Enhanced transfer of organic matter to higher trophic levels caused by ocean acidification and its implications for export production: A mass balance approach

Ongoing acidification of the ocean through uptake of anthropogenic CO2 is known to affect marine biota and ecosystems with largely unknown consequences for marine food webs. Changes in food web structure have the potential to alter trophic transfer, partitioning, and biogeochemical cycling of elements in the ocean. Here we investigated the impact of realistic end-of-the-century CO2 concentrations on the development and partitioning of the carbon, nitrogen, phosphorus, and silica pools in a coastal pelagic ecosystem (Gullmar Fjord, Sweden). We covered the entire winter-to-summer plankton succession (100 days) in two sets of five pelagic mesocosms, with one set being CO2 enriched (~760 μatm pCO2) and the other one left at ambient CO2 concentrations. Elemental mass balances were calculated and we highlight important challenges and uncertainties we have faced in the closed mesocosm system. Our key observations under high CO2 were: (1) A significantly amplified transfer of carbon, nitrogen, and phosphorus from primary producers to higher trophic levels, during times of regenerated primary production. (2) A prolonged retention of all three elements in the pelagic food web that significantly reduced nitrogen and phosphorus sedimentation by about 11 and 9%, respectively. (3) A positive trend in carbon fixation (relative to nitrogen) that appeared in the particulate matter pool as well as the downward particle flux. This excess carbon counteracted a potential reduction in carbon sedimentation that could have been expected from patterns of nitrogen and phosphorus fluxes. Our findings highlight the potential for ocean acidification to alter partitioning and cycling of carbon and nutrients in the surface ocean but also show that impacts are temporarily variable and likely depending upon the structure of the plankton food web.


Introduction
The ocean is a major sink for anthropogenic carbon dioxide (CO 2 ) by absorbing more than 2 Pg carbon per year from the atmosphere [1,2]. This uptake of atmospheric CO 2 leads to both carbonation (increasing CO 2 concentration) and acidification (decreasing seawater pH) of the surface ocean [3,4]. Changes of both environmental factors are expected to impact marine biota from the organism [5] to the ecosystem level [6,7]. Phytoplankton groups belonging to the picoeukaryotes will likely benefit from increased inorganic carbon availability [8,9], while calcifying phyto-and zooplankton groups such as coccolithophores or pteropods will likely be impaired by decreasing seawater pH and changes in seawater carbonate chemistry [10,11]. Presumed shifts in plankton community composition, e.g. to smaller (medium-sized) phytoplankton organisms [12] with different elemental stoichiometry can modify marine element cycling [13][14][15]. Recent studies have further revealed the potential of CO 2 to alter the partitioning of carbon between dissolved and particulate organic matter pools in the euphotic ocean zone [16][17][18]. Increasing proportions of dissolved organic carbon can stimulate bacterial growth and recycling of organic matter [17,19,20], but are also known to promote particle formation and organic matter export by increasing particle stickiness [19,21]. While our knowledge about the impact of CO 2 on carbon cycling in the ocean is continuously growing, the potential effects on cycling of macronutrients (inorganic nitrogen, phosphorus, and silica) through changes in the marine food webs require more in-depth investigation. In fact, the partitioning of macronutrients between different pools and trophic levels determines their turnover rates and can thereby feedback on ecosystem structure and functioning [22]. For instance changes in stoichiometry and fatty acid composition of primary producers as a consequence of increasing CO 2 have already been shown to impact mesozooplankton reproduction and development [23,24]. This implies direct consequences for element cycling within the ocean's food webs.
Calculating the mass balance of carbon and macronutrients is one of the best approaches to estimate their partitioning and cycling. However, such approaches are prone to high uncertainties in open ocean regions. Availability of essential parameters (e.g. gas exchange of CO 2 or sinking particulate matter) is often limited, while vertical mixing and lateral advection permanently exchange the investigated water masses. Pelagic mesocosms have the advantage of isolating a water mass from the surrounding ocean and hence allow us to investigate natural plankton assemblages of several trophic levels at close to natural conditions. The enclosed water bodies can be characterised with respect to element pools and plankton community, while repetitive sampling of the same water parcel allows monitoring of changes over long timescales and successive phases of plankton development.
Here we present results from a pelagic in situ mesocosm CO 2 perturbation study in Gullmar Fjord (Sweden) covering the full winter-to-summer plankton succession typical for the coastal sea in mid-latitudes. The mid-latitude regions are of particular importance to global element cycling due to the annual formation of large phytoplankton spring blooms characterised by high export efficiency [25]. We monitored the enclosed plankton communities (from viruses to fish larvae) over more than 100 days in two sets of five mesocosms representing ambient and projected year 2100 pCO 2 (partial pressure of CO 2 ), respectively [26]. Element pools of carbon, nitrogen, phosphorus, and silica (C, N, P, and Si) were measured to compute mass balances and estimates of net community production, thereby assessing the impact of ocean acidification on the partitioning and cycling of major elements within the ocean surface layer.

Mesocosm setup and maintenance
Ten "Kiel Off-Shore Mesocosms for Ocean Simulations" (KOSMOS; [27]) were deployed on January 29, 2013 in Gullmar Fjord on the west coast of Sweden (58.26635˚N, 11.47832˚E). Permission for the study location was granted by the Längsstyrelsen Västra Götalands Län (reference no: 258-39615-2012) and by the owner of the adjacent private property (Lysekil Skaftö 1:27). Sea ice drift and technical problems described in Bach et al. [26] delayed the start of the experiment until March 7 (day -2 = t -2 , i.e. 2 days before homogenization of the water column; see Sect. 2.2). Each cylindrical mesocosm bag (2 m diameter) enclosed a 17 m deep water column, sealed at the bottom end by a two meter long, funnel-shaped sediment trap (Fig 1A).
Enclosed nekton and large mesozooplankton (e.g. fish larvae or jelly fish) were removed during the initial period of the study by a full-diameter-size net (1 mm mesh) that was pulled through each mesocosm (t 6 ; Fig 2). Samples relevant for mass balancing of elements were taken over a period of more than 100 days until June 22 (t 105 ; Fig 2). Biofilm formation on the inner and outer walls of the cylindrical mesocosm bags was prevented by regular cleaning [26] (Fig 2). Settled material adhering to the inner surface of the sediment trap funnels was removed at the very end of the experiment (t 102 ). A detailed description of the study site, the initiation of the experiment, and mesocosm cleaning can be found in Bach et al. [26].

System manipulations and volume determination
The natural salinity gradient inside the mesocosms was removed by injecting air to the bottom of the enclosures in two stages (t -2 and t 0 ; Fig 2; [26]). A 'high CO 2 treatment' of initially 961 μatm pCO 2 (t 5 ) was established in five of the ten mesocosms (M2, M4, M6, M7, M8) by stepwise addition of CO 2 -saturated seawater (t -1 , t 0 , t 2 , t 4 ). The other five mesocosms served as untreated controls (M1, M3, M5, M9, M10), representing ambient CO 2 conditions. pCO 2 levels were re-adjusted four times in the high CO 2 mesocosms (t 17 , t 46 + t 48 , t 68 , t 88 ; Fig 2) to counteract the loss from outgassing and biological uptake.
Seed populations of organisms from the surrounding fjord were introduced to the mesocosms by adding 22 L of fjord water on every fourth day (Fig 2). In total, the regular fjord water additions summed up to about 1% of the mesocosms' volume [26]. In early May, we introduced green sea-urchin larvae (Strongylocentrotus droebachiensis [Müller, 1776]; t 56 ) and herring eggs (Clupea harengus [Linnaeus, 1758]; t 48 ), that hatched two weeks later on t 62 . Animal welfare of herring larvae was assured according to guidelines of the ethics permit (no 332-2012). The species C. harengus is not endangered and specimens were anaesthetised with MS-222 before handling and fixation to reduce stress to a minimum.
The volume of each mesocosm was determined by adding a known amount of calibrated sodium chloride brine solution and by measuring the salinity increase as described in Czerny et al. [28]. The brine solution was evenly dispersed inside the mesocosms on April 24 (t 46 ), elevating salinity by about 0.1 units from on average 29.2 to 29.3. Mesocosm volumes were converted from kilograms of seawater to litres using individual seawater density of each mesocosm on t 46 .

Sampling procedures and CTD operations
The mesocosm water columns and sediment traps were sampled every second day starting at t -1 with the exception of one additional inorganic nutrient sampling on t 2 (Fig 2). The sediment traps at 19 m water depth were emptied with a vacuum system following Boxhammer et al. [29]. Water column samples were taken with depth-integrating water samplers (IWS, Hydro-Bios) which collected equal amounts of water from all depth levels between 0 and 17 m. Samples sensitive for contamination or gas exchange such as inorganic nutrients (including dissolved inorganic nitrogen (DIN = nitrate (NO 3 -) + nitrite (NO 2 -) + ammonium (NH 4 + )), phosphorus (DIP = phosphate (PO 4 3-)) and silica (DSi = Si(OH) 4 )), dissolved organic matter (DOM; DOC (carbon), DON (nitrogen), DOP (phosphorus)) and carbonate chemistry samples (dissolved inorganic carbon (DIC), pH) were directly transferred from the IWS samplers into corresponding sample bottles. DOC/DON samples were gravity filtered through glass fibre filters (pore size 0.7 μm, Whatman) during transfer into pre-combusted glass vials on board of the sampling boats and acidified in the lab (HCl, 25%, analysis grade, Carl Roth) to pH 2 as described in Zark et al. [30]. DOP samples were collected in acid-rinsed polycarbonate bottles (Nalgene) and filtered in the lab trough 0.7 μm (GF/F, Whatman) into low-density polyethylene vials (LDPE, Roth) using gentle vacuum filtration (<200 mbar). Until t 55 DOP samples were only collected on 12 out of 30 sampling days (see Fig 2) and were poisoned with mercury chloride following Kattner [31]. DOP samples collected after t 55 were taken alongside the 48 hours sampling routine (apart from t 83 -t 87 ) and stored frozen at -20˚C.
Carbonate chemistry samples were taken as described in Bach et al. [26] and sterile-filtered (0.2 μm) for a maximum of three days storage (dark and cold) before analysis.
Particulate matter (PM) was sampled from seawater pooled in 10 L carboys that were subsampled within a few hours at in-situ water temperatures. Water from these carboys was used for analysis of biogenic silica (BSi), total particulate carbon (TPC), nitrogen (TPN), and phosphorus (TPP), as well as Chlorophyll a (Chla) concentrations.
Mesozooplankton was collected with an Apstein net (55 μm mesh size, 17 cm diameter opening) by vertical net hauls (17 to 0 m water depth), representing a sampled volume of about 385 L. We restricted the sampling frequency to every eighth day to minimize the impact on the mesozooplankton community (Fig 2). A subsample of 4% was used for high-resolution plankton imaging with the ZooScan method (see Sect. 2.4.6), while the majority of the sample was preserved with sodium tetraborate-buffered formalin (4% v/v) for taxonomic abundance analyses [32]. CTD casts, providing salinity and temperature profiles, were performed with a CTD60M (Sea & Sun Technology) on every sampling day between 11 a.m. and 3 p.m. (local time ; Fig 2), covering a water depth from 0.3 to 18 m.

Sample analysis
2.4.1 Carbonate chemistry measurements and calculations. DIC was determined by colorimetric titration following Johnson et al. [33], with an estimated precision of 3 μmol kg -1 (standard deviation of duplicate measurements). Measurement accuracy was ensured by calibration against certified reference materials (CRM, supplied by A. Dickson, Scripps Institution of Oceanography, USA). pH T (total scale) was determined spectrophotometrically, based on the absorption ratio of the sulphonephthalien dye, m-cresol purple [34], with a precision of 0.002 pH units (SD of duplicates), while the accuracy was set by the equilibrium constants of the indicator. pCO 2 was calculated from the combination of pH T and DIC using CO2SYS [35] with the carbonate dissociation constants (K 1 and K 2 ) of Lueker et al. [36]. The input data included salinity, temperature, and inorganic nutrient concentrations (PO 4 3and Si(OH) 4 ). , and Si(OH) 4 ,) were filtered as triplicates through 0.45 μm cellulose acetate syringe filters (Whatman) before measuring them with a QuAAtro AutoAnalyzer (Seal Analytical) as described in Bach et al. [26]. When concentrations of NO 3 -+ NO 2 and PO 4 3dropped below 0.1 μmol L -1 (t 37 and t 35 , respectively), we switched to using the nanomolar system described by Patey et al. [37]. NH 4 + concentrations were determined according to Holmes et al. [38].

Inorganic nutrient measurements.
Inorganic nutrient measurements were stopped after t 95 as concentrations were close to or below their detection limits.

DOM measurements.
Concentrations of DOC and total dissolved nitrogen (TDN) were analysed of duplicate samples using high-temperature catalytic oxidation on a Shimadzu TOC-VCPH/CPN Total Organic Carbon Analyser, equipped with an ASI-V autosampler and a TNM-1 module for TDN determination as described in Zark et al. [30]. Samples with concentrations of DOC and TDN exceeding the measurement of their duplicate by 30% or more were considered being contaminated and were excluded from the dataset. Measurements from the high and ambient CO 2 mesocosms were subsequently pooled for identification and removal of outliers using the Dixon-Dean test (

Particulate matter and Chlorophyll a measurements.
Size fractions of PM smaller and greater than 200 μm (separated with a 200 μm mesh) were collected using gentle vacuum filtration ( 200 mbar) on pre-combusted (6 h at 450˚C) glass fibre filters (GF/F, 0.7 μm pore size, Whatman) or cellulose acetate filters (0.65 μm, Whatman) for analysis of TPC, TPN, TPP or BSi, respectively. The thereby collected PM included phytoplankton, small and abundant zooplankton as well as detritus, but essentially no mesozooplankton, such as copepodites or adult copepods (see also Sect. 2.4.6). Glass fibre filters were stored at -20˚C in pre-combusted (6 h at 450˚C) glass petri dishes until analysis, while cellulose acetate filters were also frozen at -20˚C but stored in plastic petri dishes. TPC/TPN filters were oven-dried over night at 60˚C, packed in tin foil and analysed alongside blank filters on an acetanilide calibrated CN analyser following Sharp [39]. We refrained from acidifying the filters to remove inorganic C, as pelagic calcifying organisms were very low in abundance. Accordingly, all particulate C data are presented as TPC but are assumed to represent particulate organic carbon (POC). TPP collected on the filters was converted to orthophosphate as described for TDP in Sect. 2.4.3. Concentration of inorganic phosphate was then determined spectrophotometrically according to Hansen and Koroleff [40]. BSi was leached from the collected particulate matter by alkaline pulping with 0.1 M NaOH at 85˚C. After 135 minutes the leaching process was terminated with 0.05 M H 2 SO 4 and DSi was measured by spectrophotometry following Hansen and Koroleff [40]. If not indicated differently, presented PM values are the sum of the two measured size fractions (< and >200 μm). Exceptions are TPP and BSi samples that were filtered as bulk samples before t 7 and t 29 , respectively. BSi data of t 29 were removed from the dataset due to a systematic error made during size fractionation on this specific day.
Water column samples for Chla concentration analysis were filtered as described for PM, taking care to minimize light exposure during filtration. Chla content of the collected particles was extracted and analysed by high-performance liquid chromatography (HPLC) as described in Bach et al. [26].

Elemental analysis of sediment trap samples.
The sediment trap samples were collected in 5 L Schott Duran glass bottles. To separate PM from bulk seawater, particles were concentrated by flocculation and coagulation with ferric chloride (FeCl 3 ) as described by Boxhammer et al. [29]. Briefly, FeCl 3 and NaOH (for pH stabilisation) were added simultaneously to the well-stirred samples. The clear supernatant water was removed after one hour of particle sedimentation. Mean concentration efficiency of this method was 99.6% with respect to samples' TPC content [29]. The concentrated samples were centrifuged, deep-frozen at -30˚C and lyophilised for 72 hours. The desiccated material was then ground in a ball mill to a homogeneous powder of 2-60 μm particle size [29]. TPC, TPN, TPP, and BSi content of the finely ground sample material was determined from subsamples of 1-2 mg as described for PM of water column samples (see Sect. 2.4.4). The cumulative mass flux of all four elements was expressed in μmol L -1 by dividing the summed up mass flux by the calculated mesocosm volumes (Sect. 2.2).
From May 25 (t 77 ) onwards we screened the freshly taken samples for dead herring larvae that hatched inside the mesocosms on t 62 (see Sect. 2.2). All larvae found were removed for separate analysis, thus they did not contribute to the vertical flux.

Calculation of mesozooplankton biomass.
Biomass of the mesozooplankton community was calculated based on abundance data obtained from counting with a stereomicroscope [32]. The community was strongly dominated by the copepod species Pseudocalanus acuspes (Giesbrecht, 1881), which represented about 97% of the mesozooplankton counts. Therefore, we only considered copepod biomass for mesozooplankton PM. Copepod nauplii were sufficiently abundant (up to 100 ind. L -1 ) to be sampled quantitatively on PM filters (Sect. 2.4.4). Adult copepods and copepodites, however, were much lower in abundance and naturally escape sampling by the IWS. Thus they were not represented in PM analysis. To avoid double counting of nauplii biomass, only adult copepod and copepodite biomass were included in the calculation of copepod PM (PM COP ). We applied the image-based ZooScan approach to estimate biomass for the different copepod size classes [41], since biomass measurements of individual organisms have not been conducted. Therefore, subsamples from the mesozooplankton net tows (4% of the total sample) were evenly distributed on a flat-bed scanner (Perfection Pro V750, Epson) to provide high-resolution images (10.6 μm pixel size) of all particles and organisms in the sample. Subsequent image processing with ZooProcess [41] provided a large number of variables for object characterization, including several measures of size such as length or area. For estimation of copepod biomass we then converted measured area of each individual imaged organism to dry-weight (dw) by applying the empirical relationship of [42]: The dry-weight was subsequently converted to C and N content (μmol) using the data for body mass composition of zooplankton from Kiørboe [43]. For copepods, the applied C:dw and N:dw ratios were 0.48 and 0.10, respectively. The resulting conversion factors for C and N biomass per individual organism were applied to the complete time series of abundance data for adult copepods and copepodites. P content was calculated using a conversion factor of C:P of 52:1 derived from Pseudocalanus sp. caught in Oslofjord (Norway) during the same time of the year (average ratio of individuals caught between March and May) by Gismervik [44].
Similar procedures for image-based biomass estimation of mesozooplankton have been applied in previous studies and showed generally reliable results [45][46][47]. It should be noted, however, that this approach assumes constant size ranges of copepod life stages and can thus not account for shifts in size structure within a community or population.

Calculation of net changes in element pools and net community production
The relevant pools for mass balancing C, N, P, and Si are dissolved inorganic nutrients (IN C/N/P/Si ), dissolved organic matter (DOM C/N/P ), suspended particulate matter (PM C/N/P/Si ), and the sum of particulate matter collected in the sediment traps (SPM SED (C/N/P/Si) ). Mesozooplankton, strongly dominated by copepods, was treated as a separate PM pool (Sect. 2.4.6), and defined as PM C/N/P (COP) . A summary of all pools and fluxes considered in the mass balances are shown in the conceptual Fig 1B. Net changes of the element pools (IN, DOM, PM, and PM COP ) were calculated as delta (Δ) values relative to conditions at the start of the experiment. We defined the starting conditions as the average value of the first seven sampling days (t -1 -t 11 ). Averaging over this relatively long period was necessary to minimize the influence of data variability. This was well justifiable as relative changes of the element pools were small before t 13 (Sect. 3.1). However, some exceptions (listed in the following text) had to be made for distinct element pools. The first two data points of DSi (t -1 and t 1 ) were excluded due to a methodological measurement problem. The reference value of ΔDIC in the high CO 2 treatment is based on a single sampling day t 5 , since before DIC was increased by stepwise CO 2 additions (Sect. 2.2) and afterwards CO 2 rapidly outgassed to the atmosphere (super-saturation of the water column). DOC and DON data of t -1 were removed from the datasets as measurements displayed substantial unexplainable variability with strong impact on calculated starting conditions. Reference values for DOP were calculated from three data points (t -1 , t 1 and t 9 ), as those days were the only days when DOP was sampled during the initial phase of the experiment (Fig 2). The first mesozooplankton sampling on t 1 served as the reference point for net changes in PM C/N/P (COP) . The start and end point of the individual reference periods, as well as the calculated reference values of each element pool within the water column are summarized in Table 1.
Reference values of both CO 2 treatments are average values ± standard deviation (SD) of the indicated reference periods for calculation of net changes in the respective element pools (see Table 2 for abbreviations of the element pools). If start and end point of the reference period are identical, the reference period is limited to only one data point. t-tests performed on average values of all ambient and high CO 2 mesocosms are indicated by p-values (bold values indicate significant difference, p 0.05).
Net community production (NCP) is most commonly estimated by measuring the biological drawdown of DIC or NO 3 - [48,49]. In the present study, we derived NCP from the actual build-up of biogenic C, N, P, and Si following Hansell and Carlson [48] and Spilling et al. [18]. This total NCP theoretically equals the cumulative drawdown of inorganic nutrients (ΔIN) and is therefore given in moles per litre and not as a rate. We calculated net community production in (1) high temporal resolution lacking mesozooplankton contribution (Eq 2) and (2) in reduced temporal resolution but including mesozooplankton contribution (Eq 3):  Ocean acidification can influence biogeochemical cycling of C, N, and P Thus, NCP C/N/Si was calculated for every second day (Table 2), while NCP P followed the irregular sampling of DOP described in Sect. 2.3 and illustrated in Fig 2. NCP C/N/P (COP) was calculated for usually every 8 th day (Table 2) following the mesozooplankton sampling regime (Fig 2).

Data analysis and statistics
Data shown in tables and figures represent average treatment values (ambient and high CO 2 ) of the five treatment replicates. Datasets of C and Si pools encompassed the entire duration of the experiment until t 103 and t 105 , respectively. Datasets of N and P, however, ended on t 95 as this was the last day where DON and DOP data were available (see Sect. 2.4.3).
Two sample t-tests using R software [50] were performed for detection of differences in the initial concentrations of element pools between ambient and high CO 2 mesocosms (average values used for delta calculations; Table 1).
For detection of CO 2 treatment effects on net changes of element pools and calculated net community production, univariate permutational analysis of variance (PERMANOVA) tests were run in R software [50], using Euclidean distances matrices with 99,999 permutations [51,52]. PERMANOVA was chosen, as assumption of homogeneity of variances was not met for all analysed parameters in all experimental phases. CO 2 effects were evaluated for average values of each experimental phase (see Sect. 3) or in the case of sedimented PM for cumulative values at the end of the four experimental phases.

Results and discussion
The experiment was divided into four phases based on the development of Chla concentrations (Fig 3A; see also [26]): Phase I (t -1 -t 16 ), Phase II (t 17 -t 40 ), Phase III (t 41 -t 80 ), Phase IV (t 81 -t 105 ). These phases were used for the interpretation of net changes in the C, N, P, and Si pools inside the mesocosms (Fig 1B). Average pCO 2 values of the four experimental phases and the entire experiment at ambient and high CO 2 (t -1 -t 105 ) are given in Table 3.
Average volume of ambient and high CO 2 mesocosms (mean ± SD, n = 5) was determined on t 46 of the experiment. pCO 2 values (μatm) are averages of the four experiment phases (I, II, III, IV) and of the entire experiment (I-IV). The two symbols (+) and (-) represent out-and in-gassing conditions of CO 2 , respectively (presumed atmospheric pCO 2 of 395 μatm).

Temporal development of the C, N, P, and Si pools during the phytoplankton spring-bloom
The first (pre-bloom) phase of the experiment was characterised by relatively stable environmental conditions with high concentrations of DIN, DIP, and DSi (~7.0,~0.76, and~9.8 μmol L -1 , respectively; Fig 3B-3D) and short day length [26]. The enclosed water columns were entirely mixed due to thermal convection inside the mesocosm bags [26]. Enclosed plankton assemblages were relatively similar among the ten mesocosms although small differences were detected (see [26]). No significant differences in initial concentrations of inorganic nutrients as well as the other element pools (PM, PM COP , DOM) were found between CO 2 treatments apart from DIC, as a direct consequence of the CO 2 manipulation (see reference values in Table 1). Net changes in the pools of all four elements (C, N, P, Si) were relatively small underlining the pre-bloom character of Phase I (Fig 4). The decline of DOC in this early phase was not reflected in changes of any other C pool and is therefore more likely associated with sampling induced artefacts than with real changes in the DOC pool. Thus, we do not draw any conclusion from this trend.
The second phase covers the first major build-up and decrease of Chla during the phytoplankton spring-bloom (Fig 3A). Primary production was fuelled by inorganic nutrients that rapidly decreased during the bloom development (Fig 3B-3D). Small silicifiers (2-5 μm, mostly diatoms) as well as the large diatom Coscinodiscus concinnus (Smith, 1856; >200 μm) dominated the bloom-forming autotrophic community during this phase (see Bach et al. [53] for detailed information). Low DIN and DIP concentrations limited primary production after t 31 and thus terminated the exponential growth of phytoplankton (Fig 3B and 3C). The  Ocean acidification can influence biogeochemical cycling of C, N, and P decrease of DSi, however, just slowed down according to uptake kinetics of diatoms [54,55]. Thus, DSi was still available in low concentrations after the first bloom (>1 μmol L -1 at the end of Phase II; Fig 3D). Peak values of PM were reached between t 31 and t 37 with average net buildup of TPC, TPN, TPP, and BSi of~33.6,~4.7,~0.33 and~3.3 μmol L -1 , respectively (Fig 4). Sedimentation of PM of all four elements and build-up of DOM C/N/P started to increase right from the onset of the first phytoplankton bloom (Fig 4). Highest sedimentation rates were observed during the bloom peak, implying a close temporal coupling between primary production and sinking particle flux. In contrast to particulate C, N, and P, the amount of BSi removed from the water column during this period equalled the net build-up in the water column (Fig 4). BSi:C ratios in the sediment trap samples were four times higher than those in the water column (S1  Table 2 for a detailed symbol description). DIC is only included at ambient CO 2 (grey, dotted line), lacking correction for CO 2 air-sea gas exchange (see Sect. 3.2). Roman numbers denote the different phases of the experiment. Percentages indicate the approximate discrepancy between net community production and inorganic nutrient consumption during Phases III and IV. https://doi.org/10.1371/journal.pone.0197502.g004 Ocean acidification can influence biogeochemical cycling of C, N, and P Fig), suggesting a strong decoupling of the two elements when it comes to sinking from the productive surface layer; a phenomenon also observed in the open ocean [56,57]. The strong decline of Chla concentrations at the end of Phase II was much less pronounced in the PM pools (compare Figs 3A and 4). TPC, TPN, and TPP remained relatively high even though Chla strongly decreased. This suggests a highly efficient transfer of autotrophic into heterotrophic biomass and/or non-sinking phytodetritus accumulating in the water column. Indeed, bacterial as well as micro-and mesozooplankton abundances increased parallel to the Chla decrease [32,58].
Phase III encompassed the second and slightly less pronounced build-up and decrease of Chla during the spring-bloom (Fig 3A). Small diatoms and flagellates (2-5 μm), but mainly the giant diatom C. concinnus (up to 50% Chla contribution) dominated the phytoplankton community during this phase (see [53]). The shift in dominance from small diatom species (<200 μm) to the large cells of C. concinnus (>200 μm) is clearly reflected in the temporal development of the two size fractions of BSi (S2 Fig). Regenerated N and P, as well as the remaining DSi likely fueled primary production during this second bloom. Peak values of PM were reached between t 49 and t 53 with average net build-up of TPC, TPN, and TPP of~36.8,~5.1, and~0.25 μmol L -1 , respectively (Fig 4C-4H). A peak in net build-up of BSi was absent due to the high loss through sedimentation and only very low DSi concentrations available (Figs 4A, 4B and 3D). The high variability in DON and DOP concentrations likely masked consumption of both pools by the plankton community (Fig 4E-4H). We suspect that considerable proportions of DON and DOP were rather refractory and only a small fraction of these pools, composed of labile compounds, was used and turned over by bacteria and phytoplankton on time scales that could not be resolved by our 48 h sampling regime. This assumption is consistent with field observations [59,60] and is supported by relatively high background concentrations of (likely refractory) DON and DOP right after mesocosm closure (see Table 1). Labile DOP is known to be recycled within hours to days [59,61], therefore often fuelling primary production under DIP depletion. In contrast to the relatively stable concentrations of DON and DOP (Phases III and IV), DOC concentrations showed a decreasing trend during the second phytoplankton bloom, reaching values lower than the initial ones by the end of Phase III.
Adult copepod and copepodite biomass (PM C/N/P (COP) ) decreased directly after mesocosm closure (Fig 4C-4H), but increased again during the phytoplankton blooms with highest values reached during and after the second bloom peak in Phase III. Predation by herring larvae that hatched inside the mesocosms (t 62; see Sect. 2.2) and started feeding on larger mesozooplankton around t 80 were most likely responsible for the decline of PM C/N/P (COP) in the postbloom Phase IV. The relative change in the PM C/N/P (COP) pool was most pronounced in the P mass balance due to the relatively high P content of copepods. In contrast to primary producers, the PM pool of mesozooplankton is likely not properly accounted for in most mass balance approaches due to the relatively small number of organisms in the sample volumes that are used for PM analysis. We observed a temporal contribution of up to 20% to TPP build-up, which emphasizes that this pool should not be neglected.
The fourth phase (post-bloom) was characterized by typical summer conditions in the coastal mid-latitudes. Inorganic nutrient concentrations were depleted (Fig 3B-3D), the water column was stratified (see [26]), and PM concentrations had almost declined to those of the pre-bloom phase (Fig 4).

Mass balances of Si, C, N, and P
The NCP of all four investigated elements (see Eqs 2 and 3) should in theory match the consumption of their inorganic nutrients over time. This worked out well for Si, where NCP was only slightly overestimated with on average~5% during Phases III and IV (2 nd bloom and post-bloom; Fig 4A and 4B). This is well within the range one would expect from combining measurement uncertainties of three different pools (DSi, BSi, BSi SED ; Fig 1B). Interestingly, we observed a temporal mismatch of DSi consumption and NCP Si shortly before and during the onset of the spring-bloom (between Phases I and II). Wall growth, a common artefact in enclosure experiments [62][63][64], can be excluded as a sink for DSi, as mesocosm inside walls were frequently cleaned (see Sect. 2.1; Fig 2) and we have not observed a comparable pattern in the mass balances of N or P (Fig 4E-4H). Thus, we assume that this mismatch at the end of Phase I might be explained by internal storage of DSi in diatom cells, which can contribute up to 50% of their total silica content (see review by Martin-Jézéquel et al. [65]). We observed that one of the dominating diatom taxa during that time, Arcocellulus sp. (see [53]), had very fragile frustules that potentially broke during the filtration process for BSi analysis and released internally stored DSi. Apart from this specific period the Si mass balance was virtually closed.
When attempting to calculate the mass balance of C, we faced two major difficulties. These were (1) the unexplainable day-to-day variability in DOC data (up to~50 μmol L -1 within 48 hours; Fig 4C and 4D) and (2) the poorly constrained gas exchange of CO 2 with the atmosphere. Both made it ultimately impossible to calculate a reasonable mass balance of C. Achieving accurate DOC data in an experimental setup like pelagic mesocosms has shown to be challenging [64], but not impossible [9,66]. Measurement precision and accuracy in the present study was high [30], so that the variability is more likely to originate from artifacts which were induced during sampling. We refrained from smoothing the data by calculating moving averages since potential contaminations can only increase not decrease the mean and would have led to an overestimation of DOC build-up and NCP C (see S3 Fig). However, it should be noted that the strong variability in ΔDOC had a substantial impact on calculated NCP C (Fig  4C and 4D).
To correct DIC for the air-sea flux of CO 2 we have followed the approach described by Czerny et al. [67], using the injected tracer gas N 2 O to infer the exchange rate ("gas transfer velocity") of CO 2 . This technique has been shown to yield good estimates of CO 2 transfer velocity in past mesocosm experiments under relatively stable physical conditions [18,64]. However, in the present study, the hydrographic situation within the mesocosms was highly dynamic with initial thermal circulation of the entire water columns until t 37 , followed by variable thermal stratification and surface layer mixing depth (see [26]). The thermocline physically isolated the bottom layer of the mesocosms from the atmosphere and led to pulsed inputs of N 2 O into the surface layer every time the mixing depth changed. This in turn resulted in discontinuous outgassing of N 2 O, which impeded reasonable estimates of N 2 O and consequently CO 2 gas transfer velocities. Hence, DIC concentrations could not be corrected appropriately for CO 2 air-sea gas exchange. To illustrate the discrepancy of un-corrected DIC data with NCP C we have included the measured net change in DIC into Fig 4C (dotted grey line, ambient treatment). In Phase IV the cumulative sedimentation of C alone exceeds net drawdown of DIC by a factor of three. Including CO 2 gas exchange with the atmosphere is therefore clearly crucial for mass balance calculations of C or when net organic C build-up is calculated from DIC drawdown. Hence, the exclusion of the CO 2 air-sea gas exchange in DIC drawdown [68] should be seen as very critical.
Balancing the NCP of N and P with DIN and DIP drawdown was not as easy as for Si but not as difficult as for C. The offset between inorganic nutrient consumption and NCP during build-up of the first bloom (Phase II) was highly variable in the case of N (-40 to +13%) and relatively constant for P (approx. -7%; Fig 4E-4H). The offset stabilised during Phases III and IV at values of about +30% and -15% for N and P, respectively, when the phytoplankton community had taken up all inorganic nutrients. NCP N can theoretically be increased above DIN consumption by N 2 -fixation, a significant external N source in the close-by Baltic Sea [69][70][71]. However, this explanation for the overestimated NCP N was excluded in the present study, as the corresponding organisms (diazotrophic cyanobacteria) were not present in the mesocosms. The pronounced overestimation of NCP N is therefore more likely a result of accumulated measurement inaccuracies of the N pools. Similar to DOC, build-up of DON showed strong variability of up to 4 μmol L -1 within 48 hours, not reflected in any other N pool (Fig 4E  and 4F). Thus, the DON pool was the source of the largest uncertainty within the N mass balance.
The underestimation of NCP P was unexpected as uncertainties in sampling of DOM and PM (e.g. clogging of filters or bursting phytoplankton cells) rather result in a certain overestimation of the two pools, than leading to their underestimation (Fig 4G and 4H). The discrepancy in DIP consumption and NCP P might be caused by variability in DIP measurements during the reference period used for calculation of net changes (see Table 1 and Sect. 2.5). During this period, DIP concentrations varied by about 0.1 μmol L -1 within 48 hours with only low primary production going on [72]. A potential overestimation of the background concentration of DIP by about 0.1 μmol L -1 could have led to an overestimated consumption of DIP, possibly explaining the observed offset in the P mass balance.
Altogether, our study has shown that mass balance calculations of elements in marine ecosystems are challenging even in enclosed mesocosm systems with discrete measurements of all relevant parameters. Precise determination of the DOM pools and in the case of C the accurate correction of DIC by the CO 2 air-sea gas exchange turned out to be most critical. This highlights the enormous challenge of mass balancing elements in open systems (e.g. the coastal ocean, estuaries or eddies) where even more uncertainties emerge due to permanent exchange of water masses.

Impact of CO 2 on partitioning of C, N, and P
TPC was the only PM pool influenced by increased CO 2 (Fig 5A, 5F and 5K). We observed a positive trend in TPC build-up at high CO 2 during both phytoplankton blooms (Phases II and III; up to 7 and 9 μmol L -1 , respectively), although this observation was statistically non-significant due to high within-treatment variability ( Fig 5A, Table 4). The tendency of increased Cfixation was likely caused by enhanced 'carbon overconsumption' [73,74], which was also indicated by a positive trend in the C:N ratio of particulate matter at high CO 2 during both phytoplankton blooms (Fig 6A; non-significant in all Phases, see Table 5). The observed trend in the C:N ratio was most prominent in the particle size fraction larger than 200 μm, which was mainly constituted by the large diatom C. concinnus (Fig 6C). Thus, the dominance of C. concinnus during the second phytoplankton bloom (see Sect. 3.1 and [53]) explains the influence of the particle size fraction larger 200 μm on the bulk PM C:N ratio during this time (Phase III). The CO 2 -dependent C:N signal in the water column was also found in sinking PM ( Fig  6B), indicating that the excess C fixed by C. concinnus (size fraction >200 μm) was not transferred into higher trophic levels or re-mineralized by bacteria in the water column. Instead this C was removed from the water column by sinking C. concinnus cells.
In contrast to other plankton community CO 2 perturbation studies [16,17], we have not detected an increase in DOC build-up at high CO 2 , although the high variability in the present data set may have masked small differences (Fig 5B).
Surprisingly, we found that copepod biomass was significantly elevated under high CO 2 during times of regenerated production (Phase III; Table 4). Between the peak of the first phytoplankton bloom and mid of the post-bloom Phase IV (t 35 -t 89 ), PM C/N/P (COP) was increased by on average 2.7, 0.5 and 0.05 μmol L -1 with respect to TPC COP , TPN COP , and TPP COP ( Fig  5C, 5H and 5M). Enhanced primary production at high CO 2 that was revealed by Eberlein et al. [72] during the same study must have caused this amplified transfer of biomass from primary producers (phytoplankton) to the higher trophic level of mesozooplankton (see [75]). Increased abundance of zooplankton organisms without a corresponding increase in primary producers and their respective biomass has also been observed in other studies [76]. The disappearance of the CO 2 effect on copepod biomass (C, N, P) towards the end of the study (Phase IV) can likely be explained by a potential further transfer into biomass of herring larvae. This Ocean acidification can influence biogeochemical cycling of C, N, and P assumption is supported by (1) the fact that the decrease in copepod biomass in the water column (Fig 5C, 5H and 5M; Phase IV) was not reflected in the downward flux of C, N, and P ( Fig 5D, 5I and 5N) and (2) by the fact that the difference in copepod biomass build-up between the two CO 2 treatments was reflected in an increased survival rate of herring larvae at high CO 2 by the end of the study (see [77]). The amplified transfer of C, N, and P to higher trophic levels (copepods and likely fish larvae) at high CO 2 in turn has caused a prolonged retention of biomass in the water column, which significantly reduced the downward flux of N and P (Table 4). Cumulative sedimentation of both elements started to differ between treated and control mesocosms at the same time when CO 2 driven trends in TPN COP and TPP COP occurred (t 35 onwards; Fig 5H, 5I, 5M and 5N). The difference in cumulative sedimentation of N and P between treatments (Fig 5I and 5N) constantly increased until t 69 (0.7 and 0.04 μmol L -1 for N and P, respectively) and remained at this level until the end of the experiment. On t 105 the deposition of N and P at high CO 2 was reduced by on average~11 and~9%, respectively. Due to increasing within-treatment variability during the second half of the study, cumulative sedimentation of both elements was significantly different on t 69 but not on the last day of experiment (t 105 ; Table 4). Interestingly, the prolonged retention of biomass within the water column did not affect the downward flux of C (Fig 5D). We assume that the positively influenced relative C content (i.e. C:N ratio) of sinking PM under high CO 2 (Fig 6B) must have compensated for a theoretically reduced sedimentation of the element. The CO 2 related trend in the C:N ratio of sinking particles was driven by large amounts of sinking C. concinnus cells, dominating the downward flux of PM during Phases III and partly IV (see S4 Fig). Hand picked cells from unprocessed sediment trap samples (t 65 ) were found to have C:N ratios of up to 30:1 in the high CO 2 mesocosms. The large cell size of C. concinnus (>200 μm) prevented grazing by the dominating mesozooplankton species P. acuspes, which excluded transition of its biomass into higher trophic levels and allowed the diatom cells to sink out of the water column.
Our findings show that increased retention of N and P within the pelagic food web under high CO 2 can lead to a significant and equivalent reduction of their sedimentation. Ocean acidification can influence biogeochemical cycling of C, N, and P Furthermore, the plankton community composition in the present study has shown that a predator-prey size mismatch between phyto-and mesozooplankton taxa (here C. concinnus and P. acuspes) can strongly influence element cycling. Together with changes of phytoplankton C:N ratios, the observed impacts of ocean acidification on element partitioning have the potential to alter cycling of carbon and nutrients in the marine realm.

Conclusions
In this study we investigated the influence of simulated ocean acidification on the development and partitioning of the C, N, P, and Si pools in a coastal pelagic ecosystem. Our mass balance approach over 100 days, covering a natural winter-to-summer plankton succession, has highlighted important challenges and uncertainties in elemental mass balance calculations, but also revealed significant changes of element pool partitioning under realistic end-of-the-century CO 2 concentrations (~760 μatm pCO 2 ): • Even in a closed mesocosm system we experienced high uncertainties and methodological challenges for our mass balance approach that highlight potential uncertainties in balance calculations of major biogeochemical elements in the open ocean. Accurate determination of the DOM pools and the CO 2 air-sea gas exchange were most critical in the current study.
• Transfer of C, N, and P from primary producers to higher trophic levels during times of regenerated production was significantly amplified at high CO 2 , leading to prolonged Values are average values of the different phases (I-IV) in the ambient and high CO 2 treatments ± standard deviation (SD). Effects of CO 2 were assessed by PERMANOVA, giving the sum of squares (SS), the F value by permutation (Pseudo-F), and the p-value (p (perm)). Significant effects detected are highlighted in bold, while positive or negative trends are indicated by (+) and (-) , respectively. https://doi.org/10.1371/journal.pone.0197502.t005 Ocean acidification can influence biogeochemical cycling of C, N, and P retention of biomass in the water column. Retention of N and P within the pelagic food web resulted in reduced sedimentation of both elements by about 11 and 9%, respectively.
• C-fixation relative to N showed a positive trend at high CO 2 , correlating with the time of inorganic nutrient depletion and the bloom of the large diatom C. concinnus. The excess C fixed by C. concinnus was not available for higher trophic levels due to its large cell size (>200 μm) and was removed from the water column by sinking of the diatom cells. This excess C counteracted a potential reduction in C sedimentation that could have been expected from patterns of N and P fluxes.
Even though the observed impacts were temporarily variable and likely dependant on the food web structure, our findings show that ocean acidification has the potential to change the biogeochemical cycles of C, N, and P by retaining C and nutrients in the sea surface food web.