Biogeography and Edaphic Factors Structure Coastal Sediment Microbial Communities More than Vegetation Removal by Sudden Vegetation Dieback

Development of sudden vegetation dieback (SVD), a phenomenon that causes the rapid mortality of salt marsh plants, specifically Spartina alterniflora, has affected large-scale alterations in Atlantic coastal systems, through the often-complete removal of vegetation. In this study, two wetlands that differ in the time since development of SVD were compared in order to study biogeographic and temporal patterns that structure coastal wetland microbial communities and their response to disturbance. Biogeographic and edaphic factors that distinguished the two wetlands, such as differing salinity, water content, and soil carbon and nitrogen between the sites were more strongly associated with sediment microbial community structure than either sampling date or SVD development. In fact, no OTUs differed in abundance due to the season samples were collected, or vegetation loss due to SVD. This is not to say that SVD did not alter the composition of the microbial communities. The taxonomic composition of sediment communities in SVD-affected sediments was more heterogeneous between samples and a small number of OTUs were enriched in the vegetated sediments. Yet, these data suggest that coastal wetland sediment communities are predominantly shaped by environmental conditions and are generally resilient to temporal cycles or ecosystem disturbances. Importance One of the challenges of microbial ecology is predicting how microbial communities will respond to ecosystem change. Yet, few studies have addressed whether microbial responses to disturbance are consistent over space or time. In this study we employ SVD as a natural vegetation removal experiment and compare the sediment microbial communities between two geographically separated wetlands (ca 125 km). In this manner, we uncover a hierarchical structuring of the microbial communities, being predominantly governed by biogeography, with lesser effects due to disturbance, or temporal dynamics.

microbial responses to disturbance are consistent over space or time. In this study we 24 employ SVD as a natural vegetation removal experiment and compare the sediment 25 microbial communities between two geographically separated wetlands (ca 125 km). In 26 this manner, we uncover a hierarchical structuring of the microbial communities, being 27 predominantly governed by biogeography, with lesser effects due to disturbance, or 28 temporal dynamics. 29

Introduction 30
Coastal ecosystems are among the most productive on earth and have the potential 31 to sequester and store carbon at rates of up to 50 times higher than other terrestrial 32 ecosystems (1). The potency of coastal ecosystems as carbon sinks is attributable to their 33 high primary productivity, as they can produce 40% more plant biomass annually than 34 the same area of forest (2,3). This plant fixed carbon is eventually delivered to the 35 wetland sediments through litter, root exudates, or plant mortality, eventually becoming 36 the substrate for microbial metabolism (4, 5). A defining feature of wetlands is periods of 37 water saturation, with flooded sediments rapidly become anoxic. Anaerobic degradation 38 of organic matter happens relatively slowly, and the rate of organic inputs from the 39 vegetation occurs at a greater rate than losses by microbial respiration, resulting in a net 40 accumulation of carbon in wetland sediments (6, 7). This stored carbon in coastal 41 wetlands has been referred to as "blue carbon" and is an important component to 42 mitigating the atmospheric carbon concentrations that are driving climate change (1, 8-43

10). 44
Coastal wetlands along the eastern coast of North America are experiencing 45 sudden vegetation dieback (SVD), a phenomenon affecting low elevation salt marshes 46 dominated by smooth cordgrass (Spartina alterniflora). SVD presents as an initial 47 (16, 17), and drought (17) have been proposed as the cause, and likely interact or at least 53 contribute to the development of SVD. Regional differences or complex interactions 54 between factors may play a role in the difficulty in identifying a unified explanation of 55 SVD development (18). The lack of an explanatory model for SVD development has 56 driven most research to address the causes of SVD rather than the consequences of the 57 loss of vegetation on the functioning of wetland ecosystems. 58 We previously documented that sediment microbial communities differed 59 between SVD-affected sediments and stands of healthy S. alterniflora at a coastal marsh 60 in Connecticut, U.S.A (19). SVD-affected sediments harbored reduced populations of 61 bacteria in the phylum Bacteroidetes, whereas populations of sulfur-reducing bacteria 62 (predominantly within the genus Desulfobulbus) were enriched in the SVD-affected 63 sediments. Additionally, the SVD-affected patches supported 64% reduced CO 2 64 emissions compared to healthy vegetated controls (19). Taken together, these 65 observations indicate that SVD resulted in alterations in both the structure and function of 66 sediment microbial communities. Yet, there is little known regarding the spatial or 67 temporal scales at which the shifts in the microbial communities occur, or whether 68 alterations in sediment microbial communities are similar between geographically 69 isolated wetlands. 70 We predicted that the microbial communities would differ between the two field sites due 76 to biogeography, but would show similar responses to SVD, such that the sites 77 experiencing SVD would potentially be more similar to each other than they were to 78 healthy locations at the same field site. We further expected that the sediment microbial 79 communities response to SVD would be muted in fall sampling, when plants in the 80 vegetated plots were not active and producing root exudates for the microbial 81 populations. In this manner, alterations in the coastal wetland sediment communities 82 could be linked to spatial biogeography, and to temporal dynamics related to the time 83 from disturbance, and seasonal plant activity. 1A). All sampling was performed at low tide, as this is the only period where SVD 92 sediments are exposed for collection (Fig. 1B). 93 The dominant vegetation at both sites is S. alterniflora and both sites have 94 unvegetated patches due to SVD. At Hammonasset, outbreaks of SVD were first 95 documented in 1999 and many patches have remained unvegetated since. At The two sites differed significantly (P < 0.05) in all sediment chemistry variables 108 except pH (F 1,31 = 3.9, P = 0.056). In general, Narragansett had higher soil electrical 109 conductivity (EC; F 1,31 = 6.3, P= 0.017), soil moisture (F 1,31 = 141.7, P < 0.001), soil %C 110 (F 1,31 = 199.9, P < 0.001) and %N (F 1,31 = 292.3, P < 0.001), but a lower soil C:N than 111 Hammonasset (F 1,24 = 38.0, P < 0.001; Table 1). None of the measured sediment 112 variables differed among vegetation zones nor season (P > 0.05). However, we observed 113 an interaction between site and season of sampling in C:N ratios (F 1,24 = 8.5, P= 0.007), 114 where the Narragansett sediments had greater C:N during the summer sampling than the 115 Hammonasset wetland, but had greater C:N when sampled in the fall. 116 suggesting that the sediment microbial communities were significantly different between 122 the two wetlands. Vegetation status was also associated with a significant difference in 123 clustering of the datasets (PERMANOVA P=0.028). Furthermore, the interaction 124 between the date of sampling and vegetation status was not significant (P=0.995), 125 suggesting that the date of sampling did not influence the microbial community 126 composition associated with the different vegetation conditions. Date of sampling was 127 not significant factor in sample clustering (P = 0.20). Taken together, these data suggest 128 that coastal sediment microbial community composition is primarily structured by the 129 edaphic factors associated with biogeography, followed by vegetation removal by SVD, 130 with a very small contribution of sampling date. 131 132

Microbial diversity 133
To measure alpha diversity, the datasets were rarified to the same number of 134 sequences (7,689) and three diversity metrics were calculated, the number of observed 135 OTUs, Shannon's diversity index, and inverse Simpson's index (Fig. 3). The average 136 number of OTUs recovered from the Hammonasset samples was 4,594 (±521) and 4,636 137 (±378) for Narragansett. The Shannon's diversity index for both Hammonasset and 138 Narragansett samples was 7.95, and the inverse Simpson's index was 983 for 139 Hammonasset and 895 for Narragansett (Fig. 3). Furthermore, there was no apparent 140 diversity pattern between samples collected during different seasons or from different 141 vegetation zones. Together, these data indicate that sediment microbial diversity was not 142 affected by biogeography, date of sampling, or vegetation status.

Taxonomic composition of datasets 145
Sequence reads were classified to the phylum level to compare the composition of 146 the bacterial communities between the sites (Fig. 4). In general, phyla were present at the 147 two sites in similar proportions. For example, at both sites the two dominant identified 148 phyla were Proteobacteria and Bacteroidetes (Fig. 4). Both sites also harbored a relatively 149 large proportion unclassified bacterial sequences, suggesting a large fraction of 150 uncharacterized bacterial diversity in the sequence datasets. 151 A common observation across the datasets was that samples with phyla that 152 showed large deviation from the mean were predominantly from the SVD conditions. 153 Yet, these shifts were not consistent across replicate samples or with sampling date (Fig.  154 4). In this regard, no phylum level taxonomic bins were found to be significantly different 155 in relative abundance when tested for either date of sampling or vegetation status. Thus, 156 these data suggest a part of the sediment microbial community response to SVD is to 157 increase the taxonomic heterogeneity between samples, rather than a consistent shift of 158 specific taxonomic ranks. 159 160

Differentially abundant OTUs due to site 161
A total of 23 OTU's (97% sequence identity) were identified as significantly 162 different in relative abundance due to site, 9 significantly enriched at Hammonasset and 163 14 significantly enriched at Narragansett (Fig. 5). The differentially abundant OTUs 164 belonged to five phyla and could be further classified to 12 taxonomic ranks representing 165 the deepest level to which the OTUs could be reliably assigned (Table S1). There was no 166 obvious pattern in the taxonomy of the differentially abundant OTUs. In fact, several OTUs were identified to taxa that were significantly more abundant in both of the field 168 sites. For instance, two OTUs identified as significantly more abundant at Hammonasset 169 were classified to the genus Calothrix along with one of the OTUs that was enriched at 170 Narragansett (Table S1). In this respect, these data suggest that at least a portion of the 171 differentially abundant OTUs due to site may represent functionally redundant species 172 adapted to the local edaphic factors. 173 174 Differentially abundant OTUs due to sampling date 175 Samples were collected in July and October to test for temporal dynamics in the 176 sediment communities. Overall, the majority of OTUs did not show a large change in 177 relative abundance, rarely surpassing a two-fold difference between sampling dates (Fig.  178   6). None of the OTUs were identified as significantly different in abundance. Thus, these 179 data suggest that sampling date was a small factor in driving sediment community 180 structure. This further matches the ordination results in which sampling date was not a 181 significant factor in sample clustering (Fig. 2). 182 183 Differentially abundant OTUs due to vegetation status 184 OTU abundance among the samples differing in vegetation status was 185 investigated (Fig. 7). Large portions of the most abundant OTUs in the datasets were 186 present in roughly equal abundance between all three vegetation conditions (inner 187 triangle Fig. 7). Additionally, no OTUs were identified as significantly different due to 188 vegetation status at either site (ellipses Figure 7). Yet, there was a clear trend of certain OTUs showing enrichment in the SVD sediments. Thus, these data suggest that there are 191 certain OTUs that trend toward being more abundant in the vegetated samples even if 192 they did not rise to the level of significance. When taxonomy was mapped onto the 193 OTUs, there was no readily apparent pattern in the OTUs that were more abundant in the 194 vegetated sites as they were represented by multiple phyla (Fig. 7). The results of this study demonstrate edaphic factors related to geography were a 198 larger diver of sediment community composition than the date of sampling or vegetation 199 removal by SVD. Sediments from Narragansett had higher soil moisture, greater 200 electrical conductivity, and higher C and N content, indicating that tidal waters may have 201 more frequently inundated the Narragansett sites (Table 1). While sediments were 202 collected systematically along perpendicular transects from tidal creeks at both sites, it is 203 possible that Narragansett SVD patches occurred lower in the tidal frame and thus were 204 wetter, saltier, and enriched in organic matter, key factors driving microbial community 205

composition. 206
Previous studies have similarly found that geography is a large driver of bacterial 207 community composition. Regional differences in Louisiana salt marshes were at least as 208 large of a predictor of bacterial community composition as those between the rhizosphere 209 of different plants (S. alterniflora and Juncus roemerianus;(20)). Similarly, ammonia-210 oxidizing communities (bacteria and archaea) showed larger differences between regions 211 associated with soil moisture and nitrogen content than due to contamination during the sediment microbial communities, the structure of the sediment microbial communities 214 was similar between the two field sites, being composed of the same dominant taxa (Fig.  215   4), harboring similar levels of microbial diversity (Fig. 3), and only a relatively small 216 number of OTUs being identified as significantly different in relative abundance between 217 the sites (Fig. 5). A subset of the differentially abundant OTUs belonged to taxonomic 218 groups specifically enriched in a certain field site. For example, an OTU identified to the 219 genus Mariprofundus was enriched at Hammonasset (Table S1). These organisms have 220 been associated with a role in iron oxidation in marine systems (22), and may point to 221 differences in iron cycling between the sites. In contrast, several OTUs belonged to 222 taxonomic ranks identified as more abundant in both field sites (Table S2) may regulate biogeochemical processes such as iron cycling (26). Thus, these data 234 support that coastal sediment microbial activity is largely driven by biotic and abiotic 235 factors that vary at a variety of time scales. However, studies characterizing temporal patterns in microbial community assembly are notably sparse (27)(28)(29). We collected 237 samples in July and October to investigate if the sediment microbial communities showed 238 significant changes in composition related to season. Date of sampling was insignificant 239 as a factor clustering the sequence datasets (Fig. 2) and the OTUs in the datasets were 240 present in similar abundances at both time points with no OTUs being identified as 241 significantly different in abundance due to sample date (Fig. 6). Taken together, these 242 data suggest that the sediment microbial communities were largely similar in summer and 243 fall samples, suggesting a limited role for seasonal dynamics in shaping the sediment 244 communities. The samples were limited to a two-point time course therefore a finer-245 grained analysis may be required to disentangle a more nuanced response of these 246 communities to temporal cycles. Yet these observations suggest that while microbial 247 activity is responsive to the temporal dynamics in coastal wetlands, alterations in activity 248 may be a poor predictor of community composition. 249 Finally, we employed the development of SVD as a natural experiment to assess 250 the impact of an ecosystem disturbance on the sediment communities. At a landscape 251 scale, the complete loss of vegetation appears to be a dramatic disturbance that would 252 presumably translate into similarly large shifts in the sediment microbial communities. 253 We previously showed that sites at Hammonasset experiencing SVD supported 254 significantly lower populations of bacteria within the phylum Bacteroidetes and an 255 elevated relative abundance of sulfate reducing bacteria (19). Yet, in this study no OTUs 256 were identified as significantly different in relative abundance. The lack of significant 257 differences could be due to the relatively low replication per individual sample date and 258 site, the depth of sequencing, or the added variability of identifying differences between samples collected on different dates. It is important to note, that while this study did not 260 identify any OTUs significantly altered in abundance due to SVD, that does not indicate 261 that there was no effect on the sediment communities. For example, CAP analysis 262 identified a significant difference in the clustering of samples under different vegetation 263 statuses (Fig. 2), the taxonomic makeup of the sediment communities at the phylum level 264 was more heterogeneous in the SVD samples (Fig. 4), and there was a clear trend in 265 OTUs that were enriched in the vegetated sites (edge and healthy) compared to SVD sites 266 (Fig. 7). These data indicate that to the extent that there are shifts in the microbial 267 communities due to SVD, they are likely limited to relatively rare community members 268 and do not involve large shifts in relative abundance. The practical relevance of these 269 observations is that the sediment microbial communities are also likely to respond well to were analyzed for several sediment parameters. Soil pH and electrical conductivity were 317 estimated on 10g subsamples diluted with 50 mL of deionized water and quantified using 318 Orion Star A215 pH Conductivity Meter Orion with Ross Ultra pH/ATC Triode 319 (8157BNUMD) and Orion Conductivity Cell (013005MD) probes. We dried subsamples 320 at 105°C for >24hours and then weighed to estimate soil moisture. Subsamples were also 321 pulverized in a ball-mill, rolled in tins, and analyzed for %C and %N (Costech ECS 4010 322 CN Analyzer). Every ten samples we ran analytical triplicates to examine sample 323 heterogeneity and observed <20% standard deviation for all soil parameters. 324 325 DNA extraction, 16S rRNA gene amplification, and sequencing 326 Samples were processed as described previously (19, 34). Briefly, total 327 environmental DNA was extracted from sediments using the DNeasy PowerSoil Kit    Inter-sample distances were calculated with the Bray-CuBs metric using rarefied OTU count data. The percent variance explained by each of the CAP axes is indicated. Explanatory variables are indicated by arrows (data presented in Table 1).

Healthy
SVD C:N pH EC moisture N C Figure. 3. Alpha diversity of sequence datasets, separated by field site (Ham=Hammonasset, Nar=Narraganse>). Three diversity indices were calculated using OTU abundance in the rarefied datasets. The number of observed OTUs (Observed), Shannon's diversity index (Shannon), and the Inverse Simpson's Index (InvSimpson). Each point represents the value from a single dataset.