Microbial Community Succession Along a Chronosequence in Constructed Salt Marsh Soils

In this study, we examined the succession of soil microbial communities across a chronosequence of newly constructed salt marshes constructed primarily of fine-grained dredge material, using 16S rRNA amplicon sequences. Alpha diversity in the subsurface horizons was initially low and increased to reference levels within 3 years of marsh construction, while alpha diversity in the newly accumulating organic matter-rich surface soils was initially high and remained unchanged. Microbial community succession was fastest in the surface horizon (~ 24 years to reference equivalency) and became progressively slower with depth in the subsurface horizons (~ 30–67 years). Random forest linear regression analysis was used to identify important taxa driving the trajectories toward reference conditions. In the parent material, putative sulfate-reducers (Desulfobacterota), methanogens (Crenarchaeota, especially Methanosaeta), and fermenters (Chloroflexi and Clostridia) increased over time, suggesting an enrichment of these metabolisms over time, similar to natural marshes. Concurrently in the surface soils, the relative abundances of putative methane-, methyl-, and sulfide oxidizers, especially among Gammaproteobacteria, increased over time, suggesting the co-development of sulfide and methane removal metabolisms in marsh soils. Finally, we observed that the surface soil communities at one of the marshes did not follow the trajectory of the others, exhibiting a greater relative abundance of anaerobic taxa. Uniquely in this dataset, this marsh was developing signs of excessive inundation stress in terms of vegetation coverage and soil geochemistry. Therefore, we suggest that soil microbial community structure may be effective bioindicators of salt marsh inundation and are worthy of further targeted investigation.


Introduction
Tidal salt marshes are among the world's most productive ecosystems and play a vital role in the coastal landscape. They provide numerous socioeconomic benefits, including coastal flood and erosion protection [1][2][3], and habitat for wildlife including crustaceans and finfish [4][5][6]. Salt marshes additionally perform valuable biogeochemical functions, such as filtering nutrients from the landscape before they reach coastal waters and sequestering carbon through long-term burial [7][8][9]. Despite their ecological and socioeconomic value, there are several significant threats to tidal salt marsh persistence, most notably associated with historic and current land use changes (i.e., shoreline development, urbanization, and agriculture) and arising from current and projected sea level rise [10][11][12][13]. The fate of salt marshes will be determined by their ability to keep pace with accelerating sea level rise through vertical accretion and/or landward transgression [14][15][16] as well as their ability to persist through major storm events, which may increase in frequency [17][18][19]. By some estimates, 60-90% of the global salt marsh inventory may be outpaced by sea level rise by 2100 [20], although some studies have suggested that coastal marshes have greater resiliency and opportunity for upland transgression if anthropogenic barriers are minimal and sediment delivery is not restricted [21][22][23]. The UN Decade on Ecosystem Restoration has been declared from 2021 to 2030, reflecting a consensus view that there is a worldwide urgency in mitigating the degradation of ecosystems [24], including tidal salt marshes, to preserve many of the vital ecosystem services they provide, which may help to mitigate climate change and potentially offer "blue carbon" storage to offset greenhouse gas emissions [8,25,26].
Tidal salt marsh restoration and creation are important tools to replace lost ecosystem services arising from marsh loss or degradation [27]. Marsh restoration and creation strategies take a variety of forms, depending on the scale, budget, site history, and local environmental conditions. These can include restoring tidal hydrology by removing dikes, dams, and coastal defenses [28][29][30], raising marsh elevation through thin-layer application of dredged sediment [31][32][33], stabilizing shorelines and controlling erosion by constructing "living shorelines" [34][35][36], and creating new marshes using dredge material [37][38][39]. The beneficial use of dredge material to construct and restore marshes is becoming an increasingly important strategy to replace lost marshes as well as manage materials from routine maintenance of shipping channels [37][38][39].
Successfully restored wetlands are more similar to natural wetlands than their degraded, unrestored counterparts, and wetland restoration enables significant recovery of vital ecosystem functions including habitat provisioning, carbon sequestration, and nutrient removal [40][41][42][43]. The success of coastal wetland restoration projects tends to depend heavily on the effective re-establishment of appropriate hydrology, based on the geomorphic setting, and on the removal of critical stressors (e.g., invasive species, grazing, and disease) [27,44]. When these conditions are met, restored or constructed salt marshes may nevertheless differ from their natural counterparts in a variety of ways. For example, salt marshes constructed using dredged material that is initially rich in nitrogen may lead to higher rates of primary production and vertical accretion, lower allocation of biomass to roots relative to shoots, and enhanced development of organic matter accumulation and carbon burial relative to natural marshes [45][46][47][48]. The time required to reach structural and functional equivalency with natural marshes following restoration or construction can span a wide range of timescales, from tens to hundreds of years. These timescales may vary depending on a variety of environmental conditions, some of which are site-specific, such as the degree of degradation, method of restoration (i.e., planting vegetation as plugs versus allowing natural vegetation colonization), land-use history, as well as the degree of interannual variability and/or climatic events such as flooding [42,[49][50][51]. Knowledge regarding the timescales of development is useful for planning and assessing the effectiveness of salt marsh restoration and construction projects.
Microbial community composition and diversity assessed using barcoding tools can serve as highly sensitive bioindicators of ecosystem health and recovery since microbial communities respond rapidly to changing environmental conditions [52][53][54]. Studies have shown that shifts in microbial community structure may be diagnostic of changes in biogeochemical processes and the development of soil functions like respiration rates in estuarine sediments [55], litter decomposition rates in grassland soils [56], and other microbial activities related to carbon, nitrogen, phosphorus cycling in wetland soils [57]. Metrics of microbial diversity, especially evenness, have been associated with the maintenance of soil ecosystem function and stability in microcosms of vegetable fields [58] and grassland soils [59], as well as in salt marsh sediments [60]. Several previous studies have also characterized in situ microbial community composition in natural salt marshes [61][62][63][64][65]. In addition, manipulation experiments of salt marsh soils have tested the response of microbial communities to nutrient enrichment [66][67][68][69][70][71] and sand amendments [72]. However, there are only a few studies that have followed the succession of microbial communities during the development of restored or newly constructed salt marshes, especially at multiple time points. Recent studies in salt marshes have found that soil microbial communities were more similar to reference marshes 2 years [70] and 30 years [73] after restoration.
Here, we examined the trajectories of microbial community structure during soil development in newly constructed salt marshes to identify if patterns of microbial community succession are on a trajectory toward natural reference conditions or an alternative state. We took advantage of the chronosequence of marshes established as part of the Paul S. Sarbanes Ecosystem Restoration Project at Poplar Island, a large-scale island restoration project located in the mesohaline waters of the Chesapeake Bay. The project includes sequentially constructed upland and tidal marsh habitats constructed primarily using fine-grained dredge material sourced from the upper Chesapeake Bay, which allowed us to apply a space-for-time substitution to study constructed marshes of known ages. Several studies within the same restoration project have recently documented changes in surface elevation, growth and persistence of vegetation, and development of porewater geochemistry at various times following vegetation planting [47,48,74]. In this study, we compared 16S rRNA gene amplicon sequences from constructed low marshes that form a chronosequence of 16 years to amplicon sequences obtained from a nearby natural reference marsh. We then used a regression-based random forest machine learning approach to identify important predictor taxa driving changes in the development of soil microbial communities over time.

Study Site
The Paul S. Sarbanes Ecosystem Restoration Project at Poplar Island (hereafter, "Poplar Island") is an ongoing long-term island habitat restoration project located in the Chesapeake Bay, approximately 3.2 km northwest of Tilghman Island in Talbot County, Maryland (Fig. 1). Through a joint effort by the US Army Corps of Engineers (USACE), Maryland Department of Transportation Maryland Port Administration (MDOT MPA), and Maryland Environmental Service (MES), the restoration project has reconstructed the once eroded island to create wetland and upland habitats for wildlife while providing a placement site for fine-grained material from maintenance dredging of the approach channels leading to the Port of Baltimore. At the time of sampling, a total of eight tidal marshes had been developed, with the oldest marsh completed in 2003 (Cell 4D) and the newest marsh completed in 2018 (Cell 5AB).
Tidal marsh cells were sequentially constructed by placing primarily fine-grained, nutrient-rich dredged material into diked containment cells with channels and inlets that allow for tidal exchange. Additionally, locally extracted sandy material was used as the parent material in the oldest cell (Cell 4D) and overlaid on top of finegrained dredge material in specific areas within Cells 3C and 5AB to achieve target elevations [75]. After grading surfaces to meet target elevations, Spartina alterniflora and S. patens plugs were planted in low and high marsh zones, respectively [75]. A nearby natural marsh, Knapps Narrows, located approximately 5.8 km southeast of Poplar Island, was selected to serve as a reference site. Salt marsh soils at Knapps Narrows were predominantly composed of peat in at least the upper 50 cm, with some sand lenses interspersed, suggestive of sand deposition during storm events.

Sample Collection
Marsh cells at Poplar Island were constructed using two different types of parent material. Therefore, both fine-grained dredge material and sandy areas were targeted in our study. Samples were collected from S. alterniflora-dominant, low marsh areas at Poplar Island in July 2019 using a push corer (Table S1). Soil cores were collected from triplicate areas within 6 cells constructed using the fine-grained dredge material (Cells 5AB, 3C, 1B, 1A, and 3D) ranging in age from 1 to 14 years at the time of sampling (Table S1). Soil cores were also collected from duplicate areas within 2 cells which were overlain with sand (Cells 5AB and 3C; 1-and 3-year-old, respectively) and triplicate areas within a cell constructed primarily using the sandy material (Cell 4D; 16-year-old).
Each soil core was sectioned into three layers, identified visually in the field, to broadly capture: (A) the surface horizon, which was composed primarily of newly deposited organic matter from aboveground biomass; (B) the transitional horizon, which was the subsurface transitional layer consisting of the parent material penetrated by rhizomes and roots and some decaying organic matter; and (C) the parent material horizon, which was the deeper subsurface layer composed primarily of the parent material without the penetration of roots and rhizomes. Additionally, two weeks prior to sampling at Poplar Island, soil cores were collected from triplicate low marsh areas within the reference marsh using a Russian peat corer, and each core was sectioned at 10 cm intervals from the surface down to a depth of 40 cm. All soil cores were sectioned, homogenized, and aliquoted in the field immediately after sampling. For molecular analysis, samples were aliquoted into 5 mL cryovials (Nalgene), immediately frozen in a cooler under dry ice, and later the same day stored in a freezer at -80 °C. For soil and porewater analysis, samples were collected in 50 mL centrifugation tubes (Falcon) or Whirl-Pak bags, kept in a cooler with ice packs, and later stored at 4 °C.

Soil and Porewater Collection and Data Analysis
Within 1-2 days of sampling, porewater was extracted from soil samples by centrifugation (3500 rpm for 10 min) and immediately filtered through a 0.2-µm syringe filter (Tar-get2). Dissolved inorganic carbon (DIC) concentrations were measured by gas chromatography (Shimadzu GC-8A), following acidification with 0.5 N sulfuric acid. Ammonium (NH 4 + ) concentrations were measured colorimetrically using a UV-Vis spectrophotometer (ThermoScientific Evolution 60S) following the standard indophenol blue method [76]. Anion (SO 4 − and Cl − ) concentration measurements were prioritized when porewater volumes were limited and measured by ion chromatography (Dionex Integrion HPIC) after diluting porewater 25-fold. The organic matter content of soil samples was estimated as loss on ignition (LOI), by combusting samples at 550 °C for 1 h in pre-weighed ceramic crucibles and re-weighing after cooling in a glass desiccator [77].
Data analysis was conducted in R (v.4.1.2) and all plots were constructed using the package ggplot2 (v.3.3.2; [78]). The relationship between the age of marshes at the time of sampling and mean porewater concentrations were assessed by fitting exponential decay models expressed as y = C 0 e −kt for DIC and ammonium concentrations and as y = C 0 e −kt + C min for sulfate to chloride ratios, using model fitting functions within the packages drc (v.3.0-1; [79]) and stats (v.4.1.2).

DNA Extraction, Amplification, and Amplicon Sequencing
DNA was extracted from soil samples (~ 0.2 g) following the modular nucleic acid extraction protocol adapted from Lever et al. [80]. Briefly, the non-soluble DNA (nsDNA) fraction was separated from the soluble DNA (sDNA) fraction, lysed by bead-beating (Vortex Genie), and repeated freeze-thaw cycling, and purified using chloroform. Due to PCR inhibition, it was sometimes necessary to perform further purification with the CleanAll RNA/DNA Clean-up and Concentration Kit (Norgen Biotek), or serial dilution, for downstream applications.
Following the protocol outlined by the Earth Microbiome Project, DNA samples were amplified in triplicate using the modified 515F and 926R primer pair [81] with partial Illumina adaptors to target the V4-V5 region of the 16S rRNA gene and sequenced on an Illumina MiSeq platform at GENEWIZ (South Plainfield, NJ, USA). Extended methodological details are provided in the Supplemental Material.
A minimally filtered dataset was initially constructed by removing ASVs classified as eukaryotes, mitochondria, chloroplasts, or unassigned at the phylum level. Subsequently, a more stringent filtering procedure was implemented to remove ASVs occurring fewer than 10 times in 5% of the samples using the package genefilter (v.1.70; [88]).

Alpha Diversity
Faith's Phylogenetic Diversity (PD), which incorporates phylogenetic information, and the Shannon diversity index were calculated using the minimally filtered ASV dataset. To assess differences in alpha diversity indices across marsh cells, stratified by parent material type, a one-way analysis of variance (ANOVA) followed by Tukey's honest significant difference (HSD) tests or a Kruskal-Wallis test followed by Wilcoxon rank sum exact tests were performed. To determine which test would be used, Shapiro-Wilk and Levene tests were first performed to evaluate assumptions of normality and homogeneity, respectively. P values were adjusted for multiple significance testing with a Benjamini-Hochberg false discovery rate correction. To characterize how diversity changed across the chronosequence, alpha diversity indices plotted against marsh age were fit using rectangular hyperbolic curves in the form of y = (V max *X)/ (k + X), where V max is an upper limit estimate and k is a rate constant. Fit parameters were estimated using the package nlme (v.3.1-152; [89]).

Beta Diversity
Bray-Curtis dissimilarity distances were calculated based on the relative abundances of ASVs after stringent filtering and visually assessed using non-metric multidimensional scaling (NMDS). The homogeneity of dispersion and permutational multivariate analysis of variance (PER-MANOVA; 9999 permutations) were performed to test for differences among marsh age, parent material type, and depth horizon using the package vegan (v.2.5-7; [90]). If significant differences were found, post hoc pairwise comparisons were performed using the package pairwiseAdonis (v.0.0.1; [91]).
Temporal changes in microbial community dissimilarity within each depth horizon were calculated as the difference between the reference marsh (used as the baseline) and each marsh sample from Poplar Island. Community similarities were computed as the difference between the Bray-Curtis dissimilarity value and unity (1-[dissimilarity of the Bray-Curtis distance metric]). To estimate the rate of community change, linear regression models were fit to similarities plotted against marsh age for each depth horizon using the package ggpmisc (v.0.3.7; [92]). To identify which environmental variables were significantly associated with microbial community composition across the chronosequence, canonical correspondence analysis was computed using the envfit function (package vegan) with 999 permutations.

Random Forest Analysis
Random Forest (RF) is a machine learning algorithm that uses bootstrapping to build a set of decision trees from randomly sampled predictor variables to estimate their importance [93]. The RF algorithm can run both classification and regression models depending on whether response variables are factors or continuous variables, respectively. To identify important microbial taxa driving successional trajectories observed across the chronosequence, RF regression models were constructed using the randomForest package (v.4.6-14; [94]). ASV counts were agglomerated at the lowest taxonomic resolution down to the genus level and the relative abundances of these taxa were used as the predictor variables, while marsh age was used as a continuous response variable. Based on the PERMANOVA analysis, we found no detectable differences in the microbial communities between parent material types at Poplar Island, and therefore this variable was not included in the RF models. Additionally, examination of microbial community similarity indicated the surface communities of the 14-year-old marsh (Cell 3D) were likely outliers and were therefore excluded from the RF models.
For each depth horizon, two random forest models (ntrees = 9999) were run using mtry values (optimal number of predictors to be sampled at each node) generated from two different hyperparameter tuning algorithms: the tuneRF function built into the package randomForest, and the random search method using the trainControl function within the package Caret (v6.0-86; [95]). Of these two RF models, the one with the lowest mean squared error was selected. Subsequently, the number of statistically important features to be selected was determined using the package Boruta (maxRuns = 1000, getImp = getImpLegacyRfZ, and ntree = 9999; v.7.0.0, [96]). The performance of RF models for each depth was evaluated with both "leave-oneout" cross-validation method using the trainControl function within the package Caret and a significance test (1000 permutations) using the rf.significance function within the package rfUtilities; (v.2.1-5; [97]). The predictor taxa that were identified as important were visualized using heatmaps of Z-score transformed relative abundances created using the package ComplexHeatmap (v.2.4.3; [98]).

Soil Appearance and Geochemistry
Salt marsh soils were sampled at three depth horizons: surface, transitional, and parent material. The transition between the surface and transitional horizons was visibly distinct in the marshes constructed with fine-grained dredge material, where the accumulating peat at the soil surface increased from less than 1 cm in the 1-year-old marsh (Cell 5AB) to approximately 8 cm in the 14-year-old marsh (Cell 3D). This accumulating peat is primarily comprised of aboveground vegetation biomass accumulating at the surface combined with belowground biomass (roots and rhizomes) [47]. The relatively sharp transition between these horizons may be due to the compaction of the parent material which may create a barrier to soil mixing. Among sandy marsh samples, the transition between the newly accumulating peat and the parent material was less distinct, suggesting greater soil mixing associated with higher root and rhizomes production in the sandy soils.
Porewater DIC concentrations generally decreased with age at all depths, from a maximum of 12.3 ± 2.9 mM in the 3-year-old fine-grained dredge material marsh (Cell 3C) down to 4.8 ± 2.2 mM in the 16-year-old marsh (Cell 4D). An exception was observed in the 14-year-old marsh (Cell 3D), which exhibited elevated DIC concentrations (13.6 ± 4.8 mM), relative to similarly aged marshes (Fig. 2a). The exponential decay of DIC concentrations over time (excluding Cell 3D) was described by the parameters  Figure S1).
Porewater sulfate to chloride ratios (molar) were elevated at the time of vegetation planting (0.05-0.21 mM; [74]) and decreased to levels below the sea water molar ratio of 0.05 within 3 years following vegetation planting (Fig. 2c). The exponential decay of the sulfate to chloride ratios was described by the parameters C 0 = 0.11, k = 0.65, and C min = 0.025.

Alpha Diversity
Alpha diversity indices were significantly different between marsh cells, stratified by parent material type (Faith's PD: ANOVA; F (8, 63) = 5.507, p = 0. 0.00003; Shannon index: KW χ 2 = 25.407, df = 8, p = 0.0013; Figure S3). Faith's PD of the 1-year-old fine-grained dredge material marsh (Cell 5AB) was significantly different from the 7-year-old (Cell 1B), 14-year-old (Cell 3D), and reference marshes (Tukey's HSD). Similarly, Shannon diversity was different between the 1-year-old marshes (Cell 5AB) and all marshes 7 years or older (Wilcoxon rank sum exact test). Both Faith's PD and Shannon diversity indices generally increased across the chronosequence (Fig. 3, Figure S4), and curve parameters describing hyperbolic increases are provided in Table S2. In the surface soil, alpha diversity indices of the 14-yearold marsh (Cell 3D) were excluded from parameter fitting because the communities were identified as likely outliers, based on microbial community similarity (described below). Faith's PD significantly increased with age in the transitional and parent material horizons of fine-grained dredge material marshes and the surface soils of sandy marshes, with the most rapid change occurring in the first 3 years (Fig. 3). Shannon diversity exhibited the same pattern with age and depth horizon ( Figure S4).

Beta Diversity
NMDS ordination based on Bray-Curtis dissimilarity revealed clustering along the primary axis by age and along the secondary axis by depth horizon (Fig. 4). Notably, microbial communities shifted towards the reference marsh over time (Fig. 4). PERMANOVA identified significant differences in community composition associated with marsh age (p = 0.001), parent material type (p = 0.001), and depth horizon (p = 0.001) (Table S3). Pairwise comparisons by parent material type only revealed significant differences between the reference marsh and each of the parent material types at Poplar Island (p = 0.003), but no significant differences between the fine-grained dredge material and sandy substrates (p = 0.063). However, grouped datasets did not meet the assumptions of homoscedasticity and should be interpreted with caution (Table S4).
Microbial community similarity increased monotonically across the chronosequence at different rates for each depth horizon (Fig. 5). Surface soil samples from the 14-year-old marsh (Cell 3D) deviated from this trend and were therefore excluded from the linear regression analysis. Linear regression parameters projected it would take 24, 30, and 67 years for microbial communities at Poplar Island to reach reference equivalence in the surface, transitional, and parent material horizons, respectively (Fig. 5).
Canonical correspondence analysis ( Figure S5) identified the age of marshes at the time of sampling (years; p = 0.001), soil depth (cm; p = 0.005), and LOI (%; p = 0.005) as significant environmental variables associated with microbial community composition across Poplar Island and reference marshes.

Random Forest Analysis
In the surface soil, 27 important predictor taxa were identified, of which 17 increased and 10 decreased with age. The majority of the predictor taxa were affiliated with the classes Alphaproteobacteria and Gammaproteobacteria (Fig. 6). Among Alphaproteobacteria, Defluviicoccus and Candidatus Alysiosphaera increased with age. Within Gammaproteobacteria, the order Methylococcales, which includes the 3 genera, Methylocaldum, Methyloparacoccus, and IheB2-23, and an unidentified member of the family Methylomonadaceae, increased with age. Additionally, members of the genus Thiogranum, 966-1, and an unidentified member of the family B1-7BS within Gammaproteobacteria increased with age. Among Desulfobacterota, the genus-level Sva0081 sediment group increased with age.
In the transitional horizon, 54 important predictor taxa were identified, of which 46 increased and 8 decreased with age. The majority of the predictor taxa increasing with age included Bacteria affiliated with Chloroflexi and Desulfobacterota as well as Archaea affiliated with Asgardarchaeota, Thermoplasmatota, and Crenarchaeota (Fig. 7). Among Desulfobacterota, the four genera, Desulfobacca, SEEP-SRB1, SEEP-SRB2, and unidentified members of the family Dissulfuribacteraceae, increased with age, while unidentified members of the family Desulfuromonadaceae decreased with age. Within Archaea, members of the family Geothermarchaeaceae and unidentified Bathyarchaeia (phylum Crenarchaeota) and members of the genus Marine Group III, order Marine Benthic Group D and DHVEG-1, and unidentified members of the class Thermoplasmata (phylum Thermoplasmatota) increased with age.
In the parent material horizon, 74 taxa were identified as important predictors, of which 59 increased and 15 decreased with age. Predictor taxa that increased with age included Firmicutes, Desulfobacterota, and Chloroflexi, as well as Archaea affiliated with Crenarchaeota and Halobacterota (Fig. 7). Among Desulfobacterota, members of the genus Desulfatiglans, SEEP-SRB2, and unidentified members of the family Dissulfuribacteraceae increased with age, while unidentified members of the family Desulfocapsaceae and genus MSBL7 decreased with age. Within Archaea, unidentified members of the class Bathyarchaeia as well as the genus Methanosaeta within the phylum Halobacterota increased with age.  (a, b) surface, (c, d) transitional, and (e, f) parent material horizons. Panels on the left (a, c, e) are samples collected from fine-grained dredge material marshes, and panels on the right (b, d, f) are samples collected from sandy marshes. Where a time trend was evident, data were fit with rectangular hyperbolic curves, described by two parameters, PD max , a maximum PD value, and k, a rate constant. Curves are displayed if both parameters were significant (p < 0.05). Unfilled blue circles are considered potential outliers and excluded from model fitting (see "Results" section for details)

Ecological and Biogeochemical Trajectories of Newly Constructed Salt Marshes at Poplar Island
A foundational conceptual model describing the successional trajectories of constructed salt marshes was proposed by Craft et al. [49], based on extensive examinations of 8 pairs of reference and constructed salt marshes ranging in age from 1 to 28 years old. According to this model, salt marsh properties directly tied to hydrology including rates of sedimentation and organic carbon and nitrogen accumulation develop "almost instantaneously" with the establishment of appropriate hydrology and salt marsh vegetation. Platform elevations played a key factor in driving this immediate recovery of salt marsh functions, allowing these newly constructed marshes to rapidly gain elevation from sediment accumulation [49,99]. Although marshes described in Craft et al. [49] were constructed to elevations below the equilibrium to allow for inorganic sediment deposition during tidal flooding, more recently due to accelerating sea level rise, design guidelines frequently recommend constructing marshes to elevations slightly above the equilibrium to help increase the stability and productivity of marshes in the future [99][100][101]. Properties associated with biological and biogeochemical processes like primary  [49]. This time frame is theoretically set by the amount of time required to build up minimum inventories (i.e., thresholds) of soil nitrogen to support emergent vegetation (est. ~ 100 g N m −2 ) and soil carbon to support heterotrophic activity (est. ~ 1000 g C m −2 ; [49]). Finally, according to this model, properties associated with the development of mature salt marsh soils, such as depth-integrated organic carbon and nitrogen inventories, require timescales on the order of 3 or more decades to develop [49]. Comparable studies have found that time frames for created and restored marshes to attain carbon and nitrogen inventories equivalent to reference conditions may require 70 to 90 years for carbon and nitrogen pools, respectively, in marshes created on the Pamlico River estuary, North Carolina [102], 124 and 54 years for carbon and nitrogen pools, respectively, in marshes created on the Cape Fear River, North Carolina [103], and 100 years for carbon pools in created marshes within the Sabine National Wildlife Refuge, Louisiana [104] and restored marshes across Eastern England [105,106].
Marshes at Poplar Island share some characteristics with those examined in Craft et al. [49] in terms of geography (i.e., North American mid-Atlantic coast) and dominant vegetation type at low marsh elevations (i.e., S. alterniflora). Yet, the marshes at Poplar Island also differ in several important ways which demonstrably accelerated many of the biological trajectories compared with those examined in Craft et al. [49]. Seven of the eight constructed marshes from Craft et al. [49], and elsewhere [51,102,[107][108][109][110][111], were constructed using coarse-grained (sandy) material, which is initially low in organic nitrogen and carbon. By contrast, the parent material used to construct marshes at Poplar Island was primarily fine-grained dredged material from the upper Chesapeake Bay, which was initially rich in nitrogen and organic carbon (soil nitrogen (2.17 mg g −1 ), carbon (23.5 mg g −1 ), and LOI (6.9%) in the 1-year-old marsh; [112]). The parent material, therefore, exceeds the minimum threshold requirements for marsh vegetation growth from the onset [47], rather than requiring 5-15 years to accrue (sensu [49]). There is also no apparent inhibition associated with other initially high constituents such as sulfate and iron concentration [47,48,74]. Additionally, the rates of marsh accretion among the fine-grained dredge material cells at Poplar Island match or exceed rates observed in reference marshes, by the second year of vegetation growth [47]. Secondly, the marshes at Poplar Island are constructed within containment dikes (with tidal exchange via channels and inlets), which allow more organic matter from aboveground biomass to be retained within the marsh cells compared to open systems [47]. Thirdly, the marshes at Poplar Island range in size from ~ 12-20 ha, while marshes reported by Craft et al. [49] range in size from 0.2 to 1.2 ha. According to a meta-analysis of restored and Temporal changes in microbial community similarity calculated as 1-Bray-Curtis dissimilarity. Pairwise differences between the reference marsh and each marsh sample from Poplar Island were computed within each depth horizon. Black lines represent linear regression models fit to the data and gray shaded regions indicate 95% confidence intervals. Unfilled blue circles are data points that were considered outliers and excluded from the linear regression analysis (see text for further details) created wetlands, the size of wetlands affects the recovery rates of biological structure (i.e., related to the assemblage of vertebrates, macroinvertebrates, and plants) and biogeochemical functions, where larger wetlands recover faster than smaller ones [42]. As a result of these features, aboveground vegetation biomass reached a peak within 2 years after planting [47,48]. Within approximately this time frame, organic matter accumulation, denitrification, and carbon sequestration rates also reached equivalency with reference marshes, despite supporting lower belowground biomass [47,48].
Here, we reported trajectories of several summer porewater constituents across the chronosequence, which all achieved reference conditions, though at variable rates. Dissolved inorganic carbon concentrations declined somewhat more slowly than the other porewater constituents measured, reaching reference conditions within approximately 10 years. There was also one marsh (Cell 3D) that exhibited elevated values that did not match the temporal trend and is further discussed below. The decrease in DIC concentrations could be a result of declining respiration, but we suggest this is unlikely given the accumulation of organic matter content over time. Instead, we suggest Fig. 6 Heatmap of differentially abundant taxa in the surface horizon according to a random forest linear regression model. Data are Z-score transformed relative abundances of surface samples from Poplar Island and reference marshes excluding data from the 14-year-old marsh (Cell 3D), which were identified as outliers. Colored blocks on the right indicate the phylogenetic affiliation of the differentially abundant taxa (at phylum or class levels, as indicated). Abbreviations: SG, sediment group; MSG, marine sediment group; MBG, marine benthic group this trend is likely driven by an increase in hydraulic conductivity over time, which would allow for greater porewater flushing and therefore lowers DIC concentrations across the chronosequence. During salt marsh development, as organic matter increases and bulk density decreases, hydraulic conductivity and porosity typically increase [113][114][115].
As reported by Cornwell et al. [74], porewater NH 4 + concentrations were initially very high (0.22-1.44 mM) but declined to reference conditions in all marshes (0.02-0.05 mM) within 3 years of plant establishment, consistent with NH 4 + uptake by emergent marsh vegetation at depths of < 15 cm during their growing season [48]. This contrasts with porewater NH 4 + concentrations deeper in the parent material below the rooting depth which tend to remain high for longer periods (e.g., 0.36 mM and 0.16 mM at 4 years and 8 years, respectively, at > 20 cm). This prolonged source of reactive nitrogen is likely accessible to the vegetation by upward diffusion [116].
The molar ratio of sulfate to chloride declined to reference levels within 3 years. Initially high concentrations of sulfate likely reflected the oxidation of iron sulfide minerals during the drying/dewatering of the dredged material, and the rapid decline could be accounted for by the diffusion of sulfate to tidal flood water as well as sulfate reduction occurring at anoxic depths in the soils [74,117].

Trajectories of Microbial Community Structure
Characterizing the temporal shifts in microbial diversity and community composition as constructed salt marshes mature Fig. 7 Heatmaps of differentially abundant taxa in the a transitional and b parent material horizons according to a random forest linear regression model. Colored blocks on the right indicate the phylogenetic affiliation of the genera (at phylum or class levels, as indicated). Abbreviations: MBG-D, marine benthic group D may provide insights into successional changes in their biogeochemical function. Based on our space-for-time substitution approach using newly constructed S. alterniflora-dominated salt marshes, we found that alpha diversity (reported as Faith's PD and Shannon diversity) increased rapidly in the transitional and parent material horizons following the planting of marsh vegetation and establishment of tidal inundation, with both indices achieving reference equivalency within the first 3 years. By contrast, in the surface soils of fine-grained dredge material marshes, alpha diversity was initially relatively high and remained unchanged across the chronosequence. These observations likely reflect less specialized microbial communities associated with greater resource availability (i.e., oxygen and labile organic carbon; [118,119]) and negligible dispersal limitation in the surface soils, which can affect microbial communities at depth [120]. We also observed that the microbial community structure was on a trajectory toward reference conditions over time, rather than heading toward a new and/or alternative state. These timescales were 24 years in the surface soils and 30-67 years in the subsurface soils, which are near or within the range reported for the development of carbon and nitrogen soil inventories (i.e., 54-124 years, as described above). We caution that based on the records available (chronosequence of 16 years) we assumed that the rates of change were linear, but longer timescales could reveal curvilinear relationships not captured in our estimates. Nevertheless, there is a general trend of increasing community similarity toward the reference marsh with age (except for the one marsh discussed), which we suggest likely reflects the development of soil organic matter that is becoming increasingly similar to natural reference marshes.
Several of the microbial taxa identified as important in driving changes across the chronosequence have putative involvement in sulfur cycling, suggesting the development of complex soil sulfur cycling over time. Although sulfate decreased rapidly in marsh soils across the chronosequence, concentrations remained sufficiently high to support sulfate reduction. At all depth horizons, there were multiple predictor taxa affiliated with Desulfobacterota, which are likely sulfate reducers [121,122]. In the surface soil, the genuslevel Sva0081 sediment group increased across the chronosequence, which have been identified as fast-growing microaerophilic sulfate reducers that can oxidize both acetate and hydrogen [123,124]. In the transitional horizon, there were several predictor taxa affiliated with Desulfobacterota, of which most increased over time, including members of the genus Desulfatiglans, Desulfobacca, SEEP-SRB1, SEEP-SRB2, and unidentified members of the family Dissulfuribacteraceae. Desulfobacca are described as acetate-degrading sulfate reducers [125], and SEEP-SRB groups are described as sulfate reducers that can form consortia with Archaea to enable anaerobic oxidation of methane [126,127]. In the parent material horizon, increases were observed among sulfate-reducing members of the genus Desulfatiglans, SEEP-SRB2, and unidentified members of the family Dissulfuribacteraceae, while decreases were observed in unidentified members of the family Desulfocapsaceae and genus MSBL7 (family Desulfurivibrionaceae), which have been described as specializing in sulfur-disproportionating metabolisms [128,129]. Several additional predictor taxa affiliated with Chloroflexi and Firmicutes increased across the chronosequence which furthermore suggests an increasing role for fermentation and possibly organic sulfur cycling. Chloroflexi are putatively involved in carbon and sulfur cycles [130,131]. While several predictor taxa among Chloroflexi decreased in the surface soils (Caldilineaceae and JG30-KF-AS9), numerous taxa affiliated to this phylum increased in the subsurface horizons, including unclassified members of the class Anaerolineae and Dehalococcoidia. Recent work has revealed that members of these classes have genetic repertoires indicative of high metabolic plasticity with putative roles in deep sea and benthic sulfur cycling, including genes involved in organosulfur compound degradation and sulfite oxidation [130] as well as possible sulfate reduction [132,133]. Similarly, in the subsurface horizons, members of the class Clostridia (Firmicutes) increased across the chronosequence, which are broadly saprophytic fermenters and have diverse metabolic abilities including potential iron oxide reduction [68,136].
Taken together, the observed community shifts in the subsurface horizons suggest a proliferation of microbes involved in primary degradation pathways including fermentation, as well as increased sulfate reduction, with sufficient niche space for multiple sulfate reducers to concurrently proliferate and potentially support complex sulfur cycling (e.g., sulfurization of organic matter). Increases in the relative importance of fermentation and sulfate reduction could be expected based on the stimulation of anaerobic respiration associated with DOC supplied through roots during the growing season and decaying root biomass following senescence [117,134]. Sulfate reduction in brackish salt marshes may account for 50 to 95% of the total anaerobic respiration (e.g., Jack Bay, reported in [135]). Another significant proportion of respiration in salt marshes may proceed via iron respiration [135], which would be best quantified with metagenomic and metatranscriptomic sequencing to examine functional gene abundance and activity.
Nevertheless, despite microbial community evidence suggesting a possible increase in sulfate reduction, previous reports found free sulfide to be non-detectable in the porewaters of these marshes, with a single exception at Cell 3D, which is further described below [136]. Any sulfide produced is therefore likely rapidly removed via a combination of iron sulfide mineral formation and/or oxidation. We also observed an increase in the predictor taxon Thiogranum (family Ectothiorhodospiraceae) in the surface soils, an obligately chemolithoautotrophic sulfuroxidizing bacteria [137], suggesting that at least some of the sulfide may be recycled by microbial oxidation.
In the transitional and parent material horizons, large monotonic increases were also observed among Archaea. Small increases were observed within the family Geothermarchaeaceae (Crenarchaeota), which are affiliated with Nitrososphaeria, a class of putative ammonia-oxidizing archaea [138], suggesting potential increases in nitrogen cycling in the subsurface horizons. The largest increases across the chronosequence within Archaea were observed among unidentified Bathyarchaea and Crenarchaeota. At both the transitional and parent material horizons, unidentified Bathyarchaea increased from ~ 0.16% in the 1-yearold marsh (Cell 5AB) up to ~ 7% in the 14-year-old (Cell 3D) and 16-year-old (Cell 4D) marshes and ~ 9% in the reference marsh. Although their functional role in salt marsh soils is uncertain, these archaea are potentially acetogenic and/or methanogenic [139]. Increases were also observed in the parent material horizon among Methanosaeta (order Methanosarcinales). Methanosaeta are obligate acetoclastic methanogens and commonly co-occur with sulfate reducers and fermenters [132,140] and/or in syntrophic association with acetate-producing Bathyarchaea [141]. Concurrent with the development of subsurface anaerobic metabolisms, there was evidence for the development of methane oxidizers in surface soils. In the surface horizon, none of the predictor taxa driving the community shifts were affiliated with Archaea, which includes all known methanogens. Instead, there were increases across the chronosequence among aerobic methanotrophic or methylotrophic taxa affiliated with Gammaproteobacteria. Specifically, there were increases in Methylocaldum, a type Ib aerobic methane oxidizer [142,143]; Methyloparacoccus, an obligate utilizer of methane or methanol as sole carbon and energy sources [144]; genus-level IheB2-23 clade and unidentified members of the family Methylomonadaceae, a type Ia methanotroph [145]; and an unidentified member of the family B1-7BS, a probable methylotroph based on the newly described Candidatus Methylophosphatis from this family [146]. In salt marshes, increases in methane production have been reported as a consequence of slowly accumulating inventories of organic matter [27,49,147]. These observations suggest that as methane production is potentially increasing at subsurface depths across the chronosequence, a biological filter comprised of aerobic methane-and methyl oxidizers is concurrently developing in the surface soils.
The use of amplicon sequencing is a valuable first step in assessing the trajectories of microbial community structure over time and allowed us to infer the potential development of some microbially mediated processes. However, to accurately understand how functional genes are shifting over time towards reference conditions, additional tools such as metagenomic sequencing will be necessary. Geochemical measurements to characterize rates of methane and sulfur cycling are also worthy of further investigation.
Although differences in soil texture, initial nutrient content, and root depth distribution were observed between the fine-grained dredge material and sandy marshes, counter to expectation, there were no detectable differences in the microbial community structure between the different parent material types. This was somewhat surprising since it is well-established that soil texture influences the water content, nutrient retention, and oxygen penetration of soils, which in turn all play an important role in structuring microbial communities [43,148,149]. The lack of detectable differences may be due to the fact that the sandy areas were relatively limited in distribution. In Cell 3C, the sandy parent material was applied as a lens over the fine-grained dredge material, and in Cell 5AB the sandy areas were very limited in size. Within these areas of Cells 3C and 5AB, the sandy parent material receives upwardly and possibly laterally diffusing ammonium from the underlying fine-grained dredge material [74], which may be minimizing the differences in the microbial communities between the two parent material types. Additionally, our sample size for the sandy areas was smaller than the fully fine-grained dredge material areas, limiting our statistical power.

Microbes as Bioindicators of Excessive Inundation
Interestingly, the surface soil microbial composition and DIC concentrations of the 14-year-old marsh (Cell 3D) did not follow the same trajectory as the other marshes. Moreover, past surveys found detectable sulfide concentrations in the porewater only at this marsh [136]. These results suggest that marsh conditions were more anaerobic and had less hydraulic conductivity compared to the reference and similarly aged marshes. This was most likely attributed to more prolonged periods of tidal inundation. Indeed, the low marsh within Cell 3D was initially graded to a lower elevation range (0.37-0.55 m above mean lower low water (MLLW)) compared to the cells constructed subsequently (0.48-0.66 m above MLLW). Consequently, this marsh has not been gaining elevation at the same pace as the other marsh cells examined in this study. While limited areas with clumps of healthy vegetation are gaining elevation, bare areas that previously experienced diebacks are losing elevation [150]. These bare areas within Cell 3D are exhibiting signs of inundation stress, including ponding, creek bank erosion, thinning of vegetation, hummock formation, and migration of the low marsh plant, S. alterniflora, into zones built as high marsh platforms [150]. Other marshes exhibiting these traits of marsh degradation and instability tend to also exhibit an accumulation of porewater sulfide which inhibits nitrogen uptake and primary production by emergent vegetation and therefore decline in elevation [21,99,151,152].
We observed that the surface soil samples collected from Cell 3D (14-year-old) exhibited an elevated relative abundance of the Bacterial phyla Desulfobacterota, Chlorofexi, and Latescibacterota, and the Archaeal phyla Asgardarchaeota, Thermoplasmatota, and Crenarchaeota, relative to the other marsh cells. These taxa are known to be predominantly or exclusively anaerobic. The surface soil samples also exhibited a lower relative abundance of Bacteroidota, Planctomycetota, Alphaproteobacteria, and Gammaproteobacteria, which are typically aerobic. Other recent studies in salt marshes have similarly observed decreases in the relative abundance of aerobic taxa within Bacteroidota and increases in the relative abundance of anaerobic sulfatereducing bacteria in unvegetated sediments (as a result of sudden vegetation dieback) with significantly elevated water content when compared to healthy vegetated patches in a S. alterniflora dominant salt marsh [153]. Likewise, Gao et al. [154] tested the impacts of the inundation period on salt marsh soils in mesocosm experiments and found that greater soil water content, due to lower platform elevation, was associated with higher relative abundances of anaerobic taxa and lower relative abundances of aerobic taxa. Our analysis suggests that, with further testing and optimization, monitoring soil microbial communities could be an efficient tool to assess excessive inundation in salt marshes and potentially evaluate the stability of marshes following thin-layer application to supplement marsh elevation [32,33,155].

Conclusions and Future Prospects
Here we found that shifts in microbial community composition and diversity in constructed salt marshes at Poplar Island are on a trajectory towards reference conditions, highlighting the importance of environmental drivers shaping soil microbial communities and providing an independent affirmation that the marshes are developing into ecosystems with high biogeochemical similarity to reference marshes. Recent research has shown that the marshes at Poplar Island are reaching rates of carbon burial that are equivalent to, or even higher than, reference marshes within the first 2 years of vegetation establishment [47]. Likewise, rates of denitrification have been found to reach reference ranges within 2 years [48]. Our results suggest that microbial communities in the surface soils are on pace to reach reference equivalency in approximately two decades, a timescale that resembles the time required for constructed marshes to develop organic matter inventories that converge with reference conditions [49]. The trajectories of microbial community structure suggest that the processes of sulfate reduction, possibly organic sulfur cycling, and methanogenesis are likely becoming increasingly important over time in the subsurface horizons, while sulfide and aerobic methane oxidation are co-developing in the surface soils of Poplar Island marshes. Further research to quantify net fluxes of sulfide and methane and organic matter transformations in these rapidly developing marshes will improve our understanding of whether these constructed marshes built using fine-grained dredged material are similar in terms of carbon sequestration to other constructed marshes. Finally, we suggest that further targeted assessment is warranted to evaluate the prospective application of microbial communities as bioindicators for assessing the inundation period, a metric associated with marsh resiliency to sea level rise. Together with other physical, chemical, and biological indicators, microbial barcodes may provide a more comprehensive assessment of ecosystem health, which could help to better inform future tidal marsh restoration practices.