Sampling occurred in the SO around the WAP and offshore Drake Passage waters. Samples were collected from 21 stations onboard the R/V/I/B Nathaniel B. Palmer in austral spring from September – October 2016 on the cruise NBP16-08 ( Figure 1A ). Stations 1 and 13 were the same location, sampled 14-days apart. There were six stations in the Drake Passage, two in Palmer Arch, two in the Antarctic Sound, five in the Bransfield Strait, and six in the Gerlache Strait. Stations covered by ice included the two Antarctic Sound stations (Stations 10, 11), Stations 2 and 17 in the Gerlache Strait, and Station 6 in the Bransfield Strait. Water was collected from Niskin bottles on the standard CTD rosette (SBE 32 Carousel Water Sampler, SeaBird Scientific, Bellevue, WA, USA). For samples analyzed for diatom and bacterial community composition, most stations were sampled at 5 m, except for Station 4 which was sampled at 7 m and Stations 10 and 11 which were sampled at 25 m. Temperature was measured with an SBE 3plus 6800M sensor (SeaBird Scientific, Bellevue, WA, USA), and conductivity was measured with an SBE 4 – 02/O (SeaBird Scientific, Bellevue, WA, USA) and converted to salinity. Temperature and salinity at corresponding depths to DNA samples were used as metadata for the sequencing data, and CTD downcast observations binned to 1 m were used for the depth profiles of the stations ( Figure S1 ). Some CTD observations were not recorded: 6 – 24 m from Station 1, 6 – 9 m from Station 17, and 6 – 7 m from Station 20.

Macronutrient, trace metal, and siderophore samples were taken from a trace metal clean (TMC) CTD at depths of 20 – 33 m with an average sampling depth of 25 m at all stations for the shallowest surface samples, except for Station 11 which had the shallowest sample at 300 m and was therefore not included in surface sample analysis. The TMC CTD (SBE32 rosette system, Sea-Bird Scientific, Bellevue, WA, USA) was equipped with a Kevlar line and X-Niskin sampler bottles (OceanTestEquipment, Inc., Fort Lauderdale, FL, USA) (Cutter and Bruland, 2012). In the cases where our conventional and TMC CTD sample depths were mismatched, we assumed the concentrations of dissolved trace metals to be uniform in upper 30 m, consistent with more detailed mixed layer studies conducted previously ( Figure S1 ; Hatta et al., 2013; Measures et al., 2013).

Concentrations of chlorophyll a, a proxy for total photosynthetic biomass, were measured by filtering seawater collected from the same depths sampled for DNA sequencing. Seawater (100 mL) was passed through glass-fiber filters (GF/Fs) for 0.6 – 0.8 um particle retention (Cytiva, Marlborough, MA, USA) using a vacuum pump at low pressure of less than -15 in-Hg. Samples were collected in triplicate, except for Station 1 at 5 m which had duplicates. GF/Fs were extracted in 95% ethanol at room temperature in the dark for 12 – 24 hours. Samples were read while on the ship using a 10AU fluorometer (Turner Designs, Inc., San Jose, CA, USA) as both non-acidified and acidified with 10% hydrochloric acid. The 10AU fluorometer was calibrated with a commercially available standard of Anacystis nidulans algae chlorophyll a (Sigma-Aldrich, Inc., St. Louis, MO) Chappell et al. (2022).

For dissolved macronutrient analysis, water samples were collected from the TMC CTD and then filtered through 0.2 µm AcroPak™ membrane filter capsules (Pall Corporation, Port Washington, NY, USA). Filtered water was stored in acid-cleaned polypropylene (Falcon) tubes in the dark at 4°C until analyzed shipboard; a subset of samples was frozen at -20°C for additional shore-based analysis. Samples for nitrate + nitrite, soluble reactive phosphorus (SRP, ‘phosphate’), and dissolved inorganic silicic acid (dSi) were analyzed within 24-hrs of collection shipboard using a Lachat 8500 QuickChem (Hach, Loveland, CO, USA). Upon return from the cruise, nitrate + nitrite, ammonium, SRP, and dSi were analyzed using a Technicon AutoAnalyzer (AA) II (SEAL Analytical Inc., Mequon, WI, USA). Dissolved nitrate + nitrite and SRP were measured with both Lachat 8500 QuickChem and Technicon AAII and averaged. A Lachat 8500 QuickChem was used to measure dSi, with five stations having replicate measurements on the Technicon AAII. Dissolved inorganic nitrogen (DIN) is reported as the summation of nitrite + nitrate and ammonium. Only eight stations had detectable ammonium measurements, contributing 0 – 1% of the total DIN measured.

Trace metal clean techniques were used for the collection, processing, and analysis of all trace metal samples (Cutter et al., 2017). Water samples for dissolved trace metals were collected from the TMC CTD, gravity filtered through 0.2 µm AcroPak™ membrane filter capsules (Pall Corporation, Port Washington, NY, USA) into acid-cleaned low density polyethylene bottles, acidified to pH 1.8 by adding 0.024 M HCl (Optima, Thermo Fisher Scientific, Waltham, MA, USA), and stored at room temperature until analyzed. The following dissolved trace metals were measured: dFe, dissolved Mn (dMn), dissolved Co (dCo), dissolved Ni (dNi), dissolved Cu (dCu), dissolved Cd (dCd), and dissolved Zn (dZn) dissolved Pb (dPb). All sample analyses were supported by analysis of GEOTRACES and SAFe reference samples (Buck et al., 2018; Buck et al., 2019). Extraction and pre-concentration of dissolved metals were done using a seaFAST-pico system (Elemental Scientific, Omaha, NE, USA), and eluents were analyzed on a magnetic sector Element XR High Resolution inductively coupled plasma mass spectrometer (Thermo Fisher Scientific, Waltham, MA, USA) as detailed previously (Hollister et al., 2020; Burns et al., 2023). Macronutrient and trace metal data used in this analysis and additional methods information is available from NSF Biological & Chemical Oceanography Data Management Office (BCO-DMO) (Buck et al., 2018; Buck et al., 2019; Burns et al., 2023).

Concentrations of solid-phase extractable organic dFe-binding ligands were collected at each station from the TMC rosette, and were concentrated by slowly pumping (18 mL min^-1) 10 L of filtered (0.2 µm) seawater onto solid phase extraction columns (Bond Elut ENV, Agilent Technologies, Santa Clara, CA, USA). Solid phase extraction columns were cleaned with HCl-acidified (Optima, Thermo Fisher Scientific, Waltham, MA, USA) MilliQ water (pH 2) prior to use, and then flushed with three column volumes of clean MilliQ before being activated with methanol (Optima, Thermo Fisher Scientific, Waltham, MA, USA) and rinsed again with MilliQ (Bundy et al., 2018). After seawater was pumped onto the columns, each column was flushed with three column volumes of MilliQ before being stored at -20°C until analysis. Prior to analysis, each column was thawed and eluted with 12 - 14 mL of distilled methanol into TMC 15 mL tubes. Methanol eluents were dried down to approximately 500 µL on a Speed-Vac (Thermo Fisher Scientific, Waltham, MA, USA). Samples were then analyzed via liquid chromatography coupled to inductively-coupled plasma mass spectrometry and electrospray ionization mass spectrometry (LC-ICP/ESI-MS) as described in detail elsewhere (Bundy et al., 2018). For the purposes of this work, only data collected from the LC-ICP-MS step have been included, and this data represents the total concentration of solid-phase extractable Fe-binding compounds in each sample. These total concentrations represent concentrations of total siderophores for Fe.

For DNA sequencing of the diatom and bacterial communities, an average of 3.2 L of surface seawater was passed across a 3 µm, 25 mm polyester membrane filter (Sterlitech, Kent, WA, USA) using a peristaltic pump (phytoplankton size fraction, associated with diatoms). The filtered biomass samples were placed in 400 µL of AP1 buffer (DNeasy Plant Mini kit, Qiagen, Germantown, MD, USA) to minimize DNA degradation before extraction, flash frozen in liquid nitrogen, transported on dry ice to the lab, and stored at –80°C until extraction. A modified version of the DNeasy Plant Mini kit protocol (Qiagen, Germantown, MD, USA) was used to extract the DNA of the diatoms and associated bacteria captured on the filters. The modified version included an additional 1-minute bead beating step and use of a QIA-Shredder column (Qiagen, Germantown, MD, USA) with a final elution in 100 µL: RNase free water (Chappell et al., 2019). A Nanodrop spectrophotometer (Thermo Fisher Scientific Inc., Waltham, MA, USA) was used to assess the extracted DNA for quality and quantity, which determined there was an average of 1.7 ng µL^-1. A negative control for sequencing was prepared by extracting a blank 25 mm 3.0 µm polyester membrane filter. Positive controls for bacterial communities were prepared by extracting a commercially available mock community of microbial cells called the ZymoBIOMICS Microbial Community Standard (Zymo Research, Irvine, CA, USA) on a 25 mm 3.0 µm polyester membrane filter as well as using a standard of pre-prepared DNA from microbial cells of the ZymoBIOMICS Microbial Community DNA Standard (Zymo Research, Irvine, CA, USA) for the following PCR steps.

Associated bacteria were examined with primers targeting the V4 region of the 16S SSU rDNA gene using the 515F forward primer: 5’-GTGYCAGCMGCCGCGGTAA-3’ and 806R reverse primer: 5’-GGACTACNVGGGTWTCTAAT-3’ (Apprill et al., 2015; Parada et al., 2016). With the Illumina adapters added, the following full-length primers (Integrated DNA Technologies, Inc., Coralville, Iowa) were used: forward primer 5’-TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGGTGYCAGCMGCCGCGGTAA-3’ and reverse 5’-GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGGACTACNVGGGTWTCTAAT-3’. Each PCR reaction contained 2 – 8 ng of extracted DNA, 0.2 µM of each primer, 1x Bioline BIO-X-ACT Short Mix (Bioline USA Inc., Taunton, MA, USA), and DEPC-treated water for a total volume of 25 µL. The PCR cycle conditions were based on methods from the Earth Microbiome Project data site (Gilbert et al., 2010), which included an initial denaturation of 3 minutes at 94.0°C, 35 cycles of the following: denaturation for 45 seconds at 94.0°C, annealing for 1 minute at 50.0°C, elongation for 1.5 minutes at 72.0°C, and a final extension of 10 minutes at 72.0°C. The diatom-specific primers targeted the V4 region of the 18S rRNA gene: D512F forward primer: 5’-ATTCCAGCTCCAATAGCG-3’ and D978R reverse primer: 5’-GACTACGATGGTATCTAATC-3’ (Zimmermann et al., 2011). Primers with Illumina adapters were: 5’-TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGATTCCAGCTCCAATAGCG-3’ and 5’-GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGACTACGATGGTATCTAATC-3’ (Chappell et al., 2019). Each PCR reaction contained 2 – 8 ng of extracted DNA, 0.5 µM of forward and reverse primers, 1x Bioline BIO-X-ACT Short Mix, and DEPC-treated water for a total volume of 25 µL. The PCR cycle consisted of an initial denaturation step of 5 minutes at 95.0°C, 32 cycles of the following: denaturation for 1 minute at 95.0°C, annealing for 1 minute at 61.8°C, and extension for 30 seconds at 72.0°C, and a final extension for 10 minutes at 72.0°C. All PCR products were visualized on a 1% agarose gel to confirm amplification of desired sized products. These PCR products were cleaned with KAPA Pure Beads (KAPA Biosystems, Woburn, MA, USA) and visualized again on an agarose gel.

Libraries for both bacterial and diatom amplicons for Illumina sequencing were prepared by the RI Genomics and Sequencing Center (Kingston, RI, USA) by performing a second round of 8-cycle PCR with 50 ng DNA template to attach Nextera XT indices and adapters from the Illumina Nextera XT Index Kit (Illumina, Inc., San Diego, CA, USA) with Phusion High Fidelity PCR Mastermix (Thermo Fisher Scientific, Waltham, MA, USA). Again, PCR products were cleaned with KAPA Pure Beads and visualized on an agarose gel. Quality checks were performed by running certain samples on a BioAnalyzer DNA1000 chip (Agilent Technologies, Inc., Santa Clara, CA, USA). All samples were quantified using a Qubit fluorometer (Invitrogen, Waltham, MA, USA), pooled, and then quantified using qPCR in a LightCycler480 (F. Hoffman -La Roche AG aka Roche, Basel, Switzerland) with the Illumina Kit (KAPA Biosystems, Woburn, MA, USA). Samples were sequenced on the same paired-end 2x250 base pair (bp) MiSeq run using the 500-cycle MiSeq Reagent Kit (Illumina, Inc., San Diego, CA).

After sequencing, samples were demultiplexed, and bacteria and diatom reads were separated by primer sequences. Primers were trimmed from both ends of sequences with Cutadapt v1.15 (Martin, 2011). Reads without either primer sequence or less than one bp after trimming were discarded. Trimmed sequences were inputted into DADA2 v1.16 to determine amplicon sequence variants (ASVs; Callahan et al., 2016). For the 16S reads, the truncation length used was 230 bp for forward reads and 160 bp for reverse, decided based on quality scores of the raw reads, with maxN = 0, maxEE = 2 for both directions, truncQ = 2, and rm.phix as true. Only ASVs that were 253 bp in length were retained for analysis, in order to mitigate compounding differences due to read length. ASVs were assigned taxonomy using the naïve Bayes classifier (Wang et al., 2007) within DADA2, using the Silva SSU NR v138 database trimmed to the same 16S V4 region as the primer sequences using Cutadapt (Quast et al., 2013; Yilmaz et al., 2014; Glöckner et al., 2017). A 100% match of an ASV to the Silva database was required to be called a species in the DADA2 species classifier. Matching 100% to an ASV has been shown to be most reliable way of calling species (Edgar, 2018). After assigning taxonomy to the bacterial 16S V4 reads, only reads belonging to Kingdom Bacteria were retained, and any reads identified as chloroplast or mitochondria were removed. The diatom 18S V4 ASVs were created similarly in DADA2 with truncation length of 230 bp for forward reads and 220 bp for reverse reads, and other settings as reported above. Only ASVs that were 390 – 410 bp in length were retained following expectations for that region (Zimmermann et al., 2011). Taxonomy was assigned using the database pr2 v4.12.0 for 18S sequences, which was also trimmed using Cutadapt to the region of interest (Guillou et al., 2013; del Campo et al., 2018). Only ASVs belonging to Class Bacillariophyta were retained. Stations 1, 12, and 13 had triplicate filters sequenced, and a random picker was used to only continue analysis with one of the triplicates to prevent bias towards those three stations. In total, there were 1,027 bacteria ASVs and 124 diatom ASVs used in subsequent analysis.

Statistical analyses were performed in R v4.0.2 (R Core, T 2020) in R Studio v1.1.456 (RStudio, T, 2016) using the vegan v2.5.6 package (Dixon, 2003) for the test for normality, comparison of means, pairwise comparison between groups, tests of group dispersions, and analysis of similarity (ANOSIM) and ggplot2 v3.3.3 for visualizations (Wickham, 2016). Statistical comparisons of means of environmental parameters by station locations (Drake Passage, Palmer Arch, Bransfield Straight, Gerlache Straight, Antarctic Sound) including temperature, salinity, chlorophyll a concentrations, macronutrients, and trace metals were first assessed for normality with the Shapiro-Wilk test using shapiro.test() in R. If the data distribution was normal, an ANOVA was used by aov() in R, followed by a Tukey Test for comparison of groups with TukeyHSD() in R. If the data distribution was not normal, the non-parametric Kruskal-Wallis test was used with Kruskal.test() followed by a Wilcoxon Test for pairwise comparisons of groups with pairwise.wilcox.test() with a Bonferroni correction (p.adjust.method = “BH”). Only comparisons with an adjusted p value of less than 0.05 are shown. For macronutrients and trace metal analysis, Antarctic Sound was excluded since one of the two stations did not have that corresponding data. The same sample grouping by station locations was tested for group dispersions using betadisper() and permutest() in vegan on the Bray-Curtis distance matrix of the relative abundance of the bacteria and diatom ASVs produced by phyloseq using distance(method = “bray”). ANOSIM was performed using anosim() in vegan. Both were run with 999 permutations and an alpha value of 0.05. NMDS plots were made in phyloseq v1.32.0 using default parameters (McMurdie and Holmes, 2013), with stress values equal to or below 0.1 being acceptable for interpretation.

The function bioenv() in vegan was used to determine the best model formula to explain the variation in the Bray-Curtis distance of the relative abundance of the ASVs using the scaled variables, spearman correlations, and Euclidean metric. This analysis uses multivariate statistics to link BIOlogical (i.e. ASVs) to ENVironmental variables, hence it is referred to as BIOENV. Samples from Station 11 in the Antarctic Sound were excluded since they had no corresponding environmental metadata. The following parameters were inputted: temperature, salinity, DIN, SRP, dSi, dFe, dMn, dCo, dNi, dCu, dCd, dZn, and dPb. A second analysis was done including the aforementioned variables along with siderophore concentrations for stations where siderophores were measured (all stations but 7, 11, 14, 17, and 18). For both the diatom and bacteria ASV data sets, the selected BIOENV model parameters were used in the formula for canonical analysis of principal coordinates (CAP) ordination, used with method = “CAP” within the ordinate function of phyloseq, along with the Bray-Curtis distance matrix of the relative abundance of the ASVs (McMurdie and Holmes, 2013).

Relationships between diatom and bacteria ASVs were inferred using sparse inverse covariance estimation for ecological association inference (SPIEC-EASI), which accounts for the compositional nature of amplicon data and the high number of ASVs present in a typically low number of samples (Kurtz et al., 2015). SPIEC-EASI networks were performed within the R package spiec easi v1.1.0 (Kurtz et al., 2015), which performs acceptable data transformation of compositional data sets like ASVs using the centered log-ratio with unit pseudo-count to account for many zero counts. It then estimates relationships between ASVs using neighborhood selection, also known as the MB method (Meinshausen and Bühlmann, 2006), and selects the final network model by using random subsampling by Stability Approach to Regularization Selection or StARS (Liu et al., 2010). The nodes are the ASVs, and connections suggest a linear relationship. A positive linear relationship may indicate a possible mutualism between two ASVs (Steele et al., 2011). Unconnected nodes are likely ASVs which are independent from each other. The predictive power and reproducibility of the network is assessed with a desired stability threshold of 0.05. Networks were run as two groups: 1.) Drake Passage and 2.) Gerlache Strait and Bransfield Strait stations. A subset of the total ASV dataset was selected for a more simple viewing of the networks and to minimize heavy computational load required for more ASVs, a strategy employed by other studies (Kurtz et al., 2015; de Sousa et al., 2019). The top ca. 10% of bacteria ASVs (100 ASVs) and the top ca. 20% of diatom ASVs (25 ASVs) by relative abundance were used. For the Gerlache Strait and Bransfield Strait network, the lambda minimum ratio was set to 1^-2, the nlambda was set to 200, and 50 replicates were run. These same settings were used for another network of all 124 diatom ASVs across all stations, and for the network of the 292 Gammaproteobacteria ASVs across all stations the nlambda was lowered to 20. Because the Drake Passage network only had six samples, the parameters were adjusted to a lambda minimum ratio of 1^-1, nlambda of 10, and 5 replicates with a subsample ratio of 0.5. The R package igraph v1.2.6 was used to visualize the networks (Csárdi and Nepusz, 2006).

The NBP16-08 cruise sampled a naturally occurring gradient of nutrients and trace metals. Dissolved Fe concentrations near the continent ranged from 3.27 – 9.28 nM and were lower in the offshore Drake Passage waters (0.13 – 0.55 nM) ( Figure 1B and Table 1 ). There was a statistically significant difference between the dFe concentrations in Drake Passage stations and in the Bransfield Strait and Gerlache Strait stations ( Table 2 ). Other dissolved trace metal concentrations were also significantly lower in the Drake Passage compared to the Bransfield Strait and Gerlache Strait regions, including dMn, dCo, dCu, and dZn ( Table 2 ). For dMn, the Palmer Arch stations were more similar to the Drake Passage Stations than the Bransfield Strait stations, with macronutrients (DIN, SRP, and dSi) following the same pattern of lower concentrations offshore ( Table 2 ). The dFe to macronutrient ratios (dFe : DIN, dFe : SRP, dFe: dSi) were also significantly higher in the Bransfield Strait and Gerlache Strait than in the Drake Passage. The Palmer Arch dFe: dSi was significantly different from the Bransfield Strait ( Table 2 ). There was no significant difference in dNi, dPb, or siderophore concentrations between Drake Passage and the more nearshore stations ( Table 2 ). Siderophore concentrations ranged from 0.1 – 7.2 pM, with Drake Passage Station 4 being the highest followed by the Gerlache Strait Station 19 as the second highest ( Table 1 ). Overall, there was no apparent geographic pattern in siderophore concentration distributions. For example, siderophore concentrations did not correlate with dFe or other nutrients, and they did not have any clear onshore to offshore trends. They also varied in concentration with repeat occupations of the same sampling location, such as in the Drake Passage ( Table 1 ).

Throughout the cruise, low concentrations of chlorophyll a were observed which slightly increased with time over the duration of the cruise ( Figure 1C ) and did not change with depth ( Figure S1A ). There was no significant difference in chlorophyll a across sampling locations (Kruskal-Wallis, P = 0.067). The average chlorophyll a concentration across all stations was 0.31 ± 0.21 µg L^-1. At the Drake Passage stations, chlorophyll a ranged from 0.25 µg L^-1 to 1.03 µg L^-1 ( Table 1 ). Surface water temperature stayed within a three-degree range during the cruise from -1.8 to 1.6°C and was not significantly different between station location. The warmest surface temperature (1.6°C) was recorded at 25 m at Station 10 in the Antarctic Sound. From the CTD downcast, it was observed that there was a 3°C increase at 25 m at this station compared to its other depths which was unlike the other Antarctic Sound station ( Table 1 and Figure S1B ).

There was a significant difference in bacterial community composition by sampling location (Antarctic Sound, Bransfield Straight, Gerlache Straight, Drake Passage, and Palmer Arch; p = 0.001) as indicated by ANOSIM and visualized by the NMDS plot ( Figure S2 ; NMDS stress = 0.1053). For the bacterial communities, there were no significant differences in group dispersions for sampling location (p = 0.19, df = 19), indicating results from ANOSIM are reliable. Similar to the geographic separation in bacterial community composition data, ANOSIM also indicated a significant difference in diatom community composition by station location (p = 0.001) as displayed by the NMDS ( Figure S2B ; NMDS stress = 0.0931). For the diatom communities, there were also no significant differences in the group dispersions for sampling location (p = 0.56, df = 19). Interestingly, the diatom assemblage at Station 2 in the Gerlache Strait was most similar to the offshore assemblages ( Figure S2B ). This was unlike the bacterial community at Station 2 which was most similar to the other Gerlache Strait samples ( Figure S2A ).

The significant separation in bacterial and diatom communities by sampling location indicated potential underlying environmental differences in these locations, which were further explored in the BIOENV and CAP analyses. For the diatom ASVs, BIOENV determined that temperature, SRP, dSi, dMn, dCu, dZn, and dPb, with a correlation of 0.7591, were appropriate to use for subsequent analysis. When BIOENV was re-run to include siderophore concentrations in the parameters assessed, siderophores did not influence the model. Thus, we chose to use the model that included 20 stations in further analysis of the diatom communities. The bacteria ASVs were analyzed separately, and it was determined that SRP, dMn, dCo, and dCu, with a correlation of 0.7644, were appropriate to use. For the same approach to diatom analysis, the model that included 20 stations of bacteria ASVs was used in subsequent analysis.

The environmental variables resulting from BIOENV analysis, which were used in the CAP analysis, are included on the CAP plot to visualize the major contributing variables to each dimension ( Figure 2 ). For the bacterial communities, 37.9% of variation was accounted for by Dimensions 1 and 2, and samples grouped by location ( Figure 2A ). For the diatom communities, Dimensions 1 and 2 combined on the CAP plot accounted for 57.9% of variation, and samples also grouped by location ( Figure 2B ). As observed in the NMDS analysis, the diatom assemblage from the Gerlache Strait grouped closer with the Drake Passage samples ( Figure 2B ). The diatom community at Station 10 in the Antarctic Sound was influenced by temperature and was separate from the other stations ( Figure 2B ). For both diatom and bacterial communities, changes in dMn, dCu, and SRP were important contributors to Dimension 1, driving the distinct grouping of the near continent stations ( Figure 2 ). For diatoms, Dimension 1 was also driven by dPb changes, especially in the offshore stations ( Figure 2B ), in contrast to bacterial communities where BIOENV analysis deemed dPb unimportant. Additionally, dSi and dZn contributed to the variance in the diatom community analysis, especially for samples from the Bransfield and Gerlache Straits ( Figure 2B ). The dCo changes contributed to Dimension 1 for the bacterial communities, especially in the variance of the Gerlache Strait samples, but did not contribute to the diatom community analysis ( Figure 2 ).

The average number of sequences initially yielded from the sequencer was similar between the environmental samples (n = 27; 52381 ± 10706) and mock community samples (n = 2; 51011 ± 3154) for the bacteria 16S sequencing with the negative controls having fewer reads (n = 2; 1930 ± 886). After the quality control steps in DADA2 (filtering, denoising, and removing chimeras), there was an average of 42556 ± 8608 reads in the environmental samples (n = 27), 27749 ± 867 reads in mock community samples (n = 2), and 1564 ± 655 reads in the negative controls (n = 2). After processing into ASVs, the mock community samples had far fewer average sequences (n = 2; 27749 ± 867) than the environmental samples (n = 27; 41499 ± 8326) while the negative controls had the least number of sequences similar to their starting average (n = 2; 1531 ± 643). In the bacterial positive controls, the ASVs recovered matched the genera present in the mock communities. There were a smaller number of ASVs present in the positive controls which did not match the starting genera and instead matched a Gammaproteobacterium and chloroplast origin. From the environmental samples amplified with the diatom 18S primers, there were about half as many as raw sequences (n = 27; 24642 ± 5512) resulting in 19521 ± 4605 reads after quality control in DADA2. After ASV processing, there was an average of 19378 ± 4575 sequences per sample (n = 27).

Diatoms were the prevalent phytoplankton taxa across all stations as visually inspected by light microscopy and FlowCam fluid imaging ( Figure S3 ). Diatom ASVs recovered from high-throughput sequencing were grouped together by genus ( Figure 3 ). There were some diatom genera that were found at most stations and considered cosmopolitan taxa within our sampling region and some genera that were only found in certain locations ( Figure 3 ). Diatoms observed included all major classifications of raphid pennate, araphid pennate, radial centric and multipolar centric diatoms. Centric diatoms in the Thalassiosira genus and the raphid pennate diatoms in the Fragilariopsis genus were the most well-represented diatoms by the relative abundance of 18S rDNA at every station ( Figure 3 ). Fragilariopsis was best represented by the relative abundance of the 18S rDNA in the nearshore stations of Bransfield and Gerlache Straits and one Palmer Arch station ( Figure 3 ). Other diatom genera present in the 18S rDNA data at every station included Porosira, Stellarima, Chaetoceros and Corethron ( Figure 3 ). Shionodiscus, Asteroplanus, and Actinocyclus were present in nearly every station as well ( Figure 3 ). The Drake Passage stations were represented by communities that included Thalassiothrix, Pseudo-nitzschia, Hemiaulus, Rhizosolenia, Minidiscus, and Asteromphalus ( Figure 3 ). Many of these taxa in the Drake Passage stations were also present in stations in the Gerlache Strait. Notably, there were diatom ASVs which could not be identified at the genus level, especially one ASV that likely belongs to a raphid pennate diatom and represents a portion of the % relative abundance ranging from 1.2 – 12.8% with an average of 6.9% at the offshore stations (“Raphid pennate ASV2” in Figure 3 ).

Known groups of diatom-associated bacterial taxa were well represented in terms of relative 16S rDNA abundance in the associated bacterial communities, including Proteobacteria and Bacteroidetes (Amin et al., 2012). Across all stations in the phytoplankton size class, Proteobacteria were the most well-represented by relative abundance of the 16S rDNA gene along with Bacteriodota ( Figure 4A ). Within the Proteobacteria group, the class Gammaproteobacteria dominated the relative abundance while Alphaproteobacteria were also present ( Figure 4B ). There were several orders of Gammaproteobacteria that co-occurred at all stations, including Alteromonadales, Cellvibrionales, Oceanospirillales, and Thiomicrospirales ( Figure 4C ). Also represented were members of Bdellovibrionota, Planctomycetota, and Verrucomicrobiota. ASVs of Bdellovibrionota occurred with more frequency in the samples near landmasses ( Figure 4A ).

SPIEC-EASI (Kurtz et al., 2015) network analysis of the potential relationships between the diatom taxa and associated bacteria included six stations in the Drake Passage in one network and six stations each from the Gerlache and Bransfield Straits in the other network ( Figure 5 ). In the diatom-bacteria network of the Drake Passage stations, there were 61 positive edge connections (network stability = 0.05) with nine interconnected clusters (subnetworks) that contained both diatoms and bacteria in cross-kingdom connections ( Figure 5A , subnetworks #1 - 9). One subnetwork had 14 taxa interconnected but most had six or fewer connected ( Figure 5A ). In the largest subnetwork (#6) of the 14 connected ASVs, the diatom Chaetoceros debilis acted as a hub connected to members from Bacteroidia NS9, Cryomorphaceae NS10, and Alphaproteobacteria OCS116, while there was a single connection between the diatom Thalassiothrix longissima to a bacteriodetes NS9 marine group member ( Figure 5A ). In subnetwork #8, Thalassiosira minima was connected to the gammaproteobacterium Paraglaciecola sp. and the Planctomycetes Blastopirellula sp. ( Figure 5A ). Small subnetwork #9 had Thalassiosira tumida connected to a gammaproteobacterium Pseudohongiella sp. and an alphaproteobacterium Ascidiaceihabitans sp. while also connected to another diatom Shionodiscus ritscheri ( Figure 5A ). Two very small subnetworks contained Thalassiosira sp., one with it connected to Bacteroidia Lewinella sp. (#7) and another with it connected to NS6 and a member of Nitrincolaceae (#1, Figure 5A ). Other diatoms were also connected to members of Bacteroidia, including Hemiaulus sinensis to an NS9 member and Polaribacter sp. (#2, Figure 5A ). In subnetwork #5, Chaetoceros neogracilis was connected to Bacteroidia Fluviicola sp. and Cylindrotheca closterium was connected to the gammaproteobacterium Alteromonas australica ( Figure 5A ). The diatom Asteromphalus sp. was connected to a member of Magnetospiraceae in #4 and in subnetwork #3, the diatom Actinocyclus curvatulus was connected to a member of NS9 and the diatom Porosira glacialis was connected to a Nitrospinia member of LS-NOB ( Figure 5A ).

The Gerlache and Bransfield Straits subnetworks were larger, all with 11 – 16 diatom and bacteria taxa connected ( Figure 5B ). Overall, there were 102 positive edges (network stability = 0.05) with 4 subnetworks of diatom and bacteria connected ( Figure 5A ). In subnetwork #10, the gammaproteobacterium Paraglaciecola sp. was connected to two diatoms, Stellarima microtrias and Shionodiscus ritscheri, with Stellarima microtrias also connected to other gammaproteobacterium Glaciecola sp. and diatom Thalassiosira tumida ( Figure 5B ). The majority of subnetwork #11 was diatom taxa, but Thalassiothrix longissima was connected to gammaproteobacterium Alteromonas australica ( Figure 5B ). In contrast, subnetwork #12 was mostly bacteria but the one diatom Chaetoceros neogracilis was connected to Bacteroidia Fluviicola sp. ( Figure 5B ). The diatom Asteroplanus karianus was connected to a member of Cryomorphaceae and the diatom Thalassiosira minima was connected to a member of NS9 in subnetwork #13 ( Figure 5B ).

When the networks were examined by separate Kingdoms, for the Gammaproteobacteria, there were 610 positive connections (stability threshold = 0.05). For the correlation network, there was a core interconnected group of ASVs at the stations with many connections ( Figure S4 ). The diatom community was strikingly less interconnected ( Figure S5 ) and there were only 68 positive connections in the diatom network (stability threshold = 0.05).

The SO surrounding the WAP is a well-known high-nutrient low-chlorophyll area in the global ocean (Moore et al., 2002). Overall, our stations grouped primarily into offshore Drake Passage stations and those corresponding to Bransfield and Gerlache Strait locations, which allowed us to sample across a wide range of macronutrient and especially trace metal concentrations. In particular, dFe concentrations ranged from 0.13 to 9.3 nM across our study region and, based on shipboard incubation experiments conducted in parallel to our field work, dFe was the primary limiting nutrient in the Drake Passage, but was not limiting inshore (Burns et al., 2023). The low dFe and dMn concentrations offshore also suggest a limited influence from shelf sediments at our Drake Passage sites in the spring, as has been observed previously in the austral winter (de Jong et al., 2012; Hatta et al., 2013). In contrast to trace metals and macronutrients, there was no clear geographic pattern with siderophore concentrations. The majority of our siderophore measurements were < 2 pM, with the stations with the highest chlorophyll a with lower (< 1 pM) siderophore concentrations. We sampled very early in the season (austral spring) and photosynthetic biomass was low, perhaps leading to lower in situ siderophore production rates and thus the overall low siderophore concentrations in surface waters. The range of siderophore concentrations we observed is comparable to siderophore concentrations found at sites with comparable dFe : DIN ratios, including in the eastern equatorial Pacific HNLC region (Boiteau et al., 2016) and in the upper mesopelagic at station ALOHA (Bundy et al., 2018). It is also possible that siderophores are cycled rapidly amongst the microbial community, and their concentrations therefore represent a single snapshot in time. Overall, our observed concentration of siderophores was an order of magnitude less than total strong Fe ligand concentrations previously reported from the same region in the SO (Buck et al., 2010; Ardiningsih et al., 2021), which is consistent with previous work showing siderophores comprise less than 10% of the ligand pool that is measured in seawater (Bundy et al., 2018).

With a few exceptions, the diatom and bacteria assemblages generally group by similarity of sampling location: Drake Passage, Palmer Arch, Antarctic Sound, Bransfield Strait, and Gerlache Strait. The Antarctic Sound samples grouped separately from others. Notably, the Antarctic Sound samples were the only ones from 25 m depth instead of 5 m like the other locations, but unlike the other stations, the temperature at Station 10 was warmer at 25 m (1.64°C) as opposed to 5 m water temperatures at all other stations (average -1.56°C). However, at some stations in the Gerlache Strait and Drake Passage, there were also temperatures > 1.00°C between 0 – 25 m ( Figure S1B ). The diatom assemblage present at Station 2 in the Gerlache Strait was more similar to the assemblages in the Drake Passage, and Station 1 in particular, than to the other Gerlache Strait assemblages. This pattern is different for the bacterial community assemblages in which the Station 2 bacterial community was most similar to the other Gerlache Strait stations. The similarity of the diatom community to those in the Drake passage may be due to the northern location of Gerlache Strait Station 2. This location perhaps received an influx of offshore water with that corresponding diatom assemblage, as it is a mixing spot where the Antarctic Circumpolar Current and Antarctic Peninsula Coastal Current converge (reviewed in Moffat and Meredith, 2018).

The WAP diatom assemblages as assayed by 18S rDNA community composition were comprised of several different genera and included Thalassiosira and Fragilariopsis at all stations. Fragilariopsis spp. have previously been shown to be the dominant diatoms observed in spring WAP assemblages, along with the diatoms Chaetoceros spp. and Navicula spp., as determined from scanning electron microscopy (Pike et al., 2008; Annett et al., 2010). From previous 18S sequencing of diatom communities in the WAP, Fragilariopsis was also found to be the most abundant sequence (Lin et al., 2017). In Marguerite Bay, Antarctica at the long-term site for the Rothera oceanographic and biological Time Series (RaTS), Thalassiosira and Chaetoceros are abundant as observed using DNA sequencing, with fewer sequences belonging to Stellarima microtrias and unknown pennate diatoms in austral summer (Rozema et al., 2017). Chaetoceros spp. and Navicula spp. were also present in our data set although at lower relative 18S rDNA abundance than Thalassiosira and Fragilariopsis. Interestingly, at the offshore stations and Station 2 in the Gerlache Strait there was a portion of ASVs which represented 11% of the ASVs at those stations that could not be taxonomically assigned to the genus level of known diatoms. This may be a limitation of the sequences present in the taxonomic database that might lack representation of diatoms present in the offshore area. Future culturing and sequencing efforts of diatom isolates from the Drake Passage as well as in-situ visual identification through microscopy may be necessary to further understand the potential novel diversity of diatoms in the region.

The members of the diatom and the associated bacterial communities present at these stations were likely driven by the observed gradients in macronutrients and trace metals, especially since SRP, dMn, and dCu seemed to influence both groups based on our multivariate analysis (BIOENV). Low dMn has been shown to co-limit SO diatom growth with dFe (Browning et al., 2021), which results in growth inhibition and decreased carbon production (Pausch et al., 2019). We found that dMn was an important environmental variable explaining the variance in the in situ community composition of diatoms and associated bacteria, providing evidence that dMn may be an important driver of bacteria and/or diatom growth in this region. Dissolved Pb was identified as a variable that explained a significant fraction of the variance in only the diatom community composition. Although not a bioactive element, dPb may adsorb onto phytoplankton biomass (Chuang et al., 2014) and may vary depending on the surface area and morphology of diatoms. Higher dPb concentrations were associated with lower chlorophyll a in samples from the Drake Passage, perhaps reflecting less sorption to cells and the negative correlation of dPb concentrations in the CAP analysis ( Figure 2 ). Finally, dCo was a key driver of the variance in bacteria community composition. Vitamin B₁₂ requires Co (reviewed in Twining and Baines, 2013), and members of Oceanospirillales were commonly observed at stations and have been found to contribute to vitamin B₁₂ synthesis, likely assisting in relieving vitamin stress on the co-occurring diatom communities in the SO (Bertrand et al., 2015). Together, our data suggest that there are distinct diatom communities in the austral spring in the WAP: those who persist under lower trace metals and dSi in some offshore stations and those that require higher concentrations of both trace metals and dSi that are found near the continent. Likewise, there are also distinct associated bacterial communities that may provide specific benefits to the diatoms living in these different nutrient regimes.

When the connectedness of diatom and plankton-associated bacteria communities were visualized, some interesting patterns emerged, suggesting real-time mutualisms that have been observed in other environments (Steele et al., 2011) may also be occurring in this region. Bacteroidetes members were commonly found at every station in the associated assemblage in our study and were connected to diatoms in the network analysis. Genomes of Bacteroidetes members have previously been found to contain siderophore biosynthesis genes indicating this group’s potential for contributing to the overall siderophore pool in the SO (Delmont et al., 2015). The Bacteroidetes Fluviicola sp. was connected to the same diatom, Chaetoceros neogracilis in each network, regardless of station group. It was also connected to the diatom Cylindrotheca closterium in the Drake Passage network. Bacteroidetes can be beneficial to diatom growth (Amin et al., 2012), and this beneficial interaction seems to extend to multiple diatom taxa in the SO. Interestingly, in both networks, there are diatom nodes that are connected to bacterial nodes not identified at the genus level, which may indicate the importance of uncultured bacteria taxa in mutualisms with diatoms. We did not observe other connections to other genera of bacteria that were confirmed siderophore-producers in the subnetworks.

Overall, there was high diversity in the type of bacteria directly connected to diatom nodes in the networks including members of Gammaproteobacteria, Alphaproteobacteria, Bacteriodetes, Planctomycetes, and Nitrospinota. Typically, each pairing between a diatom and bacteria ASV was unique in the networks. For diatom species that are cosmopolitan in the Drake Passage and Gerlache and Bransfield Straits, their interactions with bacteria are distinct, indicating these partnerships may confer unique adaptations to each environment. For example, across both the nearshore and offshore networks, Thalassiosira spp. were connected to different bacteria, implying there are unique diatom-bacteria interactions in the WAP that form dependent on environmental conditions. Since different species or strains within the Thalassiosira genus correlate with different bacteria, this may indicate species-specific interactions even within one cosmopolitan genus in the SO. Interestingly, one strain of Thalassiosira rotula in laboratory experiments predictably recruits similar bacterial symbionts to its surface from a variety of bacterial inoculum from a similar environment (Mönnich et al., 2020). It may be that WAP bacterial seed populations are distinct between offshore and nearshore environments, or it may be that these specific associations formed in response to the environmental changes regardless of seed population. We may be examining a broader view of the diatom-bacteria interactions occurring in situ within the natural laboratory of changing oceanographic properties than has previously been observed in laboratory experiments.

The higher degree interactions betweenthe Gammaproteobacteria may indicate potential mutualisms between bacteria in the environment. Overall, bacteria seem to be interacting with each other more consistently across the stations. For example, Gammaproteobacteria ASVs were connected to other ASVs in one large subnetwork that contains almost every Gammaproteobacteria ASV. These connections may suggest metabolic partnerships occurring between bacteria in the environment. In contrast, the diatoms were less interconnected and thus unlikely to participate in direct mutualisms with other diatoms.