Recent and rapid anthropogenic habitat fragmentation increases extinction risk for freshwater biodiversity

Abstract Anthropogenic habitat fragmentation is often implicated as driving the current global extinction crisis, particularly in freshwater ecosystems. The genetic signal of recent population isolation can be confounded by the complex spatial arrangement of dendritic river systems. Consequently, many populations may presently be managed separately based on an incorrect assumption that they have evolved in isolation. Integrating landscape genomics data with models of connectivity that account for landscape structure, we show that the cumulative effects of multiple in‐stream barriers have contributed to the recent decline of a freshwater fish from the Murray–Darling Basin, Australia. In addition, individual‐based eco‐evolutionary simulations further demonstrate that contemporary inferences about population isolation are consistent with the 160‐year time frame since construction of in‐stream barriers began in the region. Our findings suggest that the impact of very recent fragmentation may be often underestimated for freshwater biodiversity. We argue that proactive conservation measures to reconnect many riverine populations are urgently needed.


| INTRODUC TI ON
We are now confronted by the sixth global mass extinction with the current rate of species losses far exceeding pre-anthropogenic background estimates (Barnosky et al., 2011). This crisis is particularly severe in freshwater ecosystems, which have shown declines of biodiversity greater than for either terrestrial or marine ecosystems (Darwall et al., 2018). Habitat loss and fragmentation are key factors leading to the genetic and demographic decline of populations that together threaten species persistence (Fischer & Lindenmayer, 2007). Over the last century, close to one million large dams and many millions of smaller in-stream barriers have been constructed globally (Jackson et al., 2001;Liermann et al., 2012).
These barriers have had devastating ecological consequences by preventing or restricting connectivity among populations, leading to higher rates of genetic drift and inbreeding. This, in turn, can lead to lower fitness due to inbreeding depression and reduced evolutionary potential due to loss of genetic diversity (Frankham, 2005;Keyghobadi, 2007). Additionally, small populations become more vulnerable to extirpation due to stochastic demographic events (Lande, 1993) and, when this occurs on a regional scale, species extinctions are the inevitable result (Hanski, 1998).
Landscape genetics provides a way to identify how human activities threaten the persistence of wild populations (Manel & Holderegger, 2013). The time lag between environmental change and any detectable genetic signal resulting from this change can, however, make it very difficult to disentangle the effects of historical from contemporary processes (Landguth et al., 2010). This is particularly the case for naturally structured populations such as those found in dendritic river networks (Coleman et al., 2018). The progression from landscape genetics to landscape genomics has increased both the spatial and temporal resolutions at which evolutionary processes can be examined, offering a more powerful framework with which to quantify the effects of very recent disturbance on populations (Allendorf et al., 2010;Grummer et al., 2019). Previous landscape genetics studies investigating the impact of in-stream barriers have often focused on larger, migratory species or assessed only one, or a few large barriers (Faulks et al., 2011;Gouskov et al., 2016;Meeuwig et al., 2010;Mims et al., 2019;Torterotot et al., 2014). For example, Muhlfeld et al. (2012) used microsatellite loci and simulations to understand the impact of placement of a single barrier on introgressive hybridization between native westslope cutthroat trout (Oncorhynchus clarkii lewisi) and non-native rainbow trout in Glacier National Park, USA. On the other hand, small-bodied but ecologically important species often receive relatively little attention from conservation managers (Olden et al., 2007;Saddlier et al., 2013).
Regional-scale efforts to improve fish passage in Australia have been successful in restoring passage along the main river channel for large-bodied species (Barrett & Mallen-Cooper, 2006;Baumgartner et al., 2014); however, these measures have proved ineffective for most small fishes (Harris et al., 2017). The cumulative impact of numerous smaller in-stream barriers (e.g., weirs, farm dams and road crossings) is likely to greatly impact small-bodied and nonmigratory fishes; however, this has been the subject of much less research at a regional scale (Coleman et al., 2018;Diebel et al., 2015; but see Nathan et al., 2019).
In this landscape genomics study, we examine the effects of recent habitat fragmentation on the southern pygmy perch (Nannoperca australis), a threatened small-bodied fish (<80 mm) that recently experienced major demographic declines and local extinctions across the Murray-Darling Basin (MDB), Australia Cole et al., 2016;Hammer et al., 2013). This ecological specialist is restricted to small streams and wetlands, is typical of many native small-bodied fishes in the region and offers a conservative model for guiding broader conservation strategies as the impacts of fragmentation are likely to be more pronounced for larger, migratory species. Since European colonization, freshwater habitat in the MDB has rapidly deteriorated due to severe water overharvesting, land clearing, habitat loss and fragmentation (Davies et al., 2010;Kingsford, 2000), and the MDB is now considered one of Australia's most vulnerable and threatened ecosystems (Laurance et al., 2011).
The MDB has very few natural in-stream barriers, but it has been heavily modified with more than 10,000 dams, weirs, road crossings, levees and barrages constructed since the late 1850s (Baumgartner et al., 2014). As such, the MDB provides a unique opportunity to examine the consequences of recent habitat fragmentation without the confounding influence of prolonged human disturbance over hundreds of years as is common to many northern hemisphere river basins (e.g., Hansen et al., 2014). Environmental factors, including human disturbance, are known to influence genetic diversity for N. australis Cole et al., 2016); however, little is known about the specific role that widespread habitat fragmentation has played in the species recent and rapid decline. We hypothesize that, after accounting for historical patterns of genetic structure, genetic differentiation among demes should increase with the number of in-stream barriers separating them. We also predict that populations most isolated by fragmentation would exhibit reduced effective population size (N e ) and lower levels of genetic diversity. Additionally, we used forward genetic simulations to investigate whether high contemporary levels of genetic differentiation could have arisen in the relatively short time since the construction of in-stream barriers began in the MDB. Our results demonstrate that recent anthropogenic habitat fragmentation has contributed to the loss of genetic diversity and population isolation observed. They also suggest that proactive conservation measures to restore connectivity (e.g., environmental flows, habitat restoration) and increase evolutionary potential (e.g., genetic rescue) are urgently required for this, and potentially many other poorly dispersing aquatic species. To minimize the number of cohorts sampled per population, we targeted adult fish of similar size from each sampling site. To avoid the inclusion of highly related individuals in the data, we estimated pairwise relatedness among individuals from each site using the dyadic likelihood relatedness estimator described in Milligan (2003) and implemented in the R package related (Pew et al., 2015). The data were then filtered to retain only variants present in at least 70% of individuals and in 70% of populations, retaining only one biallelic SNP per locus with a minimum minor allele frequency of 0.05.

| Sampling, ddRAD genotyping and SNP filtering
Population structure and other demographic parameters such as effective population size should be assessed using neutral loci (Allendorf et al., 2010;Luikart et al., 2003). To define a putatively neutral data set, F ST outlier loci were detected using a Bayesian approach with BayeScan v.2.1 (Foll & Gaggiotti, 2008) and the coalescent-based FDIST method (Beaumont & Nichols, 1996) in Arlequin v.3.5 (Excoffier & Lischer, 2010). BayeScan was run for 100,000 iterations using prior odds of 10,000. Loci different from zero with a q-value < 0.1 were considered outliers. Arlequin was run specifying the hierarchical island model with 50,000 simulations of 100 demes for each of 13 populations (based on the 13 separate catchments sampled). Loci outside the neutral distribution at a false discovery rate (FDR) of 10% were considered outliers. Loci detected as outliers by either BayeScan or Arlequin were filtered. The remaining SNPs were examined for departure from expectations of Hardy-Weinberg equilibrium (HWE) using GenoDive 2.0b27 (Meirmans & Van Tienderen, 2004). Finally, loci out of HWE at a FDR of 10% in more than 50% of populations were removed. Detailed information concerning library preparation and bioinformatics are described in Appendix S1.

| Population structure
Pairwise F ST (Weir & Cockerham, 1984) was estimated among sampling sites using GenoDive (Meirmans & Van Tienderen, 2004) with significance assessed using 10,000 permutations. Bayesian clustering analysis of individual genotypes was then performed using fast-Structure (Raj et al., 2014). Ten independent runs for each value of K (1-25) were completed to ensure consistency, and the most likely K was assessed by comparing the model complexity that maximized marginal likelihood across replicate runs. F I G U R E 1 Nannoperca australis sampling locations in the Murray-Darling Basin (MDB). Stream sections are colour coded according to F ST estimated using the StreamTree model (Kalinowski et al., 2008). Cross markers represent the location of artificial in-stream barriers. Admixture plot is based on 3,443 SNPs depicting K = 12 clusters determined by maximum marginal likelihood using fastStructure (Raj et al., 2014)

| Anthropogenic isolation of populations
If anthropogenic habitat fragmentation has affected population connectivity and dispersal, we should expect genetic differentiation to increase in response to the number of in-stream barriers separating populations. To determine whether local characteristics of the stream network (i.e., in-stream barriers and other local-scale landscape heterogeneity) better explain population differentiation than isolation by distance (IBD), we used the StreamTree model of Kalinowski et al. (2008). Genetic distances among populations were modelled as the sum of all pairwise genetic distances that mapped to each section of the stream network. This provides a distance measure that is independent of the length of each stream section and identifies the reaches that contribute most to restricting gene flow (e.g., due to dendritic structure, in-stream barriers or other local landscape effects). Model fit was assessed by plotting the StreamTree fitted distance against observed F ST and calculating the regression coefficient of determination (R 2 ). This model was then compared with a model of IBD calculated using multiple matrix regression with randomization (MMRR) following the method of Wang (2013). Pairwise population distances along the river network were calculated with ArcMap v.10.2 (ESRI, 2012). Model significance for the MMRR was assessed using 10,000 random permutations.
In dendritic river systems, hierarchical network structure and spatial hydroclimatic variation can also drive patterns of genetic diversity of stream-dwelling organisms (Fourcade et al., 2013;Hughes et al., 2009;Morrissey & de Kerckhove, 2009;Thomaz et al., 2016). To evaluate the relative contributions of anthropogenic habitat fragmentation, natural  Australia, 2011;Stein et al., 2014). These were assigned to one of five categories describing variation in temperature, precipitation, flow regime, human disturbance and topography. Variance inflation factor (VIF) analysis was then used to exclude highly correlated variables using a VIF threshold of 10 (Dyer et al., 2010). The remaining variables were reduced to principal components (PCs) using the dudi.pca function in the ADE4 R package (Dray et al., 2016), and Euclidean distance matrices were constructed based on the PCs with eigenvalues > 1 (Yeomans & Golder, 1982) retained for each category. All distance matrices were z-transformed to facilitate direct comparison of partial regression coefficients (Schielzeth, 2010). Each variable was initially tested in an independent univariate MMRR before significant factors were combined in a multivariate MMRR model with 10,000 random permutations used to assess significance.

| Habitat fragmentation, genetic diversity and population size
To test the hypothesis that the most isolated populations exhibit reduced genetic diversity, we examined the relationship between population-specific F ST and expected heterozygosity (H E ). Populationspecific F ST was estimated for each sampling site using the method of Weir and Hill (2002), and H E was calculated using Genodive.
Effective population size was estimated using the linkage disequilibrium (LD) estimator implemented in NeEstimator 2.01 (Do et al., 2014). This method is based on the assumption that LD at independently segregating loci in a finite population is a function of genetic drift and performs particularly well with a large number of loci and where population sizes are expected to be small (Waples & Do, 2010). In the absence of significant F ST , Lower Murray sites MID and MUN were considered one population and these samples were combined for the N e estimates. NeEstimator was run assuming random mating and using a P crit value of 0.075 following guidelines for small sample sizes suggested by Waples and Do (2010).

| Eco-evolutionary simulations
Simulation studies are becoming an increasingly important part of landscape genomics as a wide range of parameters can be explored for key evolutionary processes such as gene flow, genetic drift, mutation and selection (Hoban et al., 2012). In this case, we used   (Table S1).

| Population structure
High

| Anthropogenic isolation of populations
The StreamTree model was used to identify parts of the stream network that contribute more to F ST (e.g., restricted dispersal due to barriers or other local environmental conditions). Results indicated that local characteristics of the stream network better explain F ST than the null hypothesis of IBD (i.e., the resistance to dispersal for any given stream section is determined by its length). Figure 1  Although there was significant IBD within-catchment groups (i.e. the first cluster in Figure 2b, R 2 = 0.730, β = 0.0016 [0.001-0.002 95% CI], p = 6.54 × 10 −8 ), IBD was not significant in models across the whole basin, in contrast to models of stream hierarchy and barriers (see below). In addition, even when comparisons were limited to sites within catchments, the number of barriers still provided a better model than IBD (R 2 = 0.81 versus 0.73, respectively; Figure S4).  Figure 3 and Table 2). populations also harbouring the least genetic variation ( Figure S5;

| Eco-evolutionary simulations
The simulations demonstrated that contemporary population dif- generations with just one barrier (Figure 4; Table S4).

| D ISCUSS I ON
Habitat fragmentation is a key process implicated in the current and unprecedented worldwide loss of freshwater biodiversity (Fischer & Lindenmayer, 2007). Determining the contribution of recent human activities to the decline of riverine species is, however, challenging, as the genetic signal of recent disturbance can be confounded by historical patterns of dispersal shaped by hydrological network structure (Brauer et al., 2018;Coleman et al., 2018;Landguth et al., 2010). Integrating landscape genomics data with models of connectivity that account for landscape structure, we show that the cumulative effects of multiple in-stream barriers have contributed to the recent decline of a freshwater fish from the Murray-Darling Basin, Australia. Populations most isolated by recent habitat fragmentation exhibited reduced genetic diversity and increased population differentiation, and this signal remained strong after accounting for the historical effects of dendritic stream hierarchy.
Interestingly, we found no evidence for isolation by environment (IBE), despite a previous genotype-environment association (GEA) study for the same species finding several hydroclimatic variables influenced putatively adaptive genetic variation at both regional and local scales . This is likely due to the ability of GEA methods to identify signal from relatively few regions of the genome responding to selection (Forester et al., 2018). In contrast,  , the approach we used is known to perform well for small populations (Do et al., 2014;Waples & Do, 2010), and our results are consistent with expectations based on remnant habitat patch sizes, and estimates obtained in an earlier microsatellite study (Cole et al., 2016). Other previous work based on coalescent analyses of microsatellite DNA data sets has demonstrated that historical population sizes of N. australis were much larger before European colonization (Attard et al., 2016), and that populations across the MDB were also more connected until that time (Cole et al., 2016). Together, our findings support these studies and the hypothesis that the low genetic diversity, small N e and high F ST observed for contemporary populations likely reflects the combined impact of both historical and recent processes, rather than being due solely to natural demographic variability over longer evolutionary time scales. In addition, several populations sampled for this study have subsequently suffered local extirpation during prolonged drought, and the small size of most remnant populations indicate they are at high risk of extinction.
Since the 1800s, land use and hydrology in the MDB has been increasingly modified due to urbanization and irrigation (Leblanc et al., 2012). These changes have included the construction of thousands of barriers to fish passage across the basin (Baumgartner et al., 2014), and it is now considered one of Australia's most fragmented and degraded ecosystems (Davies et al., 2010;Kingsford, 2000). The focus of most barrier mitigation actions in the MDB to date has been on restoring passage across larger dams along the main river channel (Barrett & Mallen-Cooper, 2006). Although some fishways have been designed to facilitate movement of smaller fish, they TA B L E 2 Results of multiple matrix regression with randomization (MMRR) tests for the relationship between pairwise genetic distance (F ST ) and geographic distance, catchment membership, number of in-stream barriers and environmental distances  (Baumgartner et al., 2014). Furthermore, the spatial scale of dispersal for many small-bodied MDB fishes often restricts their movements to headwater streams and wetlands away from the main channel (Harris et al., 2017). Habitat loss and fragmentation associated with the thousands of smaller barriers in headwater streams have therefore likely contributed to the widespread decline of many smaller and more sedentary MDB fishes, including N. australis (Brauer et al., 2018;Cole et al., 2016;Hammer et al., 2013;Huey et al., 2017). It is perhaps surprising then, that there have been relatively few studies explicitly testing the genetic effects of anthropogenic fragmentation on small-bodied fishes in the MDB. One recent example in the neighbouring Yarra River catchment, however, combined a large empirical data set with spatially explicit simulations to examine the role of artificial barriers in driving local-scale patterns of genetic variation for river blackfish (Gadopsis marmoratus), a small and sedentary species also found in the MDB (Coleman et al., 2018). Based on eight microsatellite loci, genetic diversity was found to be lower for populations above barriers in small streams, with several isolated populations also exhibiting signs of inbreeding. In addition, their simulations demonstrated that power to detect recent impacts of barriers could be improved by increasing the number of loci used, highlighting the benefit of modern genomic data for conservation genetics.
An unprecedented severe and prolonged drought between 1997 and 2010 caused catastrophic loss of habitat and local extirpation for some N. australis populations, particularly in the lower Murray Wedderburn et al., 2012). In response, an emergency conservation breeding and restoration programme was implemented in the lower MDB (Attard et al., 2016;Hammer et al., 2013) (Davis et al., 2015).
Additionally, many species may be already depleted to the point where improved environmental conditions alone will not be sufficient to facilitate recovery. In this case, genetic rescue offers a potential solution for a broad range of threatened taxa (Ralls et al., 2018;Whiteley et al., 2015). However, despite strong evidence supporting the benefits of genetic rescue for fragmented populations, conservation managers are often reluctant to adopt these measures (Frankham, 2015). We suggest that the impacts of recent habitat fragmentation may have been underappreciated for many species, and that estimates of population structure solely attributed to historical evolutionary processes have potentially led to management frameworks that actually reinforce fragmentation and isolation at the expense of species-level genetic variation (sensu Coleman et al., 2013).

F I G U R E 4
Number of generations (log scale) for global F ST to reach 0.2 with increasing levels of habitat fragmentation for simulated N. australis metapopulations of N e = 1,000, N e = 500 and N e = 100.
Simulations were based on a stepping stone model assuming equal N e for each subpopulation and were allowed to run for 20,000 generations with a migration rate of 0.5 between adjacent demes before 300 generations with no migration. Red dashed line indicates the approximate number of generations since construction of in-stream barriers began in the MDB (160 generations) There is also increasing evidence that natural selection can influence the evolutionary trajectory of small and fragmented populations (Brauer et al., 2017;Fraser, 2017;Wood et al., 2016). Critically for conservation, this indicates that adaptive divergence of small populations can occur quickly following fragmentation (Brauer et al., 2017) and that even very recently isolated populations may harbour novel adaptive diversity. It is therefore important to build evolutionary resilience by facilitating genetic exchange among isolated populations to restore natural evolutionary processes and maintain species-level genetic variation, potentially valuable under a range of future selection regimes (Webster et al., 2017;Weeks et al., 2016).
There is a global biodiversity crisis unfolding in freshwater ecosystems with aquatic vertebrate populations declining by 80% over the last 50 years (Darwall et al., 2018). Restoring functional connectivity for aquatic communities across river basins via traditional mitigation approaches is simply not feasible within the time frame required to enable many currently threatened species to persist. There is also now strong empirical evidence that several long-established beliefs central to prevailing conservation practices are overly cautious, and that the current local-is-best approach increases the prospect of managing species to extinction (Frankham et al., 2017;Pavlova et al., 2017;Weeks et al., 2016). Given widespread fragmentation, habitat loss and the ongoing global decline of freshwater biodiversity, a rapid paradigm shift is needed to empower conservation practitioners to take action before demographic issues become critical. There are risks associated with any proactive management intervention such as translocation or genetic rescue. These risks, however, need to be weighed against the ever-increasing risk of doing nothing.

ACK N OWLED G EM ENTS
We thank the many people who helped with fieldwork, especially PhD scholarship to Chris Brauer.

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