A century of genetic variation inferred from a persistent soil‐stored seed bank

Abstract Stratigraphic accretion of dormant propagules in soil can result in natural archives useful for studying ecological and evolutionary responses to environmental change. Few attempts have been made, however, to use soil‐stored seed banks as natural archives, in part because of concerns over nonrandom attrition and mixed stratification. Here, we examine the persistent seed bank of Schoenoplectus americanus, a foundational brackish marsh sedge, to determine whether it can serve as a resource for reconstructing historical records of demographic and population genetic variation. After assembling profiles of the seed bank from radionuclide‐dated soil cores, we germinated seeds to “resurrect” cohorts spanning the 20th century. Using microsatellite markers, we assessed genetic diversity and differentiation among depth cohorts, drawing comparisons to extant plants at the study site and in nearby and more distant marshes. We found that seed density peaked at intermediate soil depths. We also detected genotypic differences among cohorts as well as between cohorts and extant plants. Genetic diversity did not decline with depth, indicating that the observed pattern of differentiation is not due to attrition. Patterns of differentiation within and among extant marshes also suggest that local populations persist as aggregates of small clones, likely reflecting repeated seedling recruitment and low immigration from admixed regional gene pools. These findings indicate that persistent and stratified soil‐stored seed banks merit further consideration as resources for reconstructing decadal‐ to century‐long records that can lend insight into the tempo and nature of ecological and evolutionary processes that shape populations over time.

Few attempts have been made, however, to reconstruct historical records of genetic variation to infer ecological and evolutionary responses of plants to environmental change from time-stratified soilstored seed banks Gugerli, Parducci, & Petit, 2005;McGraw, 1993;Morris, Baucom, & Cruzan, 2002;.
Biased representation and poor stratification are two wellrecognized concerns that have deterred use of soil-stored seed banks for reconstructing records of genetic variation and other aspects of organismal evolution (Brendonck & De Meester, 2003;Franks & Weis, 2008;Weis, 2018). Bias can arise because, for many plants, only a fraction of seeds that fall to the ground enter the seed bank (Templeton & Levin, 1979). Nonrandom attrition of buried seeds or selection acting on traits associated with germination can further bias the composition of a seed bank over time (Weis, 2018).
Mixing or weak stratification of soil layers also can confound relative and absolute aging of buried propagules (Brendonck & De Meester, 2003;Franks & Weis, 2008;Hairston & Kearns, 2002). Steps can be taken, however, to mitigate both concerns. For example, targeting a species with prolific seed production can reduce the likelihood of biased representation and false signatures of selection (Brendonck & De Meester, 2003;Weider et al., 1997). In addition, seeds from distinct depth ranges can be treated as age-relative "cohorts" (Morris et al., 2002) and, like resting stage eggs, seeds can be precisely dated when recovered from depositional environments, such as freshwater lakes and coastal wetlands, with highly stratified sediments Brendonck & De Meester, 2003;Jarrell et al., 2016;Törnqvist et al., 2004;Vavrek et al., 1991).
Prior use of the soil-stored seed bank of the foundational coastal marsh sedge Schoenoplectus americanus (Pers.) Volkart ex Schinz & R. Keller (Cyperaceae) for studying paleoecological responses to environmental change (e.g., Jarrell et al., 2016;Saunders, 2003;Törnqvist et al., 2004) indicates that it also could be a valuable resource for reconstructing historical records of genetic variation.
Depending on the condition of buried seeds, it also might be possible to characterize genetic variation over time to draw further inferences about the tempo and nature of S. americanus responses to environmental change.
In this study, we evaluated the extent to which soil-stored seed banks of S. americanus can serve as natural archives for reconstructing historical records of demographic and genetic variation. We first assessed whether sediments exhibited a time-stratified structure characteristic of recurring deposition and accumulation. We then assessed whether seed densities steadily declined with soil depths or exhibited variation consistent with shifts in the abundance of S. americanus through time (Jarrell et al., 2016). We also assessed whether genetic diversity declined with increasing soil depth, which can result from attrition or differences in germination bias (Orsini et al., 2016). In addition, we assessed whether estimates of genetic structure and pairwise measures of genetic distance varied erratically with increasing soil depth, which can also result from nonrandom attrition and bias. We did so by first reconstructing the stratigraphy of buried seeds from 210 Pb and 137 Cs dated soil cores.
We then germinated seeds to "resurrect" and genotype cohorts spanning the 20th century. Using a suite of microsatellite markers, we inferred patterns of genetic diversity and differentiation among "resurrected" cohorts, drawing comparisons to extant plants at the coring site as well as in nearby and more distant marshes across the Atlantic and Gulf coasts. In addition to offering perspective on the potential importance of nonrandom bias, this approach enabled us to infer whether patterns of temporal variation reflect immigration or local population differentiation (Holt, 1990). It also enabled us to bypass concerns about DNA contamination of buried seeds (Anderson-Carpenter et al., 2011;Gugerli et al., 2005) and assess whether soil-stored seed banks can serve as resources for assembling experimental populations to study adaptive evolution to contemporary environmental change (Franks et al., 2007).

| Soil excavation site, seed recovery and estimation of accretion rates
We excavated sediment cores from Kirkpatrick Marsh (Table 1), which is the site of the Global Change Research Wetland (GCReW) operated by the Smithsonian Environmental Research Center (Arp et al., 1993;Broome et al., 1995;Rasse et al., 2005). The GCReW has supported several studies that span 30+ years of investigation (e.g., Curtis, Drake, & Whigham, 1989;Lu et al., 2016) of ecosystemlevel responses to elevated CO 2 (Drake, 2014), nitrogen (Langley & Megonigal, 2010), invasive species (Caplan, Hager, Megonigal, & Mozdzer, 2015), and warming . As a dominant species of the GCReW plant community, S. americanus has featured prominently in much of this work. Kirkpatrick Marsh borders the Rhode River, a subestuary of Chesapeake Bay near Edgewater, Maryland (38º 51′N, 76º 32′W). Elevation of the marsh is 40-60 cm above mean low water, with 20% of high tides flooding the site (Jordan, Pierce, & Correll, 1986). Soil salinity ranges from 2 ppt to 18 ppt during the growing season (May to September), where interannual variation in growing season salinity is inversely correlated with rainfall (Saunders, 2003).
We reconstructed soil stratigraphy and seed bank profiles from a set of soil cores taken in Kirkpatrick Marsh. As described by Saunders (2003) and Saunders et al. (2006), a series of 70-cm-deep piston cores (5.1 cm diameter) were excavated between 1997 and 2000 at four-month intervals for a study quantifying depth profiles of C 3 and C 4 belowground biomass in eleven 1.5-m 2 plots in the marsh. Soil core samples from a 1.5-m 2 plot with equal amounts of C 3 (S. americanus) and C 4 (e.g., Spartina patens, Distichlis spicata) aboveground biomass were used to quantify a vertical profile of S. americanus seeds (Table 1, Figure 1). In October 2002, we removed a supplemental 30 cm diameter × 35 cm deep core (hereafter referred to as a "soil monolith") to recover additional S. americanus seeds for germination assays. The soil monolith was taken adjacent to the mixed C 3 -C 4 study plot (Plot #15; in February 2004, we removed two more piston cores (15.2 cm diameter, 30 cm apart) 2 m from where the soil monolith was taken to further quantify the vertical profile of S. americanus seeds (Figure 1), to recover more seeds for germination assays (Table 1), and to estimate accretion rates.
Following removal, all sampled soil was transported to Duke University for processing and analysis. The 2002 soil monolith was sliced into 2 cm increments perpendicular to the vertical axis for recovery and germination of S. americanus seed cohorts (Table 1).
The first 2004 piston core ("core 2004-A," 65 cm deep) was also cut into 2 cm layers, with one half of each layer dry-sieved over a 2 mm mesh (to remove large roots and rhizomes) in preparation for radionuclide analysis of 210 Pb and 137 Cs (Saunders, 2003). Soil dates from 210 Pb radionuclide data were estimated according to the constant rate of supply model (Appleby & Oldfield, 1978) to allow for variable accretion over time, as accretion rates in Chesapeake Bay marshes have fluctuated over the last 200 years (Kearney, 1996;Kearney, Stevenson, & Ward, 1994). Variability in soil dates was calculated by first-order error analysis of counting uncertainty (Binford, 1990).
The depth of peak 137 Cs activity was used as an independent marker of the depth corresponding to 1964, the year when 137 Cs reached peak concentrations in the atmosphere. The remaining soil from core 2004-A was used to recover additional seeds for germination and for reconstructing the seed bank profile (Table 1, Figure 1). The second piston core ("core 2004-B"; 40 cm deep) was used to recover additional seeds from soil horizons deeper than 8 cm for germination assays.

| Tissue sampling of extant populations
A 10 cm long tissue sample was trimmed from one to six green shoots per plot ( Table 2). As S. americanus can reproduce asexually through vegetative tillering, care was taken to sample evenly across each plot to minimize repeated sampling of the same genet. The relative location of each sample was noted according to the UTM coordinates of the plot, which were spaced ≥2.5 m apart.
The majority of the plots were established in 1987 to study ecological and physiological responses of S. americanus and cooccurring C 4 species to elevated atmospheric CO 2 (Arp et al., 1993).
Accordingly, these plots differ in CO 2 exposure regime ( Table 2).
The remaining plots were established in 1997 for the study of marsh biogeochemistry (Saunders, 2003;Saunders et al., 2006).
The vegetative composition of the plots ranged from monospecific stands of S. americanus, to mixed communities where S. americanus co-occurs with S. patens and other C 4 plant species, to stands dominated by S. patens (Arp et al., 1993;Saunders, 2003; Table 2).  Table S1). At each location, complete or nearly complete shoots with seed-bearing inflorescences were taken from plants spaced ≥3 m apart. The coordinates of individual samples from these marshes were not taken. All tissue samples were stored in coolers with ice packs for transport to long-term storage in −20°C freezers.

| Genetic data collection
We genotyped all resurrected and extant specimens at 11 microsatellite loci to examine patterns of temporal and spatial genetic variation (Blum et al., 2005). Genomic DNA was extracted from shoot tissue from all samples using DNeasy plant extraction kits (Qiagen, Inc.). The loci SCAM.4, SCAM.5, SCAM.7, SCAM.8, SCAM.11, SCAM.13, SCAM.14, and SCAM.16 described in Blum et al. (2005) were used to genotype all individuals. Three additional loci, corresponding to the following primer sets, were also used in this

| Clonality, genetic diversity, and effective population size
We first determined the number of multilocus genotypes (G) and the proportion of samples exhibiting a distinct genotype (R) for each depth cohort and sample site using the program GenAlEx v.6.41 (Table 2, Supporting Information Table S1) (Peakall & Smouse, 2006).
We also assessed the probability that shoots with identical genotypes were members of the same clone using the Pgen routine in the program GenClone v.2.1 (Arnaud-Haond & Belkhir, 2007;Parks & Werth, 1993). In addition, we calculated the probability of sampling a second occurrence of each genotype given the number of genets sampled using Psex (Parks & Werth, 1993), and we used GenClone We tested for declines in genetic diversity with increasing depthan expected outcome of attrition and germination bias (Orsini et al., 2016) -using a linear regression and a Kolmogorov-Smirnov test with two potential expected outcomes (i.e., declines in diversity and no change in diversity), both of which were implemented in R v.3.4.0 (R Core Team, 2013). We similarly tested for declines in N e with depth.
Using the R v.3.4.0 core package (R Core Team, 2013), we also examined correlations between estimates of genetic diversity and N e with seed density, which has served as a proxy measure for the relative abundance of S. americanus over time (Jarrell et al., 2016;Saunders, 2003).

| Genetic and genotypic differentiation
We used GenAlEx v.6.41 to conduct an analysis of molecular variance (AMOVA) to examine the distribution of genetic variation across depth cohorts. We also performed AMOVAs with samples grouped according to age (i.e., depth cohorts vs. extant S. americanus in Kirkpatrick Marsh) and location (i.e., among Chesapeake Bay marshes, Atlantic versus Gulf coast marshes). In addition, we conducted AMOVAs to assess whether genetic variation in extant S.
americanus reflects CO 2 exposure regime and community type (respectively) across the sampled plots in Kirkpatrick Marsh.
We assessed patterns of genetic structure according to allele frequency variation using several complementary methods. Using Genetix v.4.05 (Belkhir, Borsa, Chikhi, Goudet, & Bonhomme, 1996), we conducted a factorial correspondence analysis (FCA) of genetic variation in depth cohorts and extant S. americanus in Kirkpatrick Marsh. MSA was used to calculate and bootstrap the variance in the proportion of shared alleles 1,000 times across depth cohorts and a selection of extant populations to construct a UPGMA dendrogram using the "Neighbor" and "Consense" subroutines of PHYLIP v3.63 (Bowcock et al., 1994;Felsenstein, 2004) Figure S2) with uniform priors and starting parameters set to Brownian motion for microsatellite data. We used F ST calculations to determine theta and M values.
We also estimated genetic structure and genotypic variation

| Seed bank profile, seed ages and seed germination
Seeds of S. americanus were recovered across the full length of the sediment cores taken in Kirkpatrick Marsh ( were grouped, the highest germination rates occurred in cohorts recovered from 6-8 cm (1990 ± 1.3) and 10-12 cm (1976 ± 1.2) depths (Table 1). However, germination rates were statistically equivalent in soil depths above 14-16 cm (1947 ± 4.2), after which rates dropped by as much as 90% (Table 1). Germination rates were generally lower in our second assay than in our initial trial, particularly for seeds recovered from depths below 14 cm (Table 1).

| Genetic diversity and effective population size through time
We  (Table 2). All "resurrected" individuals exhibited distinct genotypes.
No relationship was found between measures of genetic diversity and depth according to post hoc least-squares linear regressions (all r 2 < 0.08, all p > 0.05). Genetic diversity across the length of the core could not be distinguished from a null, even distribution (p = 0.329). Similarly, N e was not related to depth (r 2 = 0.38, p = 0.16), nor did it deviate from an even distribution of N e (p = 0.081) ( Table 2).
However, the 2-4 cm depth cohort exhibited a notably larger N e than all the other depth cohorts (Table 2). Nonsignificant trends were recovered between estimates of genetic diversity and N e with seed density (r = 0.63, p = 0.26; r = −0.40, p = 0.26, respectively).

| Genetic and genotypic differentiation through time
We detected evidence of genetic structure and genotypic shifts among depth cohorts. Approximately 3% of genetic variation was attributable to differences among depth cohorts, compared to 70% of variation attributable to differences within cohorts (Supporting Information Table S2). Mantel tests illustrated that genetic differentiation increased with increasing differences in depth (i.e., time) (Supporting Information Figure S1). STRUCTURE runs at K = 3 and K = 5 also showed that the genotypic composition of depth cohorts has shifted over time (Figure 3). Both the NJ dendrogram and FCA illustrated that a distinct shift between cohorts occurred across a depth horizon corresponding to ca. 1947 (Figure 2).

| Comparison of historical and extant genetic variation in Kirkpatrick Marsh
With one exception (the N e estimate for the 2-4 cm depth cohort),   (Figure 3). A Mantel test affirmed that genetic variation is associated with geographic distance across the marsh (Supporting Information Figure S1). An AMOVA showed that 23% of genetic variation is attributable to differences among plots when grouped by community (Supporting Information Table S2). The

AMOVA of plots grouped by experimental treatment indicates that
variance is not attributable to CO 2 exposure regime (Supporting Information Patterns of fine-scale variation that appear to correspond to community type also parallel clone size and distribution across Kirkpatrick Marsh (Figure 3). Whereas all individuals from depth cohorts exhibited distinct genotypes, duplicate genotypes were detected at nearly every site where we sampled extant S. americanus (Table 2). Consequently, site-level genotypic richness ranged from 0 to 1 (Table 2). In total, we detected duplicate genotypes in 333 samples (Table 2), including about half (55 of 109) of the specimens collected from Kirkpatrick Marsh plots (Table 2). This corresponded to a probability of <1.44e-07 that shoots with identical genotypes were members of the same clone across the marsh, and a probability of <3.44e-06 of sampling a second occurrence of each genotype given the number of genets sampled in the marsh. A clonal subrange value, which corresponds to the minimum estimate of the maximum distance between two identical genotypes (i.e., reflecting the distance for which the probability of clonal identity becomes zero), was estimated at approximately 43 m in Kirkpatrick Marsh (Alberto et al., 2005;Harada et al., 1997).

| Genetic variation in extant Schoenoplectus americanus
Genetic diversity of depth cohorts and of extant S. americanus (excluding duplicate genotypes) in Kirkpatrick Marsh was comparable to levels of genetic diversity found in other Chesapeake Bay marshes (all comparisons p > 0.05). Estimates of genetic diversity also did not differ among Atlantic and Gulf coast sites (all comparisons p > 0.05), and no clear geographic patterns in diversity were observed across either coastline (Supporting Information Table S1).
We detected evidence of genetic structure across Chesapeake Bay marshes and across coastlines (Figure 3). Excluding duplicate genotypes, pairwise values of F ST ranged from 0.06 to 0.48 among sample sites in Chesapeake Bay. An AMOVA attributed 27% (p < 0.001) of genetic variance to differences among sample sites. A Mantel test indicated that genetic distance corresponds to geographic distance (Supporting Information Figure S1); however, STRUCTURE runs at optimal K values (K = 4) revealed a more complex configuration of spatial differentiation in the embayment (Figure 3).
Clusters aggregated nearby sites with one or two disjunct locations ( Figure 3). Estimates of N m between clusters ranged from 0.30 to 4.56 (Supporting Information Figure S2). We also detected a significant relationship between genetic distance and geographic distance across coastlines (Supporting Information Figure S1), and an AMOVA of all sites grouped by coast recovered a significant global F ST of 0.23 (p < 0.001, Supporting Information Table S2). STRUCTURE runs with all unique genotypes (i.e., historical and extant samples) recovered clusters reflecting biogeographic breaks (i.e., Atlantic versus Gulf coast sites), as well as regional differences along coastlines (e.g., south Atlantic versus mid-Atlantic sites). STRUCTURE runs with all unique genotypes also demonstrated that depth cohorts bear the greatest resemblance to extant S. americanus in Kirkpatrick Marsh ( Figure 3).

| D ISCUSS I ON
Here we illustrate that, like other dormant propagule pools, soilstored seed banks can serve as a resource for studying demographic and genetic variation over time. Although concerns about biased representation and stratification have discouraged interest in soilstored seed banks as natural archives, our findings indicate that both can be constrained and potentially overcome. For example, concerns about biased representation can be minimized by examining species, like S. americanus, that exhibit prolific seed production and that produce highly persistent seeds that readily incorporate into the seed bank. Similarly, concerns about stratification can be overcome by examining seed banks that develop through sedimentary deposition. Although stringent, we have illustrated that it is possible to meet these conditions under relatively ordinary circumstances (i.e., by examining a widespread species found in a common environment). We have shown that S. americanus seeds can be recovered from radionuclide-dated sedimentary layers spanning 100+ years.
Comparisons of genetic diversity among contemporary populations and depth cohorts constructed from recovered seeds also indicate that postburial attrition and potential germination bias exert little influence on genetic measures of local demography. Evidence of shifting abundance, alongside spatial and temporal patterns of differentiation, further illustrate that soil-stored seed banks can lend insight into the tempo and nature of ecological and evolutionary processes that shape populations over time.
Sediment deposition and accumulation in marshes also can result in recurring burial and storage of seeds, particularly of seeds with durable coats (Fox, 1983;Honda, 2008;Moody-Weis & Alexander, 2007) like those produced by Schoenoplectus sedges. In addition, other buried contents (e.g., diatoms) and attributes (e.g., mineral versus organic content, isotopic profiles) of marsh sediments can be examined to obtain information about past environmental conditions (e.g., inundation, salinity regimes) that determine plant performance (Kirwan & Murray, 2007;Park, Yu, Lim, & Shin, 2012). This can afford opportunities to relate proxy measures of plant demography like seed abundance with measures of environmental change over time (e.g., Jarrell et al., 2016;Saunders, 2003).
This study explores the prospects of exploiting a virtually untapped dimension of soil-stored seed banks. Prior studies have largely utilized soil-stored seed banks as resources to reconstruct records of past geological, climate-related environmental conditions (e.g., Jarrell et al., 2016;Törnqvist et al., 2004). There is also an extensive literature on the contribution of seed banks to demography and genetic diversity (e.g., Cabin, Marshall, & Mitchell, 2000;Hegazy, Kabiel, Al-Rowaily, Faisal, & Doma, 2014;Liu et al., 2014;Templeton & Levin, 1979). Little work has been done, however, on the use of soil-stored seed banks for reconstructing records of genetic variation over time. Notably, McGraw, Vavrek, and Bennington (1991) highlighted the potential to do so by germinating Carex bigelowii and Luzula parviflora seeds recovered from tundra soil. Associated common garden experiments showed that depth cohorts of both species spanning ~150-200 years exhibited heritable differences in growth and morphological traits Vavrek et al., 1991). Using protein electrophoresis, Morris et al. (2002) also detected evidence of temporal variation among plants germinated from Astragalus bibullatus seeds recovered from successively deeper soil horizons sampled from the periphery of cedar glades in central Tennessee (USA). Our work further illustrates that genetic information can be extracted from soil-stored seed banks and that it can be contextualized by a well-constrained stratigraphic record as well as complementary information on local demography (i.e., shifts in seed densities) to draw inferences about ecological and evolutionary processes that shape populations over time.
We have shown that it is possible to overcome concerns about biased representation. As work on ephippia banks has demonstrated, a priori targeting a species with prolific seed production, like S. americanus, can reduce the likelihood of biased representation (Brendonck & De Meester, 2003;Cabin, 1996;Weider et al., 1997).
Nonetheless, stochastic attrition and selection can bias the composition of dormant propagule banks over time (Weis, 2018). Biases can arise due to differences in germination at the time of seed production (Cabin, Mitchell, & Marshall, 1998;Levin, 1990;Mandák, Bímová, Mahelka, & Plačková, 2006) or if some seeds are more prone to decomposition or are less resilient to burial than others (Weis, 2018). Similarly, seed viability might vary, where some seeds are less likely to germinate after prolonged dormancy than others (Honda, 2008;Levin, 1990;Wagner & Oplinger, 2017;Weis, 2018).
The S. americanus seed profile reconstructed from Kirkpatrick Marsh suggests that decomposition may have reduced seed abundance at depths greater than 40 cm, although it is possible that the decline in abundance instead reflects environmental conditions unfavorable to S. americanus (Jarrell et al., 2016). Thus, the observed decline may reflect historical trends in relative abundance and associated metrics like seed production rather than decomposition (Jarrell et al., 2016;Saunders, 2003). Germination rates, however, were only statistically equivalent for seeds recovered from depths up to 16 cm; rates dropped at greater depths (Table 1). While this suggests that burial is an important consideration, we did not detect genetic evidence that attrition or differences in germination biased the diversity of revived depth cohorts (Orsini et al., 2016). For example, we did not detect a loss of genetic diversity with increasing depth. This differs from prior studies that have detected aggregate measures of reduced genetic diversity (Cheliak, Dancik, Morgan, Yeh, & Strobeck, 1985;McCue & Holtsford, 1998;Orsini et al., 2016) and elevated genetic diversity in soil-stored seed banks (Cabin, 1996;Mandák et al., 2006;Tonsor, Kalisz, Fisher, & Holtsford, 1993), which can arise due to selective differences in seed germination (Cabin, 1996;Levin, 1990;Mandák et al., 2006). Notably, we found that the genetic diversity of depth cohorts was comparable to the extant population, which is consistent with reports of genetic diversity in seed banks being a representative measure of local genetic variation (Honnay, Bossuyt, Jacquemyn, Shimono, & Uchiyama, 2008).
We also have demonstrated that it is possible to overcome concerns about stratigraphy. No signs of sediment mixing were evident in this study. Consistent with prior work in tundra and interior wetlands showing that dormant seeds can be recovered from age-stratified soils McGraw et al., 1991;Vavrek et al., 1991), the laminate structure and radionuclide-based age estimates of sediment sampled from Kirkpatrick Marsh demonstrated patterns of historical accumulation over a 150+ year period.
A key next step, however, will be to reduce error rates of sediment age estimates. Error rates from 210 Pb dating typically increase with depth (Table 1) (Binford, 1990;MacKenzie, Hardie, Farmer, Eades, & Pulford, 2011), and whereas 137 Cs profiles can serve as referential benchmarks, more precise age estimates might be achieved through other approaches such as 7 Be radionuclide dating (Olsen, Larsen, Lowry, Cutshall, & Nichols, 1986) or optically stimulated luminescence dating (Madsen, Murray, Andersen, Pejrup, & Breuning-Madsen, 2005). Nevertheless, the observed pattern of progressive genetic differentiation over time (i.e., as opposed to genetic homogeneity) serves as supporting evidence that mixing did not disturb the sequence of the sampled stratigraphy (Orsini et al., 2016), as diversity and autocorrelation have been found to be lower in mixed sediment compared to undisturbed seed banks (England et al., 2003).
Local and range-wide geographic comparisons offer an informative context for interpreting temporal patterns of genetic variation.
We found that S. americanus exhibits a pattern of increasing dissimilarity with greater geographic distance, which is similar to patterns exhibited by other marsh plants (Blum, Jun Bando, Katz, & Strong, 2007;Mahy, Sloover, & Jacquemart, 1998;Travis & Hester, 2005;Travis, Proffitt, & Ritland, 2004). This, alongside evidence of genetic continuity and similarity between the seed bank and spatially proximate extant individuals in Kirkpatrick Marsh (Figure 2), indicates that immigration into the marsh is low (Supporting Information Figure S2) and that recruitment consistently draws from a local propagule pool (Honnay et al., 2008). Evidence that temporal variation is nested within spatial variation also indicates that genotypes "archived" in the soil-stored seed bank are likely ancestral to genotypes in the extant population. Consistent with this, the observed patterns of hierarchically structured spatial genetic variation across the Chesapeake Bay suggest that individual or spatially proximate marsh complexes constitute (sub)populations connected by relatively low gene flow (Supporting Information Figures S1 and S2). Comparisons among marshes elsewhere on the Atlantic and Gulf coasts support this inference (results not shown), although we also detected genetic breaks corresponding to well-recognized biogeographic discontinuities in North Atlantic coastal biota (Avise, 2000;Blum et al., 2007;Wares, 2002).
Our findings suggest that genetic variation in S. americanus reflects responses to biotic and abiotic conditions within marshes.
Evidence of genetic continuity over time and low gene flow suggests that in situ (i.e., local) conditions likely exert a strong influence on genetic variation within marshes (Orsini et al., 2016). A number of factors are known to influence genetic variation in coastal marsh plants.
Like the observed patterns of spatial variation, shifts in genotypic composition across depth cohorts might reflect responses to local selective pressures. Although it is possible that the observed pattern is a consequence of stochasticity (i.e., genetic drift), relatively modest changes in stressor exposure can structure whole marsh communities (Bertness & Ellison, 1987;Pennings et al., 2005), so by extension, shifts in stressor exposure might also structure genotypic composition within foundational marsh plants over time. Work on Spartina alterniflora supports this inference. For example, evidence has been found that stressor exposure (e.g., to oil, inundation) structures genetic variation across shoreline gradients (Anderson & Treshow, 1980;Gallagher, Somers, Grant, & Seliskar, 1988;Robertson, Schrey, Shayter, Moss, & Richards, 2017), although stressor responses may also reflect plasticity and epigenetic variation (Foust et al., 2016;Proffitt, Travis, & Edwards, 2003;Robertson et al., 2017). We incidentally assessed whether stressor exposure elicits genetic differentiation in S. americanus by drawing comparisons among FACE enclosures across Kirkpatrick Marsh. Prior work has shown that exposure to elevated CO 2 increases S. americanus growth and reproduction (e.g., flowering), enough to shift the balance of competition in mixed communities toward S. americanus dominance (Arp et al., 1993;Langley & Megonigal, 2010;Rasse et al., 2005). Evidence also has been found for genotypic variation in responses of S. americanus to CO 2 exposure (Gentile, 2015), and studies conducted at other FACE sites have shown that experimental exposure to CO 2 can result in rapid adaptive responses in plants (Grossman & Rice, 2014). We did not find evidence, however, that genetic variation is associated with CO 2 exposure across the GCReW site. A more thorough assessment (e.g., SNP-based genomic analyses) might uncover signatures of responses to CO 2 exposure, although it is also possible that responses to stressors that reduce fitness and elevate mortality (e.g., increasing salinity and inundation) might supersede signatures of response to CO 2 .
Addressing some of the methodological limitations that we encountered could help foster further development and use of soilstored seed banks as natural archives. Achieving larger sample sizes, for example, would offer a stronger basis for inferring patterns of genetic variation over time. As reconstituting depth cohorts is a process of diminishing returns, future work could improve upon our efforts by sampling a larger volume of soil (i.e., by taking more and/ or larger sediment cores). This would help overcome limitations set by shifts in abundance over time (Jarrell et al., 2016) and low germination rates, particularly for reconstituting cohorts from deeper (i.e., >16 cm) soil layers. Reconstituting cohorts from finer scale depth intervals could also minimize discontinuities (i.e., time steps) and thus offer a stronger basis for examining dynamic demographic processes like population turnover (Ponnikas, Ollila, & Kvist, 2017).
It may be possible to increase sample sizes by increasing germination rates, although trials so far conducted suggest that methodological modifications may only lead to marginal improvements (Gentile, 2015). Drawing comparisons across sites (i.e., by examining depth cohorts reconstituted from cores taken at multiple locations) would clarify whether the patterns observed in this study reflect general phenomena or conditions idiosyncratic to Kirkpatrick Marsh.
Separately genotyping seed coats and germplasm would also be a key step toward understanding the limits of inferences that can be drawn from plants derived from buried seeds. This would not only clarify whether depth cohorts are representative of the seed bank, it would offer a basis for inferring relatedness and possibly a basis for reconstructing pedigrees (i.e., seed coats are typically maternally derived, whereas germplasm reflect biparental contributions).
Besides demonstrating that soil-stored seed banks can offer perspectives on demographic and genetic change over time, our work illustrates that dormant soil-stored seeds can be a resource for experimental "resurrection" approaches for studying ecological and evolutionary responses of plants to environmental change over time. In many ways, the process of reconstituting depth cohorts from soil-stored seed banks parallels the steps required to assemble experimental cohorts from dormant zooplankton ephippia and curated seed collections (Franks & Weis, 2008;Franks et al., 2007).
Thus, the literature on both can serve as guides for pursuing further work to improve use of soil-stored seed banks as a resource for "resurrection" studies. For example, besides improvement of propagation and germination methods, conducting test crosses to develop pedigreed lines could help augment sample sizes and enable the analysis of trait heritability (e.g., Franks et al., 2007), including traits that contribute to seed survival and germination. And, as has been done with zooplankton hatched from dormant ephippia, elaborating on the genomic and transcriptomic variation in responses to stressor exposure could offer greater insight into the role of drift and selection in shaping temporal patterns of genetic variation (Orsini et al., 2016). Likewise, stronger inferences could be drawn by characterizing longer time horizons (e.g., Frisch et al., 2014) and drawing comparisons to independent records of environmental change. Doing so would not only increase confidence in the use of soil-stored seed banks for the study of coastal marshes, it would also foster further interest in the use of soil-stored seed banks McGraw et al., 1991;Morris et al., 2002;Vavrek et al., 1991) for examining other ecosystems (e.g., tundra, interior wetlands) that are highly vulnerable to climate change and land use intensification.