Eight mesocosms (M1–M8) were deployed 6 km off the Peruvian coast close to San Lorenzo Island (12.0555° S, 77.2348° W). Each mesocosm consisted of a cylindrical, 17 m long polyurethane bag (2 m diameter) attached to a conical sediment trap (2 m) in an 8 m tall flotation frame. On 25 February the water was enclosed in the polyurethane bag containing approximately 55 000 L of water (Fig. 1). The experiment started on 25 February 2017 (Day 0) and was conducted for 50 d (for details on the experiment and sampling schedule, see Bach et al., 2020). After their closure, the waters inside the mesocosms remained isolated from surrounding Pacific seawater. Repetitive sampling of the water column within the mesocosms included an integrated water sample from the mixed layer (ML) and the bottom layer (BL). The lower sampling depth for the ML was adjusted during the experiment to account for a shift in the ML and oxycline depth and included 5, 10, and 13 m, respectively (Table S1 in the Supplement).

DW for the simulated upwelling event was collected along the Instituto del Mar del Perú (IMARPE) time-series transect (Graco et al., 2017). The water was sampled on Day 5 (St. 1; 30 m, St. 12.028° S, 77.22° W) and Day 10 (St. 3, 70 m, 12.04° S, 77.38° W). Samples from both stations were characterized by different nutrient ratios and therefore described as “extreme” (DIN : P of 0.2 molmol-1; DIN : Si of 0.02 molmol-1) and “moderate” (DIN : P of 1.7 molmol-1; DIN : Si of 0.2 molmol-1) DW. On 8 and 9 March 2017 (Days 11 and 12), we exchanged water enclosed in each mesocosm with water collected from Station 1 (M1, M4, M5, M8) or Station 3 (M2, M3, M6, M7). The exchange was carried out in two steps using a submersible pump (Grundfos SP 17-5R, pump rate ∼18 m3h-1). On Day 8, BW was exchanged. We installed the pump for about 30 to 40 min in each mesocosm and pumped 9 m3 out of each bag from a depth of 11 to 12 m. On Day 11, the pump was installed inside the collector bags, and 10 m3 of water were injected into a depth of 14 to 17 m. On Day 12, the procedure was repeated to exchange the surface water. In that case, approximately 10 m3 of water were removed from 8 to 9 m depth and replaced with 12 m3 of water evenly to the depth range from 1 to 9 m. This corresponded to an addition of 40 % DW to the surface (mixed) layer.

Sampling and CTD casts within the mesocosms were undertaken from small boats that departed from La Punta Harbor (Callao) and were transported to laboratories in Club Náutico Del Centro Naval and the Instituto del Mar del Perú (IMARPE) for filtration and nutrient measurements. Every 2nd day, subsamples for nutrients and natural Si isotopes were sampled from the mixed layer (ML). For δ30SidSi and δ30SibSi, 115 to 2000 mL of seawater was filtered through a 0.65 µm Polycarbonate filter (Whatman^®, 0.65 µm pore size, 47 mm). Immediately after filtration, the seawater filtrate for δ30SidSi was acidified to pH 2 and stored in the dark. Filters for δ30SibSi were dried at 40 °C. Dissolved and particulate Si samples were later processed at GEOMAR, Helmholtz Centre for Ocean Research, Kiel, Germany.

Samples for inorganic nutrients were filtered (0.45 µm filter, Sterivex, Merck) immediately after arrival in the laboratories of IMARPE. The subsequent analyses of dSi, PO34-, NO3-, NO2- concentrations were carried out using an autosampler (XY2 autosampler, SEAL Analytical) and a continuous flow analyzer (QuAAtroAutoAnalyzer, SEAL Analytical) connected to a fluorescence detector (FP-2020, JASCO). Phosphate and dSi were analyzed colorimetrically following the procedures by Murphy and Riley (1962) and Mullin and Riley (1955), respectively. Nitrate and nitrite were quantified via the formation of a pink azo dye as established by Morris and Riley (1963). Ammonium concentrations were determined fluorometrically (Kérouel and Aminot, 1997). The accuracy was monitored by including certified reference materials (CRM; BW, KANSO) during measurement sessions and ranged between ±5 % to ±10 %. For further analytical details see Bach et al. (2020). BSi filters were leached with 0.1 M NaOH at 85 °C in 60 mL Nalgene polypropylene bottles at Club Náutico Del Centro Naval. After 135 min, the leaching process was terminated with 0.05 M H2SO4, and the dSi concentration was measured spectrophotometrically following Hansen and Koroleff (1999). The bSi concentration was measured for all size fractions, as well as for size classes smaller than 20 µm and larger than 20 µm.

Seawater subsamples from the ML (25 and 50 mL) were analyzed for phytoplankton assemblages (including dead cells) in the laboratories of IMARPE (Instituto del Mar del Perú, La Punta). Cell counts were carried out according to the method of Utermöhl (1958), in which the sample was allowed to settle for approximately 24 h. Cells were counted under Nikon and Leica inverted light microscopes at ×125 magnification (10×=10.01 µm; ocular: 12.5×) and expressed in cellsL-1×106. In addition, we calculated the relative abundance (in %) considering only marine silicifiers (diatoms and silicoflagellates) and the relative abundance (in %) for siliceous and non-siliceous plankton (e.g., dinoflagellates, coccolithophores) from microscopy data. More information on all phytoplankton groups analyzed, including pigment data, can be found in Bach et al. (2020).

To estimate the biovolume of various species, we referred to the 2024 Nordic Microalgae Biovolume List (NOMP, Olenina et al., 2006). This resource is among the most comprehensive compilations currently available. Table S2 in the Supplement presents examples of encountered diatom species (e.g., Skeletonema costatum, mean diameter of 6 μm), some of the largest observed species (e.g., Actinocyclus, mean diameter 46 µm), as well as silicoflagellates (mean diameter between 23 and 30 µm) during days 13 and 17. For Dictyocha octonaria, the biovolume and diameter could not be directly retrieved; instead, the diameter was determined from SEM images (example shown in Fig. 5d), and the biovolume was estimated by assuming a half-sphere shape, which is also applied for Dictyocha fibula (Olenina et al., 2006). 1V=π12⋅d3 Where V is the Biovolume, and d the diameter. Given the broad variability in biovolume (Table S2), the values provided should be considered as approximate estimates.

For preparation of samples for δ30SidSi measurements, the seawater pH (9 to 10) was raised with 1 M NaOH to scavenge dSi with the precipitated Mg(OH)2 (MAGIC, Karl and Tien, 1992; Reynolds et al., 2006; Grasse et al., 2016). BSi filters for δ30SibSi measurements were treated according to Varela et al. (2004) including a leaching step with 0.2 N NaOH in a 90 °C water bath (for details, see also Grasse et al., 2021). In a second leach step, the filter was treated with 0.5 mL of 2.5 M HF for 48 h to dissolve lithogenic silicate (LSi). Although Ragueneau and Tréguer (1994) pointed out that up to 15 % of lithogenic silicate (LSi) can dissolve during sodium hydroxide digestion, our own measurements provide evidence that LSi dissolution in our samples, especially the samples used to determine the fractionation factor, was much lower (less than 5 %), except for 2 samples from Day 1. Most likely, coastal nearshore waters captured in the mesocosms contained higher lithogenic content, which subsequently sank out of the surface layer after the mesocosms were closed. Assuming that the samples contained up to 15 % LSi, the reported δ30Si values may be underestimated by 0.2 ‰, assuming a mean isotope signature of -1.07 ‰ for clay minerals. Lithogenic primary minerals are heavier at -0.2 ‰ (Sutton et al., 2018), resulting in an offset of 0.02 ‰. Both values are within the analytical error margin. As the lithogenic content of most samples was less than 15 %, the δ30SibSi values were not corrected. The Al/Si ratios in the bSi samples were not measured. As demonstrated by Grasse et al. (2021), the Al correction method is limited since Al/Si ratios depend on external factors and diatom cell conditions (e.g., living versus dead), rather than exclusively indicating lithogenic contamination. Furthermore, these ratios may have been heavily biased in the mesocosms, which were not trace metal-free.

The dissolved sample (dSi, bSi) was loaded onto a cation exchange column (AG50X8, 200 mesh, Biorad^®; for details, see Grasse et al., 2021). Samples with low dSi concentrations (<4 µmolL-1) were gently evaporated to double the concentration after column chemistry. According to Hughes et al. (2011), this should not affect δ30Si. Detailed preparation protocols are described in Ehlert et al. (2012) and Grasse et al. (2013).

Samples were analyzed using an Aridus II nebulizer coupled to a Nu Plasma MC-ICP-MS (Nu Instruments™, Wrexham, UK). Each analysis involved 50 to 60 cycles in sample-standard bracketing mode against NBS28. Si isotope compositions are reported in the δ notation, representing the deviation of the isotope ratio of the sample (Rsample) from that of a reference standard (Rstandard) in parts per thousand (‰). 2δ30Si=RsampleRstandard-1⋅1000 The accuracy of measurements was checked daily using solid reference standards and seawater standards. Repeated measurements of BB, Diatomite, and IRMM18 resulted in mean δ30Si of -10.66 ‰ ± 0.18 ‰ (2 s.d., n=14), +1.22 ‰ ± 0.14 ‰ (2 s.d., n=28) and -1.43 ‰ ± 0.23 ‰ (2 s.d., n=10), which are in good agreement with Reynolds et al. (2007). The seawater inter-calibration standards Aloha 1000 m and Aloha 300 m resulted in +1.25 ‰ ± 0.16 ‰ (2 s.d.; n=53) and +1.72 ‰ ± 0.10 ‰ (M2 s.d., n=8), which is in excellent agreement with the mean values obtained by the GEOTRACES Si isotope inter-calibration study (+1.24 ‰ ± 0.20 ‰; +1.68 ‰ ± 0.35 ‰, Mean ± 2 s.d.; Grasse et al., 2017).

The mesocosms are closed containers, in which dSi was replenished on day 11 (bottom water) and day 12 (surface water). The isotope fractionation factor between silicifiers and seawater (εsilicos) can, therefore, be calculated after dSi replenishment, assuming a Rayleigh-type fractionation (closed system) (Mariotti et al., 1981; De La Rocha et al., 1997; Sun et al., 2014). Although the closed system model best describes the surface layer of the mesocosm, it should be acknowledged that it is not a fully closed system. Biomass and its corresponding Si isotope signatures can be lost through sinking. In addition, dissolution can potentially lower isotope signatures and affect the overall results. Unfortunately, we were not able to calculate the Si isotope fractionation factor prior to the addition of deep water (Phase I), given that the dSi concentrations were partly increasing over time as a result of mixing and changes in the ML depth (see Fig. 2a). The Rayleigh-type model is expressed by the following equations. 3δ30SidSi=δ30Siinit+εSilicos×ln⁡(f)4δ30SibSi=δ30Siinit-εSilicos×f1-f×ln⁡(f)5f=([dSi]final/[dSi]source) where δ30SidSi, δ30Siinit and δ30SibSi are the Si isotope values of the substrate (dSi) and the initial dSi concentration on day 13 (no sampling on day 12) as well as of the product (bSi) on day 17, respectively. The remaining fraction f was calculated according to Eq. (5) using the dSi concentration from day 13 (dSi_source) and 17 (dSi_final). The time range between days 13 and day 17 was used as it marked the initial decline in dSi following the increase directly linked to deep water addition. After day 17, dSi was increasing again (Fig. 2a).

To calculate the Si isotope fractionation factors in the two different mesocosms where silicoflagellates were on both days the dominant taxa, Eq. (4) was solved for εsilicos and calculated according to the following Eq. (6) (Sun et al., 2014), which uses a Rayleigh-type model and takes into account the utilized fraction f. 6εSilicos=-(δ30SibSi-δ30Siinit)/f1-f×ln⁡(f) The offset (Δ30Si, also called the apparent fractionation factor) between δ30SibSi and δ30SidSi was calculated according to Eq. (7) (Table 1). The apparent fractionation factor is often determined in field studies investigating the marine Si cycle (e.g., Varela et al., 2016; Grasse et al., 2021) 7Δ30Si=δ30SibSi-δ30SidSi The propagated error for Δ30Si is defined as follows 82s.d.bSi2+2s.d.dSi2

Monte Carlo simulations were performed to propagate the uncertainty in the calculated ε derived from analytical errors of Si isotope measurements (Robert and Casella, 2004). 2000 Monte Carlo simulations were performed assuming a normal distribution. This allows a random generation of possible results that substitute a range of values based on a probability distribution of measured δ30Si values. The simulation was performed in Excel with the Analysis ToolPak, allowing random number generation within the 2 s.d. error (0.2 ‰). Additionally, we performed a sensitivity analysis incorporating deviations in initial dSi and δ30Si values. This included 10 % over and underestimation of the dSi source and a ±0.2 ‰ shift in δ30Si.

Nutrient concentrations, as well as δ30SidSi, were determined for both DW types (“extreme” and “moderate”, see Table 1). Both DWs showed similar dSi concentrations (DW_extreme: 17.4 µmolL-1; DW_moderate: 19.6 µmolL-1) and PO43- concentrations (∼2.5 µmolL-1) but deviated in their N : Si ratio (0.02 versus 0.2) as well as N : P ratios (0.1 versus 1.7). The δ30SidSi values for DW “moderate” and DW “extreme” were indistinguishable within error with +1.54 ‰ ± 0.12 ‰ and +1.47 ‰ ± 0.19 ‰, respectively.

The dSi concentration in the ML initially ranged between 2.1 and 4.1 µmolL-1 within the first ten days (Phase I) in M1, M2 and M7 (Fig. 2a). This variability in dSi concentrations during the first ten days was mainly the result of a shift in sampling depth on day 4 to follow the deepening of the ML depth (Fig. 2a). This resulted in an increase in dSi compared to Day 1 (Fig. 2b). After DW addition (Day 12), dSi in the ML increased to 9 µmolL-1 (day 13) in M1 and M2 and 6 µmolL-1 in M7, due to addition of DW with higher dSi (17.4 µmolL-1; DW: 19.6 µmolL-, Table 1).

Between day 13 and day 20, dSi in the ML decreased by 20 % to 50 % resulting in dSi concentrations between 3 and 8 µmolL-1 (Fig. 2a and b). From day 20 to 35, dSi increased in M2 and M7, while it remained relatively constant in M1. By the end of the experiment (after Day 35), dSi had decreased to approximately 3 µmolL-1 in M1, M2, and M7, resulting in an overall drawdown between 30 % and 70 %.

During the first 10 d of the experiment DIN : dSi ratios (DIN=NOx+NH4) ranged between 0.5 and 1.5 molmol-1 and decreased for both treatments after the DW addition (Fig. 2c). Between days 15 and 40, the DIN : dSi ratio was close to 0 and only slightly increased at the end of the experiment (after Day 38) due to surface eutrophication with NH4+ by defecating seabirds (Inca tern, Larosterna inca, Bach et al., 2020, see also video from Boxhammer et al., 2019).

DIN : P values were extremely low and showed a pattern similar to DIN : dSi. While they were higher during the first 10 d (1 to 7 molmol-1), they decreased sharply after the DW addition and were close to 0 molmol-1 after day 15 with a slight increase towards the end of the experiment (Fig. 2d). Phosphate concentrations were rather stable throughout the experiment (mostly between 1.5 and 2 µmolL-1), NOx was nearly depleted by day 17 (<0.1 µmolL-1).

At the beginning of the experiment, bSi concentrations in M1, M2, and M7 were similar, ranging from 3 to 4 µmolL-1, and mainly consisted of particles smaller than 20 µm, which accounted for 40 % to 50 % of total bSi (Fig. S1 in the Supplement). In M2 and M7 a strong increase in bSi was observed between days 3 and 4 (maximum bSi: 6.5 µmolL-1), whereas M1 remained constant over time (Table 1; Fig. 3a). This was reflected by an increase in diatom cell counts obtained from microscopy data (see Sect. 2.3; Table S1, Fig. 3b), which were highest during this time with abundances of 10 and 18×106 cellsL-1 in M7 and M2, respectively. When only considering silicifying plankton, diatoms contributed up to 100 % of the cell counts (Fig. 3c). The most common diatom species were Skeletonema costatum, Cerataulina pelagica, Cylindrotheca closterium, Guinardia delicatula, Thalassiosira sp., Leptocylindrus danicus and Entomoneis alata var. alata (new name: Entomoneis paludosa). After day 8, bSi concentrations dropped to <1 µmolL-1 together with diatom cell counts below 1×106 cellsL-1. Silicoflagellate abundances were generally low from day 1 to 10 in all mesocosms (<0.006×106 cellsL-1) and were slightly increasing before DW addition on day 10 in M2 and M7 with up to 0.038×106 cellsL-1. Two different silicoflagellate species were observed: Dictyocha fibula and Dictyocha octonaria (also known as Octactis octonaria; up to 30 µm in diameter without spines, for details, see Sect. S1 in the Supplement). For a detailed list of all diatom and silicoflagellate cell counts, see Tables S1 and S2.

After the DW addition (day 12), the bSi immediately increased, ranging between 2 µmolL-1 (M1) and 8 µmolL-1 (M7), associated with higher silicoflagellate cell counts of up to 0.4×106 cellsL-1 on day 13. On day 15, bSi concentrations decreased to <0.5 µmolL-1 until day 35 and slightly increased towards the end of the experiment, ranging between 1 and 2 µmolL-1 (with a maximum of 4 µmolL-1). Overall, bSi closely correlated with diatom and silicoflagellate cell counts, except for day 20, when we observed the highest sillicoflagellate cell counts of 1.4×106 cellsL-1. However, this does not affect the calculation of the Si isotope fractionation factor as we used data from days 13 and 17.

δ30SidSi signatures showed a large range from +0.75 ‰ to +4 ‰ during the 50 d long study period. During Phase I, δ30SidSi in all three mesocosms was similar, with values generally between +2 ‰ and +2.7 ‰ (Fig. 4a). The DW was added on day 12 to the surface layer with a low δ30SidSi signature of +1.5 ‰. Despite this relatively low δ30SidSi signature of the DW, δ30SidSi in all three mesocosms evolved to values ranging between +3 ‰ and +4 ‰ on day 14 (M7) and day 17 (M1, M2). This sharp increase in δ30SidSi was caused by the rapid decline of dSi concentration (up to 50 % removal within 1 d, see also 3.2) shortly after DW addition. After day 17, δ30SidSi values constantly decreased, with the lowest δ30SidSi values observed in M1 and M2 at the end of the experiment (+0.75 ‰ to +1.2 ‰).

δ30SibSi was only analyzed in selected samples and ranged between +0.8 ‰ and +2 ‰. After the DW addition, δ30SibSi decreased to values between -0.5 ‰ and +0.6 ‰ and then constantly increased until day 24. The δ30SibSi signal was generally lower than that of δ30SidSi, with few exceptions at the end of the experiment (days 38 and day 50).

We observed a smaller offset between δ30SidSi and δ30SibSi(Δ30SibSi-dSi) within the first ten days of the experiment than during the days after DW addition. From day 1 to 10, Δ30Si ranged between -0.2 ‰ and -1.9 ‰ (Fig. S2 in the Supplement, Table 1). Directly after the DW addition, Δ30Si decreased to -3.2 ‰ (M1), -2.9 ‰ (M2), and -2.6 ‰ (M7), respectively. After day 17, Δ30Si markedly increased towards positive values (+0.1 ‰ to +0.7 ‰) at the end of the experiment (day 50).

The Si isotope fractionation factor εSilicos was calculated after DW addition according to Eq. (6) from two independent mesocosm experiments. εsilicos was -3.47 ‰ and -3.60 ‰ for M2 and M7, respectively (Table 2). Error estimates were derived from a Monte-Carlo simulation with 2000 random combinations of δ30SidSi and δ30SibSi values, taking a 2 s.d. error of 0.20 ‰ into account as well as a sensitivity analysis.

The sensitivity analysis revealed minimum and maximum εsilicos values of -3.86 ‰ (M2) and -4.05 ‰ (M7) and -3.53 ‰ (M2) and -3.37 ‰ (M7), respectively. These values yielded an error of ±0.56 ‰ (2 s.d.), which is slightly higher than the uncertainty obtained from the Monte Carlo simulation.

At the beginning of the experiment and after the addition of DW, a constant increase in bSi was observed over more than two sampling days in M2 and M7 (Fig. 3a). These data could be used to calculate the net increase in bSi per day (Fig. S3 in the Supplement). Before the addition of DW, when mainly diatoms were abundant, the bSi increase per day was 1.1 and 1.2 µmold-1 for M2 and M7, respectively. After the DW addition, the bSi increase was 0.12, 1.31 and 2.3 µmold-1 for M1, M2 and M7, respectively, corresponding to the highest abundance of silicoflagellates (97 %). Because a small number of very large diatom cells (>100 µm) could potentially influence the silicon cycle, we quantified the biovolume of the samples (biovolume normalized to cell counts). The biovolume data revealed that silicoflagellates were the dominant group after DW addition (Fig. S4 in the Supplement, Table S2). Overall, very few diatom cells ranging between 50 and 100 µm (e.g. Actinocyclus) were observed and no cells larger than 100 µm (e.g. Coscinodiscus).

A very interesting finding was a pronounced difference in the relationship between bSi content and cell counts before and after DW addition. While we observed similar bSi concentrations (median of 3.3 and 3.3 µmolL-1), the diatom-dominated phase (days 1–9) had cell counts several orders of magnitude higher (Fig. 5a), as here cells smaller than 20 µm were dominating. The bSi/cell ratio was significantly lower for the diatom (median: 1.2 pmol Si per cell; mean: 1.4 pmol Si per cell, min=0.30, max=4.6, n=9) than for the silicoflagellate-dominated samples between days 13 and 20 (median: 17 pmol Si per cell, mean: 64.65 pmol Si per cell, min=7.18, max=181.20, n=6, Fig. 5b). To minimize the bias introduced by a small number of disproportionately large cells, we highlighted samples containing 95 % silicoflagellates. Within this subset, samples composed almost exclusively of silicoflagellates exhibited a median of 15.10 pmol per cell (mean: 15.22 pmol per cell). However, the estimated values must be treated with caution given that the error is particularly high for samples with low cell counts.

In addition to the sampling within the mesocosms, we analyzed δ30SidSi of Pacific seawater outside the mesocosms over the 50 d of the experiment. There, the dSi concentrations ranged between 4.7 and 16 µmolL-1 with no apparent pattern (Fig. S5 in the Supplement). The δ30SidSi signature was highest within the first 5 d (+1.8 ‰), then remained constant near +1.6 ‰, possibly resulting from constant admixture of “new” dSi. The lowest δ30SidSi value was observed on day 50 (+1.2 ‰, Table S1, Fig. S5). According to microscopic analysis, diatoms were the dominant phytoplankton (>99 %), species throughout, except for day 50, when diatoms only accounted for 30 % of the silicifying community (Table S1).

The mesocosm study, conducted over a 50 d period with deep water replenishment on day 12, provided insights into the evolution of δ30SidSi and δ30SibSi under closed-system conditions. During the initial 10 d, diatoms were the predominant phytoplankton group, making up nearly 100 % of the siliceous phytoplankton community and up to 59 % when including non-siliceous groups such as dinoflagellates (Bach et al., 2020). The diatom community was primarily composed of small, rapidly blooming species like Skeletonema costatum (less than 20 µm in diameter). Despite high DIN/dSi ratios (approximately 1 molmol-1) during this period (Fig. 2c), both chlorophyll α (not shown) and bSi concentrations showed a decline after day 1 in M1 and after day 4 in M2 and M7 (Fig. 3a; Table S1). This reduction in productivity could be attributed to two main factors: (a) significant light attenuation within the water column due to the high biomass standing stock in the surface layer (Bach et al., 2020), and (b) the mixing with Chl α-depleted water of the bottom layer (see method part 2.1).

During Phase I, the δ30SidSi signature remained stable (+2.1 ‰ to +2.5 ‰) indicating a balance between nutrient uptake (increase in δ30SidSi) and dissolution (decrease in δ30SidSi) in surface waters. δ30SibSi (+0.4 ‰ to +1.5 ‰) was constantly lower than δ30SidSi due to the “preferential” uptake of lighter isotopes leaving the seawater enriched in heavy isotopes (e.g., De La Rocha et al., 1997). This resulted in an apparent isotope fractionation factor between silicifiers (mainly diatoms during the first 10 d) and seawater (Δ30Si) of around -1 ‰ (ε value could not be estimated during the first 10 d, see Sect. 2.5). This observation is in good agreement with laboratory studies investigating the Si isotope fractionation factor for diatoms with an overall mean εdiatom of -1.1 ‰ (e.g., Sutton et al., 2013, 2018) as well as field estimates from the region (Δ30Si of -0.4 ‰ to -1.1 ‰; e.g. Grasse et al., 2021).

Shortly before the addition of DW, diatom cell counts dropped and did not recover after the addition of DW despite high dSi concentrations (∼6 to 9 µmolL-1) in the surface water (Fig. 2a). Instead, we observed an increase in the abundance of silicoflagellates, especially in M2 and M7, associated with higher bSi concentrations, although silicoflagellate cell counts were only 0.1 % to 8 % compared to diatom cell counts in Phase I (Fig. 5a). This abrupt community shift may have been caused by a strong decrease in DIN and therefore very low DIN : Si ratios (<0.7), creating unfavorable conditions for diatoms, which typically require a 1:1 DIN ratio for optimal growth (e.g. Brzezinski, 1985; Brzezinski et al., 2008).

Silicoflagellates often co-occur with diatoms but usually play a subordinate role, especially off the Peruvian Coast (e.g., Grasse et al., 2021; Ochoa et al., 2010). Under nutrient-rich conditions, diatoms typically outcompete silicoflagellates due to their higher growth rates that can reach doubling times of less than 24 h (Werner, 1977). In contrast, silicoflagellates replicate approximately every two days (Valkenburg and Norris, 1970). The distribution and abundance of silicoflagellates in the water column may be influenced by a number of factors (e.g. water temperature, salinity, nutrient availability and grazing), though the relationships are neither simple nor consistent (e.g., Sancetta, 1990, for more details see Sect. S1). Blooms are frequently observed in organically enriched environments, suggesting mixotrophic capability (Quéguiner, 2017), which is advantageous when nutrient availability fluctuates.

In the upper ocean, dSi uptake and subsequent dissolution in subsurface waters are the main processes leading to a close correlation between δ30SidSi and dSi concentration (Fig. 6). Since diatoms are the dominant species incorporating dSi, the isotopic signature thus generally serves as a tracer of the diatom-driven silicon cycle, with values ranging from +1.7 ‰ to +3.0 ‰ in the coastal surface ocean and +1.5 ‰ in the subsurface (Ehlert et al., 2012; Grasse et al., 2021). Interestingly, samples dominated by silicoflagellates deviated from this trend showing higher δ30SidSi at moderate concentrations (5 to 9 µmolL-1) due to greater isotope fractionation. The silicoflagellate bloom was associated with δ30SidSi of up to +4.1 ‰ and low δ30SibSi (-0.26 ‰ to +0.65 ‰); Fig. 6. This may explain some exceptionally high δ30SidSi values (up to 4.4 ‰) observed in the Eastern Equatorial Pacific (Grasse et al., 2013) that deviate from the common δ30SidSi/ln⁡(dSi) relationship.

Future studies should consider that even low silicoflagellate abundances can significantly influence surface water δ30SidSi signatures. Although, their skeletons typically comprise only 1 % to 2 % of marine sediment siliceous fractions (Riedel, 1959), observational data remain insufficient to fully assess their abundance in surface waters, which can be highly patchy and seasonally variable. While silicoflagellates likely exert minimal influence on the modern ocean δ30Si budget, their role may have been more significant in the geological past (see Sect. 4.2).

Towards the end of the experiment, a slight increase in bSi and a decrease in dSi was observed as a result of diatom growth after day 35. Orni-eutrophication during the last 10 d enabled new phytoplankton growth through the relief from N-limitation in the surface layer. Interestingly, δ30SidSi only partly shows an increase expected from enhanced dSi utilization, and also decreases towards the end of the experiment with the lowest δ30SidSi values observed during the KOSMOS experiment in M1 and M2 (+1.1 ‰ and +0.9 ‰, respectively). Relatively low δ30SidSi values (minimum value of +1.2 ‰) are also observed in surface samples from the Pacific outside of the mesocosms (Fig. S4; Fig. 6: Pacific samples; comparison with all δ30SidSi data from the region). Such low values have not yet been reported for surface waters in the Peruvian Upwelling region. The lowest δ30SidSi values observed so far (+1.7 ‰, Ehlert et al., 2012) reflect recent upwelling with a low δ30SidSi signature in subsurface waters (mean +1.5 ‰, Grasse et al., 2021; Fig. 5). Such values could instead be explained by input of small lithogenic particles characterized by low δ30SidSi (-1 ‰ to -3 ‰; Sutton et al., 2018). Even lower values (-0.33 ‰) in surface waters were observed above the Northern Kerguelen Plateau, clearly influenced by lithogenic input (Cotard et al., 2025). Very small dust particles could only have been transported into the mesocosm by wind from the nearby Island of San Lorenzo. The samples from the Pacific (outside the mesocosm) may also have been influenced by river runoff and enhanced lithogenic transport. During the sampling period, the Peruvian upwelling experienced a very rare coastal El Niño which caused unusually intense precipitation in Peru and exceptionally high freshwater runoff, resulting in elevated lithogenic inputs into coastal waters (Rodríguez-Morata et al., 2019; Geilert et al., 2023), which is the most likely explanation for the observed low δ30SidSi signatures.

Our mesocosm experiment offered a unique opportunity to determine, for the first time, the Si isotope fractionation factor of silicoflagellates, an organism group that has not previously been investigated in field-based silicon isotope studies and for which laboratory culturing remains particularly challenging (Taguchi and Laws, 1985). We calculated a mean εsilicos of -3.54 ‰ ± 0.56 ‰ (mean ± 2 s.d. obtained during the sensitivity analysis). While this value represents a robust first-order estimate, it may be biased because the mesocosm surface layer was not a fully closed system. Episodic mixing, dissolution, and biomass transfer between layers could have affected the isotope mass balance and, consequently, the calculated fractionation factor. Nevertheless, for a field-based mesocosm experiment this approach provides the best available estimate, although confirmation under controlled culture conditions is required. The calculated εsilicos is higher than the Si isotope fractionation factor for diatoms that was determined during culture experiments (-1 ‰), but lower than those of sponges (-5 ‰) and choanoflagellates (-6.5 ‰) (Hendry and Robinson, 2012; Marron et al., 2019; Sutton et al., 2018, Fig. 7).

The factors controlling Si isotope fractionation remain unclear, limited understanding of the biochemical processes governing fractionation during dSi uptake and conversion to bSi. For diatoms, fractionation is thought to occur during dSi transport across the cell membrane via sodium-coupled silicate transporters (SITs, Hildebrand et al., 1997; Thamatrakoln et al., 2006). At low dSi concentrations (<30 µmolL-1), uptake is generally SIT-mediated, while at higher concentrations, dSi enters by diffusion (Shrestha and Hildebrand, 2014; Thamatrakoln and Hildebrand, 2008). SITs have been reported for most diatom species, radiolaria, choanoflagellates and Florenciella sp., a non-siliceous stramenopile belonging to the dictyochophyte lineage (“silicoflagellates”), though the transporter gene families in silicoflagellates bearing a silicon skeleton remain unclear. A second transporter family, LSi2-like, is found in choanoflagellates, sponges, and diatoms (Marron et al., 2016). The presence of different transporter families alongside different fractionation factors has led to the suggestion of a taxonomic component to Si isotope fractionation during biosilicification (Hendry et al., 2018). Sponges and choanoflagellates (“Opisthokonts”) show high Si isotope fractionation factors (up to -7 ‰; Hendry and Robinson, 2012; Marron et al., 2019), whereas stramenopiles, such as diatoms show lower values (Fig. S6 in the Supplement). However, our intermediate fractionation factor for silicoflagellates challenges this taxonomic framework, as dictyochophytes also belong to the stramenopiles.

Alternatively, Si isotope fractionation in silicoflagellates may be controlled by nutrient availability (e.g., dSi, Fe, NO3-), affecting Si uptake, growth rate, and cellular bSi content (e.g. Paasche, 1973; Takeda, 1998). Calculated bSi/cell ratios for silicoflagellates were approximately 15 pmol per cell, roughly 15 times higher than those of diatoms in this study (1.2 pmol per cell). While the the biogenic content of small, fast-blooming diatom cells is in good agreement with literature values generally ranging between approximately 0.1 and 1 depending on light and nutrient limitation (e.g., Claquin et al., 2002), no published data for the bSi content of silicoflagellates are available for comparison. The elevated bSi/cell ratio indicates more strongly silicified cells, potentially with fewer pores which might be the reason for the strong isotope fractionation. However, further research is required to clarify the processes driving the high fractionation factor.

δ30Si data obtained from siliceous organisms (e.g., diatoms, radiolaria, and sponge spicules) preserved in sediment cores have provided valuable insight into past Si cycling (e.g., Doering et al., 2016a, b, 2019, 2021; Hendry and Robinson, 2012). However, reconstructing past dSi concentrations and utilization requires knowledge of species abundances in the sediment, in addition to careful purification of the Si isotope samples. Silicoflagellates may be particularly useful for reconstructing biogeochemical conditions under which diatoms are absent and their occurrence following diatom blooms could help constrain the role of non-diatom silicifiers in the Si cycle. They likely play an important role in understanding biogeochemical conditions under which diatoms are absent. Their occurrence following diatom blooms may also help constrain the role of non-diatom silicifiers in the silicon cycle, and knowledge of their fractionation factor could improve estimates of δ30Si of dissolved silica under nutrient-limiting conditions, allowing assessment of whether dSi was fully consumed or whether nitrate limitation prevailed. Further insights can be obtained from the fractionation factor associated with dSi uptake. Despite existing studies on silicoflagellate abundances in sediments (e.g., Bukry, 1981; Amigo, 1999; McCartney et al., 2010; McCartney et al., 2022), no studies have been conducted so far on δ30Si of silicoflagellates preserved in sediments. The obtained Si isotope fractionation factor for silicoflagellates, therefore, provides the basis for a new paleo proxy for the reconstruction of the past Si cycle. Silicoflagellates experienced periods of high abundance, particularly from the Late Cretaceous to the early Paleogene. During these periods, they may have contributed significantly to bSi production, influencing local or even basin-scale Si isotope signatures. While we can only make hypothetical assumptions about their potential impact, dedicated studies would be required to properly evaluate this possibility.