Ocean acidification and marine trace gas emissions

Abstract

The oceanic uptake of man-made CO₂ emissions is resulting in a measureable decrease in the pH of the surface oceans, a process which is predicted to have severe consequences for marine biological and biogeochemical processes [Caldeira K, Wickett ME (2003) Nature 425:365; The Royal Society (2005) Policy Document 12/05 (Royal Society, London)]. Here, we describe results showing how a doubling of current atmospheric CO₂ affects the production of a suite of atmospherically important marine trace gases. Two CO₂ treatments were used during a mesocosm CO₂ perturbation experiment in a Norwegian fjord (present day: ∼380 ppmv and year 2100: ∼750 ppmv), and phytoplankton blooms were stimulated by the addition of nutrients. Seawater trace gas concentrations were monitored over the growth and decline of the blooms, revealing that concentrations of methyl iodide and dimethylsulfide were significantly reduced under high CO_2. Additionally, large reductions in concentrations of other iodocarbons were observed. The response of bromocarbons to high CO₂ was less clear cut. Further research is now required to understand how ocean acidification might impact on global marine trace gas fluxes and how these impacts might feed through to changes in the earth's future climate and atmospheric chemistry.

Human reliance on the combustion of fossil fuels, combined with activities including deforestation and cement production, has resulted in ever-increasing atmospheric CO₂ concentrations (pCO₂). Current pCO₂ is around 380 μatm—the highest level during the last 650,000 years—and is rising at a rate unprecedented in the Holocene (1–3). The oceans constitute a critical sink for CO₂ and, since the 1980s, have absorbed around 30% of all anthropogenic CO₂ (1, 2). The uptake of this additional CO₂ is resulting in a decrease in the pH of surface waters, manifested as increasing H⁺ ion concentrations and decreasing CaCO₃ saturation states, an effect widely termed “ocean acidification” (OA) (1–3). The present average surface ocean pH of ∼8.1 is 0.1 pH units lower than in preindustrial times, representing a 30% increase in the concentration of H⁺ ions (1). Under the Intergovernmental Panel on Climate Change Special Report on Emissions Scenarios CO₂ emissions scenarios, atmospheric CO₂ concentrations are predicted to reach between 550 and 1,000 μatm by the year 2100 (4, 5), accompanied by a drop of surface ocean pH of between 0.2 and 0.5 units (1). Such rapid and dramatic changes to ocean carbonate chemistry are argued to have a detrimental impact on marine biota (2, 6).

Iodo- and bromocarbon gases in surface seawater are a major source of halogens to the marine atmosphere, where they are rapidly oxidized to produce reactive radicals. Iodine oxides play a highly significant role in the photochemical loss of tropospheric ozone, a potent oxidant and greenhouse gas (7, 8). Longer-lived halogen species are involved in the natural regulation of the protective layer of stratospheric ozone (9). Additionally, there is direct evidence that iodine oxides can contribute to particle formation (10) and to the production of cloud condensation nuclei in the coastal marine boundary layer, with potentially significant effects on global radiative forcing. Dimethylsulfide (DMS) is also produced in surface seawater and emitted to the atmosphere. Here, it undergoes rapid oxidation to produce particles which, through direct and indirect interactions with incoming solar radiation, affect planetary albedo, with the potential for climate feedbacks (11). Consequently, changes in the production rate and sea-to-air emission of marine trace gases as a result of OA may have significant impacts on atmospheric chemistry and global climate.

We participated in a community mesocosm CO₂ perturbation experiment to study the effect of OA on marine trace gas production. Six mesocosm enclosures (2-m diameter, 3.5-m depth) were deployed in a fjord in Norway, three representing pCO₂ and pH predicted for 2100 (M1, M2, M3 ∼750 μatm, pH 7.8) and three acting as present-day controls (M4, M5, M6 ∼300 μatm, pH 8.1). CO₂/air mixtures were used to manipulate seawater pH and pCO₂ inside the mesocosms, initially by aeration of the water for 2 days, followed by flushing of the headspace for the remainder of the experiment (Fig. 1 A and B). Nutrients were added to the mesocosm enclosures on May 6 to stimulate blooms of phytoplankton (final concentrations: 17 μmol l⁻¹ nitrate, 1 μmol l⁻¹ phosphate; see Methods for full details). Seawater samples were collected daily from the enclosures for trace gas analyses. The chlorophyll-a data (Fig. 1C) shows the evolution of the bloom, and we refer to two phases: May 10 to 17 (bloom), and May 18 to 23 (postbloom).

Fig. 1.

Temporal changes in pCO₂ (A), pH (B), chlorophyll a (C), and the concentrations (pM) of methyl iodide CH₃I (D), ethyl iodide C₂H₅I (E), diiodomethane CH₂I₂ (F), chloroiodomethane CH₂ClI (G) over the course of the experiment. Data for M3 and M4 only is shown from May 16 onward as the remaining four mesocosms underwent a second aeration on May 15. Chl a concentrations were significantly lower under high CO₂ for the bloom period May 10 to 17 (T = 2.45, P = 0.021). Significant differences were found for CH₃I for the bloom period (T = 2.75, P = 0.012). The postbloom period May 18 to 23 saw reductions in all iodocarbons under high CO₂: CH₃I (−67%), C₂H₅I (−73%), CH₂I₂ (−93%), and CH₂ClI (−59%). pH data calculated using TA and pCO_2. See Tables S1–S5 for details of statistical analyses.

Results and Discussion

Concentrations of chlorophyll a (Fig. 1C) began to respond to nutrient addition from May 10, with rapid increases in all enclosures from this point. Maximum concentrations of 6 to 11 mg m⁻³ were attained on May 13 and 14, coinciding with the greatest abundance of microbial plankton, and concentrations were significantly lower under high CO₂ during the bloom phase. Picoeukaryotes dominated the microbial plankton communities, while flagellates were the greatest contributor to the phytoplankton biomass (Figs. S1 and S2). Under high CO₂, numbers of coccolithophores, large picoeukaryotes, cryptophytes, and Synechococcus were significantly lower than in the present day CO₂ enclosures (Table S1). Similarly, phytoplankton biomass (g C m⁻³) was reduced in high CO₂ M1 relative to present CO₂ M6, with a 28% reduction in total biomass, and notable reductions in diatom (81%), autotrophic dinoflagellate (56%), and ciliate biomass (35%) (Table S2). The demise of the bloom began on May 14 and 15, when chlorophylla concentrations began to decline, accompanied by reductions in most components of the microbial plankton community.

The iodocarbon gases (Fig. 1 D–G) generally showed similar temporal trends and maximum concentrations were observed during the period of May 14 to 18. Although displaying characteristics associated with biological activity, the iodocarbons do not appear to be directly related to phytoplankton growth, as maximum gas concentrations occurred generally after the maxima in chlorophyll a. Furthermore, the timing of initial increases in iodocarbons did not coincide with those of chlorophyll a (May 11): iodomethane (CH₃I) and iodoethane (C₂H₅I) increased from May 8, whereas diiodomethane (CH₂I₂) and chloroiodomethane (CH₂ClI) concentrations began to rise on May 12 and 13, respectively. The data in Fig. 1 strongly suggest that lowered pH leads to a reduction in iodocarbon concentrations. During the bloom phase (May 10–17), the mean concentrations of CH₃I, C₂H₅I and CH₂I₂ were all lower under high CO₂, although only CH₃I showed significant differences (T = 2.75, DF = 22, P = 0.012). Full details of statistical analysis can be found in Tables S3 and S4. There was no difference between treatments for CH₂ClI concentrations during the bloom period. However, during the postbloom phase (May 18–23), all of the iodocarbons exhibited an effect of high CO₂ treatment with average percentage decreases of 67, 73, 93, and 59 for CH₃I, C₂H₅I, CH₂I₂, and CH₂ClI, respectively (Table 1). The differences between treatments were maintained until the end of the experiment, with the exception of CH₂I₂, which returned to its initial concentrations on May 20.

Table 1.

Summary of trace gas, dimethylsulfoniopropionate, and chlorophyll-a data

	Bloom phase May 10–17, all mesocosms			Postbloom phase May 18–23, M3 and M4			Whole experiment May 6–23 (M3 and M4 only after May 15)
	High CO2	Present day	% diff.	High CO2	Present day	% diff.	High CO2	Present day	% diff.
CH3I*	6.9	12.1	−43	3.6	10.7	−67	5.4	9.2	−41
C2H5I*	0.7	1.0	−32	0.3	1.0	−73	0.5	0.7	−32
CH2I2*	197.4	283.4	−30	4.6	63.9	−93	134.6	200.8	−33
CH2ClI*	189.3	207.2	−9	131.1	321.7	−59	136.9	191.0	−28
CHBr3*	41.2	38.1	+7	14.0	12.1	+14	39.8	34.7	+13
CH2Br2*	1.6	1.9	−17	3.29	3.3	+0.3	2.4	2.2	+8
CHBr2Cl*	0.5	0.4	+17	0.7	0.5	+29	0.6	0.5	+22
DMS†	6.1	14.1	−57	15.7	42.0	−63	5.7	14.1	−60
DMSP†	191.7	252.3	−24	182.9	184.3	−0.8	139.0	182.9	−24
Chl a‡	3.2	5.3	−40	3.6	1.8	+49	2.5	3.5	−28

Means and percentage differences of measured variables for high CO₂ treatment and present-day CO₂ treatment for the bloom phase (May 10–17) in all mesocosms, postbloom phase (May 18–23) in M3 and M4, and for the whole experiment.

*pM.

^†nM.

^‡mg m⁻³.

CH₃I and C₂H₅I concentrations under both treatments fell within open-ocean and coastal seawater measurements (12–14), indicating that conditions within the mesocosms were realistic with respect to net production of these gases. C₂H₅I is a minor iodocarbon component of seawater (12), and similarly in this study it made up <1% of the total iodocarbon pool. CH₂I₂ and CH₂ClI concentrations were somewhat elevated compared to most (but not all) oceanic measurements (13, 15) and they dominated the iodocarbon pool, in common with a number of other studies (12, 13, 16). Production of CH₃I is often referred to as “biogenic” (17): directly by macroalgae and phytoplankton (12) and indirectly through a photochemical reaction with organic matter (18). There is little available information on C₂H₅I; however, the present and previous studies have found significant correlations between C₂H₅I and CH₃I, suggesting similar production and removal mechanisms (12, 19, 20) (Fig. S3). CH₂I₂ is considered to have a primarily biogenic source (17). It is subject to rapid photolysis in surface seawater (photolytic lifetime of ∼12 min), with strong evidence that this reaction is an important source of CH₂ClI (21). For the productive mesocosm environment, the processes controlling net iodocarbon production were susceptible to lowered pH. Although chlorophyll-a concentrations were lower under high CO₂, the decrease in iodocarbon concentrations was not simply a manifestation of a general decline in biological productivity. This was most apparent during the postbloom phase when iodocarbon-to-chlorophyll-a ratios showed large differences between treatments. Mean ratios for CH₃I, C₂H₅I, CH₂I₂ and CH₂ClI to chlorophyll a under high CO₂ were 1.1, 0.1, 2.3, and 55.5, respectively, compared to higher values of 5.7, 0.5, 24.7, and 159.9 under present day CO₂. Therefore, despite elevated chlorophyll-a concentrations under high CO₂ during the postbloom phase, net production of iodocarbons was greatly reduced. Such changes may be the result of plankton community shifts. Small picoeukaryotes, cryptophytes, and Synechococcus were all significantly lower under high CO₂, and the difference between treatments was most pronounced during the postbloom phase (Fig. S1 and Table S1). If any of these organisms are involved in iodocarbon production or consumption, changes in their abundance may directly impact seawater concentrations.

The temporal changes in the concentrations of the bromocarbon gases (Fig. 2 A–C) were substantially different to those of the iodocarbons, with large peaks before the development of the bloom, followed by rapid decreases in concentrations. There appeared to be some relationship to biomass over the period of exponential growth. In addition, the bromocarbon concentrations tended to show some increase under high CO₂ (Table 1), and dibromochloromethane (CHBr₂Cl) was significantly elevated (bloom 29% increase, T= −2.82, DF = 33, P = 0.008). However, it is not clear whether these differences can be attributed to an effect of pCO₂ and further study is now needed.

Fig. 2.

Temporal changes in the concentrations (pM) of bromoform CHBr₃ (A), dibromomethane CH₂Br₂ (B), and dibromochloromethane CHBr₂Cl (C) over the course of the experiment. For the bloom period of May 10 to 17, significant differences were found for CHBr₂Cl (T = −2.82, P = 0.008). The postbloom period of May 18 to 23 saw increases in CHBr₃ (+14%) and CHBr₂Cl (+29%) in the high CO₂ M3, and CH₂Br₂ (+0.3%). See Tables S1–S5 for details of statistical analyses.

Our findings for DMS and its precursor, dimethylsulfoniopropionate (DMSP), are shown in Fig. 3. In all mesocosms, there was an overall increase in DMS concentrations and the temporal trends observed in the two treatments were markedly dissimilar. Under ambient CO₂, both DMS and chlorophyll a peaked on May 13 and 14, whereas the high CO₂ DMS displayed only a gentle rise, which did not concur with the trend in chlorophyll a (Figs. 1B and 3A). A large and statistically significant 57% reduction in DMS concentrations was observed under high CO₂ for the bloom period of the experiment (T = 4.75, P < 0.001) (Table 1). By the end of the experiment, DMS was threefold higher under present day CO₂ compared to the high CO₂ treatment. DMSP concentrations were generally lower under high CO₂ (24%) (Table 1 and Fig. 3B) and displayed significant differences between treatments for the bloom period (T = 2.18, P = 0.038). DMS and DMSP concentrations are comparable to summertime measurements from nearshore United Kingdom waters (22), and well within the range of measurements during northeast Atlantic coccolithophore blooms (23). This finding provides confirmation of the applicability of mesocosm data to the real world. Our results for DMS are in strong agreement with a mesocosm study in 2004 (24) that showed substantially lower DMS production under high CO₂ (confirmed by a series of in vitro studies). However, the results differ with those from a mesocosm experiment in 2005, which showed only small differences in DMS between elevated and ambient CO₂ treatments (25, 26). The plankton communities of the present study were dominated by flagellates and picoeukaryotes, while the 2005 study was coccolithophore-dominated. It is likely that such differences in community composition may have resulted in the different DMS dynamics seen between the two experiments.

Fig. 3.

DMS (A) and DMSP(B) concentrations (nM) over the course of the experiment. Both DMS and DMSP concentrations for the bloom period of May 10 to 17 were significantly lower under the high CO₂ treatment (DMS T = 4.75, P < 0.001; DMSP T = 2.18, P = 0.038). The postbloom period of May 18 to 23 saw reductions in DMS (−63%) and DMSP (−0.8%) in the high CO₂ M3 relative to M4. See Tables S1–S5 for details of statistical analyses.

There is considerably more knowledge of the production and removal mechanisms for DMS than for any of the halocarbons. Production of DMS in seawater arises by processes, including active exudation, viral lysis, and grazing by zooplankton, but bacterial consumption is an important removal path (8). These processes act on the phytoplankton cell constituent DMSP, a compound synthesized by a range of phytoplankton species including prymnesiophytes, such as coccolithophores, and dinoflagellates (11). In the communities of the present experiment, these groups of phytoplankton did not dominate [average % total C: prymnesiophytes 16%, dinoflagellates 8%, (Fig S2), coccolithophores 3% of total C, 9% of flagellate biomass (Fig. S4)], thus it is not easy to attribute DMSP production to any particular group of phytoplankton. Because DMSP was less markedly affected by high CO₂, this finding suggests that the change in pH may have more impact on the processes that control the conversion of DMSP to DMS rather than the initial production of DMSP itself. Our observed decrease in DMS can be related to climate-model simulations, which show that a 50% decrease in ocean DMS emissions can result in enough net cloud radiative forcing to increase surface air temperature by 1.6°C (27).

Conclusions

Through measurement of realistic concentrations of trace gases during a mesocosm experiment, we have shown that experimentally simulated OA can impact on the production of globally important marine trace gases. Although it would be inappropriate to extrapolate our data to the global oceans, the results of this study are relevant to high productivity areas of the oceans in the high-latitude oceans, coastal waters, and upwelling regions. Not only are these regions of great global importance for the production of biogenic gases, they are expected to experience the greatest changes in terms of anthropogenic ocean acidification and global climate change over the 21st century (5, 28–30).

To assess the impact on global radiative forcing and the atmosphere’s oxidative capacity, the complexity of the ocean-atmosphere system must be taken into account. The future ocean will not only be subjected to lowered pH, it will also face physical changes to the climate and ocean system. In combination, all could impact on the sea-to-air flux of trace gases (2, 5). We believe that our findings reveal important effects, and further study is now needed to investigate the response of plankton communities from a range of geographical regions. When incorporated into global ocean-atmosphere modeling studies, information of this kind will be vital to improving prediction of the earth's future climate and atmospheric chemistry.

Methods

The mesocosm experiment was carried out at the Marine Biological Station, Espegrend, Bergen, Norway (Raunefjord, 60.3°N, 5.2°E) from May 3 to 23, 2006. Six reinforced polyethylene enclosures (2-m diameter, 3.5-m depth, 0.5-m above surface of the water) with a total volume 11,000 m³ were used. The mesocosms were covered with lids consisting of a plastic frame covered in high UV-transmitting horticultural polyethylene, attached to the mesocosms through a system of ropes and karabiners. Although the lids were not intended to be leak-tight, they reduced exchange of the enclosure headspaces with the atmosphere and prevented rain entering. The transmission of photosynthetically active radiation was measured at 92%. The enclosures were attached to the southern side of a raft ≈200-m away from the shore. On May 2, the enclosures were filled with nutrient-deplete, unfiltered water, pumped directly from the fjord. The development of phytoplankton blooms was stimulated by addition of nutrients on May 6 such that initial concentrations in all bags were 1 μmol l⁻¹ phosphate and 17 μmol l⁻¹ nitrate.

Two treatments were used to assess the effect of high CO₂ on a phytoplankton bloom: “High-CO₂” (M1–M3), and “Present Day” (M4–M6). Water in the high CO₂ enclosures was aerated with ambient air enriched with CO₂ for 1 to 2 days (May 4–6) until pH ∼7.8 was attained. From May 6 onwards, the headspaces alone were flushed with high-CO₂ air, thus maintaining the atmospheric concentration and allowing the carbonate system to develop and adjust naturally. The Present Day mesocosms were treated identically, except with the use of ambient air. Discrete samples were collected daily, between 0900 hours and 1000 hours.

Between May 10 and 14, during a period of rapid phytoplankton growth, the pCO₂ gradually decreased back down to near ambient levels, with a concomitant rise in pH (see Fig. 1 A–C). Consequently, on May 15, M1 and M2 were reaerated with CO₂ to bring the pH back down to target levels. M3 was left untouched to see what would happen if the experiment had continued unaltered. The present day M4 was also left unaltered, and M5 and M6 were reaerated with air to receive the same physical treatment as M1 and M2. The experiment was continued following this additional treatment until May 23, when sampling ceased.

Mesocosm Sampling.

Water sample collection for trace gas analysis was undertaken daily (except May 17 because of technical problems) commencing on May 6 for DMS and May 8 for halocarbons. One sample was taken from all enclosures; triplicate samples were taken from M1 and M6 on alternate days to test analytical reproducibility (Table S5). Water samples for DMS, DMSP, and chlorophyll-a analysis were taken with 5-L plastic aspirators. The aspirators were inverted, and then slowly pushed into the water with taps open so that they gently filled up with the least addition of bubbles. A 200-μm mesh was used to cover the opening of the aspirator to exclude large zooplankton and detritus from entering the aspirator. While still on the raft, subsamples were transferred from the aspirators to glass-stoppered bottles using Tygon tubing, which was attached to the tap of the aspirators. To minimize exposure of the water to air during transfer, the tube was placed at the bottom of the bottle. An initial large aliquot was shaken around the bottle to rinse it, and then rejected. Water was then allowed to fill from the bottom of the bottle and sample water was allowed to overflow for an estimated three times the volume of the bottle. The glass-stopper was firmly placed onto the bottle, ensuring the presence of no headspace or bubbles. Each bottle was stored in the dark until all sampling was completed. On return to the onshore laboratory, the samples were kept in a dark, constant-temperature room, maintained at the ambient temperature of the fjord (9–11°C) until analysis.

Samples for halocarbon analyses were taken using 100-mL glass syringes fitted with a syringe extension, consisting of a nylon tube and stopcock. Samples were taken from 20-cm subsurface. Before collection of each sample, ≈50 mL of seawater was drawn up to thoroughly rinse the tubing and syringe, and then rejected. After a second rinse, the sample was gently drawn into the syringe ensuring minimum addition of air bubbles. Immediately after sampling, the glass syringes were placed in a dark box.

Seawater Trace Gas Extraction and Preconcentration.

Seawater samples were analyzed similarly to the method of Chuck et al. (14), using purge and trap gas chromatographic systems with flame photometric detector for DMS and mass selective detection (MSD) for halocarbons. Water samples (20 mL for DMS, 40 mL for halocarbons) were analyzed as soon after sample collection as possible, invariably within 2 h of collection. The purge and trap system for DMS was constructed of glass and PTFE, and that for halocarbons of glass and stainless steel. Seawater samples were purged with purified oxygen-free nitrogen at 60 mL min⁻¹ for 15 min for DMS, and 40 mL min⁻¹ for 20 min for halocarbons. The gases were trapped in a nonpacked sample loop, which was cooled in the headspace of a liquid nitrogen-filled dewar, thermostatically held at −150°C. Analytes were injected onto the GC columns by heating the trap to ∼100°C using boiling water.

Halocarbon Analysis.

Halocarbon samples were filtered before analysis, through a 47-mm GF/F filter (Whatman) between two syringes, with an inline Swinnex filtration unit. It was ensured that introduction of air was minimized during the process. Once filtered, the sample was injected with 2 μL of the surrogate analyte mixture and immediately injected into the glass purge vessel of the purge and trap system.

Samples were analyzed using an Agilent gas chromatograph (GC 6890N) and Mass Selective Detector (5975 Series MSD), with a 60-m DB-VRX capillary column installed (0.32 μm film thickness, J & W Ltd.). Following introduction of the samples into the GC, the oven was held at 40°C for 2 min, heated up to 130°C at 8°C min⁻¹, then 200°C at 60°C min⁻¹, and held for 2 min, and finally heated up to 240°C at 60°C min⁻¹ and held for a 2 min. The total runtime was 19.08 min, and the data were collected between 3 and 13 min of the run. The MSD was operated in electron ionization (EI)/ single ion mode (SIM) throughout the analyses. To monitor and correct system drift, two surrogate analytes (CD₃I and CD₃CDICD₃) were injected into each sample before analysis. For the compounds with the highest sensitivities (CH₃I and C₂H₅I) CD₃I was used, and CD₃CDICD₃ was used for the compounds with lower sensitivities (CH₂ClI, CH₂I₂, CHBr₃, CH₂Br₂, CHBr₂Cl). Surrogate analytes were prepared gravimetrically from liquid standards and diluted in HPLC-grade methanol to concentrations of 0.7 pM μL⁻¹ CD₃I and 0.9 pM μL⁻¹ CD₃CDICD_3. Both surrogate analytes were diluted into one 4-mL amber vial. Immediately before analysis, 2 μL of the surrogate analyte mixture was directly injected into 40-mL seawater samples using a Hamilton constant rate syringe. Calibration and quantification of the compounds was performed using liquid standards prepared in our laboratory. Standards were made up by dilution of the pure compounds into HPLC-grade methanol. The primary standard was prepared gravimetrically, the secondary and working standard by serial dilution. The analytical error for each compound as based on triplicate samples from M1 and M6 were as follows: CH₃I:15%, C₂H₅I:14%, CH₂I₂:15%, CH₂ClI:9%, CHBr₃:7%, CH₂Br₂:17%, CHBr₂Cl:8%.

DMS and DMSP Analysis.

DMS samples were analyzed using a Shimadzu GC-14B with flame photometric detector, with a Shimadzu C-5A Chromatopac integrator. The GC was equipped with a Chromosil 330 (Supelco) packed column, with the oven set at 60°C, the injector set at 150°C, and the detector at 175°C. A 20-mL sample was drawn into a glass gas-tight syringe (20 mL), and slow-filtered through a 25-mm GF/F filter in a Swinnex filtration unit directly into the glass purge vessel for purge and trap analysis. Following injection onto the column, the retention time for DMS was around 1.2 to 1.3 min. Calibration of DMS was performed every 3 to 5 days by cold alkaline hydrolysis of DMSP (31) (0.17–87.6 nM L⁻¹) with 10M NaOH. The analytical error of the system was 6%, as based on triplicate samples from M1 and M6.

For particulate DMSP (DMSPp), the filter paper used to filter the DMS sample was placed into a glass vial containing 15 mL of 500 nM NaOH and immediately capped with a crimp seal, cleaving all DMSP to DMS via alkaline hydrolysis. These samples were analyzed for DMSPp on return to the United Kingdom by headspace analysis on a Shimadzu GC-2020 with flame photometric detector. Samples of the headspace, ranging from 50 to 250 μL, were taken using a 250-μL Hamilton gas-tight syringe and manually injected into the GC. Calibration of the system was performed using DMSP standard diluted in 500 mM NaOH, in identical 20-mL glass vials, and crimp seals to those used for the samples and over a concentration range of 0.5 to 300 nmol L⁻¹. The analytical error of the system was 11%, based on replicate samples.

pCO₂ and pH Determination.

Discrete aqueous samples for the partial pressure of carbon dioxide (pCO₂) were taken in 500-mL volumetric flasks from May 3 to 24, 2006. The samples were analyzed within 14 h of collection using a University of East Anglia-built instrument with infrared detection (32). The analysis temperature was within 2°C of the in situ temperature. The CO₂ instrument was calibrated twice daily against secondary standards with CO₂ mixing ratios of 0, 258.40, 470.32, 682.72, and 877.19 μmol mol⁻¹. These secondary standards had been calibrated against National Oceanic and Atmospheric Administration CO₂ standards. The program CO2sys (33), with the equations of Mehrbach et al. (34) as refitted by Dickson and Millero (35), was used to correct for the sample headspace and the temperature difference between sampling and analysis. The pCO₂ was calculated both for the in situ temperature and for a constant temperature of 8°C. The accuracy of pCO₂ is estimated as better than 5 μatm. The average difference between eight replicate samples infers a reproducibility of 3 μatm.

Daily samples for total alkalinity (TA) were taken, filtered, and fixed with mercuric chloride from May 11 to 20. TA was determined by potentiometric titration with a Vindta system at the University of East Anglia. The constants of Prieto and Millero (36) and multiple least-squares fitting were used. Analysis of certified reference material suggests an accuracy and reproducibility of 4 and 2 μmol/kg, respectively. Outliers in TA were replaced by values interpolated from nearby days. TA values for May 3 to 10 have been extrapolated from TA on May 11 and an empirical relationship between TA and coccolithophorid numbers. TA varied little between treatments over this period. pH on the total pH scale was calculated from TA and pCO₂ with CO2SYS (33) with the equations of Mehrbach et al. (34) and Dickson and Millero (35). The variation in calculated pH is dominated by variation in pCO₂, with less effect from changes in TA.

Fig. 1 A and B show the changes in pCO₂ and pH throughout the course of the experiment. On May 3 and 4, the effect of bubbling with CO₂ was clear in M1, M2, and M3 with a rapid increase in pCO₂, up to 660 to 736 μatm, accompanied by a sharp drop in pH to 7.8. The rebubbling with CO₂ on May 15 of M1 and M2 is clearly seen as a sharp rise in pCO₂, up to 823 μatm, in M2. The pH dropped to 7.9 and 7.7 in M1 and M2, respectively. pCO₂ and pH in M3 remained stable for the rest of the experiment, with a mean pCO₂ of 358 μatm for this period, and a stable pH of just over 8.1. M3 could still be considered to represent a CO₂-perturbed environment, as pCO₂ levels were on average 80 μatm higher than in the Present Day enclosures, a result of the headspace still being flushed with high-CO₂ air. pCO₂ in the Present Day mesocosms was on average lower than the fjord itself, with a mean of 250 μatm for the duration of the experiment, compared to 276 μatm for the fjord. This difference arose from higher phytoplankton productivity inside the mesocosms stimulated by the addition of nutrients.

Chlorophyll a Determination.

Water samples were taken daily from the 5-L aspirators and 350 to 500 mL was filtered through 0.2-μm cellulose acetate filters. The filters were folded four times, placed in a cryovial, and shock-frozen in liquid nitrogen and stored at −80°C until analysis. To extract chlorophyll a, filters were placed in 10-mL 90% acetone and left for 24 h at −20°C, then centrifuged at 850 × g for 2 min. Samples were diluted in 90% acetone, with dilution factors of between 5 and 20, and the fluorescence was measured on a Turner Fluorometer. Calibration was performed using chlorophyll-a standard diluted in 90% acetone with a concentration range between 1 mg m⁻³ and 20 mg m⁻³.

Phytoplankton Microscopy Counts.

One hundred-milliliter samples from M1 and M6 were preserved with acid Lugol‘s iodine solution (2% final concentration) and stored in cool, dark conditions until analysis in the laboratory by inverted settlement microscopy (37). Fifty-milliliter subsamples were concentrated by sedimentation for > 24 h and all cells between 2 and 200 μm were enumerated at ×200 or ×400 magnification. Cells were identified, where possible, to species level and their linear dimensions were measured using an ocular micrometer. Cell volumes were calculated using simple geometric shapes and converted to carbon according to the equations of Menden-Deuer and Lessard (38).

Analytical Flow Cytometry.

Samples were fixed with 1% paraformaldehyde and used to enumerate microbial plankton using flow cytometry. Synechococcus, Prochlorococcus, and algae were enumerated in unstained samples with a FACSort flow cytometer (Becton Dickinson) using their specific chlorophyll/phycoerythrin autofluorescence (39). Abundance of heterotrophic bacteria was determined after staining with SYBR Green I DNA dye (40, 41). Yellow-green beads of 0.5-μm diameter (Fluoesbrite Microparticles; Polysciences) were used in all analyses as an internal standard for both fluorescence and flow rates (42). Coccolithophore cell volumes were calculated using simple geometric shapes and converted to carbon (mg C m⁻³) according to the equations of Menden-Deuer and Lessard (38).

Statistical Analyses.

All trace gas, DMSP, and chlorophyll-a data were analyzed using two-sample tests of hypotheses. Initially, tests of normality were applied, and if data failed to fit the assumptions of the test, square root transformations of the data were performed. The analysis culminated in a two-sample t test. For those data which still failed to display normality following square root transformation, nonparametric tests were applied. Summaries of statistical analyses can be found in Tables S1–S5.