Main

Model projections of global marine net primary productivity (NPP) for the year 2100 range from increases to declines by as much as 20% (ref. 1). These divergences largely centre on differences at low latitudes and result from inter-model variations in the underlying mechanisms regulating NPP1,2,3. Such uncertainties are important as they will lead to cascading impacts on ecosystems, ocean biogeochemistry and carbon cycling in models. Climate-regulated NPP changes of several per cent already occur naturally at interannual scales, driven by the impact of the El Niño/Southern Oscillation (ENSO) in the equatorial Pacific5,6,7. These changes in NPP are believed to be driven by variations in the upwelling rate of deep-water nutrients that controls their supply to phytoplankton5,6,8, with uncertainties in future projections strongly linked to changes in growth-limiting nutrients3.

Assessing limiting nutrients, their geographic boundaries and their variation over ENSO cycles is key to understanding contemporary ocean NPP and informing climate model projections4. In that way, historical variability can act as an ‘emergent constraint’ on Earth system model projections7. Available methods for determining nutrient limitation have so far centred on ship-based observations, including simple environmental predictions based on nutrient concentrations, biochemical signals linked to nutrient stress, and growth limitation tests using biomass changes following nutrient amendment9,10,11,12. These different approaches deliver unique information from population to community levels that are not directly comparable9,13. Furthermore, even with coordinated programmes and the highest throughput approaches10,12, all are ultimately restricted to documenting snapshots in time and space that are of limited utility in probing large-scale temporal changes. Attempts have been made to scale nutrient-related phytoplankton fluorescence properties to global remote sensing14,15,16, which would offer unparalleled advantages in spatial and temporal data availability, but a range of uncertainties have restricted usage15,17,18,19,20,21. Here we tackle these challenges for the tropical Pacific by building an observational dataset of nutrient limitation spanning physiological to community-level scales. Moreover, by mechanistically connecting these ecophysiological signals with coincident measurements of radiometric quantities, we directly link nutrient limitation to satellite observations, to assess large-scale variability in nutrient limitation and constrain ocean models.

Contrasting nutrient-limited physiology

Field observations were conducted along an approximately 5,000-km-long transect across the tropical Pacific Ocean (Fig. 1 and Extended Data Fig. 1). Nutrient-addition bioassay experiments and nutrient-stress biomarker proteins both revealed a consistent trend from iron (Fe) to nitrogen (N) limitation, which matched the prevailing nitrate gradient (Fig. 1a–d). Specifically, where nitrate concentrations were elevated (Fig. 1a,b), Fe amendment stimulated chlorophyll a biomass increases (sites 2 and 3; one-way analysis of variance (ANOVA) α = 0.05, Tukey’s honestly significant difference test). Conversely, where nitrate was low at sites 4 and 5, supply of N stimulated small chlorophyll a biomass increases (Fig. 1c and Extended Data Fig. 2), and the in situ abundances of several proteins reflecting N stress (urea ATP-binding cassette transporters, the signal transduction protein P-II, the global N regulator NtcA, and aminotransferase)22 for Prochlorococcus—an important contributor to phytoplankton biomass (Extended Data Figs. 3 and 4)—all increased (Fig. 1d and Supplementary Tables 1–3). In the bioassay experiments, an array of diagnostic pigments responded similarly to chlorophyll a, suggesting that most of the phytoplankton community experienced the same nutrient-limitation regime (Fig. 1e). A level of N–Fe co-stress was also identified at the predominantly N-limited sites 4 and 5, where the Fe stress biomarker flavodoxin was detected23, N + Fe supply further boosted phytoplankton pigment biomass accumulation in comparison to N alone (Figs. 1c,e and Extended Data Fig. 2), and net accumulation of Prochlorococcus cells and increases in chlorophyll a per cell were both largest following N + Fe treatment at sites 4 and 5 (Fig. 1f and Extended Data Fig. 5).

Fig. 1: Tropical Pacific nutrient limitation transition.
figure 1

a, Concentrations of nitrate and dissolved Fe (dFe), with vertical axes scaled according to assumed-average phytoplankton requirements (such that the lower line is the more deficient nutrient; N/Fe of 2,132:1 mol/mol). Circled numbers represent major sampling sites. b, Broader regional nitrate gradient. c, Summarized net chlorophyll a growth in nutrient-addition experiments. Dots are means of triplicates; overlapping lines or symbols of the same colour indicate equivalent treatments except that another nutrient was added in addition to N and/or Fe (either Co, Zn or Si; as indicated by ‘+X’ for N + Fe lines). Asterisks indicate significant chlorophyll a enhancements over non-amended control (ANOVA followed by Tukey’s post hoc test, α < 0.05; Extended Data Fig. 4). d, Abundance of Prochlorococcus nutrient-stress biomarker proteins; nESC, normalized exclusive spectral counts (for which counts have been normalized to the maximum value at each of the five sites); Aminotrans., aminotransferase; urea ABC, urea ATP-binding cassette transporter. e, Responses of individual diagnostic pigments (concentrations normalized to the highest concentration in any treatment). Grey shading indicates not determined. DV Chl-a, divinyl chlorophyll a; Chl-b, chlorophyll b; 19′-hex,19'-hexanoyloxyfucoxanthin; 19′-but, 19'-butanoyloxyfucoxanthin. f, Responses of Prochlorococcus (Pro.) with treatments limiting to the bulk community shown in c highlighted by dot colour. Red fl., red fluorescence per cell; a.u., arbitrary units. Triplicate, biologically independent replicates for each treatment are shown as individual points, with the same treatment joined with lines.

Full size image

A major shift in diel cycles of active fluorescence properties was observed across the nutrient limitation transition (‘active’ fluorescence denoting stimulation by high-intensity blue light flashes; Fig. 2; refs. 10,24,25). Active variable fluorescence (Fv) normalized to maximum fluorescence (Fm) showed both lower dawn and dusk values at Fe-limited sites and pronounced daytime and night-time reductions, with the latter matched by synchronous reductions in functional absorption cross-sections of photosystem II (PSII), σPSII (Fig. 2a,b). By contrast, under N limitation, dawn and dusk Fv/Fm was elevated and night-time reductions were smaller (for Fv/Fm) or eliminated (for σPSII). Such changes in diel active fluorescence properties in this region are well documented10,24,25,26 and have been suggested to be due to declines in Fe-rich photosystem I (PSI) and cytochrome b6f components relative to PSII, which contains less Fe (Fig. 2c; refs. 27,28,29). Such changes in photosynthetic components have not previously been demonstrated in the field, but would probably lead to a more reduced night-time plastoquinone pool in cyanobacteria, in turn triggering night-time energetic decoupling of pigments from PSII (Supplementary Discussion 1; refs. 10,24). Our metaproteomic analyses for field populations of Prochlorococcus demonstrated that proteins associated with PSII showed minor changes across the Fe-to-N limitation transition, whereas those associated with PSI and cytochrome b6f were mostly undetectable at the Fe-limited sites 1–3 but abundant at the predominantly N-limited sites 4 and 5 (Fig. 2c,d). Our data therefore provide in situ confirmation of a predicted, major nutrient-driven stoichiometric adjustment in photosynthetic components28 and its impact on phytoplankton fluorescence properties10 (Fig. 2c,d, Methods and Supplementary Tables 1–3).

Fig. 2: Phytoplankton physiological responses to the Pacific nutrient limitation transition.
figure 2

a,b, Fv/Fm (a) and σPSII (b). Light blue shading around the line in a indicates the sensitivity of Fv/Fm to the applied blank (mean ± standard deviation; n = 16 observations of separate filtrates). Yellow lines and shading indicate sunlight (relative). Daytime Fv/Fm decreases result from the well-known process of non-photochemical quenching14,15. Circled numbers represent major sampling sites. c, Schematic of the photosynthetic membrane with Fe requirements, highlighting changes in component pool sizes and pigment proteins under the two nutrient limitation regimes (Isi-PPC, Fe-stress-induced pigment protein complex; PQ, plastoquinone)30,31,32. Cyt, cytochrome; Fl, flavodoxin; Pcy, plastocyanin. Red arrows indicate relative fluorescence yields of the different components. d, Measured abundances of proteins associated with PSII and PSI. nESC, normalized exclusive spectral counts (for which counts have been normalized to the maximum value at each of the five sites). e–g, Controls on night-time Fv/Fm reductions. Data are from a diel cycle highlighted in a and b.

Full size image

We found that chlorophyll a-normalized active fluorescence (\({F/{\rm{Chl}}}_{{\rm{active}}}\)) was an even clearer diagnostic of the nutrient limitation transition than σPSII and Fv/Fm (Fig. 3a,b). Night-time \({F/{\rm{Chl}}}_{{\rm{active}}}\) values, calculated by normalizing in vivo night-time active fluorescence to solvent-extracted chlorophyll a concentrations, were on average 3.2-fold higher under the elevated-nitrate, Fe-limiting conditions in comparison to the low-nitrate zone (Fig. 3b). The ‘additional’ fluorescence per chlorophyll a produced under elevated-N, Fe-limiting conditions can derive both from a higher ratio of fluorescent PSII to minimally fluorescent PSI (Fig. 2c,d) and from Fe-stress-related pigment protein complexes, which in addition to associating with PSI30, may accumulate to the point at which a proportion have weak or no energetic connectivity to either reaction centre29,31,32,33. Rapid (44–45 h), threefold reductions in \({F/{\rm{Chl}}}_{{\rm{active}}}\) following Fe supply at the higher-nitrate sites (sites 2 and 3) provided strong evidence that this signal was Fe regulated (Fig. 3a). Notably, these \({F/{\rm{Chl}}}_{{\rm{active}}}\) reductions following Fe supply occurred irrespective of the restricted community restructuring (Figs. 1e and 3a and Extended Data Fig. 6). Alongside indistinguishable light absorption properties of bulk phytoplankton communities between the two limitation regimes (Extended Data Fig. 7), both observations suggested that this shift was not due to possible changes in phytoplankton communities with intrinsically different fluorescence properties34 that could have potentially accompanied the nutrient limitation transition (Extended Data Figs. 3 and 4).

Fig. 3: Nutrient regulation of relative fluorescence yields.
figure 3

a,b, Elevated \({F/{\rm{Chl}}}_{{\rm{active}}}\) at the high-nitrate, Fe-limited sites, which was reduced following Fe supply. In a, bars are means of all experimental treatments with or without added Fe for the Fe-limited sites 2 and 3 (left), and the predominantly N-limited sites 4 and 5 (right). Dots show individual data points, and error bars indicate the standard deviation, with two-sided, unpaired t-test P values for significantly different means indicated (n = 30 biologically independent samples for –Fe and n = 36 for +Fe). In b, \({F/{\rm{Chl}}}_{{\rm{active}}}\) is for night-time fluorescence data alone (points indicate the mean and error bars show the standard deviation; n = 5–7 biologically independent samples). c, \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{ship}}}\)–photosynthetically active radiation (PAR) response curves from the shipboard radiometry. Red lines are for the Fe-limited zone (sites 1–3) and blue lines are for the predominantly N-limited zone (sites 4 and 5); bold symbols and lines indicate PAR bin averages with model fits (\({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{ship}}}=a\times \tanh [b\times {\rm{PAR}}/a]\), where a and b parameterize the light-saturated plateau and light-limited slope, respectively; ref. 42). d, Changes in \({F/{\rm{Chl}}}_{{\rm{passive}}}\) across the Pacific Ocean transect from the shipboard radiometry and a boreal winter climatological average from the MODIS-Aqua sensor (satellite (sat)). Black dots show individual radiometric observations between 12:00 and 15:00 local time, and open circles and error bars show the mean and standard deviation (n = 36 independent radiometric observations). Days with the same letter labels have indistinguishable mean \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{ship}}}\) values (one-way ANOVA, α < 0.05, followed by Tukey’s post hoc test). For all panels, units of \({F/{\rm{Chl}}}_{{\rm{passive}}}\) are Wm−2 sr−1 μm−1 [mg Chl m−3]−1.

Full size image

Remote sensing nutrient limitation

Sunlight-stimulated chlorophyll fluorescence signals are observed at global, high-frequency scales by the Aqua Moderate Resolution Imaging Spectrometer (MODIS-Aqua). Although they offer unparalleled observational potential, these passively stimulated fluorescence signals are challenging to directly compare to the actively stimulated fluorescence previously discussed, because of known, high biophysical complexity of chlorophyll fluorescence under variable measurement conditions14,15,18,35,36. To circumvent this complexity, we additionally measured light naturally emanating from the surface ocean at hyperspectral resolution along the transect, using radiometers fitted to the bow of the research vessel. Fluorescence line height (the upwelling radiance peak height at the fluorescent wavelength, about 680 nm; Methods) and concentrations of chlorophyll a were calculated from these radiance signals, and from this, passively stimulated, shipboard \(F/{\rm{Chl}}\) was derived (\({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{ship}}}\)). We found that \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{ship}}}\) was irradiance dependent, increasing with excitation energy at low light levels and then saturating at >500 μmol photons m−2 s−1 (Fig. 3c). Such a response confirms that for the high-light conditions characterizing satellite observations of sunlight-stimulated chlorophyll fluorescence (about 500–2,500 μmol photons m−2 s−1) the impact of increasing excitation irradiances, which will increase energy available for fluorescence, is largely balanced out by dynamic non-photochemical quenching processes, which protect phytoplankton from light damage but reduce fluorescence yields in the process14,19. These two major components of \({F/{\rm{Chl}}}_{{\rm{passive}}}\) thus seem to compensate above the >500 μmol photons m−2 s−1 irradiance threshold14,15,19,37. Subsequently, midday, light-saturated \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{ship}}}\) was 3.5-fold higher in the Fe-limited part of the transect in comparison to the low-nitrate zone, quantitatively matching the changes in \({F/{\rm{Chl}}}_{{\rm{active}}}\) (Fig. 3a,b) and therefore independently monitoring the nutrient limitation transition (Fig. 3d).

Although various forms of satellite fluorescence yields have previously been calculated, their interpretation has been a major challenge owing to limited field constraints of \({F/{\rm{Chl}}}_{{\rm{passive}}}\) in the context of a myriad of potential drivers14,15,16,17,18,19,20,21. As for \({F/{\rm{Chl}}}_{{\rm{active}}}\) and \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{ship}}}\), values of a relative satellite-derived sunlight-stimulated chlorophyll fluorescence yield, \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) (see Methods for derivation), showed a similar approximately threefold decrease across the observed Fe-to-N limitation transition (Fig. 3d) and a consistent geographic pattern of higher values within the Fe-limited equatorial upwelling region and lower values outside (Fig. 4a and Extended Data Fig. 7). In addition to reduced fluorescence yields under low-N conditions, nutrient-replete conditions also decrease the pool of energetically uncoupled pigment protein complexes14,29. Consequently, whereas elevated \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) was a general characteristic of the equatorial Pacific (Fig. 4a), the broader-scale pattern revealed by the satellite data was that of distinctly lower values in the core of the upwelling zone nearer the Equator, where supply rates of both N and Fe to surface waters are elevated38 and reduced phytoplankton Fe stress has been observed25 (Fig. 4a,b).

Fig. 4: Satellite-derived Fe limitation.
figure 4

a, Regional pattern of \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\), reflecting the level of Fe limitation, for a MODIS-Aqua boreal winter climatological average. The white rectangle highlights the Niño3 region, with dot-dashed lines defining the core upwelling zone. Regions of the ocean with no data have chlorophyll a either below the satellite fluorescence detection limit or higher than our validated range. b, Longitudinally averaged changes in monthly \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) and chlorophyll a in relation to mean Niño3 region SSTs for the MODIS record. Dashed lines define the latitude range for the core of the upwelling zone. c,d, Left: time series of chlorophyll a and \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) anomalies (value minus record average) for the central Niño3 region. Right: scatter plot correlations showing type II linear regressions (ranged major axis; red lines) and the 95% confidence regions for the regression lines (grey lines). The R2 and P values (two-tailed) for the regression significance test are shown (no adjustment for multiple comparisons). e, Pacific-wide distribution of \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) (boreal winter climatological average) scaled to model limitation range (left) and model Fe limitation (right). f, Fe-limitation sensitivity to temperature changes in the central Niño3 region for the model (grey dots, R2 = 0.82; P = 9.4 × 10−281; slope = 0.06) and derived from \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) (green dots, R2 = 0.50; P = 4.39 × 10−35; slope = 0.029); regression lines and statistics as for c,d.

Full size image

Fe limitation persists over ENSO

A time series of \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) spanning the two decades of MODIS-Aqua observations can reveal how ENSO regulates nutrient limitation in this region at an unprecedented scale. As expected, sea surface temperature anomalies (ΔSST) for the Niño3 region (Fig. 4a)—an indicator of both ENSO phase and the transfer rate of deeper, colder, nutrient-rich waters to the surface—was inversely correlated with chlorophyll a anomalies (ΔChl; R2 = 0.22; P = 1.86 × 10−13)5,6. For this region, ΔSST had an even stronger positive correlation with Δ\({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) (R2 = 0.41, P = 1.04 × 10−26), with warmer, low-chlorophyll a excursions accompanied by increases in Δ\({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) that indicate stronger Fe limitation. Focusing on a narrower latitudinal band (1.5° S–1.5° N; Fig. 4a) showed that this regional trend was largely driven by equatorial changes, where upwelling strength and its variability are strongest (ΔSST versus Δ\({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\); R2 = 0.50, P = 1.4 × 10−34; Fig. 4c,d). The correlation of Δ\({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) with chlorophyll a anomalies was weaker than that with ΔSST (R2 = 0.38, P = 9.7 × 10−25), implying some decoupling between Fe-limitation levels and chlorophyll a biomass, presumably due to the latter being under additional grazing and/or photoacclimation control. Overall, for 2002–2021, we found that greater upwelling supplied more Fe (ref. 38), reducing Fe limitation25 and increasing chlorophyll a biomass. Notably, no evidence was found for Δ\({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) declines at the highest ΔSST, which would be expected to accompany an overall switch into N-limited conditions under stronger stratification. Our analysis of satellite observations therefore further showed that Fe limitation in the Niño3 region remained robust to the ENSO extremes encountered over the >18-year MODIS observational period. In the coming decades, Δ\({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) will enable observation of the nutrient-limitation response to possible stronger ENSO events comparable to the 1997/1998 El Niño8 as well as climate change1,2,3,4,5,6,7.

In addition to providing a synoptic view of ocean nutrient limitation to upscale field observations in the tropical Pacific, the parallel variations of SST and our \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) metric provide an emergent constraint for nutrient limitation in ocean and climate models. When the satellite Fe-limitation diagnostic was compared with output from a model driven by realistic ENSO dynamics (Methods), we found that spatial patterns in Fe limitation were well reproduced (Fig. 4e). However, the model demonstrated a twofold oversensitivity in the magnitude of Fe-limitation fluctuations to ENSO-driven SST changes (Fig. 4f). This suggests that the natural strength of phytoplankton Fe limitation in the equatorial Pacific is more stable in response to physical changes than models predict. This mismatch may highlight the importance of both Fe recycling39,40 and restructuring of a diverse phytoplankton community26,41 in modulating NPP responses to altered upwelled Fe supply rates. For example, a period of progressively lower Fe supply rates probably selects for phytoplankton with intrinsically lower Fe requirements and/or optimized Fe uptake, thus maintaining productivity to a greater extent than predicted within models lacking this flexibility. Better representation of such processes within ocean models in the future, benchmarked by synoptic observations of nutrient limitation such as those presented here, will lead to more realistic projections of ocean NPP to climate change.

Methods

Research cruise and sample collection

Observations and experiments were conducted on the RV Sonne (cruise SO267/2) from 28th January to 14th February 2019. Near-surface (about 2 m) seawater was collected through a towed, trace-metal-clean sampling system equipped with acid-washed tubing and a peristaltic pump. All sampling was conducted using trace-metal-clean techniques within a laboratory environment that was maintained over-pressurized with HEPA-filtered air.

Nutrient and trace element concentration analyses

Samples for determination of dissolved inorganic macronutrients (nitrate, phosphate and silicic acid) and dissolved Fe were 0.2 μm filtered (AcroPack1000 0.8/0.2 μm filter capsule, Pall). Macronutrient samples were collected into acid-washed 15-ml polypropylene vials and stored frozen at –20 °C until analysis in a laboratory on land. Samples were thawed over 24 h in a refrigerator before determination of concentrations using an autoanalyser (QuAAtro, Seal). Samples for dissolved Fe analyses were collected into acid-washed 125-ml low-density polyethylene bottles (Nalgene) and acidified under a laminar flow hood using 150 μl concentrated hydrochloric acid (10 M; Fisher Optima grade). After >6 months, samples were analysed following pre-concentration using inductively coupled plasma mass spectrometry (Element XR), adhering to the procedure of ref. 43 except that a Preplab instrument (PS Analytical) was used for pre-concentration and standard addition was used for determination of concentrations. Analysis of GEOTRACES intercalibration standard GSP91 in the same analytical run yielded concentrations matching previous determinations (Fe mean ± standard deviation = 0.166 ± 0.048, n = 6; https://www.geotraces.org/standards-and-reference-materials/)44,45.

Chlorophyll a and phytoplankton community structure

Samples for chlorophyll a analysis (100 ml) were filtered onto glass-fibre filters (Macherey-Nagel), extracted in 90% HPLC-grade acetone and measured on a calibrated fluorometer (Trilogy, Turner Designs) following the method of ref. 46. Samples for diagnostic phytoplankton pigments (2–3.5 l; experimental samples from pooled treatment replicates) were filtered onto glass-fibre filters (Macherey-Nagel) and stored frozen at –80 °C. Extraction, analysis and peak picking (Chromeleon, Thermo Fisher Scientific) followed the procedure described in ref. 26. Diagnostic phytoplankton pigments were converted to estimated contributions of different phytoplankton types using CHEMTAX47, using starting pigment ratios from ref. 48. Samples for flow cytometry analysis (2 ml) were fixed with paraformaldehyde (1% final concentration) and stored at –80 °C before analysis using a FACSort flow cytometer (Beckton-Dickinson) and gated and enumerated using CellQuest Pro software (Beckton-Dickinson) following procedures described in ref. 49 (see Supplementary Fig. 1 for an example of the gating strategy).

Underway active fluorometry and absorption measurements

A fast-repetition-rate fluorometer (Fast Ocean sensor interfaced with FastPro8, Chelsea Technologies) was connected to the ship’s underway, near-surface seawater flow-through supply for determination of diel variability in Fv/Fm = (Fm – Fo)/Fm (where Fo indicates minimum fluorescence). Before calculation of Fv/Fm, the fluorescence values Fo and Fm were first corrected for blank fluorescence (fluorescence of 0.2-μm-filtered seawater)50. Lines shown in Fig. 2a,b are 17-min rolling means of Fv/Fm and σPSII. Dots in Fig. 2e–g are 1-min binned means. The same fast-repetition-rate fluorometer was switched to bench-top mode for discrete sample analysis of the experimental samples. A separate fluorometer (Sea-Bird Eco FLNTU) housed with a self-cleaning monitoring box (SMB; -4H-Jena engineering GmbH, Jena) was additionally used to record in vivo chlorophyll a fluorescence continuously from the ship’s underway flow-through seawater supply (shown in Fig. 3b).

The absorption properties of discrete seawater samples were determined using a point-source integrating-cavity absorption meter. Samples were collected from both the towed trace-metal-clean sampling system and underway flow-through system. The measurement procedure was identical to that described in ref. 51 except that the calibration was made using a solid standard instead of the nigrosin solution52. The total absorption of sampled water was measured for unfiltered and filtered (0.2 µm) samples. The values for spectrally resolved (400–710 nm) absorption from filtered samples was subtracted from those for unfiltered samples to yield particulate absorption. Line height absorption was calculated and converted to approximate chlorophyll a concentration following ref. 53, which was then used to normalize particulate absorption spectra (Extended Data Fig. 7).

Metaproteomics

Sample collection

Samples for microbial metaproteomics were collected through direct filtration of 100–120 l of seawater diverted from the trace-metal-clean sampling system. Sample collection was always at local night-time and directly after collection of bioassay experimental seawater. Seawater was pumped through two in-line 142-mm filters (3 μm Versapor and 0.2 μm Supor 200 housed in Sartorius filtration holders). Following filtration, both filters were immediately folded, placed in separate 10-ml cryovials, soaked in RNAlater (Sigma-Aldrich) and stored at –80 °C.

Protein extraction

Analyses followed methods described in ref. 54. Proteins were extracted using a modified magnetic bead method from ref. 55. Filter sections were placed in 15 ml of protein extraction buffer (50 mM HEPES pH 8.5, 1% SDS in HPLC-grade water). Samples were heated at 95 °C for 10 min and shaken at room temperature for 30 min. Filters were removed and protein extracts were filtered through 5.0 µm Millex low-protein-binding filters. Samples were then centrifuged; supernatant was removed from the pellet and 6–7.5 ml of each sample was transferred to a Vivaspin 5K MWCO ultrafiltration unit. Protein extract was concentrated to approximately 350 µl, washed with 1 ml of lysis buffer and transferred to a 2 ml ethanol-washed microtube. Vivaspins were rinsed with small volumes of protein extraction buffer to remove all concentrated protein, and all samples were brought up to 430 µl with extraction buffer. A 30 µl volume was set aside for total protein quantification and DNA analysis.

Protein quantification

Standard curves were generated using albumin standard (Thermo Scientific). Total protein was quantified after extraction and after purification with 2 µl of sample in duplicate using the bicinchoninic acid method (Thermo Scientific Micro BCA Protein Assay Kit). Absorbance was measured on a Nanodrop ND-1000 spectrophotometer (Thermo Scientific).

Protein reduction and alkylation

A quantity of 50 units (2 µl) of benzonase nuclease (Novagen) was added to each sample and incubated at 37 °C for 30 min. Samples were reduced by adding 20 µl of 200 mM dithiothreitol (Fisher Scientific) in 50 mM HEPES pH 8.5 at 45 °C for 30 min. Samples were alkylated by adding 40 µl of 400 mM iodoacetamide (Acros) in HEPES pH 8.5 for 30 min at 24 °C, occasionally heating to 37 °C to prevent precipitation. The reaction was quenched by adding 40 µl of 200 mM dithiothreitol in 50 mM HEPES pH 8.5.

Protein purification and digestion

SpeedBead magnetic carboxylate-modified particles (GE Healthcare) were prepared according to ref. 55. A 20 µl volume (20 µg µl−1) of magnetic beads was added to 400 µl of extracted protein sample. Samples were heated at 37 °C periodically to avoid precipitation. Samples were acidified to pH 2–3 by adding 50 µl of 10% formic acid. Twice the volume (1,100 µl) of acetonitrile was immediately added. Samples were incubated at 37 °C for 15 min and then at room temperature for 30 min. Samples were placed on a magnetic rack, incubated for 2 min, and supernatant was removed and discarded. Samples were washed two times removing and discarding supernatants with 1,400 µl of 70% ethanol for 30 s on the magnetic rack. A 1,400 µl volume of acetonitrile was added to each sample for 30 s on the magnetic rack. Supernatant was removed and discarded. Samples were air dried for approximately 4 min until acetonitrile had just evaporated. Samples were removed from the magnetic rack, and beads were reconstituted in 90 µl of 50 mM HEPES pH 8.0. Purified protein was quantified as described above. Trypsin (Promega) dissolved in HEPES pH 8.0 at a concentration of 0.5 µg µl−1 was added to samples at a 1:25 trypsin-to-protein ratio and incubated at 37 °C overnight.

Peptide recovery and preparation

Acetonitrile was added to digested peptides at a concentration of ≥95% and incubated for 20 min at room temperature. Samples were then placed on the magnetic rack for 2 min and supernatant was removed and discarded. A 1,400 µl volume of acetonitrile was added to samples on the magnetic rack for 15 s. Supernatant was removed and discarded. Samples were air dried for approximately 4 min, just until acetonitrile was evaporated. Beads were reconstituted in 90 µl of 2% dimethylsulfoxide and incubated off the rack at room temperature for ≥15 min. Samples were centrifuged slowly and briefly at a relative centrifugal force of 900 to remove liquid from the tube walls. Samples were incubated on the magnetic rack for 15 min and supernatant containing peptides was transferred to a new ethanol-washed 1.5-ml microtube. This step was repeated to ensure removal of all magnetic beads. Then, 1% trifluoroacetic acid was added to samples for a final concentration of 0.1%. Samples were zip tipped with Pierce C18 tips (Fisher) according to the manufacturer’s protocol with a final resuspension in 25 µl of 70% acetonitrile, 0.1% formic acid. Samples were evaporated to approximately 10 µl in a DNA110 Speedvac (ThermoSavant). Samples with lower protein concentrations were further evaporated to minimize acetonitrile percentage in final resuspension–zip tip product to be less than 30% of total final buffer B volume. Samples were finally resuspended to a peptide concentration of 1 µg µl−1 in buffer B (2% acetonitrile, 0.1% formic acid).

Mass spectrometry

Metaproteomic samples were analysed by liquid chromatography–mass spectrometry (Michrom Advance HPLC coupled to a Thermo Scientific Fusion Orbitrap mass spectrometer with a Thermo Flex source). A 1 µg quantity of each sample (measured before trypsin digestion) was concentrated onto a trap column (C18 Reprosil-Gold, Dr. Maisch GmbH) and rinsed with 100 µl of 0.1% formic acid, 2% acetonitrile, 97.9% water before gradient elution through a reverse-phase C18 column (C18 Reprosil-Gold, Dr. Maisch GmbH) at a flow rate of 500 nl min−1. The chromatography consisted of a nonlinear 100-min gradient from 2% to 95% buffer B, with buffer A being 0.1% formic acid in water and buffer B being 0.1% formic acid in acetonitrile (all solvents were Fisher Optima grade). The mass spectrometer was set to carry out mass spectrometry scans on the Orbitrap (240,000 resolution at 200 m/z) with a scan range of 380 m/z to 1,280 m/z. Tandem mass spectrometry was carried out on the ion trap using data-dependent settings (top speed, dynamic exclusion 10 s, excluding unassigned and singly charged ions, precursor mass tolerance of ±3ppm, with a maximum injection time of 150 ms).

Metaproteomic data analysis

The raw mass spectra files were searched using SEQUEST HT within Thermo Proteome Discoverer 2.2 software using a parent ion tolerance of 10 ppm and a fragment tolerance of 0.6 Da and allowing up to 1 missed cleavage. Oxidation and acetyl dynamic modifications and carbamidomethyl static modifications were included. Percolator peptide spectral matching was used within Proteome Discoverer with a maximum Delta Cn of 0.05 and a decoy search validation based on posterior error probabilities. Processed files were then loaded into Scaffold 5.0 (Proteome Software Inc.) using prefiltered mode with a protein threshold of 1.0% false discovery rate, a peptide threshold of 0.1% false discovery rate and a minimum of 1 peptide for analysis.

The Prochlorococcus nutrient stress and photosystem proteins shown in Figs. 1d and 2d were extracted from the resultant annotated metaproteomics dataset; details are provided in Supplementary Table 1. In cases for which there was more than one detection of the same protein function attributed to Prochlorococcus (same annotation) in the total exclusive spectral counts (probably reflecting mapping of measured tryptic peptides to distinct metagenome contigs with sequence variants), the sequence with the highest counts was used. Only proteins with ≥3 exclusive spectral counts for at least one of the sites were used. A subsequent peptide-level analysis was conducted using METATRYP V2.0 (https://metatryp.whoi.edu; ref. 56), and demonstrated that the peptides found in each protein were largely unique to the Prochlorococcus taxonomic classification (Supplementary Table 1). The data reported here are in unnormalized total exclusive spectral counts, which as described in ref. 57, can be useful in avoiding biases associated with changes in biological diversity across gradients (note that a subsequent normalization of the data for visualization in Figs. 1d and 2d was applied to scale variability in protein abundances across all sites to between 0 and 1). Normalizations according to total Prochlorococcus exclusive spectral counts did not alter the interpretation of these large metaproteome signals. In addition to total exclusive spectral counts, total non-exclusive spectral counts for each of the proteins were inspected to assess for sequence variations across the transect, which could map onto a different annotation despite having the same function (Supplementary Table 2). This analysis showed that total spectral count trends for the highest count proteins showed either much lower values or largely the same trends as the exclusive spectral count trends. To further investigate whether the low abundance or absence of photosystem I and cytochrome b6f proteins found in the high-nitrate, Fe-limited region (sites 1–3) was due to strain-level differences in sequences that could have restricted their detection relative to low-nitrate strains, we carried out a further peptide analysis using METATRYP V2.0 (https://metatryp.whoi.edu; ref. 56) to assess for Prochlorococcus strain matches for peptide sequences (Supplementary Table 3). This analysis showed that all peptides matched to Prochlorococcus strains isolated in the equatorial Pacific, in addition to strains isolated in other regions.

Nutrient-addition bioassay experiments

Bioassay experiments conducted at sites 2–5 directly followed previously published protocols26. Acid-washed polycarbonate 1-l bottles (Nalgene) were filled with trace-metal-clean seawater from the sampling system described previously. Initial bottles were used to characterize initial conditions, control bottles were incubated without treatment, and treatment bottles were spiked with full factorial combinations of the nutrients N, Fe and Co. Supplementary N+Fe+Zn and N+Fe+Si treatments were also conducted to assess for potential Zn or Si serial limitation or co-limitation alongside N and Fe. All controls and treatments were conducted in triplicate. The N spike was a combined treatment of nitrate + ammonium (final amended concentration of 2 μM nitrate + 1 μM ammonium). Subsamples from incubations used for determination of macronutrient concentrations indicated that treatments with supplied nitrate were not depleted to <1 μM nitrate in any experiment over the 48-h incubation duration. Fe, Co and Zn were each added to a final added concentration of 2 nM. Silicic acid was added to a final amended concentration of 2 μM. Macronutrient spikes were previously passed through a column of prepared cation exchange resin to remove contaminating trace elements (Chelex-100, BioRad). Trace element spikes were prepared from 99+% purity solid standards dissolved in 0.01 M hydrochloric acid (Fisher Optima grade diluted in Milli-Q deionized water). Treated and control bottles were Parafilm-sealed, placed in a clear sample bag, and incubated in on-deck incubators that were continuously flushed with near-surface seawater and screened with blue screening (Blue Lagoon, Lee filters) for about 48 h. Following incubation, experiments were ended and each individual replicate was subsampled for photophysiology (fast-repetition-rate fluorometer), chlorophyll a, flow cytometry and macronutrients. The remaining samples from each triplicate replicate of a treatment were combined and filtered for HPLC pigment analysis. Calculation of chlorophyll a-based net growth rates (μChl) in Fig. 1c used the equation ln(ChlTreatment/ChlControl)/t, in which t is the duration of the experiment (2 days). The control rather than initial chlorophyll concentration was used for this calculation owing to the consistent chlorophyll reductions between initial and controls that was assumed to be a result of photoacclimation to a higher-light environment in the deck incubators over in situ conditions (Extended Data Fig. 2).

Shipborne radiometric quantities

Two sets of hyperspectral radiometer systems were used to collect radiometric quantities on the research cruise. Each set consisted of a TriOS RAMSES-ACC hyperspectral cosine irradiance radiometer that measured incoming solar irradiance ED (λ) and two TriOS RAMSES-ARC hyperspectral radiance meters. The radiance radiometers measured bulk ocean surface-leaving radiance Lsfc (θsfc, Φ, λ) and sky-leaving radiance Lsky (θsky, Φ, λ). A custom-made frame held the radiance radiometers at fixed viewing angles at the bow of RV Sonne. The zenith viewing angle of the sky-facing radiometers was θsky = 45°, and the nadir viewing angle of the sea-surface-facing radiometer was θsfc = 45°. Each set was separated by an azimuthal angle of Φ = 45° from the ship’s heading. Irradiance meters were attached to the railing at the top of the mast at least 1 m apart pointing directly upwards. Hyperspectral radiometric quantities were recorded at 5-min intervals from the ultraviolet (320 nm) to near-infrared (950 nm) regions. A correction for sea-surface-reflected glint was applied to derive water-leaving radiance (LW) and remote-sensing reflectance (RRS) as proposed in previous studies58,59. Fluorescence line height (FLH) values were calculated from LW using the approximate central wavelengths of the relevant MODIS-Aqua wavebands (667, 678 and 748 nm):

$${\rm{FLH}}={L}_{678}-{L}_{667}-({L}_{748}-{L}_{667})\times (678-667)/(748-667).$$

Chlorophyll a (Chl) concentrations were derived from RRS using the OC3 algorithm60. Although variability in chlorophyll a concentrations was relatively minor throughout the cruise track, comparing shipboard radiometry-derived OC3 and filtered and extracted chlorophyll a concentrations collected at similar time points yielded a good relationship between both methods (R2 = 0.57, P = 2.8 × 10–7, n = 33; Supplementary Fig. 2). FLH values were normalized to chlorophyll a values to generate \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{ship}}}\). Dots in Fig. 3c represent 25-min rolling averages, and in Fig. 3d, data were extracted for 12:00–15:00 local time for each day to approximately match MODIS-Aqua overpass timing (solid line indicates the mean and dots represent individual observations).

Satellite remote sensing

MODIS-Aqua level 3 normalized FLH (nFLH), chlorophyll a concentration (OCX algorithm) and SST data products were downloaded at about 9 km (1/12 of a degree) spatial resolution as standard mapped images from the NASA (National Aeronautics and Space Administration) Ocean Colour website (https://oceancolor.gsfc.nasa.gov). The nFLH product is calculated using normalized water-leaving radiances, with the normalization accounting for the viewing geometry of the ocean surface with respect to incident solar radiation14,21,61. Passive chlorophyll a fluorescence signals were observed to be at fully stimulated plateaux at the incident irradiances at the time of the satellite overpass (>500 μmol photons m−2 s−1; Fig. 3c). Therefore, we followed ref. 21 and carried out a first-order correction of nFLH values to remove the normalization, by multiplying nFLH values by the cosine of the solar zenith angle calculated at each pixel latitude for the mid-time point of the respective composite L3 image21. This conversion led to little change in trends of chlorophyll-normalized fluorescence values in the low-latitude equatorial Pacific cruise region focused on here (however, it completely changed distributions and seasonal trends at higher latitudes; Extended Data Fig. 8). It is important to note that the generality of our FLH versus irradiance observations made in the low-latitude Pacific remains to be tested in other regions, such that the latter satellite observations should be treated with caution (for example, any regions away from the tropical Pacific in regions within Fig. 4e and Extended Data Fig. 8). We also note that a conversion of nFLH back to FLH is essentially equivalent to multiplying the nFLH values by the instantaneous PAR (iPAR) at the time of image capture, as carried out in ref. 14 to correct for the predicted (now verified here for the tropical Pacific) impact of non-photochemical quenching on nFLH14.

Satellite observations of FLH were divided by satellite-derived chlorophyll a concentrations to retrieve \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\). We also excluded pixels with chlorophyll a concentrations <0.1 or >0.4 mg m−3, which bounds the estimated lower FLH detection limit18,62,63 as well as more uncertain chlorophyll a concentration retrievals64 and upper levels where the relationship between chlorophyll a and phytoplankton light absorption become increasingly nonlinear65,66. Within this chlorophyll a range, the relationship between chlorophyll a concentrations and phytoplankton light absorption is indistinguishable from linear66; therefore, an attempt to convert chlorophyll a to absorption is not necessary in this case (and could possibly introduce error due to the impact of high chlorophyll a values on chlorophyll a versus absorption equations developed using datasets from the global ocean66). For the time series analyses of equatorial Pacific \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) (that is, Fig. 4c,d,f), we used the Niño3 box where \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) was validated, and where chlorophyll a consistently falls within the applied chlorophyll a thresholds (in >90% images <5% pixels fall outside this range; including these pixels does not change trends). In Fig. 3d, \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) pixels were averaged over 100-km-radii circles around each ship track sampling point, to reduce noise in signals derived from individual pixels63. To generate a \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) range comparable to the Fe limitation term in biogeochemical ocean models in Fig. 4e,f, the range in \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) was scaled to range between 0 (no Fe limitation) and 0.78 (maximum Fe limitation predicted by the used model; see next section). The upper value of \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) used to scale to the model limitation range was taken as the \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) threshold where the fraction of pixels above this value from a global-scale monthly composite reached an asymptote (\({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\) = 0.7 Wm−2 sr−1 μm−1 [mg Chl m−3]−1; varying this threshold by 10% had <10% impact on the twofold difference in slopes between the model and observations; Fig. 4f).

Global biogeochemical ocean model

The Fe limitation term (LFe) from a state-of-the-art global biogeochemical model NEMO-PISCES67 was compared with spatial patterns and seasonality in \({F/{\rm{Chl}}}_{{\rm{passive}}}^{{\rm{sat}}}\). PISCES has a relatively complex consideration of the ocean Fe cycle, including the dynamics of phytoplankton Fe uptake, storage and limitation, as well as the parallel processes of zooplankton recycling and scavenging3,68,69. The LF term is the fractional limitation of phytoplankton maximum growth rate (range 0–1), calculated from dynamic phytoplankton Fe quotas relative to emergent minimum and optimum quotas70. The model limitation terms were extracted from the surface layer at monthly resolution for diatoms and nanophytoplankton (the phytoplankton groups in PISCES). Weighting these terms by the carbon biomass of each phytoplankton type generated an average LFe. In Fig. 4e,f and Extended Data Fig. 8 this average LFe was subtracted from 1 to generate an Fe limitation term (FeLim = 1 – LFe), so that higher values indicate greater Fe limitation (and vice versa). The model results were drawn from an online ocean model simulation forced by atmospheric reanalysis product JRA55 from 1958 to 2019 under the OMIP2 protocol. The model is forced at a 6-h time frequency over three cycles between 1648 and 201971.

Statistics

Statistics and calculations were conducted using R. Type II linear regressions (ranged major axis) in Fig. 4c,d,f were conducted using the lmodel2 package.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.