Nutrient regulation of late spring phytoplankton blooms in the midlatitude North Atlantic

The duration and magnitude of the North Atlantic spring bloom impacts both higher trophic levels and oceanic carbon sequestration. Nutrient exhaustion offers a general explanation for bloom termination, but detail on which nutrients and their relative influence on phytoplankton productivity, community structure, and physiology is lacking. Here, we address this using nutrient addition bioassay experiments conducted across the midlatitude North Atlantic in June 2017 (late spring). In four out of six experiments, phytoplankton accumulated over 48–72 h following individual additions of either iron (Fe) or nitrogen (N). In the remaining two experiments, Fe and N were serially limiting, that is, their combined addition sequentially enhanced phytoplankton accumulation. Silicic acid (Si) added in combination with N + Fe led to further chlorophyll a (Chl a) enhancement at two sites. Conversely, addition of zinc, manganese, cobalt, vitamin B12, or phosphate in combination with N + Fe did not. At two sites, the simultaneous supply of all six nutrients, in combination with N + Fe, also led to no further Chl a enhancement, but did result in an additional 30–60% particulate carbon accumulation. This particulate carbon accumulation was not matched by a Redfield equivalent of particulate N, characteristic of high C:N organic exudates that enhance cell aggregation and sinking. Our results suggest that growth rates of larger phytoplankton were primarily limited by Fe and/or N, making the availability of these nutrients the main bottom‐up factors contributing to spring bloom termination. In addition, the simultaneous availability of other nutrients could modify bloom characteristics and carbon export efficiency.

The mid-to-high latitude North Atlantic is characterized by a springtime peak in phytoplankton concentrations and sinking of organic matter, which transfers around 0.5-2 GtC yr −1 to the ocean interior (Sanders et al. 2014). Key mechanisms underlying this ecological event have been identified: in winter, phytoplankton growth is light-limited, while deep surface mixed layers act to reduce losses to grazing (Behrenfeld and Boss 2018); In spring, stratification, higher solar angles, and longer daylength allow for greater light availability for photosynthesis, enhancing growth rates (Sverdrup 1953;Mahadevan et al. 2012;Mignot et al. 2018). Together this leads to surface phytoplankton accumulation (Behrenfeld and Boss 2018). Surface macronutrient removal scales with carbon drawdown and peaks over several weeks in spring (Buesseler et al. 1992;Körtzinger et al. 2008). As in many other oceanographic regions (Agustí et al. 2015), sinking phytoplankton-which mediate the surface to deep ocean carbon flux-are dominated by bloom-forming diatoms (Billett et al. 1983;Lampitt 1985;Smetacek 1985Smetacek , 1999Alldredge and Gotschalk 1989;Turner 2002;Buesseler and Boyd 2009;Martin et al. 2011;Rynearson et al. 2013).
Multiple studies have invoked nutrient availability to explain the regulation of open ocean North Atlantic diatom blooms following relief of light limitation in early spring. However, there is a lack of consensus on which combinations of nutrients are limiting. In the high latitude Irminger Basin, community-level Fe stress has been found in summer-demonstrated by enhanced apparent photochemical efficiencies of photosystem II (PSII) (F v /F m ) and chlorophyll a (Chl a) biomass following Fe supply (Nielsdóttir et al. 2009;Ryan-Keogh et al. 2013). Phytoplankton growth following Fe supply has also been found to deplete N to secondary limiting levels, that is, where Fe + N supply leads to enhanced phytoplankton accumulation over supply of Fe only (Ryan-Keogh et al. 2013). In addition, low concentrations of Si have been implicated as limiting to diatoms, and zinc (Zn), cobalt (Co), and vitamin B 12 depletion have been suggested as potentially having coregulatory roles (Lochte et al. 1993;Sieracki et al. 1993;van den Berg 2000, 2001;Allen et al. 2005;Henson et al. 2006;Panzeca et al. 2008;Leblanc et al. 2009;Martin et al. 2011). Supporting the potential for multiple resource controls, recent GEOTRACES surveys suggest that multiple macro and micronutrients could be simultaneously approaching springtime codeficiency in North Atlantic seawater (Moore 2016). In summary, while transient blooms of diatoms might be critical for carbon and nutrient budgets in the North Atlantic, evidence is lacking to support specific constraints on their growth.
In order to constrain potential nutrient controls on phytoplankton blooms experimentally, we conducted bioassays during a cruise traversing the midlatitude North Atlantic (43-53 N) in late spring. Based on calculations of resource deficiency using available (micro-)nutrient data (Moore 2016), alongside prior experimental observations (Blain et al. 2004;Moore et al. 2006;Nielsdóttir et al. 2009;Ryan-Keogh et al. 2013), we predicted a priori that either Fe or N would be either primary limiting, colimiting, or serially limiting to both overall system productivity and diatom growth throughout the study region. As such, we tested for primary, secondary, or colimitation by these two nutrients by supplying these nutrients alone and in combination. Then, in parallel treatments with N + Fe limitation fully relived, we tested for co-/tertiary limitation of bloom formation by supplying six extra nutrients potentially depleted to limiting levels: Phosphorus (P), Si, Co, manganese (Mn), Zn, and vitamin B 12 . Finally, we tested for co-/serial limitation by more than three nutrients at the same time by supplying all nutrients in combination.

General
Fieldwork was conducted on the RRS Discovery in June 2017 (DY080), along a cruise track between Southampton (UK) and St Johns (Canada) (Fig. 1). Seawater was pumped from 2 m depth into a trace-metal-clean laboratory (over-pressurized with high efficiency particulate air (HEPA)-filtered air) from a towed water sampling device (towed fish) via acid-washed tubing and a Teflon bellows pump (Dellmeco) powered by filtered compressed air. Conductivity-Temperature-Depth casts to 500 m depth were performed throughout the cruise track. Mixed layer depths were calculated as the depth of 0.03 kg m −3 density changes relative to a 10 m reference density value (Montegut et al., 2004). Daily satellite-derived Chl a fields (4 km resolution, L3 standard mapped images) from the Visible and Infrared Imager/Radiometer Suite (VIIRS) instrument (Ocean Color Index, OCI, algorithm) were downloaded from the NASA Ocean Color website for the period April-July 2017.

Bioassay experiments
Experiments followed previously published protocols (Browning et al. 2017). Unscreened seawater, pumped from the trace-metal-clean sampling device, was used to fill a 60-liter carboy, agitated to homogenize, and dispensed in random order into 2-liter (control, N + Fe) or 1-liter (all remaining treatments) acid-washed polycarbonate bottles (Nalgene). Before and/or after filling the 60-liter carboy, samples were collected for dissolved macronutrient and trace element analysis (see below). Approximately 8-liter from the 60-liter carboy was collected at the beginning and end of the incubation bottle filling procedure in order to measure initial phytoplankton concentrations, community structure, and photophysiology (see below). Incubation bottles were spiked in triplicate with nutrient combinations to the following amended concentrations: N: 1 μmol L −1 NaNO 3 + 1 μmol L −1 NH 4 Cl; P: 0.2 μmol L −1 NaH 2 PO 4 ; Si: 1 μmol L −1 Na 2 O 3 Si; Fe: 2 nmol L −1 FeCl 3 , Mn: 2 nmol L −1 MnCl 2 ; Zn: 2 nmol L −1 ZnCl 2 , Co: 2 nmol L −1 CoCl 2 ; vitamin B 12 : 100 pmol L −1 . Macronutrient and B 12 nutrient stocks were previously passed through a prepared column of cation exchange resin to remove trace element contamination (Chelex, BioRad). Fe, Co, Mn, and Zn were prepared from 99+% purity salts, stabilized in 0.01 mol L −1 HCl (Fisher Optima grade HCl diluted in Milli-Q water). Bottles were capped, sealed with Parafilm, bagged, and placed in an on-deck incubator for 48-72 h. The incubator was screened with blue filters (Lee Filters "Blue Lagoon"), to provide an irradiance around 35% of above surface values, and was subjected to continuous seawater exchange from the ship's underway flow-through supply. Following the incubation period, bottles were removed and 500 mL of each triplicate replicate was subsampled for collection/analysis of Chl a, Fast Repetition Rate fluorometry, and flow cytometry. The remaining 500 mL from each triplicate treatment was pooled for collection of samples for high-performance liquid chromatography (HPLC)-derived pigments (0.6-1 L), particulate organic carbon/nitrogen (POC/PON) (0.2-0.5 L), biogenic silicate (BSi) (0.14-0.6 L), and microscopy (60 mL). For the control and N + Fe treatments only, samples for macronutrients were also filtered through precleaned (0.1 mol L -1 Fisher Optima grade HCl soak for 48 h followed by deionized water rinse) 0.2 μm Sterivex filter units (Millipore) under a laminar flow hood.

Macronutrients and trace elements
Dissolved trace element samples were collected in 125 mL low-density polyethylene bottles following direct in-line filtration (0.45 + 0.2 μm Sartorius Sartobran 300 filter units) from the Teflon pump-powered seawater supply. Samples were acidified within 24 h with 140 μL concentrated HCl (Optima Grade, Fisher). Samples were preconcentrated on a SeaFAST device and analyzed for Fe concentrations using an Element XR ICP-MS (directly following the method of Rapp et al. 2017). The detection limit for Fe using this method is 28.8 pmol L −1 . Macronutrient samples (nitrite plus nitrate; hereafter referred to as simply as nitrate, phosphate, and silicic acid) were collected 0.2 μm filtered, as for trace element samples, into acid-washed 50 mL polypropylene tubes and frozen directly at −20 C. Concentrations were determined upon return to a land-based laboratory following thawing for 24 h at 5 C and analysis using a nutrient autoanalyzer system (QuAAtro, SEAL Analytical). Detection limits for the macronutrient analyses, calculated as three times the standard deviation of a blank measured during the analytical run, were: nitrate + nitrite = 0.08 μmol L −1 ; phosphate = 0.03 μmol L −1 ; and silicic acid = 0.09 μmol L −1 .

Phytoplankton biomass and community structure
Samples for Chl a were filtered (100 mL) onto 25 mm diameter glass microfibre filter (GFF) filters, extracted in the dark for approximately 24 h in 10 mL 90% (by volume) acetone at −20 C, and analyzed on a precalibrated Turner Designs Trilogy laboratory fluorometer (Welschmeyer 1994). Samples for flow cytometry analysis were fixed with 1% paraformaldehyde (final concentration), mixed with a vortex, left for 10 min at room temperature in the dark, and then placed directly into a −80 C freezer. Samples were analyzed on return to shore using a FACSCalibur flow cytometer (Becton Dickenson, Oxford, United Kingdom). Cell counts were determined using CellQuest software (Becton Dickenson) following procedures described in Davey et al. (2008). Identification of divisions between pico-and nanophytoplankton was aided by analysis of 2 μm filtrates for control and N + Fe samples in each experiment (N + Fe samples typically showing enhanced fluorescence per cell).
Samples for HPLC were filtered through 25 mm diameter GFF filters and placed directly into a −80 C freezer. Following the ship's return to port, samples were transported to the laboratory on dry ice but thawed during transport. The duration in the thawed state before refreezing to −80 C was between 0 and 12 h. Samples were, nevertheless, analyzed following the method of Van Heukelem and Thomas (2001). Samples were extracted in 90% acetone in plastic vials using glass beads in a cell mill, centrifuged (10 min, 5200 rpm, 4 C), and then the supernatant was filtered through 0.2 μm polytetrafluoroethylene (PTFE) filters (VWR International) and analyzed by reverse-phase HPLC (Dionex UltiMate 3000 LC system, Thermo Scientific). Pigment standards were acquired from Sigma-Aldrich (USA) and the International Agency for 14C Determination (Denmark). Recovery of total Chl a (Chl a + divinyl Chl a) from the HPLC analysis was approximately 20% in comparison to parallel, fluorometrically determined concentrations performed on ship. However, within the HPLC pigment data set itself, Chl a correlated near-linearly with summed diagnostic pigments (DP) (Chl a = 1.1 × DP − 0.01; R 2 = 0.78; p < 1 × 10 −15 ; n = 195), suggesting, despite notable pigment degradation, a relatively internally consistent data set (Trees et al. 2000). Regardless, all HPLC pigment data were interpreted only qualitatively via approximation of phytoplankton size groups (Uitz et al. 2006).
Samples from select nutrient treatments were preserved for microscopy analysis (60 mL; amber HDPE Nalgene bottles) using 0.2 mL alkaline Lugol's solution. Cells from 10-50 mL samples were concentrated in settling chambers for 24 h, then viewed and photographed using an inverted microscope (Zeiss Axiovert) at ×100 and ×200 magnification.

POC/PON and BSi
Samples for POC/N were filtered through combusted (6 h at 450 C) 25 mm diameter GFF filters. Filters were then ovendried on ship at 50 C for 24 h. Filters were subject to sulfurous acid fuming to remove the contribution of particulate inorganic carbon, re-dried at 50 C for 24 h and then pelleted in tin boats (Elementar). Analyses were conducted using an elemental analyzer (Euro Elemental Analyser), using a set of known masses of acetanilide as standards. The contribution of particulate inorganic nitrogen to total particulate nitrogen measured was assumed to be negligible. Samples for BSi were filtered through 25 mm diameter 0.8 μm pore size polycarbonate filters (Whatman Nuclepore) and oven-dried on ship at 50 C for 24 h. Samples were digested in 4 mL 0.2 mol L −1 NaOH (Sigma-Aldrich) at 90 C for 2 h in acid-washed 15 mL polypropelene tubes (Rotilabo, Carl Roth). Following cooling, samples were neutralized with 10 mL 0.1 mol L −1 HCl (Fisher Optima grade, diluted in deionized water) and analyzed for dissolved silicic acid on a SEAL QuAAtro nutrient autoanalyzer system (SEAL Analytical). Digested blank filters yielded silicic acid concentrations below the detection limit (< 0.1 μmol L −1 ).

Phytoplankton photophysiology
A FASTOcean Fast Repetition Rate fluorometer (Chelsea Technologies Group) was used to characterize communitylevel phytoplankton photophysiology. A protocol of 100 × 1 μs blue flashes with 1 μs dark intervals (excitation phase) followed by 25 × 1 μs flashes with 84 μs dark intervals (relaxation phase) was used. Fluorescence transients were fit automatically in FASTPro8 software using the model of Kolber et al. (1998) to yield values of minimum fluorescence (F o ) and maximum fluorescence (F m ). Blank fluorescence was determined on 0.2 μm filtrates and was subtracted from F o and F m before calculation of

Results and discussion
Occupation in relation to bloom timing Satellite observations indicated that the ship-board experiments were conducted around, or post, peak springtime surface Chl a concentrations, confirming bloom climax or postbloom conditions (Fig. 1b,c).

Limitation by iron and/or nitrogen
Plankton biomass (POC) accumulated and nitrate was drawn down in untreated control bottles at all experimental sites apart from Experiment 2 where POC decreased ( Fig. 2; Table 1). Given that in situ light availability was expected to be replete (i.e., critical depths > mixed layer depths; Siegel et al. 2002), this implied plankton loss processes were reduced upon incubation (i.e., via grazing, viral lysis, dispersal), therefore allowing net plankton accumulation independent of changes in phytoplankton specific growth rates (e.g., Cullen 1991;Banse 2002;Behrenfeld and Boss 2018). However, following experimental supply of either fixed N (Experiments 1-2) or Fe (Experiments 3-6), Chl a and POC concentrations increased significantly beyond untreated controls (Fig. 2a-l). Alongside coregulation by loss processes, these responses therefore provided evidence for N or Fe limitation of the overall phytoplankton yield at all experimental sites (i.e., Liebig limitation; de Baar 1994). Furthermore, provided phytoplankton losses did not differ between treatments (i.e., via differences in grazing rates), these responses also suggested nutrient limitation of the community-level specific growth rate (i.e., Blackman limitation; Cullen 1991;de Baar 1994;Moore et al. 2013).
The geographic pattern of responses to N and Fe supply in relation to the regional nitrate gradient was consistent with previous studies in the North Atlantic (Fig. 1a). Collectively, these studies have suggested a transition from Fe limitation at higher latitudes in spring/summer (which host residual nitrate; Fig. 1a), to year-round N limitation in the subtropical gyre to the south (Graziano et al. 1996;Blain et al. 2004;Moore et al. 2006Moore et al. , 2008Nielsdóttir et al. 2009;Ryan-Keogh et al. 2013). Accordingly, transitions in the primary limiting nutrients, between either N or Fe, were in turn predictable on the basis of the initial concentrations of nitrate alone, the dissolved ratio of nitrate to Fe (i.e., N:Fe), or calculated values of Fe* (see Table 1; Browning et al. 2014Browning et al. , 2017Browning et al. , 2018Moore 2016). Conditions approaching N-Fe co-limitation were also apparent in Experiments 2 and 6 ( Fig. 2b,f; Browning et al. 2017;Ryan-Keogh et al. 2013). Specifically, serial Chl a responses observed following N + Fe supply were significantly higher than those following the supply of the primary limiting nutrient alone. Community-level photophysiological responses to nutrient supply, in terms of the apparent PSII photochemical efficiency (F v /F m ) and functional absorption cross section (σ PSII ), also exhibited a marked transition between N-and Fe-limited regimes (Fig. 2m-r; Table 1). F v /F m showed significant increases following Fe addition relative to controls at Fe-limited sites (Experiments 3-6) and restricted changes following N or Fe addition at Nlimited sites (Experiments 1-2). σ PSII responded with changes inverse to F v /F m , with the two parameters negatively correlated across the experimental dataset (R 2 = 0.76, p < 1 × 10 −15 , n = 192). These responses are consistent with prior observations, likely regulated by a mechanism involving phytoplankton accumulating fluorescent light harvesting complexes that are energetically decoupled from PSII under Fe limited, N sufficient conditions (leading to low F v /F m ) and their rapid connection to an increasing PSII pool following Fe supply (increasing

Community response to supply of limiting nutrients
In the shorter duration experiments in the eastern portion of the cruise track (Experiments 1-3; Table 1; Fig. 1 ). For all panels, arrows point to initial conditions. For panels (a-f) and (m-r), bars labeled with different letters indicate significantly different means (ANOVA p < 0.05 followed by a Tukey honesty significant difference test).

Browning et al. Nutrient regulation North Atlantic bloom
phytoplankton appeared to be initially dominated by nanophytoplankton, which, matching bulk Chl a responses, generally increased in number and intracellular Chl a content following N and/or Fe supply relative to untreated controls (Figs. 2a-c, 3a-c). The initial situation of few larger cells at these sites, including low diatom abundance (see Supporting Information), together with relatively short incubation timescales ( 48 h), collectively appeared to restrict both large absolute increases in phytoplankton biomass and accompanying changes in the size structure of phytoplankton communities following nutrient amendment (e.g., Chisholm 1992;Price et al. 1994). Conversely, the longer duration experiments (Experiments 4-6; Table 1) were initially codominated by nano-and microphytoplankton, with a greater abundance of diatoms (Fig. 2d-f; Supporting Information). At these sites, amendment with limiting nutrients (Fe or Fe + N) also enhanced abundances of pico-and nanophytoplankton (Barber and Hiscock 2006), but increases in microphytoplankton fractions were much larger, matching Chl a and POC biomass accumulation (Figs. 2d-f, 3d-f). Greater contributions of the pigment fucoxanthin (data not shown) suggested that the enhanced biomass was attributable to diatoms. This was confirmed by increased dominance of large Thalassionemataceae colonies and Guinardia chains ( 100s μm length) in microscopy samples, which were much less common in controls (with the latter diatom assemblage mostly comprising smaller [ 10s μm] individual Pseudo-nitzschia cells; see Supporting Information). The responses observed in Experiments 4-6 suggested that the pre-existing larger celled diatom community was experiencing greater Fe limitation in situ than the smaller celled community (Cullen 1991;Price et al. 1994;Ryan-Keogh et al. 2013). Following Fe supply however, adaptations enabling higher maximum potential growth rates (Marañón et al. 2012), rapid uptake and storage of nutrients under conditions of high nutrient availability (Raven 1987;Marchetti et al. 2009Marchetti et al. , 2012, together with lower susceptibility to grazer control (Smetacek 1999), presumably enabled their disproportionate accumulation and the accompanying drawdown of available nitrate (Table 1; de Baar et al. 2005). Si,P,Co,vitamin B 12 ,Zn,or Mn No clear serial biomass responses to any of the additional (micro)nutrients (P, Si, Mn, Co, Zn, B 12 ), supplied in combination with N + Fe, were observed in Experiments 1-4. This suggested that these nutrients were not deficient enough in the ambient seawater, nor the incubated seawater supplemented  with N + Fe, to markedly restrict phytoplankton accumulation over experimental timescales. In contrast, in Experiments 5 and 6 small additional Chl a enhancements were observed upon supply of N + Fe + Si (Fig. 2e,f). This indicated that serial Si limitation could have restricted diatom growth rates and/or susceptibility to grazing following supply of N + Fe alone. Indeed, assuming typical phytoplankton elemental quotas and diatom Si requirements (reported in Moore et al. 2013), initial silicic acid concentrations were more deficient than either Fe or N in the initial seawater (not shown). However, the substantial Chl a and diatom accumulation that was observed following supply of Fe + N alone (2.6-and 2.7-fold Chl a increases relative to controls in Experiments 5 and 6, respectively; initial Si < 0.3 μmol L −1 ; see also Supporting Information), implied that Si availabilityas for the other non-N or Fe nutrients tested-was not a primary limiting factor in situ (Rynearson et al. 2013). This could potentially be reconciled with a high degree of plasticity in Si requirements: despite the large biomass changes, BSi concentrations were largely invariant between Fe(+N) amendments and controls (Fig. 2). This may in turn have been related to either shifts toward less silicified diatom species (Assmy et al. 2013;Smetacek 2018) and/or reductions in silicification following Fe supply (Hutchins and Bruland 1998;Takeda 1998).

Multiple nutrient supply and non-Redfield POC accumulation
While having relatively little extra impact on Chl a responses, supply of all eight nutrients in combination appeared to have an important influence on POC accumulation at the two sites where this treatment was conducted. In Experiments 5 and 6, Chl a in this treatment was not enhanced relative to the Fe + N + Si treatment, but POC was enhanced a further 1.3-fold (Experiment 5) and 1.6-fold (Experiment 6) (Fig. 2e,f). These POC increases were not matched by an equivalent PON enhancement in Redfield proportions (C:N > 10 compared to Redfield C:N of 6.6).
Such postbloom "carbon overconsumption" likely reflects diatom response to exhaustion of dissolved inorganic N (the starting seawater concentration + that artificially supplied ;Toggweiler 1993;Körtzinger et al. 2001;Engel et al. 2002;Wetz and Wheeler 2003;Riebesell et al. 2007;Taucher et al. 2012). However, near-full nitrate depletion over the 72 h experiment duration was already apparent in the N + Fe treatments (and while not measured, presumably also in +Fe; Table 1), which did not yield such high POC (or POC:PON). Furthermore, F v /F m remained elevated in all "+Fe" treatments, demonstrating phytoplankton were not N-starved (Geider et al. 1993;Parkhill et al. 2001). Together this suggested that while N depletion was probably a necessary condition for the elevated POC:PON, it alone could not explain the large differences between treatments.
Numerous biochemical dependencies exist that could link provision of (micro)nutrients not directly restricting (N-limited) Chl a synthesis with enhanced photosynthate production (Morel and Price 2003;Saito et al. 2008). We can only speculate as to their relative importance. For instance, under conditions of high carbon fixation, the rate of inorganic carbon acquisition may be constrained by carbonic anhydrase activity, which itself might depend on the availability of activating Zn, Co, or Cd cofactors (Riebesell et al. 1993;Morel et al. 1994;Cullen et al. 1999). Likewise, vitamin B 12 catalyzes a variety of carbon metabolism and might not always be sufficiently available to maximize the rates of these reactions (Bertrand and Allen 2012;Bertrand et al. 2015

Implications for carbon export
The different POC:PON accumulation between N + Fe and "all" nutrient combinations could have important implications for our understanding of the way that phytoplankton blooms are regulated in natural systems. The "additional" C in the multinutrient treatments was likely channeled into production and exudation of C-rich, N-poor molecules into phycospheres of the dominant large diatoms (Seymour et al. 2017). In turn, this could potentially be related to stimulation of bacterial phycosphere colonization, which could increase production of enzymes liberating nutrients from dissolved organic pools and/or the synthesis of scarce vitamins (Bertrand and Allen 2012;Bertrand et al. 2015;Seymour et al. 2017). Regardless of why this occurs, in natural systems such sticky exudates coagulate into gel-like particles (Engel 2004;Verdugo et al. 2004;Engel et al. 2015), enhance particle aggregation, and speed up sinking of the high C:N organic material in the water column (Smetacek 1985;Passow and Alldredge 1994;Engel et al. 2002;Martin et al. 2011). Multinutrient regulation of sticky organic exudate production could therefore represent a link between (1) the simultaneous availability of multiple resources as N becomes depleted and (2) export efficiency of diatom blooms to the deep ocean (Billett et al. 1983;Rynearson et al. 2013;Agustí et al. 2015).

Conclusions
Our experiments suggested spatial transitions in primary limitation of phytoplankton by either Fe or fixed N throughout the mid-latitude North Atlantic in late spring. This accords with results from previous studies elsewhere in the North Atlantic (Blain et al. 2004;Moore et al. 2006;Nielsdóttir et al. 2009;Ryan-Keogh et al. 2013). Transitions in limiting nutrients fit within an overall seasonal shift in the bottom-up resource controls on phytoplankton growth (Ryan-Keogh et al. 2013): from early season light limitation, transitioning to Fe limitation, and subsequently, N limitation where winter mixed layer nitrate inventories are lower and/or Fe supply is higher (Moore et al. 2006;Achterberg et al. 2013;Ryan-Keogh et al. 2013;Achterberg et al. 2018;Hopwood et al. 2018;Sedwick et al. 2018;Birchhill et al. 2019). In combination with top-down grazer control, either or both nutrients could therefore sufficiently slow growth rates of larger phytoplankton to terminate the spring bloom (Banse 2002;Ryan-Keogh et al. 2013;Behrenfeld and Boss 2018). Substantial chlorophyll and diatom accumulation occurred following Fe + N supply, with or without provision of supplementary (micro) nutrients (Si,P,Mn,Co,Zn,B 12 ). This implied that the latter were not primary factors limiting bloom formation at the timescales (48-72 h) of our bioassay experiments. We infer that they do not terminate the spring bloom in situ either. However, simultaneous supply of multiple nutrients did lead to enhanced POC production, suggesting multinutrient availability might be important for maximizing transient carbon export, particularly if the additional POC included sticky exudates that enhance cell aggregation and accelerate sinking (Smetacek 1985;Passow and Alldredge 1994;Engel et al. 2002Engel et al. , 2015Engel 2004;Verdugo et al. 2004;Martin et al. 2011).
These observations have raised a number of further questions. First, the critical nutrient combinations have been constrained, but not fully resolved. Exactly which combinations of nutrients, in addition to Fe + N + Si, would produce the responses triggered when all eight were supplied together (Fig. 2e,f)? Our results imply the required (micro)nutrients-or at least those with differing sources, sinks and seawater chemistries-would have to be considered to accurately predict POC accumulation and its corresponding C:N ratio (Schartau et al. 2007). Second, is the high POC:PON ratio of particulate organic matter linked to transient phytoplankton blooming and carbon export? Exudation of high C:N organic matter is predicted to enhance particle abundance, aggregation, and sinking (Engel 2004;Engel et al. 2004). This would lead to faster and more efficient export of phytoplankton carbon because of (1) more rapid transfer of fixed carbon below the thermocline prior to remineralization, and (2) more carbon export per utilized N (Geider et al. 2001;Engel et al. 2004).
The short timescales that might be associated with high C:N diatom blooming and export (days-weeks; e.g., Billett et al. 1983;Lochte et al. 1993;Dale et al. 1999;Mahadevan et al. 2012;Riley et al. 2012;Rynearson et al. 2013;Agustí et al. 2015;Smith et al. 2018), severely complicates efforts to detect and observe these events, if they occur naturally, with either conventional ocean observation programs (e.g., Goldman 1993) or indirectly via remote satellite observations (the latter having a resolution that is generally too low at the relevant time scales; Kahru 2017). Planned multidisciplinary observational programs including autonomous field platforms, which are specifically designed to capture high spatial and temporal variability, might be able to overcome this (Sanders et al. 2016;Siegel et al. 2016;Behrenfeld et al. 2019;Boyd et al. 2019).