By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
Ecological restoration assists the recovery of degraded ecosystems; however, restoration can have deleterious effects such as outbreeding depression when source material is not chosen carefully and has non-local adaptations.
We surveyed 23 eelgrass (Zostera marina L.) populations along the North American Atlantic coast to evaluate genetic structure and connectivity among restored and naturally recruited populations.
While populations along the North America Atlantic coast were genetically distinctive, significant migration was detected among populations. All estimates of connectivity (FST, migration rate base on rare alleles, and Bayesian modelling) showed a general north to south pattern of migration, corresponding to the typical long-shore currents in this region.
Individual naturally recruited meadows in the Virginia coastal bays appear to be the result of dispersal from different meadows north of the region. This supports the hypothesis that recruitment into this region is typically a slow, episodic process rather than a permanent, continuous connection between the populations.
While natural recovery of populations that were catastrophically lost in the 1930s has been slow, large-scale seed-based restoration has been very successful at quickly restoring landscape-scale areal coverage (over 1600 ha in just 10 years). Our results show that restoration was also successful at restoring meadows with high genetic diversity. Naturally recruited meadows were less diverse and exhibited signs of genetic drift.
Synthesis. Our analyses demonstrate that metapopulation dynamics are important to the natural recovery of seagrass ecosystems that have experienced catastrophic loss over large spatial scales; however, natural recovery processes are slow and inefficient at recovering genetic diversity and population structure when recruitment barriers exist, such as a limited seed source. Seed-based restoration provides a greater abundance of propagules, rapidly facilitates the recovery of populations with higher genetic diversity, and when seed sources are chosen carefully protects regional genetic structure. First-order estimates indicated that the genetic diversity achieved by active restoration in 10 years would have otherwise taken between 125 and 185 years to achieve through natural recruitment events.
Large-scale disturbances are becoming more frequent in near-shore communities due in large part to human activities (Jackson et al. 2001; Lotze et al. 2006; Halpern et al. 2008; Waycott et al. 2009). These disturbed systems may not recover naturally, or recovery may be very slow. Restoration is a tool that can be used to assist or speed up recovery (Society for Ecological Restoration International Science & Policy Working Group 2004). In this study, we compare the processes of natural recovery and restoration of seagrass communities in the Virginia coastal bays.
Natural recovery of disturbed systems often occurs via dispersal and metapopulation dynamics (Kendrick et al. 2012). Metapopulations are groups of populations that are linked by effective dispersal and thus geneflow (Levins 1970; Cain, Milligan & Strand 2000), which provides the propagule source for natural recovery. Dispersal is the movement of genes from a source population into a new settlement site (Pineda, Hare & Sponaugle 2007) and is only effective when the gametes or seeds released reach an appropriate habitat, settle, survive and reproduce (Kinlan & Gaines 2003). Barriers to effective dispersal include a lack of propagules, physical barriers, limited dispersal agents and unsuitable habitat or settlement space (Van der Pijl 1982). Biotic and abiotic dispersal agents are often limited in coastal systems when compared with terrestrial systems, and coastal systems have an abundance of barriers compared with the open ocean (Kendrick et al. 2012).
Restoration augments dispersal and population connectivity by taking propagules from one population and seeding a new area, either in an effort to speed up natural recovery or to circumvent barriers to dispersal. Movement of seeds by human activities is not subject to natural controls and therefore may have unintended consequences. When source materials for restoration are collected from outside natural dispersal ranges, novel genotypes may not be locally adapted, or may interbreed with locally adapted genotypes and result in less fit progeny (i.e. outbreeding depression) (Hufford & Mazer 2003). Recognition of this as a potential issue has led to guidelines for restoration, which call for taking propagules from nearby sources where populations have similar genetic structure and variability (Broadhurst et al. 2008). For example, seed transfer zones have been developed for many terrestrial species including Douglas fir trees (Pseudotsuga menziesii F.) (Campbell 1991), Fourwing saltbush (Atriplex canescens) (Sanderson & McArthur 2004) and native grass (Festuca roemeri) (Wilson & Rannala 2003). The goal of delineating seed transfer zones is to limit the negative impacts of the introduction of novel genotypes such as genetic swamping (Hufford & Mazer 2003). In simple terms, collection should happen from an area that is potentially naturally connected via metapopulation processes. Restoration, at its best, should simply speed up the natural process of recovery. However, nearby seed sources are often limited, degraded or impractical to harvest. Furthermore, the data needed to delineate seed zones are also not always available.
In this study, we compare natural recovery and restoration by examining a system that has been highly disturbed and is recovering both naturally and through human-mediated restoration. Historical records indicate that the Virginia coastal bays were once carpeted with seagrass (Zostera marina L.); however, a widespread disease (Labyrinthula zosterae) decimated seagrass populations all over the Atlantic Ocean in the 1930s (Rasmussen 1977). While many populations slowly recovered from the impact of this disease, including populations in the nearby Chesapeake Bay (Cottam & Munro 1954; Orth & Moore 1984), seagrass populations in the coastal bays of Virginia remained locally extinct for over 60 years (Orth et al. 2006). Small naturally recovered patches of seagrass were found in the mid 1990s, and long-term water quality monitoring (www1.vcrlter.virginia.edu/home1/?q=data_wq) and modelling of light availability (Lawson et al. 2007) suggested that this area could support seagrass. Recovery was probably limited by a lack of propagules (Orth et al. 2012). As a result, large-scale restoration using seeds (density 125 000–250 000 seeds ha−1) collected primarily from adjacent Chesapeake Bay was initiated in the late 1990s to speed the recovery of seagrass in these coastal bays (Orth et al. 2012). Seeds were used because they were more practical for large-scale restoration, and collections occurred in Chesapeake Bay because flowering shoots with large numbers of viable seeds were abundant and easily accessible (R.J. Orth, pers. comm.). In this study, we use population genetic analysis (microsatellite markers) to compare natural recovery with restored seagrass meadows. Our aim was first to estimate the existing genetic structure and natural population connectivity across the region to determine potential sources (regions) for the natural recruitment. Next, we evaluated the degree to which restoration changed the natural population genetic structure of the region through the introduction of genetic diversity from an adjacent system. Finally, we contrasted the genetic diversity of naturally recruited meadows with meadows restored with seed, and made a first-order comparison of recovery efficiency, estimated as time required to re-establish genetic diversity.
Materials and methods
Eelgrass tissues samples were collected from 23 meadows ranging from Woods Hole, MA, USA to Beaufort, NC, USA. This represents the southern half of the geographical extent of this species along the eastern coast of North America (Green & Short 2003). Sampling intensity was highest around Virginia, USA, including the Chesapeake Bay and the coastal bays on the ocean side of the Delmarva Peninsula (Fig. 1), because of the unique history of extinction, recolonization and restoration of this region. Recovery in this region is recent (< 15 years), and the young meadows present in this region are either a result of large-scale restoration or natural recruitment. Naturally recruited, restored and source meadows for restoration were included in this survey as well as the previously hypothesized source of natural recruitment (Chincoteague Bay; Harwell & Orth 2002; Reynolds et al. 2012) and other representative meadows both north and south of the region. See Table 1 for a summary of sampling sites.
Table 1. Summary of multilocus genetic diversity estimates for all 23 Zostera marina populations based on seven microsatellite loci. Sites refer to locations shown on Fig. 1
N, sample size; AR, allelic richness; Na, average number of alleles per locus; He, expected heterozygosity; F, Wright's inbreeding coefficient.
Data from CH, HR6, HR7, SC, SB, FI, HC, PC and PR were analysed in Reynolds et al. 2012. Data from WGSB, and PEC were analysed in Brisbin 2010 and Peterson et al. (2013). * denotes a restoration site.
Virginia Coastal Bays
Western Great South Bay
Hog Island Bay Restoration (2006)*
Hog Island Bay Restoration (2007)*
Hog Island Bay Natural Recovery
Spider Crab Bay*
Northern South Bay*
Southern South Bay
Four Point Marsh
At each of the 23 sampling locations, entire plants were collected haphazardly from areas at least 3 m apart to avoid recollection of the same clones. The number of samples from each site varied and is reported in Table 1. Despite differences in meadow size, the area of sample collection was similar. Leaf tissue was dried using silica gel desiccant and stored at room temperature until DNA analysis was conducted. Plants from the two New York sites [Peconic Estuary (PEC) and Western Great South Bay (WGSB)] are a subset of the samples described by Brisbin (2010) and Peterson et al. (2013).
Methods follow those described in Reynolds et al. (2012): DNA was extracted from leaf tissue using DNeasy™ plant extraction kits (Qiagen, Valencia, CA, USA) following manufacturer's instructions. Extracted DNA was amplified at seven polymorphic loci previously described for this species (CT3, GA2, GA3: described by Reusch, Stam & Olsen 1999; and CT17H, CT35, CT19 and CT20: described by Reusch 2000) using standard PCR techniques (Reusch, Stam & Olsen 1999). PCR products were analysed using capillary electrophoresis on a MegaBace 1000 (GE Biosciences, Pittsburg, PA, USA).
For each population sampled, a suite of general genetic diversity measurements were calculated. Allelic richness (AR), standardized to the smallest population size by rarefaction, was computed using fstat 18.104.22.168 (Goudet 2001). The average number of alleles per locus (A), mean expected heterozygosity (He) and Wright's inbreeding coefficient (F) was calculated using GenAlEx 6.3 (Peakall & Smouse 2006). Evidence of recent bottlenecks was analysed using a Wilcoxon test of the two-phase model (TMP) in the software package bottleneck v. 1.2.02 (Cornuet & Luikart 1996).
Estimation of regional genetic structure and connectivity
Analysis of molecular variance (amova) was used to partition the genetic variation within and between populations and also between regions (southern, Chesapeake Bay, the Virginia coastal bays, and northern). amova was conducted using GenAlEx 6.3 (Peakall & Smouse 2006).
The similarity of geographically separated meadows was analysed using four measures of population differentiation. Nei's standard genetic distance, FST (estimated as Q, Weir & Cockerham 1984) and F'ST (Hendrick 2005) were calculated using the software Genodive v 2.0 (Meirmans & Van Tienderen 2004). DEST (Jost 2008) was calculated using smogd v. 1.2.5 (Crawford 2010). The clustering of populations by Nei's distance and FST was visualized using a neighbour-joining tree as implemented in the program mega 5 (Tamura et al. 2011). Tree topologies are based on the pairwise distance matrices of each parameter and thus have no inbuilt measure of support beyond the existence of branch lengths estimated by the neighbour-joining tree building algorithms. Trees were constructed as a method for visualizing the overall patterns as opposed to an inferred parentage association.
All samples were assigned to genetic clusters (K: 1–10) regardless of geographical origin using a Bayesian modelling approach conducted in structure (Pritchard, Stephens & Donnelly 2000). All model runs had a random start value, a burn-in period of 50 000, and 100 000 reps, and this analysis was repeated 10 times. The number of distinct population clusters was determined using the Evanno, Regnaut & Goudet (2005) delta K method. The geographical distribution of clusters was analysed using a linear regression analysis of each cluster against latitude, and the effect of restoration on that distribution was analysed by regressions with and without those populations. The proportion of assignments to each cluster was used to create a population distance matrix by calculating average variance between populations and dividing by the variance among populations for each cluster. The grouping of clusters was visualized using a neighbour-joining tree as implemented in the program mega v. 5 (Tamura et al. 2011; Note the tree is presented to visualize the broader relationships as described above).
The relationship between geographical distance and similarity of populations was considered using an isolation by distance analysis. For this analysis, restored populations were omitted, since the dominant movement of genes was a result of human manipulation through restoration as opposed to natural processes. Geographical and genetic distance matrixes were calculated in GenoDive v 2.0 (Meirmans & Van Tienderen 2004). Geographical distances that involved meadows in Chesapeake Bay were manually corrected to include travelling distance around the Delmarva Peninsula as opposed to straight distances, which would require travel over land. Distances were analysed using a Mantel test using ibdws version 3.23 (Jensen, Bohonak & Kelley 2005) implementing 10,000 randomizations.
Migration among populations was analysed using three techniques. Pairwise estimates of geneflow (Nm) were calculated based on rare alleles using GenePop (Raymond and Rousset, http://wbiomed.curtin.edu.au/genepop/index.html) and from FST calculated in GenAlEx 6.3 (Peakall & Smouse 2006). The direction and rate of recent migration were estimated using a multi-locus genotype-based Bayesian approach with the software BayesAss (Wilson & Rannala 2003).
Differences in natural recruitment and restored populations
In the Virginia coastal bays, there were five restored populations [HR6, HR7, Spider Crab Bay (SC), SBN and South Bay (SB)] and three small populations [HN, Southern South Bay (SSB) and Fisherman Island (FI)] that have naturally recruited into the area. Mean allelic richness, expected heterozygosities and Wright's inbreeding coefficient (F) for each of these groups were compared using a t-test.
To compare the rate of natural recovery via metapopulations and large-scale anthropogenic restoration, we made a series of first-order calculations to estimate the time necessary for natural recovery to produce populations with the same genetic diversity as those produced by large-scale restoration. The calculation was made for an 80-year period, and we first made the conservative assumption that natural recovery of allelic richness was a linear process. This assumption is conservative in that allelic richness will increase as migration events into the area increased, allelic richness would in fact increase logarithmically. However, as we only have a static measure of migration rate over a long period of time, estimated from our measures of genetic diversity, a linear model can be applied as a starting point to estimate the minimum rate of population expansion. To model population expansion, we plotted total migration rate against the allelic richness for each population. For each of the natural and restored populations, we summed the migration rates (Nm) from all of the other populations. We used the Nm calculated based on FST, as this measure gives an integrated estimate of migration event. Other measures of gene flow were not appropriate as there were few rare alleles to use in the rare allele methods of estimating migration and because the Nm calculated using the program BayesAss only estimates very recent migration events. We conducted a regression analysis between total migration (summed Nm values) and allelic richness for the natural and restored populations independently. We determined the migration rate at the point where allelic richness for the two groups was the same and hypothesized that was the inflection point of the logarithmic curve. We used the linear approach of an increase in migration rate for natural populations to estimate a time in which that would occur.
Previous studies have described the Chesapeake Bay and Virginia coastal bays as one of the most diverse regions in the species' range (Reynolds et al. 2012). The survey data reported in the present study also show that most populations are quite diverse, with mean allelic richness of 5.0 and a mean expected heterozygosity of 0.6. However, there was one population in this survey that deviated from that trend. The naturally recruited SSB in the coastal bay region had a low allelic richness of 2.13 and an expected heterozygosity of only 0.3 (Table 1). At the scale sampled, clonal diversity was 1.0.
No population illustrated significant evidence of a population bottleneck. The allele frequency distribution was a normal L-shaped distribution for each of the sampled populations, and the probability values for a two-tail test for heterozygosity excess or deficiency ranged between 0.06 and 1, suggesting no signs of severe population bottlenecks.
Estimation of regional genetic structure and connectivity
amova detected 83% of molecular variance within populations, 14% among populations, and 4% among regions. Populations in this study showed a moderate degree of genetic structure. While some geographically distinct populations did not vary genetically, many did. We used three measures of genetic differentiation within populations that varied in magnitude (0.0 > FST>0.5; 0.0 > F'ST>0.8; 0.0 > DEST>0.6); however, the pattern of differences between sites did not vary between the different measures. The highest values (FST = 0.49, F'ST = 0.78, DEST = 0.6) were between the two sites with the lowest diversity: Allen's Island (AI) in Chesapeake Bay and SSB in the Virginia coastal bays. These high values were not typical, with most values being much lower (mean FST = 0.12 SE = 0.006, F'ST = 0.30 SE = 0.012, DEST = 0.19 SE = 0.009). The sites that were most similar were the sites in Western Chesapeake Bay (with the exception of Allen's Island) and the SB restoration site, which was restored using seed from Chesapeake Bay (Table S1 in Supporting Information). Populations clustered geographically using the neighbour-joining technique with populations north of the Virginia coastal bays clustering together, Chesapeake Bay populations clustering together, and North Carolina populations clustering together. The restored populations in the Virginia coastal bays clustered between the populations in Chesapeake Bay and the regions to the north, while two of the three naturally recruited Virginia coastal bay populations were relatively unique (Fig. S1).
The Bayesian cluster analysis in structure (Pritchard, Stephens & Donnelly 2000) followed by the delta K analysis (Evanno, Regnaut & Goudet 2005) found five distinct genetic clusters among the 23 geographically separated populations sampled. The distinct genetic clusters were not distributed evenly over the geographical range. A regression analysis showed that three of the five clusters had a significant relationship with latitude, with cluster one being dominant in the south and clusters 4 and 5 being dominant in the north. Cluster 3 best described the unique, low diversity, naturally recruited population of SSB in the Virginia coastal bays, and Cluster 2 was most abundant in Chesapeake Bay (Figs 1 and S2). Adding the restored Virginia coastal bay sites to these regressions did not change any of the patters or the statistical significance (Fig. S2). The meadows that naturally recruited into the Virginia coastal bays, however, were different from one another. Two of those populations, one in Hog Island Bay (HN) and one at Fisherman Island near the mouth of the Chesapeake (FI), were similar to populations to the north: Chincoteague Bay (CB) and Woods Hole (WH). The third naturally recruited population, SSB, grouped in its own distinct cluster. When population structure was visualized with a neighbour-joining tree, populations once again clustered geographically by region: the region to the north of Virginia, Chesapeake Bay and North Carolina, while the restored populations grouped between the northern region and Chesapeake Bay and the naturally recruited Virginia coastal bay meadows mostly paired with a meadow in the northern region (Fig. S1).
There was a weak positive relationship between geographical distance and genetic distance (R2 = 0.144, P =0.01), however, a single population deviated from this pattern. Southern South Bay is the naturally recruited population that had an anomalously low diversity, and its pairwise genetic distance was constantly high despite any change in geographical distance. It was therefore left out of the isolation by distance calculations (with this site R2 = 0.04, P =0.1).
Estimates of migration calculated using rare alleles varied from Nm = 0.2 to 17.3, with SSB having the lowest connectivity to AI in Chesapeake Bay, and the two restoration sites in Hog Island Bay (HR6 and HR7) having the highest rate of intermigration. This method of estimating migration was the only method that found a significant amount of migration with the Southern South Bay site (SC = 2.3 and Chincoteague Bay (CH) = 1.7) (Fig. 2 and Table S2). When estimates of migration were based on FST values, rates ranged from Nm = 0.3 to 124.8, with the lowest rates again being between SSB and AI, and the highest rates being between two sets of paired sites in western Chesapeake Bay: Sandy Point (SP) with Four Point Marsh (FP) and SP with Brown's Bay (BB) (Fig. 2, Table S3). Unlike estimates of migration based on FST and rare alleles, migration rates calculated using Bayesian assignment statistics in the program BayesAss showed only recent (within the last few generations) migrations and also indicated directionality. These rates ranged from 0.02 to 32.2 contemporary migrants, with the lowest migration rates being from one of the most southern sites at Morgan's Island (MI) in North Carolina to the furthest north site in WH. The reverse migration was similarly small (0.8). The highest rates were from the 2007 restoration plots in Hog Island Bay (HR6) into the 2006 restoration plots in Hog Island Bay (HR7), with the reverse migration being of similar magnitude (31.0) (Fig. 2 and Table S4).
Differences in natural recruitment and restored populations
As a whole, populations that recruited naturally into this region had a lower allelic richness (3.5 ± 0.7 SE vs. 5.5 ± 0.6 SE; t = 79.07, P =0.001) and a lower expected heterozygosity (0.5 ± 0.07 SE vs. 0.7 ± 0.005 SE; t = 95.49, P =0.008) than populations that recruited into the area by restoration. All inbreeding coefficients were close to zero, and there was no difference in inbreeding coefficients between naturally recruited and restored populations (−0.07 ± 0.07 SE vs. −0.08 ± 0.03 SE; t = 0.1, P =0.9). Southern South Bay, the lowest diversity site, did not significantly skew the results. Results were similar when this population was omitted from the analysis (Fig. 3).
The linear model of the increase in allelic richness estimates that it would take 125 years to reach the same level of diversity as populations restored by seeding. The assumption that the increase in migration rates is linear with time is overly simplistic. More complex logarithmic model estimates of the time needed for natural recovery (i.e. population growth model) to reach the genetic diversity achieved by large-scale restoration were longer (185 years) (Fig. 4).
Metapopulation dynamics have been important in the natural recovery of Zostera marina populations in the Virginia coastal bays, which had previously experienced catastrophic loss. In this study, we found that seagrass was recovering naturally as a result of dispersal from meadows to the north of the disturbed sites (MD, NY and MA). Natural recovery in this system occurred slowly—nearly 80 years after the meadows were lost due to disease, only three small Z. marina populations had recruited. Our genetic analyses indicated that those naturally recruited patches had a relatively low genetic diversity and showed signs of genetic drift, probably due to a small founding population and subsequent inbreeding. Through intervention via seed-based restoration, areal coverage (see Orth et al. 2012) has increased, and the resulting genetic diversity was similar to donor meadows (Reynolds et al. 2012). In this case, restoration is less likely than natural recovery to result in populations that exhibit issues such as inbreeding, because restoration involves a large pulsed addition of seeds rather than the small and slow addition of seeds as in natural recruitment. This restoration has been successful in establishing large (1600 ha) genetically diverse seagrass populations to the coastal bays within 10 years. While this coverage may have been restored naturally, it would have taken an order of magnitude more time. And there is no evidence that restoration has altered the region by changing the overall population genetic structure, despite the collection of seeds from regions that would not likely be connected via natural dispersal. The genetic clustering of restored meadows between meadows to the north and south of the region and the migration connections between the northern and Chesapeake Bay populations (Fig. 2) suggest that the restored genetic structure of the region probably reflects the regional structure prior to the 1930s disease.
Zostera marina has mechanisms for long-distance dispersal, which allows this species to establish regional metapopulations (Kendrick et al. 2012). Long-distance dispersal in Z. marina can occur when flowering shoots break free and raft with the currents. Seeds can remain viable within rafting shoots for up to 3 weeks, meaning that seeds can be transported long distances (100–150 km), potentially into habitats that are suitable for germination and survival (Harwell & Orth 2002; Reusch 2002; Källström et al. 2008). The primary near-shore currents along the mid-Atlantic coast of North America run from north to south (Leatherman, Rice & Goldsmith 1982). Therefore, one would expect that most movement of reproductive shoots would be from north to south unless seeds were transported by other vectors such as animals, storm currents or boat propellers that do not necessarily depend on the prevailing wind or currents. Genetic evidence (distances and structure) from natural meadows supports the north–south movement of seeds into the Virginia coastal bays, with the naturally recruited meadows in Hog Island (HN) and FI pairing with an established meadow to the north (Fig. S1). Further, restored populations also share a genetic signal with the northern population in Chincoteague; suggesting that restored plots may also be receiving new recruits from the north or that natural recruited populations are mixing (Reynolds et al. 2012).
Southern South Bay was somewhat anomalous in that its genetic structure was unique; however, one estimate of migration (the rare allele method) showed connection with CB, consistent with southerly migration (Fig. 2). Because Southern South Bay does not look like any of the populations sampled (Fig. 1), the recruitment event that established this population could have been small and happened some time ago, and the population has undergone genetic drift due to founder effects so that it no longer resembles the broader diversity of its seed source. There is no evidence of a population bottleneck, so if this is the case, this is not a recent recruitment event and bottleneck effects have disappeared with time via sexual reproduction. This hypothesis is supported by the relatively low diversity (Fig. 3) and dominance of a single genetic lineage (Fig. 1) for other naturally recruiting populations. Alternatively, the source of the seeds could be from a low-diversity population in between those included in our study that we did not sample, since over the 450 km of coastline to the north of the natural recruitment, we only sampled four populations (Chincoteague, Woods Hole, Peconic, and Western Great South Bay). Many seagrass populations along this coast are declining (Waycott et al. 2009), and some of those are showing signs of genetic erosion (Campanella et al. 2009). Migration from one of these meadows also could result in the conditions observed in Southern South Bay. Fisherman Island pairs with a meadow that is nearly 700 km away (WH), which also suggests that there are probably some intermediate metapopulations that were not sampled. This may explain why the placement of Fisherman Island in the geographical structure is not consistent (e.g. it pairs with a northern population with structure analysis but not genetic distance: Fig. S1). A more targeted and detailed sampling would be required to determine the exact parental source of naturally recruited populations; however, this study shows the general flow of propagules was from north to south.
While there were some differences among different estimates of connectivity and there are limitations of each method, all of the different methods indicate metapopulation dynamics and indicate a similar pattern of general northern to southern migration. The migration rates based on Bayesian modelling (BayesAss) have the benefit of estimating directionality in a straightforward way; however, they only reflect what has happened in the last few generations. While the rare allele and FST method do not estimate directionality, because the naturally recruited meadows in the Virginia coastal bays are relatively young, we can assume that when they pair with an established meadow, dispersal is occurring from that established meadows into the young recruited meadow. These assumptions are confirmed by the BayesAss results. Most sites in this survey, spanning over 1000 km of coastline, exhibited high rates of connectivity among sampled populations. We therefore would not expect there to be significant numbers of rare alleles. The pairs of sites with significant migration rates uniquely identified by analysis using rare alleles are those that were relatively far apart [i.e. CB and Pepper Creek in Chesapeake Bay (PC)], or where the site's low diversity suggests either a very small recruitment or an older migration event that had experienced genetic drift (i.e. CB and SSB). Estimates based on overall similarity of populations (FST) show much stronger relationships, where we expect high migration rates among populations that were nearby. For example, Brown's Bay (BB) and PC are geographically close in Western Chesapeake Bay and show significant migration (Nm = 83.08), and Brown's Bay (BB) in Chesapeake Bay was used as a seed source for the SB and this is reflected in a Nm of 62.25. Migration rates and dispersal distances for this species can be high; therefore, it is logical that few rare alleles would exist, and thus, the rare-allele method would underestimate migration rates. Estimates of Nm based on FST assume that populations are of the same size and that migration occurs in both directions at similar rates, which is not true.
While populations in this region have the potential for long-distance dispersal, natural recruitment by dispersal and metapopulation dynamics is a slow process due to natural barriers in the Virginia coastal bay system. For example, the tidal channels connecting the Virginia coastal bays to the open ocean are relatively narrow, limiting the connection to the open ocean. On average, only 10% of total shoots flower (Olesen 1999), and of those flowering shoots, only a portion will break off and raft with the current at an appropriate time in their life cycle so that they have fertilized fruits that will develop into viable seeds. Previous research has shown that around 5–10% of seeds in this area germinate and develop into viable seedlings (Orth et al. 2006). Mortality of small patches can be quite high (Olsen & Sand-Jensen 1994). Barriers and limited propagules are likely to limit the founding seed population and thus naturally recruiting populations should exhibit a founder effect and genetic drift (Kendrick et al. 2012). The three natural meadows in this region showed a decreased genetic diversity (Fig. 3) and were dominated by a single genetic cluster (Fig. 1), suggesting that these meadows have been subject to genetic drift, concurrent with a small seed addition and in contrast with the large addition of seeds during restoration. Further, natural meadows showed connectivity with different established meadows to the north. This is consistent with the idea that this process is episodic, opportunistic, and slow. While the southern Virginia coastal bays have good water quality and can support seagrass expansion (Lawson et al. 2007; www1.vcrlter.virginia.edu/home1/?q=data_wq), coastal bays to the north are experiencing large declines in seagrass primarily due to water quality (Short et al. 2006) and thus may have insufficient effective population size and number of propagules which could limit the long-term capacity for natural recruitment.
Restoration was faster than natural recruitment into this dispersal-limited environment. Based on migration rates, we estimate that natural recovery would take 125–185 years to achieve the same level of allelic diversity as the populations that were restored in only 10 years. The Virginia coastal bay habitat is already being modified and ecosystem services have been enhanced by the return of seagrasses to the Virginia coastal bays through restoration (McGlathery et al. 2012). The ecological, economic, and cultural benefits of this restoration will be orders of magnitude more than that of a naturally restored region because they will be available for a longer period of time (see Fonseca, Julius & Kenworthy 2000). Full recovery by natural processes would not be accomplished for many years. The success of this restoration has been in part due to the selection of source material, a technique that maintains the diversity of these well-chosen donors (Reynolds et al. 2012), and excellent habitat conditions. While the movement of seeds over long distances by active restoration may have negative impacts on the larger ecosystem by adding in foreign genotypes, which can result in outbreeding depression (Hufford & Mazer 2003), the overall distribution of genetic structure and evidence of migration in the region surrounding this restoration suggest that there is little evidence for these problems. Indeed, Chesapeake Bay (to the south of the restoration site) shares a genetic signal with populations to the north of the restoration site (the New York sites of WGSB and PEC), and Nm values suggest that there has been historical (and even recent: Fig. 2c) migration between the two areas (Figs 1 and 2). Because the restored meadows currently group in between meadows to the north and south of the region (Figs 1 and S2) and because including the restored sites does not alter the latitudinal distribution of genetic clusters (Fig. S2), it is likely that the restored genetic structure is similar to that which existed prior to the 1930s die-back, and because of differences in the density of seeds, restoration is less likely to result in problems such as inbreeding or population bottlenecks than natural recovery. This system might have fully recovered via natural mechanisms; however, recovery would be slow and may have been hindered due to impacts to nearby systems. Large-scale restoration was a faster process, and because the connectivity of the larger system was not disrupted, the faster recovery will result in meadows that persist for a long time and thus provide more ecosystem services.
Differences in population structure between natural and restored populations will not be consistent among all species and situations. Many of the Zostera marina populations in southern Chesapeake Bay were decimated by the wasting disease of the 1930s (Orth & Moore 1984) but recovered naturally and have a high genetic diversity (Table 1, Reynolds et al. 2012). This is probably due to an abundance of seed from the more northern parts of Chesapeake Bay. These regions do not have barriers to dispersal such as those that exist with the presence of barrier islands along the coast. In other systems, such as vegetated riparian zones, restored populations can quickly recover simply by restoring water flow and thus connectivity since water flow acts as a dispersal agent (Kauffman et al. 1997). Similarly, forest recruitment onto old landfills is facilitated by insect and bird dispersal of seeds (Robinson & Handel 1993). In these two examples, the recovered populations are less likely to differ from restored populations. However, as in this study, the benefits of restoration (i.e. it is faster and more effective at creating populations with increased genetic diversity) are likely to be found in other dispersal-limited systems, such as mangroves near their geographical limit (Maguire et al. 2000) and grasslands which have been fragmented by large farms (Bakker & Berendse 1999).
We thank R.J. Orth and W.J. Kenworthy for facilitating the collection of Chesapeake Bay and North Carolina samples, and A. Calladine for assistance with graphics. Support of this study was provided by the Virginia Coast Reserve LTER project (NSF grant DEB-0621014), and by the Jones Environmental Research Endowment to the Department of Environmental Sciences at the University of Virginia.