Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

# Regulation of phosphorus bioavailability by iron nanoparticles in a monomictic lake

## Abstract

Dissolved reactive phosphorous (DRP) in lake systems is conventionally considered to predominate over other dissolved P species, however, this view neglects an important set of interactions that occurs between P and reactive iron hydroxide surfaces. This study addresses the coupling of P with dispersed iron nanoparticles in lakes, an interaction that may fundamentally alter the bioavailability of P to phytoplankton. We used diffusive gradients in thin films (DGT) and ultrafiltration to study Fe-P coupling in the water column of a monomictic lake over a hydrological year. Fe and P were predominantly colloidal (particle diameters > ~5 nm < ~20 nm) in both oxic epilimnetic and anaerobic hypolimnetic waters, but they were both DGT-labile under sub-oxic conditions, consistent with diffusion and dissolution of Fe-and-P-bearing colloids within the DGT diffusive gel. During peak stratification, increases in Fe and P bioavailability were spatially and temporally coincident with Fe nanoparticle dissolution and the formation of a deep chlorophyll maximum at 5–8 m depth. These results provide a window into the coupling and decoupling of P with mobile iron colloids, with implications for our understanding of the behaviour of nutrients and their influence on phytoplankton community dynamics.

## Introduction

Phosphorus (P) and nitrogen (N) play a central role in the biological productivity of aquatic systems1,2,3. In freshwaters, dissolved inorganic N and P act as limiting nutrients depending on their molar ratios4, but it is usually P that limits the productivity of terrestrial aquatic systems.

Excess P availability occurs primarily through human actions2,5, with the widespread use of P fertiliser, poor riparian buffering and uncontrolled sedimentation being major drivers5,6. Despite regulation of anthropogenic P inputs, the biogeochemical cycling of P between lake water columns and sediments typically maintains high P availability to phytoplankton7.

Phosphorus is present in a range of inorgaimnic and organic species across the dissolved, colloidal and particulate size ranges8. Organic P constitutes a major part of the dissolved P pool9, while particulate (>1 μm) and colloidal (1–1000 nm) fractions (from plant, animal and bacterial sources) can be present as poly-phosphates adsorbed to metal oxides, hydroxides and clays10,11,12.

Bioavailable P is usually the sum of the P species that are immediately available for biological uptake, or species that are easily transformed into available form(s) by naturally occurring processes13,14,15. Thus, the bioavailable P fraction is a moving target subject to biological, chemical and physical processes that vary through time in lakes and which requires cautious monitoring to be meaningfully estimated.

During lake stratification events, organic matter (OM) is supplied to the hypolimnion through settling of suspended particulates, or bioturbation in the upper few mm of the sediment13,16 or turbulent mixing across the thermocline17,18. To-date, two distinct phenomena have been proposed to explain the observed P fluxes from benthic sediments, (1) elevated pH (resulting from algal photosynthesis) favours the desorption of PO43− from sediment solids19,20 and (2) reductive dissolution of Iron(III) species at the onset of anaerobic conditions17. The mineralisation of OM consumes oxygen and creates steep redox gradients between iron species present in sediments and water column. These gradients are expected to affect the distribution and bioavailability of P because of well-documented interactions between dissolved inorganic phosphorus (DIP) and Fe hydroxide surfaces21,22.

Iron oxide colloids often form by hydrolysis and oxxidation of Fe(II) at oxic/anoxic boundaries in lakes17,23. Differences in the surface properties of ferrous (Fe(OH) 2) and ferric (Fe(OH) 3) colloids are likely to be significant for P binding24. The size, structure, and reactivity of this hydrolysis product can vary, e.g., because of natural organic matter coatings25,26. Previous studies of other freshwater ecosystems have reported mobile Fe colloids to be in the size range of 0.05 to 0.5 μm27,28,29,30,31. Such Fe colloids have been shown to be stable in sub-oxic conditions32,33 and can therefore potentially survive for extended periods under anaerobic conditions, with concomitant effects on solution chemistry. Hence, aquatic colloids can be rich in both P-binding surfaces and organic matter (the latter containing a significant amount of organic P)34.

Distinct from gravitoids (large particles >10 μm27, colloids (1–1000 nm) and nanoparticles (1–100 nm) remain suspended for extended periods and thus provide surface sites for adsorption or other chemical reactions within the water column10. When present within the lake water column, these dispersed surfaces have the potential to significantly modify P bioavailability to phytoplankton (Fig. 1). While the coupled release of Fe and P has been previously demonstrated in lake sediments35,36 the significance of Fe colloids for P bioavailability has yet to be elucidated in these systems.

The dearth of studies on the interaction between P and colloids in lakes can be explained by inherent difficulties associated with probing colloidal interactions in such sub-oxic (reduced) domains. This study seeks to address the lack of information on colloidal Fe-P interactions under reduced conditions, and the absence of highly spatially- and temporally-resolved data on P speciation in lake systems in general, by combining sensitive and representative measures of in-situ water column chemistry based on diffusive gradients in thin films (DGT) and ultrafiltration methods.

The use of diffusive gradients in thin films (DGT) has gained recognition as an alternative approach for rapid evaluation of labile metal/nutrient content in water and sediments. DGT solution probes are small plastic devices containing a filter membrane, ion permeable hydrogel and a binding agent incorporated in a basal gel layer. When immersed in solution, the solutes pass through the hydrogel of known thickness (Δg), also referred as diffusive layer/gel, and accumulate at the binding gel layer. Dissolved species concentrations are then determined by using Fick’s first law of diffusion from measured mass (M) over deployment time (t)37. These devices formed the principle means by which the biogeochemical dynamics of Fe and P in Lake Ngapouri were deduced. For further information on the use of DGT for evaluating bioavailability we refer readers to Zhang & Davison38,39 and Zhang et al.40.

## Results and Discussion

### Physicochemical characteristics of Lake Ngapouri

Stratification in Lake Ngapouri began in November 2015 (Fig. 2) and the lake water became markedly clearer at this point compared with the preceding winter mixed period. High turbidity during the winter may have been caused by the resuspension of sediment or a predominance of diatomaceous algae (Fig. 2), whereas increased clarity during summer can be plausibly explained by coagulation and sedimentation of the suspended particles41.

The surface water temperature of Lake Ngapouri ranged between 9 °C and 23 °C over the study period. The annual temperature cycle of the lake was characterised by two distinct phases: isothermal conditions and thermal stratification (Fig. 2).

CTD (Conductivity, Temperature, Depth profiler) casts indicate that isothermal conditions continued until September, and that a thermocline started to establish at the start of November 2015. Thermal stratification established as the summer progressed (Fig. 2).

The exact depth of the CTD casts varied due to changes in water levels over the course of this study. Therefore, the time series presented here was reduced to the minimum depth recorded in the sampling period (22 m from surface) for consistency. Since profiles for chlorophyll fluorescence were determined by CTD casts and no CTD data were collected at night time we were not able to determine the direct effect of non-photochemical quenching42. Chlorophyll fluorescence data indicate periods of high phytoplankton biomass during the winter mixed period in September 2015 and October 2016, and a peak just above the thermocline (ca. 5–8 m) during March-May 2015. This deep chlorophyll maximum (DCM) was quite similar to that observed in nearby Lake Okaro, a lake of similar trophic status with an anaerobic hypolimnion43. No response in chlorophyll fluorescence was observed after the diurnal mixing event in November 2015.

Once established, the thermocline was very stable and progressively deepened from 5 to 13 m from November 2015 to March 2016, respectively. The DO concentration markedly decreased below 5 m depth in November 2015 and established a completely anaerobic hypolimnion extending from 15 m depth to the bottom of the lake over the same interval. The epilimnion deepened to about 10 m by May 2016 and the lake experienced complete seasonal mixing in July 2016.

The lake water pH was almost uniform during winter mixed conditions, but started to decrease in deeper waters of the lake during early stratification. The onset of stratification coincided with elevated chlorophyll fluorescence near the lake surface, which coincided with higher pH conditions (exceeding pH 9 in October 2016; Fig. S2, supplementary information).

### Colloidal size distributions, morphology and chemistry through time

#### AFM Results

Particle size distributions (PSD) measured by atomic force microscopy (AFM) of colloids are shown in Fig. 3. It should be noted that AFM is biased toward small particles and does not necessarily cover the entire colloid size distribution range. Colloids were predominantly in the size range of 1 to 10 nm in both epilimnetic and hypolimnetic samples. In the epilimnion during January 2016, particles in the 2–3 nm size range constituted >45% of total particle numbers. Particles >10 nm did not form a major proportion of the PSD during this month (Fig. 3) which may be related to particle settling and gravitoids sedimentation under the prevailing stratification. On the other hand, in the hypolimnion during the same month colloid sizes shifted toward larger particle sizes (between 3 and 18 nm) with particles between 4–6 nm constituting the most significant fraction at ~40% of all measured particles (Fig. 3).

In the epilimnion during March 2016, particles were in the size range of 2–14 nm, with the greatest density between 4–5 nm (Fig. 3c). In the hypolimnion, particles remained distributed within a narrow size range, with those in the 4–7 nm range making up ~50% of the total particle number (Fig. 3e). In May, the epilimnion PSD centred on 3–9 nm, with these particles constituting more than 50% of the total particle number (Fig. 3a). This PSD was comparable to the hypolimnion PSD in January and March. The consistent size distributions between months in the hypolimnion might be explained by: (1) resuspension of the particles from the bed sediments during this stagnant phase, or (2) the recycling of nanoparticles by cycles of reduction and precipitation within the water column44. Generally, a trend was observed toward decreased particle size in the deeper anaerobic water during stratification. This could be consistent with the reductive dissolution of larger colloids and precipitation of fresh nano-scale Fe (II)/Fe (III) hydroxides, although this cannot be substantiated without higher spatial and temporal resolution data.

#### TEM Results

The Transmission Electron Microscope (TEM) analysis confirmed that Fe-bearing particles were the most prevalent class of nanoparticles detected in Lake Ngapouri. Figure 4 shows a representative TEM micrograph and EDX (Energy-dispersive x-ray analysis) spectrum, which show peaks for Fe, O with contributions from Si; all major elements commonly associated with colloidal species in freshwaters. Evidence was also found for the presence of Fe monosulfides in the hypolimnion as would be expected at the interface of Fe-reducing and S-reducing zones in the water column and sediments.

### Iron depth distributions through time

Water sample aliquots collected after ultrafiltration in a zero-grade N2 atmosphere were used to measure the fraction of Fe and P associated with colloidal particles. This analysis provides an approximation of the concentration of Fe and P in the colloidal range which excludes particles >0.45 μm and nanoparticles <~5 nm diameter (based on 100 KDa ultrafiltration).

Time series of Fe depth distributions in Lake Ngapouri (Fig. 5) reveal the dominant influence of stratification on Fe cycling. The Fe concentration started to build up in all fractions (dissolved, colloidal, CDGT and Ferrozine) over the period of early stratification and was maintained during summer. Fe < 0.45 μm was predominantly colloidal (Fig. 5b). Taken together with the AFM and TEM results, this implies that Fe was present in the hypolimnion either as simple inorganic colloids, e.g. Fe(III) hydroxides with particle sizes typically ≥5 nm, or in association with macromolecular organic colloids/aggregates. Analysis of the organic fraction of the dissolved and colloidal size range in the total organic carbon (TOC) measurement returned colloidal TOC concentrations in the 0.1–3.31 ppm range, constituting up to 100% of the organic carbon <0.45 mm (i.e., dissolved organic carbon (DOC)) in some cases.

DGT is thought to measure (1) all hydrated free-ion species, (2) a vast majority of the colloidal and complexed species that may be released into free-ion species during the deployment and (3) colloidal species that can diffuse through the APA hydrogel and bind with the basal resin layer38,45. The diffusion coefficient of solutes in the hydrogel decreases with an increase in their hydrated radii. Although exact information on the porosity of APA diffusive gel is lacking, initial studies by Zhang and Davison37 reported the pore size to be >5 nm while Fatin-Rouge, et al.46 gave indicative pore sizes in the 50–74 nm range, depending on the gel cooling rate during polymerization. Furthermore, particles >60 nm were considered locally trapped and not evenly diffused through the gel46. However, Davison and Zhang38 suggested that despite the probable existence of larger pore sizes, DGT devices can only measure particles <2 nm size because of the slow diffusion rate of larger particles. In addition, the structure and composition of solutes are important determinants. Thus, DGT can be considered to be sensitive to nano-scale particles which are sufficiently labile to be measured by the DGT resin layer, or otherwise become embedded in the resin gel matrix39. The diffusion coefficients (D) of nanoparticles in the size range determined by AFM were calculated using Stokes law (Table 1). These D values were used to calculate the time to reach 95% of the steady state accumulation rate (T95%) for colloids. D values from Zhang and Davison37 were used to calculate T95% for dissolved Fe (Table 2). The T95% values are comparable for the colloids >1 nm and demonstrate that over the course of a week (typical duration of deployment in this study), DGT is likely to have accurately recorded the concentration of DGT-labile colloidal species in the lake.

Direct measures of the initial increase in Fe concentration in the hypolimnion of Lake Ngapouri between November and January (Fig. 5) were not mirrored by an increase in CDGT Fe, which appeared to lag colloidal Fe in the upper hypolimnion (Fig. 5c). Because DGT only measures the diffusible Fe concentration, it follows that Fe released into the hypolimnion in the early period of stratification was non-DGT-labile. This supports the interpretation that Fe became more labile through time in the hypolimnion as reducing conditions intensified (Figs 2 and 5). Following March 2016, Fe concentrations measured by DGT showed improved agreement with the other Fe fractions. Indeed, in the latter months of stratification it appears that although Fe was predominantly colloidal (Fig. 5b), it was also DGT labile (Fig. 5c).

It is clear from Fig. 5 that the increase in Fe (II) started from the bottom of the lake, consistent with diffusion from the lake sediment/benthic nepheloid layer, which was characterised by lower DO concentrations than the overlying hypolimnetic water in the early period of stratification30. Dissolved Fe (II) first appeared in the lowermost sample in late November 2016 (Fig. 5), when the concentration of dissolved oxygen had fallen to 0.1 mg L−1 (Fig. 5b). From this point until the end of stratification Fe(II) was maintained at elevated levels.

Re-oxygenation of bottom waters occurred during winter in July 2016 (Fig. 2). The concentration of Fe decreased in all fractions (Fig. 5), consistent with the oxidation of Fe (II) species and the precipitation of Fe (III) hydroxides, most of which formed settleable gravitoids, and were removed from the water column. The measurements of dissolved Fe in July 2016 confirm that that while Fe was returned to the surface water by mixing, it was not DGT-labile (Fig. 5). Hence, the Fe data from July 2016 capture the suspension of Fe (III) colloids prior to aggregation and settling, a process which had continued to completion by October 2016.

Finally, the Fe(II) distribution (as determined by the Ferrozine method) was in broad agreement with the CDGT data, but over-estimated Fe in all other fractions. It is important to note that Fe (III) can co-exist with Fe (II) at circumneutral pH values, and indeed thermodynamic disequilibrium between Fe(III)/Fe(II) species can be driven by microorganisms in some cases42,43. Although Viollier, et al.47. reported 3% higher absorption of Fe(II)-Ferrozine complex in their system, the results of the present study are broadly in line with the findings of Anastacio, et al.48 who overestimated Fe(II) as a result of photochemical reduction of Fe(III) complexes. To minimize the uncertainty in Fe(II) quantification here, the samples were pipetted to pre-prepared spectrophotometric cuvettes (containing Ferrozine) and treated for a short incubation time (<1 min), yet it was practically impossible to achieve the desired oxygen-free, dark conditions during measurements in an open boat. It is likely that the overestimation of Fe by the Ferrozine method was also due to interference by other divalent cations.

### Phosphorus depth distributions through time

Mobile phosphorus has long been known to be intimately associated with particulate Fe in fluvial environments20,49 and interactions of the two have recently been shown for colloids in streams and aquifers32. During this study, the colloidal fraction of P (P(c)) was found to track the dissolved P concentration, demonstrating that P(c) was dominant over dissolved species. Given that Fe in Lake Ngapouri was predominantly colloidal (Fig. 6), one line of evidence for the interaction of Fe and P in the water column would be the colloidal association of both elements. Indeed, an analysis of P size-distributions through time also confirms that DRP in this system (Fig. 6b) is mostly in the colloidal size range. Increases in P in both dissolved and colloidal fractions also coincided with increases in the DGT-labile P concentration (Fig. 6d).

DGT probes recorded three prominent instances of P release: discrete increases between 5–10 m depth during the onset of stratified conditions, and a sustained hypolimnetic increase in P through the summer period (Fig. 6c). It is notable that these increases in DGT-labile P were mirrored by high chlorophyll fluorescence and solvent-extracted chlorophyll a at the same or adjacent depths (Fig. 2).

### Coupling and decoupling of iron and phosphorus

Generally, under winter mixed conditions, the highly oxic water column was characterised by low concentrations of Fe (from being undetectable up to 11.6 μmol L−1 towards the bottom of the lake). Low dissolved Fe also coincided with low concentrations of CDGT Fe (often close to zero) (Fig. 5). Thus, under mixed oxic conditions in the lake, the Fe data support the classical interpretation that Fe was present as colloidal Fe(III) hydroxides50 and was not DGT-labile because the Fe was held in the insoluble Fe(III) oxidation state.

During the summer-autumn stratified period the concentration of dissolved Fe (<0.45 μm) increased towards the bottom of the lake, reaching the highest value recorded in this study of 109 μmol L−1 at 20 m depth in May 2016, which coincided with a higher concentration of hypolimnetic P in the same fraction (7.51 μmol L−1). The same trend was observed in the P(c) fraction, possibly indicating the release of adsorbed or co-precipitated P from Fe(III) hydroxides as a result of Fe(III) reduction and hydrolysis of Fe(II).

Phosphorus determined by inductively coupled plasma mass spectrometry (ICP-MS) from the filtered samples (<0.45 μm) was highest at 8 m depth under completely mixed conditions in October 2016 (74 μmol L−1). Notably, there was an increase in chlorophyll fluorescence at this depth during March 2016, which suggests that the higher value of P not associated with dissolution of Fe mineral phases, was instead driven by another process, such as the mineralisation of biomass.

The coupling and decoupling of Fe and P is directly evident in the CDGT data from the lake in the two seasons (Fig. 7). In the autumn (March-May), reductive dissolution of Fe colloids in the hypolimnion is implicated as the driver of elevated CDGT Fe and P values. Dissolution of Fe colloids reached its zenith in March, and over the following two months the P: Fe ratio measured by DGT declined, probably consistent with biological uptake of P over that interval. Conversely, during the winter/spring mixed periods, phytoplankton blooms can be seen to have been entirely decoupled from Fe hydroxide reduction (Figs 2 and 5). The decoupling of P and Fe during the oxic intervals is consistent with the sequestration of P by insoluble Fe hydroxides, destabilisation of this colloidal suspension and sedimentation of the resulting Fe gravitoids. It appears that recently-sequestered Fe and P particles were likely suspended into the water column of the re-stratifying lake in the spring period (see colloidal distributions of P and Fe in the water column over these periods (Figs 5 and 6)), which could have been aided by bioturbation51,52.

At the onset of thermal stratification, the concentration of Fe2+ (inferred by the ferrozine assay) increased in the hypolimnion and gradually built up in summer through until March. In January 2015, Fe was predominantly colloidal in the hypolimnion, of which very little was DGT-labile. In addition, DGT-labile Fe was very closely related to Ferrozine Fe2+, implying that at this stage, Fe was present in a relatively inert form in the hypolimnion.

During stratification in deep lakes, when the upper layer of the euphotic zone is nutrient-depleted and phytoplankton in the lower level are more likely to be limited by light supply, a so-called ‘deep chlorophyll maximum’ (DCM) develops–typically situated around the thermocline. This phytoplankton distribution has been linked to upward diffusion of nutrients from the enriched hypolimion waters42,52,53. In this study, when the stratification was still strongly maintained in March, the DCM was observed at 11 m in association with a strong depletion of dissolved Fe, colloidal Fe and DGT-labile Fe, as well as some depletion in colloidal P (Fig. 8) at the same stratum, likely reflecting biological uptake of this important micronutrient. Interestingly, the depletion of P was only observed in the colloidal P fraction, which would be consistent with targeted dissolution of Fe colloids to release Fe (an important micronutrient for phytoplankton). In the hypolimnion in March 2015, the DGT-labile fraction of Fe exceeded the concentration of Fe in the <0.45 μm fraction. This can be explained, but not demonstrated (due to a lack of particulate data), by the removal of Fe colloids as gravitoids27.

## Conclusion

This investigation suggests that colloidal Fe oxides are generated during mixing in monomictic lakes and possibly at the interface of oxic epilimnetic and hypolimnetic waters by cycles of microbial oxidation-reduction reactions. Our results also suggest that P was associated to these iron nanoparticles and that both Fe and P increased in concentration as reducing conditions deepened in the hypolimnion. These findings are in line with those of Gottselig, et al.54 and Jiang, et al.55 who demonstrated P associated with Fe oxide colloids in the size range of 1 to 20 nm in forest stream waters and agricultural soils, respectively, using coupled flow fractionation and ICP-MS, and Hartland et al.32 who showed dominant P adsorption to iron nanoparticles in streams and shallow (reduced) aquifers.

Further to this, we suggest that freshly precipitated Fe (oxy) hydroxides are small enough to diffuse through the APA hydrogel used in DGT solution probes at a rate comparable to that of solutes (at least when integrating over several days). These colloids bind P and thereby reduce the P concentration in the <100 KDa fraction. Based on elevated chlorophyll values at the same or adjacent depths, we conclude that colloidal P (and also Fe) is available for biological uptake and that DGT-labile P provides an analogous measure of P bioavailability. This interpretation is consistent with the findings of Montalvo, et al.56 who demonstrated that the colloidal P fraction was available for uptake by terrestrial plants. Our results confirm that iron nanoparticles play an important role in P transport and cycling in the lake systems, with important implications for phytoplankton community dynamics.

## Materials and Methods

### Site description

Lake Ngapouri is a small (0.19 km2) eutrophic lake located in Waikiti valley south of Rotorua, Bay of Plenty, New Zealand. It has a maximum depth of 24.5 m. A previous study reported increased concentrations of Fe, Mn, P and As in the anaerobic hypolimnion of the lake44. Being holomictic, monomictic and high in Fe, the lake was considered to be an ideal model system for examining P-Fe interactions.

### Field Methods

#### Water sampling

Sampling of Lake Ngapouri took place every second month, from September 2015 to October 2016). Water samples were taken from 12 depths between 0 and 23 m, at the deepest point of the lake ~23 m depth). Figure 9 summarises the field-to-lab workflow.

A Van dorn water sampler (PVC BetaTM Horizontal bottle, Envco, Auckland, New Zealand) was used to collect representative water samples from the discrete depth intervals. Vertical profiles of water column chemistry were constructed by sampling every meter (relative to the lake bottom) through the thermocline, as well as at two sampling points in the epilimnion and two in the hypolimnion. A conductivity, temperature and depth profiler (CTD; SBE 19 plus V2 SeaCAT profiler, Sea-Bird Electronics Inc., Washington) equipped with sensors for pH, dissolved oxygen, phtosynthetically active radiation (Licor Inc.) and chlorophyll fluorescence (Chelsea Instruments Ltd.) was used to determine the physiochemical character of the lake prior to and after each DGT deployment. Secchi depth was also measured to determine water clarity.

#### Filtration and preservation (in field)

Samples of 15 ml lake water were filtered on site by using 0.45μm syringe filters (Sartorius Stedim, Germany) in acid-washed bottles and were acidified on-site with ultrapure HNO3 to achieve 2% acidity in preparation for ICP-MS analysis (acid was prepared using low-temperature distillation in a Savillex (Eden Prairie, MN, USA) DST-1000 acid purification system). Overloading of filters was avoided as it can affect the pore size and may cause variable removal of particulates and colloids. To maintain the integrity of the collected water samples for major cation analysis, the screw-capped bottles were sealed using laboratory film and maximum care was taken to leave no head space in the bottles, to avoid oxidation. The bottles were then covered with aluminium foil and transferred in the dark at <5 °C to avoid photolysis. Additional samples were collected in pre-cleaned (acid/deionised water) syringes to determine the size distribution of the components of interest in the water column. Syringes were capped to avoid diffusion of O2 into the samples, and were then sealed with laboratory film and stored in the dark at or below 4 °C until analysis. As a measure of the extent of oxidative loss of Fe, the Ferrozine method (Hach, Colorado, USA) was used in the field to determine the dissolved Fe(II) concentration47. This analysis was conducted by taking an aliquot immediately from the Van dorn sampler which was directly transferred to a spectrophotometric cuvette already containing Ferrozine. This allowed immediate complex formation and thus minimised the potential for oxidative loss. The Ferrozine method was also used to measure the extent of post-sampling oxidation of the samples during handling in the laboratory and during separation of colloids by ultra-filtration.

Samples for Chl-a estimation were collected from each depth by filtering 50 mL of the water sample through 0.45 μm GF/C filters (Whatman, Maidstone,United Kingdom). The filters were then folded in half (sample side inward), individually wrapped in aluminium foil, kept in a container pre-wrapped with aluminium foil as extra precaution, and kept on ice until return to the laboratory, where the samples were frozen in dark and analysed using acetone extraction method within one week of collection.

### Diffusive gradients in thin films

To determine the concentration of labile trace elements and P in the lake water column, Chelex 100 (Bio Rad Laboratories Pty, Rosedale, Auckland, NZ) and ferrihydrite DGT devices were used in this study and prepared in-house after Zhang, et al.45 and Zhang and Davison57 respectively. The precision and accuracy of CDGT values determined using DGT solution probes was confirmed before deployment in the lake and were found to be accurate to within 5% for Cd and P (Supplementary information, Tables S1 and S2). The assembled probes were transferred to airtight containers, immersed in 0.01 M NaNO3 and purged with zero grade N2 overnight to ensure removal of dissolved oxygen. This procedure was carried out in a pre-N2-purged glove box as an extra precaution. The dissolved oxygen level in the 1 M NaNO3 solution containing the DGT probes was below the level of detection of a DO probe (<1 mg L−1). The probes were then transferred to the field in airtight containers and the DO content was measured before deployment, and never exceeded 1 mg L−1.

Duplicate Chelex and ferrihydrite probes were strung onto a nylon rope at the required depth intervals, secured using cable ties, and immersed into the lake with minimum delay (Fig. S4). Probes were co-located with temperature loggers (UTBI-001, TidbiT V2 Temp Loggers (Onset Computer Corp., Cape Cod, MA, USA)) which, in addition to CTD data, were used to calculate the average temperature at each respective depth. During each deployment, a duplicate set of DGT probes which had not been purged with N2 was deployed to examine variations caused by oxygen trapped in the hydrogels (Fig. S6). The results showed no difference in concentration of analyte species bound to resin gels (Fig. S6).

After typically 4–5 days, the probes were retrieved from the lake water column, rinsed with DI water (resistivity 18.2 MΩ) and kept in sealed, pre-cleaned individual plastic bags which were transferred to the lab at <5 °C in the dark and kept under the same conditions until analysis.

### Laboratory Methods

The syringe samples were transferred to the N2 purged glove box for ultrafiltration in a stirred-cell ultrafiltration system (AmiconR, model 8400) equipped with Millipore Ultracel 100 KDa (molecular weight cut-off (MWCO)) regenerated cellulose ultrafiltration discs (Millipore Corporation, Billerica, MA, U.S.A.) which provide size exclusion above approximately 5 nm. Ultrafiltration for separation of colloids has been a focus of potential artefact10,58 and maximum care was taken to avoid these potential issues. The ultrafiltration unit was washed with 5% HNO3 and rinsed three times with deionized (DI) water before each use. Everything else, except for Chl-a estimation, was cleaned by soaking in 10% HCl overnight, rinsing with DI water, soaking in 10% HNO3 at least overnight and finally rinsing three times with DI water, and then air dried under a Class 100 laminar flow hood. Glassware was cleaned using Clean Aid and rinsed five times before Chl-a measurements. Prior to lake water filtration, the filter discs were washed with 50 mL of DI water and an aliquot was analysed as a blank for potential contamination from the filtration step. All blanks ranged between ±1 μg L−1 for P and ±1 μg L−1 for Fe.

Before ultrafiltration, the syringes were opened in the glove box and the Fe (II) content was again determined by complexation with Ferrozine47 to determine any Fe(II) loss during transportation and storage. The sampling trip was repeated when there was oxidative loss >3%. A 50 mL 0.45 μm filtered sample was then ultra-filtered under minimum back pressure (<30 kPa) of zero grade N2 and aliquots were collected in pre-cleaned (as described earlier) conical tubes that were kept open in the purging N2 glove box overnight to remove oxygen in the tubes. The filter membranes were exhaustively cleaned with 50 mL of DI water between every sample to avoid cross contamination and metal sorption to filters59,60. Aliquots were collected frequently to determine if any contamination had occurred. The filtration process was quick (5 ml/min) and the N2 flow rate was adjusted accordingly to avoid rupturing of the filter discs. The filtered fraction collected for major cation analysis was acidified in the glove box by using the same protocol as described under the “Filtration and preservation in the field” heading. The screw capped bottles were then analysed by ICP-MS calibrated using NIST-traceable certified reference materials from Inorganic Ventures (Christiansburg, Virginia).

Filter fractions were categorised on the basis of their relative pore size. However, because there is no direct relationship between molecular weight and the size of nanoparticles60,61 the size estimates based on filtration used in this study are approximate. The four fractions defined in this study are as follows: dissolved (corresponding to the sample filtered at 0.45 μm), colloidal (corresponding to the difference between 0.45 μm–100 KDa permeates), and nominally dissolved (corresponding to the 100 kDa permeate).

DGT probes were disassembled within 24 hrs of collection from the field and retrieved resins (binding gels) were immediately transferred to 1 M HNO3 in the case of Chelex and 0.25 M H2SO4 for ferrihydrite. The ferrihydrite eluent was analysed using the molybdenum blue method62. Eluents of Chelex gels were analysed by ICP-MS.

#### Characterization of colloids

Colloids from Lake Ngapouri were characterized by using transmission electron microscopy (TEM) and atomic force microscopy (AFM). TEM samples were prepared by placing ~30 μl of water sample on support films (Formvar/carbon coated 200-mesh copper, obtained from ProSciTech, QLD, Australia) and allowed to dry under a zero-grade N2 atmosphere. Freshly cleaved mica sheets (ProSciTech, QLD, Australia) were placed in 15 mL of water sample for 24 hr immediately after return to the laboratory and then allowed to dry under the same N2 atmosphere in screw-capped bottles (thereby avoiding deposition of settling particles). We recognize the possible salt crystallization on the surface of mica sheets during sample preparation steps61 and so the results are interpreted accordingly.

#### Instrumental method

Both filter fractions, i.e. dissolved (<0.45 μm) and nominally dissolved (<100 KDa) were analysed on a Perkin Elmer (Waltham MA) quadrupole ICP-MS calibrated using certified reference materials. Internal standards of known concentration were also analysed to determine instrumental drift during analysis. Blanks (DI water, 18.2 MΩ) and procedural blanks (DI water transported to the field with samples) were also analysed following every sampling campaign and reported values of most trace elements were very low. The average concentration of Fe was 0.168 μmol L−1 for the nominally dissolved fraction i.e. <100 KDa and 0.179 μmol L −1 for colloidal fraction, i.e. 0.45 μm–100 KDa permeates.

#### Flow injection analyser

Dissolved reactive phosphorus (from 0.45 μm) was analysed using standard colorimetric methods (APHA, 1998)63 on a Lachat QuickChem flow injection analyser (Zellweger Analytics Inc.). A range of standards was prepared in DI water to confirm the analytical detection limit, which was found to be 0.004 mg L−1. DI water was used as a blank. The total phosphorus concentration was determined in unfiltered lake water sample by using the same method as described above but after persulphate digestion62.

#### Atomic Force Microscopy

AFM imaging was carried out by using an XE-100 atomic force microscope (Park systems Corporation, Suwon, Korea). The analyses were carried out in a true noncontact mode under ambient conditions using silicon cantilevers with a spring constant of 42 N m−1 (PPP-NCHR, Park Systems Corp.) and images were recorded in topography mode. Images were collected from 15 to 20 arbitrary areas on the film and were recorded in 256 × 256 pixel resolution. Particles were then grouped at 5 nm intervals based on measured heights.

#### Fluorometric estimation of Chl-a

The frozen glass fibre filter was transferred to clean tubes in very dim light and Chl-a was extracted by adding 5 mL buffered acetone. The filter was then ground by using an IKA® T10 basic Ultra-Turrax homogenizer (Thermo Fisher Scientific, Auckland, New Zealand). Five mL of buffered acetone was then added to wash any left overs off the grinding tip. The samples were allowed to steep for ~24 hr, shaking at least once during this time. After the steeping period, the samples were shaken and centrifuged for 10 min at 3300 rpm, then allowed to stand for ~30 mins in the dark at room temperature before reading.

Before proceeding to measure Chl-a, glass cuvettes were visually checked for any scratches. 5 mL of buffered acetone was read as a blank, followed by addition of 150 μL of 0.1 N HCl. Five mL of the sample was decanted off and read at the “High” setting, then diluted with buffered acetone when the reading was “Over”, with the dilution factor was subsequently taken into account in calculations. 150 μL of 0.1 N HCl was then added, mixed by gently tapping the cuvette, and re-read after 90 seconds.

### Calculations

The concentration of colloidal Fe and Colloidal P was calculated according to Equations 1 and 2 respectively:

$${\mathrm{Fe}}_{(c)}={{\rm{Fe}}}_{({d})}-{{\rm{Fe}}}_{(100{\rm{KDa}})}$$
(1)
$${{\rm{P}}}_{(c)}={{\rm{P}}}_{(d)}-{{\rm{P}}}_{(100{\rm{KDa}})}$$
(2)

where Fe(c) represents the colloidal fraction and Fe(d) is filtrate through 0.45 μm filter membrane and Fe(100KDa) is the permeate filtering through 100 KDa filter membranes.

#### Calculating CDGT

CDGT concentrations were calculated for each resin disc by the following steps:

First, the mass (M) of P and Fe was calculated in respective resin gels using Equation 3

$${\rm{M}}={\rm{Ce}}({{\rm{V}}}_{{\rm{acid}}}+{{\rm{V}}}_{{\rm{gel}}})/fe$$
(3)

where Ce is concentration of P or Fe in resin gel eluent, Vacid is the volume of acid added to the resin gels, Vgel is volume of the resin gel and fe is the elution factor of P or Fe.

The DGT concentration was then calculated by Equation (4)

$${{\rm{C}}}_{DGT}=M\Delta g/(DtA)$$
(4)

where Δg is diffusive boundary layer (thickness of diffusive gel + thickness of the filter membrane), D = diffusion coefficient of the elements in gel as reported by Zhang and Davison37, t is deployment time (s) and A is exposure area (cm2). This equation was used for calculating CDGT value in the solution probes by using geometric area of standard solution DGT holders. When DGTs are immersed in water, a thin layer of water which is in line with the exposed window of the device is usually inactive and transport of material through this layer, known as Diffusive Boundary Layer (DBL), is believed to happen through diffusion only. The effect of DBL was minimised during this study by deploying devices for longer periods of time. Moreover, a 0.8 mm thick layer of diffusive gel along with 0.14 mm thick filter membrane was used for simplicity.

#### Calculating Chl-a

The concentration of Chl-a was calculated by using the following equation:

$$Chl-a({{\rm{\mu }}\mathrm{gL}}^{-1})={{\rm{F}}}_{{\rm{s}}}[({\rm{r}}/({\rm{r}}-1))\,({{\rm{R}}}_{1}-{{\rm{R}}}_{2})]\,[({{\rm{V}}}_{{\rm{e}}}{\rm{Xdf}})/{{\rm{V}}}_{{\rm{f}}}]$$

where Fs is response factor of the fluorometer, R1 and R2 are readings before and after the acidification step, respectively, r is the acidification coefficient, df is dilution factor, Ve is extraction volume and Vf is the volume of water filtered.

The depth of the thermocline was determined by calculation of dT/dz where z is the depth of water column increasing downwards42 and T is temperature. The water column was considered fully mixed when dT/dz was > −0.225 °C m−1 Data from CTD profiles were averaged over 0.125 m intervals to minimize noise. The top 5 m of the water profile was not included in thermocline calculations to avoid temporary diurnal thermoclines formed under calm conditions during daytime.

## References

1. 1.

Anderson, D. M., Glibert, P. M. & Burkholder, J. M. Harmful algal blooms and eutrophication: Nutrient sources, composition, and consequences. Estuaries 25, 704–726, https://doi.org/10.1007/bf02804901 (2002).

2. 2.

Elser, J. & Bennett, E. A broken biogeochemical cycle. Nature 478, 29–31, https://doi.org/10.1038/478029a (2011).

3. 3.

Kpodonu, A. T. N. K., Hamilton, D. P., Hartland, A., Laughlin, D. C. & Lusk, C. H. Coupled use of sediment phosphorus speciation and pigment composition to infer phytoplankton phenology over 700 years in a deep oligotrophic lake. Biogeochemistry 129, 181–196, https://doi.org/10.1007/s10533-016-0227-3 (2016).

4. 4.

Guildford, S. J. & Hecky, R. E. Total nitrogen, total phosphorus, and nutrient limitation in lakes and oceans: Is there a common relationship? Limnology and Oceanography 45, 1213–1223, https://doi.org/10.4319/lo.2000.45.6.1213 (2000).

5. 5.

Downs, T. M., Schallenberg, M. & Burns, C. W. Responses of lake phytoplankton to micronutrient enrichment: a study in two New Zealand lakes and an analysis of published data. Aquatic Sciences 70, 347–360, https://doi.org/10.1007/s00027-008-8065-6 (2008).

6. 6.

Parkyn, S. M., Davies-Colley, R. J., Halliday, N. J., Costley, K. J. & Croker, G. F. Planted Riparian Buffer Zones in New Zealand: Do They Live Up to Expectations? Restoration Ecology 11, 436–447, https://doi.org/10.1046/j.1526-100x.2003.rec0260.x (2003).

7. 7.

Wilson, T. A., Amirbahman, A., Norton, S. A. & Voytek, M. A. A record of phosphorus dynamics in oligotrophic lake sediment. Journal of Paleolimnology 44, 279–294, https://doi.org/10.1007/s10933-009-9403-y (2010).

8. 8.

Van Moorleghem, C., Six, L., Degryse, F., Smolders, E. & Merckx, R. Effect of Organic P Forms and P Present in Inorganic Colloids on the Determination of Dissolved P in Environmental Samples by the Diffusive Gradient in Thin Films Technique, Ion Chromatography, and Colorimetry. Analytical Chemistry 83, 5317–5323, https://doi.org/10.1021/ac200748e (2011).

9. 9.

Van Moorleghem, C., De Schutter, N., Smolders, E. & Merckx, R. The bioavailability of colloidal and dissolved organic phosphorus to the alga Pseudokirchneriella subcapitata in relation to analytical phosphorus measurements. Hydrobiologia 709, 41–53, https://doi.org/10.1007/s10750-013-1442-8 (2013).

10. 10.

Buffle, J. & Leppard, G. G. Characterization of Aquatic Colloids and Macromolecules. 2. Key Role of Physical Structures on Analytical Results. Environmental Science & Technology 29, 2176–2184, https://doi.org/10.1021/es00009a005 (1995).

11. 11.

Maher, W. & Woo, L. Procedures for the storage and digestion of natural waters for the determination of filterable reactive phosphorus, total filterable phosphorus and total phosphorus. Analytica Chimica Acta 375, 5–47, https://doi.org/10.1016/s0003-2670(98)00274-8 (1998).

12. 12.

Olsen, S. Recent trends in the determination of orthophosphate in water. (NV Noord-Hollandsche Uitgevers Maatschappij, 1967).

13. 13.

Boström, B., Andersen, J. M., Fleischer, S. & Jansson, M. Exchange of phosphorus across the sediment-water interface. Hydrobiologia 170, 229–244, https://doi.org/10.1007/bf00024907 (1988).

14. 14.

DePinto, J. V., Young, T. C. & Martin, S. C. Algal-Available Phosphorus in Suspended Sediments from Lower Great Lakes Tributaries. Journal of Great Lakes Research 7, 311–325, https://doi.org/10.1016/s0380-1330(81)72059-8 (1981).

15. 15.

Zhang, H., Boegman, L., Scavia, D. & Culver, D. A. Spatial distributions of external and internal phosphorus loads in Lake Erie and their impacts on phytoplankton and water quality. Journal of Great Lakes Research 42, 1212–1227, https://doi.org/10.1016/j.jglr.2016.09.005 (2016).

16. 16.

Søndergaard, M., Jensen, J. P. & Jeppesen, E. Role of sediment and internal loading of phosphorus in shallow lakes. Hydrobiologia 506–509, 135–145, https://doi.org/10.1023/b:hydr.0000008611.12704.dd (2003).

17. 17.

Davison, W. Iron and manganese in lakes. Earth-Science Reviews 34, 119–163, https://doi.org/10.1016/0012-8252(93)90029-7 (1993).

18. 18.

Imberger, J. & Hamblin, P. F. Dynamics of Lakes, Reservoirs, and Cooling Ponds. Annual Review of Fluid Mechanics 14, 153–187, https://doi.org/10.1146/annurev.fl.14.010182.001101 (1982).

19. 19.

Gao, Y., Cornwell, J. C., Stoecker, D. K. & Owens, M. S. Influence of cyanobacteria blooms on sediment biogeochemistry and nutrient fluxes. Limnology and Oceanography 59, 959–971 (2014).

20. 20.

Gao, Y. et al. High-resolution imaging of labile phosphorus and its relationship with iron redox state in lake sediments. Environmental Pollution 219, 466–474, https://doi.org/10.1016/j.envpol.2016.05.053 (2016).

21. 21.

Amirbahman, A., Pearce, A. R., Bouchard, R. J., Norton, S. A. & Kahl, J. S. Relationship between hypolimnetic phosphorus and iron release from eleven lakes in Maine, USA. Biogeochemistry 65, 369–386 (2003).

22. 22.

Hongve, D. Cycling of iron, manganese, and phosphate in a meromictic lake. Limnology and Oceanography 42, 635–647, https://doi.org/10.4319/lo.1997.42.4.0635 (1997).

23. 23.

Gunnars, A., Blomqvist, S., Johansson, P. & Andersson, C. Formation of Fe(III) oxyhydroxide colloids in freshwater and brackish seawater, with incorporation of phosphate and calcium. Geochimica et Cosmochimica Acta 66, 745–758, https://doi.org/10.1016/s0016-7037(01)00818-3 (2002).

24. 24.

Mai, T. H. N. et al. Adsorption and desorption of arsenic to aquifer sediment on the Red River floodplain at Nam Du, Vietnam. Geochimica et Cosmochimica Acta 142, 587–600, https://doi.org/10.1016/j.gca.2014.07.014 (2014).

25. 25.

Baalousha, M. & Lead, J. R. Characterization of Natural Aquatic Colloids (<5 nm) by Flow-Field Flow Fractionation and Atomic Force Microscopy. Environmental Science & Technology 41, 1111–1117, https://doi.org/10.1021/es061766n (2007).

26. 26.

Baalousha, M., Manciulea, A., Cumberland, S., Kendall, K. & Lead, J. R. Aggregation and Surface Properties of Iron Oxide Nanoparticles: Influence of pH and Natural Organic Matter. Environmental Toxicology and Chemistry 27, 1875, https://doi.org/10.1897/07-559.1 (2008).

27. 27.

Gustafsson, C. & Gschwend, P. M. Aquatic colloids: Concepts, definitions, and current challenges. Limnology and Oceanography 42, 519–528, https://doi.org/10.4319/lo.1997.42.3.0519 (1997).

28. 28.

Leppard, G. G., Buffle, J., De Vitre, R. R. & Perret, D. Ultrastructure and physical characteristics of a distinctive colloidal iron particulate isolated from a small eutrophic lake. Archiv fuer Hydrobiologie AHYBA 4, 113 (1988).

29. 29.

Pizarro, J. et al. Coagulation/sedimentation of submicron iron particles in a eutrophic lake. Water Research 29, 617–632, https://doi.org/10.1016/0043-1354(94)00167-6 (1995).

30. 30.

Tipping, E., Woof, C. & Cooke, D. Iron oxide from a seasonally anoxic lake. Geochimica et Cosmochimica Acta 45, 1411–1419, https://doi.org/10.1016/0016-7037(81)90275-1 (1981).

31. 31.

Laxen, D. P. H. & Chandler, I. M. Size distribution of iron and manganese species in freshwaters. Geochimica et Cosmochimica Acta 47, 731–741, https://doi.org/10.1016/0016-7037(83)90107-2 (1983).

32. 32.

Hartland, A., Larsen, J. R., Andersen, M. S., Baalousha, M. & O’Carroll, D. Association of Arsenic and Phosphorus with Iron Nanoparticles between Streams and Aquifers: Implications for Arsenic Mobility. Environmental Science & Technology 49, 14101–14109, https://doi.org/10.1021/acs.est.5b03506 (2015).

33. 33.

Lapworth, D. J., Stolpe, B., Williams, P. J., Gooddy, D. C. & Lead, J. R. Characterization of Suboxic Groundwater Colloids Using a Multi-method Approach. Environmental Science & Technology 47, 2554–2561, https://doi.org/10.1021/es3045778 (2013).

34. 34.

Darch, T., Blackwell, M. S. A., Hawkins, J. M. B., Haygarth, P. M. & Chadwick, D. A Meta-Analysis of Organic and Inorganic Phosphorus in Organic Fertilizers, Soils, and Water: Implications for Water Quality. Critical Reviews in Environmental Science and Technology 44, 2172–2202, https://doi.org/10.1080/10643389.2013.790752 (2014).

35. 35.

Ding, S. et al. In situ, high-resolution evidence for iron-coupled mobilization of phosphorus in sediments. Vol. 6 (2016).

36. 36.

Rydin, E. Potentially Mobile Phosphorus in Lake Erken Sediment. Vol. 34 (2000).

37. 37.

Zhang, H. & Davison, W. Diffusional characteristics of hydrogels used in DGT and DET techniques. Analytica Chimica Acta 398, 329–340, https://doi.org/10.1016/s0003-2670(99)00458-4 (1999).

38. 38.

Davison, W. & Zhang, H. Progress in understanding the use of diffusive gradients in thin films (DGT) – back to basics. Environmental Chemistry 9, 1–13, https://doi.org/10.1071/EN11084 (2012).

39. 39.

Zhang, H. & Davison, W. Use of diffusive gradients in thin-films for studies of chemical speciation and bioavailability. Environmental Chemistry 12, 85, https://doi.org/10.1071/en14105 (2015).

40. 40.

Zhang, C., Ding, S., Xu, D., Tang, Y. & Wong, M. H. Bioavailability assessment of phosphorus and metals in soils and sediments: a review of diffusive gradients in thin films (DGT). Environmental Monitoring and Assessment 186, 7367–7378, https://doi.org/10.1007/s10661-014-3933-0 (2014).

41. 41.

Buffle, J., Wilkinson, K. J., Stoll, S., Filella, M. & Zhang, J. A Generalized Description of Aquatic Colloidal Interactions: The Three-colloidal Component Approach. Environmental Science & Technology 32, 2887–2899, https://doi.org/10.1021/es980217h (1998).

42. 42.

Hamilton, D. P., O’Brien, K. R., Burford, M. A., Brookes, J. D. & McBride, C. G. Vertical distributions of chlorophyll in deep, warm monomictic lakes. Aquatic Sciences 72, 295–307, https://doi.org/10.1007/s00027-010-0131-1 (2010).

43. 43.

Özkundakci, D., Hamilton, D. & Gibbs, M. Hypolimnetic phosphorus and nitrogen dynamics in a small, eutrophic lake with a seasonally anoxic hypolimnion. Vol. 661 (2011).

44. 44.

Hartland, A., Andersen, M. S. & Hamilton, D. P. Phosphorus and arsenic distributions in a seasonally stratified, iron- and manganese-rich lake: microbiological and geochemical controls. Environmental Chemistry 12, 708, https://doi.org/10.1071/en14094 (2015).

45. 45.

Zhang, H. & Davison, W. Performance characteristics of diffusion gradients in thin films for the in situ measurement of trace metals in aqueous solution. Analytical chemistry 67, 3391–3400 (1995).

46. 46.

Fatin-Rouge, N., Starchev, K. & Buffle, J. Size Effects on Diffusion Processes within Agarose Gels. Biophysical Journal 86, 2710–2719, https://doi.org/10.1016/s0006-3495(04)74325-8 (2004).

47. 47.

Viollier, E., Inglett, P. W., Hunter, K., Roychoudhury, A. N. & Van Cappellen, P. The ferrozine method revisited: Fe(II)/Fe(III) determination in natural waters. Applied Geochemistry 15, 785–790, https://doi.org/10.1016/s0883-2927(99)00097-9 (2000).

48. 48.

Anastácio, A. S., Harris, B., Yoo, H.-I., Fabris, J. D. & Stucki, J. W. Limitations of the ferrozine method for quantitative assay of mineral systems for ferrous and total iron. Geochimica et Cosmochimica Acta 72, 5001–5008, https://doi.org/10.1016/j.gca.2008.07.009 (2008).

49. 49.

Ting, D.-S. & Appan, A. General characteristics and fractions of phosphorus in aquatic sediments of two tropical reservoirs. Water Science and Technology 34, 53–59, https://doi.org/10.1016/S0273-1223(96)00724-X (1996).

50. 50.

Hassellov, M. & von der Kammer, F. Iron Oxides as Geochemical Nanovectors for Metal Transport in Soil-River Systems. Elements 4, 401–406, https://doi.org/10.2113/gselements.4.6.401 (2008).

51. 51.

Hupfer, M., Ruübe, B. & Schmieder, P. Origin and diagenesis of polyphosphate in lake sediments: A 31PuNMR study. Limnology and Oceanography 49, 1–10, https://doi.org/10.4319/lo.2004.49.1.0001 (2004).

52. 52.

Ståhl-Delbanco, A. & Hansson, L.-A. Effects of bioturbation on recruitment of algal cells from the “seed bank” of lake sediments. Limnology and Oceanography 47, 1836–1843, https://doi.org/10.4319/lo.2002.47.6.1836 (2002).

53. 53.

Mooij, W. M. et al. Challenges and opportunities for integrating lake ecosystem modelling approaches. Aquatic Ecology 44, 633–667, https://doi.org/10.1007/s10452-010-9339-3 (2010).

54. 54.

Gottselig, N. et al. Phosphorus Binding to Nanoparticles and Colloids in Forest Stream Waters. Vadose Zone Journal 16, 0, https://doi.org/10.2136/vzj2016.07.0064 (2017).

55. 55.

Jiang, X. et al. Phosphorus Containing Water Dispersible Nanoparticles in Arable Soil. Journal of Environment Quality 44, 1772, https://doi.org/10.2134/jeq.2015.02.0085 (2015).

56. 56.

Montalvo, D., Degryse, F. & McLaughlin, M. J. Natural Colloidal P and Its Contribution to Plant P Uptake. Environmental Science & Technology 49, 3427–3434, https://doi.org/10.1021/es504643f (2015).

57. 57.

Zhang, H., Davison, W., Gadi, R. & Kobayashi, T. In situ measurement of dissolved phosphorus in natural waters using DGT. Analytica Chimica Acta 370, 29–38, https://doi.org/10.1016/s0003-2670(98)00250-5 (1998).

58. 58.

Buesseler, K. O. et al. An intercomparison of cross-flow filtration techniques used for sampling marine colloids: Overview and organic carbon results. Marine Chemistry 55, 1–31, https://doi.org/10.1016/s0304-4203(96)00046-1 (1996).

59. 59.

Wen, L.-S., Stordal, M. C., Tang, D., Gill, G. A. & Santschi, P. H. An ultraclean cross-flow ultrafiltration technique for the study of trace metal phase speciation in seawater. Marine Chemistry 55, 129–152, https://doi.org/10.1016/s0304-4203(96)00052-7 (1996).

60. 60.

Benoit, G. & Rozan, T. F. The influence of size distribution on the particle concentration effect and trace metal partitioning in rivers. Geochimica et Cosmochimica Acta 63, 113–127, https://doi.org/10.1016/s0016-7037(98)00276-2 (1999).

61. 61.

Baalousha, M. & Lead, J. R. Characterization of natural and manufactured nanoparticles by atomic force microscopy: Effect of analysis mode, environment and sample preparation. Colloids and Surfaces A: Physicochemical and Engineering Aspects 419, 238–247, https://doi.org/10.1016/j.colsurfa.2012.12.004 (2013).

62. 62.

Murphy, J. & Riley, J. P. A modified single solution method for the determination of phosphate in natural waters. Analytica Chimica Acta 27, 31–36, https://doi.org/10.1016/s0003-2670(00)88444-5 (1962).

63. 63.

APHA. (American Public Health Association Washington, 1998).

## Acknowledgements

We acknowledge the valuable contributions of Prof. Bill Perry in providing training in R, Jackson White for help with preparing graphical illustrations, and the assistance of Andrew Pearson and Chris Eager during field work. We are thankful to the University of Waikato technical staff for their assistance in completing the field work and laboratory analysis. This research was enabled in part by a Rutherford Discovery Fellowship award to AH (RDF-UOW1601) and funding through the Ministry of Business, Innovation and Employment for Enhancing the Health and Resilience of New Zealand Lakes (UOWX1503).

## Author information

Authors

### Contributions

H.S. completed all laboratory and fieldwork, created Figures 2, 4, 5 and 6, and co-wrote the manuscript with A.H. who designed the study, co-wrote the manuscript, created Figures 1, 3, 7, 8 and 9 and obtained funding for the research. N.J.L. trained H.S. in manufacture of DGT probes and contributed to the writing of the manuscript. M.B. and M.S. completed TEM and AFM analyses respectively and D.S. assisted in fieldwork sampling design and execution. M.M. and D.H. co-supervised H.S. and contributed to the writing of the manuscript.

### Corresponding author

Correspondence to A. Hartland.

## Ethics declarations

### Competing Interests

The authors declare no competing interests.

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## Rights and permissions

Reprints and Permissions

Saeed, H., Hartland, A., Lehto, N.J. et al. Regulation of phosphorus bioavailability by iron nanoparticles in a monomictic lake. Sci Rep 8, 17736 (2018). https://doi.org/10.1038/s41598-018-36103-x

• Accepted:

• Published:

### Keywords

• Iron Nanoparticles
• Deep Chlorophyll Maximum (DCM)
• Dissolved Reactive Phosphate (DRP)
• Phytoplankton Community Dynamics
• Lake Water Column