Water samples were collected in 2014 on the spring and fall AZMP missions aboard the CCGS Hudson (HUD2014004 April 4-23, HUD2014030 September 19 – October 12) at stations throughout the SS and Scotian Slope, occasionally reaching Gulf Stream influenced off-shelf waters (Figure 1). Six sections including the Browns Bank Line (BBL), the LaHave Basin Line (LHB), the Halifax Line (HL), the Louisbourg Line (LL), the St. Anns Bank Line (STAB), and the Cabot Strait Line (CSL) were extensively surveyed in 2014. In addition, during the spring leg, the station GULD04 corresponding to the ecologically rich shelf break Gully region, and the Thebaud platform (TB01) station located in close proximity to the Thebaud oil and gas platform were sampled (Figure 1A). Variations in cruise tracks led to slight differences in stations sampled in the fall and spring cruises. In total, 42 stations were sampled at 4 depths, resulting in 168 samples (92 samples from spring and 76 from fall cruises) with 64 sampling sites shared between the two 2014 cruises (Figure 1A,B).

We compared the microbial community similarity at stations along the Halifax line, a section sampled in the spring and fall AZMP cruises of 2014 and 2016. Repeat sampling of HL was carried out in 2016 AZMP missions (HUD2016003 April 9–25, HUD2016027 September 25 – October 6), resulting in an additional 65 samples for comparison with 2014 (Supplementary Table S1). At each selected station, 4 L of water was collected from each of 4 depths and pre-filtered through a 160 μm (2014) or a 330 μm (2016) mesh to remove mesozooplankton. The water was then filtered through a 3 μm polycarbonate Isopore filter (Millipore, United States) to capture particle associated (PA) bacteria and then redistributed into 4 L bottles and filtered using a vacuum (2014), or a peristaltic pump (2016) through 0.2 μm polycarbonate Isopore filters (Millipore, United States) to capture free-living (FL) bacteria. Both the 3 and 0.2 μm filters were stored at -80°C. At all stations sampled, 1 and 20 m water samples were collected, while the other two depths were selected based on the depth of the ocean floor and oceanographic features in the vertical depth profile (Supplementary Table S2). Samples were also collected from the water column oxygen minima when present, usually corresponding to a depth of 250 ± 50 m. Bacterial abundance was measured using flow cytometry. Parallel 1.8 mL seawater samples for analytical flow cytometry (AFC) were fixed with 1% paraformaldehyde (Alfa-Aesar, United States), incubated at room temperature for an hour, then stored at -80°C for later analysis. Nutrient and chlorophyll measurements were made using standard AZMP protocols (Mitchell et al., 2002).

Samples were categorized based on season (spring or fall), and depth: Surface (1–20 m), Photic (40–80 m), and Deep (100–300 m). The discrete depth categories were based on the average photic zone depth from over 10 years of observations by the Bedford Institute of Oceanography (BIO) on the SS (Johnson et al., 2014). Specifically, all surface and deep samples were consistently taken within the photic zone and aphotic zone, respectively. Samples collected at the intermediate depths in the photic zone were within the range of photic depth variability throughout the year, thus the amount of light these samples received was variable and lower than surface depths. A subset of samples from cross-shelf sections (BBL, LHB, HL, STAB1-LL; Figure 1 and Supplementary Figure S1) were used to investigate the communities at the shelf-break. These samples were grouped into categories on-shelf (BBL1, 2, 3; LHB2, 4; HL1, 2; STAB1; LL4), shelf break (BBL5; LHB6; HL4, 5.5, 6; LL7), and off-shelf (BBL7; LHB6.7; HL8, 11; LL9) based on geographic position relative to the shelf break and spatial patterns of surface temperature and salinity (Supplementary Figures S1, S2A,B,F,G). Temperature-salinity diagrams for all transects by season (not shown) were also used to confirm that the on-shelf and off-shelf groupings reflected distinct water masses of the Scotian shelf (SS).

DNA was extracted from technical duplicate 0.2 and 3 μm polycarbonate filters using the DNeasy Plant Mini Kit (Qiagen, Germany) according to the manufacturer’s instructions with some minor modifications in the cell lysis procedure. Fifty microliters of lysozyme (5 mg/mL) (Fisher BioReagents, United Kingdom) was initially added to each filter and each sample was vortexed on high for 30 s. Then 400 μL of lysis buffer AP1 (from the DNeasy Plant Mini Kit) was added to each sample tube followed by the addition of 45 μL of proteinase K (20 mg/mL) (Fisher BioReagents, United Kingdom). The samples were then incubated at 55°C with shaking for 1 h. After this incubation, 4 μL of RNase A (Qiagen, Germany) was added to the samples, which were then kept on ice for 10 min. From this point on, the extraction followed the manufacturer’s protocol, with a final elution of the DNA in 100 μL of elution buffer. DNA concentrations and purity were measured with a NanoDrop 2000 (Thermo Scientific, United States).

The samples were prepared for sequencing on an Illumina MiSeq instrument, following the Microbiome Amplicon Sequencing Workflow (Comeau et al., 2017). Each DNA sample was amplified using dual-indexing Illumina fusion primers that targeted the V6–V8 438 bp region of the bacterial 16S rRNA gene (Comeau et al., 2011). We used the forward B969F (ACGCGHNRAACCTTACC), and the reverse BA1406R (ACGGGCRGTGWGTRCAA) primer. The full primer sequences, including fusion sequences and adapters are listed in Supplementary Table S3. Each DNA sample was amplified as an undiluted template and at a 1:10 template dilution, to reduce the potential effects of PCR bias. The PCR products from diluted and undiluted templates were pooled and their quality was verified using an E-gel 96-well high-throughput system (Invitrogen, United States). Library normalization and PCR clean-up was conducted using a SequalPrep 96-well Plate Kit (Invitrogen, United States). After normalization, all samples were pooled together, and the final library pool was quantified using a Qubit with PicoGreen (Invitrogen, United States). Finally, the pooled samples were run on an Illumina MiSeq sequencer using paired-end 300 + 300 bp v3 chemistry. The MiSeq on-board software demultiplexed the reads, creating one forward and one reverse read file per sample. Raw sequence files are available at the NCBI Sequence Read Archive under accession SRP076591, and PRJNA325151.

Analytical flow cytometry (AFC) was used to characterize bacterial abundance. Samples were analyzed with a BD Accuri Flow Cytometer (BD Biosciences, United States). Measurements of bacterial concentrations were made by adding SYBR (Invitrogen, United States) stain to the sample and incubated in the dark for 15 min. The SYBR stained samples were run with a threshold of 800 at FL1 and the gates used to determine bacterial counts followed a bacterial gating strategy developed for the Accuri instrument (Gatza et al., 2013; Prest et al., 2013).

Preliminary analysis and processing of 16S rRNA gene sequences followed a QIIME version 1.8.0 (Caporaso et al., 2010) pipeline workflow (Comeau et al., 2017). The program PEAR version 0.9.6 was first used to merge the demultiplexed, paired-end sequences together (Zhang et al., 2014). After merging paired ends, sequences less than 400 bp in length or with a quality less than 30 over 90% of bases were discarded. Chimeric sequences were removed using UCHIME (Edgar et al., 2011). Operational Taxonomic Units (OTUs) were clustered based on 97% sequence similarity using sortmerna (Kopylova et al., 2012) for reference picking and sumaclust (Mercier et al., 2013) for de novo OTU picking (i.e., “open-reference” picking). This process used the reference Greengenes database version 13.8 (McDonald et al., 2012) for preliminary OTU picking, and then subsampled failed sequences using de novo picking. OTUs that were identified by less than 0.1% of reads were removed to account for bleed-through between runs on the Illumina MiSeq (Comeau et al., 2017), and the remaining OTUs were used for further analysis. In order to focus the study on the bacterial community, additional quality control measures included removing all sequences assigned to mitochondria, and chloroplasts, as well as the few reads that were assigned to Archaea, because the latter is poorly represented in sequence reads obtained from the V6-V8 variable region, in general. The Greengenes database used above, erroneously classified the family SAR86 within the class of Gammaproteobacteria as the genus “Candidatus Portiera”, which is a known endosymbiont of the white fly Bemisia tabaci (Jiang Z.F. et al., 2012). All OTUs assigned to the Candidatus Portiera classification were reclassified as belonging to the family of SAR86. A BLAST search of these misclassified sequences supported this decision. Analysis of the microbial community structure was conducted on the 9364 remaining OTUs that passed the quality control steps listed above. For indicator species analysis, only 390 OTUs that reached a relative abundance greater than 1% in at least one sample across the SS were included to avoid spurious observations.

To compare the relative abundance of OTUs between samples collected in 2014, sequence reads were rarefied to a sequencing depth of 7500 which corresponded to the duplicate sample pair with the lowest combined sequencing depth. The observations from the remaining duplicate samples were combined by averaging using the QIIME script collapse_samples.py. The comparison of the 2014 and 2016 samples from the Halifax line (HL) was conducted in a separate workflow; the samples were combined and rarefied to a sequencing depth of 5000 for downstream statistical analysis.

Statistical analyses were conducted using either base R version 3.2.1 (R Core Team, 2016) or the specific R packages described below. Figures, excluding maps, were made using gplots (Warnes et al., 2016) or ggplot2 (Wickham and Chang, 2015) packages in R. CTD data was extracted in R using the package oce (Kelley et al., 2018). Ocean Data View version 4.6.5 (Schlitzer, 2015) was used to create images featuring surface maps and transects of the SS. Density shading using the DIVA (Data-Interpolating Variational Analysis) gridding algorithm was implemented through Ocean Data View to visualize approximate spatial distribution of environmental variables and the distribution of select taxa (Barth et al., 2010).

Temperature and salinity increased with increasing distance from shore in both spring and fall (Figure 1A,B, Supplementary Figure S2, and Supplementary Table S2). Surface and photic zone temperatures were much warmer during the fall (surface average: 16.5°C, photic average: 8.4°C) than during the spring cruise (surface average: 3.3°C, photic average: 3.9°C), while temperatures at depth generally fell within the same range (5–15°C) in both seasons (Figure 1C). During both spring and fall, dissolved O₂ concentrations were higher in the surface and photic zones than at depth (Figure 1C). The average O₂ concentration in the fall was lower (241 μmol O₂/kg) than in the spring (313 μmol O₂/kg) (Figure 1C). Concentrations of ammonium, nitrate, nitrite and phosphate, in surface waters were generally higher in the spring than in the fall (Figure 1C). Nitrate, phosphate, and silicate concentrations increased dramatically with depth in both seasons, but ammonium and nitrite were relatively constant throughout the water column (Figure 1C). The average chlorophyll a concentrations were significantly higher (W = 5717, p-value < 0.001, Mann–Whitney-U Test) in the spring (maximum of 17.49 mg/m³ at STAB_01 [20 m]; average 3.97 mg/m³) than in the fall (maximum of 2.13 mg/m³ at CSL_01 [1 m]; average 0.39 mg/m³).

The physicochemical characteristics of the water column of HL in 2016 were comparable to that observed between spring and fall of 2014, reflecting the SS seasonal patterns (Figure 2) with respect to temperature, salinity, and chlorophyll concentrations (Figure 2). Surface waters exceeded 20°C throughout the section in fall of both years (Figure 2D,J), while lower water salinity down to approximately 75 m was observed on the shelf as far as the shelf break (Figure 2E,K). Although qualitatively similar, the absolute magnitude of chlorophyll a concentrations was much higher in spring of 2014 than in 2016 (Figure 2C,I). Both surface and deep water temperatures were higher than average in 2014, compared to the period 1981–2010, with the strongest positive anomalies in surface waters in the second half of the year (DFO, 2015; Hebert et al., 2015). Variation in seasonal temperature may cause slight difference in the timing of the onset of the spring bloom relative to the AZMP cruise dates in 2014 and 2016 (DFO, 2017).

Bacterial concentrations decreased steadily with depth but were significantly higher in the fall than in the spring, with an average of 1.3 × 10⁶ cells/mL and 0.7 × 10⁶ cells/mL, respectively (W = 5286, p-value < 0.001, Mann–Whitney-U Test) (Figure 1C). For all samples combined, 9364 OTUs from the FL and PA size fractions were identified representing 35 phyla (Figure 1C).

Bacterial cell density was significantly linearly correlated with temperature (adjusted R² = 0.26, F_1,122 = 44.14, p < 0.001) in a regression analysis that included all spring and fall samples, albeit with a stronger contribution from the fall samples due to a larger temperature range (Supplementary Figure S4) and overall higher cell density during that season (Figure 1C, 5G,H). The following results report on the composition and diversity of the bacterial community, independently of absolute abundance of individual taxa. We observed that the alpha diversity, measured both with number of distinct OTUs or Shannon diversity index, was positively correlated between the FL and PA fractions (Supplementary Figure S6) and overall higher in the PA fraction. Statistically significant correlation between the FL and PA fractions persisted in samples from specific depths (Surface: Adjusted R² = 0.3814, p-value < 0.0001; Photic: Adjusted R² = 0.1735, p-value < 0.01; Deep: Adjusted R² = 0.4675, p-value < 0.0001). Thus, samples exhibiting high diversity in the PA size fraction were also likely to exhibit high diversity in the FL size fraction. This relationship was stronger for species richness (observed OTUs, Supplementary Figure S6A) compared to the Shannon diversity index that takes into account both species richness and evenness (Supplementary Figure S6B). The Shannon diversity index was higher in the PA size fraction than in the FL size fraction, regardless of season (both seasons: W = 21398, p < 0.001; fall: W = 1208, p < 0.001, spring: W = 1352, p < 0.001; Mann–Whitney-U Test).

Several environmental factors were directly correlated to bacterial community diversity (Table 1). Overall, the Shannon diversity index and species richness (Chao1 index) were significantly positively correlated with salinity and other factors such as temperature and nutrients, while oxygen, chlorophyll a and latitude showed a negative correlation (Table 1; p < 0.05; Figure 7A,B). Although the strength of the correlation with diverse environmental variables varied with season, and depth, the sign of the statistically significant correlations remained the same across all categories with salinity, nutrients, depth and temperature showing a positive correlation with diversity, while oxygen, chlorophyll a, latitude and bacterial abundances were negatively correlated with diversity (Table 1).

We explored the positive correlation with temperature and salinity further. A plot of Chao1 vs. temperature revealed a negative quadratic relation with the highest species richness for both spring and fall samples converging at a temperature of ∼10°C for all the samples, although this temperature was slightly higher (13°C) for HL (Figure 7A). A closer examination of the salinity and temperature values at 50 m depth along the HL transect revealed that the intermediate temperatures, the shift in salinity and the highest bacterial species richness (Chao1) all coincided with the location of the shelf break, 200 km offshore (Figure 7C–H), a pattern that was accentuated in the fall (Figure 7D,F,H), relative to spring (Figure 7C,E,G) in both 2014 and 2016.

In general agreement with findings of previous studies (Crespo et al., 2013; Ganesh et al., 2014; Mohit et al., 2014; Rieck et al., 2015; Milici et al., 2016a), the taxa with high relative abundance in the PA fraction belonged to Deltaproteobacteria, Flavobacteriia, Verrucomicrobiae, Saprospirae (now classified with Bacteroidetes), and OM190 (from the Planctomycetes phylum), while Alphaproteobacteria and SAR406 clades were enriched in the FL fraction. Pelagibacteraceae, Rhodobacteraceae, and Bacteroidetes, the main taxa recovered in the FL fraction, are known as dominant marine bacteria in temperate coastal waters (Gilbert et al., 2012; El-Swais et al., 2015; Xiaomin et al., 2015). As expected, OTUs belonging to Pelagibacteraceae (also known as SAR11), some of the smallest but most abundant organisms on the planet (Morris et al., 2002; Giovannoni, 2017; Zhao et al., 2017), were dominant members of the bacterial communities in our study, however, with strong evidence for niche partitioning (Brown et al., 2012; Grote et al., 2012; Eren et al., 2013). Pelagibacteraceae OTUs were key indicators for FL, surface, photic, deep, off-shore, on shore, in both spring and fall. Specifically, fall and spring communities were dominated by Pelagibacteraceae ecotypes of the tropical PIa.3 (OTU#307744) and polar P1a.1 (OTU#637092) clades, respectively (Morris et al., 2005; Brown et al., 2012; Salter et al., 2015).

In contrast to seawater, particles provide a heterogeneous habitat for their associated bacteria (Alldredge and Cohen, 1987; Alldredge and Silver, 1988; Simon et al., 2002; Wright et al., 2012). Most PA bacteria are assumed to be attached either to marine snow, or other plankton, especially during blooms. Steep gradients in nutrients and oxygen create regions of microscale oxyclines in particles, resulting in selection for anaerobic bacteria and the concomitant buildup of anoxic metabolites such as hydrogen sulfide and methane, in an otherwise oxygenated environment (Shanks and Reeder, 1993). Populations of bacteria with upward of a thousand times more cells than in a comparable volume of seawater have been observed on particles (Alldredge et al., 1986; Turley and Mackie, 1994; Simon et al., 2002). Furthermore, aggregated particulate matter is more likely to contribute to transport of material from the surface to deep ocean (Boyd and Newton, 1999; Mestre et al., 2018) accounting for a larger proportion of carbon to the biological pump than FL bacteria. The relatively abundant uncultured class of Planctomycetes OM190 recovered in the PA fraction throughout the samples has been found in association with macroalgae (Lage and Bondoso, 2014). A member of the OM60 clade (OTU#630330), with high relative abundance in the PA fraction, is a likely representative of bacteriochlorophyll-containing aerobic anoxygenic phototrophs (AAPs). Most of the AAP bacteria isolated in culture belong to the OM60 clade (Zheng et al., 2016) and Congregibacter litoralis, a cultured member of this clade, is known for its ability to aggregate (Spring et al., 2009). The class of Cyanobacteria (containing Synechococcus) was also noticeably more abundant in the PA-associated fraction. The presence of Synechococcus, a small, oligotrophic, free-living genus, in the PA size fraction has been observed before, suggesting that it can also be bound to particles or hosts (Turley and Mackie, 1994; Simon et al., 2002; Crespo et al., 2013; Jackson et al., 2014; Yung et al., 2016). The differences in community composition between FL and PA fractions we observed is in agreement with previous studies, and supports the distinct functional roles of these communities (Cram et al., 2015; Milici et al., 2017). We thus advocate for size fractionation when sampling aquatic habitats.

Several bacterial OTUs were preferentially recovered from specific zones within the water column vertical profile. On the SS, surface waters were inhabited by Cyanobacteria, AAP bacteria, rhodopsin-containing bacteria, and bacteria associated with phytoplankton (e.g., Flavobacteriia). Below the photic zone, deep samples contained Deltaproteobacteria, Acidobacteria, and candidate phyla (e.g., PAUC34f), as well as select OTUs of Pelagibacteraceae. Deltaproteobacteria are diverse metabolically, with the ability to conduct sulfur oxidation, carbon fixation, C1 utilization, and heterotrophy (Sheik et al., 2014), pathways that often dominate in deeper water. Marine Group A (SAR406) and Planctomycetes, with members that participate in anammox and sulfur cycling (Francis et al., 2007; Wright et al., 2014), were significantly more abundant in deep waters. Further metagenomics and metatranscriptomic studies of the deep waters of the SS would provide information on whether these taxa are active members of the community or if they are dormant and resuspended by chance (Nemergut et al., 2013), especially in the shallow shelf region.

The boundaries between the on- and off-shelf microbial communities are delimited by the general circulation of the SS, which results in highest dispersal rates along the coast of Nova Scotia due to rapid flow of the Nova Scotia Current (NSC) and lower dispersal rates at and beyond the shelf-break current (SBC) (Rutherford and Fennel, 2018). The bacterial taxa preferentially found in either the spring or the fall most likely responded to differences in nutrients and productivity between seasons. On the SS, eukaryotic phytoplankton blooms and highly productive conditions are associated with the spring, whereas the early fall is associated with warm, stratified, and nutrient-limited conditions (Craig et al., 2014; Dasilva et al., 2014; Li et al., 2006). Bacteria preferentially found in the spring are likely copiotrophs able to achieve high growth rates in favorable conditions, whereas bacteria preferentially found in the fall could be considered oligotrophs with growth strategies adapted to low nutrients (Lauro et al., 2009). Our results show that the microbial community compositions of on-shelf and off-shelf regions are defined both by season and by their geographical location relative to the shelf break. However, we observed persistence at high relative abundance of select spring bacterial taxa during the fall, albeit only on-shelf. Conversely, a few bacterial taxa with high relative abundance throughout the SS during the fall were also significant members of the spring community, but restricted to the off-shelf region (Supplementary Table S8). These observations suggest that there may be an endemic population for select taxa throughout the SS waters, ready to grow under appropriate environmental conditions, supporting the view that the environment selects (Baas Becking, 1934). Alternatively, water exchange between the water masses, albeit low (Rutherford and Fennel, 2018), may nevertheless be sufficient to recruit bacterial taxa across the shelf break.

In the spring, the most abundant OTU was the cold-water ecotype of Pelagibacter clade P1a.1 (OTU#637092). Polaribacter and Colwelliaceae were also found in high relative abundance on-shelf only. Taxa more abundant in spring samples are likely associated with either phytoplankton directly via symbiosis or other cell-to-cell interactions, or indirectly, exploiting the dissolved organic carbon leaking from the phytoplankton blooms (Teeling et al., 2012; Georges et al., 2014; Wear et al., 2015). Many OTUs from Flavobacteriia, including those from Ulvibacter and Polaribacter, were strongly associated with the spring season and have previously been identified as important genera in the succession of phytoplankton blooms (Teeling et al., 2012; Klindworth et al., 2014; El-Swais et al., 2015). Cyanobacteria were the main taxa found more often in the fall. Minimalist oligotrophic cyanobacterial groups like Prochlorococcus and Synechococcus are known to thrive in extremely low-nutrient environments (Partensky and Garczarek, 2010; Flombaum et al., 2013). Therefore, the presence of these genera in the SS region highlights the extent of temperature stratification and oligotrophic conditions in the early fall. Notably, Prochlorococcus reached high relative abundances only in the off-shelf regions at any time during the sampling. Photosynthetic bacteria belonging to the AAP, as well as methylotrophs, were in high relative abundance in the off-shelf region of the SS, indicating that this region supported a bacterial community with a more specialized metabolism to exploit limited environmental resources.

Alpha diversity, measured as both species evenness and species richness, was significantly higher in the PA fraction than the FL fraction (Supplementary Figure S6). With some exceptions (e.g., Hollibaugh et al., 2000), a number of studies have reported similar trends (Crespo et al., 2013; Ortega-Retuerta et al., 2013; Rieck et al., 2015; Yung et al., 2016). Higher diversity in the PA fraction could be explained by the microenvironments of particles allowing for the accumulation of functionally diverse species, responding to microscale environmental gradients. The observed correlation between FL and PA diversity (Supplementary Figure S6), especially with species richness, may be a consequence of the broad range of environments sampled in this study. Although we cannot currently identify the processes that led to this correlation, our observations suggest that environmental factors affecting diversity are similar for FL and PA fractions.

Alpha diversity was strongly correlated with several environmental factors (Table 1). Teasing apart the role of individual environmental factors affecting the diversity and composition of bacterial communities is challenging because of the degree of correlation between temperature, salinity and other factors such as light, nutrients and dissolved gasses. Although correlation does not imply causality, a few statistically significant correlations are worth mentioning. In particular the negative relationship observed between chlorophyll a and diversity in our study is in agreement with previous observations that spring blooms and highly productive areas have lower bacterial diversity (i.e., are dominated by a few opportunistic species) (Wemheuer et al., 2014). The negative correlation between bacterial diversity and dissolved oxygen may also be linked to the lower diversity associated with spring bloom conditions, where high primary production leads to high dissolved oxygen concentration in the photic zone. While dissolved oxygen ranged between 130 and 350 μmol/kg, concentrations well above the suboxic levels indicative of strong oxygen minimum zones (OMZ), lower oxygen concentrations below the photic zone partially reflect the respiration of labile organic matter, likely favoring microbial communities with diverse alternative metabolic approaches to acquire energy and nutrients, and could explain the trends seen here (Wright et al., 2012). Several other studies have observed similar inverse relationships between oxygen concentration and bacterial diversity, with the most diverse bacterial communities found at low, but still measurable, oxygen concentrations (Zaikova et al., 2010; Beman and Carolan, 2013; Spietz et al., 2015; Walsh et al., 2015; Wang et al., 2015).

In our study, bacterial diversity on the SS was positively correlated with salinity over a range of 28–36 PSU. Others found a negative relationship between salinity (in the range of 12–33 PSU) and bacterial diversity in a regional surface study from the East China Sea (Wang et al., 2015), while studies targeting estuarine systems (salinity ranges between 0 and 35 PSU) observed a bimodal relationship with high diversity in both the freshest and the saltiest environments in the system, with minimum diversity in estuarine waters (Fortunato et al., 2012; Campbell and Kirchman, 2013). In contrast, a global compilation of oceanic sites (Milici et al., 2016b) found no effect between salinity ranging between 33 and 37.5 PSU and bacterial diversity. There is currently no agreement on the relationship between salinity and bacterial alpha diversity, and most likely the observed variability of this relationship reflects specific regional features rather than global trends in microbial diversity patterns, as is discussed below for our study area (Figure 7).

Although the correlation between alpha diversity and temperature was not as strong as for other environmental parameters in our study, we further examined this relationship because it has been extensively implicated as a determinant of marine microbial diversity patterns in several global surveys (Ghiglione et al., 2012; Sunagawa et al., 2015; Milici et al., 2016b), and temperature is a critically important environmental factor in the context of climate change (Wethey and Lima, 2012). In our study, the relationship between temperature and species richness was complex (Figure 7A), showing a positive correlation between temperature and Chao1 in spring and a negative one in the fall. However, combining spring and fall samples yielded a relationship fitted by a second order polynomial model with maximum species richness at intermediate temperatures of 13°C for the HL transect (Figure 7A). We showed that the intermediate temperature and the highest species richness are both coincident with the shelf break (Figure 7) at the confluence of the two main southward-flowing currents of the SS (Drinkwater and Gilbert, 2004). While several processes could lead to the observation of maximum species richness at intermediate water temperature, our results (Figure 7) support the view that the highest species richness of microbial communities on the SS is associated with the boundary between two water masses at the shelf break. Based on the overlap between on-shelf, off-shelf, and shelf break OTUs, we determined that less than 4% of the OTUs recovered from the shelf break stations were unique, while this proportion was 25% in both the on-shelf and off-shelf regions. This suggested that the increased diversity observed at the confluence of the water masses is largely an additive effect caused by mixing of the distinct on-shelf and off-shelf communities (Supplementary Figure S9). Similar processes may have led to the temperature-diversity relationship displaying highest alpha diversity of microbial communities at intermediate temperatures (∼15°C) observed in some recent global surveys of microbial diversity (Sunagawa et al., 2015; Milici et al., 2016b). However, sampling resolution at the ocean basin or global scale, may currently not be high enough spatially and temporally to provide estimates of alpha diversity of microbial communities at the transition zones where water masses of different temperature and salinity meet.

Our results support the cosmopolitan “environment selects” theory and are in line with other regional marine microbial studies from different locations (Jiang X. et al., 2012; Nguyen and Landfald, 2015; Wang et al., 2015). There was a strong correlation between the community structure of PA and FL size fractions, suggesting that biological interactions may be important in shaping bacterial communities in this region. Strong biological interactions have previously been identified as drivers of global marine community structure (Lima-mendez et al., 2015), and these relationships could be attributed to a number of underlying processes such as cross-feeding, environmental preferences, predation, or symbiosis (Fuhrman et al., 2015; Needham and Fuhrman, 2016).

Of the individual environmental variables tested, temperature and oxygen were the environmental factors most highly correlated with community similarity. Regionally, temperature has been identified as a strong driver of bacterial community structure temporally (El-Swais et al., 2015), and of eukaryotic phytoplankton community structure on the SS at a regional spatial scale (Li and Harrison, 2008; Dasilva et al., 2014), as well as globally (Rusch et al., 2007; Ladau et al., 2013; Sunagawa et al., 2015). Oxygen, previously identified as a key driver of community composition (Stewart et al., 2012; Wang et al., 2015), is extremely important in defining bacterial niches (Wright et al., 2012), and the presence or absence of oxygen in the marine environment can dramatically change the type of microorganisms and the types of metabolic processes occurring (Ganesh et al., 2014; Hawley et al., 2014). The expected trends of decreasing oxygen and increasing temperature associated with climate change (Finkel et al., 2010; Wright et al., 2012; Schmidtko et al., 2017) will have strong implications for the future of bacterial communities on the SS.

Our intensive sampling of the SS allowed us to identify the importance of the shelf break and of regional-scale circulation patterns (Rutherford and Fennel, 2018) on the bacterial community composition observed in spring and fall. The patterns of diversity observed in the bacterial community supported the view that the shelf circulation and its associated frontal zone create a physical boundary that reduces dispersal between off-shelf and on-shelf regions (Figure 7, 9, 10), leading to highest species richness near the shelf break (Figure 7) and distinct bacterial communities on either side (Figure 9 and Table 2). The Bray–Curtis decay curves of three across-shore transects show that similarity between samples decreases significantly at a paired distance of 200 km, especially in the fall (Figure 10). This decrease in the similarity of bacterial communities is most pronounced for the HL transect that covers a 400 km distance across the shelf and with stations distributed evenly across the shelf break (Figure 10B). In contrast, the high similarity of the spring bacterial communities over a distance of >600 km in the near-shore transect is notable. To explain these observations, we propose that the high flow rates of the NSC in the spring resulted in high dispersal rates along on-shelf transect that homogenized the bacterial community. The results also indicate an important role for dispersal and dispersal limitation in our study, as demonstrated by the distance-decay curves of across-shelf sections compared to those of on-shelf and off-shelf regions (Figure 10) that follow the boundary between the major currents on the SS, restricting water exchange across the shelf break. Although our results do not rule out a significant role for environmental selection, the marked decrease in community similarity between on-shelf and off-shelf stations, reflected by the marked decrease in Bray–Curtis similarity at a paired distance of 200 km distance, suggests that a physical, although fluid, barrier develops at the confluence of the prevailing currents (Rutherford and Fennel, 2018), supporting the role of dispersal limitation in the development of distinct on-shelf and off-shelf bacterial communities (Figure 10). In contrast, the results showing no decay of similarity over a much larger distance on the on-shelf and off-shelf regions imply a role for dispersal in homogenizing the bacterial community faster than the environment can modify them. The effects of dispersal were most noticeable in the spring when the NSC is stronger than in the fall (Rutherford and Fennel, 2018). Although the fluidity of oceanic environment limits the degree of dispersal limitation, thereby reducing also the effect of historical processes in shaping overall bacterial community structure (Martiny et al., 2006), dispersal in the ocean is controlled to a large extent by major ocean currents and will therefore be directional, as seen in our results (Figure 10), rather than simply related to the geographic distance (Lindström and Langenheder, 2012).

Our study, conducted at the regional scale with high resolution sampling across strong environmental gradients, covers a region of the temperate Western North Atlantic at the junction of the Arctic Gateway to the northeast and the Gulf of Maine to the southwest, with warm waters of the Gulf Stream bordering to the east. The results of our 2-year study are compatible with reports that the long-term composition of marine microbial communities is stable despite short-term variation (Cram et al., 2015; Fuhrman et al., 2015). Together, these findings indicate that similar microbial assemblages are expected on the SS on an annual basis. However, the sharp delineation of the Prochlorococcus northern distribution range and the continued reports of water temperature increase in the region, as well as earlier onset of spring warming (Wethey and Lima, 2012), point to the importance of time-series observations into the future to allow the detection of early change in the presently predictable microbial community structure in response to environmental climate change shifts such as increased temperatures, ocean acidification, or decreased salinity due to Arctic melt (Hutchins and Fu, 2017).