvae used for stocking, (b) number of fingerlings stocked to the Ovens (estimated based on number of larvae produced by each broodstock pair), (c) number of fish translocated to the Ovens, (d) number of broodstock pairs with offspring detected in the Ovens, (e) number of Ovens fish identified as stocked (i.e. those with broodstock parents assigned to them by genetic parentage analysis) and (f) summary of genetic ancestry for fish that was translocated, stocked, inferred to be stocked, and inferred to be locally born. Colours represent three cross-types: black-Dartmouth × Dartmouth (DxD), yellow-Yarra × Yarra (YxY), green-Dartmouth × Yarra (DxY). Due to lack of genetic samples for most of 2010/11 broodstock pairs, genetic outcomes for that season should not be interpreted. On (f), 92 Ovens fish inferred to be stocked include 29 DxD + 45 DxY + 18 YxY, and 200 Ovens fish inferred to be locally-born include 14 DxD + 67DxY + 119 YxY (with a low probability that one DxD and three YxY are stocked offspring of unsampled parents and six DxD are unsampled translocated fish); four translocated individuals identified in the Ovens are not plotted these were fish translocated from Dartmouth, including 27 that were mature in 2017). To ensure that loci with a signal of hemizygosity in one sex are excluded from analyses that assume codominant markers, the loci were tested for sex linkage in a panel of known-sex broodstock fish using the gl.sexlinkage dartR function: none were homozygous in one sex and heterozygous in the other. Different numbers of polymorphic loci were retained after removing monomorphic loci for various subsets of individuals (see below).

| Parentage, sibship and identity analyses
Data for 1679 individuals scored for 1204 loci were used for parentage analysis in CERVUS3 (Kalinowski et al., 2007) with error rate of 0.0001 (justification and details in Appendix B in Appendix S1, including calibration on known-parentage offspring). All Ovens fish were used as candidate offspring. Successful broodstock and translocated fish of breeding age (>2 years) in December 2017, representing potential parents of fish captured during the last round of monitoring in 2018, were used as candidate parents. We assumed that 70% of parents were sampled and 90% of loci genotyped. Simulations to calibrate likelihood thresholds for maternity and paternity assignments were run using 250 potential parents and 100,000 offspring.
COLONY2 (Jones & Wang, 2010) was used to conduct sibship and parentage analyses using the same dataset, assuming a polygamous, noninbreeding, dioecious diploid mating system, with a locus error rate of 0.0001 and 70% probability of parents being included.
Medium run length and full likelihood were used to estimate precision. We used the option to calculate allele frequencies but not to update allele frequencies after calculations, because family sizes and structure were unknown. Sibship scaling was applied.
Ovens individuals assigned as offspring of two broodstock parents from the breeding programme were inferred to be stocked, and their birth year and age at sampling calculated based on the year their parents were mated, assuming hatching on 1st of December.
To detect translocated fish, individuals sampled more than once, and potential mislabelling errors, we ran identity analysis in CERVUS3  Figure A2 in Appendix S1). Data S2). Offspring sampled in a particular age class were assumed present in all previous age classes. Next, the relative probability of offspring survival for each age class was estimated as the minimum number of surviving offspring for each age class divided by the number of stocked fingerlings produced by each pair (Table A1 in Appendix S1; estimated as the number of larvae produced by that pair divided by the total number of larvae produced that year by all broodstock fish, multiplied by the total number of fingerlings stocked that year into the Ovens).
We used generalized linear mixed models to relate the proportion of surviving offspring to parental cross-type (DxD, DxY, YxY) and genetic dissimilarity between broodstock parents (Appendix C in Appendix S1). We modelled proportional offspring survival as a binomial variable with logit link, with weights defined as the number of stocked fingerlings. We included year of stocking as a random effect to account for repeated measures on each cohort, as well as variation in survival due to environmental factors (e.g. streamflow, level of predation, population density or nutrient availability soon after stocking). We fitted separate models with each of cross-type and parental genetic dissimilarity, as well as an additive model including both. Because probability of fish survival increases with age and probability of detection differs between age classes, we also fitted these same models with an interaction between age and both fixed effects. We fitted all models with the glmer function in the R package lme4 (Bates et al., 2015) and compared models using Akaike's Information Criterion (AIC). We assessed model fit with pseudo-R 2 values calculated using the R package MuMIn (Barton, 2015) (Appendix C in Appendix S1).

| Modelling individual growth
For the 92 Ovens fish identified as stocked, we tested whether growth varied with cross-type, and either individual genetic diversity or parental dissimilarity. We fitted a standard Gompertz nonlinear regression model to length and age of Ovens fish identified as stocked and calculated residuals, reflecting the difference in their growth compared to the model average. These residuals were used as the response variable in linear mixed models in which crosstype (DxD, DxY, YxY), and genetic diversity or genetic dissimilarity between broodstock parents was fitted as fixed effects (Data S2; details in Appendix D in Appendix S1). We fitted models with all three predictors separately, as well as additive models of cross-type with genetic diversity or genetic dissimilarity, and used AIC values to identify the most parsimonious model structure (Appendix D in Appendix S1). We included year of stocking as a random effect in all models. We also used this Gompertz model to estimate age of Ovens fish unassigned to broodstock parents (i.e. inferred to be nonstocked), from their length.

| Population structure and recruitment
To investigate the possibility of local interbreeding between Dartmouth and Yarra fish, we assessed genetic ancestry of Ovens fish inferred to be locally-born (i.e. not assigned as stocked or translocated  (Keenan et al., 2013).
Then, we used the 407 loci most differentiated between Dartmouth and Yarra (F ST ≥ 0.02) in STRUCTURE analysis (Pritchard et al., 2000) to assign membership to two genetic clusters (K = 2, for Dartmouth and Yarra) for 157 unrelated Ovens fish inferred to be nonstocked (excluding all but one full-sibling detected by COLONY2), 23 similarly unrelated fish inferred to be stocked (with known ancestry), and 51 each from Dartmouth and Yarra (as above; 282 total). The sibling with the lowest level of missing data within each full-sibling group was used to represent its group. Twenty STRUCTURE replicates were run using 10 6 burnin and 10 6 MCMC samples; these replicates were summarized using CLUMPAK (Kopelman et al., 2015). For each Ovens fish of unknown ancestry included in the STRUCTURE analysis, we assigned Dartmouth, two-population or Yarra ancestry based on Q-value thresholds estimated for individuals of known Dartmouth or Yarra ancestry. For the full-siblings excluded from STRUCTURE, ancestry was assigned based on that of their included sibling. To test for local recruitment of DxY fish, we assigned locally born Ovens individuals to six ancestry classes (two sources, F 1 , F 2 , and two backcross types; details in Appendix B in Appendix S1) using NEWHYBRIDS (Anderson & Thompson, 2002).

| Parentage, identity and sibship analyses
Identity analysis revealed that among the fish sampled during monitoring of the Ovens River, six individuals were sampled in two monitoring seasons, and four had been translocated from Dartmouth (Appendix B, Table B2 in Appendix S1; seven sample mislabelling errors were also detected, which is 0.3% of 2173 fish analysed for identity). A total of 92 of 296 unique Ovens individuals (31%) were assigned both parents among known broodstock parent pairs and were thus inferred to be stocked (Figure 1e, Appendix B, Table B1 in Appendix S1). Of the 54 genetically sampled broodstock pairs that contributed to Ovens stocking (Table A1 in Appendix S1), 24 (44%) were assigned as parents of Ovens individuals sampled during monitoring (Table B1 in Appendix S1, Figure 1d), with up to 16 offspring per pair detected (median of 2.5 for parent-pairs assigned offspring).
In addition, four Ovens fish were assigned both parents among translocated Dartmouth fish, supporting local recruitment (Appendix B in Appendix S1). Forty full-sibship groups were detected by COLONY2 (Table B3 in Appendix S1; Data S1).

| Contribution of different years of the breeding programme, types of broodstock crosses and parental dissimilarity to relative survival of stocked offspring
More than 90% of 92 Ovens fish identified as stocked were born in November/December 2013, 2014 and 2015 (17, 53 and 13 individuals, respectively; Figure 1e), while only 64% of all fingerlings were released over those three years (40,500, 13,600 and 6400, respectively; Figure 1b). This suggests particularly good survival in some years.
Among the Ovens fish identified as stocked, 31% were offspring of DxD crosses, 20% YxY and 49% DxY (Figure 1e, Table 1 and Table B1 in Appendix S1), despite respective contribution of stocked fingerlings being 67% DxD, 23% YxY and 10% DxY (Figure 1b The most parsimonious model of survival included all predictor variables (parental cross-type, genetic dissimilarity between parents, offspring age class; Appendix C, Table C2 in Appendix S1). This model explained 40% of variation in relative offspring survival to all ages (Table 2). Parental cross-type and genetic dissimilarity were significant predictors of offspring survival, with survival highest in offspring of YxY pairs, followed by DxY, then DxD (Table 2). That is, cross-types characterized by the greater parental dissimilarity had better-surviving offspring. In addition to this main effect, there was a weak residual pattern that within cross-types, offspring from more genetically dissimilar pairs had slightly lower survival than those from less genetically dissimilar pairs (Table 2). However, this effect was inconsistent among cross-types (Figure 2d, Appendix C in Appendix S1). Although survival qualitatively decreased with age, there was no significant effect of offspring age class. An interaction term between age and cross-type and parental genetic dissimilarity was not significant, suggesting that all age classes had similar associations with cross-type and dissimilarity, despite age class 0 being over-represented through being inferred from captures when they were older fish. The ranking of models based on AIC and significance of cross-type and parental genetic dissimilarity as predictors of offspring survival did not change when YxY pairs were removed from the analysis (Tables C2 and C3 in Appendix S1), indicating that higher survival of YxY offspring was not driving the significant result.

| Growth of Ovens fish identified as stocked
Parental cross-type (genetic ancestry) was the only predictor included in the most parsimonious model of offspring growth (lowest AIC; Appendix D, Table D1 in Appendix S1). Models including crosstype and offspring genetic diversity (PHt) or parental genetic dissimilarity both had ΔAIC ≤ 2 (Table D1 in Appendix S1). All three models explained ~7% of the variation in residual growth, with cross-type F I G U R E 2 Variation in growth, age at sampling, genetic diversity, parental genetic dissimilarities and inferred relative offspring survival among three broodstock cross-types: black-Dartmouth × Dartmouth (DxD), yellow-Yarra × Yarra (YxY), green-Dartmouth × Yarra (DxY): (a) growth residuals plotted against genetic diversity (PHt) and (b) against age at sampling for each Ovens fish inferred to be stocked; (c) offspring genetic diversity plotted against parental genetic dissimilarity for each Ovens fish inferred to be stocked; (d) raw values of proportional offspring survival (the number of observed offspring of each age class divided by number of stocked fingerlings) plotted against parental genetic dissimilarity for each broodstock pair that contributed offspring to stocking; for each cross-type, linear trends are shown as black (DxD), yellow (YxY) or green (DxY) dashed lines, and means as crosses; the blue line shows a linear trend for offspring survival for all broodstock pairs combined. Note that (d) does not account for year of stocking, inclusion of offspring of the same pair sampled multiple times at different ages and decreased probability of offspring survival with age (this is done in the survival model; Table 2) being the only significant predictor of relative growth (Table D2 in Appendix S1). The offspring of YxY crosses stocked into the Ovens were bigger than DxD and DxY ones of the same age ( Figure 2b).
However, YxY and DxY crosses were not performed simultaneously ( Figure 1a), thus this comparison should be interpreted with caution. No other variables were significantly associated with offspring growth (Figure 2a, Table D2 in Appendix S1).

| Genetic ancestry of Ovens fish inferred to be locally born
Whereas of 296 unique Ovens individuals 92 (31.1%) were identified as stocked and four (1.4%) as translocated from Dartmouth (above), the remaining 200 (67.6%) were inferred to be locally-born (including four identified as DxD offspring of two translocated parents), although four of these 200 (three YxY and one DxD) were old enough to have been stocked in one of the three earliest seasons when one or more broodstock parents were not sampled (Table A1 in Appendix S1) and six DxD individuals were the same age as some unsampled translocated fish (below; Table 1 of Yarra origin, three were old enough to be stocked offspring of unsampled broodstock pairs, whereas the remaining 116 were too young for that, so must be local recruits (Table D4 in Appendix S1).
The proportion of Ovens fish inferred to be locally born of two-population and Yarra ancestry generally increased across years of monitoring due to an increase in the proportion of locally re-  Note: Probability of offspring survival after stocking to ages of <6 months, 1-year old, 2-year old and 3-year old, as a function of offspring age class, parental cross-type and genetic dissimilarity between members of a broodstock parent pair. This model included all age classes, with year of stocking fitted as a random factor to account for repeated measures of each cohort (see Appendix C in Appendix S1 for results of model comparisons and additional models). R2m is a likelihood ratio-based pseudo R 2 measure based on theoretically derived binomial variances. In our survival models, offspring survival was approximated by capture rates per year of life, where captured offspring were assumed to be present in earlier age classes. Because very young fish (<1-year old) were less likely to be captured during monitoring ( Figure A2 in Appendix S1), the first age class in our survival analysis was strongly influenced by captures of older fish inferred to be alive at age-0. Our assumption should not have strongly biased relative survival towards the presence of more YxY and DxY offspring, because YxY and DxY cross-types were less abundant in stocked fish than were DxD cross-type for all but one breeding season (Figure 1b). The exception was 2012/13 breeding season, in which 94% of all stocked offspring were estimated to be YxY (Table 1) and five offspring of four YxY pairs were captured during monitoring (Table B2 in Appendix S1). Nevertheless, our offspring survival model yielded the same result even when run without YxY pairs (Table C3 in Appendix S1). Moreover, the nonsignificant age interaction term suggested that the relationships between capture rates and cross-type and genetic dissimilarity were similar across all age classes. It is unlikely that very small variation in size within each age class (Figure 2b) biased capture rates in a way that could have impacted our analyses. Only YxY fish were inferred to be significantly According to theory and empirical evidence, augmenting genetic diversity through mixing suitable stocks can improve population survival and adaptive potential (Frankham et al., 2017;Weeks et al., 2015). As such, the Yarra population of Macquarie perch, founded early in the 20th century by fish from various tributaries of the Murray River, represents a successful example. Given relatively high genetic diversity of the Yarra and Dartmouth populations, reintroducing from two sources into the Ovens was intended mainly to enhance population adaptive potential, not so much to alleviate inbreeding depression, although higher fitness of offspring of two-population and Yarra ancestry could be expected under heterosis, masking deleterious alleles or balancing selection (Charlesworth, 2015;Ferchaud et al., 2018). Our findings of disproportionately many local recruits and superior survival of stocked offspring of Yarra × Yarra, followed by Dartmouth × Yarra crosstypes ( Figure 1f, Table 2) count against Dartmouth × Yarra heterosis, but support, as potential drivers, balancing selection associated with Yarra variants, or masking deleterious alleles accumulated in Lake Dartmouth. Cross-type had a strong association with survival of stocked offspring, in the expected direction of greater survival with greater interparental dissimilarity. In comparison, a negative within-cross-type effect of dissimilarity was very weak (explanation in Appendix C in Appendix S1) and inconsistent across cross-types On the other hand, higher fitness of Yarra-than Dartmouthlineage fish in the Ovens, associated with differences between riverine and lacustrine environments, could also explain superior survival and recruitment of fish of Yarra-and two-population ancestry. The Dartmouth fish for broodstock and translocations were collected from Lake Dartmouth or during annual spawning migrations into its main tributary, the Mitta Mitta River, immediately upstream. Hence, all or most of these fish are likely Lake residents, accessing riverine environments only for spawning (Tonkin et al., 2018). Although fish experience similar spring spawning temperatures and flow variability across populations (Tonkin, Kearns, Fanson, et al., 2017;Tonkin, Kearns, Lyon, et al., 2017), the conditions experienced by fish during summer are likely very different. Unlike the lacustrine population, riverine populations often experience low flow, extreme habitat contractions, prolonged exposure to high water temperatures and likely, low food availability (Tonkin et al., 2014(Tonkin et al., , 2019. These stressful conditions regulate population dynamics (Tonkin et al., 2019)  through loss of adaptation to extreme summer conditions, adaptation to lake environments and/or plasticity, Dartmouth fish may be less suitable for translocation to riverine habitats than are fish from the Yarra River. Relaxed selection pressures in the lake for stressful summer conditions, especially during Macquarie perch population growth in Lake Dartmouth after initial filling and refilling after a prolonged drought (Tonkin et al., 2014), could have promoted accumulation of mutations suboptimal in riverine environments, resulting in relatively poor performance of Dartmouth fish translocated to the Ovens River (Figure 4). Loss of adaptive traits under relaxed selection and gain of locally adaptive traits can manifest in fish after a few generations (Bell et al., 2004;Fraser et al., 2011;Lahti et al., 2009) (Pavlova et al., 2017). Very few broodstock fish could now be sourced from the Yarra due to a decline of large adults, likely contributed to by overfishing (Tonkin, Kearns, Fanson, et al., 2017). Our findings here support the recent close to the recreational fishery in the Yarra, due to its importance for conservation management of the species (Pavlova et al., 2017;Tonkin, Kearns, Fanson, et al., 2017;Tonkin et al., 2019). Attempts to recover and augment Macquarie perch populations through stocking of hatchery-bred fish and translocation of wild ones have been undertaken intermittently since the 1980s with mixed success (Lintermans, 2013;Commonwealth of Australia, 2018). The crucial role of Yarra source in the success of the Ovens population rehabilitation suggests that restoration efforts involving movement of fish from impoundments into riverine populations might be strengthened by additionally using river sources.
Our results support broader use of multiple sources for genetic augmentation during population rehabilitation and (re)introductions into former or new ranges. The need for such genetic management increases under changing climates. Frequency of catastrophic events such as prolonged heat waves, droughts, fires and floods are predicted to intensify, and so many isolated wildlife populations will be unable to adapt in situ or move to suitable habitats. In this case, they could benefit from assisted gene flow or translocation to a new habitat (Frankham et al., 2017), because more genetically diverse populations better survive climatic extremes (Reusch et al., 2005).
Given current rates of climate change and habitat loss and degradation, elevating population adaptive potential through genetic augmentation is essential for efficient conservation.

| CON CLUS IONS
In agreement with a wealth of evolutionary theory about the genetic basis of fitness and adaptation, our study adds to growing evidence that using multiple compatible sources can result in substantial fitness benefits, especially for small and isolated populations (Frankham, 2015). Although the evolutionary mechanisms behind higher fitness of individuals of mixed ancestry might not always be readily distinguishable, using multiple sources during population reestablishment provides populations with higher effective population sizes and genetic diversity, thus greater resources for selection on phenotypic variation suited to the environment. Whereas a given source population might seem preferable due to higher fitness, it might not be always available for large-scale translocations required to create initially viable populations, and other sources still contribute useful genetic variation through hybrids and backcrosses, as in the present case. Genetic monitoring will be beneficial to ensure backcrosses are produced, population sizes are increasing and effective size remains sufficient, with additional translocations performed if required (Weeks et al., 2015).

DATA AVA I L A B I L I T Y S TAT E M E N T
Genotypic data for this study are available at Bridges research repository at https://doi.org/10.26180/ 5ea17 36cc3553.