The in situ mesocosm experiment was conducted in the Gando Bay, which is located on the east coast of Gran Canaria (Figure 1A). Situated about 100 km off the West-African coast, the Canary Islands are primarily influenced by the subtropical North Atlantic gyre and to a lesser extent by the Canary current upwelling system (Barton et al., 1998; Arístegui et al., 2009). Accordingly, the waters surrounding Gran Canaria are usually characterized by warm surface temperatures and a pronounced thermal stratification of the water column, resulting in predominantly oligotrophic conditions with low nutrient concentrations and plankton biomass throughout the year (Arístegui et al., 2001). However, exceptions can occur due to mesoscale variability, e.g., island eddies that transport nutrients from the mesopelagic zone into the surface waters (Arístegui et al., 1997; Sangra et al., 2009) or upwelling filaments that carry nutrients from the West-African coast into waters surrounding the Canary Islands (Barton et al., 1998; Pelegri et al., 2005). Such events can have a profound influence on productivity in the Canaries region.

On September 23rd 2014, the research vessel Hesperides deployed nine “Kiel Off-Shore Mesocosms for Future Ocean Simulations” (KOSMOS, M1–M9; Riebesell et al., 2013a), which were moored in clusters of three in the northern part of Gando Bay (27° 55′ 41″ N, 15° 21′ 55″ W) at a depth of ~20–25 m (Figure 1).

Each mesocosm unit consisted of an 8 m high flotation frame, a cylindrical mesocosm bag (13 m length, 2 m diameter) made of transparent thermoplastic polyurethane foil (1 mm thick) that allows for penetration of light in the PAR spectrum, as well as a conical sediment trap (2 m long) that tightly seals the bottom of the mesocosm and allows for collection of sinking organic material with a vacuum pump system on a regular basis (Figure 1C).

The bags were folded and mounted onto the floatation frames prior to deployment. Once in the water, the bags were unfolded immediately and submerged below the water surface with the upper opening 1 m below sea surface. Both the upper and lower openings were covered with meshes (3 mm mesh size) to exclude patchily distributed nekton and large zooplankton like fish larvae or jellyfish from the enclosed water bodies. The mesocosm bags were then left floating in the water column for 4 days to allow for rinsing of the bags' interior and free exchange of plankton (<3 mm) between the mesocosms and the surrounding water. On September 27th, divers replaced the mesh at the bottom of the mesocosm bags with the sediment traps, while a boat crew simultaneously pulled the upper part of the bags above the sea surface. This step separated the water bodies within the mesocosms from the surrounding water and thus marked the start of the experiment (day −4 = t-4, Figure 2). The entire procedure lasted for <2 h, thereby minimizing differences between the enclosed water masses among mesocosms.

The experiment lasted for 62 days in total, starting with the closing of the mesocosms on t-4 and finishing with the last sampling of the sediment trap on t57. Day t0 (October 1st) denotes the day of the first CO₂ manipulation, corresponding to the establishment of elevated pCO₂ as the experimental treatment (see Section CO₂ Manipulation). In the fourth week of the experiment (t24), we injected deep water into the mesocosms to simulate a natural upwelling event (Section Simulated Upwelling through Addition of Deep Water). Unfortunately, one mesocosm (M6) was lost on t26, when strong currents in Gando Bay pulled some of the moorings and mesocosms ~50 m seawards. The bag of M6 was irreparably damaged during the recovery procedure. Thus, M6 was excluded from sampling and analyses after t26.

Mesocosms were cleaned from the inside and outside to minimize wall growth by benthic organisms, which would consume nutrients and eventually lead to decreasing light intensities inside the mesocosms. Therefore, mesocosm wall cleaning was conducted in regular intervals throughout the experiment (Figure 2) using a specifically designed cleaning ring for the inner surface, and scrubbers for the outside of the mesocosm bags (Riebesell et al., 2013a; Bach et al., 2016). Unfortunately, however, the conical sediment trap and parts of the lowest mesocosm segment could not be cleaned from the inside due to the narrow tapered design. These parts corresponded to ~7% of the inner surface of the mesocosm, which experienced some degree of wall growth (see Section Plankton Community Structure and Influence of Ocean Acidification).

To simulate ocean acidification in our experiment, we added different amounts of CO₂-saturated seawater to the mesocosms, following the method described in Riebesell et al. (2013a). For preparation of CO₂-saturated seawater, we collected about 1,500 L of natural seawater from Melenara Bay at ~10 m depth using a pipe and pre-filtration system connected to the PLOCAN facilities. The water was aerated with pure CO₂ gas for at least 1 h until reaching saturation and pH_NBS values of ~4.7. Afterwards, the water was filtered again (20 μm) and transferred into 20 L bottles, which were then transported by boat to the mesocosm study site.

For addition of the CO₂-saturated water to the mesocosms, we used a special distribution device (“spider”) with a large number of small tubes to distribute the water uniformly within a radius of ~1 m. By constantly pulling the spider up and down inside the mesocosms, we ensured homogenous CO₂ enrichment throughout the entire water columns. By adding different amounts of CO₂-saturated seawater to seven of the nine mesocosms, we set up an initial gradient in pCO₂ from ambient levels (~400 μatm) to concentrations of ~1,480 μatm in the highest CO₂ treatment. No CO₂ water was added to mesocosms M1 and M9, which served as a control (ambient pCO₂). To avoid an abrupt disturbance of the plankton community, this initial CO₂ manipulation was carried out incrementally in four steps over a period of 7 days between t0 and t6 (Figure 2). Two further CO₂ additions were conducted during the experiment in order to account for loss of CO₂ through air-sea gas exchange. The first time was on t21 during the oligotrophic phase to adjust pCO₂ before deep water addition, and the second time on t38 in the post-bloom phase (Figure 2).

One of the major goals of this study was to investigate whether potential effects of elevated CO₂ on natural plankton communities in the study region might differ between oligotrophic conditions and during bloom situations. Such plankton blooms regularly occur in response to upwelling of deep water, which is primarily driven by mesoscale variability (e.g., eddies) and results in transport of nutrient-rich water masses from several hundreds of meters depth to the (usually) nutrient-poor surface layer (Arístegui et al., 1997; Basterretxea and Arístegui, 2000). Besides inorganic nutrients, oceanic deep water masses usually exhibit distinct signatures of minor constituents such as dissolved organic matter and trace metals, elevated pCO₂, or seeding populations of plankton species (Pitcher, 1990; Hansell et al., 2009; Aparicio-Gonzalez et al., 2012; Tagliabue et al., 2014). All of these factors may have minor or major influences on the ecosystem in the surface layer, which go beyond the effects of the major nutrients N, P, and Si. Consequently, a “simple” addition of inorganic nutrients would not be sufficient for a realistic simulation of a natural upwelling event.

To overcome this challenge, we specifically developed a deep water collection system that allowed us to obtain the large amounts of nutrient-rich deep water required for mimicking an upwelling event in our mesocosm experiment. The goal was to replace ~20% of the mesocosm volumes with deep water, thereby ensuring a sufficiently large input of inorganic nutrients comparable to those observed during natural upwelling events in the region (Arístegui et al., 1997; Neuer et al., 2007).

The flexible walls of the deep water collector consisted of fiber-reinforced food-grade polyvinyl chloride material (opaque), which was high-frequency welded into a pear-like shape with a volume of ~85 m³ (Figure 3A). The opening (diameter of ~25 cm) was equipped with a specifically-designed water intake device (based on a modified propeller drive) and a sealing disc as a closure mechanism for the deep water collector. Operation of both components was time-controlled (programmable), thereby allowing for remotely operated collection of water at a desired depth. A screen with 10 mm mesh size covered the opening to ensure that no large particles or organisms entered the deep water collector. Furthermore, a weight of ~300 kg was attached to the deflated deep water collector before deployment to submerge it in the ocean until the target depth was reached. An acoustic trigger was installed to release the weight after completion of the water intake, thereby allowing the rise of the filled deep water collector to the sea surface, only driven by the gentle buoyancy of 24 floats attached to the main frame (Figure 3B).

On October 23rd (t22), we transported the deep water collection system to a location about 4 nautical miles north-east from the study site, where water depth was ~1,000 m and thus sufficiently deep for deployment. Transport and operation of the deep water collector was carried out with the vessel “SAPCAN IV” (chartered from Amadores harbor service, Las Palmas). Upon arrival at the target location, the deep water collector was lowered to a depth of ~650 m, where ~85 m³ of water were collected Figures 3C,D. After resurfacing of the collector, it was gently towed back to the study site, where it was anchored until addition to the mesocosms 2 days later on t24. In the meantime, defined amounts of water had to be removed from the mesocosms to create space for subsequent addition of deep water. To accomplish this, we used a submersible pump (Grundfos SP-17-5R) to remove known volumes of water from the mesocosms at ~5 m depth on October 24th (t23).

In order to reach the desired mixing ratio of deep water of about 20%, a total of ~75 m³ of deep water were distributed among the nine mesocosms. Before addition, we characterized the deep water biologically and chemically by the full set of variables also routinely sampled in the mesocosms (see Table 2). Since deep water addition had to be carried out for each mesocosm separately one after another, we anticipated that this procedure would last at least several hours. To minimize nutrient uptake and growth by phytoplankton during this time, we conducted the deep water addition during night time, thus ensuring identical starting conditions of all mesocosms for the following experimental phase. Accordingly, deep water was added in two steps during the night of October 25th–26th (t24–t25), lasting for ~9 h in total.

The actual transfer of deep water to the mesocosms was conducted by submerging a pump (the same as for water removal described above) into the deep water collector and pumping the water into the mesocosms with an injection device similar to the “spider” used for CO₂ additions (see above), but with larger tube diameters and larger volume throughput. Continuous up and down movement of this enlarged spider during addition ensured homogenous vertical distribution of deep water inside the mesocosms. In the first step, we added ~80% of the calculated amount of deep water to each mesocosm. Since the salinity of the deep water was much lower than in the mesocosms (35.7 vs. 37.7), the mixing ratio of mesocosm water with deep water could be calculated from precisely measured changes in salinity. Based on CTD profiles and salinity calculations immediately after the first deep water addition, the second addition was then used for fine-tuning and adjustment of all mesocosms to identical deep water mixing ratios and concentrations of inorganic nutrients. Furthermore, by adding defined amounts of deep water with known salinity, and measuring the resultant salinity change in the mesocosms, we could accurately estimate the total volume of seawater in each mesocosm enclosure. The volumes determined by this method were ~35 m³ on average (±5%, see Table 1).

One of our study objectives was to investigate how effects of OA on plankton communities might propagate to higher trophic levels. Accordingly, we added ~330 eggs of greater amberjack (Seriola dumerili) to each mesocosm on October 31st (t30), which was during the time of peak biomass after deep water addition on t24. The number of added eggs was determined as a trade-off between preventing potential top-down effects from becoming too strong on the one hand, and providing the presence of sufficient fish for sampling (based on expected survival) on the other hand. Eggs of greater amberjack were collected from existing broodstock, hold by the Aquaculture research group (GIA) of the University of Las Palmas de Gran Canaria (ULPGC). All protocols within the breeding facilites were approved within the EU project “AQUAEXCELL” (ethics permit number: OEBA-ULPGC04/2016). The fish eggs were gently introduced by submerging the brood containers inside the mesocosms (~3 m depth) from day t30 until t32, with calculated time of hatching at 2 days after introduction.

Unfortunately, it was not possible to determine the abundance of fish larvae on a continual basis. No larvae of S. dumerili could be found in the net catches, possibly due to their escape from the towed net. Deployment of light traps was not successful either. Some dead fish larvae were found by screening the sediment trap material on the days after hatching. While this approach did not provide robust quantitative estimates, e.g., due to the fragility and rapid decay of dead organisms, it indicated substantial mortality of fish larvae within the first few days after hatching. Furthermore, no live individuals were found in the final sampling (t56) with a 1 mm net that covered the full diameter of the mesocosms, indicating that there was no survival of fish larvae until the end of the experiment. Nevertheless, it should be kept in mind that fish larvae might have had a top-down effect on the plankton communities in the mesocosms after ~t32, even though this possible influence is most likely negligible.

We conducted out a comprehensive sampling effort for a wide range of physical, ecological, and biogeochemical variables in the mesocosms and the surrounding waters on every second day, usually lasting from 9 a.m. until noon. An exception was the period right after deep water addition (t25–t33), when a rapid response of the plankton community was observed, and most variables were sampled on a daily basis.

Our preferred method of sample collection in mesocosm studies involves use of depth-integrating water samplers (IWS, HYDRO-BIOS, Kiel), which gently take in a total volume of 5 L uniformly distributed over the desired depth. However, this method is rather time-consuming, with one IWS haul usually lasting 3–4 min. Because the sample volume of oligotrophic water required for filtrations, incubations, etc., usually amounted to at least 60–70 L per mesocosm per day, we decided to adjust our sampling strategy and applied two methods of water collection in parallel, depending on the requirements of the various measurement variables (Table 2).

For variables that are sensitive to gas exchange or contamination, we collected integrated water samples using the IWS over 0–13 m water depth and directly filled subsamples into separate containers on the sampling boats following the specific requirements and protocol for the respective variable (see Section Data Analysis and Statistics). These sensitive variables were dissolved inorganic carbon (DIC), pH, dimethyl sulfide (DMS), inorganic nutrients [nitrate (NO3-), nitrite (NO2-), dissolved silicate (Si(OH)₄), ammonium (NH4+), phosphate (PO43-)], dissolved organic carbon, nitrogen, and phosphorus (DOC, DON, DOP), and water for all in vitro incubation experiments such as primary production (¹³C and ¹⁴C), N₂-fixation, bacterial degradation of sinking organic matter, or bacterial protein production assays.

Samples for other variables (e.g., particulate organic matter) were obtained with a custom-built pump system that allowed for a much faster collection of large sampling volumes. The system consisted of a manually operated pump, a 20 L carboy, a valve with integrated pressure gauge that connected to a 20 m long plastic tube (25 mm diameter), and a special inlet with several water intakes mounted to the open end of the tube. By applying the pump, a gentle vacuum was created (<150 mbar) in order to suck in water from the mesocosms into the tube and carboy. By moving the tube and attached inlet up and down during pumping (0–13 m), integrated water samples similar to the ones obtained by the IWS could be collected. To achieve this, pumping rate and vertical movement were synchronized with the holding capacity of the sampling carboy in order to avoid overflow of water before the sample could be considered integrated, i.e., before the vertical profile was completed. The 20 L sample carboys were then stored protected from direct sunlight on deck of the boats until sampling was completed. Once on shore, the samples were stored in a dark and temperature-controlled room (set to 16°C) where subsamples were taken for a variety of ecological and biogeochemical measurements (Table 2).

Sinking particulate matter was collected in the sediment traps at the bottom of the mesocosms. Sampling of the sediment traps was carried out every second day throughout the entire study, using a vacuum system connected to the tubes, which were attached to the collection cups following Boxhammer et al. (2016).

Mesozooplankton samples were acquired with an Apstein net (55 μm mesh size, 0.17 m diameter opening) in 8-day intervals. The maximum sampling depth of net tows was 13 m to avoid contact of the Apstein net with the sediment trap material, thus resulting in an overall sampling volume of ~295 L per net tow. Mesozooplankton samples were kept dark and cool until transport to shore, where they were preserved with sodium tetraborate-buffered formalin (4% v/v) for counting and taxonomic analyses. The number of zooplankton net catches per sampling day was restricted to every 8 days to avoid “overfishing,” i.e., exerting a too strong influence on top-down control of the system by removing zooplankton biomass.

CTD casts were carried out with a hand-held self-logging CTD probe (CTD60M, Sea and Sun Technologies) in each mesocosm and in the surrounding water on every sampling day. Thereby we obtained vertical profiles of temperature, salinity, pH, dissolved oxygen, chlorophyll a, and photosynthetically active radiation (PAR). Technical details on the sensors and data analysis procedures are described by Schulz and Riebesell (2013). Potentiometric measurements of pH_NBS (NBS scale) from the CTD were corrected to pH_T (total scale) by daily linear correlations of mean water column potentiometric pH_NBS to pH_T as determined from carbonate chemistry.

To identify potential ecological effects of CO₂ on the composition of the plankton community, we carried out multivariate analysis for abundance data of the different plankton groups. Therefore, we calculated the average abundances of different plankton groups during three experimental phases: (I) the oligotrophic phase until t23, (II) the phytoplankton bloom between t25 and t35, and (III) the post-bloom phase from t37 until the end of the study. All phytoplankton data used are from HPLC and CHEMTAX analysis, whereas numbers for micro- and meso-zooplankton were obtained by microscopy. In total, we distinguished 13 plankton functional groups that we used for analysis in the present study.

To account for the different scales of abundance of the various plankton groups, ranging from picophytoplankton (<2 μm) to mesozooplankton larger than 1 mm, all abundance data were normalized by their range as: (1)Nnorm = N/(Nmax − Nmin) where N is the abundance of each individual group, and N_min and N_max refer to the highest and lowest values found in the nine mesocosms. Thereby, all data are scaled to a range between 0 and 1, while maintaining the overall sample variance, as well as the relative differences between mesocosms.

After normalization of raw data, we generated ecological distance matrices using Bray–Curtis dissimilarity, which were then used for all multivariate analyses conducted here. In a first step, we performed non-metric multidimensional scaling (NMDS) to visualize ordination of the plankton communities in the mesocosms in response to CO₂ in the different experimental phases.

For a more quantitative assessment of how CO₂ might have influenced plankton community structure, we investigated the relationship between ecological distance (Bray–Curtis dissimilarity) and environmental distance, in this case pCO₂ (using Euclidian distance). Therefore, we applied a linear regression model to environmental and ecological distance data, using the same data matrices and phases as described above for the NMDS approach. Thus, every data point in the regression analyses corresponds to a pair-wise comparison of mesocosms with respective environmental distance (differences in pCO₂) and ecological distance (Bray–Curtis dissimilarity). The latter was calculated using the same (normalized) abundance data from plankton groups as for the NMDS analysis. This method allowed us to detect whether differences in plankton community composition were related to pCO₂. A statistically significant relationship between CO₂ and plankton community composition was assumed for p < 0.05 in the linear regression. The Mantel Test serves to ensure that these patterns did not arise by chance (when p < 0.05). All multivariate statistical analysis were conducted with the Fathom Toolbox for MATLAB (Jones, 2015).

Environmental conditions in Gando Bay during the mesocosm experiment were typical for late summer/early fall in the study region. Average temperatures in the mesocosms slightly decreased from ~24.4 to 22.3°C over the course of the study, corresponding to decreasing air temperatures during early autumn (Figure 4, Supplementary Material). Vertical profiles of temperature and salinity from the CTD showed a uniform distribution of both variables, indicating that there was no stratification and that the water columns in the mesocosms were well-mixed throughout the entire study period (Figure 4). Temperature profiles of the surrounding waters in Gando Bay were very similar to those in the mesocosms.

Average salinity in the mesocosms steadily increased from ~36.95 to 38.05 during the experimental period, interrupted only by a decrease due to addition of (less saline) deep water on t24 (Figure 4). The salinity increase was driven by evaporation, which was substantial due to relatively high temperatures and usually windy conditions (see Supplementary Material). In contrast, salinity in the surrounding waters remained almost constant throughout the experimental period. Because of this salinity difference, we could easily detect the presence of holes due to damaged mesocosm walls based on daily changes in salinity: When lower salinity water from the surrounding water entered the mesocosms, the daily increase in salinity of a particular mesocosm was lower than in the other mesocosms. Based on these observations, we could observe that M4 and M9 had holes around t11 (M4) and t13 (M9). Divers sealed the holes immediately after detection, by gluing small rubber patches onto the outside of the mesocosm bags.

To what extent these water intrusions might have affected the composition of the plankton communities in the mesocosms is difficult to assess, especially since they occurred during the oligotrophic phase when plankton abundances were low and measurement variability was comparably high. However, we did not observe anomalies in any of the measured variables during or after these damages. Furthermore, neither M4 nor M9 displayed any fundamental differences in plankton community composition or succession patterns throughout the rest of the study. Thus, we are confident that the temporal water intrusions through the holes only had a minor influence on the results presented here.

The injection of different amounts of CO₂-enriched seawater into the mesocosms in the period between t0 and t6 elevated DIC concentrations from initial values of ~2,079 up to 2,342 μmol kg⁻¹ in the highest CO₂treatment (M8). The corresponding increase in pCO₂ resulted in a treatment gradient ranging from 410 to 1414 μatm after the initial CO₂ enrichment (t7, Figure 5).

Afterwards, pCO₂ in the mesocosms decreased quite rapidly due to gas exchange at the air-sea interface. Although, we did not carry out direct measurements of gas exchange, the rapid decreases in pCO₂ and DIC until t20 were not reflected in build-up of total particulate carbon (TPC, Figure 7C), suggesting that the loss of inorganic carbon can be attributed predominantly to outgassing of CO₂. This is consistent with theoretical considerations, which suggest that rates of gas exchange should be high under the environmental conditions during our study, i.e., relatively high water temperatures, high wind speeds, and constant convective mixing of the entire water column in the mesocosms (Smith, 1985; Jähne et al., 1987). During the plankton bloom between t25 and t35 (see Section Oligotrophic Phase and Plankton Bloom in Response to Deep Water Addition), the decline in DIC concentrations was further enhanced by photosynthetic CO₂ fixation.

To compensate for the loss of CO₂ and readjust the treatment gradient, we conducted two more CO₂ enrichments on t21 and t38 (Figure 5). Altogether, the CO₂ gradient could be maintained reasonably well throughout the entire study. Furthermore, vertical profiles of pH show that carbonate chemistry conditions were distributed equally over the depth of the mesocosms, ensuring that all organisms in the water column experienced similar CO₂ conditions (Figure 5C).

During the oligotrophic phase, the phytoplankton community was dominated by small phytoplankton, mostly consisting of cyanobacteria (Synechococcus), which constituted 70–80% of chlorophyll a (Figure 8). This picture changed in phase II, when deep water addition resulted in a pronounced phytoplankton bloom that was dominated by diatoms, accounting for >70% of total chlorophyll a. Microscopic analysis revealed that the dominant species were relatively large chain-forming diatoms such as Leptocylindrus sp., Guinardia sp., and Bacteriastrum sp., but also detected other species such as Nitzschia sp. at lower abundances. The remaining phytoplankton consisted mainly of dinoflagellates, Dictyocha-like flagellates (belonging to chrysophytes) in some mesocosms, and prymnesiophytes (mostly Phaeocystis sp.) throughout the experiment (Figure 8).

Microzooplankton communities in the mesocosms were mainly composed of ciliates and heterotrophic dinoflagellates, whereas mesozooplankton was dominated by different copepod species and nauplii, but also included other functional groups such as appendicularia (Algueró-Muñiz et al., in preparation).

It should be noted that underwater video footage indicated the formation of some patchy benthic growth on parts of the inner mesocosm surfaces, which could not be cleaned (i.e., the conical sediment trap and parts of the lowest mesocosm segment, see Section Mesocosm Setup, Deployment Procedure, and Maintenance). Pigment analysis of this organic material suggested that it consisted to a large part of Phaeocystis colonies. In fact, adhesion to surfaces and subsequent rapid colony formation is characteristic for Phaeocystis (Rousseau et al., 2007). However, since the affected area was rather small compared to the mesocosm volume (~10 m² uncleaned mesocosm wall surface vs. 35 m³ mesocosm volume), we are confident that this wall growth did not significantly affect the results for phytoplankton community composition and biogeochemistry presented in this study.

Altogether, the phytoplankton succession pattern observed in our mesocosms—switching from prevalence of picoeukaryotes and picocyanobacteria (Synechococcus) toward a system dominated by large diatoms and dinoflagellates—is typical for the transition from open ocean gyres to coastal upwelling regions, as well as for the species succession in mesoscale eddies (Arístegui et al., 2004; Brown et al., 2008).

The main objectives of our mesocosm campaign were to investigate (a) how ocean acidification could change plankton community composition and food-web structure in oligotrophic environments, and (b) if such potential changes might amplify or weaken during periodic upwelling events of nutrient-rich deep water. In the present paper we assess how increasing CO₂ could affect the structure of plankton community as a whole. Therefore, we included and analyzed data from different functional groups of plankton, but did not investigate patterns within these groups at more taxonomic detail, e.g., on the species level. Such questions will be investigated in more targeted studies presented within the framework of this Research Topic (Table S1).

Our analysis at the level of functional groups revealed a significant effect of CO₂ on plankton community structure, both under oligotrophic conditions (phase I) and throughout the bloom induced by simulated upwelling of deep water (phases II and III). NMDS spaces (Figure 9) show the ordination of the mesocosms according to differences in their plankton community composition. The NMDS analysis of the different phases suggests the emergence of clear differences in plankton community structure, resulting in ordination of mesocosms according to the CO₂ treatment. Notably, these differences are not attributable to the response of only one or two dominant species, but emerged from overall shifts across the entire plankton community, including various groups of phytoplankton, micro- and meso-zooplankton (Figure 9). Particularly during the bloom (phase II) and post-bloom (phase III), the two highest CO₂ mesocosms (M2, M8) appear strongly separated from the others (Figures 9B,C).

More detailed analyses of the multivariate ecological datasets reveal a significant correlation between environmental distance (i.e., differences in pCO₂) and dissimilarity among plankton communities in the mesocosms throughout the entire study (Figure 10). During the initial oligotrophic phase (A), the plankton communities in the mesocosms were generally very similar to each other (low ecological distance between 0.1 and 0.2). However, the significant positive correlation between pCO₂ (distance) and ecological distance indicates that differences between plankton communities were larger at increasing differences in pCO₂ (Figure 10A, Table 4). In other words, differences in community composition were significantly related to differences in pCO₂ already during oligotrophic condition. These findings suggest that restructuring of plankton communities can occur under prolonged low-nutrient conditions, where observed variability in biomass is generally low. This conclusion is in line with recent studies in different oceanic regions, which reported most prominent effects of OA when inorganic nutrients were depleted (Paul et al., 2015; Sala et al., 2015; Bach et al., 2016). However, our study is the first to demonstrate this for oligotrophic waters of the subtropical North Atlantic.

Interestingly, the magnitude of the pCO₂ effect on community structure became even larger during the bloom, displaying a much more distinct influence of pCO₂ on plankton community structure as visible by a much steeper slope of the correlation in Figure 10B (also see Table 4). This finding suggests that rather subtle changes in community composition arising under oligotrophic conditions can be amplified during productive phases in response to upwelling events, resulting in pronounced differences in succession patterns and food-web structure under high CO₂ conditions. The correlation between environmental distance (pCO₂) and ecological dissimilarity (community structure) was still visible, but weaker, during the post-bloom phase (Figure 10C). This might be explainable by the overall increase in variability among mesocosms, which is indicated by the generally higher ecological distance of ~0.3–0.4 even at low pCO₂ differences. Although, the correlation between environmental distance (pCO₂) and ecological dissimilarity (community structure) weakened during the post-bloom phase (Figure 10C), the CO₂ effect became visible in bulk variables such as chl-a or particulate carbon during this period (Figure 7). These effects are mostly driven by the two mesocosms with highest CO₂ concentrations (M2, M8), which also appear notably separated in the NMDS space during phase II and III (Figures 9B,C).

The results presented in the previous sections indicated a successful deep water addition to the mesocosms that broadly resembled natural upwelling events and associated phytoplankton blooms in the study region. Compared to field observations of nutrient and chlorophyll during upwelling-induced blooms in the Canary Islands region, nutrient inputs as well as rate and magnitude of biomass accumulation in our experiment were slightly elevated (Arístegui et al., 1997; Garcia-Munoz et al., 2004; Neuer et al., 2007; Lathuiliere et al., 2008).

These differences likely arise from diverging modes of deep water input between our experiment and the real ocean. Eddy-induced upwelling of deep water around the Canary Islands occurs gradually over several days or weeks, thereby constantly mixing nutrient-rich deep water with nutrient-depleted surface water (Arístegui et al., 1997; Sangra et al., 2005, 2009). In our experiment, deep water was injected in a pulsed manner (i.e., within a few hours). This created a nutrient increase that was somewhat stronger than usually observed during natural eddy-driven upwelling events in the study region. In a broader sense, these different modes of deep water input between our mesocosm experiment and eddy-induced upwelling in the real ocean can be considered analogous to batch cultures and chemostat approaches in laboratory studies, respectively.

We cannot exclude the possibility that differences between pulsed and prolonged-diluted nutrient supply could alter the build-up rates and magnitude of biomass accumulation, and thereby possibly also community composition and biogeochemical cycling. However, since plankton community structure and species succession during our study closely resembled those during natural bloom events in the Canary Islands region (Basterretxea and Arístegui, 2000; Arístegui et al., 2004), we are confident that our findings are broadly representative for natural marine ecosystems of the study area.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.