Summer comes to the Southern Ocean: how phytoplankton shape bacterioplankton communities far into the deep dark sea

During austral spring and summer, the coastal Antarctic experiences a sharp increase in primary production and a steepening of biotic and abiotic gradients that result from increased solar radiation and retreating sea ice. In one of the largest seasonally ice-free regions, the Amundsen Sea Polynya, pelagic samples were collected from 15 sites during a massive Phaeocystis antarctica bloom in 2010/2011. Along with a suite of other biotic and abiotic measurements, bacterioplankton were collected and analyzed for community structure by pyrosequencing of the 16S rRNA gene. The aims were to identify patterns in diversity and composition of heterotrophic bacterioplankton and to test mechanistic hypotheses for explaining these differences along variations in depth, water mass, phytoplankton biomass, and organic and inorganic nutrients. The overall goal was to clarify the relationship between primary producers and bacterioplankton community structure in the Southern Ocean. Results suggested that both epipelagic and mesopelagic bacterioplankton communities were structured by phytoplankton blooming in the euphotic zone. As chlorophyll a (chl-a) increased in surface waters, the abundance of surface bacterioplankton increased, but their diversity decreased. Similarity in bacterioplankton community composition between surface-water sites increased as the bloom progressed, suggesting that algal blooms may homogenize surface-water bacterioplankton communities at larger spatial scales. Below the euphotic zone, the opposite relationship was found. Mesopelagic bacterioplankton diversity increased with increasing chl-a in the overlying surface waters. This shift may be promoted by several factors including local increase in organic and inorganic nutrients from particles sinking out of the euphotic zone, an increase in niche differentiation associated with the particle flux, interactions with deep-dwelling macrozooplankton, and release from competition with primary producers. Additional multivariate analyses of bacterioplankton community structure and nutrient concentrations revealed distinct depth horizons, with bacterioplankton communities having maximum alpha and beta diversity just below the euphotic zone, while nutrient composition gradually homogenized with increasing depth. Our results provide evidence for bloom-driven (bottom-up) control of bacterioplankton community diversity in the coastal Southern Ocean and suggest mechanisms whereby surface processes can shape the diversity of bacterioplankton communities at great depth.


INTRODUCTION
In seasonally ice-free waters of the Southern Ocean, such as the Amundsen Sea Polynya (ASP), seasonal blooms of phytoplankton provide the main source of organic matter. Fixation and export of atmospheric carbon into the ocean interior occur via vertical mixing, partial decomposition, and eventual sinking of photosynthetically produced organic material. Polar oceans host an active Yager 2007, Williams et al. 2016) and diverse (Yager et al. 2001, Ghiglione et al. 2012, Delmont et al. 2014, Kim et al. 2014, Richert et al. 2015 bacterioplankton community. At any given location and time in the environmentally patchy ASP (Yager et al. 2012, Mu et al. 2014, several factors could play a role in shaping the bacterioplankton community, including interactions with the phycosphere (Delmont et al. 2014, Seymour et al. 2017, niche differentiation associated with the particle flux or marine snow (Azam 1998, Kiørboe and Jackson 2001, Azam and Malfatti 2007, and the bioavailability of particulate organic carbon , Williams et al. 2016) and dissolved organic compounds Connelly 2015, Dinasquet et al. 2017), which represent potential substrates for heterotrophic bacteria. Other factors may include irradiance (Bryant andFrigaard 2006, G omez-Consarnau et al. 2007) as well as interactions across trophic levels, in particular the local extent of primary productivity (Arrigo and van Dijken 2003, Alderkamp et al. 2012, Schofield et al. 2015 and macrozooplankton . Additionally, physical gradients that shape hydrography and separate water masses based on temperature-salinity signatures can also control the distribution of microbial communities (Galand et al. 2010, Agogu e et al. 2011, Alonso-S aez et al. 2011, Ghiglione et al. 2012, Hamdan et al. 2013. Three major water masses were present in the ASP at the time of sampling (Yager et al. 2012: Antarctic Surface Water (AASW), Winter Water (WW), and modified Circumpolar Deep Water (mCDW; Randall-Goodwin et al. 2015).
The composition of the bacterioplankton community within the ASP varies with depth (Delmont et al. 2014, Kim et al. 2014, Richert et al. 2015. A major portion of the bacterioplankton community in the seasonal surface waters of the ASP are light-adapted, fast-growing copiotrophs, including members of the Flavobacteria, Polaribacter, Gammaproteobacteria SAR92, and Oceanospirillaceae. The bacterioplankton communities within deeper mesopelagic waters are comprised of a more diverse heterotrophic community including numerous representatives of, for example, the classes Flavobacteriia, Alphaproteobacteria, and Gammaproteobacteria (Delmont et al. 2014, Kim et al. 2014, Richert et al. 2015. Metagenome data demonstrate that an oligotrophic life strategy is dominant in the ocean's free-living microbial populations (Lauro et al. 2009).
At the time of sampling (13 December 2010-15 January 2011), the polynya was characterized by the initiation and buildup of a massive phytoplankton bloom consisting mostly of the colonial microeukaryote Phaeocystis antarctica ), and the downward particle flux was mainly phytodetrital aggregates , consistent with the relatively low numbers and subsurface maxima of macrozooplankton observed during ASPIRE . The general goal was to identify how the local environment, in particular the local phytoplankton biomass, acted upon the composition and diversity of heterotrophic bacterioplankton communities during the ephemeral phytoplankton bloom characteristic of the coastal Antarctic.
The first question was how the algal bloom would affect bacterial community diversity at the surface. In their review, Horner-Devine et al. (2004) cite multiple lines of evidence for how bacterial diversity varies with primary productivity in ways similar to plants and animals, with high productivity increasing homogeneity and reducing diversity, especially for some taxonomic groups. In the ASP, bacterial biomass and activity increases with phytoplankton biomass (Williams et al. 2016), but bacterial carbon adds up to only about 1-2% of algal carbon, with microzooplankton and zooplankton biomass adding up to about 5% . Thus, the massive bloom of a near-monoculture of Phaeocystis antarctica dominates the available substrate for bacteria. Such substrate homogeneity might reduce bacterial diversity. Competition with phytoplankton for nutrients (iron in particular) might also limit heterotrophic bacterial diversity -but not abundance-by favoring a bacterioplankton community composed primarily of opportunistic or mutualistic taxa with high-affinity siderophores (Reid et al. 1993, Barbeau et al. 2001. Alternatively, epipelagic bacterioplankton communities could simply represent stochastic subsets of existing communities, in which case no relationship would be observed between bacterial community diversity and phytoplankton productivity. The second question was about the fate of the bloom and its potential impact on the mesopelagic bacterial community structure. The sedimenting flux of detritus is known to be an important factor structuring bacterioplanktonic communities below the euphotic zone (Pelve et al. 2017), and most of the ocean's surface-generated organic matter undergoes recycling in the mesopelagic ocean (Buesseler et al. 2007, Buesseler and Boyd 2009, Giering et al. 2014. Carbon budgets suggest, and measurements confirm, that significant mesopelagic bacterial activity occurs in the ASP , Williams et al. 2016) below the bloom. Here, we asked whether the deep bacterial community in the ASP shift in response to the sinking bloom, or if it was it poised from the start and steady throughout the season. A related question was whether or not the arrival of the sinking particles increases diversity.
To address these questions, physical characteristics and planktonic communities were sampled at five depth horizons that covered stations along the entire ASP and its edges during the austral summer of 2010-2011. Stations covered the full range of bloom stages (pre-bloom in the pack ice to high bloom in the central open polynya; Yager et al. 2016). Correlation and regression modeling and multivariate statistical analyses were applied to explore linkages between environmental gradients and bacterial abundance and diversity. Our results suggest that the phytoplankton bloom clearly impacts bacterioplankton community composition across the surface waters of the polynya and deep into the mesopelagic.

Sampling
Sampling was conducted during the Amundsen Sea Polynya International Research Expedition (ASPIRE) in the late spring and early austral summer (November 2010-January 2011) when the research icebreaker RVIB Nathaniel B. Palmer cruised the Amundsen Sea Polynya (71-75°S, 110-120°W) and its marginal zones (Yager et al. 2012. To the south, the polynya meets the Antarctic continent, surrounded by ice shelves (such as the Dotson Ice Shelf and Getz Ice Shelf) and the Antarctic land mass. The northern border represents the marginal ice zone and pack ice between the polynya and the open ocean. Sampling sites covered all parts of the polynya, including sites close to the continent and ice shelves, within the open polynya, close to the marginal ice zones and underneath the pack ice. Altogether, vertical sample profiles were collected from 15 stations ( Fig. 1).
At each station, probes coupled to an array of remotely controlled sampling bottles were used for parameter acquisition and sampling. This CTD (conductivity-temperature depth probe) rosette included sensors reading depth-resolved profiles of temperature (°C), conductivity (S/m), oxygen (mg/L), chlorophyll a (chl-a) concentration (mg/m 3 , via chl-a fluorescence), and photosynthetically active radiation (%PAR) for each cast (SBE 911; Sea-Bird Electronics, Bellevue, Washington, USA). Sensor profiles were monitored during the lowering of the CTD through the water column. Based on these profiles, five depths were chosen for seawater sampling during ascent of the CTD. Typically, the five depths chosen corresponded to (1) near surface (Horizon 1: H1 surf ); (2) chl-a fluorescence maximum (Horizon 2: H2 cmax ); (3) base of the high-chlorophyll layer, corresponding to the depth at which the fluorescence drops to a negligible level (Horizon 3: H3 csub ); (4) mesopelagic temperature minimum above the halocline (Horizon 4: H4 Tmin ); (5) near-bottom temperature maximum below the halocline (Horizon 5: H5 Tmax ); and (6) bottom water (Horizon 6: H6 bot , collected at a subset of stations). At each depth, seawater was collected in 12-L Niskin bottles attached to a 24-bottle rosette (SBE 32; Sea-Bird Electronics).
Grouping of biotic samples along the depth gradient was done following standard procedures in two alternate and non-exclusive ways, depending on the analysis: (i) by water mass into Antarctic Surface Water (AASW), Winter Water (WW), or modified Circumpolar Deep Water (mCDW) based on their temperature-salinity signature (following Randall-Goodwin et al. 2015) and a fourth category Mix for samples that had temperature/salinity values intermediate between WW and mCDW; or ii) by the fixed factor depth horizons at which the samples were taken (as described in this section): (1) H1 surf , (2) H2 cmax , (3) H3 csub , (4) H4 Tmin (deep Winter Water), (5) H5 Tmax (upper mCDW), and (6) H6 bot (lower mCDW). Note that grouping (ii) represents a more highly resolved classification, with Horizons 1-3 in the euphotic region and Horizons 4-6 in the mesophotic region, and that grouping (i) and some divisions between horizons in grouping (ii) are dependent upon  Table 1; Appendix S2: Table S1 for further details) within and around the Amundsen Sea Polynya, with its ice margins at the time of sampling. Background image from National Snow and Ice Data Center (NSIDC; Haran et al. 2005). halocline and thermocline boundaries and thus have variable depths among sampling stations. The extent to which these classification schemes differ in explanatory power for determining the diversity and structure of heterotrophic bacterioplankton communities is discussed below.

Bacterial community analysis
DNA was extracted from 0.2-lm filtered seawater samples using a phenol-chloroform extraction (Riemann et al. 2000). Prior to extraction, microbial cells were enzymatically digested for 30 min with lysozyme at 37°C followed by an overnight digestion with Proteinase K (both 20 mg/mL; Sigma-Aldrich, Darmstadt, Germany) at 55°C (Bostr€ om et al. 2004). The 16S rRNA genes were amplified using the bacterial primers Bakt_341F (CCTACGGGNGGCWGCAG) and Bakt_805R (GACTACHVGGGTATCTAATCC) with 454-Lib-L adapters and sample-specific barcodes on the reverse primer (Herlemann et al. 2011). Barcodes for specific samples are found in Appendix S2: Table S1. Triplicate 20-lL PCRs for each sample were carried out with 10-70 ng extracted environmental DNA as template, and Phusion Hot Start High-Fidelity DNA Polymerase (Thermo Scientific, G€ oteborg, Sweden) at the thermal conditions: initial denaturation at 98°C for 30 s, followed by 25 cycles of an initial 98°C denaturation for 30 s, subsequent annealing at 50°C for 30 s, and 30-s extension at 72°C. These 25 cycles were followed by a final 7-min extension at 72°C. Triplicate reactions were pooled, purified (Agencourt AMPure XP Kit; Beckman Coulter, Bromma, Sweden), and quantified (PicoGreen; Invitrogen, Stockholm, Sweden). Equimolar amounts of amplicons from each sample were pooled and sequenced by 454 pyrosequencing using Titanium chemistry at the SNP&SEQ SciLifeLab platform hosted by Uppsala University (Sweden).
In order to obtain a list of observed operational taxonomic units (OTUs) suitable for statistical analyses, low-quality sequences were removed from the dataset, and sequencing noise was reduced using AmpliconNoise v1.24 (Quince et al. 2011) with default parameters. Ampli-conNoise implements algorithms that remove PCR and 454-pyrosequencing errors, as well as the tool Perseus, which removes chimeric sequences. Reads that did not carry the exact primer sequence were also removed. The raw sequence data are archived at the European Nucleotide Archive (ENA) under accession no. PRJEB4866. With a length cutoff of 425 base pairs (bp), the remaining reads were processed using the Quantitative Insights Into Microbial Ecology (QIIME) software v1.3 (Caporaso et al. 2010). Sequences were clustered into OTUs at 97% pairwise identity using UCLUST (Edgar 2010). Taxonomic assignments of representative sequences from each OTU were obtained according to the SILVA111 database (Quast et al. 2013) by using the RDP classifier implemented in QIIME (Wang et al. 2007). OTUs that were assigned to the same taxa were not merged. After clustering and taxonomy assignment, the raw data table contained 214,949 reads. Singleton sequences were removed (7894 reads), as well as reads from OTUs that could not be classified (2094 reads), and eukaryotic chloroplasts, which were identified as Cyanobacteria (25,025 reads) and Archaea (1748 reads). The remaining 178,188 reads were used for further analyses.

Bacterial abundance
Samples (1.5 mL) for bacterial enumeration were fixed with EM-grade glutaraldehyde (1% final conc.; Sigma, Darmstadt, Germany), flashfrozen in liquid N 2 , and stored at À80°C. Bacteria were enumerated in a FACSCanto II flow cytometer (Becton Dickinson, Franklin Lakes, New Jersey, USA) after staining the fixed cells with SYBR Green (Invitrogen, T astrup, Denmark; Gasol and del Giorgio 2000). The sample volume was calibrated with fluorescent beads.

Nutrient and organic matter data
Water samples were collected, pre-filtered (0.45 lm), and analyzed for nitrite, nitrate, ammonia, silicate, and orthophosphate using an autoanalyzer onboard the ship (for details, see Yager et al. 2016). All samples for dissolved organic carbon (DOC), total nitrogen (TN), particulate organic carbon (POC), and particulate organic nitrogen (PON) analysis were collected, and analyzed using standard protocols (for details, see Yager et al. 2016).
Chlorophyll a fluorescence: a proxy for phytoplankton biomass and bloom stage Phytoplankton biomass was estimated by Chlorophyll a (chl-a) fluorescence in several ❖ www.esajournals.org different ways. Discrete samples were collected using Niskin bottles. Some were analyzed shipboard with a fluorometer calibrated with spinach standards; others were analyzed using HPLC . Continuous chl-a profiles were also obtained at each station using calibrated in vivo chl-a fluorescence traces from the CTD probe. When using chl-a as a proxy for phytoplankton, it was recognized that photopigment content additionally depends on taxonomy and physiological state of the phytoplankton (Longhurst and Glen Harrison 1989). Particulate organic carbon (POC) was collected from the water column at the same depths as chla. Concentrations of chl-a correlated positively with POC concentrations , with the POC:chl-a ratio in the upper water column ranging from 37 to 115 (median = 44).
For each station, the primary measure of the bloom extent was the maximum chl-a concentration observed at each station, chl-a max. A second measure was also used: integrated chl-a, calculated via numerical integration of chl-a concentration profiles (for further methodological details, see Appendix S1). Integrated chl-a represents chla concentration throughout the depth profile of each station and was positively correlated with chl-a max (Pearson's q = 0.802, P < 0.001; Appendix S1: Fig. S2). The main text presents correlations vs. chl-a max; there are corresponding regression results vs. both chl-a max and integrated chl-a provided in Appendix S2.
During analysis, unusually noisy chl-a traces were observed for Station 5 (Event 5.04; Appendix S1: Fig. S1) that might have been due to fouling of the CTD sensor. Thus, Station 5 was excluded from all regressions involving estimates of integrated chl-a. For further details, see Appendix S1.

Diversity calculations
Operational taxonomic units richness of samples was estimated throughout the water column using rarefaction estimates calculated with estimated_ob-servation_richness.py in QIIME 1.9.1 (Caporaso et al. 2010, Colwell et al. 2012. The mean number of reads per sample was 2481 AE 966 (SD) reads (range 751-5600 reads), so richness results are presented using rarefaction with sample size N = 750. Rarefaction curves for a range of sample sizes are presented in Appendix S2: Fig. S1. See Appendix S1 for further methodological details.
Diversity measures were compared and contrasted along environmental gradients using correlation and regression analyses; Pearson correlation coefficients (q) are presented in the main text and provide corresponding regression results including slope estimates in Appendix S2. Alpha diversity was quantified using both Shannon's H (Shannon 1948) and Simpson's D (Simpson 1949) measures to show that the results are qualitatively robust to our choice of alpha diversity metric. Results in the main text use Shannon's H, and corresponding results using Simpson's D are in Appendix S2. Patterns of bacterioplankton community turnover were examined using delta divergence, which expresses the divergence in diversity between pairs of samples (Scofield et al. 2012).
Beta diversity was estimated using multivariate ordination methods applied to pairwise Bray-Curtis dissimilarity, which account for both presence/absence of OTUs and shifts in their relative abundances (Bray and Curtis 1957), and contrast this with pairwise Euclidean dissimilarity in nutrient levels between samples (Ramette 2007, Legendre andBirks 2012) to examine the extent to which nutrient availability may structure heterotrophic bacterioplankton community diversity. Multivariate homogeneity of group dispersion was also compared to environmental factors that separate the groups ).

Statistical analysis
All statistical data analyses were done in RStudio using R version 3.3.2. A custom script was used (available at https://github.com/douglasgscof ield/pubs/tree/master/Richert-et-al-1) together with scripts for ecological diversity analysis available at https://github.com/douglasgscofield/dispersalDiv ersity (Scofield et al. 2012) to calculate most diversity statistics directly from the OTU table. Bacterioplankton richness and abundance was compared between depth horizons using ANOVA (function aov), and depth horizon means were compared using Tukey post hoc tests (function TukeyHSD). Samples from the deepest depth horizon H6 bot (lower mCDW) were collected from just two stations; for these comparisons, these samples were lumped with those from the H5 Tmax (upper mCDW) depth horizon. For comparisons along environmental gradients, Pearson correlation coefficients (function cor.test) and linear regressions (function lm) were calculated.
For multivariate statistical analysis, the vegan package version 2.4-2 in R was used (Oksanen et al. 2013). The number of reads per taxon was standardized by dividing by the total number of reads per sample. The nutrient data and other environmental parameters were transformed linearly to range between 0 and 1 corresponding to the minimum and maximum values across all collected samples (function decostand, method range). Pairwise dissimilarity matrices were calculated among samples for OTUs (N = 72 samples) using Bray-Curtis dissimilarity with presence-absence standardization (function vegdist). The two OTU samples from H6 bot were included within the overlying H5 Tmax depth horizon for these analyses. Pairwise dissimilarity matrices were calculated among the subset of samples having full seawater nutrient content available (N = 43 samples) using Euclidean distance (function vegdist). None of these samples were assigned to the H6 bot depth horizon. Pairwise dissimilarities were used to calculate group dispersions for OTUs and nutrients within each of the five upper depth horizons (H1 surf -H5 Tmax ) via PCoA and calculation of group-specific centroids and sample-specific distances from the centroid, as implemented in the function betadisper following the method of Anderson (2006). In brief, the centroid of a group is its spatial median, the point that minimizes the sum of distances to all sample-specific points belonging to that group. The dispersion of samples within each group is the distribution of distances of samples from the centroid so calculated. Tests for homogeneity of dispersions were performed with Tukey post hoc tests as implemented in function TukeyHSD.betadisper. Our interest was primarily in analyzing homogeneity of depth horizon-specific dispersions; apart from qualitative comparisons, we do not include analysis of the group-specific centroids per se. For the 43 samples having both OTU and complete nutrient content data, a non-metric multidimensional scaling (NMDS, function metaMDS) was performed with dissimilarity matrices calculated for OTUs and nutrients as described in this section. The NMDS plot for OTUs was annotated using environmental vectors for nutrients and for phytoplankton biomass, bacterial abundance, and depth, and the NMDS plot for nutrients using environmental vectors for phytoplankton biomass, bacterial abundance, and depth (all annotations with environmental vectors used function envfit). The dissimilarity matrices for OTUs and nutrient content were compared among these samples using a Mantel test (function mantel, Spearman rank correlation) with P value calculated via 999 row/column permutations as implemented in the function.

Bacterioplankton community richness and abundance throughout the water column
Bacterioplankton community richness varied among stations (Fig. 2a), with the highest cumulative richness at Station 4 (late spring in the pack ❖ www.esajournals.org ice near the shelf break) and the lowest at Stations 66 and 68 (also in the pack ice, but later in the season). Richness increased with depth ( Fig. 2b-c; ANOVA among depth horizons, F 4,67 = 38.45, P <0.001). It was similar between H1 surf and H2 cmax in the photic zone (Tukey post hoc test, P % 1) and also similar between mesopelagic horizons H4 Tmin and H5 Tmax + H6 bot (Tukey post hoc test, P % 1). Bacterioplankton richness at H3 csub differed from and was intermediate to those in the overlying H2 cmax and the underlying H4 Tmin (Tukey post hoc tests; P <0.001 and P <0.01, respectively). Rarefaction curves showed that the increase in bacterioplankton community richness with depth was robust to the selection of rarefaction sample number (Appendix S2: Fig. S1).

Beta diversity of bacterioplankton to corresponding nutrient distribution
In our multivariate analyses, OTU samples clustered most strongly according to depth (PCoA1 and NMDS; Fig. 6a, c) with a clear separation between the deep horizons H5 Tmax and H6 bot from upper horizons H1 surf to H3 csub . Distinct clusters appeared between H1 surf and H2 cmax clustering together and a separated H3 csub, which correspond to the switch in response of bacterioplankton community alpha diversity to chl-a shown in Fig. 5. Some outlier samples were collected at stations with reduced influence from the phytoplankton bloom, either at the Dotson Ice Shelf front (station 12.01) or beneath the pack ice (station 68.01).
Nutrients clustered as expected according to depth horizon (see also Yager et al. 2016). Samples from the three upper horizons exhibited greater variability but generally lower inorganic nutrients and higher organic compounds. Samples from the two bottom horizons showed less variability but higher inorganic and lower organic nutrient concentrations. H3 csub and H4 Tmin , which both belong to the Winter Water mass, could be separated by nutrients. The H3 csub was characterized by the highest concentration of dissolved organic carbon [DOC] and ammonia [NH 4 + ], with DOC in a range from 0.59 to 1.43 ppm and NH 4 + in a range from 0 to 1.26 lmol/L (Fig. 6c, Appendix S2: Table S5).
Both OTU diversity and nutrients displayed distinct spatial gradients among samples. Both depth horizon. Richness estimates include 95% confidence intervals calculated during rarefaction; see text for further details. (c) Bacterioplankton richness vs. depth horizon. (d) Bacterioplankton abundance vs. depth horizon. Depth horizons in panels (c) and (d) follow the legend in panel (b). Gray box plots show central tendency and quartiles. In panels (c) and (d), the horizontal lines connect depth horizons that did not differ in bacterioplankton richness or abundance, respectively, as determined via pairwise Tukey post hoc tests (P ≤ 0.05); see text for further details.
( Fig. 2. Continued) ❖ www.esajournals.org types of samples separated broadly by depth as shown by the first axes of both OTU diversity and nutrient content in both the beta dispersal and NMDS analyses (Fig. 6a, c, d, f). A positive correlation was observed between Bray-Curtis pairwise distance matrices of sample OTU diversity and sample nutrient content (Mantel test, r = 0.54, P = 0.001).
Variation in nutrient content among samples, however, showed a more linear (or monotonic) relationship with depth than did variation in OTU diversity. Variations (or lack thereof) in the bacterioplankton community cannot be explained solely by the distribution of nutrients. For example, the OTU samples in especially H3 csub separated along a second PCoA or NMDS axis (Fig. 6a, c), whereas there was much less separation between the nutrient samples (Fig. 6d, f). In contrast, the beta diversity along NMDS1 in nutrient content among the H1 surf and H2 cmax horizons (Fig. 6f) was markedly greater than the variation in OTU diversity (Fig. 6c). Nutrient distribution in the surface waters did not co-vary with bacterioplankton diversity, but with bacterial abundance, depth, and chl-a (Fig. 6f).

DISCUSSION
These results provide some insight into how phytoplankton blooms within the surface waters of the ASP structure heterotrophic bacterioplankton communities throughout the water column. The effects are not confined to the surface waters in which phytoplankton and bacterioplankton co-occur, but also continue below the phytoplankton biomass, with more diverse bacterioplankton communities appearing beneath more abundant phytoplankton blooms. One mechanism responsible for this effect may be that the local impact of the phytoplankton cascades downward as detritus particles from the bloom sink into mesopelagic bacterioplankton communities. At greater depths, the cascading effect diminishes as the particle flux declines with remineralization. In contrast to other recent studies (Galand et al. 2010, Agogu e et al. 2011, Alonso-S aez et al. 2011, Hamdan et al. 2013, hydrologic boundaries (water mass identity) were not the best explanation for bacterial community composition. As an example, the lower WW depth horizon (H4 Tmin ) contained a bacterioplankton community more similar to that of the adjacent and subtending mCDW depth horizon (H5 Tmax ) than to that of the adjacent and overlying upper WW depth horizon (H3 csub ; Fig. 6c).

Mechanisms structuring bacterioplankton communities in surface waters
A distinct subset of bacteria was found in the surface waters, dissimilar and less diverse than the underlying mesopelagic community. When local chl-a increased, a decrease in bacterioplankton community diversity occurred (Figs. 3a, 5a) along with a decrease in site-to-site community divergence (Fig. 4), all while bacterioplankton abundance increased (Fig. 3b). These results are consistent with the findings of Hyun et al. (2015), who identified a positive correlation between chl-a concentration and respiration rates in the open polynya, and with Williams et al. (2016) who reported high bacterial production and increasing exoenzyme activity associated with the bloom. All these patterns may be explained by increasing competition as chl-a increases. Bacterioplankton communities in surface waters are structured by local competitive effects that favor specializations for effective heterotrophy in the midst of a highly active phytoplankton community, favoring fast-growing copiotrophs that outcompete slow-growing oligotrophic bacteria. These results are consistent with other community structure data from the region (Delmont et al. 2014, Kim et al. 2014, Richert et al. 2015, Dinasquet et al. 2017.
These productivity-driven patterns in bacterioplankton community diversity and taxonomic composition in surface waters also agree with the findings of Horner-Devine et al. (2003,2004), who recognized taxon-specific interactions between productivity and bacterial richness, inferring that copiotrophs profit from high productivity rates and resource availability while putative oligotrophic and more fastidious microbes decrease in abundance with increasing chl-a. Direct competition for nutrients with phytoplankton is likely another contributing factor, as phytoplankton can be superior competitors vs. bacterioplankton under high nutrient availability with the opposite seen for more nutrient-poor environments (Kirchman 1994, Løvdal et al. 2007.

Cascading effects of chl-a with depth
Bacterioplankton alpha diversity at the H3 csub (base of the high-chlorophyll layer) increases as chl-a in the overlying surface waters increases (Fig. 5b), showing that the immediate influence of the bloom extends well below its occurrence and that its effect on bacterioplankton community diversity is opposite to that seen in the surface waters. Underneath the euphotic zone, downward flux of diagenetically fresh and labile organic matter fuels heterotrophs (Jiao et al. 2010) and offers a variety of niches. Within the ASP, Delmont et al. (2014) differentiated taxa into characteristic size classes such as particleassociated bacteria which form biofilms or may excrete exoenzymes (e.g., Colwellia, Pseudoalteromonas, and Cryomorphaceae) and free-living bacteria that may scavenge small organic degradation by-products (e.g., Piscirickettsiaceae, Pelagibacter, Oceanospirillales). Our results indicate that broad diversity shifts across photic, carbon, and nutrient gradients accompany this taxonomic turnover, and we expect that similar shifts would occur in similar oceanic systems regardless of the participating taxa.   6. Multivariate analyses of bacterioplankton community composition and seawater nutrient composition for 43 samples with both OTU (left column) and nutrient data (right column); see text for further details on samples and calculation of pairwise dissimilarities. Panels (a) and (d): the first two axes of principal components analysis (PCoA) with samples grouped by depth horizon (see text for depth horizon definitions). The depth horizonspecific centroid (calculated as the spatial median) is shown as a solid colored symbol following the legend in panel (d). Gray lines connect each centroid to sample-specific values for that depth horizon, and dashed lines As we descend further into the WW, further community and nutrient gradient shifts appear. The lower WW (H4 Tmin ) hosts a bacterioplankton community dissimilar to that found in the upper WW (H3 csub ), and the shift in bacterioplankton community composition revealed by NMDS analysis is pronounced. This is also shown by our bacterioplankton community beta diversity results. Winter water and surface water have similar high beta diversities among samples (Fig. 6a, b), but this similarity appears to mask fundamentally different mechanisms producing this taxonomic turnover. Beta diversity among surface waters is elevated primarily by lateral site-to-site differences in community composition within surface waters and the subtending WW from which populations in these surface communities are recruited each year (Fig. 4), while beta diversity among WW (H3 csub and H4 Tmin ) samples is also high because of community differences between samples collected at different depths along the strong gradient of WW bacterioplankton community composition (Fig. 6a-c).
Most nutrients do not show similarly strong gradients within WW (H3 csub and H4 Tmin ), with homogenous nutrient composition among deeper samples (Fig. 6f). Silicate, nitrate, and orthophosphate increased gradually with depth, indicating growth-related depletion in the euphotic surface, possibly combined with a downward flux linked to particle sedimentation (Appendix S2: Table S5; Pondaven et al. 2000, Pollard et al. 2006, Paytan and McLaughlin 2007. In contrast, the highest ammonia and DOC concentrations appear in H3 csub , where non-migrating macrozooplankton were most abundant , while depleted to below detection limits in H4 Tmin . This may explain some of the variability in bacterioplankton communities of the WW, as ammonia is the most bioavailable form of nitrogen in these waters (Zehr and Ward 2002).
Impact of bacterial community structure on marine carbon sequestration Mesopelagic bacteria and their organic matter degrading activities are critically important to the fate of sinking photosynthetically derived organic carbon (Azam 1998, Bell et al. 2005. The composition and capabilities of the bacterioplankton community in the mesopelagic WW of the ASP may directly influence the carbon sequestration potential of the phytoplankton bloom occurring in the overlying surface waters (e.g., Mu et al. 2014). On the basis of these results, it may be inferred that the major nutrient gradient underlying the cascading effects is a change in the quality of particulate organic matter (POM) and dissolved organic matter (DOM) in a continuum with depth from labile to recalcitrant as they cascade downward, and that this explains the shift in bacterioplankton communities from below surface water into the underlying mesopelagic deeper waters. Below the zone of recycling, gradual degradation of POM and DOM transits downward and causes an increased recalcitrance of POM and DOM that is accordingly more likely to undergo long-term storage than microbial recycling. Deeper water masses are known to receive such high C:N and C:P recalcitrant organic carbon (RDOM) formed by microbial modification of labile phytoplankton-derived DOM sources (Jiao et al. 2010).
This sinking phytodetritus provides a source of substrate for organisms at depth, and the cycles arising from this process have been collectively termed the biological carbon pump (Ducklow et al. 2001). One general trend in ocean carbon pump efficiency is that high-latitude regions have higher export efficiency (lower delineate the convex hull surrounding the samples for each depth horizon. Panels (b) and (e): depth horizon-specific distances of samples from the centroid in panels (a) and (d), respectively, as a measure of beta diversity of OTUs and nutrient values within each depth horizon. Solid horizontal lines connect depth horizons that did not differ in beta dispersion, as determined via pairwise Tukey post hoc tests (P ≤ 0.05); see text for further details. In panel (e), the dashed line segment indicates that H2 cmax differed from H5 Tmax , but H1 surf did not. Panels (c) and (f): the first two axes of non-metric multidimensional scaling (NMDS) ordination with samples grouped by depth horizon. Overlying the OTU and nutrient ordinations are corresponding gradients of nutrient composition (panel c) and phytoplankton biomass, bacterioplankton abundance, and sample depth (panels c and f).
( Fig. 6. Continued) ❖ www.esajournals.org carbon recycling) than low-latitude regions (Ducklow et al. 2001). Nevertheless, the highly responsive and diverse mesopelagic microbial community we observed is consistent with measurements in Antarctic polynyas showing that most phytodetrital carbon produced during bloom events, representing~85% of the yearly primary productivity, is recycled within the upper 350 m of the water column (Fischer et al. 1988, Ducklow et al. 2008.

CONCLUSION
The massive, recurring phytoplankton bloom in the Amundsen Sea Polynya clearly impacts both the epipelagic and the mesopelagic bacterial community in this key region of the Southern Ocean. The increasing homogeneity of bacterial communities as the algal bloom progresses appears to favor opportunistic heterotrophic taxa, leading to a less diverse bacterial community in the surface waters. In contrast, the bacterial communities in the mesopelagic waters just beneath the bloom become increasingly diverse as the phytoplankton build up and then sink into the subsurface. This sinking phytodetritus provides food for zooplankton and substrate for particle-associated bacteria, which in turn increase the number of biogeochemical niches available at these depths. As the particle remnants sink deeper, their impact on niche availability and bacterial diversity declines, with an increased recalcitrance of POM and DOM that is accordingly more likely to undergo long-term storage than microbial recycling.
As part of ASPIRE, this study provides additional insight to an ecosystem which is undergoing massive changes due to a changing climate. The novelty presented here is the mechanistic view on phytoplankton and bacterioplankton interaction in a polar environment characterized by strong seasonal re-occurring bloom events. Bacterioplankton as decomposers represent the major link between marine primary production from atmospheric carbon and its sequestration into the interior of the ocean.

ACKNOWLEDGMENTS
We are grateful to the captain and crew of the Research Icebreaker Nathanial B. Palmer and the ASPIRE team supporting us during the cruise, and to the Swedish Polar Research secretariat and Raytheon Polar Services who provided logistic support. The study was funded by the Swedish Research Council (grants to SB and LR) and the U.S. National Science Foundation through the ASPIRE project (ANT-0839069). The 454 pyrosequencing was done at SciLife-Lab SNP/SEQ facility hosted by Uppsala University, and some of our bioinformatic computations were performed at the Uppsala Multidisciplinary Center for Advanced Computational Science (UPPMAX). We thank Alexander Eiler for assistance with handling of the raw pyrosequencing data; Shona Wharton, who produced the GIS map; and Sharon Stammerjohn for help with understanding the ocean water masses.