Materials and Methods

Sampling. Sampling occurred from July 22-24, 2017 and coincided with the spring tides. A new moon occurred on July 23 at 5:45 (EST). While St. John tidal cycles are typically semidiurnal, during spring tide the tide cycle is diurnal. Sampling time points coincided with the low, flood, high, and ebb tides over a 48 hour (hr) window, resulting in 8 total sampling time points. Samples were collected ±1 hr from the designated time point, placed on ice, and processed within two hours of collection. Samples from nine locations across Lameshur Bay and Fish Bay on the south shore of St. John, U.S. Virgin Islands represented coral reef, seagrass bed and mangrove biomes. The Lameshur Bay mangrove location included two distinct sampling sites: an inland area that was only submerged and sampled during high tide and an outlet area that was always submerged and sampled at each time point. The majority of sampling sites were within the boundaries of the Virgin Islands National Park, which is largely undeveloped except for a small research station. The Fish Bay mangrove, Fish Bay seagrass, and Ditliff reef sites are outside the boundary of the park, and the land surrounding Fish Bay is inhabited.

At all sites, a CTD (Castaway, SonTek, San Diego, CA, USA) was deployed from the surface to the bottom depth in reef and seagrass seawater, and single point measurements were collected from mangrove seawater, to capture the temperature and salinity at each timepoint over the course of the 48 hr sampling window. Only temperature and salinity at the surface of the cast was used for analysis. Samples for inorganic nutrients were collected by filling 30ml of surface seawater into acid-washed and seawater-rinsed vials (HDPE, Nalgene), followed by freezing to 20C. Seawater (875µl) was transferred to a 2ml cryovial (Corning) for analysis of microbial abundances and fixed to a final concentration of 1% paraformaldehyde (Electron Microscopy Sciences), allowed to fix for 20 min at 4C, then flash-frozen in an LN2 dry shipper. To capture seawater microbial communities, acid-washed 4L bottles (LDPE, Nalgene, ThermoFisher Scientific, Waltham, MA, USA) were rinsed three times with seawater prior to collection of 3L of surface seawater. The specific 4L bottle used for a site at the first timepoint remained consistent across all sampling timepoints, and the bottles were rinsed with freshwater between uses. Following collection, 1L of seawater was pumped using a Masterflex peristaltic pump (Cole-Palmer, Vernon Hills, IL, USA) through Masterflex silicone tubing (L/S, platinum-cured, #96410-24 size, Cole-Parmer) to rinse the tubing. The remaining 2L of seawater was filtered through a 0.22µm Supor filter (25mm; Pall, Ann Arbor, MI, USA). For the mangrove and seagrass sites, 2L could not always be filtered completely and therefore 0.3 - 2L and 1.2 - 2L of water was filtered through the membrane, respectively. For the coral reef sites, 1.5 - 2L passed through the filter membrane. All filters were placed into 2ml cryovials (Corning, Corning, NY, USA) and flash-frozen in a liquid nitrogen dry shipper until returned to Woods Hole, MA and placed at 80C.

Flow cytometry and Nutrient analyses. Samples for microbial abundance were analyzed at the University of Hawaii with an EPICS Altra flow cytometry (Beckman Coulter Life Sciences, Inc, Indianapolis, IN) as described in Furby et al (Furby et al. 2014), with some modifications. Briefly, to obtain concentrations of cyanobacteria (Prochlorococcus and Synechococcus) and eukaryotic phytoplankton (picoeukaryotes), an unstained aliquot was run on the instrument and excited by visible wavelengths. To enumerate unpigmented cells, which is a proxy for heterotrophic bacteria and archaea (Marie et al. 1997), an aliquot was stained with a Hoechst DNA stain and run on the flow cytometer with excitation at 488nm. The number of cells per ml was estimated following analysis of fluorescence spectra using FlowJo software (v 6.4.7, Tree Star, Inc., Ashland, OR, USA).

Samples for nutrient analysis were analyzed at Oregon State University using methods described in Furby et al. (Furby et al. 2014) to measure dissolved concentrations (µM) of phosphate, ammonium, nitrite, nitrite + nitrate, and silicate.

DNA Extraction, PCR amplification, and Sequencing. DNA was extracted from the filters using a sucrose-EDTA lysis method similar to Santoro et al. (Santoro et al. 2010) that combines lysis with filter column purification. Briefly, the 25mm filter was subjected to physical and chemical lysis using 0.1mm glass beads (Lysing Matrix B, MP Biomedicals, Irvine, CA, USA), sucrose-EDTA lysis buffer (0.75M Sucrose, 20mM EDTA, 400mM NaCl, 50 mM Tris) and 10% sodium dodecyl sulfate (Teknova, Hollister, CA, USA), followed by a proteinase-K digestion (20 mg/ml Promega, Madison, WI, USA). Lysate was then purified using the DNeasy Blood and Tissue Kit (Qiagen, Germantown, MD, USA) spin column filters. Purified DNA was fluorometrically quantified using a high sensitivity dsDNA assay on a Qubit 2.0 fluorometer (ThermoFisher Scientific).

Sample DNA was diluted 1:100 in UV-sterilized PCR-grade H2O and 1µl was used in a PCR reaction. Barcoded primers recommended by the Earth Microbiome Project, 515FY and 806RB, were used to amplify the V4 region of the SSU rRNA gene in bacteria and archaea (Apprill et al. 2015, Parada et al. 2016). Triplicate 25µl reactions contained 1.25 units of GoTaq DNA Polymerase (Promega, Madison, WI, USA), 0.2µM forward and reverse primers, 0.2mM deoxynucleoside triphosphate (dNTP) mix (Promega), 2.5mM MgCl2, 5µl GoTaq 5X colorless flexi buffer (Promega), and nuclease-free water. The reactions were run on a Bio-Rad Thermocycler (Hercules, CA, USA) using the following criteria: denaturation at 95C for 2 min; 28C cycles of 95C for 20 s, 55C for 15 s, and 72C for 5 min; and extension at 72C for 10 min. Successful amplification was verified by running 5µl of product on a 1% agarose-TBE gel stained with SYBR Safe gel stain (Invitrogen, Carlsbad, CA, USA). Triplicate PCR products were pooled and purified using the MinElute PCR purification kit (Qiagen). Concentration of purified products was quantified using the high sensitivity dsDNA assay on the Qubit 2.0 fluorometer (ThermoFisher Scientific). Barcoded PCR products were diluted to equal concentrations and pooled for sequencing. Samples were shipped to the Georgia Genomics and Bioinformatics Core at the University of Georgia for sequencing on an Illumina MiSeq using paired-end 250bp sequencing.

Data analysis. All sequence processing and data analysis was performed in R Studio (v 1.1.463) running R (v 3.4.0, 2017-04-21). Sequence reads were inspected for quality, filtered, trimmed, and dereplicated in the DADA2 R package (v.1.10.0) (Callahan et al. 2016). Specific filtering parameters used included the following: truncLen = c(240, 200), maxN = 0, maxEE = c(2,2), rm.phix = TRUE, and compress = TRUE. DADA2 was then used to generate amplicon sequence variants (ASVs) and remove chimeras. Taxonomy was assigned in DADA2 using the SILVA SSU rRNA database down to the species level where applicable (v.132, (Quast et al. 2012)). ASV counts in each sample were transformed to relative abundance for further data analysis.

To understand the variability in microbial communities over time at all sites, Bray-Curtis dissimilarity was calculated between each sample in the R package vegan (v2.5.4) (Oksanen et al. 2019) and illustrated using non-metric multidimensional scaling (NMDS) in the R package, ggplot2 (v3.2.1) (Wickham 2016). Environmental vectors that significantly associated (cutoff p<0.01) with the ordination were produced using the function envfit() in the vegan R package. Pairwise dissimilarity was plotted to represent the range of dissimilarity in microbial communities over 48 hrs at each site. A higher average dissimilarity would suggest that the site experiences more variable microbial communities than a site with a lower average dissimilarity. A Kruskal-Wallis test was used to examine if there is a significant difference between sites (significance level p<0.05). To determine which pairs of sites had significantly different dissimilarities, a pairwise Wilcoxon Rank Sum test was used with a Benjamini-Hochberg correction for multiple testing and a cutoff of 0.05.

Differential abundance (DA) and of ASVs in relation to the tide was evaluated at mangrove sites using the corncob R package (v0.1.0) (Martin et al. 2019). The following analyses were conducted on a subset of the data representing mangrove communities. All ASV relative abundances were modeled in corncob using a logit-link for mean and dispersion. DA was modeled as a linear function of sea level (a continuous covariate that is representative of the tide cycle) while controlling for differential variance and the effect of site and day or night on DA. Controlling for the effect of day or night was imperative because over the 48 hr period low and flood tide only occurred during the day and high and low tide only occurred during dusk and night, respectively. The parametric Wald test was used to test the hypotheses that the relative abundance or variance of a given ASV changed significantly with respect to sea level and the Benjamini-Hochberg false discovery rate (FDR) correction was applied to account for multiple comparisons, with the cutoff at 0.05.