Seawater from the Jade Bay (North Sea) offshore the city Wilhelmshaven (Germany, 53°30′50.0″N 8°08′44.2″E) was pumped on 6th July 2016 at high tide from a depth of ~1–2 m into an outside basin to allow settlement of sediment particles for 5 days. This water was used for both of the following experiments. On 11th July 2016, water from the outside basin was filtered through a 100 μm mesh and filled into a ~700-L aquarium tank (2.2 × 0.6 × 0.5 m), made from glass. The first experiment ran from July 13–18th 2016 (Day 1–6). On 19th July 2016, the tank was refilled with water from the outside basin for a second experiment that ran from July 20th−22nd 2016 (Day 1–3). The open tank was placed in a greenhouse, which allowed exposure to diel light changes. Air (in the greenhouse) and water temperature in the tank varied from 17 to 30°C and from 20 to 26°C, respectively. Photosynthetically active radiation (PAR) was recorded using the Quantum Meter MQ-200 (Apogee Instruments, UT, USA) in a wavelength range of 410–655 nm. The PAR sensor was attached 1 m above the water surface. PAR data were measured every 30 s, and the mean of 60 measurements was logged every 30 min. For relationships between PAR and NCP, the mean PAR for the duration of each NCP incubation period (Table 1) was calculated. All dates and times mentioned throughout this work refer to UTC +01:00. The experimental set-up is illustrated in Supplementary Figure 1.

Microprofiles of O₂ and temperature in the laboratory tank were simultaneously measured using a Clark-type O₂ microsensor OX-50 and a temperature microsensor TP-200 (Unisense, Denmark), respectively, which were connected to an amplifier (Multimeter, Unisense, Denmark). The response times for OX-50 and TP-200 are < 5 and < 3 s, respectively. Profiles were measured using the MicroProfiling system (Unisense, Denmark), i.e., a motorized micromanipulator that was installed on top of the aquarium tank and controlled by a computer (Supplementary Figure 1). Signals were recorded using the software SensorTraceSuite (Version 2.6.100). The temperature sensor was calibrated each day to three random temperatures within the ranges of 0–10, 10–30, and 30–50°C measured with a mercury thermometer. The O₂ microsensor was calibrated with two points using an anoxic 0.1 M sodium ascorbate solution for 0% standard. The signal of 100% air-saturated water was taken from the 0 μm water depth (i.e., the air-water interface) of each individual profile, assuming 100% air-saturation due to equilibration with the atmosphere. The 0 μm water depth and its corresponding signal value was established for each profile as the crossing point between (a) the regression line through the approximately constant O₂ sensor values in air and (b) the regression line through the data points in the first 500 μm that deviated from the air values (Supplementary Figure 2). To reduce ambiguity, the first three data points on the airside and the waterside of the air-water interface were excluded for regression analyses. The value at 0 μm depth was converted into μmol O₂ L⁻¹ using the salinity and temperature dependent O₂ solubility (Garcia and Gordon, 1992).

To account for different vertical resolution, profiles were measured in seawater of the tank either in vertical steps of 50 μm to a depth of 5 mm (shallow profile) or in 500 μm steps to a depth of 5 cm (deep profile) through the SML in direction from air to water.

The O₂ microsensor was recalibrated at the temperature, which was measured at 0 μm water depth. As described above, the 0 μm water depth was defined by the intersecting linear regression lines of air and water temperature readings. To account for minor vertical changes of the water temperature in the tank within the upper centimeters (Supplementary Figure 3), the O₂ profile data was temperature-corrected for each individual depth according to Rahlff et al. (2017b) and Unisense supportive material.

To determine the O₂ concentration in the ULW, the mean O₂ concentration was calculated using data from the deep O₂ profiles from 1.5 to 4.0 cm water depth. To describe the temporal variability of the mean O₂ concentration, the differences between two deep profiles were calculated and expressed in μmol O₂ L⁻¹ h⁻¹. The analyzed profiles corresponded to the time points t₀, t_n, and t_end of the incubation experiments (Table 1 and Supplementary Table 1) to allow for direct comparison with metabolic activity.

For calculation of O₂ losses from the tank by diffusion we considered the Whitman film model, which describes molecular diffusion through a stagnant, laminar layer (Gladyshev, 2002). Diffusive rates (μmol L⁻¹ h⁻¹) were calculated for each O₂ gradient as follows: (1)Diffusive rate (μmol L-1h-1)=DS1V(C0-CULW)1δ where D is the diffusion coefficient (cm² h⁻¹) obtained from a straight line equation of D vs. temperature (Ramsing and Gundersen, 2000) at a salinity of 30. The mean temperature within the upper 1.1 mm [defined as the diffusive boundary layer (DBL)] was calculated from each profile to be analyzed. S is the surface area of the aquarium (12,490 cm²), V is the tank volume (673.2 L), δ is the thickness of the SML, which we considered to be the thickness of the DBL (1.1 mm) as deduced from all O₂ gradients measured. C₀ is the O₂ concentration (μmol L⁻¹ h⁻¹) at the interface (0 μm) and C_ULW is the O₂ concentration at 1.1 mm water depth.

Finally, we calculated the mean diffusive rate from the diffusive rates of two shallow profiles, which were recorded in temporal proximity to the deep profiles that in turn were used to calculate the mean O₂ concentration.

To evaluate the theoretical effect of biological metabolism (i.e., NCP) on O₂ profiles in the SML, we used a simple mathematical steady-state model with the assumption of a stagnant, homogeneous SML bordering sharply to a well-mixed ULW with constant concentrations (Figure 1). At steady state, a modified version of Fick's 2nd Law describes the concentration C_R(x) of a substance, which is both diffusing and reacting (consumption or production) as a function of depth x as

Where D is the diffusion coefficient and R is the constant reaction rate (positive for production and negative for consumption) for the gas. Integrating this equation with the boundary conditions that the concentration at the top (depth = 0 μm) of the SML, C₀, is constant at equilibration with the atmosphere and the concentration at the bottom (depth = l) of the SML, C_ULW, is constant and equal to the ULW concentration, yields the concentration profile in the SML as a function of depth x (Figure 1): (3)CR(x)=C0-R2Dx2+CULW-C0+R2Dl2lx If the reaction rate is zero, the concentration profile, C_NR(x), becomes (4)CNR(x)=C0+CULW-C0lx and the difference between the concentration profile with and without reaction is thus (5)CR(x)-CNR(x)=-R2Dx2+R2Dlx This difference is a parabola with its maximum/minimum at (6)x=l2 i.e., in the middle of the SML, where the value of this absolute difference is (7)diffmax=Rl28D The value of the concentration in the middle of the SML without reaction is (8)CNR(l2)=CULW+C02 The maximum relative difference between the concentration profiles with and without reaction is thus (9)diff rel,max=Rl24D(CULW+C0)

To sample the SML we used the glass plate technique (Harvey and Burzell, 1972), where a clean, hydrophilic glass plate is introduced vertically through the water surface until it is fully immersed. It is withdrawn at a controlled rate of 5–6 cm s⁻¹ as suggested by Carlson (1982a) resulting in the adhesion of an ~20–100 μm thick SML to the glass plate (Carlson, 1982a; Shinki et al., 2012). The thickness of the collected SML is determined by water density and viscosity, surface tension, withdrawal speed, and gravity (Levich, 1962). The SML adhering to the glass plate was scraped off into glass bottles until a volume of ~30 mL has been collected.

While the glass plate sampling technique assumes that the majority of material collected originates from the SML, dilution with bulk water (ULW) cannot be ruled out. Wurl et al. (2011a) investigated the effect of selective adhesion of sample material to glass bottles used for sample collection. They showed that only about 10% of all organic gel-like particles from a water sample adhered to the interior of a glass bottle. Assuming that this would be equivalent to unspecific adhesion of ULW material to the glass plate after immersion into the water, only 10% of organic material in SML samples may originate from the ULW. Similar approaches to test the amount or type of microorganisms from the ULW attaching to the glass plate during SML sampling are missing. However, the glass plate technique was shown to neither quantitatively (Agogué et al., 2004) nor qualitatively (Stolle et al., 2009) select microorganisms during collection. These findings overall imply that the SML samples collected with the glass plate technique very well represent the SML with only little dilution with ULW material. Samples from the ULW were collected from 30 cm water depth with a syringe connected to a hose (Supplementary Figure 1). The glass plate was cleaned with 70% ethanol before one or two dips of SML sample were taken to pre-rinse the glass plate, the funnel and the collection bottle. The syringe and the bottle for ULW sample were also pre-rinsed with sample water.

For the analysis of chlorophyll a (Chl a), 20–100 mL of each water sample were filtered onto glass microfiber filters (grade GF/F, Whatman, UK), stored at −20°C, and extracted in 90% acetone overnight at 4°C. Samples were analyzed at excitation wavelength 430 nm and emission >665 nm using a calibrated handheld fluorimeter (Aquafluor™, Turner Designs, CA, USA). The calibration equation was derived from spinach extracts (Sigma Aldrich, Germany).

The quantum yield (Fv/Fm) of phytoplankton in the ULW was measured every 10 s with a PhytoFlash sensor (Turner Designs, CA, USA) as described in Mustaffa et al. (2017). The PhytoFlash was calibrated with a Scenedesmus obliquus culture. Tank water from 20 cm water depth was pumped continuously into the PhytoFlash in a closed loop. For comparison with the PAR data, mean values of the quantum yield for every 30 min were calculated.

For determination of the abundance of bacterial and small autotrophic cells, water samples were preserved with glutaraldehyde (1% final concentration), incubated for 1 h at 4°C and snap-frozen at −80°C. Prior to counting, all ULW and SML samples were pre-filtered by gravity through CellTrics® 50 μm filter (Sysmex Partec, Germany) to avoid clogging of the flow cytometer (BD Accuri C6, BD Biosciences, CA, USA). Autotrophic cell counts were determined by measuring autofluorescent emission and scattered light (Marie et al., 2000). Bacterial cells were stained and counted as described in Rahlff et al. (2017a).

In order to evaluate if the bacterial and phytoplankton microbial community in the tank was composed of representative, naturally occurring organisms, we analyzed the community composition based on 16S rRNA and 18S rRNA gene fingerprints. Details about the methods applied and the results obtained can be found in the Supplementary Material.

Duplicates of each SML and ULW sample were filled into 3-mL Exetainers® (Labco Limited, UK) and closed gas-tight without headspace. Incubation of samples was either conducted under ambient light condition (to allow for primary production and O₂ consumption) or in dark controls (to determine O₂ consumption alone). All Exetainers® were incubated in the tank at in situ water temperature at the respective depth (Supplementary Figure 1). A fixed needle-type oxygen optode (OXF500PT, Pyro Science, Germany) was used to measure O₂ concentration within the Exetainers® at the start (t₀) and end (t_end) of the incubation period. During 4 of 5 incubation experiments, additional intermediate measurements were done, here referred to as t_n, where n refers to the number of the intermediate measurement. The incubation period ranged from ~3 to 49 h (Supplementary Table 1). The optode was calibrated each day according to the manufacturers' instructions using air-saturated water, i.e., bubbling of tank water for 15 min (100% O₂) and an anoxic 0.1 M sodium ascorbate solution (0% O₂). Using these two calibration points, the “Pyro Oxygen Logger” software (version V3.204) automatically calculates the O₂ concentration (μmol O₂ L⁻¹) based on the salinity and temperature dependence of O₂ solubility (Garcia and Gordon, 1992). The temperature was continuously recorded with a TDIP15 probe (Pyro Science, Germany).

Measurements within each Exetainer® were conducted at in situ temperature for at least 5 min. After stabilization of the O₂ signal, the O₂ concentration was recorded for 1.5 min, from which the mean O₂ concentration was calculated. The standard error (SE) for this determination of the mean was on average 0.33 μmol O₂ L⁻¹ (Experiment 1) and 0.14 μmol O₂ L⁻¹ (Experiment 2). Rates of NCP and O₂ consumption were defined as the difference in O₂ concentration between two time points in the light and dark incubations, respectively, divided by the elapsed time and are given in μmol O₂ L⁻¹ h⁻¹. Error propagation of SE when subtracting two means (M1 and M2) was calculated by (10)SE(M1-M2)= SE(M1)2+SE(M2)2  and was < 0.2 μmol O₂ L⁻¹ h⁻¹ except for three time points (Table 1).

Potential influential data in the relationships shown in Figure 3 were defined as outliers if their Cook's distance (Cook, 1977) calculated using R version 3.4.3 (R Core Team, 2017) was >4 times the mean Cook's distance of the respective dataset. These outliers were excluded from least-square regression models used for Spearman rank correlation.

Non-parametric Spearman rank correlation analyses (two-tailed, 95% confidence interval) for changes in O₂ concentration in the ULW vs. mean PAR or NCP in SML and ULW (Figure 3) were performed using GraphPad Prism (Version 5.00 GraphPad Software, San Diego CA, USA). Results of the correlation analyses and identified outliers are shown in Supplementary Table 2.

High-resolution O₂ profiling using a microsensor revealed profound dynamics of O₂ profile formation across the air-water interface. Figure 2 shows four representative shallow (< 5 mm water depth) and deep profiles (< 5 cm water depth) from two consecutive days of each experiment. In both experiments, profiles of dissolved O₂ revealed high vertical and temporal variability. The shallow profiles showed the formation of O₂ gradients in the uppermost water column with steadily increasing O₂ concentrations from the air-water interface to a water depth of ~1 mm (Figures 2A,B). The thickness of the SML was on average 1,100 μm (±295 μm). O₂ profiles below 1.1 mm water depth showed strong vertical variability with changing O₂ concentration of up to 30 μmol O₂ L⁻¹ within a range of 500 μm water depth (e.g., Experiment 2, Day 1, 10:26, Figure 2B).

From the gradients in the upper 1.1 mm we calculated diffusive rates. During Experiment 1, all rates ranged between −0.08 μmol O₂ L⁻¹ h⁻¹ and −0.37 μmol O₂ L⁻¹ h⁻¹. The diffusive rates during Experiment 2 ranged between 0.09 μmol O₂ L⁻¹ h⁻¹ and −0.58 μmol O₂ L⁻¹ h⁻¹ (Table 1).

In order to assess the temporal variability of O₂ concentration in the tank water, we calculated the temporal changes in O₂ concentrations in 1.5–4 cm water depth from the deep profiles. During the first experiment, an O₂ production up to 2.2 μmol O₂ L⁻¹ h⁻¹ during the day and O₂ consumption up to −1.2 μmol O₂ L⁻¹ h⁻¹ over night was observed (Figure 2C, Table 1). In the second experiment, increases in O₂ concentration during the day were lower (up to 0.8 μmol O₂ L⁻¹ h⁻¹) compared to the first experiment, but consumption of O₂ was enhanced (up to −4.6 μmol O₂ L⁻¹ h⁻¹) (Figure 2D and Table 1). In Experiment 1, the changes in O₂ concentration in the ULW between two profiles were clearly linked to PAR (Spearman's rank correlation coefficient (r_s) = 0.905, p = 0.005, n = 8, Figure 3A and Supplementary Table 2), but not in the second experiment (Figure 3A).

Using the data from our O₂-profiles, we calculated the theoretical biological O₂-production and -consumption with a simplified profile model, in order to obtain NCP rates required to influence O₂-gradients within the DBL. We considered the DBL to be 1.1 mm thick as deduced from all O₂ gradients. Equation (9) shows that the maximum relative difference between the concentration profiles with and without biological activity (i.e., R = reaction rate) increases with rising NCP rate and thicker DBL, whereas it decreases with increasing diffusion coefficient and absolute concentration levels. For an O₂ microelectrode, the relative precision is ~1%, and the O₂ diffusion coefficient is ~2 × 10⁻⁵ cm² s⁻¹ at 20°C and salinity of 30 as measured in the tank. Thus, to be able to distinguish between profiles with and without biological activity, if C₀ and C_ULW(x) are e.g., 245 and 278 μmol L⁻¹, respectively (Figure 2A, Day 2 (17:33), we require a NCP rate such as (11)|Rl24D(CULW+C0)|>~1% or (12)|R| >~124.5 μ mol L−1 h−1. The lowest NCP rate calculated by the model using our profile data was 103.8 μmol O₂ L⁻¹ h⁻¹, implying that a NCP rate beyond ±103.8 μmol O₂ L⁻¹ h⁻¹ would be needed to cause detectable shifts in the O₂-gradients within the DBL of 1.1 mm thickness.

During both experiments, light and dark incubations were performed to determine NCP and O₂ consumption rates, respectively. In both experiments, nearly all dark incubations of SML samples showed very strong O₂ consumption, causing O₂ concentration at t_end to be close to 0 μmol O₂ L⁻¹ (Supplementary Figure 4). Thus, O₂ consumption rates could not be reliably determined for the present study as we could not rule out limitation of O₂ for the heterotrophs. This O₂-limitation did never occur in the NCP incubations. Consequently, only NCP rates, which were not O₂-limited at the end of the incubations, were used as a measure for metabolic activity in this study.

During both experiments, NCP showed pronounced rates in SML samples ranging between −21.3 and 44.8 μmol O₂ L⁻¹ h ⁻¹, indicating strong heterotrophic and autotrophic metabolism, respectively. NCP rates in the ULW fluctuated in a narrower range between −8.0 and 3.0 μmol O₂ L⁻¹ h⁻¹ (Table 1). In the first experiment, NCP dynamics were positively related to PAR (Figure 4) indicating a strong influence of both auto- and heterotrophic processes in the tank water. Despite high PAR values during Experiment 2, NCP rates were never positive in SML or ULW, indicating a shift toward a net heterotrophic system (Figure 4).

To explore whether metabolic activity of neuston and plankton, as determined by NCP rates, could explain the temporal shifts in O₂ profiles, NCP was plotted against changes in O₂ concentration in the ULW. Both NCP rates in the SML and ULW were related to changes in O₂ concentration in the tank water (Figures 3B,C). However, this relationship was statistically significant only for NCP rates of SML samples in Experiment 1 (r_s = 0.820, p < 0.001, n = 15, Supplementary Table 2).

The abundance of auto- and heterotrophic organisms generally differed between the two experiments, and between the SML and ULW. Bacterioplankton abundance, i.e., in the ULW, decreased from 5.1 × 10⁷ cells mL⁻¹ to 3.4 × 10⁷ cells mL⁻¹ during the first experiment and slightly increased from 2.8 × 10⁷ cells mL⁻¹ to 4.1 × 10⁷ cells mL⁻¹ during the second experiment (Figure 5A). Bacterioneuston abundance increased from 6.0 × 10⁷ cells mL⁻¹ to 1.0 × 10⁸ cells mL⁻¹ during the first experiment and from 2.7 × 10⁷ cells mL⁻¹ to 8.8 × 10⁷ cells mL⁻¹ during the second experiment (Figure 5A). Thus, an enrichment of bacteria in the SML compared to the ULW was always apparent (up to 3-fold, Day 6). A single exception was Day 1 of the second experiment, where abundances in the SML were reduced by 8 × 10⁵ cells mL⁻¹ (Figure 5A). The abundance of small autotrophic cells (< 50 μm) ranged between 2.6 × 10⁵ and 1.6 × 10⁶ cells mL⁻¹ and showed no consistent differences between the SML and the ULW during the first experiment. During the second experiment, small autotrophic cells were generally less abundant compared to the first experiment and always depleted in the SML compared to the ULW (Figure 5B).

Very high Chl a concentrations were measured both in the SML (max: 450 μg Chl a L⁻¹) and the ULW (max: 14 μg Chl a L⁻¹) during both experiments (Figure 5C). Chl a concentration in the SML was up to 34-fold higher compared to the ULW during the first experiment. During the second experiment, Chl a concentration in the ULW declined while it simultaneously increased in the SML (up to 38-fold compared to the ULW), but Chl a concentration was overall lower compared to the first experiment. This decline in Chl a concentration was independent of PAR, which peaked at 874 μmol m⁻² s⁻¹ during the second experiment, indicating the shift toward a net heterotrophic system. In both experiments, high PAR values (>500 μmol m⁻² s⁻¹) were associated with lower quantum yields in the ULW (Supplementary Figure 5) especially at high light intensities around noon.

Without any influence of chemical enhancement or biological activity, exchange processes across the air-water interface in our tank system would be solely diffusion-limited (Pereira et al., 2016). In the present study, the slope direction of O₂ gradients within the upper 1.1 mm always showed higher concentration of O₂ in water compared to air (Figure 2), except for the very last day, where the tank water became a sink for O₂, most likely due to the increasing effect of net heterotrophy (Table 1). Thus, the diffusive rates calculated from the gradients in the upper 1.1 mm were generally negative, indicating a permanent diffusion of O₂ from water to air (Table 1). However, the diffusive rates (mean = −0.25 ± 0.18 μmol O₂ L⁻¹ h⁻¹, n = 18) could only insufficiently explain the temporal dynamics of changing O₂ concentration in the tank (mean = −1.02 ± 1.77 μmol O₂ L⁻¹ h⁻¹, n = 18), suggesting that the latter was rather driven by biological O₂ production and consumption.

Net community production (NCP) rates measured in ULW (mean = −1.35 ± 2.19 μmol O₂ L⁻¹ h⁻¹, n = 18) and SML samples (mean = 2.63 ± 17.60 μmol O₂ L⁻¹ h⁻¹, n = 13) were generally exceeding the diffusive rates. Moreover, in the first experiment, changing O₂ concentration in the ULW seemed to be related to plankton and neuston NCP (Figures 3B,C). In the second experiment, the overall decreasing O₂-concentration in the ULW was accompanied by general O₂ consumption in incubated samples from the SML and the ULW (Figures 3B,C). Calleja et al. (2005) found planktonic metabolism in the top 2 cm of the ocean surface in the Northeast Atlantic to influence and control direction of pCO₂ exchange but found no significant relationship for changes in pCO₂ and NCP in 5 m water depth. Making in situ measurements across the air-water interface, we can confirm a link between NCP and changing O₂ concentration within the upper 5 cm of the water column.

Even though biological communities influenced the temporally changing O₂ concentration in the tank, we could not find evidence for direct metabolic control of the neuston on in situ O₂ gradients at the air-water interface itself (Figure 2). For example, we did not observe distinct curvatures in the upper 1.1 mm of the O₂ gradients (i.e., DBL), which may result from O₂ production or consumption. We made this observation despite the fact that NCP of the neuston was on average 17-fold enhanced compared to the plankton. This is supported by our profile model, which shows that O₂ consumption / production in excess of 103.8 μmol L⁻¹ h⁻¹ within the DBL is required to measure detectable effects on O₂ gradient curvatures. However, during this experiment, the highest positive and negative NCP rates were measured in SML samples and were 44.8 μmol L⁻¹ h⁻¹ and −21.3 μmol L⁻¹ h⁻¹, respectively, and, thus, below the threshold of ±103.8 μmol L⁻¹ h⁻¹ indicated by the model.

Furthermore, signal fluctuations of the profiles varied more than 1%, which was applied in the idealized model settings. These fluctuations were probably caused by microturbulence, e.g., due to buoyancy fluxes, in the DBL. Thus, the reaction rates (i.e., NCP rates) must have been even higher in this study to be detectable over the fluctuations. It is also important to note that the profile model assumes a sharp delineation between a stagnant diffusive SML and a completely mixed ULW. In reality, there will be a gradual transition between the two phases, governed by turbulent forces. This will make higher rates necessary than those indicated by the model to recognize and quantify reaction rates within the SML. Overall, our findings suggest, that any profound contribution of neuston metabolism to O₂ gas exchange across the air-water interface is unlikely.

The O₂ gradients observed in our experimental setup indicate the formation of a thick surface film stabilizing the uppermost 1.1 mm of the water column. We observed a substantial accumulation of auto- and heterotrophic organisms in our SML samples compared to the ULW. The enrichment of bacteria in the SML is highly dependent on SML integrity (Stolle et al., 2010; Rahlff et al., 2017a), and after filling the tank, the water became increasingly settled until stagnancy, i.e., no induced mixing except microscale turbulence by buoyant fluxes and movement of the microsensors. This caused bacterial abundance to increase within the SML during both experiments (Figure 5A). Likewise, we observed high numbers of small autotrophs, Chl a concentrations of up to 38-fold in SML over ULW (Figures 5B,C). These enrichments have been previously observed in the field (Hardy and Apts, 1984). Nevertheless, in our study, autotrophic biomass and bacterial abundances were about one order of magnitude larger compared to the natural environment (Poremba et al., 1999; Sperling et al., 2012). Despite this very high biomass accumulation, we identified typical members of natural marine bacterial and eukaryotic communities with strong differences between the SML and ULW (Supplementary Figures 6, 7). Some of the bacterial taxa belonged to the families of Rhodobacteraceae and Flavobacteriaceae (Supplementary Table 3), which are common SML inhabitants (Franklin et al., 2005). They were attributed to the presence of polysaccharide microgels (Taylor and Cunliffe, 2017), which strongly enrich in the SML (Wurl and Holmes, 2008). In addition, we detected an organism related to the diatom Chaetoceros calcitrans (Supplementary Table 3), which may produce large amounts of extracellular polysaccharides (Corzo et al., 2000) and, thus, could have supported the pronounced SML formation in our study.

In our experiments, the autotrophic community responded to diurnal light availability as measured by decreased quantum yields in the ULW during high light exposure around noon (Supplementary Figure 5). Nevertheless, the strong positive NCP rates in the SML (Figure 4) suggest that inhibition of photosynthetic activity by solar radiation was lower compared to field observations (Williams et al., 1986; Ignatiades, 1990) probably due to shielding by the greenhouse. We observed light-driven autotrophic O₂ production during the day and heterotrophic O₂ consumption during the night in the first experiment. In the second experiment, phosphate-limitation (as measured from a single sample on the last day of the experiment) most likely caused a decay of the autotrophic community, resulting in net heterotrophic metabolism in the tank water (Figure 4). NCP rates in the SML were generally higher compared to ULW (Figure 4 and Supplementary Figure 4), and enhanced metabolic activity, especially of heterotrophic bacteria, is a well-known feature of the SML (Obernosterer et al., 2005; Reinthaler et al., 2008). Other studies reported SML NCP rates between −0.4 to +1.0 μmol O₂ L⁻¹ h⁻¹ for different coastal marine systems (Obernosterer et al., 2005; Rahlff et al., 2017b). The NCP rates in the SML of the present study (−21.3 to 44.8 μmol O₂ L⁻¹ h⁻¹) are several 10-folds higher, which was most likely caused by the large number of auto- and heterotrophic microorganisms.

Overall, neuston biomass and activity was strongly enhanced in the thick surface film of our tank. In the natural environment, such films with biofilm-like properties are known as slicks (Wurl et al., 2016), and can involve high biomass, e.g., during cyanobacteria blooms (Sieburth and Conover, 1965; Wurl et al., 2018). However, our study suggests that the natural neuston community will not have significant, direct effects on O₂ gas exchange, although such thicker films limit diffusion (Wurl et al., 2016).

Using microsensors, we generally observed a constant change of O₂ concentration within the uppermost 1.1 mm of the water column. This implies that the surface layer in our experimental setup had a vertical dimension of ~1.1 mm. Zhang et al. (2003) conducted a similar microsensor study and defined the SML as a “layer of sudden change” with a thickness of ~60 μm. Interestingly, these authors used natural seawater, which had been pre-filtered with a pore size of 0.45 μm, thus excluding large particles and most organisms in contrast to our study. Measuring within cyanobacterial aggregates floating at the air-water interface, Ploug (2008) could show that O₂ concentration changed strongly within few millimeters depth. Thereby, gross primary production was only detectable in the uppermost 1.1 mm, which very well resembles the dimension of the thick surface film observed in our study. Thus, we consider the SML in our study to be 1.1 mm thick, following the definition of Hunter (1997), that the vertical dimension of the SML is defined by “[…] physical, chemical, or biological properties that are measurably different from those of adjacent sub-surface waters.” Based on this, our profile model assumes a DBL thickness of 1.1 mm, and that metabolic activity measured in the SML samples is homogeneously distributed within the DBL. This may not be true as we have only measured NCP in the samples taken with the glass-plate, which collects the SML with ~40–100 μm layer thickness (Carlson, 1982a). Future studies will need to unravel if metabolic activity is changing vertically within the upper millimeters of the water column.

If we assume that the high neuston NCP rates measured in our study would only affect the ULW and not the atmosphere, the small volume of the SML compared to the remaining ~700-L tank volume would still allow for no more than 7.1% (see Supplementary Table 4) neuston contribution to O₂ concentration changes in the ULW. Conditions in the field would generally be more variable, and often preclude such strong accumulation of biomass in the SML except in slicks (Wurl et al., 2016). For example, cell abundances are typically enriched only at calmer sea states below ~5 m s⁻¹ (Rahlff et al., 2017a) and slicks persist until a similar wind speed threshold (Romano, 1996). Even though our study shows that microbial activity at conditions similar to slicks (i.e., low turbulence and high biomass) are insufficient to influence the exchange of oxygen, the increased thickness of the diffusion layer across slicks can reduce gas fluxes between the ocean and atmosphere (Salter et al., 2011; Wurl et al., 2016). Surfactants are hereby of special interest as they show a global distribution, are frequently enriched in the SML and slicks (Wurl et al., 2011b) and known to significantly reduce O₂ evasion rates (Frew et al., 1990). Surfactants could have played an important role in our tank due to the high biomass observed because phytoplankton and bacteria produce surfactants (Satpute et al., 2010). Due to limited SML sample volume we could only determine surfactant concentrations in a few samples, which show overall very high concentrations in the tank (2,947 ± 3,574 μg Teq L⁻¹; n = 16) compared to the natural environment (Ribas-Ribas et al., 2017) and 4.2-fold ± 2.3 (mean ± standard deviation, n = 8) enrichments within the SML (Supplementary Table 5). Even if we did not observe direct metabolic contributions of neuston organisms to the formation of O₂ gradients across the air-water interface, indirect influence on O₂ gas-exchange, e.g., by the organismal release of surfactants and increasing DBL thickness, cannot be excluded.

In addition, the neuston community may potentially influence the exchange of certain (trace) gases that have a low background concentration, e.g., carbon monoxide or methane (Conrad and Seiler, 1988; Upstill-Goddard et al., 2003). Whether under certain conditions, such as during extreme surface blooms of autotrophs or at sites of high organic matter availability, e.g., in aquatic foams or slicks, neuston metabolic processes could be influential for exchange processes of these trace gases, needs further investigation.

Microelectrodes have been used to investigate gradients across different marine boundary layers, e.g., sediments and microbial mats (De Beer et al., 1997), and aggregates (Ploug and Grossart, 1999). Despite the small dimensions of their tips, disturbance of the boundary layers by the introduction of microsensors has been demonstrated above marine sediment (Nøhr Glud et al., 1994). In this study, potential artifacts of the profiles measured could originate from moving the sensor through the different layers of water, and from penetrating the sensor through the water surface from the air above. Capillary forces could force the water surface to rapidly creep up the hydrophilic glass sensor tip, once the sensor tip touches the surface. This could cause a deformation of the water surface and the DBL. Such a local deformation would instantaneously raise the top of the DBL relative to the sensor tip, effectively placing the sensor tip inside the DBL. This relative shift in sensor tip position would cause a step change in the oxygen profile, both if the next data point is measured immediately and if a new diffusion steady state is awaited, which will take some tens of seconds for a 1.1 mm thick DBL (Crank, 1975). Furthermore, a deformation of the water surface in principle changes the diffusion geometry, resulting in non-linear concentration profiles inside the DBL. We could neither detect discontinuities in our oxygen profiles at the air-water interface (Supplementary Figure 8) nor systematic non-linearity in the DBL. Consequently, we conclude that artifacts due to disturbance of the water surface and the DBL by the microsensor play an insignificant role for our results.

Microprofiling by O₂ microsensors provides a very sensitive method for studying O₂ profiles at the air-water interface at high spatial resolution. The SML can be assumed to be at equilibrium with the atmosphere, and thus provides an internal reference point for calibration. This eliminates the errors associated with a separate calibration procedure that requires very careful temperature control relative to the experimental set-up. However, albeit their sensitivity, microsensors could not be used to quantify O₂ production and consumption rates in the SML directly during profiling, because NCP rates were relatively low compared to the O₂ background concentrations in the tank water. Assuming background concentrations to become negligible, the relative difference becomes correspondingly large (Equation 9). In this case, it is the absolute difference expressed in Equation (13), rather than the relative difference, that becomes limiting. Under favorable conditions, an O₂ sensor can have an absolute signal precision on the order of 0.1 μM near zero concentration. With a diffusion coefficient of 2 × 10⁻⁵ cm² s⁻¹ and a DBL of 1.1 mm, detecting the activity on the profile thus requires: (13)Rl28D>~0.1 μM or (14)R>~5 μmol L-1 h-1 This rate is in the same order of magnitude as many of the rates measured in this experiment. These rates could be quantified in situ by microprofiling through the air-water interface if the O₂ concentration in both the headspace and water phase has first been lowered artificially, e.g., by flushing with N₂. Such measurements would only be valid under the assumption that the process rates are not affected by the low concentration levels. However, this assumption seems reasonable, as values for the apparent half-saturation constant of the Michaelis-Menten equation (K_m) for microbial oceanic O₂ consumption have been shown to be in the nanomolar range (Tiano et al., 2014; Garcia-Robledo et al., 2016). Also, for compounds with a natural low ambient concentration and where a suitable microsensor exists, e.g., for nitrous oxide (N₂O), SML microprofiling can likely be used to determine reaction rates in situ. Applying microsensors in the field is challenging mostly because sea surface motion, i.e., waves, already exceeds the spatial measurement resolution of the sensor. Collecting natural SML samples onboard research vessels might help to approach in situ conditions as close as possible, but needs to consider other obstacles, e.g., the alteration of surface area of the sample, as well as onboard vibrations that interfere with sensor measurements.

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