Evolutionary restoration potential evaluated through the use of a trait‐linked genetic marker

Abstract Human‐driven evolution can impact the ecological role and conservation value of impacted populations. Most evolutionary restoration approaches focus on manipulating gene flow, but an alternative approach is to manipulate the selection regime to restore historical or desired trait values. Here we examined the potential utility of this approach to restore anadromous migratory behavior in coastal California steelhead trout (Oncorhynchus mykiss) populations. We evaluated the effects of natural and anthropogenic environmental variables on the observed frequency of alleles at a genomic marker tightly associated with migratory behavior across 39 steelhead populations from across California, USA. We then modeled the potential for evolutionary restoration at sites that have been impacted by anthropogenic barriers. We found that complete barriers such as dams are associated with major reductions in the frequency of anadromy‐associated alleles. The removal of dams is therefore expected to restore anadromy significantly. Interestingly, accumulations of large numbers of partial barriers (passable under at least some flow conditions) were also associated with significant reductions in migratory allele frequencies. Restoration involving the removal of partial barriers could be evaluated alongside dam removal and fishway construction as a cost‐effective tool to restore anadromous fish migrations. Results encourage broader consideration of in situ evolution during the development of habitat restoration projects.


| INTRODUCTION
There is an increasing recognition that human-driven evolution can shape the ecological role and conservation value of impacted populations (Hendry, Farrugia, & Kinnison, 2008;Hendry et al., 2011;Palkovacs, Kinnison, Correa, Dalton, & Hendry, 2012;Palumbi, 2001;Stockwell, Hendry, & Kinnison, 2003). This recognition has led to a growing interest in applying evolutionary principles to inform ecological restoration actions Hendry et al., 2011;Smith, Kinnison, Strauss, Fuller, & Carroll, 2014). In some cases, human activity shifts traits such that important ecological functions are altered or lost (Audzijonyte, Kuparinen, & Fulton, 2014;Audzijonyte, Kuparinen, Gorton, & Fulton, 2013;Palkovacs, Wasserman, & Kinnison, 2011). In such situations, evolutionary strategies can be applied to achieve ecological restoration. Calls to apply evolutionary restoration techniques have largely focused on managing gene flow to increase fitness in threatened populations (Aitken & Whitlock, 2013;Frankham, 2015;Leger, 2013;Whiteley, Fitzpatrick, Funk, & Tallmon, 2015). A somewhat different approach that has received less attention is to estimate the effects of anthropogenic impacts on key traits and then to manipulate selection regimes in ways that restore trait values. In this study, we develop an approach to restoration planning that considers predicted evolutionary responses to potential habitat restoration actions.
Here we apply an evolutionary restoration framework to inform the recovery of coastal California steelhead trout (Oncorhynchus mykiss; Walbaum). Steelhead trout display variability in migratory behavior.
Both within and among populations, some individuals are anadromous, spawning in freshwater and migrating to the ocean, whereas others are residents, completing their entire life cycle in freshwater (Kendall et al., 2015;Sogard et al., 2012). Populations can rapidly evolve freshwater residency when dams or other barriers impede migratory corridors (Pearse, Miller, Abadía-Cardoso, & Garza, 2014). Across a broad suite of species, the loss of anadromy has important implications for ecosystems. Anadromous fishes play a critical role in coastal watersheds by connecting ecosystems, driving nutrient dynamics, impacting food web interactions, shaping local species diversity (Flecker et al., 2010;Hocking & Reynolds, 2011;Naiman, Bilby, Schindler, & Helfield, 2002;Schindler et al., 2003;Willson & Halupka, 1995). This ecological role is fundamentally altered when human disturbance, often in the form of dam construction, causes populations to evolve freshwater residency (Palkovacs & Post, 2009;Post et al., 2008).
Currently, some anadromous steelhead populations in California are listed as either threatened (north of Point Conception, California, USA, to the Klamath River basin) or endangered (south of Point Conception) under the US Endangered Species Act. In contrast, freshwater resident populations, commonly referred to as rainbow trout, are not protected, even though many populations are native and have lost anadromy due to human habitat alteration (Clemento, Anderson, Boughton, Girman, & Garza, 2009). In an ironic twist of fate, freshwater resident rainbow trout has become the most widely distributed freshwater fish in the world due to human introductions, and these invasive rainbow trout originate largely from California hatchery stocks (Crawford & Muir, 2008;Halverson, 2008;Stanković, Crivelli, & Snoj, 2015).
Across a wide variety of fish species, anadromy and freshwater residency evolve rapidly, although individual decisions to migrate or remain resident depend on interactions among genetic, individual condition, and environmental factors (Dodson, Aubin-Horth, Thériault, & Páez, 2013;Hendry, Bohlin, Jonsson, & Berg, 2004). Anadromy may benefit some individuals by allowing them to escape stressful conditions in freshwater (i.e., reduced food supply, harmful flows, etc.) and providing opportunities for increased growth in the ocean and ultimately higher fecundity (Hendry et al., 2004). In California, females comprise a larger proportion of anadromous individuals in some populations, presumably because of the fitness benefits of greater fecundity for females (Ohms et al., 2014;Rundio, Williams, Pearse, & Lindley, 2012;Satterthwaite et al., 2010). But anadromy is costly during the migratory period and may subject individuals to increased energy expenditures and elevated risks of mortality through physiological stress and predation. Theory therefore predicts that anadromy should become less favored as freshwater growth rate increases (or marine productivity decreases), and if the risk of migrating to the ocean increases mortality (Hendry et al., 2004).
Because of widespread variation in migratory behavior within and among populations of steelhead trout, the determinants of anadromy and residency in this species have received much attention (Berejikian, Bush, & Campbell, 2014;Hale, Thrower, Berntson, Miller, & Nichols, 2013;Hayes et al., 2012;Kendall et al., 2015;Pearse et al., 2014;Phillis et al., 2016;Satterthwaite et al., 2009Satterthwaite et al., , 2010. Quantifying the proportion of anadromous steelhead vs. resident rainbow trout in a population typically requires directly observing the behavior of a large number of individuals. However, the distribution of adaptive genomic variation associated with specific traits has the potential to provide inference about the selective environments and adaptive difference among populations. In coastal California watersheds, a region of O. mykiss chromosome 5 (Omy5) has been recently identified, the Omy5 migration-associated region (MAR), with alternate alleles being tightly associated with the population prevalence of either migration or freshwater residency (Leitwein, Garza, & Pearse, 2017;Pearse et al., 2014). Many loci in the MAR are in strong linkage disequilibrium, suggesting the presence of a chromosomal inversion with loci associated with anadromous migratory traits (Leitwein et al., 2017;Pearse et al., 2014). Some of these traits include smoltification, growth rate, survival in sea water, and observed out-migration of juveniles (Doctor, Berejikian, Hard, & Vandoornik, 2014;Hecht, Hard, Thrower, & Nichols, 2015;Pearse et al., 2014;Phillis et al., 2016). In one example, a population recently translocated from below to above a waterfall has undergone a 49% reduction in the frequency of anadromy-associated alleles, a 19% reduction in smoltification, a 37% decrease in survival when exposed to sea water, and a 25% reduction in observed juvenile out-migration (Pearse et al., 2014;Phillis et al., 2016). Thus, although a single genomic locus should not be considered representative of all the adaptive genomic variation associated with this complex phenotype, variation in MAR allele frequencies does provide substantial utility for evaluating evolutionary restoration as a conservation tool (Pearse, 2016). Here we evaluated the effects of natural and anthropogenic environmental variables on the observed frequency of MAR alleles across 39 steelhead trout populations and modeled the potential for evolutionary restoration of anadromy at sites that have been impacted by anthropogenic barriers.

| Modeling overview
The overall goal of our modeling exercise was to link environmental variables such as climate, geomorphology, and migratory barriers to the frequency of MAR alleles associated with anadromy in steelhead populations across California. We then used model predictions to inform conservation strategies aimed at restoring anadromous migratory behavior to populations that have lost anadromy due to human habitat modification.

| Sample collection and genotyping
Genetic samples were collected from coastal California steelhead populations as part of earlier studies to assess population genetic structure within and among distinct population segments (DPSs; Clemento et al., 2009;Garza et al., 2014;Pearse et al., 2014). We examined

| Environmental variables
A range of environmental variables shape the contemporary evolution of anadromous migratory behavior in fishes (Table 1; Hendry et al., 2004;Quinn, 2005;Quinn & Myers, 2005). Climatological variables associated with anadromy include rainfall, runoff, streamflow, and baseflow. Geomorphological variables include streambed geology, stream order, stream gradient, riparian vegetation, elevation, stream temperature, maximum air temperature, and migration distance.
Natural and anthropogenic barriers to instream migration fall into two broad categories. Partial barriers impede but do not entirely prevent riverine migration. These barriers are surmountable under most flow conditions; however, they do impart an energetic cost to migration (Jonsson, Castro-Santos, & Letcher, 2010). Complete barriers block upstream migration entirely, but opportunities for downstream migration are possible if a fish can survive passage over a waterfall or  (Table 1). Partial barriers are those in-stream barriers that are considered passable in an upstream direction by anadromous fishes under at least some flow conditions. Complete barriers are insurmountable in an upstream direction under all flow conditions. The effects of partial barriers were considered to be additive, as they can consecutively impart an energetic cost along the migration path (Jonsson et al., 2010). In contrast, the effects of complete barriers were considered to be binary (present/absent), as they function to block all upstream movement.

| Statistical framework
In order to determine which environmental variables contributed significantly to variation in the frequency of anadromy-associated alleles, we conducted backward stepwise regressions for model selection to establish Akaike's Information Criterion (AIC) for each combination of variables. The minimum AIC value was used to select a best-fit model.
The frequency of the anadromous allele ƒ(A), at each sampling location was used as the dependent variable. Relative effect contributions for each factor were estimated as the amount of change in the population's haplotype frequency when a given factor was included or excluded from the model. Model validation was conducted in two ways.
First, observed ƒ(A) was plotted against predicted ƒ(A) using a simple liner regression. Predicted ƒ(A) values were calculated using the bestfit model. We used an R 2 value and 95% prediction interval to evaluate model accuracy. The 95% prediction interval accounts for the uncertainty of predicting a single observation in the model when compared to the 95% confidence interval, which is used to evaluate the mean values of the dataset. Second, bootstrap values were generated by taking 10,000 iterations of subsamples of the independent variables and using a p-value of <.05. We then quantified the proportions of times the term was below the p-value significance threshold and reported it as frequency of when the term was included in the model.

| Evolutionary restoration
Using the relative effect contributions determined by the best-fit model, we calculated the expected evolutionary responses (predicted ƒ(A)) for each population in a scenario where all anthropogenic barriers were removed from the downstream watershed. We then considered the change in the frequency of the anadromous allele ΔA under current versus restored scenarios as our measure of potential for evolutionary restoration. Then we assessed the potential for evolutionary restoration for each DPS, as these are the primary regional management units for coastal California.
When accurate cost estimates are available, our evolutionary restoration framework can be used to inform management of which watersheds to restore and which specific barriers to remove. This approach allowed us to compare the theoretical effectiveness of various barrier removal scenarios and to determine what types of barriers and which watersheds can yield the greatest evolutionary restoration at the lowest dollar cost. We obtained cost estimates for specific barrier types and watersheds within our study range from the Pacific States Marine Fisheries Commission (PSMFC). The PSMFC has been compiling cost estimates with the goal of incorporating them into their Passage Assessment Database, which contains all the potential barriers to anadromy along the Pacific coast. We were able to use exact cost estimates or approximate removal costs based on barrier type for Lion Canyon Creek, South Fork Bear Creek, Santa Paula Creek, Los Trancos Creek, Boulder Creek, and the Nacimiento River (PSMFC, unpublished data).

| RESULTS
The AIC best-fit model contained five terms explaining significant variation in ƒ(A; Frequency of anadromy ƒ(A) generally decreased north to south, while restoration potential ΔA generally increased from north to south ( Figure 3, Table 3). However, considerable variation among sampling locations was found in every DPS (Table 4) DPS is the least urbanized section of coastal California, although impacts from forestry and illegal marijuana cultivation (Bauer et al., 2015) are widespread (Fig. S4) Table 4 T A B L E 3 Summary of anadromous allele frequency ƒ(A) by Distinct Population Segment (DPS). The Southern California DPS contains the greatest amount of anthropogenic disturbance across all our sampling locations in the form of partial and complete anthropogenic barriers. This is reflected in the average restoration potential ΔA, which is much higher than the other DPSs

Number of barriers removed
Complete anthro bar removed?
Lion  a 18.47% increase in anadromy, which represents a substantially more costly $4,060,638 per one percent increase in anadromy. However, there are some highly urbanized watersheds that have complete anthropogenic barriers that are not large dams but improperly designed culverts or grade structures. These types of barriers prevent upstream movement just as large dams but are considerably less expensive to remove. For example, our model estimated that a 20.29% increase in anadromy-associated alleles would result from removing a partial barrier and a diversion dam in Lion Canyon Creek (SC, Fig. S1), for only $320,000. This restoration project is estimated to cost $17,551 per one percent increase in anadromy, the highest return on investment for any of the watersheds considered.

| DISCUSSION
Ecologists and evolutionary biologists have become increasingly aware that human-driven evolution can shape key traits of ecologically important species (Allendorf & Hard, 2009;Hendry et al., 2008;Palkovacs et al., 2012). A desire to return traits and their associated ecological functions to historical conditions has led to an increasing interest in evolutionary restoration Hendry et al., 2011;Smith et al., 2014). In this study, we examined the impact of We examined the impacts of climate, geomorphology, and migratory barriers on the frequency of anadromy-associated alleles.
Migratory barriers were found to have the greatest association with anadromous allele frequencies. Natural barriers (e.g., waterfalls, cascades) represent long-term migratory barriers and had the largest effect, the anadromous allele frequency being on average 31% lower when present. Complete anthropogenic barriers (mostly dams) also had a relatively large effect, with the anadromous allele frequency being an average of 18% lower when complete barriers were present. Most California dams have been operating for <100 years (Hanak et al., 2011); the large effect of complete anthropogenic barriers supports the idea that freshwater residency evolves rapidly following dam construction (Pearse et al., 2014).
While we do not have temporal information from most sites to estimate the rate of allele frequency change following barrier introduction, we can draw some inferences from below-above barrier population comparisons. There are three cases where we have estimates of neutral genetic divergence and variation at the MAR for above-and below-barrier populations (one from Scott Creek and two from the Santa Ynez River). In these cases, changes in allele frequencies at the MAR (49%-76%) are large relative to the extent of genetic divergence at neutral SNP loci (pairwise F ST values all <0.01; Clemento et al., 2009;Pearse et al., 2009). Particularly informative is a documented translocation that occurred within the Scott Creek watershed. Here, the frequency of anadromy-associated MAR alleles is 83% below a barrier waterfall and is reduced to 34% in a population translocated above the waterfall about 100 years ago (F ST > 0.3, Martinez, Garza, & Pearse, 2011). These same populations display a pairwise F ST of 0.018 at neutral SNP loci (Pearse et al., 2009), clearly demonstrating that drift is not solely responsible for the large-magnitude directional changes in allele frequencies detected at the MAR. The translocated Scott Creek population above the waterfall currently shows an anadromous allele frequency similar to populations above dams, which were probably isolated for a similar amount of time. Assuming that the below-barrier population on Scott Creek has not changed dramatically in its allele frequency over the past 100 years, we can infer that the reduction of anadromy-associated alleles occurred at a rate of approximately 0.05% per year. We anticipate that this rate of change was likely much greater in the years immediately following the translocation and has slowed markedly since then (Kinnison & Hendry, 2001).
Waterfalls, dams, and other impassable barriers are not the only types of migratory barriers found to impact the frequency of anadromy. Partial barriers impart an energetic cost to migration (Hendry et al., 2004;Kendall et al., 2015;Kinnison, Unwin, & Quinn, 2003).
When anthropogenic partial barriers were present, anadromous al-  S3) has 11 partial anthropogenic barriers, three partial natural barriers, no complete barriers, and an anadromous allele frequency of just 54% (compared to an expected allele frequency of 74% based on its migration distance and number of natural barriers). Thus, the accumulated effects of many partial barriers can have an impact equivalent to that of an impassable dam. Importantly, removal of small partial barriers is less expensive and presents fewer engineering, social, and regulatory challenges compared to large dam removal (Doyle et al., 2005;Graff, 1999).
We found a significant effect of migration distance on the frequency of anadromy-associated alleles. Migration distance has previously been found to affect anadromy in a wide variety of species, with spawning sites further from the ocean generally displaying lower rates of anadromy (Hendry et al., 2004;Kendall et al., 2015;Ohms et al., 2014). The longer the migration, the more energy must be expended to reach the spawning grounds and the higher the chance of encountering barriers, predators, and other mortality sources. Thus, our results are consistent with prior studies showing that longer migrations select for increased rates of freshwater residency.
In California, steelhead trout are managed in Distinct Population

Segments (DPSs) under the US Endangered Species Act. The Southern
California DPS had the lowest average anadromous allele frequency measured (Table 3, Fig. S1), most likely due to the high level of human disturbance in Southern California watersheds. A plethora of instream impediments have likely contributed to the overall reduction in the average frequency of anadromy-associated alleles within its sampled watersheds to just 41%, compared to an expected allele frequency of 62% based on the average migration distance and number of natural barriers. In contrast, the Northern California DPS has the lowest human population, the fewest anthropogenic barriers, and an average frequency of anadromy-associated alleles of 72% (compared to an expected allele frequency of 73%; Fig. S4). The Central California Coast DPS contains streams that range between highly altered (11-12 partial anthropogenic barriers) to relatively undisturbed (0-2 partial anthropogenic barriers; Fig. S3). While there are no complete anthropogenic barriers in our study populations for the Central California Coast DPS, the accumulation of partial barriers is associated with the reduction in anadromy-associated alleles in parts of this DPS. The South-Central California Coast contains the two populations that are outliers in the model (Fig. S3). Both of these populations are in the highly altered Salinas River watershed, which may function differently than other coastal streams due to major anthropogenic disturbances, particularly in the form of intensive agriculture. For example, Tassajera Creek (SCCC, Fig. S2) shows a lower than expected frequency of anadromy-associated alleles, perhaps due to the agricultural withdrawals that may seasonally dry the Salinas River along much of its main stem channel, creating a low-flow barrier to migration. In contrast, the Nacimiento River (SCCC, Fig. S2) shows a higher than expected frequency of alleles associated with anadromy. This river flows into the Lake Nacimiento, which may represent the destination for an adfluvial migration, where fish migrate to the lake instead of the ocean (Pearse et al., 2014). Similar adfluvial patterns were found in above reservoir populations around the San Francisco Bay Area (CCC, Fig.   S3), where there is a strong association between reservoir size and the frequency of anadromy-associated alleles (Leitwein et al., 2017).
Each of our study populations were sampled at a single time point, yet we anticipate that allele frequencies at any given site may fluctuate somewhat through time due to drift and dynamic local selective drivers such as stream flow conditions. We do not have the data from repeated sampling events to address within-site changes in allele frequencies for this study. Nonetheless, our results show that major variation in allele frequencies are predictably related to migration distance and the presence of natural and anthropogenic barriers. These strong and consistent signals would not be expected to emerge in a scenario with high temporal variability in allele frequencies due to random or site-specific factors. Thus, our overall results are likely robust to fine scale temporal shifts in allele frequencies within sites.
Our study shows that partial and complete anthropogenic barriers are strongly associated with variation in the frequencies of anadromyassociated alleles. We therefore calculated the expected evolutionary responses for each population in a scenario where all anthropogenic barriers were removed from the downstream watershed. While simulating the removal of large impassable dams yields the biggest predicted evolutionary responses, there are many social, engineering, and legal challenges for projects of this scale (Doyle et al., 2005;Graff, 1999). Smaller dams yield smaller returns, however there are many more of them, which can add up to similar effect contributions to that of a large complete barrier. Smaller scale projects can also be conducted with relative ease by local agencies or watershed stewardship groups. This strategy should be considered as an important complement to large-scale dam removal when considering the evolutionary restoration of anadromy.
In the subset of watersheds where we evaluated restoration costs, the economic potential of different barrier removal scenarios varied greatly ( Table 5). The removal of many smaller partial barriers was substantially cheaper than removing a large impassable dam, yet still achieved a similar evolutionary response. Large dam removals can cost tens of millions of dollars and take decades of planning to complete. An alternative approach to barrier removal is barrier remediation, which can be conducted on partial and complete barriers. Not all barriers were originally constructed in ways that would allow them to be modified. Nonetheless, some partial barriers such as culverts can be modified to reduce flow velocity and increase water depth, allowing unimpeded passage for anadromous fish. Some dams can have fishways installed, converting them from complete barriers into partial barriers, reducing their impacts substantially. However, fishway construction can be difficult and expensive on some larger dams and many fishways perform poorly, making dam removal the preferred restoration strategy whenever possible (Brown et al., 2013).
Ecologists have called for the use of dam removals as large-scale experiments to examine ecological processes in rivers and streams (Hart et al., 2002). Our study extends this framework to include evolution. Here we provide predicted evolutionary responses to various restoration scenarios. The next step is to monitor evolutionary change following large-and small-scale barrier removals as management experiments to test these predictions. Evolutionary experiments at this scale are rarely undertaken. Thus, barrier removal provides an important opportunity to achieve restoration objectives while testing basic hypotheses about the factors driving natural selection and evolution in wild populations.

| CONCLUSIONS
Human-induced trait change has been observed in species and ecosystems around the world, and recent efforts have been made to identify and manages these changes (Allendorf & Hard, 2009;Palkovacs et al., 2012). Most evolutionary restoration approaches have focused on manipulating gene flow Hendry et al., 2011;Smith et al., 2014). However, manipulating the environment in ways that shift selection is another method that can effectively restore historical trait values and associated ecological functions (Ashley et al., 2003;Smith et al., 2014). Our study shows that habitat modification in the form of migratory barriers such as dams and culverts are associated with the loss of anadromy-associated alleles in coastal California steelhead trout populations. While complete barriers such as dams are associated with a dramatic loss of anadromy, the accumulation of large numbers of smaller partial barriers can add up to similarly large impacts. Removing large dams is expected to result in the greatest evolutionary restoration of anadromy, however such projects can be expensive and present many social, engineering, and legal challenges (Doyle et al., 2005;Graff, 1999). Our results suggest that removal of partial barriers can be effective at restoring anadromy at a fraction of the cost. Projects involving small barrier removal present fewer technical and socio-political challenges. Thus, restoration projects involving the removal of small partial barriers could be considered alongside large dam removals and fishway construction projects as effective tools to restore anadromy to populations that have evolved increased freshwater residency. Ecosystems and Climate.

DATA ARCHIVING STATEMENT
All data for this study are permanently archived at the Southwest Fisheries Science Center Fisheries Ecology Division (https://swfsc. noaa.gov/FED/).