Effects of elevated CO2 on phytoplankton during a mesocosm experiment in the southern eutrophicated coastal water of China

There is a growing consensus that the ongoing increase in atmospheric CO2 level will lead to a variety of effects on marine phytoplankton and ecosystems. However, the effects of CO2 enrichment on eutrophic coastal waters are still unclear, as are the complex mechanisms coupled to the development of eutrophication. Here, we report the first mesocosm CO2 perturbation study in a eutrophic subtropical bay during summer by investigating the effect of rising CO2 on a model artificial community consisting of well-characterized cultured diatoms (Phaeodactylum tricornutum and Thalassiosira weissflogii) and prymnesiophytes (Emiliania huxleyi and Gephyrocapsa oceanica). These species were inoculated into triplicate 4 m3 enclosures with equivalent chlorophyll a (Chl-a) under present and higher partial pressures of atmospheric CO2 (pCO2 = 400 and 1000 ppmv). Diatom bloom events were observed in all enclosures, with enhanced organic carbon production and Chl-a concentrations under high CO2 treatments. Relative to the low CO2 treatments, the consumption of the dissolved inorganic nitrogen and uptake ratios of N/P and N/Si increased significantly during the bloom. These observed responses suggest more extensive and complex effects of higher CO2 concentrations on phytoplankton communities in coastal eutrophic environments.

At present, one of the most far-reaching global perturbations of the marine environment is caused by the massive invasion of fossil fuel CO 2 into the ocean, making it the second largest sink for anthropogenic carbon dioxide after the atmosphere itself 1 . CO 2 dissolved in seawater forms free H + ions, lowering ocean pH and shifting dissolved inorganic carbon away from carbonate (CO 3 2− ) towards more bicarbonate (HCO 3 − ) and CO 2 . This global effect of anthropogenic CO 2 emissions on ocean carbonate chemistry is of concern because it is already lowering the pH of the oceans, which may have ramifications for the growth, productivity and dominance of individual organisms or whole marine ecosystems 2 .
It has been suggested that the consequences of global ocean acidification will become more acute in the coastal zone, due to the decomposition of organic matter produced in eutrophic waters 3 . Coastal areas are complex and dynamic places in which environmental factors typically exhibit great spatial and temporal variability. For example, CO 2 partial pressures (pCO 2 ) in the inner estuary of the highly eutrophic Pearl River was found to range from 3380 to 4785 µatm in the summer, with a pH of 7.0-7.2 4 . Meanwhile, in a concurrent bloom in the outer estuary of the Pearl River pCO 2 dropped rapidly to ~200 µatm, and pH rose to as high as 8.6 5 . This is mainly due to a variety of biogeochemical processes in coastal water, not from changes of CO 2 in atmospheric concentrations. Terrigenous inputs, upwelling effects and biological activities (algal blooms, bacterial respiration) play important roles on the variations of pCO 2 and pH in the water. The coastal acidification has been predicted to be over 10% Scientific RepoRts | 7: 6868 | DOI: 10.1038/s41598-017-07195-8 p < 0.05). The mean pCO 2 levels in the HC and LC treatments were 879.15 ± 145.76 and 658.05 ± 113.98 ppmv, with a significant differences of 221.10 ppmv in the first 4 days (Fig. 2b).
Along with the growth of phytoplankton and consequent depletion of nutrients, pH value increased in all enclosures (Fig. 1c). The pCO 2 values dropped rapidly to ~200 µatm, and pH rose to over 8.5, indicating the impact of biological activity was first order. During the growth of phytoplankton, they were aerated continuously by different CO 2 enrichments. Although lower pH values and higher pCO 2 in the early phase were observed, quite high pH values (8.90 ± 0.06) and low pCO 2 (31 ± 2 µatm) in the HC enclosures were observed on day 10 ( Fig. 1).
Temporal evolution of nutrients. The initial filtered bay water used for the experiment was highly eutrophic, with nutrient concentrations in the initial waters of 30.48 ± 0.29, 53.32 ± 1. 10  , respectively (Figs 3 and 4). The composition of the dissolved inorganic nitrogen pool (DIN, NO x + NH 4 + ) indicates a very high proportion of NH 4 + (Fig. 3). DIN concentrations dropped sharply after day 3 due to phytoplankton uptake, and the final drawdown in the HC treatment was significantly larger than that of the LC treatment (Fig. 3a, p < 0.01). Early in the nitrogen uptake process the preferential uptake of NH 4 + is obvious, as the decrease of NO x started only when NH 4 + reached a  ) and pH (c) in 6 enclosures perturbed by bubbling with ambient air (400 ppmv CO 2 , Low CO 2 level) or an air/CO 2 mixture at a concentration of 1000 ppmv CO 2 (High CO 2 level) over a two week incubation. Symbols are the means and error bars are the standard errors of three replicate enclosures. minimum on day 7 (Fig. 3b). The DIN drawdown difference between treatments was due to the fact that the NO x was not exhausted under the LC scenario (Fig. 3c). The downward trends of SRP (soluble reactive phosphate) and SiO 3 2− concentrations were consistent with that of ammonium, and the SRP concentrations were consumed completely first in LC treatments by day 6 (Fig. 4). Despite the eutrophic status of the collected experimental water, it was also potentially P-limited, as the SRP was first exhausted by day 6 and the initial DIN/SRP ratio was 24.5. This ratio exhibited a sustained increase along with the decline of inorganic nutrient concentrations in the seawater (Fig. 4c). Since more NOx was used in the HC enclosures, the DIN/SRP ratios in HC treatments after day 6 were significantly lower than those of the LC enclosures. Similar results were observed for variations in the DIN/SiO 3 2− ratios (Fig. 4d). Ratios of net nutrient uptake (the difference with initial concentrations) also indicated treatment-dependent differences with significantly higher ratios of ΔDIN/ΔSRP and ΔDIN/ΔSiO 3 2− in HC enclosures during the nutrient replete period (Fig. 5, day 4-7, n = 4, paired t-Test, p < 0.01). This suggests more consumption of DIN in the HC enclosures relative to SRP and SiO 3 2− . Variations on phytoplankton communities and carbon metabolism. During the experiment, algal blooms were induced artificially in the enclosures (Fig. 6a). As there were no zooplankton grazers present, phytoplankton responses to the high CO 2 treatments were strictly driven by "bottom-up" influences. Although the initial total chlorophyll a (Chl-a) concentrations of the four phytoplankton species were the same and the initial cell number of the prymnesiophytes was higher than that of the diatoms, the diatoms had an obvious advantage in the highly eutrophic water (Fig. 6). This is evident from the observation that the consumption of SiO 3 2− was coupled to that of the other nutrients (Figs 3 and 4), as well as the increase in the diagnostic diatom pigment (Fucoxanthin, Fuco) (Fig. 6b).
Phytoplankton Chl-a biomass reached a significantly higher concentration in the HC treatment on day 9 (185 ± 59 µg L −1 ), although its response seemed to show a short time lag (Fig. 6a). Prymnesiophytes rapidly responded in the HC enclosures as indicated by their diagnostic pigment 19′hexanoyl-oxy-fucoxanthin (19-Hex), reaching a peak on day 6 (1.4 ± 0.5 µg L −1 , Fig. 6c). The temporal variations in the ratio of 19-Hex to Chl-a confirm that the contribution of prymnesiophytes to total Chl-a biomass was higher in the HC enclosures in the early phase (day 2-7), and the opposite was observed during days 8-14 (Fig. 7b). The ratios of Fuco to Chl-a exhibited no clear differences between the different treatments, both with a rising trend following a decline during the first few days (Fig. 7a). Based on the pigment ratios obtained in the mono-culture, the maximum Chl-a concentrations of diatom and prymnesiophytes in the HC enclosures were 166 ± 62 and 21 ± 33 µg L −1 , while they were 76 ± 6 and 5.5 ± 3.5 µg L −1 in the LC treatments.
The production of CaCO 3 (calcification) by prymnesiophytes in the HC enclosures (1.27 ± 0.07 µmol kg −1 d −1 ) was significantly lower than in the LC treatment (1.58 ± 0.12 µmol kg −1 d −1 ), while the production of particulate and dissolved organic carbon (POC and DOC) showed opposite trends (Table 1). These values in the HC enclosures (20.2 ± 7.6 and 29.7 ± 3.4 µM d −1 , respectively) were significantly higher than those in the LC condition Nitrite + Nitrate (c) in 6 enclosures which were perturbed by bubbling with ambient air (400 ppmv CO 2 , Low CO 2 level) or an air/CO 2 mixture at a concentration of 1000 ppmv CO 2 (High CO 2 level). Symbols are the means and error bars are the standard errors of three replicate enclosures.

Discussion
Enhanced biomass and production under HC treatment. As we mentioned above, acidification in eutrophic waters is not only accompanied by but also coupled to the development of eutrophication 3 . In natural sea water, typically the DOC concentration is less than 100 µM, and a least half of which is resistant to biological ) in 6 seawater enclosures which were perturbed by bubbling with ambient air (400 ppmv CO 2 , Low CO 2 level) or an air/CO 2 mixture at a concentration of 1000 ppmv CO 2 (High CO 2 level). Symbols are the means and error bars are the standard errors of three replicate enclosures. Nutrients ratio are shown during the nutrients replete period (day 0-7). ) during the nutrients replete period (day 4-7) in 6 seawater enclosures which were perturbed by bubbling with ambient air (400 ppmv CO 2 , Low CO 2 level) or an air/CO 2 mixture at a concentration of 1000 ppmv CO 2 (High CO 2 level). Columns are the means and error bars are the standard errors of three replicate enclosures. degradation 17 . In coastal eutrophic seawater, the concentration of labile DOC is high. In our system, the initial concentration was more than 200 µM. It is obvious that the decomposition of DOC led to the high pCO 2 in the initial water (805 µatm). Although the initial pCO 2 in the water in our study was much higher than the atmospheric CO 2 , the continued aeration in HC enclosures further exacerbated the pre-existing ocean acidification and reduced the pH (800-1000 ppm, Figs 1 and 2). The continued aeration of 400 ppm air reflects the initially high pCO 2 water gradually returning to the air sea equilibrium state (400-800 ppm) by the exchange with ambient pCO 2 air. Our results confirm the two states. The mean pCO 2 levels in the HC and LC treatments were 879.15 ± 145.76 and 658.05 ± 113.98 ppmv, with the significant differences of 221.10 ppmv in the first 4 days (Fig. 2b). Therefore, the results of this study may inform us about phytoplankton responses during a similar natural bloom in current low-pH coastal eutrophic environments under higher atmospheric CO 2 concentrations in the future.
Many previous mesocosm studies were conducted in high latitude waters (Table S1), where low temperatures lead to high solubility of CO 2 , and potentially larger biological effects of acidification 6,7 . Our work is the first mesocosm CO 2 perturbation study in a eutrophic subtropical ecosystem (24.5°N, 118.2°E) and thus could µ Figure 6. Temporal variations in concentrations of chlorophyll a (a) and diagnostic pigments of diatoms (b, Fucoxanthin) and prymnesiophytes (c, 19′hexanoyl-oxy-fucoxanthin, 19-Hex) in 6 seawater enclosures which were perturbed by bubbling with ambient air (400 ppmv CO 2 , Low CO 2 level) or an air/CO 2 mixture at a concentration of 1000 ppmv CO 2 (High CO 2 level). Symbols are the means and error bars are the standard errors of three replicate enclosures.
be comparable to the previous mesocosm experiment in the coastal waters of Korea (34.6°N and 128.5°E) 14,18 . However, the Korean experiments were performed in winter. More importantly, these previous studies were done by adding inorganic nutrients to trigger algae blooms. While, we used natural sea water after filtering out particles (< 0.01 µm). The high nutrient concentration is its own characteristics (eutrophication in Chinese coastal water). Nutrients concentrations, nutrients compositions (N/P/Si, NH 4 + /NOx, DIN/DON) and other factors (such as metals) were the same as the in-situ seawater with a salinity of 27.3.  In the HC treatment, clear enhancement of phytoplankton total Chl-a biomass and diagnostic pigment concentrations of both diatom and prymnesiophytes were observed (Fig. 6). These results are consistent with the production of particulate and dissolved organic carbon in the HC enclosures ( Table 1). The increases in pH and DIN/ DIC/pCO 2 drawdown due to phytoplankton growth during the experiment were also larger in the HC scenario than those of the LC treatment (Figs 1-3). Similar to our results, increasing inorganic carbon concentrations have been shown to enhance primary production 19 and carbon assimilation in various photoautotrophs. Of course, studies have reported similar results for diatoms 20 and coccolithophores 21 in mono-culture. Growth of a natural phytoplankton community was also stimulated by elevated CO 2 in an Arctic mesocosm experiment, leading to enhanced nutrient uptake and higher biomass build-up right after dissolved inorganic nutrient addition 16 . The enhanced production and exudation of organic matter in particular stimulated microbial loop activities and altered food web structure 7 .
It has been suggested that CO 2 could limit carbon fixation by marine phytoplankton and by large diatoms in particular, as the free CO 2 concentration is usually below that required for half saturation of Ribulose-1, 5-bisphosphate Carboxylase Oxygenase (RUBISCO), the core carbon-fixing enzyme in photosynthesis. There is experimental support for this idea 20,22 , even though most phytoplankton can utilize cellular C-concentrating mechanisms (CCM) based on the active uptake of CO 2 and/or HCO 3 − from the environment 23 . The quite high pH values (8.90) and low pCO 2 in water (31 µatm) on day 10 in our HC treatments also indicate that there might be insufficient CO 2 for phytoplankton (Figs 1 and 2). For diatoms (Phaeodactylum tricornutum), previous studies showed that growth and photosynthetic carbon fixation rates were enhanced by the enrichment of CO 2 under low or moderate levels of light 24 , though photosynthetic inorganic carbon affinity was down regulated by 20% under the high CO 2 condition 22 . Therefore, it is reasonable to infer that the growth of phytoplankton in eutrophic water, in particular that of large diatoms 25 , can be enhanced by high CO 2 ( Fig. 6 and Table 1).
However, there are also reports of neutral responses and even negative effects 26 . Species-specific CO 2 responses could result from taxonomic differences among phytoplankton in the physiological mechanisms of CO 2 uptake 27 . In addition to increased pCO 2 in seawater under HC, lowered pH can lead to acidic stress. The lower calcification rates (Table 1) reflect a more rapid dissolution of CaCO 3 or reduced rates of calcification in the low pH enclosures 28 . The difference on the ratios of PIC:POC between treatments suggest two opposing effects on the production of organic and inorganic carbon, respectively ( Table 1). The lowered pH may also reduce the ability of some species to tolerate high light stress, resulting in increases in respiratory carbon loss 29 . Thus, whether or not phytoplankton will benefit from increased CO 2 remains controversial 26 , since species-specific behavior and different physiological processes in different waters or experimental conditions are involved 29,30 . In eutrophic coastal waters where diatoms dominated in the phytoplankton community, our results support the general proposition that increasing CO 2 may promote phytoplankton photosynthesis and growth 7, 15 . Species composition under high CO 2 . Changes in dissolved aqueous CO 2 may affect phytoplankton community structure 31 . A meta-analysis of published experimental data emphasized that the differing responses to elevated pCO 2 caused sufficient changes between phytoplankton types in competitive fitness to significantly alter community structure 32 . Our results indicate a rapid response of prymnesiophytes in the HC treatment, while a short time-lag was observed in the stimulation of diatom growth (Figs 6 and 7). This response was the opposite of the expected trend, by which large phytoplankton should grow faster than small phytoplankton in nutrient replete conditions while the latter have the advantage in oligotrophic environments. This unexpected result could be due to the greater contribution of NH 4 + to total DIN compared to NO 3 − , as prymnesiophytes began to decline in day 6 when NH 4 + was depleted (Figs 3 and 6). Results of an Arctic mesocosm experiment also indicated a positive effect of HC on pico-and nano-eukaryotes during the nutrient replete phase 16 . The authors of this study suggested that if cells are small enough that their dissolved inorganic carbon supply needs can be met at least partly by diffusion, higher seawater CO 2 concentrations could stimulate photosynthetic carbon fixation and growth in these species 16 .
Our study indicates that diatoms have an overall advantage in eutrophic HC waters compared to prymnesiophytes, although the initial Chl-a concentrations of both groups were the same and the initial cell numbers of prymnesiophytes were even higher than those of diatoms (Fig. 6). As mentioned above, the decreased affinity for HCO 3 − or/and CO 2 and down-regulated CCM in diatom can save CCM-operational energy 22,33 , so that increased CO 2 availability can be beneficial in terms of energetics. In contrast, prymnesiophytes could have a competitive advantage over diatoms in low CO 2 environments 34 . Similarly, it was observed a clear succession from prymnesiophytes to diatoms when CO 2 concentrations increased from 150 to 750 ppmv 31 . Another experimental study in the North Atlantic Ocean, however, showed a shift away from diatoms and towards coccolithophores under HC, warmer conditions 35 . It has been suggested that paleooceanographic data showing lower Si:N utilization ratios by phytoplankton during the last glacial maximum could be due to community shifts towards non-siliceous species such as prymnesiophytes caused by decreased CO 2 in the glacial atmosphere (180 ppmv) 31 , although other studies have attributed lower Si:N utilization ratios and export of biogenic Si to relaxed iron limitation of diatoms during glacial periods 36 . In addition, other factors are likely to be also important, such as light. Prymnesiophytes can grow fast and make dense blooms, but they also like high light conditions 37 . In a mesocosm where Chl-a reaches such high concentrations, light must be much reduced through self-shading, possibly not favoring prymnesiophyte growth.
Effects on stoichiometry. Overconsumption of carbon relative to nutrient uptake has been reported in several studies of the planktonic response to increased CO 2 9, 20, 38 . For the same uptake of inorganic nutrients, net community carbon consumption under increased CO 2 exceeded present rate by 27 and 39% in 700 and 1050 μatm respectively 1 . However, many other studies have found contrary trends or no effect of high CO 2 on C:N ratios 10,13,14,18,39 . Differing effects of HC on stoichiometric uptake ratios (Table S1) may be attributed to the different biogeochemical demands of the dominant plankton functional types, and life stage-specific biogeochemical requirements 13 .
Our results indicate a clear increase in N:P and N:Si consumption ratios under HC treatments throughout the experiment, supporting higher Chl-a and production (Figs 3-6 and Table 1). In other words, assimilations of SRP and SiO 3 2− may be reduced relative to that of nitrogen in eutrophic coastal environments under high CO 2 . A previous study reported that C:P, in contrast to the C:N response, increased significantly in the post-bloom phase 18 . An Arctic mesocosm project with higher CO 2 had higher POC/POP and PON/POP during the nutrient rich phase 16 . Based on our results, the differences in N:P and N:Si consumption ratios between the two treatments (Figs 4 and 5) were completely due to changes in nitrate drawdown (Fig. 3C). Previous study revealed that elevated pCO 2 strikingly reduced NO 3 − uptake and assimilation in the diatom Thalassiosira pseudonana at both high and low light, as indicated by both short-term and steady-state net NO 3 − uptake rates, which was further supported by the reduced gene transcription, protein expression and enzymatic activity of nitrate reductase at high CO 2 33 . In parallel, diminished NO 3 − uptake at elevated pCO 2 resulted in lower PON and total protein content 40 . This could be more important for larger diatoms as they require a greater fraction (by 3.5-fold compared with small ones) of their total cellular nitrogen to RUBISCO for maintaining carbon fixation, hence a higher nitrogen cost in larger diatoms for RUBISCO leads to higher nitrogen requirements 25 . In addition, accumulating evidence suggests that the nitrogen cycle may respond strongly to higher pCO 2 through increases in global N 2 fixation 41 and possibly denitrification 42 , as well as decreases in nitrification 43 .
In contrast, to date, most studies have found negligible or minor effects of projected future changes in pCO 2 on most phytoplankton phosphorus requirements 10 . If indeed N:P and N:Si consumption ratios are elevated under HC environment, the P and Si limitation often observed in fresh and coastal waters will be possibly eased by lower consumption of SRP and Si under HC conditions, relative to the same uptake of DIN (Figs 3 and 4). This is also likely to be the cause of higher Chl-a biomass in HC environments with the same initial nutrient concentrations (Fig. 6).
Moreover, the ongoing increase in wind-driven upwelling 44 and anthropogenic nutrient inputs in coastal systems 45 may increase nutrient inputs and blooms in coastal waters 46 . The sustained increase in nitrate loading from the Mississippi River 47 , the Pearl River 48 and the Changjiang River 49 has resulted in rapidly rising nutrient ratios (N:P and N:Si) since the 1950s, due to increased use of agricultural fertilizers. Associated with the higher Chl-a biomass, species succession and eased P and/or Si limitation discussed above, our study further suggests that future climate and land use changes may result in even more serious and complicated interactive effects of eutrophication, ocean acidification and hypoxia in coastal waters 3, 50 .

Conclusions
Shallow coastal areas are vulnerable to the effects of human development, and can receive massive loads of fresh water, nutrients, and organic and inorganic carbon. In this study, a mesocosm CO 2 perturbation study was conducted to investigate the effect of rising CO 2 on a model plankton community in a eutrophic subtropical bay in China. Although the initial DIC and further decomposition of organic matter led to the pCO 2 in the coastal water being much higher than that in the air, further enrichment of CO 2 appeared to be conducive to the production and biomass of both diatoms and prymnesiophytes. Diatoms had a clear advantage in this highly eutrophic water under the elevated CO 2 concentration. However, prymnesiophytes seemingly responded rapidly in the HC enclosures, whereas a time lag was observed in diatom growth. Compared with the low CO 2 treatments, the N/P and N/Si consumption ratios significantly increased during the growth of phytoplankton at higher CO 2 partial pressure. These results indicate complex effects induced by ocean acidification in phytoplankton stoichiometry, production and community structure in eutrophic coastal waters which may have serious consequences for these biologically and economically important ecosystems.

Material and Methods
Experimental setup and sampling. The Xiamen University mesocosm facility for ocean acidification impacts study (FOANIC-XMU, http://mel.xmu.edu.cn/dynamicfile.asp?id=76) was deployed in Wuyuan Bay, Xiamen, China (24.5°N, 118.2°E) 51 . The dimensions of the floating platform are 28 × 10 m, and the facility includes 9 mesocosm enclosures immersed in the seawater along the south side of the platform to avoid shading (Supplementary Material, Fig. S1). The enclosures are 3 m deep and 1.5 m wide, with 50 cm projecting above the sea surface. The volume of the enclosures was 4 m 3 , and they were composed of a 0.9 mm thick cylindrical transparent thermoplastic polyurethane plastic membrane that is partially transparent to UV. The mesocosms are covered with plastic domes to reduce the contamination risk and prevent rainfall from diluting the experiments.
In order to minimize the influences of other groups of organisms such as grazers, and remove non-living suspended particles that would potentially affect the later measurements of biogenic elements, in situ seawater was filtered through a water purifier (MU801-4T, Midea) which was equipped with 0.01 μm pore size cartridges, and simultaneously injected into the enclosures. Then 0.2 g L −1 of NaCl solution was added into each mesocosm to determine the exact volume by comparison of the salinity before and after salt addition. Ocean acidification conditions were induced gradually with aeration using CO 2 enrichlors (CE-100, Wuhan Ruihua Instrument & Equipment, China). The pCO 2 of seawater in 6 of the enclosures was perturbed by bubbling with free air (ambient CO 2 , ~400 ppmv, denoted LC) or using an air/CO 2 mixture at the concentration of 1000 ppmv (denoted HC). The air with different CO 2 concentrations was delivered into the seawater at a flow rate of ~5 L min −1 with 6 mm diameter plastic tubing, and dispersed by an air stone disk placed in the center of each mesocosm's bottom. The bubbling was continued for the duration of the whole experiment.
The species interactions of natural populations are typically extremely complicated, depending on the abundance and intrinsic properties of various species as well as other abiotic and biotic factors. Therefore, four well-studied phytoplankton species were used in this study, including the diatoms Phaeodactylum tricornutum (CCMA 106, from Center for Collections of Marine Algae, Xiamen University) and Thalassiosira weissflogii (CCMP 102, from The National Center for Marine Algae and Micobiota, USA), and the coccolithophores Emiliania huxleyi (CS-369, from Commonwealth Scientific and Industrial Research Organization, Australia) and Gephyrocapsa oceanica (NIES-1318, National Institute for Environmental Studies, Japan), were first mono-cultured indoors at 400 and 1000 ppmv CO 2 . They were acclimated for 10 days (about 10-15 generations) at 20 °C and 150 μmol m −2 s −1 (cool white fluorescence) irradiance using the same eutrophic bay seawater later used for the experiments, collected and filtered in situ without nutrient additions. After the acclimation period, these species were inoculated into each enclosure with equivalent chlorophyll a (Chl-a), respectively, at a total final abundance of 5.07*10 4 cells L −1 . The initial abundances of Phaeodactylum tricornutum, Thalassiosira weissflogii, Emiliania huxleyi and Gephyrocapsa oceanica were about 10000, 700, 20000 and 20000 cells L −1 , respectively. The initial cell concentrations were set up based on their differences in size and Chl-a per cell to yield similar initial biomass levels. The growth of these phytoplankton groups was studied for 15 days (15 to 30 June, 2013) under present and elevated partial pressure of CO 2 (pCO 2 = 400 and 1000 ppmv, respectively) in triplicate 4 m 3 enclosures.
Environmental factors. The salinity, temperature and pH profiles in the mesocosm were measured daily at 10:00 AM with a CTD (RBR) or a SeaFET (Satalantic). Due to aeration, these profiles indicated the water in each enclosure was homogenized (Supplementary Material, Fig. S2). The subsamples for chemical and biological determinations were taken at the middle of the mesocosms with a water sampler. The total amount of water for sampling from the enclosures was less than 5% of the initial volume. The pH values were determined using the pH indicator meta-cresol purple with a spectrophotometer (Agilent 8453), with a measurement accuracy of ± 0.0005. The CO 2 partial pressure in the seawater (pCO 2 ) was calculated by the program CO2SYS 52 from dissolved inorganic carbon (DIC) and total alkalinity (TA) 53,54 .
Nutrient samples for NO 3 − + NO 2 − (NO x ), NH 4 + and soluble reactive phosphate (SRP) were filtered immediately after collection through 47 mm GF/F filters and were then frozen at −20 °C. Samples for SiO 3 2− were filtered through acid-cleaned 0.45 µm pore size acetate cellulose filters and were kept refrigerated at 4 °C. Filtrates for NH 4 + and SiO 3 2− were preserved with 100 μL CCl 4 . All the nutrient samples were measured in our land-based laboratory at Xiamen University within 15 days after the mesocosm experiment. Nutrient samples for NO X , SRP and SiO 3 2− were analysed using a four-channel continuous-flow Technicon AA3 Auto-Analyzer (Bran-Lube GmbH) 55 . NO x was analyzed using the copper-cadmium column reduction method 5 . SRP and SiO 3 2− were measured using typical spectrophotometric methods 56 . NH 4 + was run with a 722 type spectrophotometer (Xiamen Analytical Instrument Co., China) according to the indophenol blue spectrophotometric method 5,57 . The precision for nutrient analyses in this study was ≤3%, and the detection limits for NO x , SRP, SiO 3 2− and NH 4 + analyses were 0.03, 0.03, 0.05 and 0.16 μM, respectively. Phytoplankton biomass and community structure. Photosynthetic pigment concentrations were measured by high-performance liquid chromatography (HPLC) using a Shimadzu 20 A HPLC system fitted with a 3.5 μm Eclipse XDB C 8 column (4.6 × 150 mm, Agilent Technologies, Waldbronn, Germany), as in our previous studies 8,58 . Briefly, seawater samples for phytoplankton pigment analysis (0.2-2 L, according to biomass) were filtered through 25 mm GF/F glass fiber filters (under a vacuum pressure < 10 kPa under dim light), and then they were immediately frozen (−80 °C) until analysis in the laboratory within 30 days. Phytoplankton pigments were extracted in N, N-dimethylformamide and analyzed. The concentrations of chlorophyll a (Chl-a), diagnostic pigments of diatoms, (fucoxanthin, Fuco) and prymnesiophytes (19′hexanoyl-oxy-fucoxanthin, 19-Hex) were detected, and quantified using standards that were purchased from DHI Water & Environment, Hørsholm, Denmark.
Pigment samples of the four phytoplankton species were collected in exponential growth phase during mono-culture under indoor conditions to obtain the ratios of diagnostic pigments to Chl-a for CHEMTAX analyses 8,58 . These initial pigments ratios might be quite different from those of the same species grown under sunlight in the mesocosms. To confirm the phytoplankton community structure based on pigments information, a 0.5 L water sample was collected and preserved with Lugol's iodine solution for microscopic observation. The Chl-a concentrations were also compared to the cell number, and these two values were significantly correlated (Supplementary Material, Fig. S3, R 2 = 0.76, n = 27, p < 0.01).
Carbon metabolism. Net calcification rate was estimated by the TA anomaly method 59 . TA was measured by potentiometric titration using the Gran procedure 60 . In addition, TA has to be corrected for the effect of primary production, i.e., since an uptake of 1 mol NO 3 − or PO 4 3− increases TA by 1 mole 59 , and an uptake of 1 mol NH 4 + decreases TA by 1 mole 61 . Thus, the net calcification rate can be estimated as equation (1) For particulate organic carbon (POC) sampling, seawater was filtered through a pre-combusted 47 mm GF/F filter (Whatman). Prior to analysis, the filters were frozen at −20 °C and POC (via acid fuming) was determined with a PE-2400 SERIES IICHNS/O analyzer according to the JGOFS protocols 56 . Dissolved organic carbon (DOC) samples were filtering onto pre-combusted 25 mm GF/F filters and stored in pre-combusted EPA vials at −20 °C until analysis using a Shimadzu TOC-V Analyzer 62 . The overall POC and DOC production rates for 15 days (day 0-day 14) were calculated according to the following equations: Since particulate inorganic carbon (PIC) samples were limited, we assumed that all PIC is in the form of CaCO 3 . Accordingly, the PIC value in this study was calculated with the following formula 63  , ΔDOC, ΔPOC, ΔPIC calculated denote the changes of these parameters, and Δt denotes the elapsed time. A one-way ANOVA was used for statistical analysis following a test for homogeneity of the variances. The significance level was set at p < 0.05. ANOVA results were compared using the Tukey HSD method.