Simulated ocean acidification reveals winners and losers in coastal phytoplankton

The oceans absorb ~25% of the annual anthropogenic CO2 emissions. This causes a shift in the marine carbonate chemistry termed ocean acidification (OA). OA is expected to influence metabolic processes in phytoplankton species but it is unclear how the combination of individual physiological changes alters the structure of entire phytoplankton communities. To investigate this, we deployed ten pelagic mesocosms (volume ~50 m3) for 113 days at the west coast of Sweden and simulated OA (pCO2 = 760 μatm) in five of them while the other five served as controls (380 μatm). We found: (1) Bulk chlorophyll a concentration and 10 out of 16 investigated phytoplankton groups were significantly and mostly positively affected by elevated CO2 concentrations. However, CO2 effects on abundance or biomass were generally subtle and present only during certain succession stages. (2) Some of the CO2-affected phytoplankton groups seemed to respond directly to altered carbonate chemistry (e.g. diatoms) while others (e.g. Synechococcus) were more likely to be indirectly affected through CO2 sensitive competitors or grazers. (3) Picoeukaryotic phytoplankton (0.2–2 μm) showed the clearest and relatively strong positive CO2 responses during several succession stages. We attribute this not only to a CO2 fertilization of their photosynthetic apparatus but also to an increased nutrient competitiveness under acidified (i.e. low pH) conditions. The stimulating influence of high CO2/low pH on picoeukaryote abundance observed in this experiment is strikingly consistent with results from previous studies, suggesting that picoeukaryotes are among the winners in a future ocean.


Introduction
The seasonal succession of plankton involves the occurrence and disappearance of plankton taxonomic and functional groups in an annually repeated pattern [1].The major biomass build-up during the spring bloom is traditionally seen as the starting point of the succession in temperate regions, although the initiation of the bloom already takes place in early winter [2,3].Succession patterns differ among oceanographic regions and are controlled by a multitude of abiotic factors such as turbulence, nutrients, and light [4], as well as biotic interactions, including competition, predation, and infection [1,4,5].
Changes in the marine carbonate system due to the net influx of anthropogenic CO 2 into the ocean's surface layer (i.e.ocean acidification (OA)) could alter phytoplankton succession because taxonomic groups shaping the succession pattern are differently sensitive to changing carbonate chemistry.Phytoplankton species which benefit from CO 2 fertilization may become more dominant in future communities while those which are unresponsive to increasing CO 2 or even detrimentally affected by decreasing pH could become less important or be replaced by other species [6][7][8][9].Uncovering the potential for CO 2 -induced community shifts is important as these can re-organize the energy flow through food webs and alter biogeochemical element fluxes [10,11].
In this study we investigated the influence of projected end-of-the century carbonate chemistry conditions (average pCO 2 = 760 μatm) on a natural winter-to-summer phytoplankton succession in a temperate coastal environment.Our study is part of the BIOACID II long-term mesocosm study which took place in Gullmar Fjord (Skagerrak, west coast of Sweden) from January to July 2013.It belongs to a series of papers covering various components of the plankton community in and outside the mesocosms.A summary on the main foci of these contributions is provided in the overview paper accompanying this PLOS collection [12].The focus in the present contribution is primarily on how CO 2 affects phytoplankton functional and taxonomic groups during the winter-to-summer succession.

Mesocosm deployment, manipulation, and maintenance
On the 29 th of January 2013, ten "Kiel Off-Shore Mesocosms for Future Ocean Simulations" (KOSMOS, M1-M10; [13]) were moored by research vessel Alkor in Gullmar Fjord on the Swedish west coast (58˚15' 50" N, 11˚28' 46" E).Study site, key events, deployment, and mesocosm manipulation procedures are described in detail in the abovementioned overview paper [12].In brief, each mesocosm was composed of an 8 m tall floatation frame and an 18.7 m long cylindrical polyurethane bag with a diameter of 2 m.The bags were folded and installed in the floatation frame before mesocosm deployment by Alkor.After deployment, bags were unfolded and lowered underwater to allow water exchange with the fjord.Water inside the bags was isolated from the fjord water by attaching 2 m long conical sediment traps to the bottom and pulling the upper end of the bag about 1.5 m above the surface [13,14].The mesocosm bags with the attached sediment traps reached ~19 m deep after the closing procedure.
Extended sea ice coverage prolonged the time between mesocosm deployment and the closing procedure.They were closed for the first time on the 12 th of February and CO 2 was manipulated in the high CO 2 mesocosms (M2, M4, M6, M7, M8) shortly thereafter.However, due to technical difficulties with the sediment traps we had to stop the experiment after 19 days on the 3 rd of March.Afterwards, mesocosms were lowered below surface to allow water exchange while the sediment traps were repaired on land.After four days we restarted the experiment by closing the mesocosms for the second time on the 7 th of March.The second experiment lasted for 113 days until the 28 th of June.In the present paper we only describe results from the second experiment.We will use the "experimental day nomenclature" which is consistent among all papers associated with this mesocosm study.Here, the 7 th of March is day -2 and the 28 th of June is day 111.
The mesocosms enclosed a volume ranging from 47.5 (M3) to 55.9 (M2) m 3 [12].The water was gently mixed directly after enclosure by bubbling the water column for 5 minutes with compressed air.A second bubbling procedure two days after enclosure (day 0) was necessary to fully eliminate the salinity stratification.All mesocosms were cleaned by divers from the outside approximately every second week and from the inside with a cleaning ring approximately every 8 th day.A mesh (1 mm mesh size) was attached to the cleaning ring on day 6 to remove large zooplankton (e.g.jelly fish) or nekton (e.g.fish) from the water column as these organisms were considered to be unevenly distributed among mesocosms.However, only very few organisms, mostly jelly fish and some fish larvae, were removed during this operation.
Five of the ten mesocosms were enriched with CO 2 -aerated seawater at the beginning of the experiment (M2, M4, M6, M7, M8) while the other five mesocosms remained unperturbed and served as controls (M1, M3, M5, M9, M10).High CO 2 concentrations had to be re-established on 5 occasions (days 17,46,48,68,88) during the study to compensate for CO 2 gas loss at the air-sea boundary layer of the mesocosms.
Due to the long duration of the experiment, we added 22 L of unfiltered fjord water to each mesocosm on every 4 th day thereby allowing plankton species which were not present during closing to enter the mesocosm [12].Green sea-urchin (Strongylocentrotus droebachiensis) and herring (Clupea harengus) larvae were added to each mesocosm on day 56 and day 63, respectively, to study the influence of OA on their development.Both species were added in relatively low densities (~90 herring eggs and 110 sea urchin larvae per m 3 ) to minimize potential topdown-effects [12].Please note that the OA response of these particular organisms will be addressed in detail in other publications.
Ethical statement: Herring welfare was assured by performing the experiment according to the ethical permission (number 332-2012) issued by the Swedish Board of Agriculture "Jordbruksverket").The species (Clupea harengus) used is not endangered and was obtained from a local registered and licensed fisherman (licence ID = 977 224 357).

Sampling, filtration, and measurements
All mesocosms were sampled every second day for usually 1-3 hours starting at 9 a.m.(local time) from small boats.The water column was sampled with integrating water samplers equipped with pressure sensors (IWS, Hydrobios) which collect 5 L of seawater evenly from the water column while being lowered from 0-17 m.Water from two IWS hauls were transferred with a tube into a 10 L carboy.The carboys were brought back to shore and stored in a temperature-controlled room set to in situ water temperature.Subsamples for particulate matter (PM), flow cytometry, light microscopy, and pigment analysis were taken from carboys shortly (usually within 1 hour) after they arrived in the temperature-controlled room.Each carboy was rotated gently before subsampling in order to avoid sedimentation bias.
PM samples were filtered (500 mL, Δpressure -200 mbar) on glass fiber (GF/F) filters and immediately photographed at full magnification with a CANON 60D and a EF-S 60 mm f/2.8 Macro lens.These pictures were manually processed to count the abundance of the large (>200 μm) diatom Coscinodiscus concinnus.
Pigment samples were filtered (800 mL, Δpressure -200 mbar) on GF/F filters.These filters were folded, put into 2 mL cryovials, and stored at -80˚C for 4-7 months until analysis.Pigments were extracted in 90% acetone and their concentrations quantified by means of reverse phase high performance liquid chromatography (HPLC) [15].The contribution of distinct phytoplankton taxa to total chlorophyll a (chla) was calculated with the CHEMTAX software which classifies phytoplankton based on taxon-specific pigment ratios [16].For the calculation, we used pigment ratios provided by Mackey et al. [16].Pigment data from all mesocosms were aggregated in one data sheet and evaluated in the same analysis run (iterations = 86, root-mean-square error = 0.22).Thus, the in-and output pigment ratios used in the CHEM-TAX analysis were identical in all mesocosms (S1 Table ).
Light microscopy samples were transferred from the carboys into 250 mL brown glass bottles and fixed with acidic Lugol's solution (1% final concentration).Phytoplankton were counted and identified 1-24 months after sampling at 200 and 400 times magnification with an Zeiss Axiovert inverted microscope [17].Light microscopic species counts comprised auto-, mixo-, and heterotrophic protists within a size range from ~10-500 μm.A species list with approximate organism sizes is provided in S2 Table.
Cells smaller than 10 μm were abundant in the experiment but could often not be taxonomically identified.The exceptions were small silicifying species (<5 μm) which were identified with scanning electron microscopy.Therefore, ~10 mL of water sample were gravity filtered on 0.2 μm polycarbonate filters and further processed as described by Bach et al. [18].
Flow cytometry subsamples for phytoplankton analyses were transferred from the 10 L carboys into 50 mL beakers directly after return of the sampling boats in the harbor.Subsamples (650 μL per mesocosm) were immediately analyzed within 3 hours using a Accuri C6 (BD Biosciences) flow cytometer [12].The flow rate estimated by the Accuri C6 was controlled and verified regularly by weighing sample before and after measurements and calculating the volume difference.Phytoplankton populations were distinguished based on the signal strength of the forward light scatter (FSC), the red fluorescence from chlorophyll a light emission (FL3), and the orange fluorescence from phycoerythrin light emission (FL2) [19].The FSC signal strength is positively correlated with particle size and can be used to distinguish phytoplankton size classes [20].To constrain the size range we conducted sequential fractionations using polycarbonate filters of different pore sizes (0.2, 0.8, 2, 3, 5, 8 μm) and gravity filtration [21].The following groups were defined (S1 Fig) : Synechococcus (0.8-3 μm), Cryptophytes (Crypto, 2-8 μm), Picoeukaryotes (Peuks, 0.8-3 μm), and four groups of Nanoeukaryotes with increasing size (Nano I-III between 2-8 μm with Nano III being the largest; Nano IV, >8 μm).The clusters of Peuk and Nano I-IV populations changed their appearance slightly in the course of the study.We accounted for these changes by adjusting gate shapes in the cytograms on 7 occasions (days 14, 30, 46, 66, 79, 94, 108).Importantly, gates and gate adjustments were identical in all mesocosms at every point in time.This was essential to keep the flow cytometry data comparable among replicates and treatments.The abundance of each flow cytometry group (in cells mL -1 ) was calculated as the number of events within a gate divided by the analyzed volume.
Bacterial abundance samples were withdrawn from the 10 L carboys directly after sampling and immediately fixed with glutaraldehyde (0.5% v/v; 30 minutes), flash-frozen in liquid nitrogen, and stored at -80˚C for 4-7 months until measurements with the Accuri C6 flow cytometer.Bacteria samples were prepared for analysis by thawing samples at 37˚C and staining them with SYBR green I for 15 minutes at 20˚C in the dark.Bacteria could be distinguished from non-living particles by the green fluorescence of the stained DNA and from phytoplankton by the lack of red fluorescence [22].Please note that archaea are also included in this analysis but they are typically much less abundant in surface waters than bacteria and therefore not specifically considered here as a separate group [23].

Statistical data analysis
We postulated that the high CO 2 treatment can affect the temporal trend of a dependent variable in four different ways: by amplification or weakening of the peak amplitudes ( To detect and visualize such potential responses we applied generalized additive mixed-effect modelling (GAMM; R packages "mgcv" and "nlme" [24][25][26]) and analyzed the data with the following procedure.Three different GAMM models were fitted to each dependent variable as a function of time (in our case as a function of the day of experiment; e.g.chla(day of experiment)).CO 2 was set as categorical explanatory variable modifying the absolute values of the dependent variable and/or its trend shape over time [25,27].The first model assumed no difference between treatments.The second model assumed a more or less constant offset in temporal trends but no change in trend shape (i.e.phenology).The third model assumed differences in phenology.In each model, the mesocosm number was set as random effect to account for any unknown effects of individual mesocosms.We accounted for heteroscedasticity and temporal autocorrelation of residuals in the models to ensure that model assumptions were satisfied [26].In most cases best fitting results were gained with a 3 rd order autoregressive structure and a variance configuration accounting for within-treatment variance.Statistically significant GAMM models were compared by their coefficient of correlation (R 2 ).The model with the highest R 2 was chosen as the one describing the response best.With this approach, CO 2 effects were detected by determination of the most appropriate GAMM model (model 1 = no CO 2 effect; model 2,3 = CO 2 effect detected).Importantly, model 2 was in no case gaining the highest R 2 value.Thus, when a CO 2 effect was detected, it was always an effect on phenology (model 3).The response scenario (Fig 1) was determined in case of a detected CO 2 effect by visually inspecting the phenology of the GAMM model fits shown in S2 Fig.

Basic chemical parameters and the different phases of phytoplankton blooms in control and high CO 2 mesocosms
The partial pressure of CO 2 was elevated to ~1000 μatm in the high CO 2 mesocosms during the first six days of the study.CO 2 -outgassing at the air-sea interface in the high CO 2 mesocosms was countered by regular additions of CO 2 -aerated water while pCO 2 was not manipulated in the control treatment.The pCO 2 levels averaged over the entire experimental period were 759 (±11) and 384 (±19) μatm in the high CO 2 and control environments, respectively [12].
Inorganic nutrients were up-welled by winter mixing before the study started and enclosed with comparable concentrations in all mesocosm bags when isolating the water from the surrounding fjord (S3 Table; see also [12]).The subsequent temporal development of nutrient and Si(OH) 4 concentrations were also quite low at the first chla peak (0.07 ±0.07 and 2.67 ± 0.30 μmol kg -1 on day 33, respectively) but those nutrients were not the primarily limiting ones after the spring bloom [12].Si(OH) 4 remained above detection limit for some days to weeks after day 33 before reaching detection limit.PO 4 3-fluctuated at a very low level (max. 0.2 μmol kg -1 ) from day 33 until the end of the experiment [12].The average inorganic nutrient concentrations (NO ) for each of the four phases are provided in S3 Table .A graphical representation of the inorganic nutrient dataset is provided in the overview paper [12] accompanying this study.
The temporal development of Chla was similar in control and high CO 2 mesocosms (Fig 2).Chla concentrations were initially low (~0.3 μg L -1 ) and showed a relatively slow increase until day 17.Afterwards, chla started to increase rapidly until reaching the first of two major peaks around day 33 (Fig 2).The first bloom declined after day ~33 with chla concentrations dropping to ~1.5 μg L -1 on day ~40.This temporal minimum of chla also marks the initiation of the second phytoplankton bloom which peaked around day 55 ( Fig 2).Peak chla concentrations were on average only slightly lower than in the first bloom with peak1/peak2 chla ratios ranging from 1.5 (M10) to 0.9 (M3) [12].The second chla peak declined more slowly and reached baseline values around day 77 (Fig 2).After day 77, chla concentrations remained at low concentrations (~0.5 μg L -1 ) and no further chla peak developed (Fig 2).
Based on the observed chla development we divided the experiment in 4 major phases (Fig 2).Phase I is the time before the first major chla build-up and characterized by relatively low chla (day -2-16).Phase II comprises the build-up and decline of the first major chla peak (day .Phase III includes the second major chla peak (day 41-77).Phase IV is the post-bloom period where chla was relatively low and fairly stable (day 78-111).

Succession of functional and taxonomic phytoplankton groups in control and high CO 2 mesocosms
The plankton community composition was similar among mesocosms at the beginning of the study (see [12] for a detailed analysis).Likewise, the succession of phytoplankton groups was similar in all mesocosms so that the following description of the temporal development refers to both the control and the high CO 2 treatment.
Initially, picoeukaryotes were abundant (Fig 3) and contributed about half of the total chla concentration (Figs 4 and 5).Their abundance started declining, however, around day 10 (Fig 3B and 3J) which is also reflected in a slight decline in chla (Fig 2).The major spring bloom forming groups distinguished by flow cytometry and filter counts were Nano I-IV, Crypto, and C. concinnus.Their abundances were very low at the beginning but they grew exponentially from the first days until the peak of the bloom around day 33.This is difficult to see on linear cell abundance plots (Fig 3A -3H), but becomes clearly visible when using a logarithmized y-axis (Fig 3I -3P).HPLC pigment measurements and CHEMTAX analysis revealed    The tendency towards a more diverse phytoplankton community continued in phase IV (the post bloom phase) where Prasino-, Crypto-, Cyano-, Chloro-, Prymnesio-, and Dinophyceae became more important (Fig 5).Notably, auto-and/or mixotrophic Dinophyceae were quasi absent during the two diatom-dominated blooms but emerged quickly thereafter (phase IV, Fig 4).The marked increases of Cyano-and Cryptophyceae in phase IV that was revealed by the CHEMTAX analysis was reflected in the increase of Synechococcus (a Cyanophyceae genus) and Crypto groups measured with the flow cytometer (compare Figs 3A and 4D).The high consistency among both independent methods increases the confidence in our results.

CO 2 effects on the phytoplankton community
CO 2 significantly influenced the development of chla (Table 1), however not consistently.An effect of CO 2 was absent during the first chla peak in phase II but clearly identifiable during the second bloom in phase III.Here, chla build-up was significantly amplified under high CO 2 conditions (Fig 2 ).A shift in the timing (i.e.temporal occurrence) of chla peaks was not apparent.Thus, our results point towards a type A response of chla (increase in bloom amplitude; Fig 1A) during the second phytoplankton bloom in phase III.
The GAMM analyses revealed temporal CO 2 effects in 6 of the 8 taxonomic phytoplankton groups distinguished with CHEMTAX (Table 1).Diatom, Prasinophyceae, and Chlorophyceae biomass was significantly higher under high CO 2 (Table 1).The positive effect on diatoms occurred for a relatively short period during the second phytoplankton bloom in phase III, similar to bulk chla (compare Figs 2 and 4A).Prasinophyceae were stimulated during a minor peak in phase I and throughout phase III (Fig 4B).Chlorophyceae were close to detection limit during most of the experiment but showed a positive response to high CO 2 during a peak in phase IV (Fig 4E).Auto-and/or mixotrophic Dinoflagellates (Dinophyceae) experienced positive CO 2 effects during the end of phases II and IV.Prymnesiophyceae were impaired by high CO 2 from the end of phase II until the middle of phase IV. ( were negatively affected under high CO 2 during phase III and positively affected during phase IV (Table 1), although it must be recognized that the effects were very small and Cyanophyceae were close to detection limit during phase III (Fig 4D).
The GAMM analyses of flow cytometry and filter count data revealed significant temporal CO 2 effects on 5 out of 8 groups (Table 1) with the clearest CO 2 response observed for Peuks (Fig 3).Importantly, Peuk abundance was already significantly higher by about 9% (~1500 cells mL -1 ; t-test p = 0.0018) in the high CO 2 treatment on the first day of the experiment and thus already before the first CO 2 addition [12].The reason for this was a carry-over effect from a failed OA experiment we carried out in the same mesocosms before our successful experiment started (see section 2.1).In this previous experiment we already observed a positive CO 2 effect on picoeukaryotes [12].However, due to technical problems we had to finish this experiment, lower the mesocosms below surface and dismount the sediment traps until we could restart four days later (section 2.1; [12]).In the four days in between the two studies, water exchange with the fjord was almost but not entirely complete so that some of the CO 2 -induced picoeukaryote signal was transferred into the successful second experiment that is described in the present paper.The small initial difference was lost at the end of phase I (day 17) where abundances in the control and the high CO 2 treatment were insignificantly different.The first CO 2 effect on picoeukaryotes that developed during the second experiment started to appear right at the peak of the first chla bloom (day ~33).At this time, Peuk net growth was positive under high CO 2 and slightly negative in the control mesocosms (Fig 3B and 3J).Opposite net growth rates were observed until day ~47 and generated an offset in Peuk abundance between control and high CO 2 mesocosms which prevailed until the end of phase III (Fig 3B and 3J).A second divergence in Peuk abundance occurred during phase IV where they bloomed under high CO 2 (~38,000 cells mL -1 on day 91) but not in the control (Fig 3B and 3J).
The abundance of C. concinnus was significantly elevated under high CO 2 , mainly during the second phytoplankton bloom in phase III (Fig 3H and 3P).Synechoccocus abundance was slightly lower under high CO 2 during phase III and marginally higher during a short period in phase IV (Fig 3I , Table 1).Nano IV abundance was lower under high CO 2 during the beginning of phase IV but the effect was very small (Fig 3O).The detected CO 2 effect on Nano I (Table 1) must be regarded carefully.Here, short consecutive abundance peaks constrained the generation of adequate GAMM fits (S2 Fig) so that a reliable determination of CO 2 effects was not possible.

The potential influence of ocean acidification on phytoplankton blooms
We observed no detectable CO 2 effect on chla during the first bloom in phase II (Fig 2 ) where phytoplankton utilized inorganic nutrients that were initially available from winter upwelling.This outcome is consistent with results from the majority of previous mesocosm OA experiments under nutrient replete conditions.So far, ten studies with mesocosm volumes ! 100 L reported no response of CO 2 on maximum chla build-up [28][29][30][31][32][33][34][35][36][37], while only five detected either positive [35,[37][38][39] or negative [40] impacts.
A positive CO 2 effect on chla build-up was observed during the second bloom in phase III.The CO 2 effect did not appear to be particularly pronounced (Fig 2) but this may not reflect the actual chla difference appropriately because we also observed significantly higher mesozooplankton biomass under high CO 2 during this period [20,27].Thus, part of the chla difference may have been grazed off.
Inorganic nutrient concentrations were close to detection limit during the second bloom so that the bloom was fueled by other nutrient sources.Published results on mesocosm OA experiments conducted under inorganic nutrient deplete conditions are less numerous so far, making it even more difficult to reveal a general response pattern.The seven mesocosm experiments we are aware of (volume ! 100 L; published until June 2017) either observed a stimulation of chla concentrations under high CO 2 [35,41] or reported no effect [37,39,42].Accordingly, the chla response pattern observed in our study aligns reasonably well with the general tendency currently taking form in the literature-i.e.rather no chla response to OA under nutrient replete conditions and perhaps a slight tendency towards a positive response when inorganic nutrients are low [42].However, more data and thorough meta-analyses that consider the individual features of experiments are needed to confirm or disprove this impression.

CO 2 effects in the phytoplankton community
CO 2 effects on individual phytoplankton groups were identified in 10 out of 16 parameters analyzed with GAMM (Table 1; Nano I was not considered here).As for chla, most of the detected effects were present only during certain stages of the succession and effect sizes appeared to be small in most cases (Figs 3 and 4).Uncovering the origin of group-specific CO 2 responses is challenging in ecologically realistic experiments because the multitude of unconstrained factors allows for a multitude of potential explanations.For example, changing CO 2 or pH can lead to a direct (i.e.physiological) response of the investigated taxon with direct consequences for its competitiveness within the natural community.In this case, results from physiological laboratory investigations can be used to explain certain patterns.However, observed responses can equally well be evoked indirectly, via CO 2 effects on other players in the food web which influence the investigated taxon through trophic cascades.Indirect effects are considered to be very important but hard to prove as they require a comprehensive understanding of the various interactions in the food web [43].In the following, we aim to present what we consider to be the most likely explanations for the observed CO 2 responses in some of the investigated phytoplankton groups.We would like to emphasize, however, that explanations different to the ones provided here are possible in each case.
4.2.1 Diatoms.Diatoms were dominating the phytoplankton community and their temporal development is largely identical to the development of bulk chla.(compare Figs 2 and  4A) This suggests that the positive CO 2 effect on chla during phase III is primarily a positive CO 2 effect on diatoms.Diatoms were represented by very small species (~2-8 μm) such as Minidiscus sp. or Arcocellulus sp. and by the large species C. concinnus (>200 μm).The abundance of C. concinnus was significantly higher in the high CO 2 treatment by about 70 cells L -1 during the peak of the second bloom in phase III (average between days 45 to 55 of 255 and 324 cells mL -1 in the control and the high CO 2 mesocosms, respectively).To approximate the relevance of this difference in terms of chla, we measured chla content of C. concinnus cells on days 45 and 49, multiplied the average chla cell -1 value with measured cell numbers, and compared this to bulk chla concentrations.Based on that, C. concinnus contributed about 50% (ranging from 36% in M5 to 66% in M4) to the bulk chla concentration during phase III.The difference of ~70 cells L -1 between control and high CO 2 explains about half (i.e.0.5 μg L -1 ) of the CO 2 -induced difference in bulk chla (i.e.~1 μg L -1 ; Fig 2).The CHEMTAX diatom trend suggests, however, that the entire 1 μg L -1 difference is due to differences in diatom biomass (compare Figs 2 and 4A).Thus, the remaining 0.5 μg L -1 must have been due to biomass differences in the small diatom species like e.g.Arcocellulus sp.(Fig 6).Unfortunately, there is no biomass data on any of the small diatoms available but due to their approximate size they must have been included in the Nano I and/or Nano II populations quantified with the flow cytometry.Here, we do not find any CO 2 -related differences (Fig 3) meaning that the CHEMTAX and the flow cytometry data are conflicting in this particular case.We found no explanation for this other than uncertainties in the associated measurements and the abovementioned biomass estimation of C. concinnus.
The elevated C. concinnus abundances observed in the high CO 2 treatment occurred during phase III where inorganic nutrients were depleted.This is consistent with results from a recent laboratory study where the diatoms Thalassiosira weissflogii and Dactyliosolen fragilissimus also reached higher population densities under high CO 2 when nutrients were exhausted [44].The authors hypothesized that less resources were necessary for inorganic carbon acquisition under high CO 2 thereby allocating resources to growth which leads to higher population densities [44].Interestingly, CO 2 stimulation was shown to be much more pronounced in larger diatoms as these are considered to be more diffusion limited [44][45][46].This may explain why we found a clear positive CO 2 effect in the large (i.e.>200 μm) diatom C. concinnus.
The line of reasoning presented above points towards a direct (i.e.physiological) effect of CO 2 on the growth of C. concinnus.An indirect effect through food web interactions seems less likely, also because C. concinnus was too large to be grazed by any of the present zooplankton species [20] including herring larvae where no C. concinnus was found in the gut content.Thus, our results support the hypothesis that large diatom genera like Coscinodiscus could become more competitive in an acidified ocean under nutrient deplete conditions through facilitated inorganic carbon acquisition [44,45].In contrast, our observations on small diatoms are inconclusive, mainly because our data is not resolved with the necessary detail on diatom community structure in the small size range.

Dinoflagellates.
Dinoflagellates are a diverse group of protists which acquire energy through photo-or heterotrophy or a combination of both known as mixotrophy [47].Here, we determined dinoflagellate contribution to chla with CHEMTAX and therefore only considered photosynthesizing species with a pigment setup characteristic for Dinophyceae [16].This excludes heterotrophic species and mixotrophic ones which acquire plastids from other phytoplankton taxa (e.g.Dinophysis which sequesters cryptophyceae chloroplasts from its prey [47]).Dinophyceae were growing early in the experiment but started to decline at the beginning of phase II in the control mesocosms.High CO 2 did not affect the maximum biomass but delayed their decline by two weeks (Fig 4F).Based on microscopy counts we identified Heterocapsa triquetra as the most likely species responsible for the observed trends in the CHEMTAX data during phase I and II since it was the only dinoflagellate species found in noticeable quantities during this time.H. triquetra is primarily phototrophic but can apply phagotrophy under nutrient-limiting conditions to acquire nitrogen and phosphorous [48].Culture experiments suggested that the growth rate of H. triquetra is unaffected by pH in the range between 8.7-7.5 [49].This argues against a direct CO 2 effect on H. triquetra growth rate and points towards an indirect effect, for example through reduced grazing pressure under high CO 2 during phase II.
Dinophyceae were not detected for most of phase III but started to increase again during phase IV.They reached higher biomass in the high CO 2 treatment towards the end of the experiment (Fig 4F).Horn et al., investigated dinoflagellate abundance in the same mesocosm study by means of light microscopy and found the same CO 2 trend in phase IV [50].In their analysis they focused on species which are traditionally considered as heterotrophic although still aware that many species are at least facultative mixotrophic [50].The CO 2 effect detected by Horn et al. was caused by elevated abundances of athecate dinoflagellates (<30-55 μm) [50] represented primarily by Gyrodinium and/or Gymnodinium sp.(H.Horn, pers.comm.).The authors hypothesized that the positive CO 2 effect on these mixotrophic species was caused by increased availability of picoeukaryote (Peuk) prey [50].Our data supports this hypothesis since Peuk abundance was indeed elevated under high CO 2 before the onset of the second dinoflagellate bloom and then rapidly declined to very low numbers when Dinophyceae started to grow (Fig 4).The elevated availability of picoeukaryotic prey under high CO 2 may have enabled Dinophyceae to reach higher biomass on the last days of the experiment.

Prymnesiophyceae.
Prymnesiophyceae had a minor contribution to total chla (Fig 5).Their biomass peaked in the aftermath of the two major phytoplankton blooms and was lower in the high CO 2 treatment throughout almost the entire experiment (Fig 4H ).A recent synthesis of OA studies with natural plankton communities found a consistently negative CO 2 effect on Prymnesiophyceae (aka Haptophyceae) biomass with only few exceptions [7].Negative effects were often driven by calcifying Prymnesiophyceae (coccolithophores) [7], which are known to be sensitive to low pH [51,52].However, non-calcifying genera like Phaeocystis or Chrysochromulina also responded negatively to increasing CO 2 [7].Unfortunately, we were unable to identify the species or species assemblage causing the negative CO 2 response in our study but the high consistency among the various mesocosm experiments with taxonomically very different Prymnesiophyceae species points towards a physiological carbonate chemistry sensitivity that is rooted in the core physiological apparatus of this taxon.An (indirect) CO 2 effect on Prymnesiophyceae through food web interactions seems rather unlikely because in this case we would have expected a more variable response among previous studies and also a less consistent negative CO 2 effect in our "long-term" study.Collectively, the evidence from multiple experiments suggests that Prymnesiophyceae face the risk of playing a less important role in plankton communities in an acidified ocean.

Picocyanobacteria (Cyanophyceae).
Picocyanobacteria were present throughout the entire study although they played a minor role in terms of biomass and occurred in high abundances only at the end of the experiment (Figs 3A and 5).They were represented most likely by the genus Synechococcus and not Prochlorococcus because the latter is not occurring above 40˚N [53] and its marker pigments (divinyl chlorophyll a and b [54]) were not detected.The temporal development of Synechococcus counted with the flow cytometer, and Cyanophyceae, determined with CHEMTAX agree well with each other (compare Figs 3A and 4D) suggesting that Synechococcus was the only cyanobacterium genus present in noticeable amounts.CO 2 had a weakly negative effect on its abundance (and Cyanophyceae biomass) during phase III and a marginally positive one during phase IV (Figs 3A and 4D).Previous experiments with pelagic communities revealed variable responses of Synechococcus abundances to simulated OA (positive, negative, neutral) which was attributed to the enormous cryptic diversity of this genus [7,55,56].
Alternatively, indirect CO 2 effects could explain their variable responses.In our experiment, the negative CO 2 effect manifested shortly after inorganic nutrients were exhausted (~day 33) and the major spring bloom was on the decline (Fig 2 ; phase II).We observed no significant CO 2 effect on predominant microzooplanktonic grazers such as ciliates and heterotrophic dinoflagellates during this period [50] but detected a positive effect on picoeukaryotes appearing precisely when Synechococcus responded negatively to CO 2 (Fig 3).Indeed, picoeukaryote genera like Micromonas can be mixotrophic and feed on spherical particles with a size of at least 0.9 μm in diameter [57][58][59].Thus, enhanced grazing on Synechococcus by picoeukaryotes under high CO 2 could potentially explain their negative CO 2 response during phase III (mixotrophy of picoeukaryotes is discussed further in section 4.2.5;please note that Dinophyceae, represented by the mixotroph H. triquetra (section 4.2.2), also respond positively to CO 2 during this time but this species does not feed on Synechococcus [48]).A similar antagonistic CO 2 response between picoeukaryotes and Synechococcus has also been observed in a previous mesocosm study in Raunefjord (Norway) [60].Here, Paulino et al. speculated that picoeukaryotes were better nutrient competitors under high CO 2 relative to Synechococcus [60].Findings by Paulino et al. are in contrast to the findings by Schulz et al. who observed a synergistic response of picoeukaryotes and Synechococcus in a follow-up OA mesocosm experiment at the Raunefjord study site (both Synechococcus and picoeukaryote abundance was stimulated by high CO 2 [7]).However, in the case of the Schulz et al study, picoeukaryotes were dominated by Chlorophyceae whereas Paulino et al. argue that they were dominated by Prasinophyceae (Micromonas) in their particular study [7,60].In accordance with both studies, we observed an antagonistic response during phase III where picoeukaryotes were dominated by Prasinophyceae whereas a synergistic response occurred during phase IV where picoeukaryotes were predominantly composed of Chlorophyceae (see section 4.2.5).Thus, CO 2 effects on Synechococcus may be coupled to the taxonomic composition and the trophic interactions with their picoeukaryotic competitors.4B and 4E; see also section 4.2.4).This is in line with previous studies who also determined Prasino-and Chlorophyceae as the predominant picoeukaryotes [7].
The small but significant difference in Peuk abundance between control and high CO 2 at the first day was a remnant of a preceding CO 2 experiment (see section 3.3).We cannot fully exclude that this initial difference was also causing the differences observed later in the experiment but several reasons make this unlikely.Most importantly, the difference was small (1500 cells mL -1 ; less than 9% of the population) and could be equalized quickly under the assumption of realistic picoeukaryote growth rates [12].Indeed, mean Peuk abundances equalized between control and high CO 2 already quite early in the experiment and it lasted more than two weeks until deviating Peuk abundances between control and high CO 2 treatment reestablished (day ~33; Fig 3B and 3J).We would have expected a continuous offset between control and high CO 2 rather than a reoccurring one in case the initial difference was responsible for the deviating trends later in the experiment.Furthermore, Peuks belonged primarily to the Prasinophyceae class at the beginning of the study while a large fraction belonged to the Chlorophyceae at a later stage.It seems rather unlikely that an initial difference in one class triggered the same response in another class later in the experiment.To conclude this line of arguments we would like to point out the following: Even in the unlikely case that the positive CO 2 responses of Peuks observed in this study were triggered by the small initial difference, our interpretations would still be valid.This is because the initial difference itself is not a coincidence but the result of a positive CO 2 effect on Peuks occurring in the preceding experiment which was stopped due to technical problems (section 2.1; [12]).
Stimulation of phytoplankton growth rate and abundance by elevated levels of CO 2 has frequently been observed in cell cultures and natural assemblages [61].The phenomenon is typically explained by a CO 2 fertilization of the often rate-limited carbon fixing enzyme Rubisco [62].This straight-forward hypothesis may also be true for picoeukaryotes where in-vitro experiments documented accelerated growth rates of important picoeukaryote genera like Ostreococcus and Micromonas under above ambient pCO 2 (i.e.~500-1000 μatm; [63,64]).It is surprising, however, that we found a particularly pronounced CO 2 -stimulation on abundance in the smallest eukaryotic phytoplankton group.In theory, we would expect that primarily larger species like C. concinnus benefit more from high CO 2 because they are more diffusionlimited due to their lower surface to volume (S/V) ratio (section 4.2.1;[46,65]).This counterintuitive result indicates that additional (or complementary) mechanisms may have determined the specific stimulation of Peuks.We propose three of these mechanisms in the following.All three are related to nutrient acquisition when inorganic nutrients are limiting.).These conditions primarily select for phytoplankton with high abilities to gather nutrients from the environment [66].In general, smaller phytoplankton groups are considered to be more capable in nutrient acquisition than larger ones due to their relatively high S/V ratio [67].Thus, Peuks may be the phytoplankton group who could capitalize best on the CO 2 fertilization of photosynthesis under nutrient-limiting conditions since they were superior nutrient competitors.
2. The second mechanism follows the same underlying logic as described in the first one but takes a pH dependency of remineralization rather than a CO 2 dependency of photosynthesis into account.Previous OA studies with auto-and heterotrophic bacteria reported accelerated rates of extracellular enzymes involved in organic matter remineralization under low pH [68][69][70][71][72].If the same pH dependency also applies for eukaryotic phytoplankton, they should have an advantage in the extraction of nutrients from organic sources under acidified conditions.This would be once more particularly beneficial for picoeukaryotes due to their increased S/V ratio relative to larger species.
3. The third mechanism we propose is related to the mixotrophic abilities of picoeukaryotes.
Recent field studies revealed the potential of photosynthesizing picoeukaryotes to satisfy part of their nutrient requirements through phagocytosis of bacteria in oligotrophic regimes [58,59,73].CO 2 fertilization of photosynthesis in the OA treatment may have raised the nutrient requirements of Peuks and therefore stimulated grazing on bacteria.This hypothesis is supported by the abundance ratio of Peuks to heterotrophic bacteria which was significantly elevated under high CO 2 during the picoeukaryote bloom under inorganic nutrient deplete conditions in phase III (S3 Fig).This antagonistic pattern was also observed in a previous mesocosm study in oligotrophic post-bloom conditions [74].It has been hypothesized that bacterivory by mixotrophs can serve as alternative nutrient source when inorganic nutrients are limiting and simultaneously weaken heterotrophic bacteria as nutrient competitors [75].Indeed, laboratory experiments with mixotrophic phytoplankton showed that photosynthesizing cells can adjust phagotrophic rates to changing nutrient concentration and/or light intensity in order to sustain optimal nutrient supply [48,76].Thus, it is possible that such an adjustment of phagotrophic rates also occurs when the nutrient demand of Peuks is altered by changing carbonate chemistry.
Another noticeable positive CO 2 effect on Peuk abundance was observed later in the experiment during phase IV (around day 90; Fig 3J).This Peuk bloom was different to the previous ones in that it was dominated by Chlorophyceae and not Prasinophyceae.Accordingly, the positive CO 2 effect on picoeukaryotes seems to be related to their size and their role in the food web rather than on their taxonomic classification.
The positive effect of end-of-the-century CO 2 partial pressures on picoeukaryote abundance is a strikingly consistent result in ocean acidification studies with plankton communities [7].This has been shown from eutrophic to oligotrophic regimes [41,60], from high to lower latitudes [35,37,77], from winter to summer [this study], and from marine to freshwater environments [78,79].Other climate change related consequences such as ocean warming, freshening, and enhanced stratification also seem to favor picoeukaryotes [80][81][82][83].Thus, multiple Fig 1A); by inducing a shift in peak timing (Fig 1B); by a changing peak amplitude and timing (Fig 1C); by changing the entire pattern of the response curve (Fig 1D).

Fig 1 .
Fig 1.Four potential scenarios how phytoplankton bloom development could be altered by ocean acidification explained with the example of chla concentration.Blue and red lines illustrate control and "treatment", respectively.(A) Change in bloom amplitude.(B) Change in bloom timing.(C) Change in bloom amplitude and timing.(D) Change in bloom pattern.https://doi.org/10.1371/journal.pone.0188198.g001

Fig 2 .
Fig 2. Chla development over time.Red and blue lines display the average of five high and five ambient CO 2 mesocosms, respectively.Shaded areas represent standard deviations from means.Vertical grey lines (Roman numbers I to IV) separate the four experimental phases.https://doi.org/10.1371/journal.pone.0188198.g002

Fig 3 .
Fig 3. Development of phytoplankton groups quantified by flow cytometry and filter counts.Red and blue lines display the average of five high and five ambient CO 2 mesocosms, respectively.Shaded areas represent standard deviations from means.Data are displayed on linear (A-H) and logarithmic y-axis (I-P).Note: the exponent in A-H after a group name needs to be multiplied with the y-axis numbering (e.g. 5 Syn x 10 4 !50000 Synechoccocus cells mL -1 ).Vertical grey lines (Roman numbers I to IV) separate the four experimental phases.https://doi.org/10.1371/journal.pone.0188198.g003

Fig 4 .
Fig 4. Development of phytoplankton classes based on CHEMTAX pigment taxonomy.Red and blue lines show the average of five high and five ambient CO 2 mesocosms, respectively.Shaded areas represent standard deviations from means.The y-axis shows the amount of chla contributed by each class.Vertical grey lines (Roman numbers I to IV) separate the four experimental phases.https://doi.org/10.1371/journal.pone.0188198.g004

4 . 2 . 5
Picoeukaryotes (Prasinophyceae and Chlorophyceae).The abundance of picoeukaryotes (Peuks) was positively affected by high CO 2 at different stages of the winter-to-summer succession.The Peuk clusters determined by means of flow cytometry were most likely dominated by Prasino-and Chlorophyceae as the combined pattern closely resembles the Peuks trend over time (compare Figs 3B with

1 .
The largest differences in Peuk net growth between control and high CO 2 were observed after day 33 when NO 3 -+NO 2 -concentrations were close to detection limit (days ~33-47 and ~80-90 in Fig 3J; nutrient concentrations are shown in S3 Table

Table 1 . Summary of statistical results.
The temporal development of phytoplankton was analyzed by means of GAMM.A CO 2 effect was detected when the GAMM model with the best fit (highest R 2 value) accounted for a CO 2 dependency of the phenology.In the case of Nano I a CO 2 effect was detected by the GAMM analysis but not considered further due to an unsatisfactory model fit.