Reactive oxygen species (ROS, including superoxide, hydrogen peroxide, and hydroxyl radical) are critical for enabling abiotic reaction paths between organic carbon and atmospheric oxygen in surface waters. Abiotic ROS production in seawater is usually attributed to photoprocesses involving the direct reduction of oxygen by photoexcited natural organic matter or by Fe(II) generated by photoinduced ligand to metal charge transfer (Zepp et al., 1998; Powers and Miller, 2014). In the latter case, oxidizable ligands can include a wide variety of organic molecules and some ligands that are not ordinarily considered reductants, including water and chloride (Sulzberger and Laubscher, 1995; Sima and Makanova, 1997; Voelker et al., 1997). However, there are other abiotic sources of reductive equivalents that can reduce Fe(III) to Fe(II) without the need for sunlight; including hydrogen sulfide, polysulfides, some forms of organic carbon (e.g., polyhydroxylated phenols, organothiols etc.), and superoxide (Poulton et al., 2004; Carey and Taillefert, 2005; Larsen et al., 2006; Ma et al., 2006; Konovalov et al., 2007; Rickard and Luther, 2007; Gartman et al., 2011; Johnston et al., 2011; Chirita and Schlegel, 2012; Wan et al., 2014; Chirita and Schlegel, 2015; Havig et al., 2015; Peiffer et al., 2015; Duinea et al., 2016). Our own interest in Fe(II) is associated with the tidally driven efflux of anoxic porewater and this has led us to investigate the potential for ROS formation as a consequence of the non-photochemical reduction of Fe(III) by sulfide and other reduced sulfur species (Figure 1).

The oxic portion of the biosphere is a metastable mixture of different oxidation states of carbon, sulfur and oxygen energetically poised from equilibrium by the net rate differentials between photosynthetic carbon fixation and its metabolic or abiotic oxidation (Watson et al., 1978). Sedimentary carbon burial widens this gap by imposing a significant mass transfer limitation on the rate of carbon transport between the lithosphere, atmosphere and hydrosphere. It also restricts oxygen transport, forcing microbial metabolism of buried material to rely on alternative electron acceptors such as sulfate or carbon dioxide. The anaerobic microbial metabolism of buried carbon results in the reduction of ~11.3–75 × 10¹² moles of sulfate to sulfide per year in marine sediments and coastal marshes (Bottrell and Newton, 2006; Luther et al., 2011; Bowles et al., 2014). This range is compiled from recent efforts to reconcile older sulfate reduction estimates based on spatial averaging to more recent measurements correlating the global carbon flux to sediment with sulfate reduction (see the recent work by Bowles et al., 2014 and references therein). Based on the more conservative estimate of 11.3 × 10¹² moles of microbially produced sulfide/yr, and applying accepted percentage outcomes for the fate of sulfur in the sulfur cycle, approximately 2.3 × 10¹² moles of this is immobilized annually in the process of pyrite burial. The remaining 9.0 × 10¹² moles sulfide is reoxidized and returned to the water column (primarily as sulfate). The direct oxidation of sulfide by dioxygen is thermodynamically favorable but kinetically unfavorable and requires the intercession of a catalyst such as sedimentary Fe(III) or Mn(IV) (Equations 1–9; Vazquez et al., 1989; Roden et al., 2004; Carey and Taillefert, 2005; Bottrell and Newton, 2006; Aller R. C. et al., 2010; Johnston et al., 2011; Kubo et al., 2011; Luther et al., 2011; Lin et al., 2012; Rickard, 2012; Bowles et al., 2014; Murphy et al., 2014).

Here we focus on the impact of sulfide oxidation on the ferric/ferrous iron system given its relative geographical importance, kinetic facility, and potential for generating ROS. The non-photochemical interaction between the carbon, oxygen and sulfur cycles as ROS sources is interesting because the potential ROS generation capacity is so large, as indicated by the number of moles of Fe(II) produced/yr globally by the reoxidation of microbially produced hydrogen sulfide (vide supra; Bottrell and Newton, 2006; Luther et al., 2011; Bowles et al., 2014). The initial oxidation of sulfide and bisulfide by Fe(III)_aq or hydrous ferric oxides [represented collectively Fe(III)OH_x in the following equations] results in a mixture of Fe(II)-containing species as summarized in Equations (1–6) (Rickard, 1975; Afonso and Stumm, 1992; Rickard et al., 1995; Luther et al., 1996; Rickard and Luther, 1997, 2007; Poulton et al., 2004). These may undergo rapid oxidation by dissolved O₂, generating the reactive oxygen species superoxide (O2·-) and hydrogen peroxide (H₂O₂) while regenerating Fe(III)OH_x to continue sulfide oxidation (Equations 5, 6). The fate of HS· is unknown, although dimerization to produce H₂S₂ or reaction with excess HS⁻ to produce polysulfides are possible (Rickard et al., 1995; Rickard and Luther, 2007; Wan et al., 2014). The self-reaction of superoxide with its conjugate acid HOO·(Equation 8) is the kinetically favored outcome for superoxide at typical seawater pH unless there is a significant quantity of Fe(II) present, in which case superoxide may be reduced directly by Fe(II) to yield Fe(III) and hydrogen peroxide (Equation 9; Bielski, 1978; Bielski et al., 1985). (1)Fe(III)OHx+HS-⇆Fe(III)S-+H2O (2)Fe(III)S-⇆Fe(II)S· (3)Fe(II)S·+H2O⇆Fe(II)+HS·+ HO- (4)Fe(II)+HS-⇆[Fe(II)SH]+ (5)Fe(II)+O2⇆Fe(III)(OH)x+O2·- (6)[Fe(II)SH]++O2⇆Fe(III)(OH)x+ O2·-+ Soxidized (7)O2·-+H2O⇆HOO×+HO-(pKa=4.83) (8)O2·-+HOO·+H2O⇆H2O2+HO-+O2 (9)O2·-/HOO·+Fe(II)+H2O⇆H2O2 +HO-+Fe(III) If one conservatively assumes a 1:1 stoichiometry between Fe(III) and HS⁻ this implies a 1:1 conversion of the reductive equivalents in sulfide to superoxide. Given the most significant loss of superoxide is through formation of hydrogen peroxide (Equations 7–9) this implies an annual global sulfide-driven ROS formation potential of 9.0–4.5 × 10¹² moles, depending on whether the terminal ROS is superoxide or hydrogen peroxide and based on the assumption that 75–90% HS⁻ oxidation is abiotic (Bottrell and Newton, 2006, and references therein). There are relatively few estimates of annual photochemical ROS production in surface waters to compare this to, but recent work by Powers and Miller suggests the marine average is between 2.9 and 10.9 × 10¹² moles ROS in the top meter of the oceans/yr (Powers and Miller, 2014). There are other sources of sulfide not considered in this estimate such as volcanic and hydrothermal vents so the overlap between potential sulfide-driven ROS production and the estimated photochemical ROS production demonstrated here is not quantitative (Field and Sherrell, 2000; Moore et al., 2009a; Yucel et al., 2009; Gartman et al., 2011). There are also other sources of Fe(II), such as the direct reduction of Fe(III) by facultative anaerobes (Perry et al., 1993; Coates et al., 1998; Dollhopf et al., 2000; Sekar and DiChristina, 2014). Nonetheless, the estimated values are intriguingly close, certainly within the same order of magnitude, and that serves as justification for studying potential mechanisms for sulfide-driven ROS generation.

Here we report an investigation of the sources and mechanisms of ROS formation in sediment suspensions containing dissolved hydrogen sulfide (and bisulfide) and oxygen. These conditions are rarely observed in the open water column but are often encountered at the sediment/water interface. Specific examples include conditions associated with bioturbation, undersea mudflow, dredging, wave-driven mixing, and the trailing edges of the tidal prism (Luther et al., 1991; Precht et al., 2004; Aller and Blair, 2006; Rickard and Luther, 2007; Moore et al., 2009b; Aller J. Y. et al., 2010; Aller R. C. et al., 2010; Michaud et al., 2010; Santos et al., 2012). This work is a continuation of an investigation of ROS generation associated with the oxidation of reduced transition metals at sediment surfaces (Burns et al., 2010, 2011a,b; Murphy et al., 2014). It reports a test of the hypothesis that the conditions of frequent episodic anoxia set the stage for pulsed ROS production in marine littoral zones, focusing on the roles of hydrogen sulfide, ferric oxides, and pH on superoxide and hydrogen peroxide production (Figure 1). The oxidation of reduced sulfur species by dissolved Fe(III) and hydrous ferric oxides is much more rapid than by more crystalline iron oxides such as goethite, lepidocrite, or magnetite, and a central hypothesis tested by this work was that hydrous ferric oxides would play a correspondingly more important role in ROS formation than other ferric iron sources (Kostka and Luther, 1995; Poulton et al., 2004; Ma et al., 2006; Rickard and Luther, 2007; Cai et al., 2010; Luther, 2010; Luther et al., 2011). Sediments were collected from across the marine littoral zone in South Carolina, from the beach face to the forested inland edge of a saline Spartina alterniflora-dominated estuary. The addition of pulses of HS⁻ to aerated suspensions of collected sediments resulted in rapid Fe(II) production with concomitant superoxide and hydrogen peroxide formation. After a brief initiation phase superoxide essentially reached a steady state in the tested systems while hydrogen peroxide was more dynamic and sensitive to instantaneous concentration of Fe(II). This pump-and-probe experimental strategy of interrogating sediments for ROS production capacity was applied to all sediments tested and marsh sediments were additionally exposed to multiple sequential pulses of HS⁻ and hydrogen peroxide. All sediments tested consumed HS- but only marsh sediments produced significant ROS.

Hydrogen sulfide was added to separate, aerated suspensions of all sediments studied or a sediment-free control at pH 7.00, 7.50, 8.00, and 8.25. The systems were monitored for changes in sulfide, Fe(II), superoxide, and hydrogen peroxide for 120 min following sulfide addition. Sulfide consumption followed two profiles; an extremely rapid decay (95+%) in the first 30 s with a slow decay thereafter or an overall slow decay that in many cases was not statistically different than the control (Figure 3). Sediments from the marsh fell in the former category at all pHs studied, whereas the sediments from the intracoastal waterway, dune, or beach swash zone displayed the latter. Marsh sediments were also the only samples to experience significant increases in Fe(II) (as indicated by Ferrozine) over the timescale of the experiments (Figure 4), obtaining a maximum Fe(II) of 45.0 × 10⁻⁶ M very rapidly (<60 s) at pH 8.25. In contrast, intracoastal waterway sediment yielded the second highest apparent Fe(II) concentration of 7.3 × 10⁻⁶ M at ~600 s. These results justified focusing primarily on the marsh system. For marsh sediments the instantaneous concentration of Fe(II) at a given time never exceeded an amount corresponding to an Fe(II) yield above 15% of HS⁻ consumed at the same time. Given that total iron and hydrous ferric oxides were both in stoichiometric excess to HS⁻ in marsh sediment the low yield of Fe(II) was attributed to competitive losses of HS⁻ through formation of polysulfides (Giggenbach, 1972; Hoffmann and Lim, 1979; Fukuto et al., 2012; Cunha et al., 2016). The apparent half-life for Fe(II) was also considerably slower than would be expected based on existing Fe(II) oxidation models (e.g., over 1000 s; King et al., 1995; Santana-Casiano et al., 2005; Trapp and Millero, 2007; Burns et al., 2011a,b; Peiffer et al., 2015). These results can be explained by two exclusive models of the system; one where Fe(II) oxidation was slowed by the presence of a stabilizing ligand or one where Fe(II) oxidation was kinetically facile and the measured concentration was actually the product of simultaneous Fe(III) reduction and Fe(II) oxidation. The two models were resolved by examination of the concentration vs. time profiles for superoxide (Figure 5) in marsh sediments. Both pH conditions experienced a sudden increase in superoxide upon the addition of hydrogen sulfide, indicating at least some of the total Fe(II) was available for oxidation by dioxygen (Equation 5). The half-lives for superoxide were calculated based on dismutation (Equation 8) using the maximum observed superoxide as initial concentration (Equation 10): (10)t12=1k[O2−]0 where at pH 7.00 k = 5.01 × 10⁵ M⁻¹s⁻¹ and superoxide = 204 × 10⁻⁹ M and at pH 8.25 k = 1.78 × 10⁴ M⁻¹s⁻¹ and superoxide = 514 × 10⁻⁹ M (conditional k-values obtained from a comprehensive review by Bielski et al., 1985). T under these conditions was 9.8 and 109.0 s at pH 7.00 and 8.25, respectively. Given the apparent stability of superoxide in the experiments (Figure 5) and the truism that dismutation sets the minimum rate of superoxide decay in aqueous systems we concluded that superoxide was continually replenished by the oxidation of Fe(II) in both cases; i.e., the kinetically facile model was correct. This meant the instantaneous concentration of Fe(II) at any point was a function of the relative rates of parallel Fe(III) reduction by reduced sulfur species (including HS-, polysulfides, bisulfite etc.) and the parallel oxidation of Fe(II) by dioxygen, superoxide, and hydrogen peroxide (Warneck and Ziajka, 1995; Lichtschlag et al., 2013; Lohmayer et al., 2014; Wan et al., 2014; Cunha et al., 2016). Since only a few of the required bimolecular rate constants are known for these reactions and the concentration of the secondary sulfide oxidation products was not measured it was not possible to convert the instantaneous Fe(II) concentration to a direct estimate of the ROS formation potential of the system.

The stoichiometry of the conversion of reductive equivalents from HS⁻ to hydrogen peroxide was investigated by comparing the outcome of integrated superoxide production vs. instantaneous hydrogen peroxide. The production of superoxide between any two time points (time a and b) was estimated by calculating the loss from time a to time b by dismutation and adding the difference between the new concentration and that observed (Equation 11): (11)O2produced net−=([O2−]b−[O2−]a)                               +   ([O2−]a−1kt+1[O2−]a) where k = the conditional pH dependent dismutation value, t = the elapsed time between times a and b (0.5 s in this study) and assuming measurement times were close enough so not all superoxide was consumed between times a and b. The latter assumption was supported by the half-life calculations (Equation 10, vide supra) that demonstrated superoxide had a half life over an order of magnitude longer than the sampling interval for the superoxide technique. The approach allowed the overall estimation of the total moles of superoxide produced during the experiment. The accumulated superoxide was summed over the duration of the experiment (7200 s) and plotted against time to obtain a nearly linear estimated net increase in superoxide produced (r² = 0.998 and 0.991 for pH 7.00 and 8.25 respectively, Figure 6). Given that superoxide production came at the cost of Fe(II) oxidation, the negatives of the slopes in Figure 6 were the rates of Fe(II) consumption at the two pHs and the reaction was zero order in Fe(II). This was consistent with the model of ferric and ferrous iron playing the role of a kinetically saturated catalyst for HS⁻ oxidation. The net loss of Fe(II) in Figure 4 indicated that Fe(II)/Fe(III) cycling was not perfectly efficient and is speculatively a result of Fe(III) removal through precipitation. The stoichiometry of the process was investigated by assuming all superoxide produced was consumed by dismutation and comparing the measured hydrogen peroxide yield to that predicted from dismutation (Figures 7, 8 for pH 7.00 and 8.25, respectively). At pH 7.00 superoxide was consumed very rapidly with little hydrogen peroxide production and at pH 8.25 superoxide was converted nearly quantitatively to hydrogen peroxide. The insert plot of hydrogen peroxide predicted vs. measured in Figure 8 is notable for a slope of nearly 1. However, this model was inadequate for predicting hydrogen peroxide yield at pH 7 (Figure 7 and insert). The overall yields of hydrogen peroxide based on HS⁻ consumed were <0.3% at pH 7.00 and ~1.6% at pH 8.25. Presumably this was a result of a pH dependent reaction (or reactions) that consumed either superoxide or hydrogen peroxide more effectively at the lower pH, such as the reaction of hydrogen peroxide with bisulfite or Fe(II), or more likely their reaction with an FeS species (Rush and Bielski, 1985; Warneck and Ziajka, 1995; Chirita and Schlegel, 2012; Giel et al., 2013; Theil et al., 2013; Duinea et al., 2016). It was also possible that at the lower pH some hydrogen peroxide or superoxide was consumed by reactions involving natural carbon in the sediments or through the Fenton reaction (Bielski, 1978; Bielski et al., 1985; Rush and Bielski, 1985; Deguillaume et al., 2005).

The effects of sequential reductant and oxidant additions were measured in marsh sediments at pH 8.25. Additional hydrogen sulfide or hydrogen peroxide additions occurred at time = 1800, 3000, and 4500 s after the initiating hydrogen sulfide pulse, with each addition sufficient to bring the system to a nominal concentration of 300 × 10⁻⁶ M HS⁻ or experience a net increase of 10 × 10⁻⁶ M hydrogen peroxide. Each additional HS⁻ spike oxidized rapidly and the nominal concentrations were only directly observed in sediment-free controls. In sediment suspensions measured sulfide fell typically by 95% within the first 30 s after addition. The concentration of Fe(II) roughly followed the time profile of HS⁻ over multiple additions, indicating the sediments sustained their ability to oxidize sulfide with very short reoxidation times (Figure 9). The ROS response of this system was also monitored. The sequential addition of HS⁻ spikes to these samples resulted in an apparent decrease in superoxide immediately after each addition, however hydrogen peroxide tended to increase in concentration after the pulse while superoxide fell or plateaued (Figure 10). Hydrogen peroxide did not rise to higher levels than previously observed, indicating that consumption was occurring simultaneously with production. Regardless of brief changes in the relative slope of the time profile of superoxide or hydrogen peroxide, the introduction of multiple HS⁻ pulses reduced the apparent plateau concentrations of superoxide by ~25% and hydrogen peroxide by ~50%. It is possible these reductions were an outcome of the accumulation of partially oxidized S species in the system such as S₈, which coat sediment surfaces and inhibit their ability to act as catalysts or directly scavenge oxidants (Rickard et al., 1995; Rickard and Luther, 2007).

In contrast added hydrogen peroxide pulses did not have a statistically significant effect on measured HS⁻ or Fe(II) under these conditions (Figure 11). However, superoxide appeared to experience a (50–100) × 10⁻⁹ M increase after each addition (Figure 12). The most interesting result from this experiment was the changing slope of hydrogen peroxide post-spike; it was evident that the initial spike was consumed rapidly while the latter two appeared to demonstrate more of a step function-like increase. Presumably this was due to consumption of oxidizable sulfur species, at least on the time scale of this study, so that later additions were more stable.

This work demonstrated the potential for the global sulfide reoxidation flux to participate in ROS production in parallel to more recognized photoproduction of ROS. Specifically addition of hydrogen sulfide to oxic muds resulted in the rapid production of Fe(II) species, superoxide and hydrogen peroxide. The choice of hydrogen sulfide re-oxidation as an initial strategy to parameterize non-photochemical ROS production was justifiable because the speciation and redox characteristics of the sulfur cycle are well-established. It was particularly important that sulfate is very soluble in seawater and that there are no naturally occurring, known abiotic mechanisms for reducing sulfate to hydrogen sulfide in oxygenated solutions. In contrast, although the Fe cycle is well-studied, Fe(III) can be readily directly reduced by organic carbon or some ROS in the water column and its solubility is limited. These factors made it much more difficult to analyze reported Fe fluxes and extrapolate the number of reductive equivalents potentially transferred to O₂.

The production of ROS was not quantitative and between 50 and 5% of the sulfide consumed appeared to contribute to ROS production in this system (based on estimated superoxide production and measured peroxide). These estimates are based on the assumption hydrogen peroxide was conservative in this system on the timescale of the experiment. Based on the sulfide re-oxidation budget posited in the introduction this suggests global HS⁻ derived superoxide input on the order of 4.5–2.25 × 10¹² moles/yr near the sediment/water column interface. In environments where the oxic/anoxic mixing zone is very near the sediment surface this implies an ROS production intensity comparable to photoderived ROS production in surface waters. Hydrous ferric oxides played the most significant role in promoting ROS formation over short time scales. The time scale of the experiments shown corresponds to previously measured efflux of the anoxic portion tidal prism through estuarine muds during the falling tide, suggesting that hydrous ferric oxides will be important sources of ROS in those ecosystems. They are also likely to be influential for ROS production in other episodic events, such as bioturbation, storm-driven agitation, dredging etc. It is notable that elevated levels of antioxidant enzymes are frequently observed in biota at environmental compartments that fall in this category, including hydrothermal vents (Bebianno et al., 2005; Geszvain et al., 2012; Genard et al., 2013), cold seeps (Bernhard and Bowser, 2008), and the surface sediments of many coastal salt marshes (Abele et al., 1998a,b; Company et al., 2006). These observations range from single-celled (planktonic) to complex multicellular organisms (limpets, worms) and suggest ROS may have an unexpected ecological importance even in niches that are reliably aphotic because of depth.

SAM, SM, and JF conceived and designed the experiments; SAM, SM, DD, and BS performed the experiments; SAM, SM, and JF analyzed the data; SAM, TS, and JF co-wrote the paper. All authors reviewed the document prior to submission.