Natural microbial assemblages were incubated in six 650 L polythene tanks (minicosms) housed in a temperature-controlled shipping container (Fig. 1). All minicosms were acid washed with 10 % vol : vol AR HCl, thoroughly rinsed with MilliQ water, and given a final rinse with seawater from the sampling site before use. The minicosms were filled with seawater taken amongst decomposing fast ice in Prydz Bay at Davis Station, Antarctica (68∘35′ S, 77∘58′ E) on 19 November 2014. Water was transferred by helicopter in multiple collections using a 720 L Bambi Bucket to fill a 7000 L polypropylene holding tank. Seawater was gravity fed into the minicosm tanks through Teflon-lined hosing fitted with an in-line 200 µm Arkal filter to exclude metazooplankton. All minicosms were filled simultaneously to ensure uniform distribution of microbes in all tanks.

The ambient water temperature at the time of sampling in Prydz Bay was -1.0 ∘C. Tanks were temperature controlled to an average temperature of 0.0 ∘C, with a maximum range of ±0.5 ∘C, through the cooling of the shipping container and warming with two 300 W aquarium heaters (Fluval) that were connected to a temperature control program via Carel temperature controllers. The contents of each tank were gently mixed by a shielded high-density polyethylene auger rotating at 15 rpm, and each tank was covered with a sealed acrylic lid.

Each tank was illuminated on a 19 : 5 h light : dark cycle by two 150 W HQI-TS/NDL (Osram) metal halide lamps (transmission spectra; ). The light output was filtered by a light-scattering filter and a one-quarter colour temperature (CT) blue filter (Arri) to convert the tungsten lighting to a daylight spectral distribution; attenuating wavelengths were < 500 nm by ∼ 20 % and > 550 nm by ∼ 40 % .

Similar to , the fugacity of carbon dioxide (fCO2) in each tank was raised to the target concentration in a stepwise manner over the first 5 days of the incubation (Fig. 2, see below). During this acclimation, phytoplankton growth in the tanks was slowed by attenuating the light intensity to 0.9 ± 0.2 µmolphotonsm-2s-1 using two 90 % neutral density (ND) filters (Arri).

At the conclusion of this CO2 acclimation period, the light intensity was increased for 24 h through the replacement of the two 90 % ND filters with one 60 % ND filter. The final light intensity was achieved on day 7 with a one-quarter CT blue and a light-scattering filter, which proved to be saturating for photosynthesis (see below).

Unless otherwise specified, samples were taken for analyses on days 1, 3, and 5 during the CO2 acclimation period and every 2 days from day 8 to 18.

Samples for carbonate chemistry measurements were collected daily from each minicosm in 500 mL glass-stoppered bottles (Schott Duran) following the guidelines of . Subsamples for dissolved inorganic carbon (DIC; 50 mL glass-stoppered bottles) and pH on the total scale (pHT; 100 mL glass-stoppered bottles) measurements were gently pressure filtered (0.2 µm) with a peristaltic pump at a flow rate of ∼ 30  mLmin-1, similar to .

DIC was measured by infrared absorption on an Apollo SciTech AS-C3 analyser equipped with a Li-cor LI-7000 detector using triplicate 1.5 mL samples. The instrument was calibrated (and checked for linearity) within the expected DIC concentration range with five sodium carbonate standards (Merck Suprapur) that were dried for 2 h at 230 ∘C and prepared gravimetrically in MilliQ water (18.2 MΩcm-1) at 25 ∘C. Furthermore, daily measurements of certified reference material batch CRM127 were used for improved accuracy. Volumetrically measured DIC was converted to µmolkg-1 using calculated density derived from known temperature and salinity. The typical precision among triplicate measurements was < 2 µmolkg-1.

The pHT was measured spectrophotometrically (GBC UV–vis 916) in a 10 cm thermostated (25 ∘C) cuvette using the pH indicator dye m-cresol purple (Acros Organics; 62625-31-4, lot A0321770) following the approach described in , which included changes in sample pH due to dye addition. Contact with air was minimised by sample delivery, dye addition, and mixing via a syringe pump (Tecan; Cavro XLP 6000). Dye impurities and instrument performance were accounted for by applying a constant off-set (+0.003 pH units), determined by the comparison of the measured and calculated pHT (from known DIC and total alkalinity (TA), including silicate and phosphate) of CRM127. Typical measurement precision for triplicates was 0.001 for higher and 0.003 for lower pH treatments. For further details see .

Carbonate chemistry speciation was calculated from measured DIC and pHT. In a first step at salinities measured in situ (WTW197 conductivity meter), practical alkalinity (PA) was calculated at 25 ∘C using the dissociation constants for carbonic acid determined by as refitted by . Then, total carbonate chemistry speciation was calculated from measured DIC and calculated PA for in situ temperature conditions.

The fCO2 in the minicosms was adjusted by additions of 0.22 µm filtered natural seawater that was saturated by bubbling with AR-grade CO2 for ≥ 30 min. In order to keep fCO2 as constant as possible throughout the experiment, pH in each minicosm was measured with a portable NBS-calibrated probe (Mettler Toledo) in the morning before sampling and in the afternoon to estimate the necessary amount of DIC to be added. The required volume of CO2-enriched seawater was then transferred into 1000 mL infusion bags and added to the individual minicosms at a rate of about 50 mLmin-1. After reaching target levels, the mean fCO2 levels in the minicosms were 343, 506, 634, 953, 1140, and 1641 µatm (Table S1 in the Supplement).

The average light intensity in each minicosm tank was calculated by measuring light intensity in the empty tanks at three depths (top, middle, and near-bottom) and across each tank (left, middle, and right) using a Biospherical Instruments Laboratory Quantum Scalar Irradiance Meter (QSL-101). The average light irradiance received by the phytoplankton within each tank was calculated following the equation of (Table 1). Incoming irradiance (Io‾) was calculated as the average light intensity across the top of the tank. The average vertical light attenuation (Kd) was calculated as the slope from the regression of the natural log of light intensity at all three depths, and mixed depth (Zm) was the depth of the minicosm tanks (1.14 m).

Changes in vertical light attenuation due to increases in Chl a concentration throughout the experimental period were calculated from the equation in ; Kd(biomass) = 0.0451157 × Chl a (mgm-3). Total light attenuation Kd(total) in each tank at each sampling day was calculated by addition of Kd and Kd(biomass).

No nutrients were added to the minicosms during the experiment. Macronutrient samples were obtained from each minicosm following the protocol of . Seawater was filtered through 0.45 µm Sartorius filters into 50 mL Falcon tubes and frozen at -20 ∘C for analysis in Australia. Concentrations of ammonia, nitrate plus nitrite (NOx), soluble reactive phosphorus (SRP), and molybdate reactive silica (Silica) were determined using flow injection analysis by Analytical Services Tasmania following .

Samples for POM analysis, particulate organic carbon (POC), and particulate organic nitrogen (PON) were collected following the method of . Equipment for sample preparation was soaked in Decon 90 (Decon Laboratories) for > 2 days and thoroughly rinsed in MilliQ water before use. Forceps and cutting blades were rinsed in 100 % acetone between samples. Seawater was filtered through muffled 25 mm Sartorius quartz microfibre filters until clogged. The filters were folded in half and frozen at -80 ∘C for analysis in Australia. Filters were thawed and opposite 1/8 subsamples were cut and transferred into a silver POC cup (Elemental Analysis Ltd). Inorganic carbon was removed from each sample through the addition of 20 µL of 2N HCl to each cup and drying at 60 ∘C for 36 h. When dry, each cup was folded shut, compressed into a pellet, and stored in desiccant until analysed at the Central Science Laboratory, University of Tasmania using a Thermo Finnigan EA 1112 Series Flash Elemental Analyzer.

Seawater was collected from each minicosm and a measured volume was filtered through 13 mm Whatman GF/F filters (maximum filtration time of 20 min). Filters were folded in half, blotted dry, and immediately frozen in liquid nitrogen for analysis in Australia. Chlorophyll a (Chl a) pigments were extracted, analysed by HPLC, and quantified following the methods of . Chl a was extracted from filters with 300 µL of dimethylformamide plus 50 µL of methanol, containing 140 ng apo-8′-carotenal (Fluka) internal standard, followed by bead beating and centrifugation to separate the extract from particulate matter. Extracts (125 µL) were diluted to 80 % with water and analysed on a Waters HPLC using a Waters Symmetry C8 column and a Waters 996 photodiode array detector. Chl a was identified by its retention time and absorption spectra compared to a mixed standard sample from known cultures , which was run daily before samples. Peak integrations were performed using Waters Empower software, checked manually for corrections, and quantified using the internal standard method .

Primary productivity incubations were performed following the method of based on the technique of . This method incubated phytoplankton for 1 h, minimising respiratory losses of photo-assimilated 14C so that the uptake nearly approximated gross primary productivity e.g.. Samples were analysed for total organic carbon (TO14C) content, thereby including any 14C-labelled photosynthate leaked to the dissolved organic carbon (DO14C) pool .

For all samples, 5.92 MBq (0.16 mCi) of 14C-sodium bicarbonate (NaH14CO3; PerkinElmer) was added to 162 mL of seawater from each minicosm, creating a working solution of 37 kBqmL-1. Aliquots of this working solution (7 mL) were then added to glass scintillation vials and incubated for 1 h at 21 light intensities ranging from 0–1412 µmolphotonsm-2s-1. The temperature within each of the vials was maintained at -1.0 ± 0.3 ∘C through water cooling of the incubation chamber. The reaction was terminated with the addition of 250 µL of 6N HCl and the vials were shaken for 3 h at 200 rpm to remove dissolved inorganic carbon. Duplicate time zero (T0) samples were set up in a similar manner to determine background radiation, with 250 µL of 6N HCl added immediately to quench the reaction without exposure to light. Duplicate 100 % samples were also performed to determine the activity of the working solution for each minicosm. For each 100 % sample, 100 µL of working solution was added to 7 mL 0.1 M NaOH in filtered seawater to bind all 14C. For radioactive counts, 10 mL of Ultima Gold LLT scintillation cocktail (PerkinElmer) was added to each scintillation vial, shaken, and decays per minute (DPM) were counted in a PerkinElmer Tri-Carb 2910TR Low Activity Liquid Scintillation Analyzer with a maximum counting time set at 3 min.

DPM counts were converted into primary productivity following the equation of (Table 1) using measured DIC concentrations (varying between ∼ 2075 and 2400 µmolkg-1) and normalised to Chl a using minicosm Chl a concentration (see above). Photosynthesis versus irradiance (PE) curves were modelled for each treatment following the equation of using the phytotools package in R . Photosynthetic parameter estimates included the light-saturated photosynthetic rate (Pmax⁡), maximum photosynthetic efficiency (α), photoinhibition rate (β), and saturating irradiance (Ek).

Chl a-specific primary productivity (csGPP14C) was calculated following the equation of using average minicosm light irradiance (I‾). Gross primary production rates (GPP14C) in each tank were calculated from modelled csGPP14C and Chl a concentration (see above). Calculations and units for each parameter are presented in Table 1.

Community photosynthesis and respiration rates were measured using custom-made mini-chambers. The system consisted of four 5.1 mL glass vials with oxygen sensor spots (Pyro Science) attached on the inside of the vials with non-toxic silicon glue. The vials were sealed, ensuring that any oxygen bubbles were omitted, and all vials were stirred continuously using small Teflon magnetic fleas to allow homogenous mixing of gases within the system during measurements. To improve the signal-to-noise ratio, seawater from each minicosm was concentrated above a 0.8 µm, 47 mm diameter polycarbonate membrane filter (Poretics) with gentle vacuum filtration and resuspended in seawater from each minicosm CO2 treatment. Each chamber was filled with the cell suspension and placed in a temperature-controlled incubator (0.0 ± 0.5 ∘C). Light was supplied via fluorescent bulbs above each chamber and light intensity was calibrated using a 4π sensor. Oxygen optode spots were connected to a FireSting O2 logger and data were acquired using FireSting software (Pyro Science). The optode was calibrated according to the manufacturer's protocol immediately prior to measurements using a freshly prepared sodium thiosulfate solution (10 % w / w) and agitated filtered seawater (0.2 µm) at experimental temperature for 0 and 100 % air saturation values, respectively. Oxygen concentration was recorded until a linear change in rate was established for each pseudoreplicate (n=4).

Measurements were first recorded in the light (188 µmolphotonsm-2s-1) and subsequently in the dark, with the initial steeper portion of the slope used for a linear regression analysis to determine the post-illumination (PI) respiration rate. Gross community production (GCPO2) was then calculated from dark PI respiration (RespO2) and net community production (NCPO2) rates and normalised to Chl a concentration (csGCPO2, Table 1). Chl a content for each concentrated sample was determined by extracting pigments in 90 % chilled acetone and incubating in the dark at 4 ∘C for 24 h. Chl a concentrations were determined using a spectrophotometer (Cary 50; Varian) and calculated according to the equations of , modified by .

The photosynthetic efficiency of the microalgal community was measured via Chl a fluorescence using a pulse-amplitude-modulated fluorometer (WATER-PAM; Walz). A 3 mL aliquot from each minicosm was transferred into a quartz cuvette with continuous stirring to prevent cells from settling. To establish an appropriate dark adaptation period, several replicates were measured after 5, 10, 15, 20, and 30 min of dark adaptation, with the latter having the highest maximum quantum yield of PSII (Fv / Fm). Following dark adaptation, minimum fluorescence (F0) was recorded before the application of a high-intensity saturating pulse of light (saturating pulse width = 0.8 s; saturating pulse intensity >3000 µmolphotonsm-2s-1), and maximum fluorescence (Fm) was determined. The maximum quantum yield of PSII was calculated from these two parameters . Following Fv / Fm, a five-step steady-state light curve (SSLC) was conducted with each light level (130, 307, 600, 973, 1450 µmolphotonsm-2s-1) applied for 5 min before recording the light-adapted minimum (Ft) and maximum fluorescence (Fm′) values. Each light step was spaced by a 30 s dark “recovery” period before the next light level was applied. Three pseudoreplicate measurements were conducted on each minicosm sample at 0.1 ∘C. Non-photochemical quenching (NPQ) of Chl a fluorescence was calculated from Fm and Fm′ measurements. Relative electron transport rates (rETRs) were calculated as the product of effective quantum yield (ΔF / Fm′) and actinic irradiance (Ia). Calculations and units for each parameter are presented in Table 1.

To investigate the effects of CO2 on carbon uptake, two inhibitors for carbonic anhydrase (CA) were applied to the 343 and 1641 µatm treatments on day 15: ethoxzolamide (EZA; Sigma), which inhibits both intracellular carbonic anhydrase (iCA) and extracellular carbonic anhydrase (eCA), and acetazolamide (AZA; Sigma), which blocks eCA only. Stock solutions of EZA (20 mM) and AZA (5 mM) were prepared in MilliQ water, and the pH was adjusted using NaOH to minimise pH changes when added to the samples. Before fluorometric measurements were made, water samples from the 343 and 1641 µatm CO2 treatments were filtered into ≥ 10 and <10 µm fractions and aliquots were inoculated either with 50 µL of MilliQ water adjusted with NaOH (control) or a 50 µM final concentration of chemical inhibitor (EZA and AZA). Fluorescence measurements of size-fractionated control- and inhibitor-exposed cells were performed using the WATER-PAM. A 3 mL aliquot of sample was transferred into a quartz cuvette with stirring and left in the dark for 30 min before the maximum quantum yield of PSII (Fv / Fm) was determined (as described above). Actinic light was then applied at 1450 µmolphotonsm-2s-1 for 5 min before the effective quantum yield of PSII (ΔF / Fm′) was recorded. Three pseudoreplicate measurements were conducted on each minicosm sample at 0.1 ∘C.

Bacterial abundance was determined daily using a Becton Dickinson FACScan or FACSCalibur flow cytometer fitted with a 488 nm laser following the protocol of . Samples were pre-filtered through a 50 µm mesh (Nitex), stored at 4 ∘C in the dark, and analysed within 6 h of collection. Samples were stained for 20 min with 1:10000 dilution SYBR Green I (Invitrogen) , and PeakFlow Green 2.5 µm beads (Invitrogen) were added to the sample as an internal fluorescence standard. Three pseudoreplicate samples were prepared from each minicosm seawater sample. Samples were run for 3 min at a low flow rate (∼ 12 µLmin-1) and bacterial abundance was determined from side scatter (SSC) versus green (FL1) fluorescence bivariate scatter plots. The analysed volume was calibrated to the sample run time and each sample was run for precisely 3 min, resulting in an analysed volume of 0.0491 and 0.02604 mL on the FACSCalibur and FACScan, respectively. The volume analysed was then used to calculate final cell concentrations.

Bacterial productivity measurements were performed following the leucine incorporation by microcentrifuge method of . Briefly, 70 nM 14C-leucine (PerkinElmer) was added to 1.7 mL of seawater from each minicosm in 2 mL polyethylene Eppendorf tubes and incubated for 2 h in the dark at 4 ∘C. Three pseudoreplicate samples were prepared from each minicosm seawater sample. The reaction was terminated by the addition of 90 µL of 100 % trichloroacetic acid (TCA; Sigma) to each tube. Duplicate background controls were also performed following the same method, with 100 % TCA added immediately before incubation. After incubation, samples were spun for 15 min at 12 500 rpm and the supernatant was removed. The cell pellet was resuspended into 1.7 mL of ice-cold 5 % TCA and spun again for 15 min at 12 500 rpm and the supernatant was removed. The cell pellet was then resuspended into 1.7 mL of ice-cold 80 % ethanol, spun for a further 15 min at 12 500 rpm, and the supernatant was removed. The cell pellet was allowed to dry completely before addition of 1 mL of Ultima Gold scintillation cocktail (PerkinElmer). The Eppendorf tubes were placed into glass scintillation vials and DPMs were counted in a PerkinElmer Tri-Carb 2910TR Low Activity Liquid Scintillation Analyzer with a maximum counting time of 3 min.

DPM counts were converted to 14C-leucine incorporation rates following the equation in and used to calculate gross bacterial production (GBP14C) following . Bacterial production was divided by total bacterial abundance to determine the cell-specific bacterial productivity within each treatment (csBP14C). Calculations and units for each parameter are presented in Table 1.

The minicosm experimental design measured the microbial community growth in six unreplicated fCO2 treatments. Therefore, subsamples from each minicosm were within-treatment pseudoreplicates and thus only provide a measure of the variability of the within-treatment sampling and measurement procedures. We use pseudoreplicates as true replicates in order to provide an informal assessment of differences among treatments, noting that results must be treated as indicative and interpreted conservatively.

For all analyses, a linear or curved (quadratic) regression model was fitted to each CO2 treatment over time using the stats package in R , and an omnibus test of differences between the trends among CO2 treatments over time was assessed by ANOVA. This analysis ignored the repeated measures nature of the data set, which could not be modelled due to the low number of time points and an absence of replication at each time. For the CCM activity measurements, differences between treatments were tested by one-way ANOVA followed by a post-hoc Tukey's test to determine which treatments differed. The significance level for all tests was set at < 0.05.

The fCO2 of each treatment was modified in a stepwise fashion over 5 days to allow for acclimation of the microbial community to the changed conditions. Target treatment conditions were reached in all tanks by day 5 and ranged from 343 to 1641 µatm, equating to an average pHT of 8.10 to 7.45 (Fig. 2, Table S1), respectively. The initial seawater was calculated to have an fCO2 of 356 µatm and a PA of 2317 µmolkg-1, from a measured pHT of 8.08 and DIC of 2187 µmolkg-1 (Fig. S1 and Table S2 in the Supplement). One minicosm was maintained close to these conditions (343 µatm) throughout the experiment as a control treatment.

The average light irradiance for all CO2 treatments is presented in Table S3. During the CO2 acclimation period (days 1–5) the average light irradiance was 0.9 ± 0.2 µmolphotonsm-2s-1 and was increased to 90.5 ± 21.5 µmolphotonsm-2s-1 by day 8. The average vertical light attenuation (Kd) across all minicosm tanks was 0.92 ± 0.2. Increasing Chl a concentration over time in all CO2 treatments increased Kd(total) from 0.96 ± 0.01 on day 1 to 3.53 ± 0.28 on day 18, resulting in a decline in average light irradiance within the minicosms from 86.61 ± 20.5 to 35.97 ± 9.3 µmolphotonsm-2s-1 between days 8 and 18.

Nutrient concentrations were similar across all treatments at the beginning of the experiment (Table S2 in the Supplement) and did not change during the acclimation period (days 1–5). Ammonia concentrations were initially low (0.95 ± 0.18 µM) and fell rapidly to concentrations below the limits of detection beyond day 12 in all treatments (Fig. S2 in the Supplement). No differences in drawdown between CO2 treatments were observed, and thus it was excluded from further analysis. NOx fell from 26.2 ± 0.74 µM on day 8 to concentrations below detection limits on day 18 (Fig. 3a), with the slowest drawdown in the 1641 µatm treatment. SRP concentrations were initially 1.74 ± 0.02 µM and all CO2 treatments followed a similar drawdown sequence to NOx, reaching very low concentrations (0.13 ± 0.03 µM) on day 18 in all treatments (Fig. 3b). In contrast, silica was replete in all treatments throughout the experiment falling from 60.0 ± 0.91 µM to 43.6 ± 2.45 µM (Fig. 3c). The drawdown of silica was exponential from day 8 onwards and followed a similar pattern to NOx and SRP, with the highest silica drawdown in the 634 µatm and the least in the 1641 µatm treatment.

Particulate organic carbon (POC) and nitrogen (PON) concentrations were initially low at 4.7 ± 0.15 and 0.5 ± 0.98 µM, respectively, and increased after day 8 in all treatments (Fig. 4a, b). The accumulation of POC and PON was effectively the reciprocal of the drawdown of nutrients (see above), being lowest in the high CO2 treatments (≥ 1140 µatm) and highest in the 343 and 643 µatm treatments. Rates of POC and PON accumulation were both affected by nutrient exhaustion, with declines in the 343 and 634 µatm treatments between days 16 and 18. POC and PON concentrations on day 18 were highest in the 953 µatm treatment. The ratio of POC to PON (C : N) was similar for all treatments, declining from 8.0 ± 0.38 on day 8 to 5.7 ± 0.28 on day 16 (Fig. 4c). The slowest initial decline in the C : N ratio occurred in the 1641 µatm treatment, displaying a prolonged lag until day 10, after which it decreased to values similar to all other treatments. Nutrient exhaustion on day 18 coincided with an increase in the C : N ratio in all treatments, with C : N ratios >10 in the 343, 634, and 953 µatm treatments and lower C : N ratios (8.6–6.7) in the 506, 1140, and 1641 µatm treatments.

Chl a concentrations were low at the beginning of the experiment at 0.91 ± 0.16 µgL-1 and increased in all treatments after day 8 (Fig. 5a). Chl a accumulation rates were similar amongst treatments ≤ 634 µatm until day 14, with a slightly higher Chl a concentration in the 506 and 634 µatm treatments on day 16 compared to the control treatment. By day 18, only the 506 µatm treatment remained higher than the control. Chl a accumulation rates in the 953 and 1140 µatm treatments were initially slow but increased after day 14, with Chl a concentrations similar to the control on days 16–18. The highest CO2 treatment (1641 µatm) had the slowest rates of Chl a accumulation, displaying a lag in growth between days 8 and 12, after which the Chl a concentration increased but remained lower than the control. Rates of Chl a accumulation slowed between days 16 and 18 in all treatments except 1641 µatm, coinciding with nutrient limitation. At day 18, the highest Chl a concentration was in the 506 µatm exposed treatment and lowest at 1641 µatm.

The omnibus test among CO2 treatments of trends in Chl a over time indicated that the accumulation of Chl a in at least one treatment differed significantly from that of the control (F5,23=5.5, p=0.002; Table S4). Examination of individual coefficients from the model revealed that only the highest CO2 treatment, 1641 µatm, was significantly different from the control at the 5 % level.

During the CO2 and light acclimation phase of the experiment (days 1–8), all treatments displayed a steady decline in the maximum photosynthetic rate (Pmax⁡) and the maximum photosynthetic efficiency (α) until the levels on day 8 were approximately half of those at the beginning of the experiment, suggesting cellular acclimation to the light conditions (Fig. S3a, b in the Supplement). Thereafter, relative to the control, Pmax⁡ and α were lowest in CO2 levels ≥ 953 µatm and ≥ 634 µatm, respectively. Rates of photoinhibition (β) and saturating irradiance (Ek) were variable and did not differ among treatments (Fig. S3c, d). The average Ek across all treatments was 28.7 ± 8.6 µmolphotonsm-2s-1, indicating that the light intensity in the minicosms was saturating for photosynthesis (see above) and not inhibiting (β < 0.002 mgC(mgChla)-1(µmolphotonsm-2s-1)-1h-1).

Chl a-specific primary productivity (csGPP14C) and gross primary production (GPP14C) were low during the CO2 acclimation (days 1–5) and increased with increasing light climate after day 5. Rates of csGPP14C in treatments ≥ 634 µatm CO2 were consistently lower than the control between days 8 and 16, with the lowest rates in the highest CO2 treatment (1641 µatm; Fig. 6a). Rates of GPP14C in treatments ≤ 953 were similar between days 8 and 16, with the 343 (control), 506, and 953 µatm treatments increasing to 46.7 ± 0.34 µgCL-1h-1 by day 18 (Fig. 5b). Compared to these treatments, GPP14C in the 634 µatm treatment was lower on day 18, only reaching 39.7 µgCL-1h-1, possibly due to the concurrent limitation of NOx in this treatment on day 16 (see above).

The omnibus test among tanks of the trends in CO2 treatments over time indicated that GPP14C in at least one treatment differed significantly from the control (F5,23=4.9, p=0.003; Table S5 in the Supplement). Examination of the significance of individual curve terms revealed that this manifested as differences between the 1140 and 1641 µatm treatments and the control group at the 5 % level. No other curves were different from the control. In particular, GPP14C in the 1641 µatm treatment was much lower until day 12, after which it increased steadily until day 16. Between days 16 and 18, a substantial increase in GPP14C was observed in this treatment, subsequently resulting in a rate on day 18 that was similar to the 1140 µatm treatment (36.3 ± 0.08 µgCL-1h-1) although these treatments never reached rates of GPP14C as high as the control.

The productivity of the phytoplankton community increased over time in all CO2 treatments; however, there were clear differences in the timing and magnitude of this increase between treatments (Fig. 6b). A CO2 effect was evident on day 12 when Chl a-normalised gross O2 productivity rates (csGCPO2) increased with increasing CO2 level, ranging from 19.5–248 mgO2(mgChla)-1h-1. After day 12, the communities in CO2 treatments ≤ 634 µatm continued to increase their rates of csGCPO2 until day 18 (97.7 ± 17.0 mgO2(mgChla)-1h-1). The 953 and 1140 µatm CO2 treatments peaked on day 12 (90.4 and 126 mgO2(mgChla)-1h-1, respectively) and then declined on day 14 to rates similar to the control treatment. In contrast, the 1641 µatm treatment maintained high rates of csGCPO2 from days 12–14 (258 ± 13.8 mgO2(mgChla)-1h-1), coinciding with the recovery of photosynthetic health (Fv / Fm; see below) and the initiation of growth in this treatment (see above). After this time, rates of csGCPO2 declined in this treatment to rates similar to the control. Despite these differences in csGCPO2, there was no significant difference in the gross community production (GCPO2) among CO2 treatments (Fig. 5c).

The community maximum quantum yield of PSII (Fv / Fm) showed a dynamic response over the duration of the experiment (Fig. 7). Values initially increased during the low-light CO2 adjustment period but declined by day 8 when irradiance levels had increased. Between days 8 and 14, differences were evident in the photosynthetic health of the phytoplankton community across the CO2 treatments, although by day 16 these differences had disappeared. Steady-state light curves revealed that the community photosynthetic response did not change with increasing CO2. The effective quantum yield of PSII (ΔF / Fm′) and NPQ showed no variability with CO2 treatment (Figs. S5 and S6 in the Supplement). There was, however, a notable decline in overall NPQ in all tanks with time, indicating an adjustment to the higher light conditions. Relative electron transport rates (rETR) showed differentiation with respect to CO2 at high light (1450 µmolphotonsm-2s-1) on days 10–12. However, as seen with the Fv / Fm response, this difference was diminished by day 18 (Fig. S7 in the Supplement).

There was a significant decline in the effective quantum yield of PSII (ΔF / Fm′) with the addition of the iCA and eCA inhibitor EZA to both the large (≥ 10 µm, p=0.02) and small (<10 µm, p < 0.001) size fractions of the phytoplankton community exposed to the control (343 µatm) CO2 treatment (Fig. 8). The addition of EZA to cells under high CO2 (1641  µatm) had no effect on ΔF / Fm′ for either size fraction. However, in the case of the small cells under high CO2 (Fig. 8b), ΔF / Fm′ was the same as that measured in the control CO2 in the presence of EZA. The addition of AZA, which inhibits eCA only, had no effect for either CO2 treatment in the large-celled community. In contrast, there was a significant decline in ΔF / Fm′ in the smaller fraction in the control CO2 treatment (p < 0.001), but no effect of AZA addition under high CO2. Again, the high CO2 cells exhibited the same ΔF / Fm′ as those measured under the control CO2 in the presence of AZA.

During the 8-day acclimation period, bacterial abundance in treatments ≥ 634 µatm increased with increasing CO2, reaching 26.0–32.4 × 107 cellsL-1 and remaining high until day 13 (Fig. 9a). Between days 7 and 13, bacterial abundances in CO2 treatments ≥ 953 were higher than the control. In contrast, abundance remained constant in treatments ≤ 506 µatm (20.6 ± 1.4 × 107 cellsL-1) until day 11. Cell numbers rapidly declined in all treatments after day 12, finally stabilising at 0.5 ± 0.2 × 107 cellsL-1. An omnibus test among CO2 treatments of the trends in bacterial abundance over time showed that changes in abundance in at least one treatment differed significantly from the control (F5,185=9.8, p < 0.001; Table S6 in the Supplement). Examination of individual coefficients from the model revealed that CO2 treatments ≥ 953 µatm were significantly different from the control at the 5 % level.

Gross bacterial production (GBP14C) was low in all CO2 treatments (0.2 ± 0.03 µgCL-1h-1) and changed little during the first 5 days of incubation (Fig. 9b). Thereafter it increased, coinciding with exponential growth in the phytoplankton community. The most rapid increase in GBP14C was observed in the 634 µatm treatment, resulting in a rate twice that of all other treatments by day 18 (2.1 µgCL-1h-1). No difference was observed among other treatments, all of which increased to an average rate of 1.1 ± 0.1 µgCL-1h-1 by day 18. Cell-specific bacterial productivity (csBP14C) was low in all treatments (1.2 ± 0.5 fgCL-1h-1) until day 14, with slower rates in treatments ≥ 953 µatm, likely due to high cell abundances observed in these treatments (Fig. S8 in the Supplement). It then increased from day 14, coinciding with a decline in bacterial abundance. Rates of csBP14C did not differ among treatments until day 18, when the rate in the 634 µatm treatment was higher than all other treatments (0.5 pgCcell-1L-1h-1).

The results of this study suggest that exposing phytoplankton to high CO2 levels can decouple the two stages of photosynthesis (see also the discussion below). At CO2 levels ≥ 1140 µatm, Chl a-specific oxygen production (csGCPO2) increased strongly yet displayed the lowest rates of Chl a-specific carbon fixation (csGPP14C; Fig. 6). This mismatch in oxygen production and carbon fixation is likely due to the two-stage process in the photosynthetic fixation of carbon reviewed in. In the first stage, light-dependent reactions occur within the chloroplast, converting light energy (photons) into the cellular energy products, adenosine triphosphate (ATP), and nicotinamide adenine dinucleotide phosphate (NADPH), producing O2 as a by-product. This cellular energy is then utilised in a second, light-independent pathway, which uses the carbon-fixing enzyme RuBisCO to convert CO2 into sugars through the Calvin cycle. However, under certain circumstances the relative pool of energy may also be consumed in alternative pathways, such as respiration and photoprotection . Increases in energy requirements for these alternate pathways have been demonstrated, where measurements of maximum photosynthetic rates (Pmax⁡) and photosynthetic efficiency (α) display changes that result in no change to saturating irradiance levels (Ek) . This “Ek-independent variability” was evident in our study, in which decreases in Pmax⁡ and α were observed in the high CO2 treatments, while Ek remained unaffected (Fig. S3 in the Supplement).

This highlights an important tipping point in the phytoplankton community's ability to cope with the energetic requirements of maintaining efficient productivity under high CO2. While studies on individual phytoplankton species have reported decoupling of the photosynthetic pathway under conditions of stress, no studies to date on natural phytoplankton communities have reported this response. Under laboratory conditions, stresses such as nutrient limitations or a combination of high CO2 and light climate have been shown to induce such a response in which isolated phytoplankton species possess higher energy requirements for carbon fixation. In our study, the phytoplankton community experienced a dynamic light climate due to continuous gentle mixing of the minicosm contents, and although nutrients were not limiting, the phytoplankton in the higher CO2 treatments did show lower csGPP14C rates (Fig. 6a), which could be linked to higher energy demand for light-independent processes. Since nutrients were replete and not a likely source of stress, it follows that CO2 and light were likely the only sources of stress on this community.

Increased respiration rates could account for the decreased carbon fixation rates measured. Thus far, respiration rates are commonly reported as either unaffected or lower under increasing CO2 . This effect is generally attributed to declines in cellular energy requirements via processes such as the down-regulation of CCMs, which can result in observed increased rates of production . Despite this, decreased growth rates have been linked to enhanced respiratory carbon loss at high CO2 levels (800–1000 µatm) . The contribution of community respiration rates to csGCPO2 was high and increased with increasing CO2 (Fig. S4 in the Supplement). However, respiration rates were generally proportional to the increase in O2 production (i.e. the ratio of production to respiration remained constant across CO2 conditions), making it unlikely to be a significant contributor to the decline in carbon fixation. Instead, high respiration rates were possibly a result of heterotrophic activity.

It has been suggested that the negative effects of ocean acidification are predominantly due to the decline in pH and not the increase in CO2 concentration e.g.. A decline in pH with ocean acidification increases the hydrogen ion (H+) concentration in the seawater and is likely to make it increasingly difficult for phytoplankton cells to maintain cellular homeostasis. Metabolic processes, such as photosynthesis and respiration, impact cellular H+ fluxes between compartments, making it necessary to temporarily balance internal H+ concentrations through H+ channels . Under normal oceanic conditions (pH ∼ 8.1), when the extracellular environment is above pH 7.8, excess H+ ions generated within the cell are able to passively diffuse out of the cell through these H+ channels. However, a lowering of the oceanic pH below 7.8 is likely to halt this passive removal of internal H+, requiring the utilisation of energy-intensive proton pumps and thus potentially reducing the energy pool available for carbon fixation. While not well understood, these H+ channels may also perform important cellular functions, such as nutrient uptake, cellular signalling, and defense . Our results are consistent with this idea of a critical pH threshold, as significant declines in GPP14C were observed in treatments ≥ 1140 µatm (Fig. 10), which are the CO2 treatments for which the pH ranged from 7.69–7.45 (Fig. 2).

Despite the initial stress of high CO2 between days 8 and 12, the phytoplankton community displayed a strong ability to adapt to these conditions. The CO2-induced reduction in Fv / Fm showed a steady recovery between days 12 and 16, with all treatments displaying similarly high Fv / Fm at day 16 (0.68–0.71; Fig. 7). This recovery in photosynthetic health suggests that the phytoplankton community was able to acclimate to the high CO2 conditions, possibly through cellular acclimation, changes in community structure, or most likely, a combination of both. Cellular acclimations were observed in our study. A lowering of NPQ and a minimisation of the CO2-related response to photoinhibition (rETR) at high light intensity suggested that PSII was being down-regulated to adjust to a higher light climate (Figs. S6 and S7 in the Supplement). Decreased energy requirements for carbon fixation were also observed in the photosynthetic pathway, resulting in increases in GPP14C and Chl a accumulation rates (Fig. 5). Acclimation to increased CO2 has been reported in a number of studies, resulting in shifts in carbon and energy utilisation . Numerous photophysiological investigations on individual phytoplankton species also report species-specific tolerances to increased CO2 , and a general trend toward smaller-celled communities with increased CO2 has been reported in ocean acidification studies globally . Changes in community structure were observed with increasing CO2, with taxon-specific thresholds of CO2 tolerance . Within the diatom community, the response was also related to size, leading to an increase in abundance of small (< 20 µm) diatoms in the higher CO2 treatments (≥ 953 µatm). Therefore, the community acclimation observed is likely driven by an increase in the growth of more tolerant species.

It is often suggested that the down-regulation of CCMs helps to moderate the sensitivity of phytoplankton communities to increasing CO2. The carbon-fixing enzyme RuBisCO has a low affinity for CO2 that is compensated for through CCMs that actively increase the intracellular CO2 . This process requires additional cellular energy and numerous studies have suggested that the energy savings from down-regulation of CCMs in phytoplankton could explain increases in rates of primary productivity at elevated CO2 levels e.g.. In Antarctic phytoplankton communities, showed that the energetic costs of CCMs are low and any down-regulation at increased CO2 would provide little benefit. We found that the CCM component carbonic anhydrase (CA) was utilised by the phytoplankton community at our control CO2 level (343 µatm) and was down-regulated at high CO2 (1641 µatm; Fig. 8). Yet we saw no promotion of primary productivity that coincided with this down-regulation. Thus, our data support the previous studies showing that increased CO2 may alleviate energy supply constraints but does not necessarily lead to increased rates of carbon fixation .

Furthermore, size-specific differences in phytoplankton CCM utilisation were observed. The absence of eCA activity in the large phytoplankton (≥ 10 µm; Fig. 8a) suggests that bicarbonate (HCO3-) was the dominant carbon source used by this fraction of the phytoplankton community . This is not surprising as direct HCO3- uptake has been commonly reported among Antarctic phytoplankton communities . On the other hand, the small phytoplankton (<10 µm; Fig.  8b) seem to have used both iCA and eCA, implying that carbon for photosynthesis was sourced through both the extracellular conversion of HCO3- to CO2 and direct HCO3- uptake . Despite these patterns, CCM activity in this study was only determined via Chl a fluorescence and therefore direct measurement of light-dependent reactions in photosynthesis. This imposes limitations to the interpretability of this particular data set, as CA is involved primarily in carbon acquisition, which occurs during photosynthetic reactions that are independent of light.

The presence of iCA has also been proposed as a possible mechanism for increased sensitivity of phytoplankton to decreased pH conditions. found that the presence of iCA caused strong intracellular acidification and inhibition of carbon fixation when a CO2-tolerant iCA-expressing algal species was transferred from ambient conditions to very high CO2 (40 %). Down-regulation of iCA through acclimation in a 5 % CO2 treatment eliminated this response, with similar tolerance observed in an algal species with low ambient iCA activity. Thus, the down-regulation of iCA activity at high CO2, as was seen in our study, may not only decrease cellular energy demands but may also be operating as a cellular protection mechanism, allowing the cell to maintain intracellular homeostasis.

Contrary to the high CO2 treatments, the phytoplankton community appeared to tolerate CO2 levels up to 953 µatm, which identified a CO2 threshold. Between days 8 and 14 we observed a small and insignificant CO2-related decline in Fv / Fm, GPP14C, and Chl a accumulation among the 343–953 µatm treatments (Figs. 7 and 10). Tolerance of CO2 levels up to ∼ 1000 µatm has often been observed in natural phytoplankton communities in regions exposed to fluctuating CO2 levels. In these communities, increasing CO2 often had no effect on primary productivity or growth , although an increase in primary production has been observed in some instances . These differing responses may be due to differences in community composition, nutrient supply, or ecological adaptations of the phytoplankton community in the region studied. They may also be due to differences in the experimental methods, especially the range of CO2 concentrations employed , the mechanism used to manipulate CO2 concentrations, the duration of the acclimation and incubation, the nature and volume of the mesocosms used, and the extent to which higher trophic levels are screened from the mesocosm contents see.

Previous studies in Prydz Bay report a tolerance of the phytoplankton community to CO2 levels up to 750 µatm . Although these experiments differed in nutrient concentration, community composition, and CO2 manipulation from ours, when taken together, these studies demonstrate consistent CO2 effects throughout the Antarctic summer season and across years in this location. The most likely reason for this high tolerance is that these communities are already exposed to highly variable CO2 conditions. CO2 naturally builds beneath the sea ice in winter when primary productivity is low , and is rapidly depleted during spring and summer by phytoplankton blooms, resulting in annual fCO2 fluctuations between ∼ 50 and 500 µatm . Thus, variable CO2 environments appear to promote adaptations within the phytoplankton community to manage the stress imposed by fluctuating CO2.

Changes in POM production and the C : N ratio in phytoplankton communities can have significant effects on carbon sequestration and change their nutritional value for higher trophic levels . We observed a decline in particulate organic matter production (POM) at CO2 levels ≥ 1140 µatm (Fig. 10), while changes in organic matter stoichiometry (C : N ratio) appeared to be predominantly controlled by nutrient consumption (Fig. 4). Increases in POM production were similar to Chl a accumulation, with declines in high CO2 treatments (≥ 1140µatm) due to low rates of primary productivity. Carbon overconsumption has been reported in some natural phytoplankton communities exposed to increased CO2, resulting in observed or inferred increases in the particulate C : N ratio . While in our study the C : N ratio did decline to below the Redfield ratio during exponential growth, it remained within previously reported C : N ratios of coastal phytoplankton communities in this region . However, as we did not analyse the elemental composition of dissolved inorganic matter, carbon overconsumption cannot be completely ruled out . Therefore, it is difficult to say whether or not changes in primary productivity will affect organic matter stoichiometry in this region, particularly as any resultant long-term changes in community composition to more CO2-tolerant taxa may also have an effect .

In contrast to the phytoplankton community, bacteria were tolerant of high CO2 levels. The low bacterial productivity and abundance of the initial community is characteristic of the post-winter bacterial community in Prydz Bay where growth is limited by organic nutrient availability . Whilst an increase in cell abundance was observed at CO2 levels ≥ 634 µatm (Fig. 9a), it was possible that this response was driven by a decline in grazing by heterotrophs instead of a direct CO2-related promotion of bacterial growth. The subsequent decline in abundance was likely due to top-down control from the heterotrophic nanoflagellate community, which displayed an increase in abundance at this time . Bacterial tolerance to high CO2 has been reported previously in this region and has also been reported in numerous studies in the Arctic , suggesting that the marine bacterial community will be resilient to increasing CO2.

While we detected an increase in bacterial productivity, this response appeared to be correlated with an increase in Chl a concentration and available POM rather than CO2. Bacterial productivity was similar among all CO2 treatments, except for a final promotion of productivity at 634 µatm on day 18 (Fig. 9b). This promotion of growth may be linked to an increase in diatom abundance observed in this treatment . The coupling of bacterial growth with phytoplankton productivity has been reported by numerous studies on natural marine microbial communities . Thus, it is likely that the bacterial community was controlled more by grazing and nutrient availability than by CO2 level.