Biogeosciences Ocean acidification shows negligible impacts on high-latitude bacterial community structure in coastal pelagic mesocosms

The impact of ocean acidification and carbonation on microbial community structure was assessed during a large-scale in situ costal pelagic mesocosm study, included as part of the EPOCA 2010 Arctic campaign. The mesocosm experiment included ambient conditions (fjord) and nine mesocosms with pCO2 levels ranging from ∼ 145 to ∼ 1420 μatm. Samples for the present study were collected at ten time points (t−1, t1, t5, t7, t12, t14, t18, t22, t26 to t28) in seven treatments (ambient fjord (∼ 145), 2 × ∼ 185, ∼ 270, ∼ 685, ∼ 820, ∼ 1050 μatm) and were analysed for “small” and “large” size fraction microbial community composition using 16S rRNA (ribosomal ribonucleic acid) amplicon sequencing. This high-throughput sequencing analysis produced ∼ 20 000 000 16S rRNA V4 reads, which comprised 7000 OTUs. The main variables structuring these communities were sample origins (fjord or mesocosms) and the community size fraction (small or large size fraction). The community was significantly different between the unenclosed fjord water and enclosed mesocosms (both control and elevated CO2 treatments) after nutrients were added to the mesocosms, suggesting that the addition of nutrients is the primary driver of the change in mesocosm community structure. The relative importance of each structuring variable depended greatly on the time at which the community was sampled in relation to the phytoplankton bloom. The sampling strategy of separating the small and large size fraction was the second most important factor for community structure. When the small and large size fraction bacteria were analysed separately at different time points, the only taxon pCO2 was found to significantly affect were the Gammaproteobacteria after nutrient addition. Finally, pCO2 treatment was found to be significantly correlated (non-linear) with 15 rare taxa, most of which increased in abundance with higher CO2.

Abstract.The impact of ocean acidification and carbonation on microbial community structure was assessed during a large-scale in situ costal pelagic mesocosm study, included as part of the EPOCA 2010 Arctic campaign.The mesocosm experiment included ambient conditions (fjord) and nine mesocosms with pCO 2 levels ranging from ∼ 145 to ∼ 1420 µatm.Samples for the present study were collected at ten time points (t−1, t1, t5, t7, t12, t14, t18, t22, t26 to t28) in seven treatments (ambient fjord (∼ 145), 2 × ∼ 185, ∼ 270, ∼ 685, ∼ 820, ∼ 1050 µatm) and were analysed for "small" and "large" size fraction microbial community composition using 16S rRNA (ribosomal ribonucleic acid) amplicon sequencing.This high-throughput sequencing analysis produced ∼ 20 000 000 16S rRNA V4 reads, which comprised 7000 OTUs.The main variables structuring these communities were sample origins (fjord or mesocosms) and the community size fraction (small or large size fraction).The community was significantly different between the unenclosed fjord water and enclosed mesocosms (both control and elevated CO 2 treatments) after nutrients were added to the mesocosms, suggesting that the addition of nutrients is the primary driver of the change in mesocosm community structure.The relative importance of each structuring variable depended greatly on the time at which the community was sampled in relation to the phytoplankton bloom.The sampling strategy of separating the small and large size fraction was the second most important factor for community structure.When the small and large size fraction bacteria were analysed separately at different time points, the only taxon pCO 2 was found to significantly affect were the Gammaproteobacteria after nutrient addition.Finally, pCO 2 treatment was found to be significantly correlated (non-linear) with 15 rare taxa, most of which increased in abundance with higher CO 2 .

Introduction
The acidification of our oceans, caused predominantly by dissolution of anthropogenic carbon dioxide (CO 2 ) in seawater, has the potential to affect the physiology of marine microbes.Therefore, because marine microbes play a major role in global biogeochemical cycles, this increase may Published by Copernicus Publications on behalf of the European Geosciences Union.
have unforeseen consequences on ocean biogeochemistry (Falkowski et al., 2008;Worden and Not, 2008).Experimental manipulation of the partial pressure of carbon dioxide (pCO 2 ) in marine mesocosms has demonstrated speciesspecific physiological responses to elevated dissolved CO 2 concentrations.For example, delayed or decreased coccolithophore calcification (Delille et al., 2005), a significant increase in photosynthetic capacity (Fu et al., 2008), higher CO 2 and N 2 fixation (Hutchins et al., 2007), and a decreased abundance of picoeukaryotes (Newbold et al., 2012) have been observed.However, the response of bacterial communities to elevated pCO 2 concentrations is less defined, with mixed reports of both significant increases in bacterial protein production (Grossart et al., 2006), and no significant changes in microbial community structure (Tanaka et al., 2008;Allgaier et al., 2008;Newbold et al., 2012).For example, during the 2008 PeECE III mesocosms study, elevated pCO 2 had no significant impact on bacterial abundance, diversity, or activity; however, the community structure of the small size fraction bacteria was significantly altered by the induced phytoplankton bloom (Allgaier et al., 2008;Arnosti et al., 2011;Riebesell et al., 2008).
While these existing studies have observed little impact of elevated pCO 2 on microbial community structure, they were all performed with molecular techniques that offered limited taxonomic resolution (e.g.High-Performance Liquid Chromatography, Denaturing Gradient Gel Electrophoresis, Terminal Restriction Fragment Length Polymorphism).To improve that resolution, this study employed high-throughput amplicon sequencing of 16S rRNA to characterize microbial taxonomic community dynamics.High-throughput amplicon sequencing provides an efficient method to obtain a deep molecular overview of microbial community structure, without having to cultivate environmental isolates (Agogué et al., 2011;Gilbert et al., 2009;Hubert et al., 2007;Huse et al., 2008;Margulies et al., 2005;Sogin et al., 2006).In this study, the variation of microbial assemblages was characterised through time, across a gradient of pCO 2 , in a largescale in situ pelagic mesocosm experiment in the coastal Arctic Ocean.In addition to characterizing the detailed response of the microbial community structure to elevated pCO 2 , the analysis of the 16S rRNA database provided insight on the effect of isolating the water column in a mesocosm, and to investigate the community structure response to elevated pCO 2 .

Location and carbonate system manipulation
The European Project on Ocean Acidification (EPOCA) supported a large mesocosm experiment in the Arctic which was conducted in the water of Kongsfjorden, Svalbard, Norway (78 • 56.2 N, 11 • 53.6 E) during the months of June and July 2010.Throughout the experiment, diverse environmental parameters were measured to explore the effect of ocean acidification (OA) on multiple biological processes.Briefly, nine mesocosms containing about 45 m 3 of seawater and reaching down to 15 m depth were deployed from Ny-Ålesund and pCO 2 was manipulated by the addition of CO 2 -saturated seawater in seven mesocosms, resulting in initial pCO 2 ranging from ∼ 286 to ∼ 1420 µatm.Two of the mesocosms were not manipulated and served as controls with starting pCO 2 of ∼ 185 µatm.Additionally, samples were taken directly from the fjord (initial pCO 2 ∼ 145 µatm) in which the mesocosms were suspended and from which the mesocosm water originated.These samples were used to monitor any natural changes in pCO 2 that may occur in the ambient water during the course of the experiment and were also important for detecting deviations in pCO 2 between the fjord and the untreated mesocosms with time.To promote phytoplankton growth, all nine mesocosms were subjected to nutrient additions (nitrate (NO 3 ), phosphate (PO 4 ) and silicate (Si)) on day (t) 13, creating pre-nutrient (t−1 to t12) and post-nutrient (t13 to t30) periods (Fig. 1).Detailed information about the experimental set-up, the mesocosms deployment, the carbonate chemistry, and the nutrients additions can be found in this issue in Riebesell et al. (2013), Czerny et al. (2013a, b), Bellerby et al. (2013), andSchulz et al. (2013), respectively.

Sampling, filtration and sample selection
A total of 10 L of water was collected using integrated water sampler (Hydrobios, Kiel, Germany) between 0 and 12 m water depth, from the fjord (∼ 145 µatm), and six mesocosms (starting pCO 2 = 2× ∼ 185, ∼ 270, ∼ 685, ∼ 820, ∼ 1050 µatm) on t−1, t1, t5, t7, t12, t14, t18, t22, t26 and t28 (Fig. 1).Only six of the mesocosms were chosen for this study due to time, personnel and equipment constraints.The collected water was first pre-filtered on a 20 µm sieve, and sequentially filtered through a 10 µm, a 3 µm filter to isolate associate-particle bacterial fraction (large size fraction) and through a 0.2 µm filter to isolate the small size fraction (Durapore ® 47 mm, Millipore).To avoid nucleic acid degradation, processing of the samples from filtration to flashfreezing (in liquid nitrogen) was performed within 30 min of the sampling event.Samples were then stored at −80 • C until DNA/RNA extraction.

DNA extraction, PCR, and Sequencing
Total nucleic acid was extracted from the 0.2 and 3 µm filters using the "Total RNA and DNA purification -NucleoSpin ® RNA II RNA/DNA buffer" kits from Macherey-Nagel (Macherey-Nagel GmbH & Co. KG, Düren, Germany).Standard protocol with minor modifications was followed.Changes to the protocol included making the filters brittle by immersing the samples in liquid nitrogen while still  in the cryovials to facilitate disruption and homogenization.The filters were crushed with RNase-free plastic pestles and lysozyme was directly added to the broken filter pieces while still in the cryovial.Both the RNA and DNA were isolated during the experiment.However, the RNA was kept for further purposes.DNA quality and quantity were assessed by micro-volume spectrophotometer nanodrop ND-1000 (Pe-qLab GmbH, Erlangen, Germany) measurements.All samples were kept at −80 • C until further analysis.
Polymerase chain reaction (PCR) and sequencing were performed following the Illumina HiSeq2000 and MiSeq V4-16S rRNA protocol (Caporaso et al., 2012).Briefly, the V4 region of the 16S rRNA gene was amplified with regionspecific primers that included the Illumina paired-end flowcell adapter sequences (Illumina Inc., CA, USA).The barcode was read using the custom index sequencing primer in an additional cycle (12 bp).Each sample was amplified in triplicate, and was pooled afterwards.Each 25 µL PCR reaction contained 12 µL of MoBio PCR Water (certified DNA-free), 10 µL of 5 Prime HotMasterMix, 1 µL of Forward Primer (5 µM initial concentration), 1 µL Golay Barcode Tagged Reverse Primer (5 µM initial concentration), and 1 µL of template DNA.The reactions were heated to 94 • C for 3 min for their initial denaturation, followed by 35 cycles in series of 94 • C for 45 s, 50 • C for 60 s, and 72 • C for 90 s.The amplicons were quantified using Quant-it ™ Picogreen ® (Invitrogen by Life Technologies ™ , CA, USA), and pooled in equal amounts (ng) into a 1.5 mL tube.Once pooled, the entire amplicon pool was cleaned up with the MO-BIO UltraClean ® PCR Clean-Up Kit (MO-BIO Laboratories, Inc., CA, USA).Finally, the pooled samples were quantified using a Qubit ® fluorometer (Invitrogen by Life Technologies ™ , CA, USA), and the molarity was estimated based on amplicon length.From this estimate, dilutions were made down to 2 µM and the standard Illumina sample preparation for sequencing was followed.Pooled amplicons were sequenced using custom sequencing primers, Read 1, Read 2, and Index.These sequencing primers were designed to be complementary to the V4 amplification primers to avoid sequencing of the primers.Amplicons were sequenced in a paired-end, 100 bp × 100 bp cycle run on the Illumina HiSeq2000, at a concentration of 4 pM with a 10 % PhiX spike.An entire control lane devoted to PhiX is also useful when sequencing low base diversity samples, like amplicons, and was included in the present analysis.

Sequence data analysis
All sequence analyses were performed using Quantitative Insights Into Microbial Ecology v. 1.5.0 (QIIME; Caporaso et al., 2010).QIIME defaults were used for quality filtering of raw Illumina data (including chimeras).Unique operational taxonomic units (OTUs) were picked against the Greengenes (McDonald et al., 2012) database and pre-clustered at 97 % identity; sequences that did not hit the reference collection were discarded.Representative sequences were aligned to the Greengenes core set with PyNAST (Caporaso et al., 2010).All sequences that failed to align were discarded.A phylogenetic tree was built from the alignment, and taxonomy was assigned to each sequence using the Ribosomal Database Project (RDP) classifier (Wang et al., 2007) retrained on Greengenes.Samples were rarefied to an even depth of 81 181 sequences and only the OTUs that appeared at least twice in the dataset were included in the further analyses; 106 singleton OTUs were not included in this analysis.

Statistical analysis
Multivariate analysis of microbial community structure was carried out in CANOCO 4.54 (ter Braak and Šmilauer, 2002), where the count of each OTU (97 % similarity) was used as a measure of abundance.All analyses had samples as scaling focus, and all species data were Hellinger-transformed using the program PrCoord 1.0 (Legendre and Gallagher, 2001;ter Braak and Šmilauer, 2002).Analysis of variance (ANOVA) followed by a Tukey test was done to test for significant differences between treatments (i.e.control vs. fjord, fjord vs. mesocosm, control vs. mesocosm) within each abundant phylum.Detrended correspondence analysis of the transformed OTU abundance data showed axis lengths < 3.0, suggesting a linear treatment of the data (Ramette, 2007).Redundancy analysis (RDA), with manual forward selection and Monte Carlo permutation tests (999 permutations), was used to evaluate effects of environmental variables (salinity, temperature, pH, chlorophyll a, etc.) on the microbial community composition.An indirect gradient analysis (PCoA) www.biogeosciences.net/10/555/2013/Biogeosciences, 10, 555-566, 2013 was used to plot the distribution of samples in ordination space, with important environmental variables (as indicated by forward selection) overlaid as supplementary data.Microbial community composition differences were assessed by UniFrac (Lozupone and Knight, 2005) distance using QIIME (Caporaso et al., 2010).
In order to assess whether or not particular taxa were significantly influenced by pCO 2 , a Bonferroni-corrected g-test was done using QIIME to remove significance due to chance.All analyses were considered to have a significant difference if p < 0.05 after Bonferroni correction.
Contour plots presenting mean abundance count plotted against pCO 2 and time (days) of the three most abundant genus of the OTUs significantly correlated to pCO 2 were created using Ocean data view (Bremen, Germany).

Experimental timeline
Phytoplanktonic bloom evolution was identified using the daily measured chlorophyll a (chl a) concentration (µg L −1 ) (Fig. 1).The chl a protocol and patterns are presented in Schulz et al. (2013).Briefly, all treatments (fjord included) underwent a natural bloom between t0 and t11, with highest chl a concentrations on t6.Subsequently, a second and third strong phytoplankton bloom happened only in the mesocosms following nutrient addition on t13.The second bloom had its highest chl a concentration on t19 and the third one, which varied greatly between mesocosms, reached its highest concentration on t27.These 3 blooms were represented as four general phases in phytoplankton chlorophyll phases defined by Schulz et al. (2013): phase 0 occurred from the start of the experiment on t−4 until adjustment of CO 2 was completed on t4; phase 1 started with the end of CO 2 addition on t4 until the nutrient additions on t13; phase 2 included the end of the first bloom on t13 to the end of the second bloom on t22; and phase 3 started from the end of the second bloom on t22 and lasted until the end of the experiment, on t30 (the chl a minimum of the third bloom was not recorded) (Fig. 1).Detailed fluctuations of chl a, nutrient concentrations, pH and pCO 2 are presented in this issue in Schulz et al. (2013) and Bellerby et al. (2013).

Community-structuring variables
The significant structuring variables for the total community during the post-nutrient addition period (t13-t30) of the experiment were (in order of explanatory importance) "fjord vs. mesocosm origin" (i.e.whether the sample was from water contained in a mesocosm or from the open fjord), sampling strategy (i.e.physical fractionation into small and large particle sizes), Si concentration, PO 4 concentration, mean primary production 14C-POC (PP), temperature (T ), and pH (Fig. S1 and Table 1).The microbial community in the small size fraction (0.2-3 µm) from the fjord and all the analysed mesocosms was dominated by Proteobacteria (in order of abundance: Gamma (γ )-, Alpha (α)-and Beta (β)-proteobacteria) throughout the experiment.However, Proteobacteria began dropping in abundance gradually after t7, coincidentally with the increase in the abundance of Bacteroidetes (Fig. 2).In the large size fraction (3-12 µm) Bacteroidetes dominated consistently, while a fourth group comprised of the "Cyanobacteria and eukaryotic chloroplasts" (which included Chlorophyta, Haptophyceae, Rhodophyta and Stramenopiles) were also abundant (Fig. 2).The group classified as "others" in the small size fraction was composed predominately of Cyanobacteria at the beginning of the experiment, and of Actinobacteria towards the end (Fig. S2).In the large size fraction, the "others" group was extremely variable until t7.For example, at t−1 the fjord "others" group was dominated by the Verrucomicrobia while the mesocosms "others" groups was dominated by Actinobacteria; by t5 Firmicutes dominated in most mesocosms, while being almost absent from the fjord.At t7, the Actinobacteria was the dominant taxa in the "others" group in all treatments for the remainder of the experiment.At the end (t28), some Verrucomicrobia increased in the control, ∼ 270, and ∼ 685 µatm mesocosms (Fig. S2).
Once the community was analysed with regard to filter size fraction (small vs. large size fraction), the structuring community variables varied.The fjord had a significantly different assemblage from the mesocosms in the small and large size fraction before (origin 3 %-4 %) and after (origin 48 %-12 %) mesocosm nutrient addition (Table 2); however, the fjord and mesocosm communities were not significantly different until after t5.The microbial community in the fjord small size fraction was not significantly different from the mesocosms communities in the pre-nutrient addition phase and only the γ -proteobacterial abundance was significantly different (p < 0.05) between fjord and mesocosm in the post-nutrient addition phase.The fjord large size fraction microbial community was significantly different from the mesocosms during both the pre-and post-nutrient addition phases.In particular, the "Cyanobacteria and eukaryotic chloroplasts" group was significantly different between fjord and mesocosms pre-and post-nutrient addition; while the Bacteroidetes, α-proteobacteria and "others" were only significantly different post-nutrient addition (Fig. 3 and Table 3).Furthermore, the significant variables that correlated with community structure changes in the small size fraction were dimethyl sulphide (DMS-16 %), bacterial production (bp-15 %), density (d-12 %) for the pre-nutrient period (t−4 to t12), and origin (48 %), pCO 2 (10 %), day (10 %) for the post-nutrient period (t13-t30; Table 2).For the large size fraction, these variables were oxygen (O 2 -7 %), DMS (7 %),  2).Therefore, the differences in the microbial community structure between the fjord and mesocosms were primarily due to the addition of nutrients to the mesocosms, and not to pCO 2 manipulation, as the control mesocosms were not significantly different from the elevated CO 2 mesocosms post-nutrient addition.

pCO 2 effect on microbial community
Although the pCO 2 treatment was not identified as a major community structuring variable, the relative abundances of 15 rare taxa (% abundance across time and treatment was <0.22%; Table 4) were significantly correlated to pCO 2 levels.From these 15 rare taxa in both small and the large size fractions, 12 showed a significant but slight increase with pCO 2 , having their maximum abundances in either the medium (∼ 685 and ∼ 820 µatm) or the high (∼ 1050 µatm) pCO 2 mesocosms.The remaining three decreased, with their highest abundances in the lowest (∼ 185 µatm) pCO 2 meso-  and ∼ 824 µatm mesocosms toward the end of the experiment (t22).Fluviicola was present from the beginning of the experiment, but decreased precipitously after CO 2 was added and then recovered in abundance after t10, reaching its highest abundance in the 1050 µatm mesocosm between t12 and t22 (Fig. 4).

Mesocosms and structuring effects
In this study, a large-scale mesocosm experiment was used to investigate the impacts of OA on the microbial community structure in a coastal, high latitude marine pelagic ecosystem.The experimental design provided the opportunity to test for the effects of four different pCO 2 concentrations (∼ 270, ∼ 685, ∼ 820, ∼ 1050 µatm) against two negative controls (∼ 185 µatm) over a six-week period.In addition, mesocosm-specific experimental artefacts were monitored by sampling the fjord microbial community throughout the course of the experiment.The microbial community structure post-nutrient-addition (t13) was significantly corre-lated with seven variables, the most influential of which was sample origin (fjord or mesocosm).The overall community structure was not significantly different between mesocosms (including control versus elevated pCO 2 ) over the course of the experiment.The significant effect of the mesocosm enclosures on microbial community structure could be due to the mesocosms themselves (isolating a microbial community from the surrounding fjord community) or since the effect was not significant before nutrient addition, more likely due to the addition of nutrients into the mesocosms at t13.The sampling strategy separating the community into size fractions was the second most important variable in explaining differences in community structure.Before nutrient addition, the communities in the small size fraction were not significantly different between the fjord (ambient), control mesocosms, and the elevated pCO 2 mesocosms.However, after the addition of nutrients, γ -proteobacterial abundances were significantly different between fjord and mesocosms, and probably reflected the utilization of metabolites released by decaying phytoplankton in the post-bloom system.In particular, the overall abundance of Bacteroidetes in the small and large size fractions increased in post-blooms conditions,  ) showing the relationship in between each treatment pre-and post-bloom condition for (a) small and (b) large size fraction bacteria of phyla with significant differences.Significant values are p < 0.05.possibly also as a result of the dissolved organic carbon (DOC) released by a decaying algal bloom and aggregation of dying phytoplankton, respectively.The γ -proteobacteria and Bacteroidetes generally include many phytodetritusassimilating organisms (Teske et al., 2011;Abell and Bowman, 2005;Pinhassi et al., 2004) and this would explain their increase in abundance during the demise of the bloom.Despite the observation that Bacteroidetes showed bloomrelated dynamics, and contradictory to the findings of Zhang et al. (2012), no significant difference in the Bacteroidetes abundance (in either fraction) was found between the control and elevated pCO 2 mesocosms, suggesting that elevated pCO 2 did not impact the relative abundance of Bacteroidetes.However, their abundance in the fjord was significantly lower than in the mesocosms, suggesting that the nutrient addition or influence of the mesocosm enclosure did have an impact.The large size fraction in the mesocosms also showed differences in the relative abundance of dominant phyla following nutrient addition (t13).It has previously been established that particle-associated assemblages were predominantly connected to phytoplankton development (Riemann et al., 2000;Allgaier et al., 2008).Furthermore, differences in the "Cyanobacteria and eukaryotic chloroplasts" group were measurable before nutrient addition.However these differences appear to be related to the natural phytoplankton bloom (which occurred in the fjord and mesocosms) that reached its maximum on t7.The "post-nutrient addition" differences were significant between the fjord and mesocosms for almost every abundant phyla throughout the different phytoplankton phases; suggesting that nutrient addition influenced autotrophic and heterotrophic microbial community structure.However, no significant differences were found between the control and the elevated mesocosms, which suggests that high pCO 2 level was not an important communitystructuring variable for the large size fraction in this experiment.Silica was the third most important structuring variable and is potentially related to diatom abundance (de Kluijver et al., 2010).The recycling of Si from decaying diatoms, after a phytoplankton bloom, is carried out by a diverse fast growing bacteria related to cytophagales (from Flavobacteria; Riemann et al., 2000).Indeed, an increase in the abundance of Bacteroidetes, which contains the Flavobacteria, was observed in the post-nutrient addition phase.
However, no single environmental variable could account for the microbial community composition of the large and small size fractions for all of the phases of the mesocosm experiment (Fig. 1).Rather a shift was observed between pre-and post-nutrient addition with DMS concentration as the most influential variable for the small size fraction under pre-nutrient addition, while origin (Fjord vs. mesocosm) was most influential under post-nutrient addition conditions.Oxygen and Si were the most significant structuring variables for the large size fraction for the pre-and post-nutrient addition, respectively.Variables associated with phytoplankton bloom dynamics were most important for structuring the community, especially when looking at the taxonomic shifts between fjord, control mesocosms and elevated pCO 2 mesocosms.The differences were greater after t13 because of the two subsequent phytoplankton blooms that were triggered by the nutrient addition.The differences were most evident in the large size fraction, probably due to the association of the bacterial community with phytoplankton aggregates.Therefore, it is possible to state that nutrients, and the phytoplankton blooms, were the main drivers of microbial community structure in this experiment, which is in agreement with previous (Allgaier et al., 2008;de Kluijver et al., 2010) and present studies (Sperling et al., 2013).

Elevated pCO 2 effect
The effect of elevated pCO 2 on microbial community structure has also been investigated in previous (Newbold et al., 2012) or present mesocosms (Zhang et al., 2012), where no evidence of a major pCO 2 effect on the general bacterial community was found.However, other work suggests that only the community structure of the small size fraction bacteria is significantly affected by elevated pCO 2 (Allgaier et al., 2008).The extensive database of 16S rRNA sequence obtained in this study provided the high resolution necessary to study subtle but significant changes in community structure hinted at in prior studies.In agreement with Allgaier et al. (2008), the effect of elevated pCO 2 in this experiment was slight and only impacted the small size fraction bacteria after nutrient addition, which corresponded to post-nutrient addition and post-bloom conditions (after t13) in this study.This increased post-bloom CO 2 effect was previously observed in other mesocosms experiments (Arnosti et al., 2011;de Kluijver et al., 2010), confirming a possible increased CO 2 effect under nutrient (N, P, Si) limitation.
While pH was shown to be a weak driver of microbial community structure in our experiment, the direct impact of pCO 2 was found to be non-significant, except for 15 rare taxa, which did show a response to elevated CO 2 .Therefore, the level of taxonomic resolution afforded by this study suggests that, in this ecosystem, rare organisms may be disproportionately affected by acidification.The most abundant of these 15 rare taxa was Methylotenera (genus) and had its highest mean abundance in the medium pCO 2 mesocosms (∼ 685 µatm).Species from this genus are generally aerobic, ubiquitous bacteria found in a wide range of O 2 , salinity, temperature and pH.Methylotenera can colonize multiple pH range (5 to 8.5) but grows optimally at pH 7.5 (Kalyuzhnaya et al., 2006;Bosch et al., 2009), suggesting that pH may strongly influence for distribution of this taxa.Indeed, the pH close to this value from t5 until the end of the experiment in the mesocosms with a pCO 2 over ∼ 685 µatm.The highest abundance was found from t22 until t28 where the pH was 7.9 and 7.94.A lower pH was found (pHT 7.57-7.80) in the ∼ 1050 µatm mesocosm but this was not accompanied by an increase in Methylonera abundance, potentially because the pCO 2 concentration itself was toxic to this species at this stage or this could represent mesocosm variability, suggesting a need for improved replication.Functionally, the species included in this genus have been described as bacteria that require organic compounds containing no carbon-carbon bonds (C 1 compounds) like methylamine and/or methanol as energy sources (Lidstrom, 2006;Kalyuzhnaya et al., 2006Kalyuzhnaya et al., , 2010)).These organic compounds play an important role in the global carbon cycle because they are greenhouse gases whose emissions are on a scale similar to methane (Chistoserdova et al., 2009).Further investigation of the behaviour of these C 1 -compounddegraders in response to elevated CO 2 are, therefore, important for understanding biotic influences on climate dynamics.The second most abundant group of the 15 pCO 2 -correlated rare taxa was Colwellia, which is a versatile group with broad temperature range tolerance.For example, the psychrophilic Arctic marine strain Colwellia psychrerythraea grows at a range of temperature from −1 to 10 • C (optimal growth 8 • C), Colwellia chuckchiensis at a range from 0 to 30 • C and Colwellia asteriadis spp. at a range from 4 to 25 • C.These organisms are also capable of colonising a wide range of pH from 4 to 10 (Yu et al., 2011;Choi et al., 2010;Methé et al., 2005).C. psychrerythrea is considered a model organism for psychrophiles and shows multiple molecular adaptations to the cold, like enzymes for cryoprotection, for dissolving high-molecular-weight organic compounds (ex.carbon), for stability in extreme environments (extracellular polymeric substances) and for cold-active processes (Methé et al., 2005;Huston et al., 2004).These features make Colwellia spp.key participants in carbon and nutrient cycling in the cold marine environments.Since some methanogenic enzymes were previously found in Colwellia spp.(Methé, et al., 2005) one can speculate that these compounds were found in greater abundance toward the end of the experiment.This would also support the presence of the Methylotenera, which increased in abundance towards the end of the experiment.Finally the genus Fluviicola, the third most abundant OTU correlated with pCO 2, was dominant in the elevated CO 2 mesocosms (∼ 1058 µatm).Interestingly, Fluviicola was present at the beginning of the experiment but decreased shortly after CO 2 treatment started.The abundance increased under elevated pCO 2 , but stayed low in medium pCO 2 mesocosms and absent in the controls, for both size fractions.Little is known about this genus, making speculations about its ecological role difficult.

Conclusions
In summary, multiple parameters were found to significantly influence the structure of the bacterial community in Svalbard mesocosms.The most influential factors were the origin of the sample (fjord or mesocosms) and nutrient addition.Furthermore, the relative importance of sampling strategy (small versus large size fraction), Si, PO 4 , primary production, temperature, and pH in structuring the community depended greatly on the time at which the community was sampled in relation to the phytoplankton blooms.The direct impact of pCO 2 was found to be significant for only 15 rare taxa and should be further investigated as analysis of low abundance community members is known to be problematic in 16S surveys (Bokulich et al., 2013).If confirmed, this limited pCO 2 effect could have evolutionary consequences creating a shift in the taxa dominance and/or diversity, profoundly affecting the structure of entire community in a high CO 2 world.However, it should be noted that the pCO 2 conditions in which these organisms dominated were super-elevated compared to predicted outcomes for the surface ocean under current climate change scenarios.Furthermore, the evolutionary response of the unicellular eukaryote Emiliania huxleyi to elevated CO 2 was studied by Lohbeck et al. (2012) and showed that only 500 asexualgenerations were necessary to permit evolution either via adaptive changes from diverse genotype selection or via new mutations.It would be interesting to investigate how the bacterial communities from the present mesocosms experiment would evolve faced to extended elevated CO 2 exposure, allowing a longer population growth.
Future work should focus on exploring the functional responses of the community (metagenomics/metatranscriptomics) to evaluate how elevated pCO 2 or OA influence these processes over a longer time period.

Fig. 1 .
Fig. 1.Chlorophyll a (µg/l) concentrations measurements plotted against days, where arrows marked time points analysed in the present study.Figure derived from Schulz et al. (2012).

Fig. 1 .
Fig. 1.Chlorophyll a (µg L −1 ) concentration measurements plotted against days, where arrows mark time points analysed in the present study.Figure derived from Schulz et al. (2013).

Table 1 .
Redundancy analysis showing the significant structuring variables for the whole bacterial community during the postnutrient addition period (t13-t30).Significant values are p < 0.05.

Table 3 .
Analysis of variance (ANOVA

Table 4 .
Bonferroni-corrected g-test of significance (p < 0.05) listing 15 taxa significantly correlated with CO 2 , for both small and large size fraction; bold highlights mark the taxa presented in Fig.4.Greengenes OTU identifiers refer to prokMSA ids in the Greengenes database.