Diel patterns of oceanic dimethylsulfide (DMS) cycling: Microbial and physical drivers



[1] Dimethylsulfide (DMS) is a biogenic gas with potential climatic effects, and its marine emission depends on the interplay between microbial activity and physical forcing in the oceanic upper mixed layer. We investigated the diel cycling patterns of DMS and its precursor dimethylsulfoniopropionate (DMSP) in four experiments (28 to 48 h long) performed in mesotrophic to ultraoligotrophic Mediterranean and Sargasso Sea waters. Samples taken every 4 or 6 h were analyzed for dimethylated sulfur pools and incubated to measure DMS and DMSP cycling rates, as well as primary and bacterial production. In all four experiments, DMS budgets showed pronounced day versus night variability. In the three summer experiments, gross community DMS production (GPDMS) increased by twofold to threefold from nighttime to daytime, peaking 0–4 h after solar noon. This excess GPDMS was balanced by higher photochemical and microbial sinks during the day, effectively buffering DMS concentrations. In the only winter experiment, GPDMS exhibited opposed temporal dynamics and peaked at nighttime in parallel to total DMSP consumption. Community DMSP to DMS conversion yields were generally <10% throughout the winter experiment and at night in summer, and increased to >15% (even >50%) during the day in summer, presumably due to phytoplankton radiative stress. Our data suggest that (1) diel variability should be taken into account in process studies, diagnostic, and prognostic models of DMS cycling and (2) the community DMS yield is a key variable that defines characteristic DMS cycling regimes.

1 Introduction

[2] The dimethylated sulfur compound dimethylsulfoniopropionate (DMSP) is produced in varying amounts by diverse marine phytoplankton groups [Stefels et al., 2007]. Besides osmoregulation, intracellular DMSP serves distinct physiological functions related to environmental stress, particularly in response to high radiation levels and nutrient starvation [Stefels et al., 2000; Sunda et al., 2002, 2007]. Planktonic DMSP metabolism is the main source of the volatile dimethylsulfide (DMS), produced either through direct enzymatic DMSP cleavage by phytoplankton or through DMSP transformations mediated by microbial food web interactions [Simó, 2004; Stefels et al., 2007]. The latter involve bacterial DMSP catabolism, which may channel DMSP to DMS or, more often, divert it to other volatile and nonvolatile compounds [Moran et al., 2012].

[3] DMS emission plays a potentially important role in aerosol formation, growth, and chemistry, to the point that it might influence cloud microphysics in remote marine regions [Andreae and Rosenfeld, 2008; Lana et al., 2012] and induce significant radiative forcing at regional scales. Furthermore, marine DMS emission has been proposed to drive a plankton-climate feedback loop [Charlson et al., 1987], a hypothesis that remains controversial [Quinn and Bates, 2011]. Because DMS is generally greatly supersaturated in the ocean with respect to the atmosphere, its emission depends largely on sea-surface DMS concentrations and wind speed. However, DMS ventilation generally represents a nondominant sink compared to other competing DMS consumption pathways in the oceanic upper mixed layer (UML): microbial (bacterial) DMS consumption and photochemical oxidation [Simó, 2004; Toole et al., 2006]. Thus, it follows that the interplay between biotic and abiotic DMS sinks and gross community DMS production within the UML determines how much DMS ends up in the marine troposphere.

[4] Day-night alternation and the underwater light exposure regime (the latter modulated by meteorological variability and vertical mixing) exert an obvious rhythmic forcing on photochemical and photobiological processes [Doney et al., 1995]. In turn, these may couple or decouple, to amplify or buffer the diel oscillations of biogeochemical fluxes. Since light-driven processes are key in the biogeochemical cycling of DMS and its precursor DMSP [Toole et al., 2006; Galí et al., 2011; Lizotte et al., 2012; Miles et al., 2012], diel variability is to be expected; yet studies aimed at understanding this relevant time scale are surprisingly scarce. Diel studies of DMS(P) cycling can also provide significant insight into coupled physical-microbial ecosystem functioning, given the important role of DMSP in planktonic metabolism [Kiene et al., 2000; Simó, 2004; Seymour et al., 2010], and the relevance of DMS and DMSP as model compounds. Here we report four diel cycle experiments conducted in different oceanographic settings, revealing a marked diel variability of biotic and abiotic DMS cycling which translated into highly variable but nearly balanced surface ocean DMS budgets.

2 Methods

2.1 Overview

[5] We performed four diel cycle studies of 28–48 h duration, as summarized in Table 1. The wintertime study (CMEDwin) took place in March 2004 at the Blanes Bay Microbial Observatory (BBMO), a thoroughly studied NW Mediterranean coastal site. The remaining three studies were conducted in summer during oceanographic cruises to the Sargasso (Biocomplexity cruise on board the R/V Seward Johnson) and the Mediterranean Sea (Modivus cruise, R/V García del Cid). In the Sargasso Sea cruise, an anticyclonic eddy (named A2) was followed for 6 days by deploying Lagrangian drifters, and the diel cycle experiment (SARGsum) was conducted during the last 2 days. In the Mediterranean Sea cruise, a first experiment (CMEDsum) was conducted 1 km offshore from the regular BBMO sampling site, and a second at an open ocean station (MEDsum), located approximately 120 km south of the BBMO where the water depth was approximately 2000 m.

Table 1. General Description of the Diel Cycle Experiments
ExperimentMed Winter (Coastal)Med Summer (Coastal)Med Summer (Oceanic)Sargasso Summer (Oceanic)
  1. a

    See Bailey et al. and Gabric et al. [2008] for details on the trajectory of the Lagrangian drifters in SARGsum.

Date22–24 March 200418–20 September 200723–25 September 200731 July–1 August 2004
Positiona41°39.9′N, 2°48.3′E41°39.1′N, 2°48.0′E40°39.1′N, 2°51.0′E29°0.8′N, 63°36.0′W
Trophic statusmesotrophicoligotrophicoligotrophicultraoligotrophic
Bottom depth (m)24302000>4000
Experiment duration (h)48484828
Time between samplings (h)6444
Sampling depth (m)0.5453
Type of incubationdarkdark-lightdark-lightdark

[6] During CMEDwin, surface seawater was sampled from a boat every 6 h and brought to the lab within 2 h for further processing. In the ship-based experiments, sampling was done every 4 h from Niskin bottles attached to a conductivity-temperature-depth (CTD) rosette, and the samples were immediately processed. All the diel cycle experiments proceeded similarly: After taking aliquots to determine dimethylated sulfur concentrations and ancillary microbial parameters at time zero, incubation bottles were filled with unfiltered water and immediately incubated to measure biological sulfur cycling rates plus autotrophic and heterotrophic microbial activities. All the biological process incubations (unless otherwise noted) lasted for approximately 6 h, and, thus, successive incubations overlapped. The samples were incubated in the dark at in situ temperature, either in a thermostatted chamber (CMEDwin) or in a tank with running seawater from the ship underway intake. In CMEDsum and MEDsum, additional incubations of whole and 0.2 µm filtered seawater were done in UV-transparent polytetrafluoroethylene (Teflon) bottles during the day, to determine the response of microbial and photochemical processes to sunlight (see sections 2.4 and 2.5).

[7] Some of the data used here, corresponding to the 2004 Sargasso Sea cruise, were already published (as noted throughout the paper). The reader is referred to the works by Blomquist et al. [2006], Bailey et al. [2008], and Gabric et al. [2008] for a comprehensive description of the oceanographic setting and DMS cycling during that cruise.

2.2 Oceanographic and Meteorological Data

[8] Vertical profiles of conductivity (salinity) and temperature in the upper 200 m were obtained every 4 h with a CTD probe (Seabird 9/11 plus) equipped with sensors of chlorophyll fluorescence, turbidity, beam attenuation, and dissolved oxygen. The SEASOFT software (Seabird) was used to calculate seawater density (σt) and to bin the profiles at 1 m intervals. Binned profiles were used to calculate the mixed layer depth (MLD), defined as the depth where the temperature difference to that at 4 m was >0.02°C (Figure 1; 4 m was the shallowest depth available in all CTD profiles). This fine MLD criterion is likely to capture the actively mixing layer and diurnal stratification events [Brainerd and Gregg, 1995]. In CMEDwin, temperature profiles were obtained with a manual thermometer. Due to the coarser vertical resolution, we used a 0.05°C criterion and 2 m as the reference depth, which rendered more robust MLD values.

Figure 1.

Total shortwave irradiance (first row), and vertical-temporal variability of temperature (T; second row) and chlorophyll (Chl) fluorescence profiles (Flu.; third row) in the summer experiments. The white line is the depth of the mixing layer, and the black horizontal line the average 10% penetration depth of 340 nm radiation. Arrows indicate time points of the DMS profiles shown in Figure 7. See Table 1 for a description of experiment locations and nomenclature.

[9] Underwater irradiance profiles in six bands in the ultraviolet region (UV; 305, 313, 320, 340, 380, and 395 nm) and one integrated band for photosynthetically available radiation (PAR; 400–700 nm) were obtained by deploying a PUV 2500 radiometer (Biospherical) 2 to 4 h before and after the solar noon. Diffuse attenuation coefficients of downwelling cosine irradiance (Kd) were calculated from the linear regression between logarithm-transformed irradiance and depth (within the UML or a deeper, optically homogeneous surface layer). Meteorological data were recorded by either shipboard stations or a nearby land-based station (CMEDwin; Malgrat de Mar, Catalan Meteorological Service, SMC). During sunlit incubations, the PUV 2500 radiometer was placed inside the incubator, with the optical window located at the same depth as the incubation bottles (~10 cm), to take a continuous record of downwelling irradiance. Care was taken to use black incubation tanks to avoid upward light reflection. By combining the time series of spectral irradiance, CTD-derived mixing depths, and underwater light attenuation, we obtained 5 min and 4 h resolution time series of spectral irradiance and radiation doses within the UML. These were used to estimate in situ DMS photolysis (section 2.6) and the effects of sunlight on microbial processes.

2.3 Chemical Analyses and Determination of Microbial Abundance

[10] DMS was analyzed by purge and trap gas chromatography coupled to flame photometric detection (GC-FPD), with a detection limit of approximately 3 pmol and analytical precision better than 5% in the low nanomolar range. Total DMSP (DMSPt), comprising the particulate (DMSPp) and dissolved (DMSPd) pools, was analyzed as the DMS evolved by alkaline DMSP cleavage in whole water samples, as described elsewhere [Galí et al., 2013a]. Chlorophyll a (Chl a) was determined fluorometrically after filtration onto GF/F filters and overnight extraction in acetone. Flow cytometry was used to enumerate heterotrophic bacteria after DNA staining with SybrGreen I [Gasol and del Giorgio, 2000], and picophytoplankton and nanophytoplankton populations in live samples (Prochlorococcus, Synechococcus, picoeukaryotes, and nanoeukaryotes) [Marie and Partensky, 2006]. Microphytoplankton species were identified and counted with an inverted microscope in the three Mediterranean Sea experiments.

2.4 Microbial Activity Parameters Determined With Radioisotope Additions

[11] Particulate primary production (PPp) was estimated from the uptake of H14CO3 addition following standard procedures [Morán et al., 1999]. Incubation irradiance levels were adjusted with neutral density screens, so that we could estimate the PPp rates corresponding to the average UML light levels. Since different incubation setups were employed in CMEDwin (artificial light source), and (C)MEDsum and SARGsum (on-deck incubators), we will use the estimated PPp rates to explore possible relationships between microbial activities, but not to make quantitative inferences. Bacterial heterotrophic activity was estimated with the 3H-leucine incorporation method [Kirchman et al., 1985]. Triplicate samples plus one killed control were incubated in the dark for 2 h and processed by the centrifugation method of Smith and Azam [1992]. Size-fractionated microbial assimilation of DMSP-sulfur into macromolecules (>0.22, >0.65, and >3 µm fractions) was measured by trace additions of 35S-DMSP, only in CMEDsum and MEDsum. The samples for DMSP-sulfur assimilation were incubated in 50 mL quartz flasks (thus, under full spectrum sunlight during the day) and processed as explained elsewhere [Ruiz-González et al., 2012a].

2.5 GC-based DMS(P) Cycling Incubations

[12] The inhibitor method [Wolfe and Kiene, 1993; Simó et al., 2000] was employed to estimate gross community DMS production (GPDMS) and bacterial DMS consumption rates (BCDMS). Briefly, GPDMS was calculated as the rate of DMS accumulation over time in bottles amended with 200 nmol L−1 dimethyldisulfide (DMDS, final concentration). Net DMS production rates (NPDMS) were determined in unamended bottles incubated in parallel, so that BCDMS could be determined as the difference between GPDMS and NPDMS. Only initial and final concentrations were measured; previous work had shown that linear changes in concentration occur, particularly in short-term (<10 h) incubations [Simó et al., 2000; Saló et al., 2010]. Darkened 2.9 L amber glass bottles were used for the dark incubations. Light incubations were done in 2.3 L polytetrafluoroethylene (Teflon, Nalgene) bottles, which transmitted 65%, 77%, and 100% of UVB, UVA, and PAR, respectively [Galí et al., 2013a]. The bottles were additionally covered by one layer of neutral screen (62% transmission) to better approximate the mean spectral irradiance in the UML. Although no treatment replicates existed, the large incubation volumes ensured a community-inclusive incubation which minimized the experimental error [Galí et al., 2011; 2013a]. GPDMS and NPDMS rates obtained in sunlight were corrected to account for photochemical DMS loss [Brimblecombe and Shooter, 1986], using the photolysis rate constants measured in each experiment (see below) scaled to the radiation dose received by each incubation, following Galí et al. [2011].

[13] Total microbial DMSP consumption was measured with the “net-loss curve” approach of Simó et al. [2000] in the same whole water dark incubations where net DMS production was determined (no DMDS added). Briefly, the logarithm of fractional DMSPt loss (final/initial) was divided by the incubation time. The resulting DMSPt consumption rate constant (h−1) was multiplied by the DMSPt concentration in situ, yielding the consumption rate (nmol L−1 h−1). Note that this measurement quantifies the amount of DMSPt consumed by the whole community, that is, the sum of particulate and dissolved DMSP consumption. The community DMS yield can be calculated as the quotient between GPDMS and DMSPt consumption. Thus, the community DMS yield includes the DMS yield from microbial DMSPd transformation (in the sense of Kiene and Linn [2000]) and from particulate DMSP transformation. The DMSPt consumption rate was further combined with in situ DMSPt variations to estimate, by budgeting, the DMSPt production [Simó and Pedrós-Alió, 1999a].

[14] Pseudo first-order DMS photolysis rate constants (kphoto,incub; h−1) were determined by incubating 0.2 µm filtered water under subsurface irradiance in the on-deck incubator. In the Mediterranean Sea cruise, bulk photochemical DMS loss was estimated as described in Galí et al. [2011], with the difference that gaseous DMS was added at concentrations up to 60 nmol L−1 to ensure better detection. In this DMS concentration range, first-order kinetics should be expected [Kieber et al. 1996]. In the CMEDwin experiment, no photolysis measurements were available. Thus, we decided to use an average wintertime photolysis rate constant obtained from published studies done at the same site [Vila-Costa et al., 2008; Galí et al., 2013a]. In the Sargasso Sea cruise, kphoto,incub were obtained with the 35S-DMS tracer method (details in Bailey et al. [2008, and references therein]).

2.6 Sulfur Cycling Budgets in the UML

[15] Sulfur cycling budgets were calculated by extrapolating the measured DMS cycling rates (GPDMS, BCDMS, and photolysis) and the parameterized DMS ventilation to the whole UML. We assumed homogeneous DMS concentrations and biological activity throughout the UML. At any time interval, the DMS budget equation has the form [Gabric et al., 2001]:

display math(1)

[16] Photolysis and ventilation rates were calculated as the product of in situ DMS concentration and the respective rate constants (k, which are a function of meteorological forcing) and averaged throughout the UML as explained in the following paragraphs. To this end, the 1 min resolution irradiance and wind speed time series were binned to 5 min, and the DMS and MLD measured every 4 h were linearly interpolated to match the same time interval. GPDMS and BCDMS rates were directly obtained from the incubations, lagged by 2 h from the sampling time. This time lag was chosen for all the experiments because it represented a good compromise between the sampling time and the length of the incubations. Lagged correlations showed that the best fit between diagnosed and in situ net DMS production generally occurred at around 2 h lag, and always between 0 and 4 h. We used the rates determined in sunlight when possible (CMEDsum and MEDsum). Finally, the DMS budgets were calculated for the day versus night period and also at 1 h temporal resolution. The latter required temporal interpolation of biological DMS cycling rates (originally measured every 4–6 h).

[17] To obtain the UML-averaged photolysis rate constants (kphoto,UML), experimentally determined kphoto,incub were normalized to the average shortwave irradiance received during the incubation and then scaled to the UML-averaged scalar irradiance at time t (Eo,UML; for clarity, time is omitted from the notation):

display math(2)

[18] Eo,UML was calculated as the vertical integral of exponentially decreasing irradiance in the UML [Vallina and Simó, 2007]. We assumed that photolysis propagated downward with the attenuation coefficient of 340 nm (instead of that centered at PAR), because this wavelength has been shown to represent reasonably the spectral peak of DMS photolysis within the mixing layer in UV-transparent waters [Toole et al., 2003; Galí et al., 2013a]. Above-water total shortwave irradiance was converted to 340 nm spectral irradiance at the water subsurface using an empirical equation of quadratic form deduced from simultaneous pyranometer and PUV 2500 measurements (R2 = 0.99, n = 694, minute-averaged irradiance). The scalar irradiance was calculated from downwelling irradiance (Ed,UML), corrected by the cosine of underwater radiance (μ) to better take into account the tridimensional light field:

display math(3)

[19] Our calculations indicated that a value of μ = 0.80 was a good approximation for the wavelengths and the optical characteristics of the waters studied [Bannister, 1992; Morel and Gentili, 2004].

[20] DMS sea-air flux was calculated as the product of seawater DMS concentrations and the transfer velocity (kw), after accounting for the temperature-dependent Schmidt number [Saltzman et al., 1993]. kw (m h−1) was parameterized as a linear function of wind speed (U; m s−1), according to Marandino et al. [2009]:

display math(4)

[21] We chose this parameterization based on recent evidence for a linear wind speed dependence of kw for wind speeds up to 14 m s−1 [Huebert et al., 2010; Marandino et al., 2009]. kw was set to 0 in the few occasions when U was <0.52 m s−1, which would produce negative kw. In our data set, 0–3.4% of minute-averaged U was <0.52 m s−1, and 83–98% of the values fell in the 1–9 m s−1 wind speed range (for which equation (4) was obtained). The use of the Marandino et al. [2009], Huebert et al. [2010], or Nightingale et al. [2000] parameterizations had a relatively small impact on DMS fluxes. In SARGsum, our mean kw estimate of 0.0152 m h−1 was within the uncertainty of the measurements reported by Blomquist et al. [2006] (at an average wind speed of 7.8 m s−1). The surface flux was divided by MLD to obtain average volumetric ventilation rates.

[22] Vertical DMS transport was estimated only in the oceanic experiments and treated as a constant term in the budgets due to the lack of measurements with adequate spatial-temporal resolution. Turbulent diffusion can be estimated as the product of the vertical DMS gradient and vertical diffusivity (Kz) at the base of the UML. In MEDsum, we combined the only DMS profile available with the output of model simulations (O. Ross, personal communication, 2012), including depth-resolved vertical eddy diffusivity, to make order-of-magnitude estimates of vertical DMS diffusion. In SARGsum, we used the estimates made by Bailey et al. [2008] for the entire Lagrangian study of the anticyclonic eddy A2. In CMEDwin and CMEDsum, no vertical DMS profiles were measured, so no estimations were made.

3 Results

3.1 Oceanographic Settings and Lagrangian Nature of the Sampling

[23] The four experiments covered 1 order of magnitude in phytoplankton biomass (measured as Chl a), from the average 0.56 µg L−1 in CMEDwin to the 0.11–0.14 µg L−1 in the Mediterranean summer and 0.05 µg L−1 in the Sargasso Sea eddy. Algal biomass was dominated by diatoms, prasinophytes, and haptophytes in CMEDwin (a rather typical situation at BBMO in late winter [Gutiérrez-Rodríguez et al., 2011]), whereas Synechoccoccus and photosynthetic picoeukaryotes, in varying proportions, dominated the summer assemblages (supporting information table 1). Given the high DMSPt:Chl a ratios found in summer, it seems plausible that haptophytes made up an important proportion of picoeukaryotic phytoplankton. Bacterial counts decreased in parallel to the trophic status but to a minor extent. Sizable inorganic nutrient pools could be measured only in CMEDwin. Means and ranges of meteorological and oceanographic variables and sulfur cycling rates are summarized in Table 2.

Table 2. Summary of the Ecosystem Setting and Sulfur Cycling During the Diel Cycle Experimentsa
  1. a

    Means and ranges (min-max) of usually 8–11 measurements are reported. When <5 measurements are available only the means are reported. SST: sea-surface temperature; Ed hourly max: maximum hourly total shortwave irradiance; Kd,PAR and Kd,340: vertical attenuation coefficients of PAR and 340 nm radiation, respectively; MLD: mixed layer depth; SRD: solar radiation dose index [Vallina and Simó, 2007]; UVBUML hourly max: maximum hourly UVB dose in the upper mixing layer; see text for other abbreviations.

  2. b

    Estimated according to PAR attenuation and Blanes Bay climatology (M. Galí et al., unpublished data).

  3. c

    This value is suspected to be unrealistically low.

  4. d

    According to Sargasso Sea climatology [Steinberg et al., 2001].

Meteorological and Oceanographic Data
SST (°C)12.6 (12.1–12.9)23.6 (23.3–23.9)24.7 (24.4–25.3)27.5 (27.3–27.7)
Salinity37.9 (37.1–38.4)37.7 (37.6–37.9)37.5 (37.4–37.6)37.0 (36.8–37.1)
Wind speed (m s−1)1.5 (0.5–3.2)3.9 (0.3–9.9)4.7 (0.04–14.6)7.8 (0.8–13.5)
Ed hourly max (W m−2)725913818660
Kd,PAR (m−1)0.120.08 (0.05–0.12)0.040.035
Kd,340 (m−1)0.26b0.18 (0.12–0.29)0.100.055
340 nm 10% penetration depth (m)9b13 (8–17)2234
MLD (m)10 (4–24)7 (3–13)14 (5–22)21 (9–30)
SRD (W m−2)9812313967
UVBUML hourly max (W m−2)0.420.80.50.55
Nitrate + nitrite (µmol L−1)2.70.630.09<0.02d
Phosphate (µmol L−1)<0.01d
Microbial Community Descriptors
Dominant phyto. (biomass)Diat ~ (Peuk + Neuk)Syn > (Peuk + Neuk)Syn > (Peuk + Neuk)(Peuk + Neuk) > Syn
Chl a (µg L−1)0.560.140.110.05 (0.035 - 0.061)
PPp (µmol C L−1 d−1)0.760.48 (0–0.7)0.5 (0–0.9)0.02c
Het. Bacteria (109 cells L−1)
LIR (nmol Leu L−1 d−1)0.70.8 (0.2–1.6)0.5 (0.2–0.8)0.24 (0.10–0.46)
Dimethylated Sulfur Concentrations and Ratios
DMS (nmol L−1)1.5 (1.1–1.9)3.1 (2.2–4.0)4.0 (2.3–4.7)3.1 (2.8–3.4)
DMSPt (nmol L−1)36.1 (13.3–77.1)18.1 (15.5–27.4)16.9 (14.4–19.8)8.0 (6.6–9.3)
DMS:DMSPt0.05 (0.01–0.12)0.18 (0.10–0.25)0.20 (0.14–0.31)0.40 (0.34–0.50)
DMS:Chla (nmol µg−1)2.814.136.474 (58–102)
DMSPt:Chla (nmol µg−1)33135155180 (143–206)
Biotic DMS(P) Cycling
K DMSPt prod (day−1)2.4 (0–5.5)1.0 (0–2.0)0.9 (0.3–2.5)0.8 (0.3–1.5)
K DMSPt cons dark (day−1)2.1 (0.5–5.6)0.9 (0.3–1.6)1.0 (0.5–1.8)0.8 (0.3–1.6)
K GPDMS dark (day−1)1.6 (0.2–3.9)0.6 (0–1.6)0.5 (0–1.0)1.0 (0–1.8)
K GPDMS light (day−1) 0.8 (0–2.3)0.7 
K BCDMS dark (day−1)1.3 (0–2.8)0.5 (0–0.8)0.5 (0.2–1.0)0.7 (0–1.6)
K BCDMS light (day−1) 0.4 (0–0.7)0.4 
Abiotic DMS Cycling (UML Mean)
K DMS photo (day−1)0.43 (0–2.1)0.19 (0–0.93)0.13 (0–0.70)0.05 (0–0.28)
K DMS vent (day−1)0.04 (0–0.07)0.28 (0.01–1.4)0.13 (0–0.75)0.19 (0.02–0.53)

[24] The coastal experiments (CMEDwin and CMEDsum) displayed higher hydrographic variability. This was evident in the excursions displayed by the temperature-salinity diagrams during the final 12 h of CMEDwin and the initial 12 h of CMEDsum (supporting information figure 1). Conversely, the oceanic experiments showed more compact temporal T-S traces. This was expected in SARGsum, which was Lagrangian by design. Yet we also observed quasi-Lagrangian behavior in MEDsum and in the last 36 h of CMEDsum. The T-S diagrams were used to exclude from further calculations and budgets those measurements that were judged as excessively affected by water mass transitions (the first two time points of CMEDsum).

[25] Contrasting dynamics in the vertical stratification regime were found in winter versus summer and coastal versus oceanic experiments (Table 2 and Figure 1). In CMEDwin, we found a quasi-mixed vertical profile (temperature differences <0.5°C from top to bottom depths), which sometimes showed a very subtle surface thermal stratification. In CMEDsum, MLD reflected a water-mass transition event by which the deep water intrusion observed in the two initial samplings was replaced by a thicker and warmer surface layer in a matter of hours. Both oceanic experiments (MEDsum and SARGsum) displayed similar stratification regimes, with a marked seasonal thermocline and a shallower actively mixing layer, which underwent diurnal thermal stratification and nighttime convective overturning (Figure 1).

3.2 Autotrophic and Heterotrophic Activities

[26] Experiment-averaged PPp was highest in CMEDwin (0.76 µmol C L−1 d−1), with lower values in CMEDsum and MEDsum (~0.50 µmol C L−1 d−1), and lowest values in SARGsum (0.02 µmol C L−1 d−1; this rate is suspected to be unrealistically low). Leucine incorporation rates (LIR; a proxy for bacterial heterotrophic production) were highest at the coastal sites (0.71 and 0.81 nmol leu L−1 d−1 during CMEDsum and CMEDwin, respectively), with lower values during MEDsum and SARGsum (0.48 and 0.24 nmol leu L−1 d−1, respectively). While the diel variation in primary production (PPp) was obviously light dependent, we found that LIR also underwent pronounced diel variations, with a large nighttime increase in the three summer experiments (by ninefold at CMEDsum and 4.5-fold at both MEDsum and SARGsum; Figure 2). The day versus night difference in LIR was significant in CMEDsum and SARGsum (Kruskal-Wallis two-group test, p <0.05) and suggestive in MEDsum (p = 0.11). The diel pattern of LIR likely reflected an increase in the activity per cell, because bacterial numbers underwent much smaller variations (with coefficient of variation about the mean, CV, of ±13%; T. Lefort et al., manuscript in preparation, 2013).

Figure 2.

Diel changes in particulate primary production (PPp) and leucine incorporation rates (LIR) in the four diel cycle studies; in Figure 2a the estimated mixing-layer average PPp during the daytime is shown. The gray shade represents the quotient between mixing layer averaged irradiance and the daily maximum irradiance at the water subsurface, for the 340 nm radiation band. Error bars represent the standard error of triplicate LIR measurements.

3.3 DMS and DMSPt Pools

[27] Experiment-averaged DMSPt concentrations ranged between a minimum of 8 nmol L−1 in SARGsum and a maximum of 36 nmol L−1 in CMEDwin, while DMS ranged between 1.5 (CMEDwin) and 4 nmol L−1 (MEDsum). Thus, compared to the order-of-magnitude differences in Chl a between experiments, DMSPt and especially DMS spanned a narrower range (Table 2). DMS and DMSPt pools underwent minor variations during the diel cycles (Figure 3), with CV < 20%. The exception was the broad variability observed for DMSPt during CMEDwin (CV = 53%), with marked DMSPt peaks at night. Regular diel variations of DMS could only be observed in CMEDsum, with midday troughs and midnight peaks. In MEDsum, lowest DMS levels occurred at 04:00 on the two consecutive nights. However, the sharp DMS decrease of the second night could not be explained by measured cycling processes or hydrographic variability, so it might be a spurious point. In SARGsum, DMS concentrations were very stable, with a slight surface depletion at solar noon.

Figure 3.

Diel changes in total DMSP (DMSPt) and DMS concentrations. The gray shade represents the quotient between mixing-layer averaged irradiance and the daily maximum irradiance at the water subsurface, for the 340 nm radiation band, and is proportional to UML-averaged DMS photolysis. Error bars correspond to duplicate analyses.

3.4 Gross DMS Production and Community DMSP Metabolism

[28] Gross DMS production (GPDMS) displayed broad and regular diel variations in the dark incubations (Figure 4), with opposite patterns found in winter and summer. Nevertheless, the four diel cycles displayed surprisingly similar GPDMS rates, with experiment averages ranging 0.07–0.11 nmol L−1 h−1. In CMEDwin, GPDMS peaked at night and was lowest in the morning, with a significant difference between day and night samples (Kruskal-Wallis p <0.05). Conversely, in the summer experiments, GPDMS peaked at solar noon or shortly after, decreased during the night, and started increasing again around dawn, so that GPDMS was significantly higher during the day (p <0.05 in all three experiments). In a few incubations (three in CMEDsum and one in MEDsum), the measured GPDMS was slightly negative though not distinguishable from zero. Since negative GPDMS cannot occur, in these cases, it was set to zero, which had a marginal effect on experiment averaged rates (<5% increase), and no effect on the statistics.

Figure 4.

Diel changes in gross DMS production rates; measurements done in the dark and in sunlight are represented with filled and open symbols, respectively. The gray shade represents the quotient between mixing-layer averaged irradiance and the daily maximum irradiance at the water subsurface, for the 340 nm radiation band. Error bars correspond to the analytical error of single incubations.

[29] In the sunlit incubations (performed in CMEDsum and MEDsum only), GPDMS was generally higher (up to threefold) than in their dark counterparts, with an average stimulation of 52%. Considering the whole 48 h sampling, light-driven stimulation resulted in a mean GPDMS 30–36% over the mean dark GPDMS. These figures compare well with the ranges of light-driven stimulation reported by Galí et al. [2013a] (for the short term) and Galí et al. [2011] (for a 1 day period).

[30] Mean DMSPt consumption rate constants (k; Table 2) were highest in CMEDwin (0.087 h−1, equal to 2.4 day−1), and lower in the other three experiments (~0.04 h−1, ~0.9 day−1), suggesting a more dynamic DMSPt metabolism in the late winter coastal setting. No significant differences in day versus night DMSPt consumption k were observed. DMSPt consumption rates, calculated as the product of the k (h−1) and the initial DMSPt concentration of the sample, were highest in CMEDwin (2.4 nmol L−1 h−1) followed by MEDsum and CMEDsum (~0.7 nmol L−1 h−1) and SARGsum (0.3 nmol L−1 h−1). DMSPt consumption also underwent day-night variations (Figure 5). These were more marked in CMEDwin and SARGsum, both displaying higher nighttime DMSPt consumption (quasi-significant differences were found in CMEDwin; p = 0.05). Conversely, DMSPt consumption was slightly higher during the day in CMEDsum and MEDsum, peaking around noon in CMEDsum. Net DMSPt synthesis (one incubation in SARGsum) or no DMSPt consumption (one incubation in MEDsum) was occasionally measured (both in a 04:00 sample), precluding the calculation of DMSPt consumption and the DMS yield.

Figure 5.

Diel changes in total DMSP consumption, DMSPt to DMS conversion yield, and DMSPd-sulfur assimilation yield; measurements done in the dark and in sunlight are represented with filled and open symbols, respectively (see text). For clarity, only average errors (analytical error of single incubations) are shown in boxes for each variable and experiment (not available for DMSP assimilation). The gray shade represents the quotient between mixing-layer averaged irradiance and the daily maximum irradiance at the water subsurface, for the 340 nm radiation band.

[31] The experiment-averaged DMS yields (the quotient of GPDMS to DMSPt consumption, in %) were minimum in CMEDwin (4%) and maximum in SARGsum (47%), with intermediate values (around 9–13%) in CMEDsum and MEDsum. Distinct diel patterns of DMS yield emerged depending on the covariation (or the lack thereof) between GPDMS and DMSPt consumption. While no obvious diel patterns were found in CMEDwin, daytime peaks of DMS yield were observed in the three summer experiments, though with variable amplitude. Significant day-night differences in DMS yield were found in the three summer experiments (Table 3).

Table 3. Day Night Differences in DMS Yield from DMSPt Consumption (%)a
  1. a

    Values of p <0.05 are interpreted as a significant difference (nonparametric Kruskal-Wallis test; in bold).

Overall mean (range)4.0 (0.7–8.1)9.3 (0–22.5)13.2 (0–30.2)46.7 (10.4–100)
Daytime average ± sd3.1 ± 1.317 ± 619 ± 870 ± 22
Nighttime average ± sd5.1 ± 0.94 ± 74 ± 616 ± 5
Day/Night factor0.
p (Kruskal-Wallis test)0.220.0310.0140.034

[32] DMSPd-sulfur assimilation into macromolecules (microbial biomass), as measured with 35S-DMSP additions, represented a minor sink for DMSP-sulfur, with an average assimilation yield of 4.8% and 2.7% in CMEDsum and MEDsum, respectively, and little diel variability (Figure 5). Thus, assimilation might have been an even smaller sink with respect to the totality of DMSP cycled (DMSPt consumption). DMSPd-sulfur assimilation was dominated by the >3 µm organisms or particles (60–83%), followed by the 3–0.65 µm and the 0.65–0.22 µm fractions.

3.5 Bacterial DMS Consumption

[33] As found for GPDMS, experiment-averaged dark bacterial DMS consumption (BCDMS) spanned quite a small range (0.06 to 0.09 nmol L−1 h−1), but exhibited distinct diel patterns from one experiment to another (Figure 6). Dark BCDMS peaked during the central hours of the day in MEDsum or at dusk in SARGsum, with predawn minima in both experiments. In MEDsum, dark BCDMS was significantly higher during the day (p <0.05). In contrast, dark BCDMS showed no clear diel variations in CMEDwin and CMEDsum. In those experiments where BCDMS was simultaneously measured in the dark and in the light (CMEDsum and MEDsum), inhibition (four samples) or no clear light effects (four samples) were generally observed.

Figure 6.

Diel changes in microbial DMS consumption; measurements done in the dark and in sunlight are represented with filled and open symbols, respectively. The gray shade represents the quotient between mixing-layer averaged irradiance and the daily maximum irradiance at the water subsurface, for the 340 nm radiation band. Error bars correspond to the propagated analytical error of nonamended and dimethyldisulfide-amended single bottle incubations.

3.6 DMS Photolysis

[34] In CMEDsum, MEDsum, and SARGsum, where the subsurface photolysis rate constants were determined experimentally, Kphoto,max were 0.049, 0.040, and 0.025 h−1, respectively (SARGsum value taken from Gabric et al. [2008]). In CMEDwin, a value of 0.13 h−1 was used, with an associated uncertainty of about ±50%. The daytime averages of DMS photolysis in the UML (that is, after accounting for underwater radiation fields, in situ DMS concentrations, and mixing depths) were 0.044, 0.041, 0.047, and 0.008 nmol L−1 h−1 in CMEDwin, CMEDsum, MEDsum, and SARGsum, respectively. The value obtained for SARGsum (0.10 nmol L−1 d−1) is consistent with that obtained by Bailey et al. [2008] (0.20 nmol L−1 d−1), since the former corresponds to an overcast day and the latter to the less cloudy conditions that prevailed during the sampling of the anticyclonic eddy.

3.7 DMS Ventilation

[35] Average sea-air transfer velocities (kw) were 0.013, 0.068, 0.053, and 0.152 m h−1 in CMEDwin, CMEDsum, MEDsum, and SARGsum, respectively. These values correspond to mean sea-air fluxes ranging from 0.02 µmol m−2 h−1 in CMEDwin to 0.47 µmol m−2 h−1 in SARGsum (equivalent to 0.5–11.3 µmol m−2 d−1). Converted to volumetric fluxes in the UML, these figures become 0.002, 0.020, 0.038, and 0.025 nmol L−1 h−1 in CMEDwin, CMEDsum, MEDsum, and SARGsum, respectively.

3.8 Vertical DMS Transport

[36] Vertical DMS profiles were measured at the two oceanic stations, both showing a peak in the thermocline (30–40 m) of ~7 nmol L−1 (Figure 7). Thus, the thermocline layer will be a source of DMS to the UML. The DMS gradient at the base of the UML was ~0.30 in SARGsum and ~0.50 µmol m−4 in MEDsum. Combining this gradient with vertical eddy diffusivities of the order 10−5 to 10−4 m2 s−1 (estimated to be similar at both sites) yielded an upward DMS flux of 0.08 ± 0.06 and 0.12 ± 0.08 nmol L−1 d−1 in SARGsum and MEDsum, respectively (following equation (3) of Bailey et al. [2008]). These figures (note the day−1 units) are negligible compared to the other budget terms. Nevertheless, Bailey et al. [2008] found that accounting for the entrainment of DMS due to diurnal stratification/mixing cycles increased the upward flux to 0.34 ± 0.06 nmol L−1 d−1. Entrainment might have been of similar magnitude in MEDsum, because it showed mixing dynamics and vertical DMS distribution similar to SARGsum.

Figure 7.

Vertical DMS profiles in the Mediterranean and Sargasso Sea oceanic experiments. Horizontal lines represent the average mixing layer depth at the time the profiles were measured. Error bars in the Sargasso Sea profile represent the range of two profiles measured at 04:00 local time on consecutive days.

4 Discussion

[37] The four ocean settings displayed a biogeochemical gradient that was reflected in the dimethylated sulfur compound concentrations and their microbial cycling rates (Table 2). Combined with the diel sampling scheme, this provided an opportunity to test hypotheses regarding the biotic/abiotic regulation of DMS cycling.

4.1 Factors Controlling the Fate of DMSP and Gross DMS Production

[38] DMSP cycling occurs through a tangled network of microbial interactions [Simó, 2004; Stefels et al. 2007] and follows diverse biochemical pathways [Curson et al., 2011; Moran et al., 2012]. We took an integrative approach where all the DMSP cycling pathways eventually releasing DMS were considered [Simó et al., 2000]. This has the advantage of better constraining gross DMS production for budgeting and modeling exercises, and the drawback that the contribution from the operational dissolved and particulate pools can only be guessed from indirect evidence. In our approach, gross DMS production can be modulated by changing either DMSPt consumption, DMS yields, or both. In CMEDsum, MEDsum, and SARGsum, GPDMS was driven by DMS yields (Table 3) rather than by DMSPt consumption (Figure 8). This was most evident in SARGsum, where the lower DMSPt consumption during the day was more than compensated by a sharp increase of the DMS yield (with four measurements >50%). In the case of CMEDsum and MEDsum, the increase during the day of both DMS yield (by a 4 to 4.8 factor) and DMSPt consumption (by a 1.2 to 1.6 factor) contributed to the observed increase in GPDMS. In CMEDwin, conversely, the nighttime increase in DMSPt consumption, combined with low DMS yields over the full diel cycle, resulted in nighttime GPDMS peaks (Table 3 and Figures 4, 5, and 8).

Figure 8.

Day (upper values, in bold) versus night budgets of DMS and DMSP cycling in the upper mixed layer. Values in parentheses are the standard deviation (sd) of biological process incubations (3 < n < 6) or that of 5 min resolution budgets in the case of DMS photolysis and ventilation. Order-of-magnitude estimates of vertical DMS transport are also given, with no day-night distinction.

[39] The summertime diel GPDMS pattern could result from different processes which are not mutually exclusive. A feasible scenario is that light-stressed phytoplankton upregulate DMS production enzymes to cope with oxidative stress [Sunda et al., 2002] or as an overflow mechanism to channel excess reducing power [Stefels, 2000]. Nutrient limitation could have acted synergistically with light exposure to enhance DMS production in summer stratified waters [Sunda et al., 2007]. Besides the evidence derived from culture studies [Archer et al., 2010; Green et al., 2012], seasonal field studies support a strong relationship between light exposure and DMS production by phytoplankton. Vila-Costa et al. [2008] showed that bacterial DMS production or microzooplankton grazing were unlikely to explain the summer increase in GPDMS at BBMO. Levine et al. [2012] found that potential DMS production in the “phytoplankton” (>1.2 µm) size fraction co-occurred with strongest UV exposure in the Sargasso Sea. At the diel time scale, to our knowledge, only one study has reported a daytime increase of GPDMS, in the St. Lawrence estuary [Merzouk et al., 2004]. However, this increase was attributed to dinoflagellate vertical migration, which we did not find in our study, at least, in CMEDsum and MEDsum (supporting information table 2). Recently, Galí et al. [2013a] showed, by means of experimental manipulation, that GPDMS displays spectral irradiance dependence similar to that of phytoplankton photoinhibition or photodamage caused by UV radiation. Thus, the response of phytoplankton to UV-PAR exposure seems to have a strong influence on GPDMS.

[40] Supporting the phytoplankton stress hypothesis, we observed that bulk chlorophyll fluorescence was depleted in surface waters during the hours of strongest irradiance in the oceanic experiments (Figure 1), indicating reduced photosynthetic competency and/or dissipation of excess irradiance through photoprotective mechanisms [Sakshaug et al., 1997]. In CMEDsum and MEDsum, both dark and light GPDMS were significantly correlated to primary production (Spearman's rank r = 0.74 and 0.67 in CMEDsum and MEDsum, respectively; p < 0.05; supporting information figure 2). In SARGsum, the UV dose in the UML during the 4 h prior to sampling was the best predictor of dark GPDMS (Spearman's r = 0.99; p <0.01; supporting information). Since we observed not only diel GPDMS variations in dark incubations, but also direct light-driven stimulation, we propose that the interplay between cumulative and instantaneous light exposure drove GPDMS in summer. The cumulative component might be related to UV-induced irreversible damage, and the instantaneous component, to photosynthesis-related physiology.

[41] Besides direct algal DMS production from the particulate pool, DMS production can also arise from DMSP release by phytoplankton and its transformation by bacterial enzymes or by dissolved algal enzymes (note that in some algal strains, physical disruption is required to put DMSP in contact with its cleavage enzymes [Wolfe and Steinke, 1996]). DMSP release is generally thought to occur through active DMSP exudation and cell disruption by grazing, viral lysis, or autolysis [Stefels et al., 2007]. As noted by Galí et al. [2013a], UV-triggered cell damage and death [Llabrés and Agustí, 2006] can also enhance DMSP leakage from phytoplankton cells. Due to a number of counteracting factors, DMS production by free-living bacteria has limited potential to respond positively to solar radiation [Slezak et al., 2007]. However, bacteria can be exposed to elevated DMSP concentrations in the phycosphere [Seymour et al., 2010]. We speculate that light-induced DMSP leakage might enhance bacterial DMSP degradation and eventually DMS production in the vicinity of algal cells.

[42] The role of circadian rhythms should also be considered. The diel gene expression patterns of several plankton microbes are regulated by molecular clocks [e.g., Corellou et al., 2009], which help cells anticipate environmental variability and adequately phase sensitive processes like cell division [Vaulot and Marie, 1999]. In a recent metatranscriptomics study, Poretsky et al. [2009] identified a switch between housekeeping genes at night and photophysiology-related genes during the day. Phytoplankton DMSP metabolism might be regulated in such a manner if it effectively prevented radiative stress. Regarding DMSP synthesis, it is notable that net in situ DMSPt production was frequently observed during the night (modest in CMEDsum, large in CMEDwin). By budgeting net in situ DMSPt variations and DMSPt consumption, we infer that DMSP synthesis occurred at night in all the experiments, though to a different extent (Figure 8). This contrasts with other works where DMSP synthesis paralleled carbon fixation [Simó et al., 2002]. Nevertheless, budget-derived DMSP production rates bear a large uncertainty and must be regarded with caution (see below). Regarding DMS production, it is interesting that in CMEDsum and MEDsum, GPDMS rose significantly in the predawn sample (04:00) or in the 08:00 sample, when surface waters had barely seen any radiation. Green et al. [2012] recently reported that DMS production during the light period doubled that of the dark period in an Emiliania huxleyi culture, and that DMS production often started to increase during the dark period (see Figure 5 of that paper). It is intriguing, however, why phytoplankton should start cleaving DMSP instead of accumulating it intracellularly before radiative stress sets in.

[43] The role of microzooplankton grazing in the diel patterns of GPDMS is uncertain too. Recently, Ruiz-González et al. [2012b] reported on diel patterns of protozoan grazing on picoeukaryotic phytoplankton in a study done at the same site (BBMO) and in similar conditions as the CMEDwin experiment. In that study, picoeukaryotes divided early in the night, causing a subsequent peak in the amount of picoeukaryote cells ingested by heterotrophic flagellates. In CMEDwin, similarly, the nighttime increase and subsequent predation of DMSP-bearing cells might have caused higher DMSPt consumption at night (Figure 8). However, this was not reflected in the (biomass-specific) DMSPt consumption k. In summer, significant DMSPt consumption went on at night (k ~ 0.8 day−1 at all three sites), which can be reasonably assigned to microzooplankton grazing. All in all, it seems plausible that in low radiative stress conditions, most GPDMS arose from the combination of protozoan grazing and bacterial DMSP metabolism, with low associated DMS yields [Kiene and Linn, 2000; Slezak et al., 2007; Del Valle et al., 2012]. During the day, the combination of grazing-induced and light-induced DMS(P) release and, possibly, more abundant DMSP-cleaving enzymes in phytoplankton, could have contributed to increase GPDMS.

[44] Although our measurements clearly support the existence of distinct DMS production regimes at the diel scale, they suffer from some experimental uncertainties. First, it is unknown whether solar radiation would have affected community DMSPt consumption to the same extent as GPDMS, therefore modifying the DMS yields obtained from dark incubations. Another source of uncertainty comes from the “net-loss curve” method used to estimate DMSPt consumption. In the absence of dark DMSP synthesis during incubations, this method should quantify the gross DMSP consumption rate. However, we observed dark DMSP synthesis in a few incubations (see results; see also Simó et al. [2000]), as well as net in situ DMSPt production (Figure 3). These facts suggest that dark DMSP synthesis might have occurred in other incubations, even when the overall balance at the end of the incubation resulted in net DMSPt consumption. Hence, our DMSPt consumption estimates would represent a lower limit and, in consequence, DMS yields would represent an upper limit. On the other hand, there are findings that support our methodology. The seasonal study done by Vila-Costa et al. [2008] in Blanes Bay showed, by comparing the results of the net-loss curve method and those of 35S-DMSPd amended incubations, that DMSPd consumption accounted on average for 52% of DMSPt consumption (and this proportion is probably an upper limit due to the overestimation of DMSPd concentrations [Kiene and Slezak, 2006]). The DMSPt consumption rate constants reported by Vila-Costa et al. ranged between 0.0 and 2.1 day−1 (mean of ~0.9 day−1), similar to the range observed in our study (0.8–2.4 day−1; experiment means). These figures are similar to the specific turnover rates of phytoplankton biomass across oceanic regimes (0–2.5 day−1, mean ~ 0.5 day−1), mainly caused by microzooplankton grazing [Calbet and Landry, 2004]. Altogether, these facts support the notion that DMSPt turnover is similar to that of its phytoplankton producers and that our DMSPt consumption estimates are not too far from gross DMSPt consumption. Better methods are needed to track the synthesis and catabolism of algal DMSP by microbial communities.

4.2 Factors Controlling Bacterial DMS Consumption

[45] According to current knowledge, UV inhibition should tend to decrease BCDMS during the day. However, this effect should show up with different intensity depending on (1) whether the samples are incubated in the dark or in sunlight and (2) the length of the incubation compared to the recovery time. In some of the incubations, inhibition due to light was observed or, from another point of view, the relief from UV inhibition elicited some recovery. In contrast, in other samples, no clear light effects were measured. The relative dark-light inhibition we observed should be viewed with some caution since the inhibitor method is less precise than other methods (e.g., 35S-DMS tracer) for determining BCDMS.

[46] The response of BCDMS in the three summer experiments contrasted with that of bulk bacterial heterotrophic production (LIR), which was markedly depleted during the day. In fact, the two variables were uncorrelated (supporting information figure 2). A study by Toole et al. [2006] suggested similar photoinhibition patterns of BCDMS and LIR in the water column, which makes the pattern we encountered more surprising (especially in MEDsum). However, no other BCDMS measurements across diel cycles have been published to our knowledge. Thus, the factors controlling the activity of the bacterial DMS consumers might be different from those controlling more widespread activities like leucine uptake. Recent research has shown that few bacterial taxa can grow on DMS as the sole carbon source [Vila-Costa et al., 2006], whereas a broader diversity of bacteria can oxidize DMS to DMSO to obtain energy provided that they have an alternative C source to grow upon [Vila-Costa et al., 2006; Green et al., 2011; Hatton et al., 2012]. Interestingly, Del Valle et al. [2007] showed that DMSO was the fate of most DMS consumed in the UML in the Sargasso Sea (in the same cruise as SARGsum). It has been postulated that in oceanic environments, far from coastal carbon sources, pelagic bacteria rely on the labile carbon excreted by phytoplankton [Gasol et al., 1998]. It is feasible therefore that the diel patterns of BCDMS observed in MEDsum and SARGsum were regulated by the interplay between labile carbon supply and UV exposure, besides other unknown factors.

4.3 Short-Term Microbial/Meteorological Variability and DMS Budgets

[47] Since the pioneering work of Bates et al. [1994] in the northeast Pacific, DMS budgets in the UML have been calculated in contrasting oceanic regimes and with different methodologies. The budgets tend to agree in that DMS removal is dominated by BCDMS [Simó, 2004], followed by photolysis and sea-air flux. In our data set, BCDMS also represented on average the main DMS sink, from 48% (CMEDsum) to 80% (CMEDwin). Photolysis accounted for a rather constant proportion of DMS removal, between 12% (SARGsum; note that this value corresponds to an overcast period) and 17% (both CMEDwin and CMEDsum). Sea-air flux accounted for a more variable fraction, from 2% (CMEDwin) to ~35% (CMEDsum and SARGsum). In CMEDsum and SARGsum, the important role of sea-air flux resulted from the combination of elevated water temperature and high wind speed (and shallow stratification in CMEDsum).

[48] The original contribution of our study shows that contrasting day-night patterns in biological DMS(P) cycling might be widespread (Figures 8 and 9) and that, despite this, DMS budgets can be nearly balanced in the short term (e.g., CMEDwin or SARGsum). If subtle budget imbalances occur, distinct diel patterns can emerge (e.g., CMEDsum) with either positive or negative day-night DMS trends. As noted by Galí et al. [2013a], an important buffering effect occurs through the compensation between light stimulation of GPDMS and photolysis. At this point, it has to be acknowledged that biological DMS cycling processes, if light dependent, should have been depth dependent as well. Recent in situ experiments have shown that the samples incubated in the middle optical depth within the UML display GPDMS rates similar to those obtained by integrating vertically the rates from samples incubated at different depths within the UML [Galí et al., 2013b]. Our data indicate that besides photolysis, BCDMS may also be able to respond to and compensate the excess daytime GPDMS (MEDsum and SARGsum; Figure 9), a finding that deserves further attention. Due to these stabilizing effects, important day-night changes in DMS emission are not expected, unless wind speed varies regularly from day to night. Indeed, quasi steady-state DMS budgets can be disrupted by storms or vigorous phytoplankton blooming.

Figure 9.

Idealized DMS budgets in the upper mixing layer (1 h resolution), reflecting the combined effect of meteorological variability and biological responses. The budgets were calculated assuming that vertical DMS transport was negligible, and that DMS ventilation was shared within the mixing layer within 1 h (which is probably unrealistic).

[49] The uncertainties in our measurements resulted in a variable agreement between our budgets and net DMS variations in situ over days, nights, and the whole sampling periods (detailed information in supporting information table 3). Some uncertainty likely arose from the poorly resolved vertical DMS transport. In MEDsum, budget-predicted DMS concentration trends were rather close to the day, night, and overall DMS trends in situ. In CMEDwin, our budgets magnified the net daytime DMS loss and the net nighttime DMS production, but the overall DMS trend was successfully captured. In CMEDwin, the sign of day and night net DMS trend was well reproduced, but the overall increasing trend of DMS was not captured. Interestingly, our results in SARGsum can be compared to those obtained by Bailey et al. [2008] in the same eddy A2 with a different approach. By budgeting simultaneous measurements of net in situ DMS evolution and biotic and abiotic DMS sinks, Bailey et al. diagnosed a mean GPDMS of 0.68 ± 0.09 nmol L−1 in the UML, which is three to four times lower than our value (2.56 ± 1.25 nmol L−1 d−1). It appears that neither value is totally correct. Bailey et al. recognized that their budgeting scheme might have produced a slight underestimation of GPDMS. In addition, they used a constant bacterial consumption rate constant derived from the 04:00 CTD casts, the time when the lowest BCDMS occurred according to our results (Figure 6). This might have caused further underestimation of GPDMS. On the other hand, our results would suggest net DMS accumulation, which did not occur in situ. If we add up the excess net in situ DMS change of 0.45 nmol L−1 d−1 deduced from our budget and the 0.34 nmol L−1 d−1 of DMS transport into the UML [Bailey et al., 2008], we get an overestimation of GPDMS of about 0.8 nmol L−1 d−1. Further work is warranted to determine the relative importance of the source and the sink terms in modulating oceanic DMS concentration.

4.4 Insights Into Dimethylated Sulfur Cycling Regimes

[50] Toole and Siegel [2004] proposed a classification of DMS cycling regimes, distinguishing between a bloom regime typical of high latitudes, where DMS correlates to phytoplankton biomass, and a stress regime typical of low latitudes, where DMS is negatively correlated to phytoplankton biomass (the “summer paradox” areas of Simó and Pedrós-Alió. [1999b]). The decoupling between DMS and plankton biomass [see Lizotte et al. 2012] remains difficult to represent in prognostic DMS models [Le Clainche et al., 2010]. Part of this decoupling results, indeed, from the covariation between strong DMSP producers and high light conditions, which occurs seasonally [Vila-Costa et al., 2008] and spatially [Bell et al., 2010]. In our data set, this is illustrated by the DMSPt:Chl a ratio, which varied by approximately sixfold, from 33 nmol µg−1 (CMEDwin) to 180 nmol µg−1 (SARGsum). However, some recent modeling studies had to invoke an additional factor, namely, radiative stress, to be able to reproduce DMS seasonality [Toole et al., 2008; Vallina et al., 2008; Vogt et al., 2010]. The more than tenfold variability we observed in community DMS yield may well result from this “stress factor.” In this regard, it is remarkable that we found similar GPDMS at sites where Chl a spanned 1 order of magnitude and DMSPt varied by fourfold to fivefold.

[51] Relating community DMS yields to other variables that can be more easily measured can provide useful information for the biogeochemical models. For instance, DMS yields and experiment-averaged DMS:DMSPt ratios spanned a similar range in our dataset: 4% versus 0.05 (CMEDwin), 9–13% versus 0.18–0.20 (CMEDsum and MEDsum), and 47% versus 0.40 (SARGsum), suggesting that the ratios might serve as a proxy for the process-based yields. In a recent study, the GPDMS:DMSPt ratio was proposed as a useful shortcut to predict gross DMS production [Herrmann et al., 2012], arguing that it is relatively invariant across ecosystems with a value about 0.06 ± 0.01 day−1. Our results suggest that this index, conceptually close to the DMS yield, is far from constant, since we have observed GPDMS:DMSPt ratios (mean ± sd) as variable as 0.07 ± 0.06 day−1 (CMEDwin), 0.10 ± 0.10 day−1 (CMEDsum), 0.15 ± 0.15 day−1 (MEDsum), and 0.29 ± 0.14 day−1 (SARGsum).

[52] In our view, oceanic DMS cycling could be divided into high and low DMS yield situations, both of which can potentially occur in environments with high or low biomass and/or DMSPt and can even switch from one to another in the short term. In low oxidative stress situations (nighttime, winter mixing), bacterial DMS production may prevail, with low associated yields [Kiene and Linn, 2000]. In high stress situations, much higher DMS yields can be attained [Simó and Pedrós-Alió, 1999b; this work]. During the last years, much research has focused on understanding the so-called bacterial switch [Simó, 2001; Moran et al., 2012] by which bacteria divert a lower or higher amount of dissolved DMSP to the DMS production pathway(s). Now there is compelling evidence that an important share of gross DMS production originates from the cycling of “particulate” DMSP and that the fate of this DMSP is modulated by solar exposure. Therefore, ecosystem-level studies of DMS(P) cycling should evaluate total DMSP turnover, ideally, simultaneously with DMSPd turnover [Vila-Costa et al., 2008]. Understanding the distinct modes of operation of the oceanic DMS cycle will require disentangling the intricate microbial interactions and their interplay with plankton photophysiology and physical forcing through all relevant time scales.


[53] We thank the crews and scientists who made possible the Biocomplexity and Modivus (Sargasso and Mediterranean Sea) cruises on board the R/V Seward Johnson and Garcia del Cid, respectively, as well as the BBMO sampling team. We are indebted to R. P. Kiene for kindly providing 35S-DMSP and to Irene Forn for invaluable help in the Sargasso Sea cruise. We also thank Oliver Ross for providing physical model simulations, and Jaume Piera, David J. Kieber, and Sergio M. Vallina for thoughtful comments. M.G. acknowledges the receipt of a CSIC JAE scholarship. This work was supported by the (former) Spanish Ministry of Science and Innovation through the projects MODIVUS (CTM2005-04795/MAR) and SUMMER (CTM2008-03309/MAR) and by the NSF Biocomplexity funding program through OPP-0083078. The constructive comments of three anonymous reviewers helped improve the manuscript. This is a contribution of the Research Groups on Marine Biogeochemistry and Global Change and on Aquatic Microbial Food Webs, supported by the Generalitat de Catalunya.