Copiotrophic marine bacteria are associated with strong iron‐binding ligand production during phytoplankton blooms

Although marine bacteria were identified nearly two decades ago as potential sources for strong iron‐binding organic ligands detected in seawater, specific linkages between ligands detected in natural water and the microbial community remain unclear. We compared the production of different classes of iron‐binding ligands, dissolved iron and macronutrient concentrations, and phytoplankton and bacterioplankton assemblages in a series of iron amended 6‐d incubations. Incubations with high iron additions had near complete macronutrient consumption and higher phytoplankton biomass compared with incubations with low iron additions, but both iron treatments were dominated by diatoms. However, we only detected the strongest ligands in high‐iron treatments, and strong iron‐binding ligands were generally correlated with an increased abundance of copiotrophic bacteria, particularly Alteromonas strains. Ultimately, these robust correlations suggest a potential linkage between copiotrophic bacteria and strong iron‐binding ligand production after iron fertilization events in the marine environment.

The concentrations, chemical forms, and spatial/temporal distributions of oceanic iron are important factors shaping the ecology of marine phytoplankton and the overall productivity of marine ecosystems (Boyd et al. 2007). Dissolved iron (dFe) that has been regenerated from less kinetically labile forms, such as particles, is important in sustaining surface ocean productivity in iron-limited regions (Bowie et al. 2001). The mechanisms by which particulate iron is recycled back to dissolved iron and into the wider microbial food web, a process herein referred to as iron remineralization, are largely uncharacterized and represent major unknowns in global iron budgets (Tagliabue et al. 2016). The activity of microzooplankton (Barbeau et al. 2001) and viruses (Strzepek et al. 2005) are known to facilitate iron remineralization, and heterotrophic bacteria can also play a fundamental role (Boyd et al. 2010) as has been observed for other nutrients (Bidle and Azam 1999).
DFe is predominantly bound by organic ligands of mostly uncharacterized composition (Gledhill and van den Berg 1994;Rue and Bruland 1995). The concentrations and conditional stability constants, a measure of iron-binding strength, of natural marine ligands are typically measured using an electrochemical technique known as competitive ligand exchange-adsorptive cathodic stripping voltammetry (CLE-ACSV). Historically, organic ligands have been operationally partitioned into two classes based on their binding affinity for Fe (Gledhill and van den Berg 1994;Rue and Bruland 1995), but additional classes can be specified by varying the concentrations of a well-characterized competing ligand during electrochemical titrations, in a methodology termed Multiple Analytical Windows (MAW) (Bundy et al. 2014(Bundy et al. , 2015. Here, we partitioned organic ligands into three distinct classes in decreasing order of binding strength (L 1 > L 2 > L 3 ) using MAW and novel data processing techniques.
Some portion of the strong organic iron-binding ligands from seawater (Mawji et al. 2008) or marine cultures (Boiteau and Repeta 2015) are siderophores, bacterial secondary metabolites produced as an iron acquisition strategy. Intriguingly, the binding strengths of L 1 ligands measured in the field can be similar to or higher than those of known siderophores, leading to hypotheses that at least some portion of L 1 ligands in seawater may in fact be siderophores Bruland 1995, 1997;Witter et al. 2000). Furthermore, electrochemical studies have demonstrated a positive correlation between L 1 ligand concentrations and biological activity as reviewed in Gledhill and Buck (2012), but the precise biological sources of those ligands are difficult to ascertain through electrochemical methods alone. Ultimately, it is clear that strong L 1 organic ligands are present in marine systems and are probably connected to the metabolic activities of marine bacteria and phytoplankton, but direct functional and mechanistic linkages between L 1 and biological community structure at this point remain elusive. Here, we couple highthroughput 16S rRNA marker gene surveys with MAW CLE-ACSV chemical analysis in incubation experiments in order to explore how phytoplankton and bacterioplankton assemblage composition is connected to the production of different classes of iron-binding ligands.

Oceanographic setting and incubation setup
We collected trace-metal clean seawater for incubation experiments from 35 m depth in August 2012 approximately 250 km from the Southern California coast (33.8798 N, 123.3068 W). We sampled using 5 L Teflon-coated externalspring Niskin bottles on a rosette deployed on nonmetallic hydroline. After sampling, we moved Niskin bottles to a positive pressure clean van and dispensed unfiltered water into an acid-cleaned 50 L carboy. We added 10.0 lmol L 21 , NO 2 3 1.0 lmol L 21 PO 32 4 , and 9.3 lmol L 21 Si(OH) 4 to the dispensed water and divided it between six 2.7 L polycarbonate bottles. Prior to use, we pretreated all macronutrients with Chelex resin to remove contaminating trace metals. We added 1 nmol L 21 FeCl 3 to Low Fe bottles and 5 nmol L 21 FeCl 3 to High Fe bottles then moved all bottles to an ondeck flow-through incubator screened to 30% incident light levels. We sampled bottles at 0 and 144 h (6 d) for dFe, dFebinding ligands, chlorophyll a (chl a), macronutrients (NO 2 3 , PO 32 4 , and Si(OH) 4 ), pigment concentrations, phytoplankton cell counts, and DNA and for only nutrients and chl a at 72, 120, and 144 h (days 3, 5, and 6). We sampled by rapidly moving incubation bottles to a trace metal clean van, subsampling bottles in a laminar flow hood, then returning bottles to the incubator.

Nutrients, pigments, and phytoplankton
We filtered subsamples (0.2 lm Acropak) of incubations, stored the flow through at 2808C, and measured Si(OH) 4 , PO 32 4 , and NO 2 3 (nitrate 1 nitrite) onshore using a Lachat QuickChem 8000 flow injection analysis system. We collected samples for chl a and phytoplankton pigments onto GF/F filters, extracted the chl a filters in acetone for 24 h, and analyzed chl a concentrations fluorometrically onboard. We stored filters for accessory pigments in liquid N 2 until analysis by high performance liquid chromatography onshore. We preserved phytoplankton for cell counts in formalin and counted cells onshore using light microscopy. We classified cells by morphology into categories of Chaetoceros spp., Pseudo-nitzschia spp., other diatoms (>10 lm), dinoflagellates, flagellates (<10 lm), and ciliates.

Dissolved Fe and dFe-binding organic ligands
We analyzed total dFe using flow-injection analysis (FIA) as described by King and Barbeau (2007), and measured the concentration and conditional binding strengths of organic dFe-binding ligands using MAW CLE-CSV methods (Bundy et al. 2014(Bundy et al. , 2015(Bundy et al. , 2016. We resolved ligand classes using the ProMCC software and first attempted to fit our titration data to three different ligand classes in a mass balance optimized speciation model (Omanović et al. 2015). If the model did not converge after 500 iterations, we fit the model again but for only two ligand classes. We characterized ligand classes as L 1 when logK cond

DNA extraction, library preparation, and sequencing
We filtered 100 mL of seawater onto 0.22 lm Sterivex-GV filters (Millipore) without prefiltration and froze filters at 2808C. We extracted DNA onshore using the methods described by DeLong et al. (2006). We PCR amplified the V3-V4 region of the 16S ribosomal RNA gene from our samples using Q5 polymerase (New England Biolabs) and primers, S-D-Bact-0341-b-S-17/S-D-Bact-0785-a-A-21, (Klindworth et al. 2013) modified with Illumina adapters. We sequenced samples using 300 bp paired-end sequencing on an Illumina MiSeq machine running v3 chemistry, which generated an average of 1.2 3 10 5 reads per sample (n 5 6) (Supporting Information Table S1).

OTU processing and analysis
Briefly an OTU or "Operational Taxonomic Unit" is a grouping of sequenced 16S ribosomal RNA read fragments with 97% nucleotide similarity. This classification cutoff has been traditionally assumed to correspond to microbial "species." We merged paired reads from each sample using USEARCH v8.1.1861_i86linux32 (Edgar 2013), allowing no more than 1.0 total expected errors for all bases in the merged read (default parameters). We pooled, de-replicated, and grouped merged reads into OTUs at 97% sequence identity and removed chimeras by comparison to the UCHIME gold standard reference database. We then used the RDP Na€ ıve Bayesian classifier (Wang et al. 2007) to taxonomically classify OTUs and remove unclassifiable OTUs at the kingdom level and OTUs matching eukaryotic, cyanobacterial, or archaeal sequences (20% of reads). We used the PhyloSeq package in R to process the resulting OTU abundance matrix (McMurdie and Holmes 2013) and tested for differential abundance of OTUs between treatments using DESeq2 (Love et al. 2014;McMurdie and Holmes 2014). See Supporting Information for data analysis procedures. All unix shell and R scripts for reproducing analyses can be downloaded from https://dx.doi.org/10.6084/m9.figshare.3184534.v1.

Nutrients
The initial water for incubation experiments contained both low nitrate concentrations (0.94 lmol L 21 ) and dFe concentrations (0.31 nmol L 21 6 0.04 nmol L 21 ) (Supporting Information Tables S1, S2), indicating that the initial phytoplankton community was likely limited by NO 2 3 (King and Barbeau 2007). For the nutrient amended incubations, we selected a low Low Fe treatment (1 nmol L 21 dFe, NO 2 3 : dFe 5 9) to simulate initial conditions approaching phytoplankton Fe-limitation and a High Fe treatment (5 nmol L 21 dFe, NO 2 3 : dFe 5 2) to simulate Fe-replete conditions. King and Barbeau (2007) determined that a nitrate to dFe ratio approaching 10 (lM NO 2 3 : nM Fe) strongly indicated insufficient Fe to support complete phytoplankton nitrate consumption and consequently could be a reliable proxy for community Fe limitation in the field. After 6 d, average ratios of NO 2 3 : dFe were elevated in Low Fe (NO 2 3 : dFe ffi 6) relative to High Fe treatments (NO 2 3 : dFe ffi 0.2) suggesting that Low Fe incubations were likely Fe stressed (Supporting Information Table S2).

Phytoplankton assemblage
The chl a concentration was initially 0.75 lg L 21 , and increased to 9.91 6 0.18 lg L 21 in Low Fe treatments and 12.50 6 3.17 lg L 21 in High Fe treatments (Fig. 1). In both treatments, most of the biomass gain was due to an increase in the abundance of diatoms as indicated by the increase in fucoxanthin in Low Fe and High Fe incubations (Supporting Information Table S3) and direct observation in the cell counts (Supporting Information Table S4). Before Fe addition, the eukaryotic phytoplankton assemblage was composed primarily of Pseudo-nitzschia species (67%) and other large diatoms. Pseudo-nitzschia spp. responded to Fe addition in both treatments, increasing to 78% of the assemblage on average in Low Fe bottles and 82% in High Fe bottles. Cell counts of specific taxonomic groups were not statistically different between High and Low Fe treatments when normalized to total cell counts, indicating that phytoplankton assemblage composition was largely unchanged between Fe treatments although total biomass increased in High Fe samples.

Dissolved Fe-binding ligands
We determined several classes of dFe-binding ligands in each experimental treatment, and Low Fe and High Fe treatments displayed different patterns after 6 d. L 2 and L 3 ligands were detected in situ in the initial water mass, while the strongest L 1 ligands were conspicuously absent (Table 1). After 6 d, Low and High Fe treatments displayed similar L 2 concentrations to initial CCE waters. L 3 concentrations increased at day 6 in High Fe compared with Low Fe treatments (Student's t-test, p < 0.05). However, the most striking result was the presence of strong L 1 ligands exclusively in High Fe treatments at day 6, while no Fe-binding ligands of this strength were detected in Low Fe treatments. In some treatments, the calculated logK of the L 1 class was similar to that of the L 2 class (within logK of 0.2). Although our rigorous data processing methodology resolved three High Fe ligand classes, it is possible that the L 1 and L 2 classes as defined in our study may have partially overlapped in High Fe samples. Regardless of overlapping binding strengths, there would still be significant ligand concentration difference between Low and High Fe treatments. These results suggest that L 1 production was dependent on Fe treatment and the resulting biological dynamics in these incubations.

Heterotrophic bacterial assemblage
We examined the composition of the heterotrophic bacterial assemblage in High and Low Fe treatments using 16S rRNA marker gene sequencing. We first explored broad patterns of taxonomic diversity using a nonmetric multidimensional scaling (NMDS) analysis based on diversity and abundance of bacterial OTUs. This analysis indicated that Low Fe treatments were of a similar taxonomic composition, while High Fe treatments were largely variable between Concentrations of DFe and DFe-binding ligands (L x ) in the six incubations after 6 d and at the start of the experiment (Day 0). Initial values are from in situ seawater used to start the incubations and do not include biological replicates. DFe is the total dissolved Fe concentration in nM, DFe SD is the standard deviation of three technical FIA replicates for the dissolved Fe measurements, L (x) is the concentration in nM of the three ligand classes defined in this study, LogK(x) displays log 10 of the conditional stability constant measured for each ligand class determined at the 95% confidence interval in chemical speciation mode in ProMCC with less than 10% root mean square error. replicates but also taxonomically distinct from Low Fe incubations (Fig. 2). We then fit continuous environmental variables to the OTU ordination and tested for correlations at a significance level of a 5 0.1. Increasing NO 2 3 concentrations were positively correlated with the bacterial OTU composition in Low Fe incubations (R 2 5 0.75, p < 0.05), while L 1 concentrations were strongly positively correlated with OTUs more abundant in High Fe treatments (L 1 , R 2 5 0.77, p < 0.05). The alpha diversity, or the total species richness in each incubation, was significantly lower in High Fe treatments (Supporting Information Table S5).
Because we were struck by the differences in L 1 concentrations between Fe treatments and the general alignment of the OTU composition with increasing L 1 , we searched for potential indicator OTUs that were abundant and overrepresented in High Fe incubations. We narrowed our search to OTUs that each comprised greater than 1% of the total reads in each sample and also had average abundances that were at minimum 1.5 times greater in High Fe samples than in Low Fe samples. Nine OTUs matched these criteria, and all were similar to known copiotrophic strains particularly from the orders Alteromonadales and Rhodobacterales (Fig. 3, Supporting Information Table S6). We also tested for differentially abundant OTUs using a mixture model framework (Love et al. 2014) which identified ten OTUs (   Significant differential abundances of OTUs between high and low Fe conditions after 6 d. Asterisks denote taxa considered to be copiotrophic responders to High Fe treatments. Only OTUs with false discovery rate corrected p values less than 0.05 and with base means greater than 100 are displayed. Base mean indicates the mean OTU abundance across all samples. Log 2 Fold Change is the fold change from high Fe to low Fe samples (i.e., a negative value indicates enrichment in high Fe samples) and OTUs are ordered by increasing fold change, FDR adjusted p value is the False Discovery Rate adjusted p value from negative binomial Wald Tests for significant differential abundance between Fe treatments, and the remaining columns designate taxonomic class, family, and genus of each OTU.
High and Low Fe treatments. Three of these OTUs were enriched in High Fe incubations and had strong similarity to copiotrophic strains. Some highly abundant copiotrophic OTUs, for example Alteromonas OTU13, co-occurred almost exclusively with a single replicate from High Fe treatments (Fig. 3) and were missed by the mixed model approach due to their large variability.

Discussion
After 6 d of incubation, both High and Low Fe treatments stimulated phytoplankton biomass and produced diatom blooms. However, the phytoplankton assemblages in High and Low Fe treatments were largely of the same taxonomic composition, while the Fe-binding ligand pool, macronutrient concentrations, and the heterotrophic microbial assemblage differed substantially. The strongest L 1 ligands were particularly distinct between Fe treatments, and the L 1 ligand classification, defined in this study as having a logK 12, is comparable with the binding affinities of siderophores found in cultures of marine bacteria (Vraspir and Butler 2009). Siderophores are a probable constituent of the operationally defined L 1 class defined here, but we stress that the L 1 class should not be considered to be entirely composed of siderophores.
One possible explanation for the emergence of strong L 1 ligands in High Fe treatments is that they were produced actively by phytoplankton. However, sequenced eukaryotic phytoplankton genomes lack siderophore biosynthesis pathways and do not appear directly take up siderophores (Morrissey and Bowler 2012). An alternative explanation for differences in L 1 is from variation in the composition and activities of bacteria. The bacterial assemblage in High Fe incubations was broadly aligned with L 1 concentrations, and these incubations were dominated by a handful of OTUs related to copiotrophic strains (Supporting Information  Table S6). This copiotroph enrichment was also reflected by the reduced species richness in High Fe treatments (Supporting Information Table S5). Although we only detected L 1 in High Fe incubations, replicates had variable L 1 concentrations as well as variable abundances of copiotroph OTUs (Figs. 2, 3), which may reflect unsampled ecological dimensions from our study. This could include differences in the concentrations and types of bacterial growth substrates in each replicate (Goutx et al. 2007;Mayali et al. 2015), potentially antagonistic microbial interactions (Long and Azam 2001), and/or that multiple copiotrophic groups may produce strong organic ligands. Despite the intravariability of High Fe replicates, broad patterns in the composition of the heterotrophic microbial assemblage appeared to align with the High/Low Fe treatment experimental structure (Fig. 2).
Our observations are consistent with other studies reporting an enrichment of generalist, copiotrophic bacterial taxa after dissolved organic matter additions, and biogenic particle additions to seawater (Nelson and Carlson 2012;Mayali et al. 2015). Other studies have demonstrated that the activities of mostly rare but transiently abundant and transcriptionally active marine copiotrophic bacteria have disproportionately large impacts on biogeochemical cycling (Pedler et al. 2014). Copiotrophic microbial taxa tend to have large genomes with greater regulatory capacity and a greater diversity of Fe acquisition pathways including siderophore biosynthesis and/or siderophore uptake (Hogle et al. 2016). Copiotrophs can also rapidly adapt to patchy nutrient conditions and exploit diverse growth substrates. It may be that copiotrophs also have a large Fe demand in order to fuel their rapid growth and carbon consumption after episodic nutrient pulses. For example, hydroxamate siderophore production by Pseudoalteromonas haloplanktis is sensitive to carbon availability (Sijerčić and Price 2015), and the Alteromonas sp. ALT199 genome contains a petrobactin biosynthesis pathway (Pedler et al. 2014).
We suggest that the most likely scenario accounting for the emergence of L 1 in High Fe treatments is that copiotrophic bacteria were directly producing strong Fe-binding organic ligands in order to scavenge iron from lysing algal cells. The almost complete drawdown of nitrate in all High Fe replicates and the chl a collapse in High Fe C suggests that after 6 d the phytoplankton assemblage in High Fe treatments had entered or was beginning to enter the initial phases of remineralization. We postulate that phytoplankton senescence was coupled with a release of dissolved organic matter (DOM) which stimulated endemic bacterial copiotrophs (Buchan et al. 2014). We suspect the differences between Low and High Fe treatments in the bacterial assemblages were primarily driven by DOM, while the differences in L 1 concentrations reflected siderophore and other strong organic ligand production by copiotrophs responding to DOM enrichment. Although this study provides only correlative evidence and limited temporal resolution, our results do suggest a potential link between specific copiotrophic bacteria and strong dFe-binding ligand production and further study is warranted.
Large scale Fe fertilization studies have demonstrated increases in strong dFe-binding ligands when microbial communities were presumably Fe replete (Rue and Bruland 1997) as well as during phytoplankton bloom decline after Fe fertilization (Kondo et al. 2008). Our experiments suggest that spikes of L 1 production in these mesoscale studies may have been related to shifts in copiotrophic bacterial abundance during phytoplankton blooms. We hypothesize that high Fe conditions initially stimulate Fe-limited photoautotrophs, particularly diatoms, but not endemic heterotrophic bacteria due to organic carbon limitation (Kirchman 1990;Church et al. 2000). As the diatom bloom exhausts inorganic nutrients and progresses to senescence, newly released diatom-derived DOM associated with algal senescence stimulates ambient carbonlimited copiotroph bacteria (Seyedsayamdost et al. 2011;Sule and Belas 2013). Because aerobic respiration requires many Fe-containing enzymes (Hogle et al. 2014), rapidly growing copiotrophs may quickly shift to Fe limitation relative to excess algal-derived organic carbon. The most abundant local Fe source in this context would likely be algal-derived metalloproteins, and copiotroph L 1 production may function to extract Fe from these algal sources and make it bioavailable to the wider bacterial community. If this phenomenon is widespread it may serve as a significant source of strong organic ligands in marine waters. Strong ligand production by heterotrophic bacteria during early bloom senescence may thus be important for overall iron recycling efficiency in microbial ecosystems and may serve to reduce Fe loss due to particle export in the upper ocean.