Fe(II) stability in coastal seawater during experiments in Patagonia, Svalbard, and Gran Canaria

The speciation of dissolved iron (DFe) in the ocean is widely assumed to consist almost exclusively of Fe(III)-ligand complexes. Yet in most aqueous environments a poorly defined fraction of DFe also exists as Fe(II), the speciation of which is uncertain. Here we deploy flow injection analysis to measure in situ Fe(II) concentrations during a series of mesocosm/microcosm/multistressor experiments in coastal environments in addition to the decay rate of this Fe(II) when moved into the dark. During five mesocosm/microcosm/multistressor experiments in Svalbard and Patagonia, where dissolved (0.2 μm) Fe and Fe(II) were quantified simultaneously, Fe(II) constituted 24 %–65 % of DFe, suggesting that Fe(II) was a large fraction of the DFe pool. When this Fe(II) was allowed to decay in the dark, the vast majority of measured oxidation rate constants were less than calculated constants derived from ambient temperature, salinity, pH, and dissolved O2. The oxidation rates of Fe(II) spikes added to Atlantic seawater more closely matched calculated rate constants. The difference between observed and theoretical decay rates in Svalbard and Patagonia was most pronounced at Fe(II) concentrations < 2 nM, suggesting that the effect may have arisen from organic Fe(II) ligands. This apparent enhancement of Fe(II) stability under post-bloom conditions and the existence of such a high fraction of DFe as Fe(II) challenge the assumption that DFe speciation in coastal seawater is dominated by ligand boundFe(III) species.

 An additional paragraph, and a few extra lines concerning the specific purpose of this manuscript, are added in the introduction (new lines 75-87). (Pasted below) To what extent the behaviour of dissolved Fe(II) in the ocean is influenced by organic material, rather than solely by inorganic chemical processes, is however challenging to determine given the sub-nanomolar Fe(II) concentrations present in pelagic environments, and the diurnal variability expected in surface Fe(II) concentrations (Johnson et al., 1994;Miller and Kester, 1994). The clear effects of biologically derived organic material on Fe(II) stability however do indicate that aquatic micro-organisms may directly or indirectly influence the speciation and stability Fe(II) in solution around them and thus moderate the bioavailability and bioaccessibility of dissolved Fe to cellular uptake systems (Croot et al., 2001;Shaked et al., 2002;Samperio-Ramos et al., 2018a). In order to gain insight into the potential interactions between Fe(II) and marine micro-organisms in dynamic surface waters, here we adapted flow injection apparatus to measure in situ Fe(II) concentrations both in a series of mesocosm experiments (Gran Canaria, Patagonia, Svalbard) and in adjacent ambient waters. The experimental gradients in these experiments included pH, zooplankton density and dissolved organic carbon (DOC) concentration facilitating an investigation of Fe(II) stability under a diverse range of biogeochemical conditions. By allowing ambient Fe(II) concentrations to decay in the dark, the stability of in situ Fe(II) concentrations was assessed and compared to calculated inorganic oxidation rates.
 Minor corrections are made to the labels for figures 1 and3, and Tables 3 and 4 (now Tables 1  and 2) Sincerely, Mark Hopwood, Corresponding author

Introduction
The micronutrient iron (Fe) limits marine primary production across much of the surface ocean (Martin and Fitzwater, 1988;Martin et al., 1990;Kolber et al., 1994). Fe is required for the synthesis of the photosynthetic apparatus of autotrophs 30 (Geider and Laroche, 1994), is an essential element in the enzyme nitrogenase required for N 2 fixation (Moore et al., 2009), and is important for phosphorous (P) acquisition from dissolved organic P compounds as part of the enzyme alkaline phosphatase (Mahaffey et al., 2014). Fe is thus one of the key environmental control factors that concurrently regulate marine microbial community structure and productivity (Boyd et al., 2010;Tagliabue et al., 2017). The distribution of dissolved Fe (DFe) in the ocean (Tagliabue et al., 2017;Schlitzer et al., 2018) and the magnitude of the dominant 35 atmospheric (Mahowald et al., 2005;Conway and John, 2014), hydrothermal (Tagliabue et al., 2010;Resing et al., 2015) and shelf sources (Elrod et al., 2004;Severmann et al., 2010) are now moderately well constrained. Furthermore, dissolved Fe(III) speciation has also been explored in depth and it is evident that organic Fe(III)-binding ligands are a major control on the concentration and distribution of DFe in the ocean Hunter and Boyd, 2007;Gledhill and Buck, 2012). Small organic ligands (L) capable of complexing Fe(III), can maintain DFe concentrations of up to ~1-2 nM in oxic 40 seawater which is an order of magnitude greater than the inorganic solubility of Fe(III) under saline, oxic conditions Millero, 1999, 2002). Characterising these ligands in terms of their concentrations and affinity for Fe(III) was therefore a major objective for chemical oceanographers over the past two decades using a variety of related titration techniques (Gledhill and Van Den Berg, 1994;Rue and Bruland, 1995;Hawkes et al., 2013). 99% of DFe in the ocean is hypothesized to be present as Fe(III)-L complexes (Gledhill and Buck, 2012) and this observation explicitly or implicitly underpins the 45 formulation of DFe in global marine biogeochemical models (Tagliabue et al., 2016).
There are however two specific marine environments in which this widely quoted "99%" statistic is incorrect. The first is oxygen minimum zones, where low O 2 concentrations extend the half-life of Fe(II) with respect to oxidation and thus permit high nanomolar concentrations of Fe(II) to accumulate in the water column accounting for up to 100% of DFe (Landing and 50 Bruland, 1987;Lohan and Bruland, 2008;Chever et al., 2015). The second is surface waters where photochemical processes initiate the redox cycling of DFe and permit persistence of measurable (>0.2 nM) concentrations of dissolved Fe(II) in spite of rapid oxidation rates (Barbeau, 2006;Croot et al., 2008). Fe(II) is reported to account for 20% of surface DFe concentrations in the Baltic (Breitbarth et al., 2009), 12-14% in the Pacific , and 5-65% in the South Atlantic and Southern Ocean (Bowie et al., 2002a;Sarthou et al., 2011). A significant fraction of DFe is therefore likely 55 present globally as Fe(II) in oxic surface waters. Fe(II) concentrations at depth are less well-characterised, although there is some evidence of picomolar Fe(II) concentrations occurring throughout the pelagic water column suggesting that 'dark' Fe(II) production is also a widespread phenomenon Sarthou et al., 2011;Sedwick et al., 2015). The kinetic lability of dissolved Fe(II) relative to dissolved Fe(III) (Sunda et al., 2001), the positive effect of redox cycling maintaining DFe in solution in bioavailable forms-irrespective of whether Fe(II) itself is bioavailable- (Croot et al., 2001;60 Emmenegger et al., 2001), and the potentially widespread presence of Fe(II) as a high fraction of DFe in surface waters (O'Sullivan et al., 1991;Sarthou et al., 2011) raise interest in the role of Fe(II) in the marine biogeochemical Fe cycle.
Fe(II) ligands, akin to Fe(III) ligands in seawater-but likely with different functional groups and binding constants (Boukhalfa and Crumbliss, 2002), are widely speculated to affect the oxidation rate of Fe(II) in seawater (Santana-Casiano et al., 2000;Rose and Waite, 2003;González et al., 2014). Yet characterising the concentration and properties of organic Fe(II) ligands in natural waters using titration approaches, as successfully adapted to determine Fe(III)-speciation (Gledhill and Buck, 2012), has proven challenging (Statham et al., 2012) due to practical difficulties in stabilizing Fe(II) concentrations 70 without unduly affecting Fe(II) speciation. Nevertheless a broad range of cellular exudates have been demonstrated to affect Fe(II) concentrations in seawater, both via enhancing Fe(II) formation rates and retarding the Fe(II) oxidation rate (Rijkenberg et al., 2006;González et al., 2014;Lee et al., 2017).
To what extent the behaviour of dissolved Fe(II) in the ocean is influenced by organic material, rather than solely by 75 inorganic chemical processes, is however challenging to determine given the sub-nanomolar Fe(II) concentrations present in pelagic environments, and the diurnal variability expected in surface Fe(II) concentrations (Johnson et al., 1994;Miller and Kester, 1994). The clear effects of biologically derived organic material on Fe(II) stability however do indicate that aquatic micro-organisms may directly or indirectly influence the speciation and stability Fe(II) in solution around them and thus moderate the bioavailability and bioaccessibility of dissolved Fe to cellular uptake systems (Croot et al., 2001;Shaked et al., 80 2002;Samperio-Ramos et al., 2018a). In order to gain insight into the potential interactions between Fe(II) and marine micro-organisms in dynamic surface waters, here we adapted flow injection apparatus to measure in situ Fe(II) concentrations both in a series of mesocosm experiments (Gran Canaria, Patagonia, Svalbard) and in adjacent ambient waters. The experimental gradients in these experiments included pH, zooplankton density and dissolved organic carbon (DOC) concentration facilitating an investigation of Fe(II) stability under a diverse range of biogeochemical conditions. By 85 allowing ambient Fe(II) concentrations to decay in the dark, the stability of in situ Fe(II) concentrations was assessed and compared to calculated inorganic oxidation rates.

Mesocosm set up and sampling (MesoPat/MesoArc/Gran Canaria)
The setup for the same series of incubation experiments from which we discuss results here is reported in detail in a 90 companion paper (Hopwood et al., 2018a). However, for ease of access, a shorter version is reproduced here. Briefly, all experiments used coastal seawater which was either pumped from small boats deployed offshore, or from the end of a floating jetty. Two of the outdoor mesocosm experiments (MesoPat and MesoArc) were conducted using the same basic design in different locations. For these mesocosms, 10 identical 1000-1500 L tanks (high density polyethylene, HDPE) were filled ~95% full with coastal seawater passed through nylon mesh to remove mesozooplankton. Fresh zooplankton 95 (copepods) were collected at ~30 m by horizontal tows with a mesh net, stored overnight in 100 L containers and non-viable 4 copepods removed by siphoning prior to making zooplankton additions to the mesocosm tanks. After filling the mesocosms, the freshly collected zooplankton were added to 5 of the tanks to create contrasting high/low grazing conditions. Macronutrients (NO 3 /NH 4 , PO 4 and Si) were added daily. Across both the 5-high and 5-low grazing tank treatments, a DOC gradient was created by addition of glucose to provide carbon at 0, 0.5, 1, 2 and 3 times the Redfield Ratio (Redfield, 1934) 100 of carbon with respect to added PO 4 . At regular 1-2 day intervals throughout each experiment, mesocosm water was sampled through silicon tubing immediately after mixing of the tanks using plastic paddles with the first 2 L discarded in order to flush the sample tubing.
A 3 rd outdoor mesocosm experiment, Gran Canaria, (Taliarte, March 2016) used 8 cylindrical polyurethane bags with a depth 105 of approximately 3 m, a starting volume of ~8000 L and no lid or screen on top (for further details see Filella et al., 2018 andHopwood et al., 2018a). After filling with coastal seawater the bags were allowed to stand for 4 days. A pH gradient across the 8 bags was then induced (on day 0), by the addition of varying volumes of filtered, pCO 2 saturated seawater using a custom-made distribution device (Riebesell et al., 2013). A single macronutrient addition was made on day 18.

Microcosm (MicroPat) and multistressor (MultiPat/MultiArc) set up and sampling 110
MicroPat, a 10-treatment microcosm mirroring the MesoPat 10 tank mesocosm (treatment design as per MesoPat, but with 6 × 20 L containers per treatment rather than a single HDPE tank) and two 16-treatment multistressor experiments (MultiPat/MultiArc) were conducted using artificial lighting in temperature-controlled rooms. Coastal seawater, filtered through nylon mesh, was used to fill 20 L HDPE collapsible containers. The 20 L containers were arranged on custom made racks with a light intensity of 80 µmol quanta m -2 s -1 , approximating that at ~3 m depth. Lamps (Phillips, MASTER TL-D 90 115 De Luxe 36W/965 tubes) were selected to match the solar spectrum as closely as possible. A diurnal light regime representing spring/summer light conditions at each fieldsite was used and the tanks were agitated daily and after any additions (e.g. glucose, acid or macronutrient solutions) in order to ensure a homogeneous distribution of dissolved components. In all 20 L scale experiments, macronutrients were added daily. One 20 L container from each treatment set was 'harvested' for sample water each sampling day. 120 The experimental matrix used for the MultiPat/MultiArc experiments duplicated the MesoPat/MesoArc design, with an additional pH manipulation: ambient and low pH. The pH of 'low' pH treatments was adjusted by a single addition of HCl (trace metal grade) on day 0 only with pH measured prior to and after the addition. Sample water from 20 L collapsible containers was extracted using a plastic syringe and silicon tubing which was mounted through the lid of each collapsible 125 container. Throughout, where changes in meso/micro/multi experiments are plotted against time, 'day 0' is defined as the day the experimental gradient (zooplankton, DOC, pH, pCO 2 ) was imposed. Time prior to day 0 was intentionally introduced during some experiments to allow water to equilibrate with ambient physical conditions after mesocosm filling. Fe(II)

Comment [HM3]:
Removed Tables 1 and 2 and associated captions/descriptions (2.5 pages of material) concentration varies on diurnal timescales and thus during each experiment where a time series of Fe(II) or DFe concentration was measured, sample collection and analysis occurred at the same time each day. 130

Trace elements
Trace metal clean low density polyethylene (LDPE, Nalgene) bottles were prepared via a three stage washing procedure (1 day in detergent, 1 week in 1.2 M HCl, 1 week in 1.2 M HNO 3 ) and then stored empty and double bagged until use. Total dissolvable Fe (TdFe) samples were collected without filtration in trace metal clean 125 mL LDPE bottles. Dissolved Fe 135 (DFe) samples were collected in 0.5 or 1 L trace metal clean LDPE bottles and then filtered through acid-rinsed 0.2 µm filters (PTFE, Millipore) using a peristaltic pump (Minipuls 3, Gilson) into trace metal clean 125 mL LDPE bottles within 4 h of sample collection. TdFe and DFe samples were then acidified to pH <2.0 by the addition of HCl (150 μL, UpA grade, Romil) and stored for 6 months prior to analysis. Samples were then diluted using 1 M distilled HNO 3 (SpA grade, Romil, distilled using a sub-boiling PFA distillation system, DST-1000, Savillex) and subsequently analyzed by high resolution 140 inductively coupled plasma-mass spectrometry (HR-ICP-MS, ELEMENT XR, ThermoFisherScientific) with calibration by standard addition. To verify the accuracy of Fe measurements the Certified Reference Materials NASS-7 and CASS-6 were analysed after the same dilution procedure with the measured Fe concentration in close agreement with certified values (6.21 ± 0.77 nM certified 6.29 ± 0.47 nM, and 26.6 ± 0.71 nM certified 27.9 ± 2.1 nM). The analytical blank was 0.13 nM Fe. The field blank (de-ionized, MilliQ, water handled and filtered as if a sample in the field) was ~0.5 nM and varied slightly 145 between field experiments, yet was always <16% of DFe concentration.
Fe(II) samples (unfiltered) were collected in trace metal clean translucent 50 or 125 mL LDPE bottles and analyzed via flow injection analysis (FIA) using luminol chemiluminescence without preconcentration (Croot and Laan, 2002) exactly as per Hopwood et al., (2017a). Fe(II) samples during the MesoPat/MesoArc/MicroPat/MultiPat/MultiArc experiments were 150 analysed immediately after sub-sampling from each individual mesocosm/microcosm/multistressor container. In Gran Canaria the warmer seawater temperature and distance between the experiment location and laboratory precluded immediate analysis. Therefore, prior to sampling, 10 µL 6 M HCl (Hiperpur-Plus) was added to the LDPE bottles in order to maintain the sampled seawater at pH 6 and thus minimize oxidation of Fe(II) between sample collection and analysis. For Gran Canaria only, opaque LDPE bottles were used to prevent further photochemical formation of Fe(II). The pH modification is 155 outlined in detail by  and is not thought to significantly affect in situ Fe(II) concentrations during the short time period between collection and analysis. Fe(II) was then quantified within 2 h of sample collection. In all cases Fe(II) was calibrated by standard additions (normally from 0.1-2 nM) using 100 or 600 µM stock solutions. Stock solutions were prepared from ammonium Fe(II) sulfate hexahydrate (Sigma-Aldrich), acidified with 0.01 M HCl and stored in the dark. A diluted Fe(II) stock solution (1-2 µM) was prepared daily. The detection limit varied slightly between FIA 160 runs from 90 pM (Gran Canaria) to 200 pM (Arc/Pat experiments).
Wavelength dispersive X-ray fluorescence (WDXRF) was conducted on triplicates of particulate samples collected by filtering 500 mL of seawater through 0.6 µm polycarbonate filters. After air-drying overnight, samples were stored in PetriSlide boxes at room temperature until analysis at the University of Bergen (Norway). Analysis via WDXRF 165 spectroscopy was exactly as described by Paulino et al., (2013) using a S4 Pioneer (Bruker-AXS, Karlsruhe, Germany).

Macronutrients and chlorophyll a
Dissolved macronutrient concentrations (nitrate, phosphate, silicic acid; filtered at 0.45 µm) were measured spectrophotometrically the same day as sample collection (Hansen and Koroleff, 2007). Nutrient detection limits inevitably 170 varied slightly between the different mesocosm/microcosm/multistressor experiments; however this does not adversely affect the discussion of results herein. Chlorophyll a was measured by fluorometry as per Welschmeyer (1994).

Carbonate chemistry
pH (except where stated otherwise, 'pH' refers to the total scale reported at 25ºC) was measured during the Gran Canaria 175 mesocosm using the spectrophotometric technique of Clayton and Byrne (1993) with m-cresol purple in an automated Sensorlab SP101-SM system and a 25ºC-thermostatted 1 cm flow-cell exactly as per González-Dávila et al., (2016). pH during MesoPat/MicroPat/MultiPat was measured similarly as per Gran Canaria using m-cresol. During MesoArc/MultiArc pH was measured spectrophotometrically as per Reggiani et al., (2016). For calculation of Fe(II) oxidation rates constants as per Santana-Casiano et al., (2005), pH free was calculated from measured pH using the sulphate dissociation constants derived 180 from Dickson (1990). compare Fe(II)/H 2 O 2 FIA data to discrete DFe/TdFe samples the mean of 7 FIA datapoints, corresponding to 14 minutes of sample intake and analysis time, was used.

Fe(II) decay experiments
A series of experiments was conducted during Meso/Micro/MultiPat, Meso/MultiArc (n=79), and under laboratory conditions using filtered Atlantic seawater (n=46) to investigate the change in Fe(II) concentration when water was moved 195 from ambient light into the dark. Fe(II) decay experiments were conducted inside the temperate controlled rooms hosting the MultiPat/MultiArc experiments. As such, a constant temperature was maintained throughout these experiments. Sub-samples for Fe(II) analysis or decay experiments were always collected when the mesocosms had been untouched (i.e. no sampling or additions) for >12 h, thus Fe(II) species could not plausibly have been directly perturbed by any external manipulation of the mesocosm/microcosm/multistressor experiments. After collection of unfiltered 1-2 L samples in transparent 2 L HDPE 200 containers, the PTFE FIA sample line was placed into the sample bottle and continuous analysis for Fe(II) and H 2 O 2 begun.
After a stable chemiluminescence response was obtained (typically 2-4 minutes after first loading the sample), the sample bottle was moved to an Al foil lined dark laminar flow hood and analysis continued for >1 h or until Fe(II) concentration fell below the detection limit (~0.2 nM). The time at which the sample was moved into the dark was designated t = 0.
Subsamples for the determination of DFe and TdFe were retained from this time point. 205 Theoretical decay rate constants (k') for these experiments were calculated using the formulation presented in Santana-Casiano et al., (2005) with measured pH, temperature, dissolved O 2 and salinity as per Eq. (1) where T is temperature (°K), pH is pH free and S is salinity (psu). O 2 saturation was calculated as per Garcia and Gordon (1992) and then k' was adjusted for measured O 2 concentrations as per Eq. (2). Measured rate constants (k meas ) were derived from the gradient of ln[Fe(II)] 210 against time for each decay experiment from at least 5 sequential datapoints (Fe(II) concentration was obtained at 2 minute intervals) for t 0-15 minutes. Dissolved oxygen was measured using an Oxyminisensor (World Precision Instruments). Salinity and temperature for each experiment were measured using a hand-held LF 325 conductivity meter (WTW). Fe(II) decay experiments under laboratory conditions used aged, filtered (0.2 µm) Atlantic water. This water was previously stored filtered in 1 m 3 trace element clean 215 HDPE containers for in excess of 1 year and maintained in the dark at experimental temperature for 3 days prior to commencing any experiment.

Quantifying the potential for Fe contamination during a mesocosm experiment
During MesoArc a 'bookkeeping' exercise was conducted for the mesocosm experiment by the sub-sampling of all solutions added to the incubated seawater. Aqueous additions consisted of: HCl solution (used to apply the pH gradient), 220 macronutrient solution, glucose solution and zooplankton. A short (1-2 h) 1 M HCl (trace metal grade) leach was applied to equipment placed within the mesocosm and also to the HDPE mesocosm containers prior to filling to provide a quantitative estimate of 'leachable' Fe. Atmospheric deposition of Fe into the tanks when open was estimated by deploying open bottles of de-ionized water within the vicinity of the mesocosms for fixed time intervals of 1 h in triplicate on 3 occasions and recording the approximate extent of time when the mesocosm lids were removed. All additions to the MesoArc mesocosm 225 experiment were volume weighted as per Eq. (3) using the mean (mid-experiment) mesocosm volume (V mesocosm ), and assuming that all additions were well mixed and TdFe behaved conservatively.

'Bookkeeping' Fe additions for a 1000 L mesocosm experiment (MesoArc) 230
In order to provide a rigorous assessment of Fe contamination during one experiment, Fe inputs were tracked in all additions to MesoArc and scaled to the mesocosm volume (initially 1200 L, declining by 15% over the experiment duration).Volume weighting all additions (Table 1) to the MesoArc mesocosm experiment as per Eq. (3) produced a total mean concentration of 48 nM TdFe (Fig. 1). In addition to the uncertain variability arising as the mesocosms were filled, approximately 8% (3.6 nM) of TdFe within the MesoArc experiment could be attributed to inadvertent addition (Fig. 1)   When MesoArc is compared to the two other mesocosms with a similar design (MesoPat and MesoMed) the TdFe inputs and the relative contribution of inadvertent TdFe addition were: 66.9 nM TdFe with 4.8% arising from inadvertent addition for MesoPat and 13.3 nM with 24% TdFe arising from inadvertent addition for MesoMed (Fig. 1). Systematic contamination was in all cases a minor, yet measurable, source of TdFe for these inshore mesocosms. Strictly, the inadvertent input of TdFe varied between different treatments within each mesocosm experiment due to, for example, the variable volume of glucose 245 solution used to create a DOC gradient. However, these differences caused small or negligible changes in TdFe addition ( on TdFe measurements at time zero from the MesoPat multistressor/microcosm and DSi measurements on experiment day 0 or 1 from multiple mesocosms.

General trends in Fe biogeochemistry; the MesoArc and MesoPat mesocosms
Concentrations of both DFe and H 2 O 2 (as per Hopwood, 2018) were measured at the highest resolution for the baseline treatments (no DOC addition, no zooplankton addition) during the mesocosm experiments. For MesoPat (Fig. 2), the initial 255 concentration of DFe and H 2 O 2 was estimated by using a Go-Flo bottle to sample at a depth of 10 m in the fjord (at which approximate depth the mesocosms were filled from). The apparent rise in H 2 O 2 between day 0 and day 1 (Fig. 2) likely reflects the result of increased formation of H 2 O 2 after pumping of water from ~10 m depth into containers at the surface.
NO 3 was added daily, hence concentrations increased prior to the onset of a phytoplankton bloom. The decline in DFe likely reflects biological uptake and/or scavenging onto particle (>0.2 µm) or mesocosm container surfaces. Less frequent temporal resolution was available for treatments other than the 'baseline' no DOC/zooplankton addition 265 treatment, but the decline in DFe during the MesoPat mesocosm was apparent across all measurements. In addition to TdFe measurements from unfiltered water samples, particulate (>0.6 µm) Fe concentrations were also determined from wavelength dispersive X-ray fluorescence. WDXRF data were normalised to phosphorus (P) in order to discuss trends in the elemental composition of particles and are thus presented as the Fe:P [mol Fe mol -1 P] ratio. The initial Fe:P ratio in particles varied between the mesocosm fieldsites: MesoPat 0.34 ± 0.09 and MesoArc 0.62 ± 0.07. A similar trend however was 270 observed during all experiments; a general decline in Fe:P across all treatments with time. Particulate Fe:P ratios on the final day of measurements were invariably lower than the initial ratio: MesoPat 0.09 ± 0.04, MicroPat 0.05 ± 0.01, MultiPat 0.07 ± 0.03, and MesoArc 0.17 ± 0.08. All of these ratios are high compared to literature values reported for offshore stations where the ratio for cellular material ranged from 0.005 to 0.03 mol Fe mol -1 P (Twining and Baines, 2013). However, this may simply reflect elasticity in Fe:P ratios which increase under high DFe conditions (Sunda et al., 1991;Sunda and 275 Huntsman, 1995). Alternatively, it could reflect the inclusion of a large fraction of lithogenic material, which would be expected to have a higher Fe:P ratio than biogenic material (Twining and Baines, 2013).
Particles from ambient waters outside the mesocosms were collected and analysed at the Patagonia and Svalbard fieldsites in order to assist in interpreting the temporal trend in Fe:P. Suspended particles from Kongsfjorden (Svalbard) exhibited a Fe:P 280 ratio of 3.01 ± 0.06 mol Fe mol -1 P and suspended particles in Comau fjord (Patagonia) varied more widely with a mean ratio of 0.54 ± 0.41. Kongsfjorden surface waters are characterised by extremely high TdFe concentrations originating from particle rich meltwater plumes (Hop et al., 2002) and thus the 3.0 Fe:P ratio can be considered to be a lithogenic signature.
After ambient water was collected for the mesocosm experiments, the steady decline in particle Fe:P ratios throughout the experiments likely resulted partially from a settling or aggregation of lithogenic material. At the same time, a decline in the 285 ratio of dissolved Fe:PO 4 during each experiment, due to the daily addition of PO 4 and minimal addition of new Fe, may also have led to reduced Fe uptake relative to P.

Fe(II) time series (Gran Canaria)
A key focus of this work was to determine the fraction of DFe present as Fe(II). During the Gran Canaria mesocosm, a detailed time series of Fe(II) concentrations was conducted. The timing of sample collection was the same daily (14:30 290 UTC) in order to minimise the effect of changing light intensity over diurnal cycles on measured Fe(II) concentrations. Over the duration of the Gran Canaria mesocosm, Fe(II) concentrations fell within the range 0.10-0.75 nM (Fig. 3a). On the first measured day (day -2) Fe(II) ranged from 0.13 nM (700 µatm pCO 2 treatment) to 0.63 nM (1450 µatm pCO 2 treatment) with an overall mean (± standard deviation) concentration of 0.41  0.12 nM. From day 9 to 20 strong variations were observed between treatments. Following nutrient addition on day 18, a phytoplankton bloom was evident in chlorophyll a data from 295 day 19 or 20 with chlorophyll a peaking on day 21 or later (Hopwood et al., 2018b). An increase in Fe(II) was then evident from days 20-29 under bloom and post-bloom conditions (Fig. 3b).

discontinued after leakage and exchange with surrounding seawater occurred on experiment day 3 and so no data is shown.
Contrasting days 1 and 29, Fe(II) concentration in all of the mesocosms except the 700 µatm pCO 2 treatment experienced a measurable increase (+0.4, +0.4, +0.2, +0.2, +0.2, 0.0 and +0.3 nM). The 700 µatm pCO 2 treatment was also anomalous with respect to slow post-bloom nitrate drawdown and elevated H 2 O 2 concentration (100 nM H 2 O 2 greater than other treatments under post-bloom conditions, Hopwood et al., 2018b). Overall, despite the large gradient in pCO 2 (400-1450 µatm and a 305 corresponding measured pH range of 8.1-7.7), Fe(II) showed no significant correlation with pH (Pearson Product Moment Correlation p 0.32) (Fig. 3a).

Fe(II) decay experiments (Meso/micro/multiPat and Meso/multiArc)
In a companion text presenting H 2 O 2 results from the same series of experiments (Hopwood et al., 2018a) Initial Fe(II) concentrations ranged from 0.3-16 nM. Generally a decline in Fe(II) was observed immediately after transferring this sampled water to a dark box, yet this was not always the case. The Fe(II) concentration more often than not remained measurable (> 0.2 nM) for the entire duration of the decay experiment. One hour after the transfer of water from 320 ambient conditions into the dark, Fe(II) was below detection on only 2 out of 79 occasions, and on average 55% of the initial Fe(II) concentration at t = 0 remained.
In order to account for the many physio-chemical parameters that affect Fe(II) oxidation rates, theoretical pseudo-first order rate constants (k') were calculated for each decay experiment assuming pseudo-first order kinetics (correlation coefficients 325 are noted for each linear regression-Supplementary Material). The rate constant, k (Eq. 1), thus accounts for the major effect of variations between experiments of salinity, temperature, pH and O 2 in a single constant (Fig. 4). Before comparing k meas and k, an estimate of the uncertainty should also be made as differences between the two values may arise due to the relatively large combined error from propagating the uncertainty in S/T/pH free /[O 2 ], and in analytical error on Fe(II) measurements. The accuracy of Fe(II) measurements is challenging to quantify for a transient species with no appropriate 330 reference material. In this case, the exact Fe(II) detection method used here was previously compared to another variation of the luminol chemiluminescence method (with pre-concentration, Bowie et al., 2002b) and k meas was determined with ±20% difference between two methods (Hopwood et al., 2017). The uncertainty on k meas is therefore assumed to be ±20% rather than the generally smaller uncertainty than can be calculated from a linear regression of ln[Fe(II)]. The uncertainty in calculated k was assessed by calculating the change resulting from the estimated uncertainty on measured salinity (±0.1), 335 temperature (±0.5°C), pH free (±0.05) and O 2 (±10 µM). The combined uncertainty is ±35% for k. Reduced uncertainties are possible with closed thermostat systems where the uncertainty on all physical/chemical parameters (S/T/pH/O 2 ) would be reduced, however our objective here was to measure the decay rates of in situ Fe(II) concentrations and thus the first priority was to commence measurements after sub-sampling rather than to stabilize physical/chemical conditions. 340 k, an additional set of experiments was conducted using aged, filtered Atlantic seawater (Fig. 4). The background concentration of Fe(II) in this water was below detection (<0.2 nM) and the initial DFe concentration relatively low (0.98 ± 0.39 nM). In a series of 46 decay experiments, Fe(II) spikes of 2-8 nM were added and then the decay in the dark monitored as per the Meso/micro/multi Arc/Pat in situ experiments. 345  Assembling and maintaining mesocosm scale experiments under trace-element clean conditions is a logistically challenging 355 exercise (e.g. Guieu et al., 2010) and thus it was desirable to conduct a thorough assessment of the extent to which Fe concentrations were subject to inadvertent increases during at least one experiment. All of the incubation experiments herein were conducted using coastal or near-shore waters. This is reflected in the low salinities of the MesoPat (27.5-28.0) and

Figure 4: A comparison of k meas and calculated k (both M -1 min -1 ) for Fe(II) decay experiments. (a) Rate constants for Fe(II) decay experiments from Meso/micro/multiPat (green), Meso/multiArc (blue) and spikes to aged Atlantic seawater (colourless). k meas =k (±35%) plotted for clarity (solid black line). (b) The difference between observed and calculated values of k (Δk = k meas -k) is shown
MesoArc (33.7-33.8) mesocosms. Both of these fieldsites were fjords with high freshwater input. Comau fjord (Patagonia, MesoPat) is situated in a region with high annual rainfall and receives discharge from rivers including the River Vodudahue. 360 Kongsfjorden (Svalbard, MesoArc) receives freshwater discharge from numerous meltwater fed streams and marine terminating glaciers in addition to melting ice. Correspondingly high DFe and TdFe concentrations were thereby found in surface waters; universally >4 nM DFe. The Gran Canaria (initial S 37.0) mesocosm cannot be considered to have had a coastal low salinity signature from large freshwater outflows, but was still conducted using near-shore waters which would generally be expected to contain higher Fe concentrations than offshore waters due to sedimentary sources of Fe (see, for 365 example, Croot and Hunter, 2000). Despite the inshore basis of the MesoArc mesocosm, Fe contamination was a small, but significant, fraction of the TdFe added to the starting water (8%, 3.6 nM, Fig. 1). It is not anticipated that this small TdFe addition will have had any adverse effect on the Fe redox chemistry results presented herein for the Meso/micro/multi Arc/Pat experiments.

Fe speciation within the mesocosms 370
Throughout all of the Meso/micro/multi Arc/Pat experiments, Fe(II) consistently constituted a large fraction of DFe ( Table   2). The presence of 24-65% of DFe in mesocosms as Fe(II) is not unexpected, as the photoreduction of Fe(III) species by sunlight is well characterized (Barbeau, 2006;Wells et al., 1991). Yet it also raises questions about how Fe speciation is modelled in these waters. DFe in the ocean is widely assumed to be characterised as "99% complexed by organic species" (Gledhill and Buck, 2012) on the basis of extensive research using voltammetric titrations to determine the strength 375 and concentration of Fe binding ligands (Gledhill and Van Den Berg, 1994;Rue and Bruland, 1995).
Yet these approaches exclusively measure Fe(III)-L species (Gledhill and Buck, 2012

385
Here we should note that the method utilized during these incubation and diurnal experiments, flow injection analysis with a PTFE line inserted directly into the experiment water, is relatively well suited for establishing the in situ concentration of Fe(II) (O'Sullivan et al., 1991). Such an experimental set up ensures no unnecessary delay is introduced between the collection and analysis of a sample. When using an opaque sampler, such as a Go-Flo bottle typically deployed at sea for collection of trace element samples (Cutter and Bruland, 2012), the collection process inevitably displaces near-surface 390 water from its ambient light conditions for a time period that constitutes >1 half-life of Fe(II) in warm, oxic seawater.
Measured near-surface Fe(II) concentrations on samples from a rosette system would therefore always be expected to underestimate in situ Fe(II) concentrations (O'Sullivan et al., 1991).
Fe(II) concentration was also quantified in ambient waters adjacent to the mesocosms and found to constitute a lower 395 fraction of DFe (2-11%). Most of the decay experiments, from which initial Fe(II) concentrations are reported (Table 2), were conducted at the end of mesocosm/microcosm/multistressor experiments and thus it is not possible to assess the development of Fe(II) stability throughout a phytoplankton bloom. Nevertheless, the high fraction of DFe present as Fe(II) in these experiments ( Table 2) relative to that observed in ambient waters is consistent with the increase in Fe(II) concentrations observed in Gran Canaria after the initiation of the phytoplankton bloom (day 19 onwards, Fig. 3b). The 400 Meso/micro/multi Arc/Pat experiments had macronutrient additions daily, whereas the Gran Canaria experiment had macronutrient addition only on day 18. The conditions within the Meso/micro/multi Arc/Pat experiments during the time period which decay experiments were conducted were therefore typical of those during, or shortly after, a phytoplankton bloom. Whilst chlorophyll a was not quantified for ambient waters, for which Fe(II) data are reported (Table 2), sampling in Svalbard (MesoArc, July 2015) and Patagonia (MesoPat, November 2014) occurred during relatively low productivity 405 phases relative to the annual cycle in primary production at these fieldsites (Hop et al., 2002;Iriarte et al., 2013). The

Comment [HM6]:
Description edited ambient concentrations of Fe(II) measured at the mesocosm experiment fieldsites are therefore not necessarily directly comparable to Fe(II) concentrations measured after nutrient addition in the corresponding mesocosm experiments.

Fe(II) decay experiments
Fe(II) oxidation rates are relatively well constrained in seawater with varying temperature, salinity, pH, H 2 O 2 and O 2 410 concentration from extensive series of experiments where the change in concentration of an Fe(II) spike was monitored with time and the rate constants for oxidation with O 2 and H 2 O 2 then derived from first order kinetics (King et al., 1995;Millero et al., 1987). Whilst dissolved O 2 is the dominant oxidizing agent for Fe(II), H 2 O 2 is also of importance as an Fe(II) oxidizing agent in surface seawater (Millero and Sotolongo, 1989;King and Farlow, 2000;González-Davila et al., 2005).
The unusually low concentration of H 2 O 2 within the Meso/micro/multi Arc/Pat experiments due to the enclosed HDPE 415 mesocosm design and/or synthetic lighting (Hopwood et al., 2018a) was therefore fortunate from a mechanistic perspective as it allows the simplification that O 2 was the only major oxidising agent. The much lower H 2 O 2 concentrations (1-79 nM) present, compared to ambient surface waters, throughout the Meso/micro/multi Arc/Pat experiments should mean that Fe(II) decay rates during these experiments more closely match the oxidation rate constants used to derive Eq. 1 (which were derived for low-H 2 O 2 conditions). 420 The decay experiments reported here still however differ in two critical respects from controlled oxidation rate experiments used to derive rate constants. First, the speciation of Fe(II) may differ. It is debatable to what extent Fe(II)-L species, analogous to Fe(III)-L species, exist in surface marine waters due to the absence of reliable techniques to probe Fe(II)organic speciation (Statham et al., 2012). Yet there is consistent evidence that organic material affects Fe(II) oxidation rates 425 (see below). Second, these decay experiments measure the change in Fe(II) concentration between light and dark conditions and not specifically the oxidation rate. If photochemical Fe(II) production was the sole Fe(II) source, and oxidation of Fe(II) via H 2 O 2 and O 2 were the only Fe(II) sinks, then the decay rate measured here would approximate the oxidation rate determined under controlled laboratory conditions. However, there are possible biological sources of Fe(II) (Sato et al., 2007;Nuester et al., 2014), the possibility of biological uptake of Fe(II) (Shaked and Lis, 2012) and cross-reactivity with 430 other reactive trace species (e.g. reactive oxygen species and Cu, Rijkenberg et al., 2006, Croot andHeller, 2012) to consider. These complexities make Fe(II) more challenging to model in natural waters compared to controlled conditions. This is especially the case at the low Fe(II) concentrations relevant to the surface ocean where Fe(II) concentrations range from below detection up to ~1 nM (Gledhill and Van Den Berg, 1995;Sarthou et al., 2011).

435
Contrasting k with k meas during Fe(II) decay experiments (Fig. 4), it is immediately apparent that the Fe(II) present within Meso/Micro/Multi Arc/Pat experiments was generally much more stable than would be predicted for an equivalent inorganic spike of Fe(II) added to water with the same physical/chemical properties i.e. in most cases k meas < k. Three plausible hypotheses can be conceived for the offset: i.
The measured rates here refer to relatively low initial Fe(II) concentrations (0.3-16 nM) compared to the concentrations at which rate constants have been derived (typically ~20-200 nM) and the difference arises simply because the rate constants are not calibrated for low nanomolar starting concentrations.
ii. There is 'dark' production of Fe(II) in the experiments i.e. on-going formation of Fe(II) counter-acts the first order decay of Fe(II) via oxidation. 445 iii.
The speciation of Fe(II) in seawater is more stable with respect to oxidation than the species for which the rate constants are calculated.
For the series of experiments using spikes of Fe(II) in Atlantic seawater, k meas is consistently closer to k than for any in situ experiments (Fig. 4a). Nevertheless, some datapoints for spiked Atlantic seawater still fall outside the ±35% uncertainty 450 boundary. Yet, as the spiked experiments closely matched the initial Fe(II) concentrations in the in situ decay experiments, the higher Fe(II) concentrations generally used to establish the rate of Fe(II) decay in laboratory experiments cannot be a major factor in the discrepancy between k meas and k. Furthermore, differences in the formulation of k' between studies are relatively minor (Millero et al., 1987;King et al., 1995;Santana-Casiano et al., 2005).

455
Calculating the difference between calculated and measured k (Δk), it is evident that the largest differences were associated with the lowest initial Fe(II) concentrations (Fig. 4b). This is consistent with both hypothesis II and III. Assuming that the dominant source of Fe(II) is photochemistry, the effects of both a secondary 'dark' Fe(II) source and a limited fraction of Fe(II) existing in a more stable form with respect to oxidation would be most evident at the lowest initial Fe(II) concentration. Sources of Fe(II) other than photochemistry are plausible and may include, for example, zooplankton grazing 460 due to the reduced pH and O 2 within organisms' guts (Tang et al., 2011;Nuester et al., 2014). Mesozooplankton addition was one of the three experimental variables manipulated during the Arctic/Patagonia experiments. However, no clear trend was evident with respect to Δk and the zooplankton addition status of the experiments. Mean Δk ± SD (×10 -2 ) for the high/low zooplankton treatments over all experiments were 4.66 ± 5.79 and 4.08 ± 5.63, respectively. A dependency of Δk on the initial Fe(II) concentration (Fig. 4b), with [Fe(II)] t=0 likely very sensitive to multiple experimental factors such as the 465 time of day that the sample was collected and the exact time delay between sample collection and the first timepoint for each Fe(II) decay experiment, would however make determining the relative importance of any other underlying causes challenging. In order to gain further insight into the potential role of zooplankton in Fe(II) release under dark conditions, a series of incubations was conducted with addition of the copepod Calanus finmarchichus to cultures of the diatom Skeletonema costatum (Hopwood et al., 2018a). No change in extracellular Fe(II) or H 2 O 2 concentrations were evident 470 across a gradient of copepods from 0-10 L -1 . Whilst this suggests the role of high/low zooplankton treatments was minimal in short-term changes to ambient Fe(II) concentrations, the potential release of Fe(II) by zooplankton may of course be species specific; different results may have been obtained with different zooplankton-prey combinations.
The high magnitude of Δk in some cases at low initial Fe(II) concentrations (Fig. 4) is consistent with the theory that Fe(II) 475 binding ligands are responsible for the observed stability of Fe(II) in some natural waters (Roy and Wells, 2011;Statham et al., 2012). The Fe(II)-binding capacity of any ligands present in a specific sample would be expected to become saturated as Fe(II) concentrations increased. Similarly, strong Fe(II) binding sites would be occupied first, and thus the strength of the Fe(II)-L interaction would decrease with increasing Fe(II) concentration even prior to ligand saturation. The effect of Fe(II) ligands on the oxidation rate of an added Fe(II) spike would therefore become less evident as Fe(II) concentration increased 480 because the fraction of Fe(II) present as Fe(II)-L species would decline i.e. Δk would approach zero. This has an important methodological implication. The effect of cellular exudates, or natural organic material extracts, on Fe(II) oxidation rate is more often than not tested by adding reasonably high nanomolar Fe(II) spikes to solution and then following the Fe(II) decay with time (see, for example, Lee et al., 2017). By raising the initial Fe(II) concentration, such an approach may however systematically under-estimate the effect of organic material on Fe(II) stability at in situ Fe(II) concentrations. 485 The effect of organic material on Fe(II) is difficult to generalize as organic compounds can accelerate, retard or have no apparent effect on Fe(II) oxidation rates via oxygen (Santana-Casiano et al., 2000). However, there are now sufficient studies of Fe(II) behaviour to distinguish between the broad effects of allochthonous and autochthonous material. Extracts from the green algae Dunaliella tertiolecta (González et al., 2014), cyanobacteria Synechococcus (Samperio-Ramos et al., 2018b) and 490 Microcystis aeruginosa (Lee et al., 2017), coccolithophore Emiliania huxleyi (Samperio-Ramos et al., 2018a), and diatoms Chaetoceros radicans (Lee et al., 2017) and Phaeodactylum tricornutum (Santana-Casiano et al., 2014) have all been found to retard Fe(II) oxidation rates. Furthermore, the effect of cellular exudates on the reaction constant appears to scale with increasing total organic carbon (Samperio-Ramos et al., 2018b). In contrast to the stabilization apparent in some cellular exudates, allochthonous material generally, although not universally, has the opposite effect with an acceleration of Fe(II) 495 oxidation rates reported both in coastal environments (Lee et al., 2017) and using terrestrially derived organic leachates (Rose and Waite, 2003). The generally positive effects of cellular exudates on Fe(II) stability with respect to oxidation determined in single-species studies is consistent with the stability of Fe(II) observed in almost all experiments here (Fig. 4) and this suggests that microbial cellular exudates are indeed a stabilizing influence on Fe(II) concentrations at a broad scale in coastal marine environments. Stabilization of Fe(II) by freshly produced exudates could explain the sustained increase in 500 Fe(II) concentrations across all pCO 2 treatments under post-bloom conditions in Gran Canaria (Fig. 3b) and the high fraction of DFe present as Fe(II) during all Meso/micro/multi Arc/Pat experiments (Table 2).
Apart from the influence of organic Fe(II) ligands on Fe(II) stability arising from the slower oxidation rates of some organically complexed Fe(II) species, Fe(II) binding organics may also have a role in the generation of superoxide (O 2 -) 505 which is speculated to be a dominant mechanism for the formation of Fe(II) in the dark (Rose, 2012). Experiments with 65-130 nM of protoporphyrin IX demonstrated increased formation of Fe(II) in the dark with both increasing porphyrin concentration and increasing irradiation of seawater prior to the onset of darkness (Rijkenberg et al., 2006). Whilst the rates of this process are challenging to investigate at the sub-nanomolar porphyrin and Fe(II) concentrations expected in most marine environments, the dark formation of Fe(II) mediated by reactive oxygen species interacting with Fe(II)-organic 510 complexes could potentially be important in both the diurnal cycling of Fe in the surface ocean and the non-photochemical formation of Fe(II) in the dark of the ocean's interior (Rose, 2012). From a mechanistic perspective, it is challenging to establish definitively from the experiments herein whether apparent Fe(II) stability arises from reduced oxidation rates due to Fe(II) complexation, or dark Fe(II) formation via a mechanism, such as that proposed for superoxide, which involves Fe(II)organic complexes. Both hypothesis are consistent with field observations and it is also possible that both processes operate 515 in parallel, which may contribute to the variability in Δk and the variable quality of the assumed pseudo-first order kinetic fit for decay experiments.

Conclusions
The existence of a high fraction (24-65%) of DFe as Fe(II) during mesocosm experiments, and the apparent stability of low 520 concentrations of Fe(II) suggests that the classic characterisation of '99% of dissolved Fe existing as Fe(III)-L complexes' (Gledhill and Buck, 2012) is inadequate to describe DFe speciation in coastal surface waters. Fe(III)-ligand complexes may overwhelmingly dominate Fe speciation in the ocean as a whole, but in coastal waters a dynamic redox cycle operates maintaining considerable concentrations of Fe(II) in solution. The stabilizing effects on Fe(II) with respect to oxidation reported here were strongest at low (<2 nM) Fe(II) concentrations suggesting that the Fe(II) stabilization mechanism is 525 caused by a process akin to complexation where the magnitude of the effect is capped by a factor other than physical conditions.
Exudates stabilizing Fe(II) may be a poorly characterized component of the aptly named 'ferrous wheel' (Kirchman, 1996;Strzepek et al., 2005) and contribute to the efficient recycling of DFe within marine surface waters. Irrespective of whether 530 Fe(II) is more or less bioavailable relative to Fe(III), the formation of Fe(II) is a mechanism for increasing DFe and thus increasing DFe availability to biota. Mechanisms such as the stabilization of Fe(II) by cellular exudates during and after phytoplankton blooms may therefore facilitate DFe uptake to a greater extent than would be possible in the absence of Feredox cycling. Both Fe(III) and Fe(II) speciation and concentration must therefore be defined in order to comprehensively understand the role of Fe as a driver of marine primary production. 535

Author Contributions
All authors contributed to the design of the study and the interpretation of data. MH, CS, JG, NS, ØL and TT conducted analytical work. MH coordinated the writing of the manuscript with input from other authors.

Acknowledgements
The Ocean Certain and KOSMOS/PLOCAN teams assisting with all aspects of experiment logistics and organisation are 540 thanked sincerely for their efforts. Labview software for operating the H 2 O 2 FIA system was designed by P Croot, M Heller, C Neill and W King. Financial aid from the European Commission (OCEAN-CERTAIN, FP7-ENV-2013-6.1-1; no: 603773), DFG (Collaborative Research Centre 754 Climate-Biogeochemistry Interactions in the Tropical Ocean) and Ministerio de Economía y Competitividad of the Spanish Government (EACFe project, CTM2014-52342-P) is gratefully acknowledged. 545