Iron Limitation Modulates Ocean Acidification Effects on Southern Ocean Phytoplankton Communities

The potential interactive effects of iron (Fe) limitation and Ocean Acidification in the Southern Ocean (SO) are largely unknown. Here we present results of a long-term incubation experiment investigating the combined effects of CO2 and Fe availability on natural phytoplankton assemblages from the Weddell Sea, Antarctica. Active Chl a fluorescence measurements revealed that we successfully cultured phytoplankton under both Fe-depleted and Fe-enriched conditions. Fe treatments had significant effects on photosynthetic efficiency (Fv/Fm; 0.3 for Fe-depleted and 0.5 for Fe-enriched conditions), non-photochemical quenching (NPQ), and relative electron transport rates (rETR). pCO2 treatments significantly affected NPQ and rETR, but had no effect on Fv/Fm. Under Fe limitation, increased pCO2 had no influence on C fixation whereas under Fe enrichment, primary production increased with increasing pCO2 levels. These CO2-dependent changes in productivity under Fe-enriched conditions were accompanied by a pronounced taxonomic shift from weakly to heavily silicified diatoms (i.e. from Pseudo-nitzschia sp. to Fragilariopsis sp.). Under Fe-depleted conditions, this functional shift was absent and thinly silicified species dominated all pCO2 treatments (Pseudo-nitzschia sp. and Synedropsis sp. for low and high pCO2, respectively). Our results suggest that Ocean Acidification could increase primary productivity and the abundance of heavily silicified, fast sinking diatoms in Fe-enriched areas, both potentially leading to a stimulation of the biological pump. Over much of the SO, however, Fe limitation could restrict this possible CO2 fertilization effect.


Introduction
The Southern Ocean (SO) exerts a disproportionate control on the global carbon cycle over glacial-interglacial timescales [1,2] and contributes significantly to the oceanic sequestration of anthropogenic CO 2 [3]. Besides abiotic drivers such as ocean circulation and sea-ice cover, biological carbon uptake and drawdown also control the air-sea-flux of CO 2 in the SO [1,2]. These biological processes are mediated by phytoplankton communities, dominated mainly by silicifying diatoms [4].
The surface waters of the SO are rich in major nutrients such as nitrate and phosphate, but in vast areas of this region primary production is limited by low iron (Fe) availability [5]. Both laboratory and in-situ fertilisation experiments have demonstrated that the growth of SO phytoplankton is strongly enhanced by the addition of Fe [6,7,8]. As Fe is a key nutrient for biochemical pathways including photosynthesis and nitrate assimilation [9], limiting Fe concentrations lead to decreased photochemical efficiencies and photosynthetic rates [10,11]. One important source of iron in open-ocean waters is the melting of sea-ice, which causes seasonal and localized phytoplankton blooms and strong vertical particle fluxes [12,13]. These factors make the marginal sea-ice zone a biogeochemically important region of the SO [12].
The effects of seawater carbonate chemistry on SO phytoplankton have received increasing attention over recent years. Laboratory studies suggest that Antarctic phytoplankton can be growth-limited by CO 2 supply under present-day CO 2 concentrations [14,15]. Field data from continental shelf waters of the Ross Sea have demonstrated CO 2 -dependent changes in primary productivity and phytoplankton assemblages [16,17]. In these prior studies, phytoplankton assemblages were not demonstrably Fe-limited (e.g. high F v /F m reported in [17]), making the extrapolation of results to the open SO waters difficult. Recently, pH-dependent shifts in Fe speciation have been reported [18], suggesting a strong potential for ocean acidification (OA) to reduce Fe bioavailability as seen in experiments with Arctic phytoplankton assemblages [19].
Given the Fe-limited status of much of the SO, there is a great need to investigate combined effects of OA and Fe limitation in this region. Here we present results from a CO 2 -Fe-incubation experiment (190, 390 and 800 matm pCO 2 under Fe-enriched and Fe-depleted conditions) using an open ocean phytoplankton assemblage from the Weddell Sea, an important region for SO primary productivity. The aim of this study was to investigate the interactive effects of OA and Fe availability on species composition, primary production, as well as iron uptake and photophysiology of Fe-limited phytoplankton assemblages.

Experimental Setup
A ship-board incubation experiment was designed using a CO 2 -Fe-matrix-approach to examine potential interactive effects between CO 2 and Fe availability on SO phytoplankton communities. A natural phytoplankton assemblage from the Weddell Sea (66u509S, 0uW) was sampled during mid Dec. 2010 on the RV Polarstern ANT-XXVII/2 cruise. The permission for field work according to the Antarctic Treaty was issued by the Umweltbundesamt (Germany). Seawater was collected from 30 m depth using a ''torpedo fish'' towed outside the ship's wake [20]. To eliminate large grazers, we filtered seawater through an acid-cleaned 200 mm mesh. Water containing the natural phytoplankton community was transferred into acid-cleaned 4L polycarbonate bottles and incubated in growth chambers at 361uC with a constant daylight irradiance of 4065 mmol photons m 22 s 21 (Master TL-D 18W daylight lamps, Philips, adjusted by neutral density screens). The applied irradiance was based on several light measurements in the SO at the sampling depth (data by Mitchell; e.g. DOI: 10.1594/PANGAEA.132802). To provide sufficient time for changes in the phytoplankton assemblages to occur and achieve ecologically relevant information, experiments lasted between 18 and 30 days depending on experimental treatment (18-20 days in case of Fe-enriched and 27-30 days in case of Fedepleted treatments). In order to prevent significant changes in chemical conditions due to phytoplankton growth, incubations were diluted with 0.2 mm filtered seawater when nitrate concentrations were about 10 mmol kg 21 . Dilution water was obtained from the initial sampling location and filtered through acidcleaned 0.2 mm filter cartridges (AcroPak 1500, PALL). Experiments were run with triplicate treatments of two Fe levels (Feenriched and Fe-depleted; see below) and 3 pCO 2 levels (190, 390 and 800 matm).
Tubing, bubbling systems, reservoir carboys, incubation bottles and other equipment were acid-cleaned prior to the cruise using trace metal-clean techniques: After a 2-day Citranox detergent bath and subsequent rinsing steps with Milli-Q (MQ, Millipore), equipment was kept in acid (5N HCl for polyethylene and 1N HCl for polycarbonate materials) for 7 days, followed by 7 rinses with MQ. Equipment was kept triple-bagged during storage and experiments. Incubation bottles were stored under acidified conditions (addition of 500 mL 10N suprapure quarz destilled HCl, Carl Roth, in 500 mL MQ) and rinsed twice with seawater prior to the start of the experiment.
In order to mimic different pCO 2 conditions, the incubation bottles were continuously sparged with air of different CO 2 partial pressures (190, 390 and 800 matm) delivered through sterile 0.2 um air-filters (Midisart 2000, Sartorius stedim). Gas mixtures were generated using a gas flow controller (CGM 2000 MCZ Umwelttechnik), in which CO 2 -free air (,1 ppmv CO 2 ; Dominick Hunter) was mixed with pure CO 2 (Air Liquide Deutschland). The CO 2 concentration in the mixed gas was regularly monitored with a non-dispersive infrared analyzer system (LI6252, LI-COR Biosciences) calibrated with CO 2 -free air and purchased gas mixtures of 150610 and 1000620 ppmv CO 2 (Air Liquide Deutschland).
To promote phytoplankton growth, 1 nM Fe (FeCl 3 , ICP-MS standard, TraceCERT, Fluka) was added to the Fe-enriched treatments. In the Fe-depleted treatments, 10 nM of the hydroxamate siderophore desferrioxamine B (DFB, Sigma) was added to bind and thereby reduce the bioavailable Fe [21,22]. No additional macronutrients were added to the incubation bottles. Abiotic control bottles, used to assess changes in Fe chemistry, contained filtered seawater (0.2 mm) exposed to each treatment condition over the duration of the experiment.

Chemical parameters
Nutrients were determined colorimetrically on-board with a Technicon TRAACS 800 Auto-analyzer on a daily basis over the course of the experiments, following procedures improved after [23]. Samples for total alkalinity (TA) were 0.6 mm-filtered (glass fibre filters, GF/F, Whatman), fixed with 0.03% HgCl 2 and stored in 150 mL borosilicate bottles at 4uC. TA was estimated at the Alfred Wegener Institute (Germany) from duplicate potentiometric titration [24] using a TitroLine alpha plus (Schott Instruments). The calculated TA values were corrected for systematic errors based on measurements of certified reference materials (CRMs provided by Prof. A. Dickson, Scripps, USA; batch #111; reproducibility 65 mmol kg 21 ). Dissolved inorganic carbon (DIC) samples were filtered through 0.2 mm cellulose-acetate filters (Sartorius stedim), fixed with 0.03% HgCl 2 and stored in 5 mL gas-tight borosilicate bottles at 4uC. Also in the home laboratory, DIC was measured colourimetrically in triplicate with a QuAAtro autoanalyzer (Seal) [25]. The analyser was calibrated with NaHCO 3 solutions (with a salinity of 35, achieved by addition of NaCl) with concentrations ranging from 1800 to 2300 mmol DIC kg 21 . CRMs were used for corrections of errors in instrument performance (e.g. baseline drift). Seawater pH was measured potentiometrically on the NBS scale (pH NBS ; overall uncertainty 0.02 units) with a two-point calibrated glass reference electrode (IOline, Schott Instruments). Values for pH were reported on the pH total scale for better comparability with other datasets. Following suggestions by Hoppe et al. [26], seawater carbonate chemistry (including pCO 2 ) was calculated based on TA and pH using CO 2 SYS [27]. The dissociation constants of carbonic acid of Mehrbach et al. (refit by Dickson and Millero) were used for calculations [28,29]. Dissociation constants for HSO 4 were taken from Dickson [30]. Iron chemistry was analyzed using the competitive ligand exchange adsorptive cathodic stripping voltammetry using the ligand 2-(2-thiazolylazo)-p-cresol (TAC, 10 mmol kg 21 [31]). Total dissolved (,0.2 um) Fe concentrations were analyzed following a 45 min UV-photo-oxidation step (acid washed quartz tubes closed with Teflon lids) and concentrations were determined in triplicate using standard additions of a freshly made FeCl 3 standard (ICP-MS standard, TraceCERT, Fluka, 1-4 nM).

Biological Parameters
To determine the taxonomic compositions at the end of the experiment, duplicate aliquots of 200 mL unfiltered seawater were preserved with both hexamine-buffered formalin solution (2% final concentration) and Lugols (1% final concentration). Preserved samples were stored at 4uC in the dark until further analysis by inverted light microscopy (Axiovert 200, Zeiss). Additionally, species dominating the final communities were identified using scanning electron microscopy (Philips XL30) according to taxonomic literature [32]. Average biovolume of the dominant species was calculated based on representative cell size measurements [33]. Values were in good agreement with the MAREDAT database [34]. For analysis of particulate organic carbon (POC), cells were collected onto precombusted GF/F-filters (15 h, 500uC), which were subsequently stored at 220uC and dried for .12 h at 60uC prior to sample analysis. Analysis was performed using an Automated Nitrogen Carbon Analyser mass spectrometer system (ANCA-SL 20-20, SerCon Ltd.). Samples for determination of chlorophyll a (Chl a) concentration were filtered onto 0.45 mm cellulose acetate filters (Sartorius stedim) and stored at 220uC until analysis onboard. Chl a was subsequently extracted in 10 mL 90% acetone (overnight in darkness, at 4uC) and concentrations determined on a fluorometer (10-000 R, Turner Designs), using an acidification step to determine phaeopigments [35].

Physiological assays
Primary production of the final phytoplankton assemblages was determined in 100 mL incubations after addition of a 10 mCi (0.37 MBq) spike of NaH 14 CO 3 (PerkinElmer, 53.1 mCi mmol 21 ). From the incubations, 0.5 mL aliquots were immediately removed and mixed with 10 mL of scintillation cocktail (Ultima Gold AB, PerkinElmer) to determine the total amount of added NaH 14 CO 3 . For blank determination, samples were filtered and acidified immediately after 14 C spikes. After 24 h of incubation under acclimation light intensity, samples were filtered onto GF/ F-filters, acidified with 6N HCl and left to degas overnight. Filters were then transferred into scintillation vials, to which 10 mL of scintillation cocktail was added. After $2 h, the samples were measured on a ship-board liquid scintillation counter (Tri-Carb 2900TR, PerkinElmer), using automatic quench correction and a maximum counting time of 5 minutes.
Maximum Fe uptake capacity of the final phytoplankton assemblages was estimated after 2-4 h dark-acclimation in 500 mL acid-cleaned PC-bottles. In case of the Fe-treatments, cells were gently concentrated by filtration over an HCl-cleaned 2 mm membrane filter (Isopore, Millipore), rinsed and resuspended in 500 mL filtered seawater from the initial sampling location in order to dispose all DFB. From this, 50 mL were taken for Chl a measurements. Subsequently, 1 nM of 55 Fe (Perkin Elmer, 33.84 mCi mg 21 as FeCl 3 in 0.5 N HCl) was added to each sample (both Fe-and -enriched). Generally, 2 mL were taken from all samples to determine the initial amount of 55 Fe. Subsequently, cells were exposed for at least 24 h to the acclimation light intensity. At the end of the incubation time, the sample was filtered onto GF/F-filters and rinsed 5 times with oxalate solution (gravity filtered, 2 min between rinses) and 3 times with natural seawater [36]. Each filter was then collected in a scintillation vial, amended with 10 mL scintillation cocktail (Ultima Gold A, PerkinElmer) and mixed thoroughly (Vortex). 55 Fe counts per minute were estimated for each sample on the ship-board liquid scintillation counter (Tri-Carb 2900TR, PerkinElmer), and converted into disintegrations per minute considering the radioactive decay and custom quench curves. 55 Fe uptake was then calculated taking into account the initial 55 Fe concentration and the total dissolved Fe concentration (background and added). Fe uptake rates were normalized to POC using Chl a:POC ratios from the respective treatments.
Photophysiological parameters were measured using a Fluorescence Induction Relaxation System (FIRe; Satlantic, Canada; 37). Samples were 1 h dark-acclimated prior to measurements to ensure that all photosystem II (PSII) reaction centers were fully oxidized and non-photochemical quenching (NPQ) was relaxed. The duration of the dark acclimation was chosen after testing different time intervals (data not shown). Samples were then exposed to a strong short pulse (Single Turnover Flash, STF), which was applied in order to cumulatively saturate PSII. Afterwards, a long saturating pulse (Multiple Turnover Flash, MTF) was applied in order to fully reduce the PSII and the plastoquinone (PQ) pool. The minimum (F 0 ) of the STF phase and maximum (F m ) fluorescence of the MTF was used to calculate the apparent maximum quantum yield of photochemistry in PSII (F v / F m ) according to the equation (F m 2F 0 )/F m . This parameter was calculated for all bottles on a regular basis (every 6-7 days). Values of these parameters as well as of the functional absorption cross section of PSII (s PSII [Å 2 quanta 21 ]) were derived using the FIRePro software provided by Satlantic [37]. Additional fluorescence measurements were performed under increasing irradiances (21,41,66,88,110 and 220 mmol photons m 22 s 21 ) provided by an external actinic light source (warm white 350 mA LEDs, ILL3A003, CML Innovative Technologies). After 5 minutes acclimation to the respective light level, the light-acclimated minimum (F q 9) and maximum (F m 9) fluorescence were estimated. The effective quantum yield of photochemistry in open reaction centers of PSII was derived according to the equation (F m 92F q 9)/ F m 9 [38]. Relative electron transport rates (rETR) were then calculated as the product of effective quantum yield and applied growth irradiance of 40 mmol photons m 22 s 21 . Using the Stern-Volmer equation [39], NPQ of Chl a fluorescence under growth irradiance was calculated as F m /F q 921. NPQ was relaxed (values ,0.1) at lowest light levels for all treatments (data not shown). All measurements were conducted at the growth temperature.

Statistics
All data is given as the mean of the replicates (n = 3) with 1 standard deviation. To test for significant differences between the treatments, Two Way Analyses of Variance (ANOVA) with additional normality tests (Shapiro-Wilk; passed for all data shown) were performed. The significance level was set to 0.05. Statistical analyses were performed with the program SigmaPlot (SysStat Software Inc).

Seawater chemistry
The initial carbonate system (pH: 7.9360.01; DIC 2210617 mmol kg 21 ; TA: 2303614 mmol kg 21 ) shifted to experimental treatment levels (average pH of 8.3960.02, 8.1360.02, and 7.8060.03 for the three CO 2 treatments) within the first 2 days of the experiment (Figure 1 A). The semi-continuous dilutebatch approach led to stable seawater carbonate chemistry over the course of the experiment (Figure 1 A). Compared to abiotic controls, drift was ,8% and ,5% for TA and DIC, respectively (

Photophysiology
Over the course of the experiment, we observed significant Fedependent differences in the apparent maximum quantum yield of PSII reaction centres (F v /F m ; Figure Table 2). In contrast, the pCO 2 treatments had no effects on F v / F m under either of the Fe treatments. Relative electron transfer rates from PSII (rETR) were significantly higher in Fe-enriched treatments, and also showed pCO 2 -dependent increases under both Fe-enriched (5.760.8 at low and 7.560.9 at high pCO 2 ; p,0.001) and Fe-depleted conditions (4.260.4 at low and 5.360.1 at high pCO 2 ; p,0.001). Significant interactive effects between Fe-and pCO 2 -treatments were also observed in photoprotective non-photochemical quenching (NPQ; Table 2; p,0.001). Under Fe-limitation, NPQ decreased from 0.2260.03 at low pCO 2 to 0.1160.00 at high pCO 2 , while under Fe-enriched conditions, NPQ was independent of pCO 2 (0.0660.01 at all pCO 2 levels; p,0.001).
The ratios of Chl a to POC of the final phytoplankton assemblages (Table 1) were significantly higher (p,0.001) in Feenriched (0.02360.003 at low and 0.01760.004 at high pCO 2 ) compared to Fe-depleted treatments (0.01260.001 at low and 0.01060.001 at high pCO 2 ). Chl a:POC ratios furthermore decreased significantly with increasing pCO 2 levels (p = 0.012), irrespective of the Fe-status.

Fe and C uptake
For all pCO 2 levels, carbon-normalized Fe uptake capacities at the end of the experiment were 10-fold higher in Fe-depleted compared to Fe-enriched treatments (Figure 2; p,0.001), but with no significant CO 2 effect. Combining data across pCO 2 treatments, mean Fe uptake capacities were 7.5063. 35

Species composition
We observed pronounced shifts in the diatom-dominated phytoplankton assemblages in association with the CO 2 -dependent changes in primary productivity ( Figure 4; Table 3). Shifts in species composition did not lead to changes in average cell size in the different assemblages (data not shown). After Fe-enrichment, Pseudo-nitzschia cf. turgiduloides was the most abundant species under low and intermediate pCO 2 (3965% and 4069%, respectively) , whereas Fragilariopsis cylindrus dominated communities under high pCO 2 levels (7265%). Furthermore, Chaetoceros cf. simplex abun-  dances increased with rising CO 2 (from 1160% at low pCO 2 to 1761% at high pCO 2 ). Phytoplankton composition changes were also observed under Fe limitation, but the nature of these species shifts differed significantly from those seen under high Fe levels ( Figure 4; Table 3). Under Fe-depleted conditions, Pseudo-nitzschia cf. turgiduloides dominated the low pCO 2 treatment (55616%), while Synedropsis sp. was the most prevalent species under high pCO 2 (7862%).

Confirmation of Fe limitation
Dissolved Fe concentrations in Fe-depleted abiotic controls remained at initial concentrations, showing that experimental manipulations and CO 2 bubbling resulted in no significant Fe contamination (Table 1). In Fe-enriched abiotic controls, dissolved Fe concentrations decreased over the course of the experiment. This can be attributed to precipitation and absorption of colloidal iron in the absence of significant concentrations of Fe-binding ligands [40]. Fe limitation of phytoplankton in the DFB treatments was confirmed by significant differences in F v /F m between Feenriched and -depleted treatments (Table 2; Figure 1 B). F v /F m values of Fe-depleted treatments are comparable to those observed in naturally Fe-limited phytoplankton communities [41]. In line with previous findings on SO phytoplankton, also other photophysiological parameters like NPQ and rETR differed significantly between Fe treatments (Table 2, [42,43]). Moreover, the significantly higher Fe uptake capacity in Fe-depleted treatments likely reflects the induction of high-affinity uptake systems and/or the selection of phytoplankton communities with greater Fe affinities [21,22]. All of these observations confirm Fe limitation in the Fe-depleted treatments.

OA response under Fe-enriched conditions
The observed CO 2 -dependent increase in primary productivity under Fe-enriched conditions (Figure 3) confirms that CO 2 fixation in SO phytoplankton can be limited by carbon supply under current CO 2 concentrations [14,16]. This hypothesis is further supported by the decrease in Chl a:POC ratios and the increase in rETRs with increasing pCO 2 (Table 2). Similarly, Ihnken et al. [44] observed ETR max in Fe-sufficient Chaetoceros muelleri to increase with increasing CO 2 . These findings suggest that the Calvin cycle, the major sink of photosynthetic energy [9], is the rate-limiting step of photosynthesis under low pCO 2 levels. NPQ was not affected under high pCO 2 , Fe-enriched conditions. Under the applied irradiance of 40 mmol photons m 22 s 21 , however, NPQ values were generally very low (Table 2) suggesting that there is little requirement for dissipation of light. For one of   the dominant species in the Fe-enriched treatments, Fragilariopsis cylindrus, a significant induction of NPQ was only found at irradiances larger than 200 mmol photons m 22 s 21 [43]. We can therefore conclude that even when photosynthesis was carbon limited under low pCO 2 , the applied irradiance was too low to induce NPQ under Fe-enriched conditions and other pathways were operated as electron sinks (e.g. midstream-oxidases [9]). The changes in physiological responses in Fe-enriched phytoplankton assemblages were accompanied by a pronounced shift in the species composition ( Figure 4; Table 3) from Pseudo-nitzschia cf. turgiduloides under low and intermediate pCO 2 to Fragilariopsis cylindrus under high pCO 2 levels. Likely mechanisms for this floristic shift include species-specific differences in carbon acquisition [15,45], as well as pH-mediated differences in cellular physiology, e.g. changes in electrochemical membrane potentials and ion transport processes [46]. Pseudo-nitzschia has also been observed to dominate in bloom situations after Fe fertilization [7], where pH increases due to biomass build-up and drawdown of CO 2 . This is in line with results from CO 2 manipulations on SO phytoplankton assemblages, which were dominated by Pseudonitzschia at high pH [16]. In a laboratory study under Fe-enriched conditions, growth and rETRs of Pseudo-nitzschia subcurvata were unaffected by pCO 2 , suggesting that this species shows little to no responses to CO 2 fertilization [15]. Our field experiment suggests that Fragilariopsis cylindrus, in contrast, benefited from increased pCO 2 . Even though information of OA responses for this species is lacking, the related Fragilariopsis kerguelensis showed enhanced rETRs with increasing pCO 2 (S. Trimborn, unpublished data). We thus speculate that F. cylindrus increased its photosynthetic activity under elevated pCO 2 , thereby outcompeting the otherwise faster growing Pseudo-nitzschia. Under OA, also the relative abundance of Chaetoceros sp. increased by 50% (Table 3), which is consistent with previous findings on SO phytoplankton assemblages [16] as well as growth responses of Chaetoceros debilis to increased pCO 2 [15].

OA response under Fe-limitation
The observation that under Fe-limitation, productivity was not stimulated by increasing pCO 2 ( Figure 3) may indicate that Fe acts as the main limiting factor suppressing the effects of other nutrients such as inorganic carbon. Alternatively, the apparent insensitivity of primary production to OA may arise from antagonistic physiological responses to pCO 2 and pH.
Under Fe limitation, elevated pCO 2 significantly increased rETRs and decreased NPQ, while the functional absorption cross section was not affected ( Table 2). These results suggest a greater electron sink associated with the Calvin cycle and thus a decreased need for energy dissipation under OA [44]. It is also known that linear electron transport (LET) towards the Calvin cycle is not the sole sink for photosynthetic energy and that, depending on the ATP demand of the cells, alternative electron pathways can play an important role (e.g. Mehler reaction, MOX pathway, pseudocyclic electron flow around PSI) [9]. Increased pCO 2 , however, is known to decrease photorespiration and/or the need for carbon concentrating mechanism (CCMs), which would lead to a decrease in cellular ATP demand [47]. The observed increase in rETRs in Fe-limited cells at high pCO 2 levels ( Table 2) may therefore rather be linked to higher LET rates than to alternative electron pathways. The higher LET, enabled by the enhanced CO 2 fixation in the Calvin cycle, could counteract the generally greater need for photoprotection under Fe limitation [42,43]. This could explain the opposing CO 2 effects on NPQ under Fedepleted and Fe-enriched conditions. The CO 2 effect, apparent in photophysiology, is potentially masked in primary production by co-occurring pH effects on Fe bioavailability. According to the observed decline in Fe bioavailability with decreasing pH [18,19], Fe-depleted phytoplankton would experience the greatest Fe stress under high pCO 2 .
In this study, Fe-limitation was achieved by the addition of the chelator DFB. Even though DFB has been shown to form strong complexes with Fe and thereby decrease Fe availability by .90% [36], phytoplankton can still access DFB-bound Fe to some extent [48,49]. Since the phytoplankton assemblages in our DFBtreatments were strongly Fe-limited (as demonstrated by photophysiology, Fe and C uptake), bioavailability of Fe must have been largely reduced. Fe bioavailability also seems to be slightly reduced with increasing pCO 2 (Figure 2), as has been observed in natural phytoplankton communities (i.e. without any added chelators) and in studies using different chelators such as EDTA and DFB [18,19,50]. Also, Maldonado et al. [49] suggest that the in-situ organic Fe-complexes observed in the SO have similar bioavailability compared to DFB. Although the chemical nature of in-situ organic Fe-binding ligands is not fully resolved, hydroxamate siderophores have been reported [51]. It is thus possible that at least some of the organically bound Fe exhibits a similar pHdependent bioavailability as induced by DFB, and thus may allow for the extrapolation of our results to field situations. In order to study the bioavailability of Fe associated with in-situ Fe-binding organic ligands under OA scenarios, future experiments without added chelators should be conducted. As the nature of Fe-binding ligands remains largely unknown and can vary spatially [52,53], one should address the Fe bioavailability of various compounds (e.g., humic acids, saccharids, exopolymeric substances) that were reported to affect iron biogeochemistry [36,54,55]. Furthermore, organic ligands control the bioavailability and the physicochemistry of trace metals in general [52,56]. As some of those (Co, Cd, Zn) are also essential for phytoplankton physiology (e.g. for the activity of the carbonic anhydrase) [57], joint measurements of other trace metals as well as their ligands would be desirable.
Although primary productivity was not sensitive to OA under Fe limitation, we did observe CO 2 -dependent species shifts, with Pseudo-nitzschia sp. dominating under low and Synedropsis sp. under high pCO 2 ( Figure 4). The low abundances of F. cylindrus in Fe-depleted treatments probably reflect the rather high sensitivity of this species towards Fe limitation [43], which could be due to low Fe uptake capacities observed for this species [21]. In contrast, Pseudo-nitzschia has been shown to be an efficient user of Fe under limiting concentrations [58] and sporadic Fe input events [59]. Interestingly, the final proportion of Pseudo-nitzschia declined strongly with increasing pCO 2 , irrespectively of the Fe status ( Table 3), suggesting that its growth rates must have been significantly lower than those of the dominant species (F. cylindrus and Synedropsis sp. under Feenriched and -depleted conditions, respectively). This observation is in line with a recent study on P. pseudodelicatissima, whose growth rates were not affected by OA under either Fe-deplete nor -replete conditions [50]. As Pseudo-nitzschia generally does not seem to benefit from increased pCO 2 levels ( [15,16,50], this study), one could expect OA to have a negative effect on the abundances of this genus under on-going climate change. At present, nothing is known about the Fe and CO 2 requirements of Synedropsis [60]. However, a possible appearance of Synedropsis in phytoplankton assemblages or incubation experiments might have been overlooked in past studies, as their delicately silicified frustules are very prone to dissolution [61] and not distinguishable from Pseudo-nitzschia by light microscopy (Figure 4).
Our results clearly demonstrate a strong difference in CO 2dependent community structure between Fe-enriched and Fedepleted conditions (Figure 4). To explain these shifts, more information on species-specific differences in Fe requirements, uptake, as well as allocation strategies [21,62,63], and inorganic carbon acquisition [15,16] is needed for all dominant species.

Biogeochemical implications
The findings of our study suggest that the effects of OA on primary production and community structure are strongly modulated by the prevailing Fe concentrations. Our results, and those of others [16,64], indicate a potential stimulation of the biological pump as a result of increased pCO 2 in Fe-replete regions. Under Fe enrichment and increasing pCO 2 levels, we observed a shift from weakly silicified Pseudo-nitzschia towards more heavily silicified Fragilariopsis. Pseudo-nitzschia remineralizes quickly in subsurface waters [65], while Fragilariopsis is a more efficient vector of carbon export [12]. Thus, enhanced primary production, in concert with potentially higher export efficiencies, could lead to a stronger downward flux of organic matter in Fe-replete areas under OA.
The described feedback by 'CO 2 -fertilisation', however, may not operate over the broad expanse of the Fe-limited SO. These regional differences in CO 2 -sensitivity might be even more pronounced in terms of carbon export efficiencies, as under Fedepleted conditions no functional shifts in species composition were observed. Here, all assemblages were dominated by weakly silicified species such as Pseudo-nitzschia cf. turgiduloides [54] or Synedropsis sp. [60]. Frustules of both species are delicate and only preserved in shallow waters or under special circumstances such as large aggregation events in combination with anoxia [61,65]. Irrespective of their potential for carbon export, all species dominating our incubations are ecologically important [61,66,67]. Furthermore, both F. cylindrus and P. turgiduloides are not only characteristic sea-ice algae but also dominate phytoplankton assemblages in open waters [66,67]. In fact, most genera being characteristic for SO phytoplankton assemblages were present in initial and final phytoplankton assemblages (Table 3). Overall, species in the incubations resemble a mixture of typical open-ocean and sea-ice associated species [68][69][70][71]. Hence, our interpretations may not be restricted to sea-ice influenced habitats only.
Our results suggest that the potential 'CO 2 fertilization' effect critically depends on the availability of Fe, determining how strongly the biological pump will serve as a carbon sink in the future SO. Realistic projections of primary production and CO 2 sequestration thus remain difficult as long as scenarios for Fe input as well as its bioavailability to phytoplankton remain poorly constrained [18,72]. The results of this study furthermore highlight the need to assess combined effects of important environmental factors in order to understand and predict responses to single stressors such as OA. In this respect, irradiance levels should also be considered as a potentially interacting factor. Indeed, the level of energization has been shown to strongly influence the strength of phytoplankton responses to OA [73], suggesting that also the interaction between Fe and CO 2 availability could be modulated by light conditions. To thoroughly assess consequences of OA, multifactorial perturbation experiments (including factors such as different Fe sources or grazing) should target physiological as well as ecological responses of SO phytoplankton assemblages.