Surface seawater samples (~5 m) were collected in the South Atlantic Bight (SAB) off the coast of Georgia (GA), USA onboard the R/V Savannah in June 4–6, 2016 and locally off the Skidaway River Estuary (SRE) dock (July 20, 2016) and the Altamaha River (August 5, 2016) (Table 1). All sampling equipment (tubing, filters and containers) were acid-cleaned (overnight in 0.01 M HCl), Milli-Q water (Millipore; >18 MΩ) rinsed (5 times), and dried prior to sampling. Samples were filtered either directly from Niskin bottles or pumped through 0.2 μm filters (Whatman; Polycap-75AS) into 10 or 20 L acid-cleaned polyethylene carboys (Nalgene) and stored at 4°C until use, for a period of up to 1 month. Our experience indicates that storage for seawater samples does not impact sample optics or the photochemical production rates for H₂O₂ (Powers et al., 2015). While all equipment in contact with seawater samples was acid cleaned as described above and nitrile gloves were used in handling samples, the R/V Savannah and the laboratory was not equipped with special trace metal processing facilities and thus, limited trace metal contamination could be present.

H₂O₂ was measured using a FeLume chemiluminescence (CL) system (Waterville Analytical) and modifications to the methods of King et al. (2007). Briefly, samples were mixed with 2 μM 10-methyl-9-(p-formylphenyl)-acridinium carboxylate trifluoromethanesulfonate (AE; provided by Dr. James J. Kiddle, Western Michigan University), which reacts with H₂O₂ to chemiluminesce at alkaline pH. A syringe was used to flush (≥1 mL) and fill a 195 μL Teflon sample loop (VICI 10-port valve) that was subsequently mixed with the AE reagent using a peristaltic pump (Rainin). The sample/carrier—AE mix was transported to a spiral flow cell (Global FIA) located in front of a photomultiplier tube (PMT; Hamamatsu HC135 PMT, 900 V, 400 ms integration) where it mixed with 0.1 M carbonate buffer to initiate CL. Decreased carbonate buffer pH (10.4) using 0.2 M HCl as the carrier prevented interference by precipitated Mg(OH)₂ and helped decrease background CL. A ~2 mM H₂O₂ stock solution was prepared from 30% H₂O₂ (J. T. Baker) and measured weekly for stability using its absorbance (molar absorptivity 38.1 M⁻¹ cm⁻¹ at 240 nm; Miller and Kester, 1988) on a single beam UV-Vis spectrophotometer (Agilent) with Milli-Q as the absorbance blank. Standards were prepared in seawater samples matching those used in irradiations and were checked for low or undetectable H₂O₂ starting values against the same seawater with added catalase (20 μL of 100 units L⁻¹ in 20 mL seawater; C100 Sigma-Aldrich). Under these conditions, the H₂O₂ detection limit, defined as three times the standard deviation of the blank, was 2 nM.

Prior to irradiation experiments, each seawater sample was brought to room temperature and refiltered through a 0.2 μm membrane filter (described above) and portioned into two acid-clean and combusted (450°C for 5 h) 1 L borosilicate bottles (Kimax). For each sample, one bottle was left unamended and the other was treated with an addition of the enzyme superoxide dismutase (SOD; S7571-15KU, Sigma) immediately prior to irradiation. SOD was stored at −20°C when not in use, and a fresh stock solution of ~1.5 × 10⁷ U L⁻¹ SOD (calculated from the manufacturer's reported activity) was prepared in Milli-Q and stored at 4°C for a period of up to 2 weeks. SOD activity was verified by monitoring O2- photoproduction in the absence of SOD and subsequently observing no O2- production in the presence of added SOD, using methods for superoxide described previously (Powers et al., 2015). 1.5 × 10⁴ Units L⁻¹ SOD (based on the manufacturer's assay) was used for all 15-min irradiation experiments with added SOD.

All irradiations were performed using a Suntest CPS solar simulator equipped with a 1.5 kW xenon lamp (Atlas). For time series and initial rate determinations, the solar simulator was fitted with a window glass filter to simulate sunlight reaching the earth's surface (~295 nm and above). All samples were distributed into 15 matched 10 cm cylindrical quartz spectrophotometric cells, filled leaving no headspace, and sealed using gas-tight caps fitted with Teflon-faced, butyl rubber septa. Cells were placed vertically below the solar simulator into a black, water-cooled, aluminum irradiation block fitted with a one inch gray PVC lid with 1.6 cm diameter apertures directly above the cell windows to eliminate residual off-axis photons. The Al block maintained the samples at 20°C and allowed no transfer of light between cells during irradiation. For H₂O₂ AQY determination, wavelength dependent experiments were also performed by removing the window glass filter and aligning the block, and therefore each cell, under a variety of evenly spaced Schott long-band-pass filters (280, 295, 305, 320, 380, 425, and 480 nm 50% cutoff). One dark control was kept in the same irradiation block as other samples but with all irradiance blocked. Additional dark control sample cells were kept covered in the water bath. The spectral downwelling irradiance entering each cell, E₀(λ) [mol(photons) m⁻² s⁻¹ nm⁻¹], was quantified at 1 nm intervals with a UV-Vis portable spectroradiometer (OL756; Optronic Laboratories) fitted with a quartz fiber optic cable and 2″ diameter integrating sphere. The OL756 was calibrated with a NIST standard lamp (OL752-10 irradiance standard; Optronic Laboratories).

Accurate and precise measurements of CDOM absorbance (A) are critical in determining the photons absorbed in these irradiation experiments. Absorbance spectra from 250 to 800 nm were measured using an UltraPath liquid waveguide cell (World Precision Instruments; Miller et al., 2002) for the clear Gulf Stream sample, with a high powered UV-Vis light source (Hamamatsu) and a fiber optic assembly attached to a MAYA2000-Pro UV-Vis spectrometer (Ocean Optics). The liquid waveguide cell of the UltraPath provides a 2 m pathlength, with absorbance sensitivity 20 times greater than standard spectrophotometers holding 10 cm cuvettes. Blanks, cleaning protocols, and refractive index corrections for salinity changes were performed as in Stubbins et al. (2006). For all samples, absorbance in each 10 cm irradiation cell was also measured from 190 to 1100 nm with a single beam UV-Vis spectrophotometer (Agilent) both prior to and post irradiation at room temperature using Milli-Q water as the absorbance blank. Absorbance offsets (O) were determined by averaging raw absorbance spectra over the 700–800 nm range and were subsequently subtracted from absorbance spectra. All corrected absorbance (A_corr = A−O) were then converted to Napierian absorption coefficients [a_g(λ); m⁻¹] with the equation: (1)ag(λ) = 2.303Acorr(λ)/L over the 250–700 nm range, where L (m) is the pathlength of the spectrophotometric or waveguide cell.

The photon dose absorbed by CDOM for each wavelength interval over the course of an experiment was calculated using the following equation (Hu et al., 2002): (2)Qa(λ) = E0(λ)(ag(λ) / at(λ)) S[1−exp(−at(λ) L)] where Q_a is the mol photons absorbed s⁻¹ by the sample, λ is wavelength (nm), E₀ is irradiance entering the top of the cell (mols photons m⁻² s⁻¹), S is the irradiated surface area of the spectrophotometric cell (m²), a_g and a_t are the absorption coefficients for CDOM and for the total solution respectively (m⁻¹), and L is the pathlength of the spectrophotometric cell (m). The experimental photochemical production under each cutoff filter is defined according to the equation: (3)dP/dt = AQY×Qa where the dP/dt is the measured production rate of the species in question (H₂O₂ here) per experiment. The intentional variation of Q_a using multiple irradiation spectra gives production results that allow use of an in-house MATLAB® routine for iterative, non-linear fitting to solve for the single AQY spectrum that best describes the production measured under all of the cutoff filters for a single experiment with the following equation (Ziolkowski and Miller, 2007; Xie et al., 2009; White et al., 2010; Powers and Miller, 2015b):

where m₁ and m₂ are fitting parameters and 290 is a reference wavelength (nm). The statistical robustness of the determined AQY spectrum was analyzed using the r² of a linear regression between measured [H₂O₂] in each sample cell vs. that predicted for each cell by the resulting AQY spectrum and spectral photon dose using Equation (3) (integrated from 280 to 600 nm) and by calculating the normalized root mean square error (nRMSE) between the two.

Using H₂O₂ measurements to provide O2- production rates for use in AQY calculations requires a known O2-:H₂O₂ reaction stoichiometry as discussed above. Like Petasne and Zika (1987), we used SOD to kinetically force the well-known 2:1 dismutation reaction described by Bielski et al. (1985). To ensure that all photochemically generated O2- in our samples was catalyzed by SOD via the disproportionation reaction to H₂O₂, we measured photochemical H₂O₂ generation with a range of added SOD concentrations (Figure 1). SRE water, potentially containing higher levels of possible reactants with O2-, was used to determine the concentration of SOD needed to kinetically out-compete all other possible reactions with O2- and ensure the 2:1 dismutation stoichiometry during a 15-min irradiation. An asymptote for H₂O₂ production was noted in our SOD addition series, indicating that 3.0 × 10³ Units L⁻¹ is all that it takes to ensure maximal H₂O₂ production. Consequently, we chose to add 1.5 × 10⁴ Units L⁻¹ SOD (5 × this activity) to each sample in all subsequent irradiation experiments containing SOD.

It is also essential that SOD is not denatured, and maintains its activity for the duration of its exposure to sunlight. Previous tests performed during our work in the Gulf of Alaska (Powers et al., 2015) using SOD with direct O2- measurements confirmed that SOD is not deactivated by extended exposure to maximum output of the Suntest CPS solar simulator, exhibiting about 3 × the photon flux used in the AQY experiments performed here. Seawater in water jacketed beakers with quartz covers containing added SOD of the activity levels shown in Figure 1 showed no measurable O2- production throughout 60 min of constant exposure (Powers and Miller, 2014; Powers et al., 2015), indicating that SOD catalyzed dismutation continues at maximal rates in extended, full sun exposure. Consequently, results from our SOD-added experiments are assumed to represent H₂O₂ production exclusively via the SOD catalyzed disproportionation reaction.

One last critical issue for any accurate AQY calculation made using a single time point measurement is the assumption that photochemical production is linear to the point at which the sample is sacrificed and measured. As noted previously, H₂O₂ photochemical efficiency decreases with prolonged exposure (Kieber et al., 2014; Powers and Miller, 2015a). Therefore, to ensure H₂O₂ photoproduction was always a linear function of photon dose during our AQY experiments, we performed broadband irradiation of SRE water and plotted the resulting data against Q_a(λ) integrated from 280 to 600 nm (mol photons absorbed) at each time point over the course of the irradiation (Figure 2). After ~2 h of irradiation, amounting to a photon dose of 0.15 mmol photons absorbed, H₂O₂ photoproduction slowed. Therefore, for all seawater irradiations used in AQY determinations in this study, we limited sample photon doses to 0.15 mmol photons absorbed or less.

Laboratory determined AQY spectra for all SAB seawater samples decreased monotonically from low UV to visible wavelengths and fell within a relatively narrow range (Figure 3; Table 1A). H₂O₂ AQY spectra for samples with no SOD treatment ranged from between 2.9 and 4.2 × 10⁻⁴ (mol H₂O₂ /mol photons absorbed) at 300 nm to between 0.24 and 0.37 × 10⁻⁴ (mol H₂O₂/mol photons absorbed) at 400 nm. The Altamaha River sample had the highest photochemical efficiency, with AQY values of 10 × 10⁻⁴ (mol H₂O₂/mol photons absorbed) and 1.0 × 10⁻⁴ (mol H₂O₂/mol photons absorbed) at 300 nm and 400 nm, respectively. With added SOD, the magnitude of the H₂O₂ AQY spectrum for all samples (riverine to Gulf Stream waters) increased in the UVB and to a lesser extent, in the UVA, resulting in higher slope coefficients (m₂ values, Table 1). This increase was largest for Gulf Stream waters and lowest for the Altamaha River sample. The AQY spectrum determined for each sample was also used to predict photoproduced H₂O₂ in each sample cell (i.e., all light treatments). The resulting measured vs. predicted H₂O₂ correlations had r² ≥ 0.95 and normalized root mean square errors (nRMSE) <8% for all but the untreated Gulf Stream sample (r² = 0.91; nRMSE = 9.2%).

Superoxide photoproduction rates were determined by assuming that the 2:1 dispropotionation stoichiometry is valid in the presence of added SOD and the O2- production rates are double those required to produce observed H₂O₂ (i.e., O2- = 2 × H₂O₂). AQY spectra for O2- were then determined by repeating the non-linear curve fitting routine using 2 × the H₂O₂ data for each sample cell containing added SOD and their respective Q_a(λ) spectra (Table 1B). Ratios of O2- to H₂O₂ photoproduction were then determined in two ways: a broadband estimate and an AQY estimate with both listed in Table 2. The broadband estimate was determined as the average ratio between 2 × H₂O₂ generated with added SOD to H₂O₂ generated with no treatment for samples under the 280 and 305 nm cutoff filters. The AQY estimate was determined from the slope of H₂O₂ with no SOD treatment compared to that with added SOD for all spectral treatments during the AQY experiment (Figure 4). The O2-:H₂O₂ ratio was then determined as two divided by the slope of these regressions. Because the addition of SOD increased the H₂O₂ AQY spectra most in the UVB (i.e., <320 nm), O2-:H₂O₂ ratios were always higher under 280 cutoff filters when compared to the ratio determined for samples under the 305 nm cutoff filters for all samples but the Altamaha River (Table 2). Linear regressions of H₂O₂ production with and without added SOD are shown in Figure 4. The slope coefficients for these regressions decrease with decreasing CDOM absorption [Table 2, a_g(325) values listed in Table 1] ranging from 0.59 for the Gulf Stream sample to 1.07 for the Altamaha River sample. Results indicate that reductive and oxidative sinks are comparable for the Altamaha sample but, for the Gulf Stream sample, about ~40% of the superoxide generated does not lead to H₂O₂ formation.

H₂O₂ AQY values fell in a narrow range for SAB seawater samples (i.e., from 1.7 to 2.3 × 10⁻⁴ at 325 nm) despite a much larger range in CDOM absorption coefficients (0.24 to 6.5 m⁻¹ at 325 nm). Previous studies have found that the H₂O₂ AQY spectra were similar in various marine waters (Yocis et al., 2000; Kieber et al., 2014), suggesting that differences in AQY values between studies could be due to the large activation energy (E_a) for this wavelength and temperature dependent photoproduct (average E_a from 17 kJ mol⁻¹ at 290 nm to 32 kJ mol⁻¹ at 400 nm; Kieber et al., 2014). In contrast, Powers and Miller (2014) found lower H₂O₂ AQY spectra in oligotrophic waters compared to coastal waters. Because linear accumulation of H₂O₂ was not confirmed in this study, differences in AQY spectra could reflect differences in irradiation time and a lack of reciprocity for H₂O₂. However, higher H₂O₂ photochemical efficiency has generally been observed in freshwaters (Cooper et al., 1988; Moore et al., 1993; O'Sullivan et al., 2005) when compared to both coastal and open ocean seawater (e.g., Yocis et al., 2000; Kieber et al., 2014; Powers and Miller, 2014).

In our study, the H₂O₂ AQY for the Altamaha River was 5.4 ± 0.7 times larger at 325 nm than all SAB seawater samples at the same wavelength, despite having a_g(325) values ranging from ~2 (SRE samples) to over 50 (Gulf stream) times larger (Table 1). The increased H₂O₂ AQY spectra for riverine water may reflect its terrestrial DOC source and a low exposure history. Zhang et al. (2014) demonstrated that the addition of phenol donors substantially increases the photoproduction of H₂O₂ from CDOM and Sharpless et al. (2014) have shown that prolonged irradiation tends to decrease CDOM electron donating capacity while having very little effect on electron accepting capacity. Perhaps the increased photoreactivity of the Altamaha River sample reflects an increased presence of phenolic compounds/electron donors, compared to seawater samples, even including the darker estuarine waters of the SRE. However, the reader is reminded that other factors in natural waters also impact H₂O₂ photoproduction efficiency including pH (Dalrymple et al., 2010) and metal chemistry (Shaked et al., 2010), as well as temperature (Kieber et al., 2014). Therefore, further mechanistic studies would be required to tease out the possible causes for variability in AQY spectra in various natural waters.

The best approach to evaluate superoxide photochemistry is to measure it directly. However, its instability in aqueous solutions creates a number of analytical limitations (Heller and Croot, 2010c), especially with regard to the determination of AQY spectra. Common methods to generate superoxide standards involve either UV-C generation of superoxide via a ketone/alcohol photolysis (McDowell et al., 1983; Rose et al., 2008; Powers et al., 2015) or additions of KO₂ to alkaline solutions using trace metal clean techniques (Heller and Croot, 2010a,c; Diaz et al., 2013). Both methods generate μM concentrations of O2- that allow direct monitoring via UV absorbance at 240 nm. Superoxide from standards is added to seawater in nanomolar amounts but doesn't last very long and subsequent monitoring of superoxide decay, most commonly via flow injection analysis with chemiluminescent detection (FIA-CL; Burns et al., 2012), is required for calibration of the system. Measuring photochemical superoxide production in seawater samples is subject to similar analytical difficulties arising from O2- instability in seawater. FIA-CL requires precise and continuous pumping of the sample to a flow cell below the detector. Currently our irradiation system for AQY determinations does not facilitate direct measurement of O2- from our individual, optically isolated, sealed spectrophotometric cells. Pumping samples would be limited to 30 ml and create an unacceptable optical condition, introducing a changing gas volume that would create undefined changes in refraction and reflection. Thus, as a practical alternative to direct measurements of O2- for deriving AQY spectra, we calculated O2- from the analysis of H₂O₂ with SOD additions. The effective use of SOD in this way is supported by reports of its stability in 0.2 μm filtered seawater (Petasne and Zika, 1987) and under solar radiation (Cooper and Zika, 1983; Draper and Crosby, 1983; Petasne and Zika, 1987) but it does lead to further consideration of both formation and decay mechanisms for superoxide in natural waters.

Previous work, using added phenol donors and borohydride reduction of CDOM prior to photochemical irradiations, has shown that the likely mechanism for superoxide photoproduction involves reaction of phenol electron donors with photo-excited triplet states of aromatic ketones within the DOM pool, generating ketyl radicals that react with O₂ to form superoxide (Zhang et al., 2014, 2012). Additionally, Zhang et al. (2014) noted that the AQY for phenol loss was 2- to 10-fold higher than that observed for H₂O₂ production, consistent with superoxide formation rates greater than twice that for H₂O₂. In other words, their results are consistent with dominant superoxide decay involving oxidative pathways in the Suwannee River Fulvic and Humic Acid (SRFA and SRHA, respectively) solutions studied. A recent study by Zhang and Blough (2016) determined photoproduction rates of reducing intermediates using a nitroxyl radical probe to estimate the production rates of superoxide in SRFA and SRHA solutions. They observed production rates of reducing intermediates that were 13 times greater than those for H₂O₂ during polychromatic irradiations, similar to irradiation conditions in our study. Their results suggest that up to 67% of O2- decays via oxidative pathways in these samples. Interestingly, we observed changes in the H₂O₂ AQY spectral shape with added SOD that may suggest additional oxidative pathways. Samples in our study showed increased efficiency in the UVB (<320 nm) and often the UVA (~340–360 nm) portion of the AQY spectrum (Figure 3). Consequently, all O2- AQY spectra calculated using SOD additions, had steeper slopes than those observed for H₂O₂ without SOD. Since SOD should not alter photochemical formation mechanisms for O2-, this apparent increase in efficiency at low wavelengths could represent a situation where SOD effectively outcompetes a natural phototransient oxidative sink generated by low UV wavelengths. This explanation is, of course, highly speculative and may simply represent natural variability in our AQY data. It does, however, suggest the presence of potential transient pathways for O2- generated by UV solar radiation. The formation of O2- sinks during photochemical irradiation still requires further study because direct O2- photoproduction measurements made to date have been determined from measured O2- steady-state concentrations during irradiation and dark decay post irradiation (e.g., Powers and Miller, 2015a; Powers et al., 2015) and raises the question “Do measurements of dark decay capture the complete story for O2- redox pathways in sunlit surface waters?”

One potential proxy for partitioning superoxide sinks could be CDOM absorbance. The literature contains positive correlations between CDOM absorption coefficients and pseudo-first order decay constants for O2- in coastal waters (Goldstone and Voelker, 2000) and negative correlations between H₂O₂ quantum yields and the E2/E3 ratio (i.e., absorbance at 254 nm divided by that at 365 nm) (Dalrymple et al., 2010). We could not verify the E2/E3 trend with our small sample size (r² = 0.16) but did find a negative correlation between O2-:H₂O₂ ratios and a_g(325) (r² = 0.69), albeit largely biased by a small sample set and high CDOM absorbance for the Altamaha sample. This approach may be of limited use even though our preliminary data suggests that the proportional oxidative sink for superoxide in blue water (O2-:H₂O₂ ratio of 3.4) is far larger than coastal waters with higher CDOM. On the other hand, ratios reported for open ocean waters in the Gulf of Alaska (Powers et al., 2015) had a large range of O2-:H₂O₂ ratios (2.2–9.8) in surface and mesopelagic (1000 m) waters despite low variability in CDOM absorption coefficients [a_g(325) from 0.13 to 0.39 m⁻¹]. Clearly, CDOM absorption should not be the only factor considered when estimating the relationship between O2- and H₂O₂ photoproduction. Work by P. Croot's group (e.g., Heller and Croot, 2011, 2010a; Wuttig et al., 2013) has shown dominant pathways for O2- decay to include copper (Southern Ocean), manganese, or CDOM (Tropical Atlantic). These factors and others (e.g., metal speciation, pH, temperature), that contribute to O2- photoproduction and its subsequent decay to either O₂ or H₂O₂ require further study.

From a practical point of view, it will require more than one stoichiometry if H₂O₂ AQY data is to be used to infer O2- photoproduction data for large-scale photochemical models, even in the blue waters that dominate the world's oceans. For now, we can only say with confidence that for the majority of the world's oceans, an oxidative sink for O2- is likely the dominant pathway for superoxide decay. Based on our limited data, instead of assuming that the 2:1 stoichiometry predicted from O2- dismutation is the same as it ever was, a more reasonable estimate of open ocean O2-:H₂O₂ stoichiometry is probably closer to 4:1. Previous calculations of O2- photochemistry based on H₂O₂ measurements (or AQY data) that use a 2:1 O2-:H₂O₂ stoichiometry probably are vast underestimates of true O2- photoproduction rates in the surface ocean. Until future work allows us to understand the specific relationship (or lack thereof) between O2- reactions and H₂O₂ production, global estimates of O2- from H₂O₂ remain highly uncertain. Until such time, measurements of H₂O₂ photoproduction in the presence of sufficient SOD represent a convenient way to predict O2- AQY spectra. Here we report only four O2- AQY spectra from a range of natural waters, but encourage further use of this approach to amass the data required to address the variability in O2- AQY spectra in blue waters that will support new and better global models of the photochemical redox chemistry driven by superoxide production.

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.