Temporal Dynamics of South End Tidal Creek (Sapelo Island, Georgia) Bacterial Communities

Site description and sample collection.South End Creek, Sapelo Island, GA (31°40′N, 81°28′W), has a semidiurnal tidal cycle, with a tidal range of 4 to 6 m, and empties into Doboy Sound on the western shore of Sapelo Island, approximately 1 mile north of the southern extent of the island. The creek mouth faces southwest toward mainland Georgia. The study site was not significantly influenced by freshwater inputs.

A 74-h, five-tidal-cycle survey of physical, chemical, and biological parameters was conducted in South End Creek. A total of 37 observations were made. Hourly observations were made for 25 h on 20 and 21 October 2007; nine observations were made on 21 and 22 October approximately every 3 h at high tide, low tide, maximum ebb, and maximum flood; and three observations were made on 23 October at high and low tides. The sampling schedule was based on tide charts for Old Tower, Sapelo Island (31°23′N, 81°17′W), the nearest location for which tidal data were available (Nobeltec Corporation).

Bacterial communities were captured on 0.2-μm Supor membrane filters and frozen until further processing. For chlorophyll a, creek water was filtered through Pall A/E glass fiber filters and immediately stored in opaque canisters and frozen. Samples for bacterial cell counts were preserved with 3% formaldehyde (wt/vol). Stage height, temperature, pH, dissolved oxygen (DO) concentration, percent saturation of DO, salinity, conductivity, specific conductance, and turbidity were measured in the field at the time of whole-water collection. Stage height was measured with a weighted tape from three established reference points at one-fourth, one-half, and three-fourth low-tide creek width. Probe measurements and water samples were taken at 60% of the total depth from the surface at the center of the creek (one-half low-tide width). Temperature, DO concentration, salinity, conductivity, and specific conductance were measured using a YSI-85 sonde; pH was measured using a YSI-63 sonde (YSI Inc., Yellow Springs, OH) and turbidity by using a DRT-15CE turbidimeter (HF Scientific, Inc., Fort Myers, FL).

Chemical and biological analyses.Total chlorophyll a was measured by fluorometry as described by Carpenter et al. (3). Real-time photosynthetically active radiation data were obtained from the Georgia Coastal Ecosystem Long-Term Ecological Research (GCE-LTER) network data set, courtesy of D. Hurley (available at http://gce-lter.marsci.uga.edu/portal/sinerr/2007/data/ ).

Bacterial cells were separated from aggregates by using sodium pyrophosphate and mechanical disaggregation. Bacterial cells were enumerated by flow cytometry and calibrated by epifluorescence microscopic counts. We stained samples with SYBR green 1 (Molecular Probes, Eugene, OR), collected cells on 0.2-μm polycarbonate black filters (GE Water & Process Technologies), and enumerated 10 fields and at least 300 cells per slide (Axioplan-2 microscope; Carl Zeiss MicroImaging, Inc., Thornwood, NY). Flow cytometry was performed using a BD FACSCalibur flow cytometer and the associated software (BD Biosciences, San Jose, CA).

Sample storage time and conditions are known to reduce bacterial cell numbers, but losses are consistent across the samples such that back modeling is possible to determine the original cell abundance (20). In this study, analysis is focused on relative cell abundances between samples, not absolute cell counts. It is assumed that cell loss was consistent across samples.

Regression analyses.Patterns among environmental parameters were described using both linear and nonlinear regression. A simple damped sine function (y ∼ sin(x)/x) was used for harmonic regression analyses of the environmental parameters with stage height (as a proxy for tide) and was found to agree with overall linear analyses. One turbidity observation was removed after outlier detection (normal Q:Q and residual plots) in the R environment for statistical computing (http://cran.r-project.org/ ). Because the frequency of observation changed with time in the study, regression analyses of both the 0- to 24-h sampling period (hourly observations) and the 0- to 74-h sampling period (all observations, including 1-, 3-, and 6-h frequencies) were conducted.

Bacterial community fingerprinting and statistical analyses.Total DNA from frozen filters was extracted using a Bio 101 FastDNA kit (QBiogene, Carlsbad, CA), per manufacturers' instructions with minor modifications (23, 24). The universal 6-carboxyfluorescein-labeled primer 1406f (5′-TGYACACACCGCCCGT-3′) and bacterial-specific primer 23Sr (5′-GGGTTBCCCCATTCRG-3′) were used to selectively amplify the intergenic spacer region (ITS) between the 16S and 23S bacterial ribosomal genes (7, 23). PCR was conducted on a Mastercycler gradient thermocycler (Eppendorf, New York, NY), with 5 to 10 ng of extracted DNA as a template. PCR conditions were as follows: 2 min denaturation at 94°C, 30 cycles of 35 s denaturation at 94°C, 45 s annealing at 55°C, 2 min elongation at 72°, and a final extension for 2 min at 72°C. Automated ribosomal intergenic spacer analysis (ARISA) was used to fingerprint the bacterial community, with published modifications (23, 24). Amplified community DNA fragments were separated by denaturing capillary electrophoresis on an ABI 3730xl DNA fragment analyzer (Applied Biosystems, Foster City, CA). Fingerprint profiles were analyzed and aligned using GeneMarker 1.4 (SoftGenetics, State College, PA). ITS fragment sizes were analyzed using an internal ROX (6-carboxyl-X-rhodamine) standard of 100 to 1,250 bp, with standard fragments for every 100-bp length within the range. ARISA fragments were manually quality controlled and binned into operational taxonomic units (OTUs) based on profile alignments. In each profile, the normalized peak height in each fingerprint was used to represent that OTU's proportion in the community (15, 24). The signal-to-noise ratio was determined using a moving window of 3 standard deviations around the mean peak height, similar to the methods of Abdo et al. (1). Wider “bins” were assigned to longer fragments because of increased variation in capillary electrophoresis time with increasing ITS size. Generally, fragments of less than 500 bp were grouped into 1-bp bins, fragments of between 500 and 700 bp bins were 3 bp, and fragments over 700 bp were grouped into 5- to 8-bp bins.

Correspondence analysis (CA) was used to search for patterns among tidal-stage groups. CANOCO for Windows software package 4.1.5 (19) was used for these analyses. Constrained correspondence analysis (CCA) was used to explore how well the measured environmental variables explained community variation, and partial correspondence analysis (pCA) was used to explore patterns in the communities after accounting for the variation explained by tidal stage.

CANOCO CA returns a CorE value describing the correlation of the environmental variables with the axes. The following environmental variables were included in the CA: stage height, water temperature, pH, DO concentration, DO percentage, salinity, conductivity, specific conductance, turbidity, chlorophyll a concentration, photosynthetically active radiation, cumulative time, and hour of the day. Because most of the environmental variables did not correlate well with the axes, only those variables that were correlated with axis 1 or 2 by >0.25 were shown.

Analysis of similarity (ANOSIM) was used to first test for significant differences between community groups (tidal or diel groups) (4). ANOSIM is a permutated analogue of analysis of variance that generates an R test statistic to describe the degree of distinction between groups, as well as a P value for significance. The R statistic ranges from 0 (complete overlap in composition; groups are essentially the same) to 1 (no overlap in composition; groups are completely distinct); negative R statistics suggest greater across-group similarity than within-group similarity. Bray-Curtis similarity was used for the ANOSIM, and cumulative pairwise Bray-Curtis similarity was calculated to represent both time and community resemblance from the first community observation (i.e., Bray-Curtis similarities between observations 1 and 2, 1 and 3, 1 and 4, etc.).

Categorical tidal groups were assigned conservatively. Communities observed closest in time to high or low tide were assigned as such, and all communities observed in between were grouped as ebb or flood according to the direction of the water movement. For example, observations from 1 hour past high tide to 1 hour before low tide were classified as ebb. The 34 total community observations were grouped into tidal classification as follows: 6 high tide, 4 low tide, 11 flood, and 13 ebb communities. Also, categorical “diel” groups were defined by whether the observation was collected between sunrise and sunset (day [15 total observations]) or between sunset and sunrise (night [16 total observations]). To be conservative, observations closest to sunset and sunrise were omitted from the analysis.

Physical and chemical characteristics of South End tidal creek.As expected, harmonic patterns were observed in the environmental variables through time (Fig. 1). In order to characterize these environmental patterns, linear and harmonic regressions of stage height (as a proxy for tide) with the environmental parameters were performed (Table 1). Generally, linear and harmonic regressions agreed, though higher r² values were observed with harmonic regressions. Strong relationships between stage height and several environmental parameters (DO concentration, percent oxygen saturation, and pH) were detected. Weaker, but significant, relationships were found between tide and conductivity, salinity, and turbidity.

• Open in new tab
• Download powerpoint

FIG. 1.

Chemical, physical, and biological environmental time series of South End tidal creek, Sapelo Island, GA, on 20 to 24 October 2007. Shaded area represents night. NTU, nephelometric turbidity units.

Because of differences in observation frequency, we analyzed two sampling intervals, 0 to 24 and 0 to 72 h. The relationships between stage height and most environmental parameters between hours 0 and 24 were stronger than those during hours 0 to 72, likely because of the more-resolved sampling frequency.

Significant linear correlations between three environmental parameters and cumulative time were observed. Chlorophyll a concentration, salinity, and conductivity all increased with time (adjusted r² values were 0.18, 0.48, and 0.26, respectively; all P values were <0.006). These trends represent a uniform change over the 3-day sampling period. The range of chlorophyll a concentration increased with time (∼0 to 10 μg/liter to 0 to 17 μg/liter), while the ranges of salinity and conductivity remained constant (Fig. 1).

Variability of regression strength and significance may be attributed to the uneven spacing of temporal observations, the progression of the tidal cycle with time, or environmental variables not accounted for by this study (e.g., atmospheric pressure, currents). For example, spring tide occurred 4 days after our study ended, and the lunar phase (waning crescent) caused an increase in tidal range as this spring tide approached. The proximity of our observation period to spring tide likely influenced the results, especially in those environmental parameters significantly related to cumulative time.

Environmental drivers of tidal creek bacterial communities.By ANOSIM, tidal creek bacterial communities were separated by qualitative tidal cycle categories (Table 2). Communities observed during high and low tides were most distinct, while ebb and flood communities were not significantly different. Communities also differed in richness and overlap of common OTUs across tidal-cycle categories (Fig. 2). There was no significant distinction in communities by diel (day/night) groupings (R = −0.019; P < 0.61).

• Open in new tab
• Download powerpoint

FIG. 2.

Representation of total observed OTUs common across South End tidal creek (Sapelo Island, GA) bacterial communities. Average Bray-Curtis similarity per group is given in parentheses. (A) Ebb and flood community overlap. (B) Ebb and flood, high-tide, and low-tide community OTU overlap. Here, ebb and flood communities are combined, because they are not statistically distinct (see Table 2).

CA and CCA were conducted to explore the relationship between whole-water communities and the measured physical and chemical parameters (Fig. 3). Though all measured environmental parameters were included in the analysis, only eight of these had CANOCO CorE values greater than 0.25. These are included in Fig. 3, and their correlations are listed in Table 3. DO concentration was most strongly correlated to community variation, while tide (measured by stage height) was the second most correlated.

• Open in new tab
• Download powerpoint

FIG. 3.

(A) CA of whole-water bacterial communities from South End tidal creek, Sapelo Island, GA, including environmental variables with correlation of >0.25 to axis 1 or 2. DO concentration was most correlated with axis 1, and colinear environmental variables tide (measured by stage height) and conductivity were the next most strongly correlated. (B) CCA suggests that the environmental parameters measured were most important in explaining community variation. (C) pCA with tidal-stage height as a covariable. spec., species; environ, environment; Chl a, chlorophyll a.

Because there was a difference in communities by categorical tidal-cycle groupings (high tide, low tide, ebb, flood), remaining variation in bacterial community composition after accounting for tide was explored. A pCA on whole-water communities was conducted, using stage height as a covariable (Fig. 3C). After accounting for the variation explained by stage height, DO concentration and conductivity remained correlated to community composition, suggesting that despite their colinearity with tide, these variables are separately important in structuring the bacterial community. Though all other measured variables were less correlated in the pCA, cumulative time became slightly more correlated to the second axis (Table 3). The influence of cumulative time on the bacterial community dynamics is further represented in Fig. 4. Increasing variation is observed with time from the first observation, possibly because of the decreased temporal resolution in hours >24 in the study. Also interesting is the harmonic pattern of community similarity, as both low-tide communities and high-tide communities are more similar to like communities; this within-group resemblance decreases progressively with time, following the general trend of the entire community.

• Open in new tab
• Download powerpoint

FIG. 4.

Cumulative pairwise Bray-Curtis similarity of each bacterial community with the initial time 1 observation (1030 h, 20 Oct. 2007). For example, the first point is the Bray-Curtis similarity between observations 1 and 2, the second is between observations 1 and 3, the third is between observations 1 and 4, etc. High- and low-tide observations are marked with solid and dashed vertical lines, respectively. Community variation increases and cumulative similarity decreases with time from the initial observation.

A theoretical framework to describe microbial community assembly is based on the concept that there exist multiple, hierarchical drivers that structure communities (e.g., see reference 11); this concept is borrowed from macro-scale ecology (22). The results suggest that in the tidal creek system and within the hierarchical framework, tidal stage is the strongest driver of bacterial community dynamics. This is supported by the strong correlation of bacterial community variation to stage height, which was used as a proxy for tidal stage (Fig. 3), and also by ANOSIM (Table 2). There may be no drivers stronger than tide, as the CCA results (which constrain the ordination to be explained by the measured environmental parameters, including tide) (Fig. 3B) agree well with the results of the unconstrained CA. However, after accounting for the compositional variation explained by tide using pCA, DO and conductivity were consistently correlated to community variation despite their colinearity with tide (Fig. 3 and Table 3). If the observed relationships between DO and conductivity and bacterial community composition were due to only their covariation with tide, their influence would be negligible in the pCA. Rather, both remain reasonably correlated to the first axis.

It is not surprising that the results suggest that DO acts as a driver of bacterial community dynamics. The fluctuations of DO concentration in salt marshes have been extensively modeled and studied (e.g., see reference 2). These DO fluctuations have also been recognized as being primarily driven by microbial processes; microbes are responsible for both the depletion of oxygen in the marsh and the photosynthetic production of oxygen in the open ocean. Other studies have shown DO to be important for microbial community dynamics in other systems (12, 16), though these specific drivers have not been measured directly with community compositions in tidal creek systems. Shiah and Ducklow have shown temperature to be the most important driver of bacterial production, growth, and respiration on multiple temporal scales, including diurnal observations, in tidal creeks (18). This study did not measure DO or consider community composition and was conducted in a north temperate salt marsh that experiences a wider range of seasonal temperature fluctuations than does Sapelo Island, GA. Despite these differences, it is reasonable that temperature is a stronger driver of bacterial community dynamics seasonally but that on a diurnal scale, DO may be equally or more important. Given the results of this study, future studies of tidal-stage-influenced systems should consider DO concentration as a standard environmental parameter to be measured.

There is a potential influence of cumulative time on tidal creek bacterial community dynamics. Because of the increased explanatory importance of cumulative time in the pCA (Fig. 3C and Table 3), it is likely that temporal successions or trajectories may be slowly occurring in concert with tidal cycle. Interestingly, conductivity and salinity were moderately correlated both with cumulative time and bacterial community variation. Conductivity, which is in part measured by the number of ions relative to salinity, was more strongly correlated to community variation. Previous studies have found salinity to be important in determining estuary community composition (4b, 4c). During this study, the range of salinity was only 3 ppt; this gradient may not have been biologically relevant over hourly observation resolution but may become more influential as temporal perspective broadens.

The observed similarity of bacterial community composition in ebb and flood waters was interesting because the origins of the waters were very different. While ebb waters drain the salt marsh, flood waters originate from Doboy Sound and the Atlantic Ocean. In-stream processes may have affected the bacterial communities. The lack of distinction between ebb and flood community compositions could be explained by the variation in creek velocity with tide and the corresponding capacity of the creek to carry suspended sediments. During ebb and flood tide, higher velocities increase suspended sediment load and their associated bacterial communities. During slack tide, velocity decreases, approaches zero, and reverses direction. Lower fluid velocities have less capacity to carry suspended loads, and a fraction of suspended-sediment bacteria may settle out of the water column during slack tides. Slack high and low tides may allow suspended pelagic bacteria, from the ocean or sound and from the marsh, respectively, to be detected. It is possible that ebb and flood communities were indistinguishable because in-creek suspended bacterial communities are distinct from communities more influenced by marsh, estuary, or ocean conditions.

Bacterial community dynamics and scale in South End tidal creek.As do all ecologists, microbial ecologists must choose an appropriate scale of observation for the question asked and the study subject (1a, 7c, 14). This question of scale is further complemented by the ongoing debate as to the applicability of traditional ecological theory to microbial systems (14).

Multiple studies have examined various temporal scales for microbes. Kent et al. have shown that lake bacterial communities must be observed minimally twice weekly to explore interactions with phytoplankton (11), and Jones et al. have observed bacterial community response to stimuli in as quickly as 1 to 3 days (10). Many studies have shown strong interannual or seasonal dynamics in riverine, oceanic, and lake bacterial communities (4c, 7a, 17). Researchers must often make assumptions about the frequency of aquatic microbial surveys and may not expect small-scale (e.g., hourly or less) changes to impact overall results. In reality (and depending on the question), a microbial community fingerprint, such as those produced by denaturing gradient gel electrophoresis, ARISA, or terminal restriction fragment polymorphism techniques, only portrays a snapshot of a microbial community at the observation time. It is often unknown whether a fingerprint observed an hour later or earlier would inspire a different conclusion.

In this study, the influence of tidal stage was the strongest measured parameter that explained variation in bacterial community composition. Therefore, observations taken an hour apart did not appear to be distinct, but observations taken 6 hours apart were distinct. For future studies of tidal creek bacterial community composition and dynamics, this result can direct questions of temporal resolution when designing experiments or deciding on an appropriate sampling regime.

Assumptions and caveats of the study.As with any PCR-based molecular technique, our results may be biased by our fingerprinting technique, ARISA. One particular issue with ARISA is that multiple OTUs may share the same ITS length but be phylogenetically distinct. Oppositely, some OTUs may have multiple rRNA genes and, thus, may be represented by more than one ITS length, leading to redundancy in the community data set. One way to address this concern is to build and sequence clone libraries from targeted 16S-ITS regions to identify the ARISA OTUs, as in the case of Newton et al. (13). However, Jones et al. have shown that overall ARISA microbial community patterns are robust to methodological differences, and despite these biases support the same ecological trends (10a). Because the focus of this study is community dynamics rather than specific population abundances, and because the occurrence of overrepresented ARISA OTUs or inaccurately merged OTU groups is yet unknown for aquatic systems, readers should interpret results for general trends and not for absolute abundance differences.