Changes in free-living bacterial community diversity reflect the magnitude of environmental variability

Authors

  • Alice. C. Ortmann,

    Corresponding author
    1. Department of Marine Sciences, University of South Alabama, Mobile, AL, USA
    2. Dauphin Island Sea Lab, Dauphin Island, AL, USA
    • Correspondence: Alice. C. Ortmann, Department of Marine Sciences, University of South Alabama, Mobile, AL 36688, USA. Tel.: +251 861 2141 ext 7526; fax: +251 861 7540; e-mail: aortmann@disl.org

    Search for more papers by this author
  • Natalie Ortell

    1. Department of Marine Sciences, University of South Alabama, Mobile, AL, USA
    2. Dauphin Island Sea Lab, Dauphin Island, AL, USA
    Search for more papers by this author

Abstract

A 2-year study was undertaken to compare patterns in the diversity of free-living bacteria in a river-dominated estuary and offshore, on the shelf, to determine whether changes in the free-living bacterial community could be related to differences in environmental seasonality and variability. Although the environmental conditions inshore were significantly more variable than those on the shelf and demonstrated clear seasonal patterns, there were no significant differences in the alpha diversity of the communities based on richness, evenness, or phylogenetic diversity. Comparison of communities using Bray–Curtis similarity indicated no significant differences in the magnitude of change between sequential samples from inshore and on the shelf. Seasonal differences were detected both inshore and on the shelf. However, analysis using the weighted UniFrac distance indicated significantly lower overall change between shelf samples with no significant seasonal differences. These findings suggest different patterns of change between the two sites. Inshore, changes in the relative abundance of distantly related bacterial species reflect the larger environmental variability, while on the shelf, changes in the relative abundance of closely related bacterial species or strains may result in a more functionally stable community. Thus, the magnitude of environmental change can alter patterns of bacterial diversity in marine systems.

Introduction

Long-term studies of the diversity of marine bacterial communities have led to the observation of seasonal patterns in community composition (Fuhrman et al., 2006; Nelson et al., 2008; Andersson et al., 2009; Carlson et al., 2009; Gilbert et al., 2009, 2010, 2012; Ngugi & Stingl, 2012). The patterns suggest that environmental and biotic factors drive the assembly of communities from specific species that are detected year after year. The repeated detection of the same OTUs over time suggests that these communities are not composed of functionally redundant organisms, but have specialized members (Fuhrman et al., 2006; Fortunato et al., 2011; Kara et al., 2013). Several factors have been identified as driving the observed seasonal patterns, including depth of the mixed layer (Carlson et al., 2009; Treusch et al., 2009), day length (Gilbert et al., 2010, 2012), temperature and nutrient concentrations (Fuhrman et al., 2006), and depth and salinity (Fortunato et al., 2011). These studies indicate that specific groups are responding to the environment to occupy specific ecological niches (Vergin et al., 2013). Most of these studies have been carried out in oceanic, deep coastal waters, upwelling zones, or in temperate regions where strong seasonal variations in temperature occur. In contrast, the northern Gulf of Mexico (nGOM) is characterized as relatively shallow with a wide shelf where summer hypoxia commonly occurs (Rabalais et al., 2002; Park et al., 2007). The nGOM is subtropical, but unlike coastal California which would be defined as a Mediterranean subtropical climate with dry summers, the nGOM has a humid subtropical climate, lacking a true dry season (Peel et al., 2007).

Few studies have investigated the diversity of the microbial community in the nGOM, and none have characterized the microbial diversity over time (Jones et al., 2010; Olapade, 2010; King et al., 2012). The most comprehensive survey sampled surface and deep water from Texas to Alabama coasts at 17 stations in March 2010 (King et al., 2012). This study found little spatial variability in the community composition, with the largest differences identified between surface (< 100 m) and deep water (> 100 m) samples. The SAR 11 group of the Alphaproteobacteria and Bacteroidetes dominated the surface waters, with an increase in Gammaproteobacteria, Thaumarchaeota, and Firmicutes with depth. The majority of the 17 stations were in the nGOM, with salinities > 30 ppt. One station, in the Mississippi River plume, had much higher abundances of Betaproteobacteria, indicating that coastal microbial communities may differ. This was also observed in a study carried out at shallow sites around Apalachicola Bay, where the sites within the bay were more similar to each other than sites on the nGOM side of the barrier islands (Olapade, 2010). These studies provide a snapshot of diversity over a short period of time and do not address the potential seasonality of the bacterial communities.

Over a 2-year period, surface water was collected from four stations, two within Mobile Bay, AL, and two in the nGOM, to determine whether the bacterial community showed seasonal patterns and how patterns in community shifts are related to environmental change. Based on discharge, Mobile Bay is the fourth largest estuary in the United States, second only to the Mississippi River in the nGOM (Schroeder & Wiseman, 1999). This results in large variations in the environmental conditions within the bay and nearshore areas. Alpha- and beta-diversity metrics were used to test the hypothesis that larger environmental variability inshore would result in larger changes in the diversity of the free-living bacterial communities inshore compared with shelf locations.

Materials and methods

Sampling

Surface samples were collected from four locations from within Mobile Bay, AL, to the shelf in the nGOM (Fig. 1). Samples were collected starting in July 2009 through December 2011, approximately every other month. Temperature, salinity, and dissolved oxygen (DO) were measured with a Seabird 19 + CTD. Nutrients [dissolved inorganic nitrogen (DIN), dissolved organic nitrogen (DON), math formula, and dissolved silicate (DSi)] were determined using a Skalar SAN+ autoanalyzer after filtering water through a 0.7-μm glass fiber filter (Whitledge et al., 1981). Concentrations of chlorophyll a (chl a) were determined fluorometrically after extraction of pigments with acetone/DMSO from a 0.07-μm filter (MacIntyre & Cullen, 2005). The abundances of heterotrophic nanoflagellates (HNF) were determined using epifluorescence microscopy and DAPI staining (Sherr et al., 1993), while prokaryotes and viruses were enumerated using a BD FACSCalibur flow cytometer after staining with SYBR Green (Life Technologies, CA) (Ortmann et al., 2012). River discharge for the 7 days prior to sampling was estimated based on discharge from the Tombigbee and Alabama rivers (N. Ortell & A.C. Ortmann, unpublished data). Average river discharge for wet (spring and winter) and dry (summer and autumn) seasons was calculated along with standard errors.

Figure 1.

Map of Mobile Bay, AL, and the nGOM where surface water samples were collected. The shelf samples include T20, in 20 m water, and T35, in 35 m water. Inshore samples include MB and DI, in 3 and 10 m water, respectively. Created in arcgis.

Differences between inshore and shelf samples were tested in jmp 9.0 (SAS Institute, NC) using the Welch anova for unequal variances for each variable. Environmental variables were normalized in Primer 6 (Primer-E Ltd, UK) after DIN and math formula concentrations were log(x + 1) transformed. Pairwise comparisons of samples were calculated using the Euclidian distance, and these values visualized using multidimensional scaling (MDS). Analysis of similarity (ANOSIM) was carried out on each location separately to quantify seasonal differences in the environment. Values were considered significant if R-values > 0.3 and P-values < 0.05. Euclidian distances between sequential samples were extracted from the matrix to calculate the magnitude of environmental change over 2 months and determine whether differences were significant between inshore and shelf samples. Samples from July 2009 were not included in this analysis as the time between samples was 4 months.

DNA extraction and sequencing

Samples were originally collected to determine rates of grazing and viral lysis of prokaryotes and phytoplankton using dilution experiments (N. Ortell & A.C. Ortmann, unpublished data). Because of this, all samples were prefiltered through 150-μm nitex screening and then gently filtered through a 142-mm-diameter 0.7-μm glass fiber filter before cells were collected on a 142-mm-diameter 0.2-μm polyvinyl membrane (Durapore, Millipore, CA). 20–24 L was processed for each sample. The filters were folded inward into quarters and stored frozen, at −20 °C, in sterile sample bags until processing. To extract the free-living bacterial DNA, the 0.2-μm filter was cut with a sterilized razor blade, and c. 1/16 of the filter placed in a microfuge tube. Lysis buffer (Massana et al., 1997) was added along with 1 mg mL−1 lysozyme and the mixture was incubated at 37 °C for 45 min. Proteinase K (0.2 mg mL−1) and SDS (1%) were added to the tube, and it was incubated for 1 h at 55 °C. The DNA was then extracted using phenol/chloroform and precipitated with isopropanol and ammonium acetate.

For each sample, three separate 20 μL PCRs were carried out using a mixture of 5 forward and 4 reverse primers targeting the V6 hypervariable regions of the bacterial 16S rRNA gene (Sogin et al., 2006; Huse et al., 2008). Primers were modified to include a 5-bp barcode (forward primers) and adapter sequences for the Ion Torrent PGM sequencer (Life Technologies). Reactions were carried out using Phusion Master Mix (Thermo Fisher Scientific, MA). After an initial 30 s at 98 °C, 35 cycles of 98 °C for 10 s, 63 °C for 15 s, and 72 °C for 15 s were followed by 5 min at 72 °C. The three reactions were combined, and amplification was checked by running an aliquot on an agarose gel. Enzymes and dNTPs were removed from the PCRs using the MinElute PCR Clean-up kit (Qiagen, CA), and the concentration of the PCR products was measured using a NanoDrop 1000 (Thermo Fisher Scientific).

Sequencing was carried out using an Ion Torrent PGM with barcoded samples pooled on 316 chips. Amplicons were pooled based on the NanoDrop concentrations to obtain equimolar quantities of each sample. The pooled samples were quantified using the High Sensitivity DNA kit on a Qubit fluorometer (Life Technologies) and diluted to obtain the desired number of molecules for template preparation following the One Touch/ES system protocols. The amplified samples were loaded onto 316 chips and sequenced using the 100-bp Sequencing Kit V2.0 following the manufacturer's protocols.

Sequence analysis

The fastq files for each chip were downloaded from the Torrent Server and split into fasta and qual files using mothur (Schloss et al., 2009). These files were then processed using qiime (Caporaso et al., 2010a). Individual sequences were matched to their sample according to the barcodes and filtered to remove sequences that failed to (1) be longer than 70 bases; (2) have quality scores > 20; (3) have no ambiguous bases; and (4) have homopolymer runs with < 6 bases. Mismatches between barcodes and forward primers were not allowed. Both primers were removed along with the barcode. The fasta files from all chips were concatenated, producing a single fasta file for downstream analysis. Sequences were clustered into OTUs at 97% similarity using the qiime pick_otus_through_otu_table.py workflow (Kuczynski et al., 2011). Taxonomy was assigned using the May 2013 release of Greengenes (DeSantis et al., 2006) with a minimum confidence score of 0.8 and the naïve Bayesian RDP classifier (Wang et al., 2007). The OTU table was filtered to remove all OTUs represented by a single sequence as well as any OTU that failed to align with a 16S rRNA gene database using pynast (75% min ID, 46 bp) (Caporaso et al., 2010b). OTUs identified as chloroplasts were also removed from the OTU table. Sequences are available from the SRA at the NCBI under accession numbers SRR847546-SRR847560, SRR847562-SRR847596 under study PRJNA201435.

Rarefaction analysis was used to determine whether sufficient sampling effort was carried out. For each sample, 10 OTU tables at 10 different sampling levels from 10 to 65 000 reads were generated and alpha-diversity metrics calculated and plotted against sampling depth. To compare the diversity of the samples, all further analysis was carried out after subsampling the OTU table to even depth. For alpha- and beta-diversity analysis, a depth of 8900 sequences per sample was chosen as the smallest sample included 8976 sequences. For each sample, different measures of alpha diversity were calculated including estimates of richness [phylogenetic distance (PD), Chao1] and evenness (Shannon diversity). Coverage was estimated using Good's coverage and the Gini index. Alpha-diversity values were analyzed in jmp 9.0 (SAS Institute) to determine whether there were significant differences inshore compared with the shelf.

Beta diversity was calculated using two different approaches. The OTU table, with 8900 sequences per sample, was exported into Primer 6 (Primer-E Ltd, UK), and counts were fourth-root transformed. The transformed values were used to calculate the Bray–Curtis similarity between each pair of samples. In qiime, the weighted UniFrac distances were calculated using the same OTU table and the phylogenetic tree generated by the make_phylogeny.py script using FastTree (Lozupone & Knight, 2005; Price et al., 2009). The distance matrix was then imported into Primer 6 for analysis. The pairwise distances or similarities between subsequent time points and locations were extracted from the distance matrices and compared to determine whether there were significant differences in the variance and means between inshore and shelf samples. The three different distance matrices for each location were compared with each other using the RELATE function in Primer 6. To determine whether seasonal differences were detectable, anosim was carried out for each location using both Bray–Curtis similarities and weighted UniFrac distances.

Because these samples were prefiltered through a 0.7-μm filter, particle-attached bacteria may have been removed, altering the total diversity of the bacterial community. To estimate how the diversity may have been altered, a secondary analysis was carried out that included 16 samples collected from the same locations from January 2012 to June 2012. These samples were not prefiltered, and 0.5–1.0 L of water was processed for each sample. DNA extraction, PCR amplification, and sequencing were carried out as above, but samples were rarefied to 3500 reads per sample. anosim was carried out to determine whether there was a significant difference between the prefiltered communities and the unfiltered communities using three different beta-diversity metrics: Bray–Curtis similarity, unweighted UniFrac distance, and weighted UniFrac distance. Patterns were visualized with MDS.

Results

Environmental variability

Fifty-two samples were collected from four stations over the sampling time. The abundance of HNF, prokaryotes, and viruses was not obtained for inshore samples in November 2009 due to the loss of the samples; therefore, these samples were not included when calculating the distance matrix for environmental samples. Samples from the most southerly station (T35) were not collected until May 2010 following the Deepwater Horizon oil spill. Both shelf sites received some oiling from the spill, with the majority occurring in June (Graham et al., 2010). When the July 2010 samples were collected, a light sheen was reported at both locations, but no samples were analyzed for hydrocarbon concentrations (R. Condon, personal communication). No impact of oil on the free-living bacteria on the shelf was detected, and impacts are assumed to be minimal.

The inshore environment was more variable throughout the sampling period compared with the shelf environment, although at times, the conditions were similar (Fig. 2). Individual testing of the variables measured indicates that salinity, nutrients, and microbial biomass were more variable inshore compared with the shelf (Table 1). Salinity, although significantly lower inshore compared with the shelf, varied by 32.83 ppt inshore compared with only 14.61 ppt on the shelf. Nutrients and microbial biomass were significantly higher inshore, with the exception of math formula , which was more variable inshore, but not significantly higher compared with the shelf.

Figure 2.

MDS plot of the Euclidian distance between samples based on environmental variables. Samples from different locations overlap; however, the variability is higher for the inshore samples, as indicated by the wider distribution of points. Inshore samples from November 2009 are not included in this analysis.

Table 1. The range of values of environmental variables along with the means is shown for inshore (n = 28) and shelf (n = 24) samples
VariableInshoreShelf
RangeMean (standard deviation)RangeMean (standard deviation)
  1. Bold indicates variables where the locations had significantly different variance and significantly different means (Welch anova).

  2. a

    The variance inshore was significantly higher than the shelf, but there was no difference in the means.

  3. b

    These values do not include estimates from November 2009 inshore samples.

Temperature (°C)6.44–31.7421.98 (7.68)11.7–30.9923.24 (5.88)
Salinity (ppt)0.7033.53 14.88 (9.83) 20.9835.59 29.76 (4.35)
DO (mg L−1)5.3212.59 8.34 (1.85) 5.328.82 6.88 (0.91)
DIN (μM)0.0815.94 4.77 (5.25) 0.267.81 1.30 (1.63)
DON (μM)10.68–74.5525.00 (14.44)9.63–53.9718.95 (13.01)
math formula (μM)a0.081.45 0.30 (0.27) 0.050.88 0.20 (0.17)
DSi (μM)3.79106.34 47.26 (29.99) 0.1430.33 6.17 (7.12)
Chl a (μg L−1)1.0124.81 8.24 (6.24) 0.276.44 1.71 (1.50)
HNFs L−15.6 × 103–2.3 × 1051.1 × 105 (6.8 × 104)b6.8 × 103–2.2 × 1058.7 × 104 (5.6 × 104)
Prokaryotes (mL−1)2.7 × 1051.2 × 1074.9 × 106 (3.2 × 106)b6.2 × 1055.6 × 1062.4 × 106 (1.4 × 106)
Viruses (mL−1)7.9 × 106–1.6 × 1086.2 × 107 (2.0 × 106)b2.0 × 106–1.5 × 1084.6 × 107 (3.4 × 107)

anosim indicated significant differences between inshore samples based on season (Table 2). Pairwise tests detected significant differences between summer & spring and summer & winter or between spring & autumn. The smallest differences were between spring & winter. Both winter and spring were characterized by lower temperatures, lower salinity, and higher nutrients compared with the other seasons. For shelf samples, the global test indicated no significant difference between seasons, so no pairwise comparisons were carried out.

Table 2. anosim R-values for comparison of environmental parameters across spring (March–May), summer (June–August), autumn (September–November), and winter (December–February)
ComparisonInshoreShelf
  1. Bold indicates where values are considered significant (R > 0.3, P < 0.05).

Global test 0.39 0.11
Spring–autumn 0.55  
Spring–summer 0.58  
Spring–winter0.17 
Autumn–summer0.24 
Autumn–winter0.27 
Summer–winter 0.63  

Sequence analysis

After processing, the final OTU table included 126 781 OTUs across 52 samples. During processing, c. 96% of the reads passed the preliminary quality control, but only c. 60% of the total OTUs were included in the OTU table after filtering. This step removed OTUs represented by a single read (57% of OTUs), those that failed to align with 16S rRNA gene templates (c. 3%) and those related to chloroplasts (< 0.1%). The average number of sequences in each sample after processing was 67 256 reads, but there was large variation across the samples (range: 8976–243 409 reads per sample). Rarefaction analysis based on PD and Chao1 suggests that richness was not saturated at sampling depth of 65 000 reads (Supporting Information, Fig. S1A and SB), while the evenness of the community appeared saturated in samples with n > 13 000 (Fig. S1C and D).

To compare across samples, a sequencing depth of 8900 reads per sample was chosen. Good's coverage (0.89 ± 0.02) and the Gini index (0.99 ± 0.004) indicate that at this depth, most of the diversity was sampled with an n = 8900; however, the communities were highly uneven (Wittebolle et al., 2009). The unevenness of the communities was confirmed by the high number of OTUs represented by a single read in the resampled table. In the rarefied OTU table, 64.7% ± 3.0 of the OTUs were represented by a single read.

Comparison of the prefiltered bacterial communities with the additional unfiltered communities indicated no significant difference based on any of the beta-diversity metrics used (Table S1, Fig. S2). This suggests there is no clear bias in the bacterial diversity due to prefiltration; however, the communities described below will be considered representative of the free-living bacterial community.

Alpha diversity

Three different measures of alpha diversity were calculated for the samples. Wilcoxon signed-rank tests indicated no significant differences between the richness (Chao1), evenness (Shannon diversity), or the taxonomic diversity (PD) of the inshore samples compared with the shelf samples (Fig. 3). There was no obvious pattern in the magnitude of the change in alpha diversity between sample times. The largest changes were observed in January 2011, when low diversity was detected at both locations. Environmental variables do not indicate anything unusual about these samples. PD and Shannon were significantly lower in winter (anova, P < 0.05); however, removal of the January 2011 points resulted in no significant seasonal differences.

Figure 3.

Alpha-diversity estimates for samples over time. Means and standard deviations are shown for (a) phylogenetic distance, (b) Chao1 estimate, and (c) Shannon diversity. No significant differences were detected between locations.

Beta diversity

Beta diversity was calculated using two different metrics: one which measures turnover of species, defined as OTUs (Bray–Curtis similarity); and one that takes into consideration the phylogenetic relatedness of the species (weighted UniFrac) (Lozupone & Knight, 2005). The Bray–Curtis metric does not use phylogenetic information and treats all species as completely unrelated to each other. When communities have species that are distantly related, the weighted UniFrac distances will show similar patterns as the Bray–Curtis distances because the distances between species on a phylogenetic tree are large. When communities are composed of closely related species, the smaller distances between species on a phylogenetic tree will result in smaller pairwise distances using the weighted UniFrac metric relative to the Bray–Curtis metric. To measure the changes between sampling times, pairwise values for the metrics were compared for sequential samples along with changes in the environment (Fig. 4). The distance between samples was significantly higher and more variable inshore compared with the shelf, indicating larger changes in the environment inshore (Fig. 5).

Figure 4.

Pairwise changes in (a) environmental, (b) Bray–Curtis similarity, and (c) weighted UniFrac distances for inshore and shelf samples. Values are means of all pairwise comparisons for each set of samples with the standard deviation of the mean.

Figure 5.

Mean and standard deviations of environmental and community changes based on different metrics. Asterisks indicate the location with significantly higher means based on a Welch anova.

Differences between samples were significant based on Welch anova of Bray–Curtis similarities (P = 0.05) and weighted UniFrac distances (P = 0.003; Fig. 5). The differences indicated larger changes in the bacterial communities between samples inshore compared with the shelf. For both measures of community change, the variance inshore was significantly higher than on the shelf (F-test, P < 0.01).

Analysis of beta-diversity matrices using anosim identified significant seasonal differences in the diversity of inshore communities, but the patterns for inshore communities depended on the metric used (Table 3). Inshore, the Bray–Curtis similarity and weighted UniFrac distances both show significant differences in community structure across seasons. Most of the pairwise comparisons between seasons indicated significant differences. On the shelf, seasonal differences were only detected based on Bray–Curtis similarities. As for the inshore analysis, most seasons were significantly different from each other (Table 3). The global test using the weighted UniFrac distances was not significant for inshore samples, indicating no seasonal differences. No pairwise comparisons were carried out for this analysis.

Table 3. anosim R-values for seasonal comparisons of Bray–Curtis similarities and weighted UniFrac distances for inshore and shelf samples
ComparisonInshoreShelf
Bray–CurtisWeighted UniFracBray–CurtisWeighted UniFrac
  1. Bold indicates where values are considered significant (R > 0.3, P < 0.05).

Global test 0.41 0.36 0.37 0.285
Spring–autumn 0.48 0.49 0.49  
Spring–summer 0.38 0.49 0.36  
Spring–winter 0.33 0.37 0.30  
Autumn–summer0.28 0.30 0.34  
Autumn–winter0.200.290.19 
Summer–winter 0.56 0.70 0.59  

To compare the patterns in beta diversity with patterns in environmental variability, the matrices were compared using the RELATE function in Primer 6 (Table 4). There was a stronger relationship between the environment and free-living bacterial community inshore for both beta-diversity metrics than on the shelf. There was a weak relationship between the Bray–Curtis similarity matrix on the shelf and the environmental matrix, but the relationship between the weighted UniFrac distance matrix and the environment was not significant. The two beta-diversity metrics were well correlated in both locations, although the relationship was higher inshore compared with the shelf.

Table 4. Rho and P-values (in parentheses) from the RELATE function measuring the degree of similarity between the environmental similarity matrix and community composition as well as a comparison of the two beta-diversity matrices
ComparisonInshoreShelf
Bray–Curtis similarity to environment0.482 (0.001)0.254 (0.02)
Weighted UniFrac distance to environment0.357 (0.001)0.130 (0.13)
Bray–Curtis similarity to weighted UniFrac distance0.857 (0.001)0.753 (0.001)

Discussion

The inshore stations represented a variable environment with seasonal differences driven by river flow and temperature. River discharge affects both the salinity and the nutrient concentrations throughout the bay and resulted in significant differences in seasonal conditions. During lower river flow in summer and autumn (1174 m3 s−1 ± 602), salinity increased in the bay, while nutrients decreased and the conditions were more similar to the shelf environment. These samples overlapped with the shelf samples in the multivariate plots. High river flow during winter and spring (2883 m3 s−1 ± 743) reduced salinity, resulting in the bay being close to freshwater at times. The main difference between winter and spring was temperature. Temperature did vary substantially throughout the sampling period; however, the lowest observed temperatures, in January 2010 (6.4 °C ± 0.007), were anomalously low. Temperatures in January 2011 and 2012 averaged 14.1 °C (± 2.7) inshore. On the shelf, the main factor showing any large variability was temperature, but this was not sufficient to generate significant seasonal differences based on the measured environmental variables.

Based on the higher environmental variability and seasonal differences inshore compared with the shelf, it was hypothesized that the diversity and turnover of the free-living bacterial community would be higher inshore compared with on the shelf. Surprisingly, there was no difference between locations based on alpha-diversity measures, while differences based on beta-diversity measures were observed using phylogenetic-based analyses.

Three different alpha-diversity metrics, which measure different aspects of diversity, showed similar patterns over time, both inshore and on the shelf. Both locations had free-living bacterial communities that were dominated by a few abundant taxa along with a large number of rare taxa representing a diverse rare biosphere. Unlike in the Delaware Bay estuary, where a change in diversity and evenness was observed across the salinity gradient (Campbell & Kirchman, 2013), the system here showed no spatial patterns. No patterns were observed in the growth rate of prokaryotes across the same gradient (N. Ortell & A.C. Ortmann, unpublished data), suggesting that the similar productivity on the shelf supported as diverse a free-living bacterial community as was found inshore.

Although values for PD and Shannon were significantly lower in winter in both locations, these differences were driven by the low values in January 2011. Removing the January 2011 estimates from the analysis resulted in no significant seasonal differences. The lack of seasonal patterns in PD, Chao1, and Shannon diversity contrasts with a study in the English Channel where higher diversity was observed during winter, with lower diversity in summer (Gilbert et al., 2009, 2012). In the English Channel, a proxy for day length (DX1) was found to be the main factor explaining > 65% of the variability in community richness (Gilbert et al., 2012). A global analysis of bacterial diversity data also indicated a winter peak in diversity in temperate regions, with day length being a main driver of diversity along with proximity of the thermocline and phosphate concentrations (Ladau et al., 2013). Day length is a proxy for season, indicating that season was a strong driver of community richness. Because multiple factors and interactions vary over season, it appears that no single variable is responsible for changes in the bacterial community. In this study, day length was not included in the environmental analysis as it did not differ between inshore and the shelf, but temperature, which showed the largest seasonal pattern in both locations, was not correlated with any measures of alpha diversity. Because day length in the nGOM increases only 4 h from December to June compared with 8 h in the English Channel, it is possible that the smaller range of seasonal variability in subtropical waters results in no significant seasonal patterns in alpha diversity.

The larger variation in the environment inshore did translate to larger changes in the structure of free-living bacterial communities using beta-diversity metrics. Patterns of change on the shelf depended on which metric is used. Both Bray–Curtis and weighted UniFrac metrics detected larger variation in the magnitude of community changes inshore compared with the shelf. The mean change between consecutive samples was significantly higher inshore. There was no clear correlation between the environmental change (Euclidian distance) and the change in the free-living bacterial community (Bray–Curtis similarity or weighted UniFrac distance). This could be due to the fact that both measurements integrate change over time, but that the time scale or response time (lag) may differ (Steele et al., 2011; Hatosy et al., 2013).

Strong seasonal patterns based on anosim of Bray–Curtis similarities were detected both inshore and on the shelf, although there was no significant difference between autumn & winter communities in either location. In contrast, while seasonal variability was detected inshore when communities were analyzed using anosim and the weighted UniFrac distance, no significant seasonal pattern was observed on the shelf. These differences suggest different patterns of community change inshore and on the shelf.

Inshore, the Bray–Curtis and weighted UniFrac pairwise distances show similar patterns, suggesting that the species that change in abundance over time were distantly related. As the environment changed, the relative abundance of bacterial species adapted to the new conditions increased, while poorly adapted species decreased in abundance. The low relatedness between these species would suggest that they had different functional and metabolic needs, which would translate to different functional roles in the ecosystem. On the shelf, different patterns were observed for the weighted UniFrac distances compared with the Bray–Curtis distances (Table 4). The smaller pairwise distances based on the weighted UniFrac metric relative to the Bray–Curtis distances suggest that the species that changed in abundance were related to each other. Under the more stable environmental conditions on the shelf, communities may change at the species or strain level. This pattern was observed in stable human-controlled aquaculture systems where most of the diversity was at the strain level (Rodriguez-Brito et al., 2010). Closely related bacterial species likely have more similar physiological and metabolic requirements compared with more distantly related species. On the shelf, where the environment was more stable, with little seasonality, the free-living bacterial community may be more functionally stable. Thus, the magnitude of environmental change altered the patterns of bacterial diversity in the surface water of the nGOM.

Acknowledgements

This study was supported by a BP Rapid Response Marine Environmental Sciences Consortium grant to A.C.O. and BP-Northern Gulf Institute Year 1 and 2 grants to A.C.O. Thanks to R. Condon, L. Wang, and B. Christiaen for critical comments to improve the manuscript, FOCAL for providing samples, and everyone involved in processing samples.

Ancillary