The mesocosm experiment was conducted in the Facility for the Study of Ocean Acidification Impacts of Xiamen University, located in the subtropical coastal region Wuyuan Bay (24°31′48″ N, 118°10′47″ E), during 9th April–15th May 2018. Nine cylindrical mesocosm bags with 3 m in depth and 1.5 m in diameter, made of transparent thermoplastic polyurethane (TPU), were fixed within steel frames and sheltered by cone lids. The in situ seawater were filtered (pore size of 0.01 μm) by water purifier (MU801-4T, Midea, China) and pumped simultaneously into 9 mesocosm bags for 3000 L within 36 h. Bag 2, 4, 6, and 8 were set as control, aerated with ambient air pCO₂ (denoted as AC). To adjust seawater to projected 1000 μatm CO₂ in 2100, bag 1, 3, 5, 7, and 9 were added with approximately 11 L CO₂ saturation seawater (denoted as HC) (Riebesell et al., 2013). Subsequently, AC and HC mesocosm bags were aerated (5 L min^–1) with ambient air of ∼410 ppmv CO₂ and pre-mixed air-CO₂ of 1000 ppmv CO₂, respectively, till the end of the experiment. After aerating to homogenize the carbonate system for about 2–3 h, 720 L of in situ seawater filtered by 180 μm mash was added simultaneously into 9 mesocosm bags, so that each bag was inoculated with 80 L in situ seawater containing a natural microbe community. Subsequently, the mesocosm samples were collected from 0.5 m depths of each bag at 10:00 a.m. every 1–3 days, for physical, chemical and biological analysis.

The intensity of solar irradiance was measured with a real-time solar irradiance monitoring device (EKO, Japan). Salinity and temperature were measured using a salinometer and digital thermometer, respectively. The pH was determined instantly using a pH meter (Orion Star A211, Thermo Scientific), which was calibrated against NIST-traceable pH buffers. Samples for dissolved inorganic carbon (DIC) measurements were poisoned with final concentration of 1‰ saturated HgCl₂ solution and stored in 40 mL borosilicate glass vials until further analysis. The DIC was determined by acidifying 0.5 mL samples and then measuring CO₂ concentration using a DIC analyzer (AS-C3, Apollo, United States). The other parameters of carbonate system were calculated from DIC and pH_NBS using CO2SYS program (Lewis and Wallace, 1998) based on known levels of nutrients, temperature, and salinity.

Samples for nutrient measurements were filtered through 0.45 mm cellulose acetate membrane by suction filter. The filtrate was divided into 2 of 125 mL high density polyethylene bottles: one was stored at −20°C and measured concentrations of NO₃^–, NO₂^–, PO₄^3–, and SiO₃^2– using an auto-analyzer (AA3, Seal, Germany) at room temperature, and another added with 1‰ chloroform was stored at 4°C and measured concentration of NH₄⁺ with indophenol blue spectrophotometry (Tri-223, Spectrum, China) at 25°C.

Water samples of 250–1000 mL were filtered onto GF/F membranes by suction filter with low vacuum pressure (<0.02 MPa), and soaked in pure methanol at 40°C for 1 h. Then the extracts were centrifuged at 6000 × g for 10 min, and the absorption spectra of supernatants from 400 to 800 nm were measured using a UV-VIS Spectrophotometer (DU800, Beckman, United States). The Chl a concentration was calculated according to the equations from Ritchie (2006).

Samples filtered by 20 μm mash were collected into 2 mL centrifuge tubes and fixed with glutaraldehyde to a final concentration of 0.5% at room temperature for 15 min, then frozen with liquid nitrogen and stored at −80°C for further measurement. Viruses and bacteria were counted according to forward scatter, side scatter, and fluorescence intensity after samples were stained with SYBR Green I (Molecular Probe, United States), using the Epics Altra II flow cytometer (Beckman Coulter, United States) and Accuri C6 flow cytometer (Becton and Dickinson, United States), respectively (Chen et al., 2019). For the abundance measurement of both viruses and bacteria, fluorescent beads (Molecular Probes) with a diameter of 1 μm were added as an internal standard.

Water samples of 500–2000 mL were collected into polyethylene bottles and fixed by adding with 1% lugol’s iodine. Then the samples were static placed, and concentrated to 25 mL using siphons within 3 days. The concentrated samples were observed using microscopy and plankton chamber to analyze phytoplankton abundance and diversity based on morphological characteristics.

Water samples of 120 mL were collected and transferred immediately into 6 glass scintillation vials (3 vials were used for 12 h inoculation and 3 vials were used for 24 h inoculation) before sunrise. The samples were inoculated with 100 μL of 5 μCi (0.185 MBq) NaH¹⁴CO₃ solution (ICN Radiochemicals, United States) for 12 h and 24 h with similar levels of sunlight and temperature as in the mesocosms. After 12 and 24 h incubation, the cells were filtered onto glass-fiber filters (25 mm, Whatman GF/F, United States), and the filters were stored at −20°C. Later on, the filters were thawed and exposed to HCl fumes overnight and dried at 60°C for 6 h to remove unincorporated labeled carbon, then 3 ml of scintillation cocktail (Hisafe 3, Perkin-Elmer) was added. The incorporated radioactivity was measured by liquid scintillation counting (LS 6500, Beckman Coulter, United States). The respiratory carbon loss during night was calculated as the difference between carbon fixation of 12 h (day-time primary productivity) and 24 h (daily net primary productivity).

Water samples of 150–500 mL for biogenic silica measurement were collected onto 2 μm polycarbonate membranes, and rinsed with filtered seawater. Each membrane sample was put into 15 mL centrifuge tube, dried and stored in dryers until analysis in the laboratory. The biogenic silica contents were measured by molybdenum-blue-ness spectrophotometry (Brzezinski and Nelson, 1995). Briefly, the samples were placed in a boiling water bath for 40 min, after mixed with 4 mL of 0.2 mol L^–1 NaOH. Then the samples were cooled on ice, and neutralized with 1 mL of 1 mol L^–1 HCl. The supernatant, obtained by centrifuging samples at 6000 × g for 10 min, were diluted with MilliQ water, and then added with 2 mL of 0.006 mol L^–1 ammonium molybdate and stood for 10 min. Finally, the samples were chromogenically reacted with reducing reagent for 3 h. The absorbance of samples was measured at 810 nm using a UV-VIS Spectrophotometer (DU800, Beckman, United States). For the calculation of biogenic silica contents, 0–30 μmol L^–1 Na₂SiF₆ were used as standards.

The statistical analysis was performed using SPSS Statistics, R and Origin. Independent-samples t-test was conducted to check the significant effects of increased pCO₂ at the level of p < 0.1, because the variations were larger within ecological replicates than biological replicates. To further detect the pCO₂ effects on Chl a concentration, the abundance of viruses and bacteria, BSi concentration, primary productivity and night respiration, the generalized additive mixed-effect modeling (GAMM, R packages “mgcv”) was applied to analyze the data with the following procedures (Bach et al., 2017). Three GAMM models (Model I: pCO₂ did not affect the dependent variable, Model II: pCO₂ influenced the constant offset in temporal trends, Model III: pCO₂ changed trend shape) were fitted to each independent variable as a function of time, in which pCO₂ and mesocosm number were set as explanatory and random variables, respectively. To ensure model assumptions were satisfied, we considered heteroscedasticity and temporal autocorrelation of residuals, as well as autoregressive structures. The model with the highest coefficient of correlation (R²) was defined as the best that described the pCO₂ effects.

Throughout the mesocosm experiment, most of the days were sunny, with daily mean PAR (photosynthetically active radiation) ranged 690–1075 μmol photons m^–2s^–1 (Supplementary Figure 1). The average salinity was 30.4‰, and temperature increased gradually from 21.9 ± 0.1 to 26.1 ± 0.1°C during the period of 9th April–15th May, 2018 (Figure 1A).

The differences of pH_NBS, pCO₂ and DIC concentration between AC and HC treatments were maintained, respectively, by 0.13–0.31, 230–600 μatm, and 58–130 μmol kg^–1 throughout the experiment (p < 0.1, Figures 1B–D and Supplementary Table 1). The pH_NBS of HC and AC treatments dropped, respectively, from the initial values of 7.79 ± 0.03 and 8.01 ± 0.01, to 7.70 ± 0.05 and 7.87 ± 0.03 on day 4. Correspondingly, the DIC concentration and pCO₂ increased by 26 μmol kg^–1 and 206 μatm in HC treatment, and by 41 μmol kg^–1 and 221 μatm in AC treatments, respectively. Then the pH_NBS increased to peak values of 7.94 ± 0.13 in HC treatments on day 10 and 8.13 ± 0.03 in AC treatments on day 12, and the corresponding DIC concentration (pCO₂) decreased to 1920 ± 52 μmol kg^–1 (735 ± 246 μatm) in HC treatments and to 1824 ± 8 μmol kg^–1 (427 ± 30 μatm) in AC treatments. Subsequently, the pH_NBS decreased by 0.21 units in HC treatments and by 0.18 units in AC treatments on day 28, and the corresponding DIC concentration (pCO₂) increased by 36 μmol kg^–1 (501 μatm) in HC treatments and by 44 μmol kg^–1 (263 μatm) in AC treatments. Afterward, the carbonate system of both HC and AC treatments sustained relatively stable levels till the end of the experiment.

The nutrient concentrations experienced drastic changes in the early stage of the mesocosm experiment, and maintained stable afterward (Figure 2). The initial concentrations of NO₃^–, NO₂^–, NH₄⁺, PO₄^3–, and Si(OH)₄ in all mesocosm bags were 6.73–10.65, 0.49–0.79, 6.07–8.71, 0.18–0.47, and 5.42–9.50 μmol kg^–1, respectively. The concentrations of NH₄⁺ and PO₄^3– decreased rapidly to nearly 0 in both HC and AC treatments on day 2. However, the concentrations of NO₃^–, NO₂^–, and Si(OH)₄ increased at first and then decreased. The NO₃^– concentration increased to 20.62 ± 14.87 μmol kg^–1 in HC treatments and to 12.48 ± 2.77 μmol kg^–1 in AC treatments on day 2, and then decreased dramatically to 0.10–4.73 μmol kg^–1 on day 4. The NO₂^– concentration increased up to day 2 and then maintained relatively stable till day 6, with the values of 1.99 ± 1.68 μmol kg^–1 in HC treatments and 1.20 ± 0.38 μmol kg^–1 in AC treatments. Subsequently, the NO₂^– concentration decreased to 0.39 ± 0.48 μmol kg^–1 in HC and to 0.44 ± 0.61 μmol kg^–1 in AC treatments on day 8. The Si(OH)₄ concentration increased to 10.60 ± 2.50 μmol kg^–1 in HC treatments and to 12.68 ± 3.03 μmol kg^–1 in AC treatments on day 2, then decreased progressively to 1.08 ± 1.00 μmol kg^–1 in HC and to 0.64 ± 1.28 μmol kg^–1 in AC treatments on day 10. After reaching the very low values, the concentrations of all nutrients were stable till the end of the mesocosm experiment. No significant differences in nutrient concentrations were observed between HC and AC treatments during the whole mesocosm experiment (p > 0.1), except that the average concentrations of both NO₃^– and NO₂^– were higher (p = 0.289 and 0.095, respectively) in HC treatments than in AC treatments during the initial stage of the experiment.

As shown in Figure 3, the temporal changes in autotrophic biomass were indicated by Chl a concentration, during the whole mesocosm experiment. The Chl a, with initial concentration of 0.17 μg L^–1, increased to 7.91 ± 4.02 μg L^–1 on day 12 in AC treatments and 6.41 ± 2.80 μg L^–1 on day 11 in HC treatments. Then it decreased to 0.91 ± 0.31 and to 1.03 ± 0.73 μg L^–1 on day 22 in AC and HC treatments, respectively. Thereafter, Chl a concentration maintained stable till day 25 in AC treatments and to day 28 in HC treatments. Subsequently, it increased slightly to 1.34 ± 1.42 μg L^–1 in AC treatments and to 1.00 ± 0.75 μg L^–1 in HC treatments till the end of the experiment. Based on the ln scale of Chl a concentration, phytoplankton growth kinetics in both AC and HC treatments was classified into four phases: exponential phase (Phase I) from days 0 to 8, stationary phase (Phase II) from days 8 to 14, decline phase (Phase III) from days 12 to 25 (AC treatment) or day 28 (HC treatment), and exponential phase (Phase IV) of another growth cycle from day 25 (AC treatment) or day 28 (HC treatment) to day 31. The average value of Chl a concentration was slightly higher in HC treatments than in AC treatments during days 8–16, although the GAMM results showed that HC did not influence the development of Chl a concentration (Table 1).

The abundance of viruses and bacteria underwent dynamics with different patterns, in both HC and AC treatments (Figure 4). The viruses abundance with initial values of 2.01–4.31 × 10⁶ particles mL^–1 increased, with the growth of autotrophic plankton, to 1.65 ± 0.20 × 10⁸ particles mL^–1 in HC treatments on day 10 and to 1.00 × 10⁸ ± 0.34 × 10⁸ particles mL^–1 in AC treatments on day 11, respectively. After reaching the peak, the viruses abundance decreased to 3.22 ± 0.94 × 10⁷ particles mL^–1 in HC treatments and to 2.70 ± 0.56 × 10⁷ particles mL^–1 in HC treatments on day 16. Subsequently, the viruses abundance sustained relatively stable values till the end of the experiment in the AC treatments. In the HC treatments, however, it did not change significantly during days 16–22 but increased to 5.46 ± 2.75 × 10⁷ particles mL^–1 during days 25–31. The GAMM analyses revealed that elevated pCO₂ influenced the viruses trend shape (Table 1 and Supplementary Figure 2), with higher abundance in HC during days 6–14 and days 25–31, relative to AC.

The initial abundance of bacteria was 1.78 ± 0.25 × 10⁶ cells mL^–1 in HC treatments or 1.85 ± 0.96 × 10⁶ cells mL^–1 in AC treatments. In the early stage of the experiment, the bacteria abundance increased rapidly to 5.94 ± 0.94 × 10⁶ cells mL^–1 in HC treatments on day 2, which was higher than that in AC treatments of 4.44 ± 0.54 × 10⁶ cells mL^–1 (p = 0.059), and then the bacteria abundance decreased till day 4 under both HC and AC conditions. With the exponential growth of autotrophs (days 4–8), the bacteria abundance increased, however, it decreased in the stationary phase of phytoplankton (days 8–12). A constant offset in bacteria abundance existed between AC and HC, with higher bacteria abundance in HC throughout the experimental period, indicated by GAMM analyses results (Table 1 and Supplementary Figure 2).

Using microscopy, a total of 34 genera were identified, in which 26 genera belonged to diatoms and 8 genera to dinoflagellates. From all of the mesocosm bags, the dominant genera were Chaetoceros, Thalassiosira, Rhizosolenis, Eucampia, and Prorocentrum, and the dominant species were Chaetoceros curvisetus, Rhizosolenis fragilissima, Eucampia zoodiacu, Cerataulina pelagica, and Guinardia delicatula.

Phytoplankton community structure underwent dynamic succession at phylum levels (Figure 5). Diatoms accounted for a proportion of above 90% in both treatments during the period of days 0–16, subsequently decreased to 79.2 ± 13.7% in HC treatments on day 20, being higher than that of 52.2 ± 26.9% in AC treatments, although no significant differences were observed (p = 0.138). The relative abundance of diatoms appeared to decrease to 39.8 ± 28.8% in AC on day 28, but maintained stable by 81.8 ± 14.5% in HC treatments, resulting in more significant differences in relative abundance of diatoms between HC and AC treatments (p = 0.054). Correspondingly, dinoflagellates dominated by Prorocentrum minimum and Peridinium conicoides, maintained relatively stable abundance with variations less than 10% in HC treatments, but increased to 60.2 ± 28.8% in AC treatments.

The dynamic changes of community structure were more complicated at genus and species levels than at phylum level (Figure 6 and Supplementary Figure 3). The diatom Guinardia, dominated by G. delicatula, was the most abundant with relative abundance of 53.4% in terms of total diatom + dinoflagellate species, with cell density of about 9.9 × 10⁴ cells mL^–1, after in situ community was inoculated into mesocosm bags. However, it decreased rapidly to average relative abundance of 3.2% (density of 2.6 × 10³ cells mL^–1) and to 1.0% (density of 1.6 × 10³ cells mL^–1) on day 3 in AC and HC treatments, respectively, and no longer existed from day 6. The diatom Eucampia in both HC and AC treatments, mainly composed of E. zoodiacu, maintained density of approximately 6 × 10⁴–9 × 10⁴ cells mL^–1 during days 0–16, then dropped to 0.2 × 10⁴–3.7 × 10⁴ cells mL^–1 on day 20 and maintained stable afterward. The relative abundance of Eucampia increased from initial value of 29% to 61–76% and to 71–75% during days 3–6, and subsequently decreased to 7–16% and to 1–13% on day 10, in HC and AC treatments, respectively. The diatom Chaetoceros grew exponentially from 1.4 × 10⁴ cells mL^–1, and reached the peak on day 10 with average concentration of 2.9 × 10⁶ cells mL^–1 in HC and 3.4 × 10⁶ cells mL^–1 in AC treatments, which accounted for 65.3 and 71.9% of diatom + dinoflagellate assemblages, respectively. Subsequently, the dominant genus in both HC and AC treatments became the diatom Thalassiosira and Rhizosolenis with the average relative abundance of 53 and 34% on day 16, respectively. Then the abundance of Thalassiosira and Rhizosolenis decreased progressively. During the late stage of these experiment, HC stimulated the growth of the diatom Nitzschia on day 20, but AC stimulated the growth of the dinoflagellate Propocentrum on day 28. As a consequence, the diatom Nitzschia had significantly higher relative abundance during days 20–28 (p = 0.019), with the dinoflagellate Propocentrum accounting for lower proportion on day 28 (p = 0.078), in HC relative to AC treatments.

The temporal variations of day-time primary productivity, daily net primary productivity and night respiratory carbon loss normalized to water volume showed similar changing patterns with autotrophic biomass levels indicated by Chl a concentration (Figure 7). The day-time primary productivity and daily net primary productivity increased with the growth of phytoplankton, from initial values of 24–77 and 21–63 μg C L^–1, to the peak values of 573 and 392 μg C L^–1 on average, respectively, as Chl a concentration reached the maximum. Subsequently, the day-time primary productivity and daily net primary productivity decreased progressively, with the decay of algal blooms, with lowered values of 6.5–204 and 6.3–111 μg C L^–1 after day 22. Respiration carbon loss of phytoplankton during night period increased from 15 μg C L^–1 on average during days 2–8 to 150 μg C L^–1 on average on day 9. Subsequently, the night period respiration of phytoplankton maintained a rate of 140–180 μg C L^–1 during the stationary phase of autotrophs, and decreased progressively with the decline of autotrophic blooms. Finally, the respiration stayed 7–9 μg C L^–1 during days 22–28, and increased slightly at the end of the mesocosm experiment. The daily net primary productivity and day-time primary productivity normalized to Chl a increased immediately from the initial values of 17.3 ± 4.0 and 26.8 ± 9.6 μg C (μg Chl a)^–1, to 88.7 ± 23.9 and 100.1 ± 30.3 μg C (μg Chl a)^–1 on day 6, and then sustained relatively stable levels with the average values of 48.2 and 67.9 μg C (μg Chl a)^–1. The night-period respiration normalized to Chl a, with lower values of 12.3 μg C (μg Chl a)^–1 during days 0–9, increased drastically and maintained higher values of 12.3 μg C (μg Chl a)^–1. Neither primary productivity nor night respiration of phytoplankton was influenced significantly by HC (Table 1), normalized to water volume or to Chl a, at any time points examined during the experiment.

Throughout the mesocosm experiment, the differences in temporal development of biogenic silica (BSi) contents were not observed between HC and AC treatments (Table 1 and Figure 8). With development of phytoplankton blooms, the BSi concentration in the water increased and reached peak values of 11.3 ± 0.5 μmol BSi L^–1 on day 12. Correspondingly, the BSi contents normalized to Chl a increased from the initial values of approximately 0.05–2.1 ± 0.8 μmol BSi (μg Chl a)^–1 on day 4, and then maintained relative stable till day 14. Afterward, the BSi concentration in the water decreased, while the BSi contents normalized to Chl a increased, with the decay of phytoplankton blooms till the end of the experiment.