Severe consequences of habitat fragmentation on genetic diversity of an endangered Australian freshwater fish: A call for assisted gene flow

Abstract Genetic diversity underpins the ability of populations to persist and adapt to environmental changes. Substantial empirical data show that genetic diversity rapidly deteriorates in small and isolated populations due to genetic drift, leading to reduction in adaptive potential and fitness and increase in inbreeding. Assisted gene flow (e.g. via translocations) can reverse these trends, but lack of data on fitness loss and fear of impairing population “uniqueness” often prevents managers from acting. Here, we use population genetic and riverscape genetic analyses and simulations to explore the consequences of extensive habitat loss and fragmentation on population genetic diversity and future population trajectories of an endangered Australian freshwater fish, Macquarie perch Macquaria australasica. Using guidelines to assess the risk of outbreeding depression under admixture, we develop recommendations for population management, identify populations requiring genetic rescue and/or genetic restoration and potential donor sources. We found that most remaining populations of Macquarie perch have low genetic diversity, and effective population sizes below the threshold required to retain adaptive potential. Our simulations showed that under management inaction, smaller populations of Macquarie perch will face inbreeding depression within a few decades, but regular small‐scale translocations will rapidly rescue populations from inbreeding depression and increase adaptive potential through genetic restoration. Despite the lack of data on fitness loss, based on our genetic data for Macquarie perch populations, simulations and empirical results from other systems, we recommend regular and frequent translocations among remnant populations within catchments. These translocations will emulate the effect of historical gene flow and improve population persistence through decrease in demographic and genetic stochasticity. Increasing population genetic connectivity within each catchment will help to maintain large effective population sizes and maximize species adaptive potential. The approach proposed here could be readily applicable to genetic management of other threatened species to improve their adaptive potential.


| INTRODUCTION
A primary goal of conservation management is to improve the adaptive potential of populations, that is, the ability of populations to adapt and persist in the face of environmental changes (Sgrò, Lowe, & Hoffmann, 2011). Genetic diversity underpins adaptive potential, but is lost in small populations through genetic drift, which can lead to loss of fitness, accumulation of genetic load and ultimately population extinction (Harrisson, Pavlova, Telonis-Scott, & Sunnucks, 2014;Reed & Frankham, 2003;Willi, Van Buskirk, & Hoffmann, 2006). Through reduction or cessation of gene flow, habitat fragmentation reduces species ranges to small populations at high extinction risk, contributed to by environmental, demographic and genetic factors (e.g. inbreeding depression, genetic load and inability to rapidly adapt to environmental changes) and their interactions (Benson et al., 2016;Fountain, Nieminen, Sirén, Wong, & Hanski, 2016;Frankham, 2005;Lopez, Rousset, Shaw, Shaw, & Ronce, 2009). Given that taxa of conservation concern have typically undergone habitat fragmentation and/or strong population contractions, genetic issues should be addressed in recovery or management plans, although this is often not done (Pierson et al., 2015(Pierson et al., , 2016. Effective and efficient allocation of conservation resources is best assisted by information about the distribution of genetic variation across landscapes (Hoffmann et al., 2015). Spatial patterns of genetic diversity and genetic differentiation reflect stochastic and environmental effects on key demographic and evolutionary processes (e.g. population size, gene flow, adaptive potential) linked to viability. Thus, genetic data can inform about the conservation status of populations, improve predictions about population responses to environmental change and management interventions, and support the development of effective conservation strategies for enhancing population viability (Willi & Hoffmann, 2009). For a given population or species, a direct demonstration of genetic problems, such as loss of fitness due to decline in genetic diversity or inbreeding, presents a strong case for genetic management. In contrast, unavailability of evidence for genetic problems in a specific case often prompts an argument against genetic intervention until such evidence is collected. This position ignores evidence acquired from numerous wild and captive systems that small and isolated populations will most likely be facing genetic problems with negative demographic consequences (Keller & Waller, 2002;Saccheri et al., 1998;Woodworth, Montgomery, Briscoe, & Frankham, 2002), reflecting cultural rather than evidence-based decisions concerning genetic management (Love Stowell, Pinzone, & Martin, 2017). Obtaining data on population fitness is challenging and expensive and may take years. Postponing decision-making until such data are available, or disregarding genetic problems completely, is likely to result in managing small populations to extinction (Frankham et al., in press;Love Stowell et al., 2017;Weeks, Stoklosa, & Hoffmann, 2016). Instead, the augmentation of gene flow is a powerful conservation management option for counteracting loss of genetic diversity resulting from drift in small populations, with the potential to promote positive demographic outcomes (Frankham, 2015;Hufbauer et al., 2015;Whiteley, Fitzpatrick, Funk, & Tallmon, 2015). Genetic augmentation to alleviate detrimental effects of inbreeding and/or genetic load in small, isolated populations (i.e. genetic rescue) and/or increase levels of genetic diversity and adaptive potential (i.e. genetic restoration) can be achieved through reconnection of habitat or human-assisted translocations (Attard et al., 2016;Whiteley et al., 2015). Strong precedents supporting the overwhelmingly positive and long-lasting effects of genetic rescue on fitness exist across a broad range of taxa (Frankham, 2015(Frankham, , 2016. By bringing in new genetic diversity, genetic augmentation is expected to improve adaptive potential and probability of persistence of small and isolated populations, even if they do not yet suffer from inbreeding depression and genetic load (Willi et al., 2006).
Another common argument against genetic augmentation is that gene flow between divergent populations may "swamp" local adaptation and distinctiveness, or contribute to outbreeding depression ; Le Cam, Perrier, Besnard, Bernatchez, & within a few decades, but regular small-scale translocations will rapidly rescue populations from inbreeding depression and increase adaptive potential through genetic restoration. Despite the lack of data on fitness loss, based on our genetic data for Macquarie perch populations, simulations and empirical results from other systems, we recommend regular and frequent translocations among remnant populations within catchments. These translocations will emulate the effect of historical gene flow and improve population persistence through decrease in demographic and genetic stochasticity. Increasing population genetic connectivity within each catchment will help to maintain large effective population sizes and maximize species adaptive potential.
The approach proposed here could be readily applicable to genetic management of other threatened species to improve their adaptive potential.

K E Y W O R D S
adaptive potential, effective population size, genetic rescue, genetic restoration, inbreeding depression, Macquarie perch Macquaria australasica, management, population persistence Evanno, 2015;Love Stowell et al., 2017). Such concerns frequently lead to recommendations to maintain isolation of differentiated populations, but without appropriate assessment of the risks and benefits of gene flow (Farrington, Lintermans, & Ebner, 2014;Frankham et al., 2011;Roberts, Baker, & Perrin, 2011). Genetic differentiation per se is not a clear indicator of risk that populations are differently adapted and gene flow might be harmful: some differentiation is inevitable under restricted dispersal, particularly among small populations, and is often caused by recent human impacts (Cole et al., 2016;Coleman, Weeks, & Hoffmann, 2013;Faulks, Gilligan, & Beheregaray, 2011).
Even where genetic differentiation is associated with adaptive divergence, locally adapted traits can still be maintained in a population in the presence of gene flow from a differently adapted population (Fitzpatrick, Gerberich, Kronenberger, Angeloni, & Funk, 2015), unless the level of gene flow is so large that it overwhelms natural selection (Lowe & Allendorf, 2010). The potential benefits of gene flow are large, and any risks of outbreeding depression must be weighed against the risk of extinction due to inaction, for example from inbreeding depression, genetic load or inability to adapt to a novel selective pressure (Becker, Tweedie, Gilligan, Asmus, & Whittington, 2013;Fisher, Garner, & Walker, 2009;Frankham et al., in press;Lintermans, 2013).
In developing plans for genetic rescue and/or restoration, four key questions must be considered:

1.
What level of genetic variation should trigger genetic rescue/ restoration?

2.
Which populations can be used as sources for translocations?

3.
What is the risk of outbreeding depression? and 4. How much gene flow is required to rescue the populations at risk and restore their adaptive potential? Recent reviews provide guidelines: 1. Effective population size (N e ) ≥100 is required to prevent inbreeding depression in the short-term and limit loss of fitness to ≤10% over 5 generations. A much larger global effective population size, N e ≥ 1,000, is required to retain adaptive potential, by maintaining an equilibrium between the accumulation of genetic variation for a selectively neutral trait by mutation, and the loss variation by random genetic drift (Frankham, Bradshaw, & Brook, 2014). Based on a combination of theory and empirical evidence, the numbers from previously proposed 50/500 rule (Franklin, 1980) were found to be insufficient to prevent inbreeding depression/avoid loss of genetic variation for fitness (Frankham et al., 2014).

2.
Any population only recently diverged from a genetically depleted one can be used as a source, but the magnitude of genetic rescue is greater if the source population is outbred and/or differentiated (Frankham, 2015).

3.
The risk of outbreeding depression will be low if populations have the same karyotype, were isolated for <500 years and are adapted to similar environments .

4.
If the risk of outbreeding depression is low, then up to 20% gene flow from the source population will improve population adaptive potential; ongoing monitoring within an adaptive management framework (i.e. iterative process of decision-making based on monitoring of outcomes) is recommended to avoid genetic swamping and outbreeding depression (Hedrick, 1995;Whiteley et al., 2015).
The Macquarie perch Macquaria australasica Cuvier 1830 is an endangered Australian endemic freshwater fish species that has undergone substantial population decline and range contraction and fragmentation.
Prior to European colonization in the late 18th century, the Macquarie perch was widespread, occurring in the inland Murray-Darling Basin (MDB), and across three coastal drainage basins, Hawkesbury-Nepean (HNB), Shoalhaven and Georges (Figure 1). Although distinct taxa within Macquaria australasica are not formally recognized, morphological, allozyme and DNA data all indicate that the inland and coastal forms of Macquarie perch diverged in the Pleistocene and probably represent different species (Dufty, 1986;Faulks, Gilligan, & Beheregaray, 2010a, 2015Faulks et al., 2011;Knight & Bruce, 2010;Lintermans & Ebner, 2010;Pavlova et al., 2017). Over the last two centuries, Macquarie perch suffered strong declines, particularly in the lowland reaches of rivers, and as a result the species is now restricted to isolated locations in the headwaters of Lachlan, Murrumbidgee and Murray tributaries in the MDB, Hawkesbury and Nepean rivers in the HNB, and Georges River in the Georges Basin ( Figure 1) (Gilligan, McGarry, & Carter, 2010;Knight & Bruce, 2010;Lintermans, 2007). A highly differentiated form in the Shoalhaven Basin is thought to have become extinct by 1998, which underscores the urgent need to understand and manage remaining genomic diversity within the species (Faulks et al., 2010a). Many anthropogenic factors have been implicated in the species' decline across its range, including loss of 95% of habitat, fragmentation by barriers to movement, increased river sedimentation, alien fish species, overexploitation and altered flow regimes reducing previously connected habitats to small fragments (Cadwallader, 1978;Ingram, Douglas, & Lintermans, 2000;Ingram et al., 1990;Pollard, Ingram, Harris, & Reynolds, 1990). Loss of physical and genetic connectivity across the species' range is expected to have contributed to its ongoing decline, although establishing that directly is challenging given the lack of historical genetic samples.
In attempts to reverse its decline, the Macquarie perch has been subject to various management actions including habitat restoration, threat amelioration, wild-to-wild translocations and captive breeding with subsequent stocking of captive-bred fish to re-establish extinct populations, augment threatened ones and support recreational fisheries (Ho & Ingram, 2012;Lintermans, 2012;Lintermans, Lyon, Hammer, Ellis, & Ebner, 2015). Despite substantial financial investment and effort, these management actions have so far failed to reverse the decline of the species. Indeed, water scarcity under recent droughts has made several of the remaining populations exceedingly small and vulnerable to environmental, demographic and genetic stochasticity. Historically, vulnerable or extinct populations would have been recolonized from neighbouring populations when flow returned, but this is no longer possible due to insufficient physical connectivity | 535 PAVLAP et PVl. between remnant habitat patches. As others have stated (Love Stowell et al., 2017;Whiteley et al., 2015), we argue that although the main reasons for population decline (e.g. habitat availability and degradation) should be addressed by conservation management, boosting genetic diversity through assisted gene flow might be necessary to improve population fitness and/or adaptive potential and decrease extinction probability of small populations. Management decisions need to be made now based on the best available information. When direct information on genetic problems in a specific case is unavailable, drawing on accumulated evidence from empirical, theoretical and simulation studies of other systems is a more evidencebased approach than making the unvalidated assumption that there are no genetic problems (Frankham et al., in press;Love Stowell et al., 2017;Weeks et al., 2016).
Here, we explore patterns of genetic diversity and genetic differentiation across the entire range of the Macquarie perch and expand on earlier genetic work that revealed deep divisions within the named species, hybridization among inland and coastal forms, strong population substructure and environmental factors affecting the distribution of population genetic diversity (details in Appendix S1; Faulks et al., 2010aFaulks et al., , 2011Pavlova et al., 2017). For the first time, we estimate effective population sizes across the species range, evaluate them against established thresholds (Frankham et al., 2014) and demonstrate by simulations that within a few decades, smaller Macquarie perch populations are likely to suffer erosion of genetic diversity and inbreeding unless genetic diversity is restored and maintained by translocations.
In addition, we identify novel environmental variables associated with individual-based heterozygosity and improve earlier estimates of population history, key divergence times, population genetic diversity and differentiation, and population structure at basin-and catchment scales by analysing additional samples and populations, additional microsatellite loci and longer mitochondrial sequences. Our results will contribute to developing separate management plans for appropriate evolutionary units, the selection of broodstock source populations for the current hatchery breeding programmes and the identification of critical locations for stocking and translocations. The processes of habitat loss, degradation and fragmentation that have affected the previously common and widespread Macquarie perch are similar to those affecting many species including, but not limited to, other threatened MDB species that have experienced recent population extinctions, for example trout cod Maccullochella macquariensis, southern pygmy perch Nannoperca australis and southern-purple spotted gudgeon Mogurnda adspersa Lintermans, 2007;Trueman, 2011); hence, the workflow here could be readily applicable to genetic management of other threatened species to improve their adaptive potential. and a Shoalhaven Basin population (a single known genetic sample from Kangaroo River). Individuals were captured using various electrofishing, netting or angling techniques. Most fish were measured for length and weight and immediately released at the site of capture after collection of fin clips. Fin tissue was preserved in 100% ethanol and stored at −20°C. DNA was extracted using a QiPg n DNeasy tissue extraction kit, unless available from other projects (Faulks et al., 2010a(Faulks et al., , 2011Nguyen, Ingram, Lyon, Guthridge, & Kearns, 2012).
Twenty individuals with scores missing for more than three loci were excluded, as well as a single Dartmouth individual that returned a F I G U R E 1 Geographic distribution of Macquarie perch samples analysed for this study (top), mitochondrial control region haplotype network showing distribution of haplotypes across populations (middle; Shoalhaven haplotype not included) and individual memberships in two or 12 genetic clusters inferred from microsatellite data (major structure inferred by K = 2 and K = 12 setrucetur analysis of all HNB and MDB samples; see Appendix S9; Shoalhaven individual not included). Two colour stripes above setrucetur plots are colour codes for basins (top) and populations (bottom), as on map. Yarra and Cataract Dam populations are translocated from the MDB; Cataract River comprises hybrids between endemic HNB lineage and fish dispersed from Cataract Dam (details in Appendix S9). On the haplotype network, populations are coloured as on the map (population labels are also given for each colour); each circle is a unique haplotype (number within-haplotype ID, small black circle -missing haplotype); the size of each circle is proportional to the haplotype frequency; connections between circles are single substitutions unless marked with a number (of substitutions). In five locations, a single haplotype is fixed (1 in Abercrombie, 10 in Cotter, 41 in Murrumbidgee, 32 in Glenbrook, 42 in Wheeny) triploid genotype (confirmed by repeated genotyping of re-extracted DNA). Final analyses included genotyping scores for 871 individuals from 20 populations (Appendix S2; Supporting Information); only 0.3% of scores were missing. Tests for deviation from Hardy-Weinberg and linkage equilibria were performed using g n L 4.2 (Rousset, 2008) (Appendix S3).

| Estimating population history and key divergence times
To evaluate the level of mitochondrial haplotype-sharing across populations, indicative of past gene flow, a median-joining network (Bandelt, Forster, & Roehl, 1999) was constructed in n etwLrk v4.6.1.0 (www.fluxus-engineering.com). Times of divergence among major mitochondrial lineages were previously estimated from complete mitogenomes (Pavlova et al., 2017; summarized in Appendix S1); however, some lineages were not represented. Using mitochondrial haplotypes, we estimated the times of lineage divergence with B Pset 1.8.2 (Drummond, Suchard, Xie, & Rambaut, 2012), assuming the HKY+G+I substitution model, coalescent tree prior and Bayesian skyline (with 10 groups and piecewise-constant skyline model) as population size prior. A strict molecular clock was used, based on results of complete mitogenome analysis (Pavlova et al., 2017); the control region clock rate, estimated based on a wide clock rate prior in that analysis, was used here as a prior (i.e. lognormal clock rate with mean (in real space) = 0.05 and SD = 0.5; this corresponds to 95% of rates being sampled between 0.017 and 0.118 substitutions per site per million years). Three replicates were run for 100 million steps sampling every 100,000; runs were checked for convergence, combined and analysed in etrPc r 1.6.0 (Rambaut & Drummond, 2007) after discarding first 10% of samples from each run as burn-in.

| Genetic differentiation and tests for evolutionary timescale of divergence
PrV Quin 3.5 (Excoffier & Lischer, 2010) was used to calculate pairwise population F ST , R ST (microsatellites) and Φ ST (mtDNA sequences) between catchments. SPAGeDi ver 1.5 (Hardy & Vekemans, 2002) with 10,000 permutations of microsatellite allele sizes was used to test if populations have been evolving independently long enough to evolve new microsatellite alleles through mutation; locus AB009 was excluded, because it was monomorphic in all but one population.

| Assessing population structure at basin-and catchment scales
setrucetur (Pritchard, Stephens, & Donnelly, 2000) was used to identify genetic clusters from microsatellite genotypes (data on population of origin were not applied). We first tested for general population structure across the species range using all 870 MDB and HNB individuals. Twenty replicates of 10 6 burn-in iterations followed by 3 × 10 6 iterations were run for each K from 1 to 16 (the maximum was set according to preliminary runs showing no meaningful structure above K = 12) using the admixture model with correlated allele frequencies.
Runs were summarized using the web server cVumPk (Kopelman, Mayzel, Jakobsson, Rosenberg, & Mayrose, 2015). The largest number of genetic clusters resulting in coherent geographic groups was considered to best represent population structure.
Three subsets of the data were further analysed using setrucetur : (i) a data set of 671 individuals of MDB origin (10 MDB populations, Cataract Dam and Yarra) was run for K max = 16 to detect major population subdivisions within MDB; (ii) a data set of 375 individuals from the Murray and Yarra catchments (Dartmouth, Hollands, Sevens, Hughes, King Parrot and Yarra populations) was run for K max = 7 to explore the origin of two previously detected genetic clusters in Dartmouth and Yarra (Nguyen et al., 2012); and (iii) a data set of 134 individuals from six HNB populations (all except the introduced Cataract River and Cataract Dam samples) was run for K max = 7 to test for presence of structure within the HNB.

| Analyses of genetic diversity, proxy for historical effective population sizes
For mtDNA sequencing data, PrV Quin was used to calculate haplotype diversity (Hd), nucleotide diversity (π), the number of segregating sites (S) as well as Tajima's D (Tajima, 1996) and Fu's Fs (Fu, 1997)

| Analyses of contemporary effective population sizes
We estimated N e using two single-sample methods, both of which assume that population size is stable over a few generations. First, we used the approximate Bayesian computation method based on eight summary statistics (including linkage disequilibrium, LD) implemented in Ln sPm (Tallmon, Koyuk, Luikart, & Beaumont, 2008); 50,000 populations were simulated using a prior N e range of 4-1,000 individuals. Ln sPm was also run with a prior N e range of 4-500, to confirm estimate consistency. Second, an LD-based method was used (LDNe; Waples & Do, 2008), implemented in n setimPetLr V2.0 (Do et al., 2014); random mating model and P Crit = 0.02 were applied.
LDNe estimates were interpreted only for samples of ≥50 individuals, as smaller samples are not informative when true N e > 100 . When estimated from a single cohort of an iteroparous species, an LD-based estimate of N e reflects the harmonic mean of the effective number of breeders in one reproductive cycle (N b ) and effective size per generation (N e ), but when estimated from mixedage sample, it approximates effective size per generation, albeit consistently downwardly biased (50-90% of true N e , least biased when the number of cohorts in a sample is close to the length of the generation time) (Waples, Antao, & Luikart, 2014). Whereas N b reflects short-term effective population size relevant to inbreeding, N e is responsible for shaping long-term evolutionary processes ; thus, both would be required for evaluating two parts of the 100/1,000 N e threshold (Frankham et al., 2014). Using simulations, Waples, Luikart, Faulkner, and Tallmon (2013) showed that the N b /N e ratio can be approximated using two life-history traits.
According to their formulae, for Macquarie perch N b /N e = 1.156 (assuming age at maturity of 3 years and adult life span of 23 years (Appleford, Anderson, & Gooley, 1998;Lintermans & Ebner, 2010)); thus, an LD-based estimate of N e should reasonably approximate both N b and N e , provided that other biases can be accounted for.
Macquarie perch samples including ≥7 cohorts would be least biased, assuming a generation time of 7 years (calculated in ALret x (Lacy, Miller, & Traylor-Holzer, 2015), below). According to the range of individual fish lengths in each population (Supporting Information) and the modelled relationship between length and age for Macquarie perch

| Identifying environmental variables associated with individual-based genetic diversity
We used an individual-based Bayesian modelling approach to identify environmental variables that were correlated with spatial distribution of genetic variation (Harrisson, Yen, et al., 2016). The model was based on the 16 populations from the MDB and the HNB (translocated Yarra and Cataract Dam and admixed Cataract River samples were excluded).
The response variable was homozygosity-by-locus (HL), calculated for each individual using the Rhh package (Alho, Valimaki, & Merilä, 2010) in R 3.2.0 (R Development Core Team 2014). HL measures the proportion of homozygous loci per individual, attributing greater weight to homozygosity at more variable loci (Aparicio, Ortego, & Cordero, 2006). Low HL of an individual indicates low genetic similarity of its parents, which is expected in populations with high genetic diversity. Using an individual-based measure of genetic diversity, rather than a population-based one (such as AR), enabled us to consider environmental variation within populations (Appendix S4) rather than population averages, which should resolve environmentalgenetic associations at finer spatial scales.
Eleven environmental variables reflecting flow regime, connectivity, habitat and climate, previously shown to be indicators of Macquarie perch presence and population health (Appendix S4), were originally considered for inclusion in the model as predictors of individual genetic diversity (  Fivefold cross-validation was used to estimate the predictive capacity of the fitted model and to test whether the model was likely to be identifying true relationships. To account for possible correlation of HL among individuals from the same genetic clusters due to shared history, cross-validation test data sets comprised nine population clusters, with approximately 120 individuals in each test data set (1-5 population clusters; Appendix S5). We used microsatellite clusters, rather than phylogenetic groups, because recent history (e.g. drift in small populations) could have a strong effect on HL, which can differ across distinct genetic clusters. Model predictions for cross-validation were based on environmental variables only; clustering variables (basin and site) were used in model-fitting but not to make predictions.

| Simulating genetic rescue/genetic restoration of isolated Macquarie perch populations
To demonstrate genetic rescue and/or genetic restoration effects of translocations to small populations of Macquarie perch, we simulated the outcomes of two management scenarios (do-nothing vs. 50 years of translocation) for two pairs of populations using an age-structured population model in ALret x 10.1.6 (Lacy & Pollak, 2014). ALret x simulates effects of deterministic forces, and those of demographic, environmental and genetic stochasticity on population parameters, such as probability of survival, time to extinction, population size, allelic diversity, heterozygosity and inbreeding (approximated as homozygosity at modelled loci). Age-specific mortality was modelled based on Todd and Lintermans (2015) (details in Appendix S6). Inbreeding depression was modelled by assuming that each founder individual has unique recessive alleles (6.29 lethal equivalents per individual, the empirically derived ALret x default value). Fifty percent of inbreeding depression was assumed to be due to recessive lethal alleles: inbred offspring with two copies of the same recessive alleles die before reproduction. The translocation scenario assumed that six individuals (3 of each sex) were moved from larger to smaller populations each year for 50 years to approximate the impacts of repeated, modestsized translocations; designing the most cost-effective translocation programme is beyond the scope of this paper. All translocated adults were assumed to survive translocation. Genetic results were summarized for 1 mitochondrial and 19 microsatellite markers simulated based on observed allele frequencies. The first pair of populations had initial population sizes of 3,000 and 500 (modelled using allele frequencies of Dartmouth and King Parrot, respectively), and the second pair had initial population sizes of 300 and 100 (modelled using Cataract Dam and Murrumbidgee). Five hundred forward-in-time simulations were run for 100 years for each scenario using parameters outlined in Appendix S6. To compensate for the inability of ALret x to model large female fecundities, calculations were restricted to adults (Lacy et al., 2015).

| Population history, divergence time estimates and genetic differentiation
The mitochondrial haplotype network was consistent with historical isolation between the MDB and the HNB, historical isolation of the Lachlan catchment from the rest of the MDB, isolation of Adjungbilly-the only known remnant Macquarie perch population in the lower Murrumbidgee catchment-from other populations, and isolation among HNB populations ( Figure 1). Lack of haplotype-sharing among any HNB populations (Appendix S7) suggests that strong genetic drift operates in this basin, consistent with earlier findings from a mitogenome study (Pavlova et al., 2017). In contrast, haplotype-sharing across Murray River tributaries (Dartmouth, Hollands, Sevens, Hughes, King Parrot) indicated historical connectivity within the southern MDB, consistent with a continuous reconstructed historical distribution in the southern MDB (Trueman, 2011). Similarly, haplotype-sharing between Lachlan and Abercrombie (northern MDB) supported previously inferred contemporary connectivity between these two populations (Faulks et al., 2011). The translocated fish from the MDB population into Cataract Dam apparently bore T A B L E 1 Details of the environmental model predicting homozygosity-by-locus (HL, an inverse of genetic diversity): final set of seven variables (first column) and variables highly correlated to them (second column), predicted relationships with HL, probability of inclusion in the model (value >0.7 indicate a strong association with HL, shown in bold) and direction of the effect (positive means that a variable increases HL; negative means that a variable decreases HL). Environmental variables were sourced from National Environmental Stream Attributes Database.

Analysis of six HNB populations (all except Cataract River and Cataract
Dam; N = 134) assigned all HNB populations to different genetic clusters, except a single individual in the Little River had Q > 90% in the Kowmung cluster, suggesting a rare long-distance dispersal event or undocumented translocation (details of all analyses are in Appendix S11). setrucetur analyses did not support the previously reported two clusters within Dartmouth and Yarra (Nguyen et al., 2012).

| Analyses of genetic diversity and effective population sizes
Mitochondrial (Hd and π) and nuclear (AR and He) genetic diversity   Figure 3).  Effect of harvest (for translocation) on population size, heterozygosity, number of alleles or inbreeding of both source populations was minimal, but harvest increased the probability of extinction and decreased mean time to extinction for the smaller Cataract Dam (and to a lesser extent, the larger population at Dartmouth; Table 2, Appendix S6).

| The need for specieswide genetic management to prevent further loss of genetic diversity and boost adaptive potential
Using novel genetic data, we investigated whether genetic augmentation is likely to be beneficial and necessary for long-term F I G U R E 2 Estimates of mitochondrial and nuclear genetic diversity and effective population sizes (N e , Ln sPm: lower-upper 95% confidence limits, bars show mean N e ) for 19 populations (coloured as on Figure 1). Crosses indicate populations for which mtDNA was not sequenced here (estimates from partial control region (Faulks et al., 2010a) were the following: Wollemi Hd = 0.111, π = 0.0006; Erskine Hd = 0, π = 0; Kowmung Hd = 0.708, π = 0.0076). Inbreeding coefficient F e for each population is calculated as F e = 1 − He/He Dartmouth ; adjusted N e is calculated as 4.5× Ln sPm N e (see Section 4; lower and upper bounds are not adjusted) Assuming levels of inbreeding depression similar to those reported for other wild populations, our simulations support a decline in population viability unless there is active genetic management. We believe that these combined lines of evidence provide sufficient support for genetic augmentation to be implemented now, despite a lack of data on fitness decline, which may take too long to collect in order to be useful for preventing extinctions of many populations (Weeks et al., 2016).

Significant investment in habitat restoration and protection has
failed to lead to detectable recovery of some Macquarie perch populations (Mark Turner, GBCMA pers. comm.). Although no data on fitness loss exist, it is reasonable to assume that inbreeding depression might have contributed to this failure. Our simulations suggest that inbreeding depression may contribute to population declines and impede population recovery in the near future. Our estimates of N e suggest that most populations will have limited capacity to adapt to rapid environmental changes. Thus, genetic problems should be addressed in conjunction with other key threats to population persistence. Restoring the historical, within-basin genetic connectivity of Macquarie perch via assisted gene flow will improve genetic diversity and adaptive potential across the species' range. Our simulations indicate that small-scale translocations (6 adults/year) will result in rapid and long-lasting increase in genetic diversity (limited only by genetic diversity and divergence of the source(s)) and decrease in inbreeding and probability of population extinction. Although large source populations should not be impacted by appropriately limited harvest (e.g. for translocation or captive breeding), smaller sources may become more vulnerable to extinction, loss of genetic diversity and inbreeding. Therefore, where possible, management should consider reciprocal translocations and multiple sources.
Our ALret x model has a number of limitations: (i) offspring that do not survive to reproduction were not modelled (due to software limitation), but, given extreme fecundity (up to 110,000 eggs/female; Cadwallader & Rogan, 1977), strong purifying selection acting on juveniles could purge deleterious mutations, reducing inbreeding depression; (ii) individuals at the start of simulations were assumed to be unrelated, which, if incorrect, artificially increases the time when inbreeding depression starts to impact viability; (iii) mutations were not modelled, and thus, novel genetic diversity was not accounted for; (iv) adaptive capacity was not addressed; (v) there are large uncertainties around environmental effects and population parameters, including an assumption that all individuals survive translocation.
Nevertheless, our simulations are in agreement with previous simulation studies (Allendorf & Ryman, 2002;Rieman & Allendorf, 2001) that under management inaction, populations of 100-500 adults will lose genetic diversity over time (>10% of alleles and >10% of heterozygosity in 100 years), and the loss will be faster in smaller populations.
F I G U R E 3 Fitted relationships between HL and environmental variables for variables with probability of inclusion greater than 0.7. The x-axis shows standardized values for each variable (e.g. a value of 1 means 1 standard deviation above the mean value for that variable), and the y-axis shows deviation from the mean HL value (in units of HL) for a given standardized value of the predictor variable. Grey shading is one standard deviation of the fitted effect. Models included basin and site as clustering variables, so these fitted effects account for differences among sites or between basins  Although in simulations inbreeding depression did not appear to impact the largest population (initial N = 3,000, ALret x N e = 257) for at least 50 years (~7 Macquarie perch generations), it started to impact three smaller populations (initial N ≤ 500, ALret x N e < 100) within the first 10 years (i.e. <2 generations), supporting the short-term inbreeding threshold N e of 100 (Frankham et al., 2014). Inbreeding can be approximated for a small population by scaling its genetic diversity by the diversity of a known outbred population using the effective inbreeding coefficient F e : F e = 1 − He inbred /He outbred , where He inbred is

heterozygosity (for neutral variation) of a population in question and
He outbred is heterozygosity of an outbred population (Frankham, 1998).
If Dartmouth, one of the largest natural populations of Macquarie perch in the MDB, is used as an outbred reference, then F e coefficients for Wollemi, Wheeny, Glenbrook, Adjungbilly, Cotter and upper Murrumbidgee are >0.45, and for Little is 0.25 (i.e. is as high as F e for the offspring of full siblings; Figure 2). At F e of 0.2, inbreeding depression is typically observed for populations of naturally outcrossing species (Frankham, 1995;Szulkin & Sheldon, 2007;Walling et al., 2011;Woodworth et al., 2002). Consistently, our simulations show that within a decade, inbreeding depression can be expected to act in Macquarie perch populations of ≤500 adults, unless translocations are performed. Although theoretically one effective migrant per generation may be sufficient to provide inbreeding connectivity (i.e. significantly reduce harmful effects of inbreeding; Mills & Allendorf, 1996), as well as adaptive connectivity (i.e. potential of highly advantageous alleles to spread in a recipient population; Rieseberg & Burke, 2001), it is not expected to drive F e below 0.2 (Lowe & Allendorf, 2010). In any case, well before F e climbs to 0.2, many populations will already be suffering inbreeding depression. A long-term study of inbreeding depression in the wild using genomic estimates of inbreeding detected major negative impacts on lifetime reproductive success at F even below drop F e values sufficiently (e.g. below 0.1) to rescue genetically depauperate populations from strong inbreeding depression (Frankham et al., in press;. Contemporary N e related to inbreeding is best estimated using two or more temporal genetic samples (inbreeding N e ; Luikart, Ryman, Tallmon, Schwartz, & Allendorf, 2010), or using a sample from a singlecohort and adjusting estimates using life-history traits . N e relevant to adaptive potential is best estimated from two or more genetic samples taken several generations apart and measuring the change in allele frequencies over time (variance N e ; Luikart et al., 2010). Our data were not suitable for temporal analyses: although sampling was spread over years, temporal samples spanning generations at individual sites were lacking, but sampling was random, com- global N e > 1,000 necessary to maintain population adaptive potential and evolvability in the face of environmental change . Measuring N e dynamics using multiple temporal genetic samples will allow managers to monitor the effects of assisted gene flow on N e . When genomic data including those on physical linkage or genomic position of genetic markers become available for Macquarie perch, contemporary N e and its recent change could be estimated from contemporary samples (Hollenbeck et al., 2016).

| Designing translocation strategies while considering risks of outbreeding depression and loss of local adaptation
Selecting appropriate sources of admixture requires simultaneous consideration of the timescale of population differentiation and the risks of inbreeding versus outbreeding depression and loss of local adaptation Love Stowell et al., 2017;Weeks et al., , 2016 (Pavlova et al., 2017). In accordance with mitochondrial differentiation, the microsatellite genotype of the single individual from the Shoalhaven Basin population differed markedly from those of the HNB and MDB lineages (Appendices S8 and S9).
The fact that the Shoalhaven individual was homozygous at 90% of loci suggests that loss of genetic variation and individual inbreeding accompanied extinction of the endemic Shoalhaven lineage. The widespread and not sex-limited hybridization beyond first generation of HNB and MDB lineages in lower Cataract River provides no evidence for outbreeding depression, but evidence for absence of genomic incompatibilities and unimpeded fitness of hybrids is lacking (Verhoeven, Macel, Wolfe, & Biere, 2011). Separate management of Macquarie perch lineages from different basins, which is current practice, is warranted until an informed risk-benefit analysis suggests otherwise (Frankham et al., 2011). Likewise, long-term (58-191 KY;Pavlova et al., 2017) divergence between the northern (here: Wollemi, Wheeny) and southern HNB populations (here: the remaining HNB populations) suggests that genetic rescue/restoration be conducted independently within these regions, unless an informed risk-benefit analysis indicates otherwise.
In some cases, nearby (presumably similarly adapted) source populations do not exist, so there may be no option but to source individuals for translocation from a more strongly divergent population (Love Stowell et al., 2017;, as suggested for threatened southern pygmy perch from the MDB based on genotype-environment association analyses .  (Frankham, 2015;Frankham et al., in press). This suggestion is in line with calls for focussing on preservation of genetic diversity at the species level, rather than population, subspecies or evolutionary significant unit (Love Stowell et al., 2017;Weeks et al., 2016).
In cases when it is unclear whether mitochondrial divergence confers selective advantage on local mitogenomes (such as Adjungbilly, inferred to have diverged from the other populations of the Murrumbidgee catchment ~8-100 KY ago), genetic rescue could involve within-catchment translocation of males. Assuming that on this timescale mitochondrial divergence did not drive mito-nuclear coevolution (Wolff, Ladoukakis, Enriquez, & Dowling, 2014), this translocation approach would preserve any potential selective advantage of local mtDNA while providing nuclear gene flow for increasing N e (Weeks & Corrigan, 2011).
In contrast to cases of long-term divergence, admixture of populations that have diverged only recently (including due to recent anthropogenic barriers to movement) should be associated with very low risk of outbreeding depression or loss of local adaptation (Frankham, 2015), whereas managing these populations separately will decrease their adaptive potential and increase species extinction risk (Coleman et al., 2013;Weeks et al., 2016). In the HNB, extreme divergence between the two northern Macquarie perch populations (Wollemi and Wheeny) is likely due to recent isolation by anthropogenic barriers (Faulks et al., 2011) and strong effects of genetic drift in very small populations (mean adjusted N e < 100). In the MDB, historical records indicate that prior to arrival of Europeans in Australia, Macquarie perch were common along the length of the Murrumbidgee River and in many tributaries of the Murray River (including lowland zone of the Central Murray River), suggesting that the Murrumbidgee and Murray catchments were previously connected by gene flow (Trueman, 2011).
Historically translocated populations, that have been isolated for sufficient generations to accumulate allelic differentiation, represent viable sources for translocations back into their population of origin if they contain diversity lost from the recipient population (Frankham, 2015  Although we found some potentially meaningful environmental correlates of individual genetic diversity, the data were limited in their ability to distinguish whether environmental factors or population history (genetic drift/gene flow/mutation) are the main determinants of genetic diversity, mainly because the species is limited to very few small and isolated upstream reaches, without much environmental variation across sites in each (Appendix S4).

| Summary of management recommendations
We found that most Macquarie perch populations are associated with low levels of genetic diversity and effective population size. Given that Macquarie perch continue to experience ongoing declines in spite of habitat restoration and threat amelioration efforts, these populations are at risk of extinction, at least in part due to negative genetic effects (e.g. inbreeding depression, low adaptive potential). To elevate genetic diversity, N e and population sizes, we recommend reconnecting populations within the following regions with assisted gene flow: Genetic rescue/restoration is a powerful management tool for a number of reasons: (i) it can have a large and long-lasting impact on genetic diversity and adaptive potential for relatively little additional work and expense (Frankham, 2015), (ii) it does not necessarily require ongoing intervention, although if populations remain small, assisted gene flow will need to be performed periodically (Hedrick, Peterson, Vucetich, Adams, & Vucetich, 2014), (iii) it could work at a range of scales from single species rescue (Hedrick & Fredrickson, 2010) to landscape reconnections (Smith, van der Ree, & Rosell, 2015). Population viability analysis can assist in developing objective and measurable criteria of metapopulation connectivity (Carroll, Fredrickson, & Lacy, 2014;Lacy, 2000), provided sufficient knowledge of species biology exists. Empirical results from other systems could be used to help provide priors for unknown parameters (e.g. inbreeding depression) in population viability models used to predict population persistence under management inaction and various scenarios of genetic intervention. Translocations of juveniles, in addition to adults, over several consecutive years have been advocated when there are limited adult fish available in a source populations (Todd & Lintermans, 2015). For these cases, we argue for translocating genetically diverse sets of individuals, achieved by collecting a small number of juveniles from each of many sampling sites. Numbers of translocated juveniles would need to account for high juvenile mortality (e.g. for Macquarie perch ~56 1-year-old juveniles would be required to produce 6 adults; Appendix S6; Todd and Lintermans (2015)).
Even if direct evidence of genetic problems (fitness loss) in specific cases is regarded as desirable, it should not be required to justify genetic intervention for small populations. Genetic augmentation alone often may not be sufficient to reverse population declines, but it must accompany other threat management actions for small populations (Frankham et al., in press;Love Stowell et al., 2017). Performed and monitored carefully within a risk-assessment framework (e.g. after considering the risk of outbreeding depression), in most instances genetic augmentation will be beneficial, and not harmful, for long-term population viability. Primmer and three anonymous reviewers, whose comments helped us to improve the manuscript.

DATA ARCHIVING STATEMENT
Data for this study are available as Appendices and Supporting Information (collection details, lengths, microsatellite genotypes, mitochondrial control region sequences and HL values for 872 Macquarie perch samples from 20 populations and 11 raw and standardized environmental variables for 16 populations). Sequence data are also deposited to GenBank (accession KT626048-KT626381).