Microbial and Reactive Transport Modeling Evidence for Hyporheic Flux-Driven Cryptic Sulfur Cycling and Anaerobic Methane Oxidation in a Sulfate-Impacted Wetland-Stream System

Abstract

This study reexamines the common expectations that in freshwater systems, sulfur plays a minor role in carbon cycling, and aerobic processes dominate methane oxidation. In anoxic sediments of a sulfate-impacted wetland-stream system in Minnesota (USA), a reactive transport model calibrated to geochemical observations predicted sulfate reduction to be the major terminal electron accepting process, and it showed that anaerobic oxidation of methane predominantly coupled with sulfate reduction attenuated methane concentrations near the sediment-water interface. Consistent with model results, 16S rRNA microbiome analysis revealed a high relative abundance of taxa capable of dissimilatory sulfate reduction. It further supported the conclusion that high simulated sulfate reduction rates could be maintained by a “cryptic” sulfur cycle coupled to iron and methane. Low relative abundance of known iron reducing bacteria raised the possibility of abiotic ferric iron (Fe) reduction driving sulfide reoxidation to intermediate-valence sulfur forms; widespread potential for microbially mediated disproportionation, oxidation, and reduction of sulfur intermediates provided mechanisms for completing redox cycles; and archaea comprising up to 25% of the microbial community could include consortia capable of anaerobic oxidation of methane. These biogeochemical processes were found to be controlled by hyporheic fluxes. Lower-magnitude fluxes in wetland compared to channel sediments created sharper geochemical gradients that generated greater heterogeneity in microbial distributions and reaction rates. Changes in upward flux caused fluctuations in sulfate concentrations that led to alternating simulations of methane production and transport. Our work supports the importance of hyporheic flux-driven iron-sulfur-methane cycling in sulfate-impacted wetlands and prompts further investigations under freshwater conditions.

Key Points

• Reactive transport model and microbiome analysis quantify and describe hyporheic flux-driven Fe-S-CH processes in sulfate-impacted wetlands
• Cryptic S cycling, coupled to Fe reduction and anaerobic CH oxidation and involving S intermediates, may drive high sulfate reduction rates
• Dynamic hyporheic fluxes control spatial distributions of microbial compositions and spatiotemporal distributions of reaction rates

1 Introduction

Although wetlands cover only 5–8% of the Earth's land surface, they are the single greatest natural source of the potent greenhouse gas methane to the atmosphere (Meng et al., 2012) while also acting as a sink for atmospheric carbon dioxide and storing the largest reservoir of global organic carbon in their sediments (Köchy et al., 2015). This makes their potential to contribute either a positive or negative feedback to global climate change a critical issue—but one that remains unresolved (Kirschke et al., 2013; Meng et al., 2016). Two complications include uncertainties about the influence of “cryptic” sulfur (S) cycling and anaerobic oxidation of methane (AOM) on wetland carbon budgets, especially under dynamic hydrologic conditions.

The role of S and AOM have generally been overlooked in freshwater environments because of the typically low sulfate (SO ) concentrations there compared to in marine settings (Segers, 1998), where AOM occurs ubiquitously coupled to SO reduction (Hansen et al., 1998; Hoehler et al., 1994; Milucka et al., 2012; Reeburgh, 1980). Low freshwater SO concentrations have led to expectations that dissimilatory sulfate reduction (SR) should be inconsequential in anoxic sediments, particularly in the presence of more thermodynamically favorable terminal electron acceptors such as iron and nitrate. However, this paradigm has been changing, as “cryptic” sulfur cycling has been discovered to sustain high SR rates (Berg et al., 2019; Elsgaard & Jørgensen, 1992; Hansel et al., 2015; Holmkvist et al., 2011; Jorgensen, 1990; Mills et al., 2016), which has implications for carbon mineralization rates. Similarly, studies on AOM prompt further investigation into its role in freshwater methane (CH ) consumption, which can amount to 1–90% of gross production (Segers, 1998). Although nearly all freshwater CH consumption has generally been assumed to occur aerobically at the oxic-anoxic interface or near the rhizosphere, AOM has been found to occur in anoxic wetland sediments coupled to the reduction of more thermodynamically favorable electron acceptors, including SO (Beal et al., 2009; Gupta et al., 2013; Haroon et al., 2013; Riedinger et al., 2014; Segarra et al., 2013; Zehnder & Brock, 1980).

Complementing commonly recognized dissimilatory SO reduction (Pester, 2012), various pathways proposed for “cryptic” sulfur redox cycling include reoxidation of sulfide (HS ) to intermediate-valence sulfur forms (e.g., S and thiosulfate) through coupling with abiotic ferric iron (Fe(III)) reduction (Flynn et al., 2014; Hansel et al., 2015; Holmkvist et al., 2011; Lohmayer et al., 2014; Pester, 2012), simultaneous reduction and oxidation of S intermediates by disproportionating bacteria to HS and SO (Elsgaard & Jørgensen, 1992; Jorgensen, 1990; Milucka et al., 2012), microbially mediated reduction of S intermediates (Flynn et al., 2014; Lohmayer et al., 2014), and reoxidation of sulfite to SO (Antler et al., 2013; Brunner & Bernasconi, 2005; Mills et al., 2016) (many of these processes are depicted in Figure 1). These processes can potentially support fast S redox cycling even when there are only small net changes in SO and HS concentrations, and hence, they are often referred to as “cryptic.” In addition to having significant implications for soil carbon degradation, recent studies have identified SR as the likely dominant terminal electron accepting process where freshwater AOM occurs (Schubert et al., 2011; Segarra et al., 2015; Timmers et al., 2016; Weber et al., 2017). In fact, in an experiment with freshwater sediments, “cryptic” S cycling coupled to Fe reduction appeared to provide a hidden source of SO to drive AOM (Weber et al., 2016) (as depicted in Figure 1), consistent with previous findings in marine environments of coupled interactions of AOM and “cryptic” S cycling through S intermediates (Milucka et al., 2012).

Figure 1

The conceptual model at Second Creek comprises a dominant sulfur cycle that likely includes “cryptic” reactions involving sulfide reoxidation through sulfur intermediates and disproportionating reactions. This sulfur cycle is linked to the iron cycle, not only through precipitation of FeS, but possibly also through abiotic Fe reduction coupled to sulfide reoxidation. The sulfur cycle is also possibly linked to methane through coupled sulfate reduction and anoxic methane oxidation. The rate of coupled Fe-S-CH cycling depends on hyporheic flux, which controls the concentrations of sulfate entering the sediments either from the overlying surface water or from deeper groundwater through dispersion and advection.

Although the potential importance of AOM in anoxic wetlands sediments has been noted across latitudes, climates, and hydrologic settings (Gauthier et al., 2015; Gupta et al., 2013; Segarra et al., 2015; Smemo & Yavitt, 2007), uncertainties in rates and mechanisms persist. Laboratory experiments showed freshwater AOM rates ranging from a mere 0.5% (Blazewicz et al., 2012) to at least 10%, and possibly as great as 100%, of aerobic methane oxidation rates (Gauthier et al., 2015; Martinez-Cruz et al., 2018), and in situ measurements have shown AOM rates to be equivalent to those in marine settings (Segarra et al., 2015). Another unresolved issue is the relative importance of AOM for lowering CH emissions from wetlands (Segarra et al., 2015) compared to substrate competition between sulfate-reducing bacteria and methanogens (Blazewicz et al., 2012; Gauthier et al., 2015; Smemo & Yavitt, 2007). Unlike substrate competition among the microbial community, AOM has the capacity to consume CH during its transport through sediments. Therefore, clarifying AOM's importance and driving mechanisms, particularly coupled with SO and “cryptic” S cycling, is a priority in order to improve our ability to predict C budgets. In hydrologically dynamic wetland systems, this understanding must critically be extended to evaluate how “cryptic” S cycling and its interactions with CH respond to varying hyporheic flux.

In this study, we hypothesize that currently underconstrained “cryptic” S cycling and AOM serve as major biogeochemical processes in wetlands and, further, that the variable rates and mechanisms observed in past studies can be attributed to dynamic conditions of hyporheic zones. Hyporheic zone mixing of surface water and groundwater creates steep redox gradients and promotes hot spots and hot moments of fluctuating microbial activity (Boano et al., 2014; Briggs et al., 2015; Danczak et al., 2016; Feris et al., 2003; Krause et al., 2013; McClain et al., 2003; Vidon et al., 2010; Zarnetske et al., 2012), but little is known about how these dynamic fluxes impact “cryptic” S cycling and AOM in wetlands. We evaluated our hypotheses using a combination of hydrologic and geochemical field observations, reactive transport modeling, and microbiome analysis at a SO -impacted wetland-stream system in northern Minnesota (USA). Not only do SO -impacted environments have the most urgent need for clarifying the impact of S on overall wetland biogeochemistry, but they also provide the clearest test bed for examining S cycling in freshwater sediments. Previous reactive transport models have included AOM coupled to SO reduction for marine environments (see review by Regnier et al., 2011, and He et al., 2019), but similar applications in freshwater ecosystems are minimal. Additionally, uncertainties about specific mechanisms and their rates have limited the incorporation of “cryptic” S cycling in any model applications. We implemented a reactive transport model with an implicit representation of AOM coupled to S cycling based on hydrologic and geochemical observations, and we further constrained our model interpretation based on mechanistic insights provided by microbial analyses. With this two-pronged approach, our work rigorously assesses in situ abiotic-biotic S-Fe-CH processes in a hydrologically dynamic wetland-stream system.

2 Study Site

Our study site “Second Creek” is a first-order SO -impacted wetland-stream system in northeastern Minnesota located at the southern extent of the North American boreal zone (Figure 2). Biogeochemical cycling is of particular concern in boreal wetlands, because they store 20–50% of global organic carbon (Hugelius et al., 2013; Meng et al., 2016; Mitsch et al., 2013), and future climate change has the potential to promote increased organic matter degradation and methane emission rates in them (Yvon-Durocher et al., 2011; Hodgkins et al., 2014). Second Creek consists of a narrow stream channel 2–3 m wide that is flanked by wetlands 20–30 m wide on either side. The site was included in a geochemical field survey of Minnesota wild rice lakes and streams investigating the impact of surface water sulfate on wild rice (Zizania palustris), a culturally, ecologically, and economically important species in the Great Lakes region (Myrbo et al., 2017). Water pumped from mine waste rock pits (man-made lakes hosted in bedrock) upgradient to the north and northeast of the site serves as a point source of elevated SO concentrations to the surface water.

Figure 2

The Second Creek study site, located in northeast Minnesota (upper inset). Geochemical sampling (porewater and sediments) and hyporheic flux monitoring (with stream gauges, piezometers, and temperature probes) took place in the channel and in the wetland on the west flanking side, as indicated by the symbols. The channel and wetland sites were located about 2 m apart. Lower inset: view upstream. This figure is based on Figure 1 in Ng et al. (2017). The “Wetland” sampling location in this study corresponds to the “West Wetland” in Ng et al. (2017).

Despite successful efforts to curb sulfur pollution, global SO emissions are predicted to increase due to industrial activities in developing countries (Smith et al., 2005; Ward, 2009), making directly relevant the high sulfate conditions found at Second Creek. Further, recent studies highlighting the prominence of sulfur reactions in low-concentration settings (Hansel et al., 2015; Pester, 2012; Segarra et al., 2015) demonstrate even more widespread need for the hydrobiogeochemical process understanding that can be gained at this study site.

Previous work by Ng et al. (2017) at Second Creek showed that surface water SO concentrations reach 10 mM under typical hydrologic conditions, and a lower but still moderate concentration of SO in groundwater (average of 0.3 mM) flows from the underlying aquifer consisting of glacial outwash and till to the streambed. Streambed sediments at Second Creek include up to 31 mg/g Fe based on HCl extractions (Myrbo et al., 2017), which likely in part consist of solid ferric oxyhydroxides that reductively dissolve to generate observed porewater Fe concentrations of up to 1.5 mM (Ng et al., 2017). Sediment Fe also likely includes precipitated FeS formed from reduced Fe and sulfide, as supported by acid volatile sulfide extractions of up to 7 mg/g (Myrbo et al., 2017; Ng et al., 2017) and low concentrations of dissolved metals other than Fe (Myrbo et al., 2017).

Using a reactive transport model calibrated to geochemical data collected over two distinct hydrologic regimes in the summer of 2015, Ng et al. (2017) determined high SR rates that exceeded Fe reduction rates, contrary to the expected thermodynamic ordering. Also, during an unusual, prolonged 2-month flood in the summer of 2015, high SO concentration surface water penetrated deeper than usual below the sediment-water interface and coincided with attenuated porewater concentrations of dissolved CH . In the current study, we use a similar modeling approach but augment it with analyses of microbial community distribution and potential metabolic function assessments to evaluate whether “cryptic” S cycling could be driving high SR rates as well as coupled AOM at Second Creek. Further, we assess whether these processes could be occurring under more typical unflooded stream conditions in the summer of 2016, when groundwater advects up through wetland and stream sediments over the entire season.

3 Methods

3.1 Hydrology

We determined daily hyporheic flux from 1 June to 30 September 2016 using measurements from temperature probes, piezometers, and a stream gauge. A Schlumberger Micro-Diver (DI601) pressure transducer was deployed in a manually installed stream gauge to record surface water head, and Schlumberger Baro (DI500) pressure transducers were deployed in three manually installed shallow piezometers to record subsurface head levels at 35-cm depth; calculations accounted for the contribution of air pressure recorded by an additional Schlumberger Baro sensor. Two of the piezometers were deployed at the main monitoring and sampling sites in this study (Figure 2): in the center of the stream channel and in the west-flanking wetland (sufficiently close to the channel such that inundated conditions persisted throughout the summer). To assess variability within Second Creek, the third piezometer was deployed between the first two locations in the channel near its west bank. Vertical temperature probes collocated with the piezometers measured porewater temperature at 0, 5, 10, 15, 20, and 30 cm depths below the sediment-water interface. The temperature probes were constructed using wooden dowels, housed thermistors potted at the dowel surface in waterproof epoxy, and were connected to an open-source “ALog” data logger (an intermediary version between that presented in Wickert, 2014, and Wickert et al., 2018).

Time series data of temperature profile and hyporheic zone head gradients in the center channel, west wetland, and west channel were incorporated in the heat-transport inverse model 1DTempPro (Voytek et al., 2014) to estimate the time series of hyporheic flux at each location. For parameters required by 1DTempPro model, we used sediment porosity of 0.51 previously measured by Myrbo et al. (2017), thermal conductivity of 0.56 W/m/°C, saturated heat capacity of  J/m /°C estimated assuming 80% sediment organic matter and 20% siliciclastic material in the approach by Farouki (1986), and heat dispersivity of 0.1 m assigned based on typical solute dispersivity for the tens of centimeters spatial scale considered here (Zheng & Bennett, 2002). Sensitivity tests with a range of parameter values showed this choice of dispersivity and saturated heat capacity produced good fits with observed temperatures at all locations, while results appeared relatively insensitive to the saturated heat capacity value (details in supporting information section S1).

3.2 Aqueous Chemistry

On 14 June and 15 August 2016, we sampled porewater at 1.56-cm depth intervals from 1.56 cm above the sediment-water interface to about 40- to 50-cm depth below the interface using passive porewater equilibrators (“peepers”) following the method described in Ng et al. (2017). In June, duplicate peepers were deployed in the channel and wetland monitoring/sampling locations (Figure 2). In August, duplicate peepers were deployed again in the channel; intended peeper deployment in the wetland in August was later found to be mislocated in channel sediments and failed to capture actual wetland conditions. For all peeper samples, pH was analyzed in the field using a Thermo Scientific Orion STAR A329. Fe was quantified in the field using the phenanthroline method (Eaton et al., 2005). Methane samples were collected and analyzed using the same methods described in Ng et al. (2017). Porewater sulfide concentrations (including all reduced and dissolved inorganic sulfide) at Second Creek are generally low (<2  M) due to precipitation of FeS (Ng et al., 2017). For select peeper samples, we acidified cation samples of about 12-ml volume with one drop of 6N HCl before analysis using a Thermo Scientific iCAP 6500 dual view ICP-OES (for Al, Ba, Ca, Fe, K, Li, Mg, Mn, Na, P, Si, and Sr) at University of Minnesota, Twin Cities; anion samples were analyzed using a Thermo Dionex ICS 5000+ ion chromatography system (for F , Cl ,NO ,Br , and SO ) at University of Minnesota, Twin Cities.

3.3 Sediment Microbiology

On 14 June and 15 August 2016, duplicate sediment cores were collected at the channel and wetland monitoring/sampling locations (Figure 2) using an HTH sediment gravity corer (Pylonex, Sweden) with 7-cm-diameter polycarbonate tubing, as described in Ng et al. (2017). We aimed to collect cores close to the corresponding peeper deployments, but spatial heterogeneities should be considered. In particular, although August wetland peepers were likely mislocated in channel sediments (see above), August wetland sediment cores appeared to be correctly located in wetland sediments.

From each core, triplicate sediment samples were collected at various depths from 0 cm to approximately 30 cm below the sediment-water interface, at intervals ranging from 2 to 11 cm, to capture the vertical heterogeneity within the microbial community. Once collected, sediment samples were immediately flash-frozen in a dry ice-ethanol bath in the field and stored on dry ice during transportation to the laboratory, where they were stored at 80 °C until processing. In the laboratory, DNA was extracted from each sample using the FastDNA Spin kit for Soil (MP Biomedicals) with the following modifications. Polyadenosine (Poly (A); 200  g per sample) was added to the lysis buffer to reduce inhibition by metal cations (Webster et al., 2003). Two homogenization steps on the FastPrep instrument (MP Biomedicals) were carried out with a 5-min incubation on ice in between. The initial centrifugation step to remove sediment and cell debris was extended to 15 min, and the binding matrix incubation was extended to 10 min. Elution was carried out by resuspending the binding matrix in 100- l nuclease-free sterile water and incubating at 55 °C. Extracts were quantified using the Qubit double-stranded DNA BR assay kit (Life Technologies) with a Qubit 2.0 fluorometer, and those exceeding 100 ng/ l were diluted to that concentration. Extracts from triplicate samples were then pooled and used in DNA amplification and sequenced at the University of Minnesota Genomics Center using a dual-indexing approach modified from the Earth Microbiome Project protocols (Gohl et al., 2016). Negative extraction kit controls were also extracted in triplicate using identical procedures to assess any microbial taxa contained within the kits themselves. To assess the bacterial and archaeal communities, the 16S rRNA gene region was amplified using the V4 hypervariable region primers 505F and 806R (Gohl et al., 2016). After amplification, DNA samples were then diluted (1:10 or 1:100) to reduce amplification inhibition and subjected to paired-end sequencing with a MiSeq 600 Cycle v3 kit (Illumina, San Diego, CA).

To assess taxonomic makeup of the bacterial and archaeal communities, raw DNA sequences were processed using the mothur software package (v 1.39.5) (Kozich et al., 2013a) according to the protocols described in Kozich et al. (2013b). This entailed merging paired-end reads into contigs, quality screening the contigs, aligning sequences to the SILVA 16S rRNA sequence database, and screening and removing chimeras. Bacterial and archaeal sequences were further cleaned by classifying against the Silva reference database in mothur using the method developed by Wang et al. (2007). Processed sequences were then clustered into operational taxonomic units (OTUs) using a 97% sequence similarity cutoff (OTU_0.03) using the OptiClust algorithm in mothur (Westcott & Schloss, 2017) and assigned consensus taxonomy. The number of sequences of each OTU present in the negative controls (polymerase chain reaction and DNA extraction kits) was manually subtracted from the sequence abundance of that OTU in the experimental samples using R Version 3.4.2 (R Core Team, 2015) to account for contamination and to limit noise associated with the sequence data sets (Nguyen et al., 2015).

16S rRNA gene sequences were also utilized to infer metabolic pathways using the paprica v.0.4.0 metabolic inference pipeline (Bowman & Ducklow, 2015). Paired end reads were aligned to the Silva database using mothur, as described above, and parsed into separate bacterial and archaeal files. All reads shorter than 100 bases or that did not align were discarded. Quality-filtered reads were subsampled to the size of the smallest library for each data set (1,094 reads for archaea and 24,354 reads for bacteria), and metabolic structure was determined with paprica. Paprica uses the pplacer phylogenetic placement software to place 16S rRNA gene reads on reference trees created from all completed bacterial and archaeal genomes (Matsen et al., 2010); the reference genome database was compiled on August 2018. Pathways present in negative controls were manually subtracted from the experimental samples using R. Statistical analyses and plotting of paprica results were also performed using R.

3.4 Reactive Transport Model

Multicomponent reactive transport models integrate aqueous and mineral phase processes in a physically dynamic environment (Hunter et al., 1998; Li et al., 2017; Mayer et al., 2002; Steefel et al., 2005) and are effective tools for examining redox processes in hyporheic zone settings (Arora et al., 2016; Bardini et al., 2012; Boano et al., 2014; Engelhardt et al., 2014; Lautz & Siegel, 2006; Liu et al., 2017; Runkel & Kimball, 2002; Zarnetske et al., 2012;). We used the reactive transport model code PHT3D (Prommer et al., 2003), which combines the numerically robust transport solvers from MT3DMS (Zheng & Wang, 1999) and the comprehensive geochemical reaction algorithms from PHREEQC-2 (Parkhurst & Appelo, 1999); the MT3DMS solvers use groundwater flow inputs generated by MODFLOW (Harbaugh, 2005). Our model implementation was based on the formulation presented in Ng et al. (2017); here we summarize the main features and describe differences in the current work.

For the model domain, vertical 1-D profiles comprising 1.5-cm-thick layers, starting at the sediment-water interface, represent the hyporheic zone of the channel and wetland. The full computational domain extends to 3-m depth to be able to set deeper groundwater boundary conditions, although other model constraints include observations within the top 50 cm. While a 1-D domain cannot capture the full spatial heterogeneity or potential complex flow pathways at Second Creek, our implementation of a wetland and channel profile provides a representative range of variability, and our focus on the vertical component likely captures the sharpest gradients due the distinct surface water and groundwater chemistries.

The model simulated saturated groundwater flow, which is appropriate for the fully inundated conditions recorded by our surface water gauges over the monitoring period. Solute transport included advection based on simulated groundwater flow, and dispersion using a dispersivity of 0.01 m and tortuosity of 0.5. Though bioirrigation and sediment mixing potentially influence solute transport in sediment systems, to our knowledge, no previous studies have investigated the impacts of these processes at our site, which provides little information on which to base their representation in the model. The aquatic vegetation at the site also has the potential for influencing geochemical conditions through their exudation of both oxygen and organic substrates (Armstrong, 1980; Bais et al., 2004; Colmer, 2003; Grayston et al., 1997). However, these effects were also not included in the model. Though roots may extend >10-cm depth in the sediment profile, oxygen and root exudates only extend about 3–6 mm away from the root surface (Jensen et al., 2005) and therefore do not have a large impact at greater distances from root surfaces. Further, we specifically located cores away from plant roots to minimize rhizosphere impacts on our measurements.

Simulated aqueous geochemical components include total concentrations of inorganic C, dissolved CH , dissolved oxygen (DO), Al, Ba, Ca, Cl, Fe, S, K, Mg, Mn, Na, P, S, Si, and pH. Different simulated terminal electron accepting processes (TEAPs) coupled to organic carbon oxidation include aerobic respiration, reductive dissolution of ferric oxyhydroxide represented by the model mineral phase Fe(OH) , SO reduction, and methanogenesis. Nitrate reduction was not included due to the low measured porewater concentrations (<0.1 mM). The model represents coupled redox reactions using a partial equilibrium approach (Jakobsen & Postma, 1999), in which the oxidation of sediment organic carbon, stoichiometrically represented as CH O, is assumed to be the rate-limiting step. The different terminal electron accepting reactions are assumed to be fast in comparison to the oxidation half step and immediately equilibrate in response to electrons being released by organic carbon oxidation. We represented CH O decay using a calibrated first-order kinetic parameterization (Table 1), which is automatically followed by secondary oxidation reactions through the partial equilibrium approach. TEAP ordering and redox zonation are dictated by the relative equilibrium parameters for the different reduction reactions. The standard PHT3D/PHREEQC-2 geochemical database provided equilibrium constants for aqueous reactions, including for sulfur and carbon/methane redox reactions, while equilibrium parameters for Fe mineral dissolution were calibrated, as discussed further below. We found that the organic-rich sediments did not significantly deplete in organic carbon over the summer simulation, and thus, we did not attempt to include a replenishment process.

Table 1. Parameter Values Used in the Calibrated Model Simulations

	Channel		Wetland
Parameter	Calibrated	Observed	Calibrated	Observed	Approach
Average hyporheic flux, positive up (mm/day)	13.5	20	9.5	12	Calibrated to 2016 geochemical observations
Sediment organic carbon concentration (mol L )	0.34	N/A	0.34	N/A	Calibrated to 2016 geochemical observations
pre-June sediment organic carbon lower depth limit (cm)	45	N/A	65	N/A	Calibrated to 2016 geochemical observations
June–Sept sediment organic carbon lower depth limit (cm)	65	N/A	80	N/A	Calibrated to 2016
geochemical observations
top-of-domain sulfate concentration (mM)	3	3.0 to 7.1	7	3.5 to 5.8	Calibrated to 2016 geochemical observations
Aerobic 1st order decay constant (s )	1.00	N/A	1.00	N/A	Calibrated in Ng et al. (2017)
Anaerobic 1st order decay constant (s )	2.61	N/A	2.61	N/A	Calibrated in Ng et al. (2017)
Fe(OH) equilibrium dissolution logK	2.119	N/A	2.119	N/A	Calibrated in Ng et al. (2017)
FeS equilibrium dissolution logK	2.349	N/A	2.349	N/A	Calibrated in Ng et al. (2017)
Cation exchange capacity (mol Lw )	0.005	N/A	0.005	N/A	Calibrated in Ng et al. (2017)

• Note. Dynamic properties such as organic carbon, flux, and sulfate boundary conditions were calibrated to 2016 geochemical observations. Other model inputs (sediment and organic carbon decay parameters) are from Ng et al. (2017), calibrated to 2015 geochemical observations.
• ⁽¹⁾L_w refers to liters of water.
• ⁽²⁾ Organic carbon concentrations were not measured in summer 2016. Summer 2015 measurements of total carbon, assumed to approximate organic carbon, ranged over 0.003 to 0.27 mol L in the channel and 0.3 to 0.54 mol L in the wetland.
• ⁽³⁾ First-order decay of sediment organic carbon concentration, represented stoichiometrically as CH O, is parameterized as (where brackets represent concentration).
• ⁽⁴⁾ Dissolution reaction: .
• ⁽⁵⁾ Dissolution reaction: .

Using partial equilibrium, the model does not explicitly represent AOM coupled to SR or “cryptic” sulfur cycling reactions, such as Fe reduction coupled to HS reoxidation, or disproportionation of S intermediates. However, we aim to implicitly represent the major drivers for cycling among these redox pairs through the use of calibrated partial equilibrium reaction parameters. Though lacking in mechanistic details, this representation was chosen over a detailed kinetic implementation (e.g., using a Monod formulation) for a number of reasons. First, specific reaction networks and associated kinetic parameters have not yet been constrained at Second Creek. This prompted DNA-based analysis in this work, but as explained further in section 4.3, these only provide information about potential (not actual) activity, and they presented various possible S cycling networks through different intermediate-valence sulfur forms. Testing the different pathways with the model was made difficult by the uncertainties in the kinetic parameterizations for each individual reaction. The kinetic properties of reactions in “cryptic” sulfur cycling and AOM are not well known, which means that without measurements on individual kinetic pathways, the model is highly underconstrained (each reaction requires at least two parameters—maximum specific growth rate, half-velocity constant, and possible inhibition terms). Lastly, Monod kinetic formulations have their drawbacks, including their potential for producing unrealistic redox zonation and pH simulations without the additional application of thermodynamic limitations (Curtis, 2003; Jin & Bethke, 2009). Given the lack of reaction-specific data for our study, such level of detail in the model would be difficult to support.

In light of the challenges of implementing a full kinetic model at this time, we opted to build off the partial equilibrium model from Ng et al. (2017), which used a much smaller number of calibrated reaction parameters (compared to a Monod approach) that were determined through inverse modeling with in situ geochemical conditions. Use of an equilibrium instead of a kinetic implementation for the reduction half-reactions is a simplification that we consider to be reasonable in our study for representing net SR rates at depth, which we propose are primarily driven by cryptic sulfur cycling and transport processes rather than competition between different TEAPs. The lack of a fully kinetic representation likely leads to simulation errors for other redox reactions, such as coupled Fe reduction-AOM rates, but these are peripheral to our main focus on sulfur. The calibrated reaction parameters in our partial equilibrium model included aerobic and anaerobic first-order decay constants for organic carbon (relative to organic carbon) and equilibrium dissolution logK values for Fe(OH) (to simulate reductive dissolution) and FeS (to simulate precipitation of reduced Fe and sulfide) (Table 1). Rather than capturing theoretical or true thermodynamic properties, these are effective parameters calibrated to represent the diverse organic matter stoichiometries, the mix of mineralogical compositions, and likely, to some degree, kinetic effects over the calibration period. By evaluating results from a more tractable, calibrated partial equilibrium model, and further comparing them with potential microbial activity based on taxonomic analysis and inferred metabolic functions, this work provides an important step in identifying and quantifying wetland AOM and “cryptic” sulfur processes, which can prompt more detailed measurements and explicit model implementations in future studies.

The 2016 model utilized previously calibrated parameters from summer 2015 (Ng et al., 2017) for properties not expected to substantially change between the two years, including the reaction parameters listed above and the cation exchange capacity used together with divalent cation exchange equilibrium constants taken from a northern Minnesota modeling study by Ng et al. (2015) (Table 1). As in Ng et al. (2017), the model was initialized with Fe(OH) throughout the column. Though we did not attempt to simulate their production, ferric oxides can occur even at depth in largely anoxic profiles due to radial oxygen loss from plant roots and by occasional downward flux of oxygenated water from the surface into the subsurface. New, manually calibrated inputs for the 2016 model included time-averaged hyporheic flux, sediment organic carbon concentration and distribution, and top-of-domain SO concentrations. Hyporheic flux and top-of-domain SO were calibrated rather than were directly taken from measurements because of their uncertainties and large spatiotemporal changes even within a season; instead, temperature probe flux estimates and surface water SO data were indirectly used to guide their respective calibration. Calibration of sediment organic carbon, which was not measured in 2016, included not just adjusting concentration but also a lower depth limit, based on the assumption that most accumulates at the top of the sediment column.

For other model inputs, other major cation and anion top boundary conditions were set to average 2016 observed surface water measurements. Following Ng et al. (2017), the top boundary condition for DO was set to a suboxic level (0.1875 mM) based on reducing conditions observed immediately below the sediment-water interface. Average groundwater concentrations measured upgradient (west) of the wetland stream in summer of 2015 (Ng et al., 2017) were used for bottom boundary conditions at 3-m depth in the 2016 model, because of the unavailability of summer 2016 measurements. Observations in 2015 showed groundwater conditions to be relatively stable over time, and model matches with 2016 observed porewater concentrations indicate that 2015 groundwater data provided reasonable bottom boundary conditions. All boundary conditions are provided in the supporting information (Table S1).

In summer 2016, dissolved CH was measured during all time periods in both the wetland and channel sediments, serving as an important constraint on the redox condition of the system. To ensure proper representation of CH , an update to the summer 2015 model implementation from Ng et al. (2017) included irreversible CH outgassing if the sum of partial pressures (from CH , CO , and N ) reached atmospheric pressure.

The model calibration procedure covered two simulation intervals for summer 2016. To spin up the model leading up to the first sampling time on 14 June 2016, observed aqueous concentrations for major cations and anions on 14 June 2016 were used as initial conditions (see Table S1 in the supporting information) for a 60-day simulation that started with no dissolved CH ; this represented spring conditions, when slower biogeochemical reactions under cooler conditions are less likely to create methanogenic conditions. The sediment organic carbon distribution and average flux for this 60-day spin-up were calibrated by matching the final model results to June 2016 aqueous chemistry observations, with dissolved Fe, SO , and CH serving as the major calibration targets. A second calibration period began with the spun-up June 2016 conditions and ended in 15 August 2016, with the August observations serving as calibration targets to estimate late summer organic carbon distributions for representing the response to warmer temperatures and faster biogeochemical reactions. August wetland geochemical observations were not used in the calibration, however, because these were found to be mislocated in channel sediments (as described earlier). After the calibration, we implemented fully transient simulations spanning 14 June 2016 (the first sampling period) to 30 September (the end of the hydrologic monitoring period). These were initialized with the spun-up porewater conditions in June and used the calibrated organic carbon parameters as well as time-varying daily flux rescaled to match the calibrated average flux.

4 Results and Discussion

4.1 Hyporheic Flux

Hydraulic head gradients measured in the channel center, west wetland, and west channel over the summer of 2016 indicated variable magnitude but a nearly consistent upward flux direction throughout the summer, with the exception of a single downward flux event lasting less than 24 hr during a major rain event in late August. Inverse modeling of temperature profile observations together with head gradients yielded about 1.6 times greater magnitude flux in the channel center than in the west wetland (time average of 20 mm/day vs. 12 mm/day) (Figure 3). Unexpectedly, the west channel results showed the lowest average flux magnitude (4 mm/day); this could have resulted from analysis error due to lateral flux near the stream bank (the estimation method assumes 1-D vertical flux) and/or high spatial heterogeneity typical of the hyporheic zone (Boano et al., 2014). We focused on the channel center and west wetland in this study, which we hereafter refer to as the “channel” and “wetland,” respectively, for naming simplicity. However, the west channel flux results raised caution about the accuracy of flux measurements and justified further calibration of the flux magnitudes in the model. Both the (center) channel and (west) wetland exhibited similar temporal changes (coefficient of variation over time of 18% in the channel and 16% in the wetland) (Figure 3).

Figure 3

Daily hyporheic flux (mm/day) estimated over the summer of 2016 using inverse modeling of vertical porewater temperature profiles and hydraulic gradients measured at three locations. “(Center) Channel” and “(West) Wetland” correspond to the monitoring/sampling sites shown in Figure 2; “Channel West” is situated between the two in the main channel. The black dotted line is the reference for no flux (0 mm/day); positive (negative) values indicate upward (downward) flux. The gap in late July corresponds to a low surface water level time when our first surface water gauge location no longer had standing water. We then moved the surface water gauge toward the channel center for the remainder of the summer. Green dashed vertical lines indicate the geochemical and microbial sampling dates (14 June and 15 August 2016).

For a comparison of these summer 2016 temperature probe estimates, summer 2015 flux observations (Ng et al., 2017) only included measurements from single-day deployments of seepage meters. The 2015 measurements during upward flux times were limited to the channel, and these yielded much lower estimates (0.2 to 0.7 mm/day) than the 2016 temperature probe estimate (average 20 mm/day). This discrepancy could in part be explained by errors in the seepage meter measurements due to installation difficulties and inaccuracies cited by Ng et al. (2017). The ratio of seepage meter measurements may be more reliable if seepage meter uncertainties are consistent across deployments. Seepage meters deployed in both the channel and wetland during a 2-month flood period in 2015 indicated that during that time, the channel flux could be 1.4 to 7 times greater magnitude than the wetland flux, which brackets the 1.6 times greater flux in 2016 estimated in the channel compared to wetland using the temperature probes. Further calibration with the reactive transport model using 2015 geochemical data resulted in a 2015 upward channel flux magnitude of 8.64 mm/day following the flood, which is much higher than the 2015 raw measurements from seepage meters and closer to the average 2016 upward flux estimate of 20 mm/day.

4.2 Simulations of Sulfate, Iron, and Methane Processes

Hyporheic flux time series estimates and geochemical measurements were first used to rigorously calibrate the reactive transport model and produce quantitative simulations of Fe, S, and CH interactions. Subsequent qualitative comparison of model results with microbial analyses in section 4.3 provides insights into the physical and biotic mechanisms driving these processes.

Table 1 shows the key model parameters controlling simulated S, Fe, and CH processes—average hyporheic flux, sediment organic carbon distribution, and top boundary condition sulfate concentration, which were calibrated based on observations of dissolved Fe, SO , and CH in the channel and wetland over the summer of 2016. Supporting information includes details of the calibration procedure (section S2).

4.2.1 Calibrated Simulations of Geochemical Profiles

Duplicate profile measurements displayed a range of spatial trends with depth (symbols in Figure 4), and so we made an effort to calibrate the model to one of the duplicate profiles for consistency, rather than attempt to capture aspects of both. Channel simulations captured the overall increase in dissolved CH from June to August and the relatively consistent Fe and SO profiles observed over the summer, in response to organic carbon oxidation and reductive dissolution of ferric oxide (Figure 4, top). It can be seen that organic carbon concentrations were high and did not appreciably deplete over the summer with the calibrated decay parameters. Wetland simulations also corresponded with observed Fe , SO , and CH concentrations in June—best matches were with the duplicate profiles with lower Fe and CH measurements at 40 to 50 cm depth (Figure 4, bottom), and like in the channel, August wetland simulations showed dissolved CH at greater concentrations and deeper in the profile compared to June. Greatest discrepancies between the model and observations occurred for CH near the sediment-water interface; in the channel, CH in the upper 5 cm was undersimulated (in both June and August), while in the wetland, CH at 5 to 15 cm depth was oversimulated (in June, when simulations were compared against observations). The observed cooccurrence at these depth intervals of SO and CH represents a state out of thermodynamic equilibrium, which could not be simulated with the simple partial equilibrium model implementation. Despite these mismatches, the model captured the overall trends in each of the observed concentration profiles (other than the August wetland profiles, which were likely mislocated in channel sediments, as described in section 3.2 and supporting information section S2). Results for other aqueous and sediment components are shown in Figures S3 and S4 in the supporting information.

Figure 4

Organic carbon (OrgC), ferric oxide (FeOH ), dissolved iron (Fe ), sulfate (SO ), and dissolved methane (CH ) profiles in the channel (top) and wetland (bottom) below the sediment-water interface ( ). Calibrated model simulations (lines) are compared with duplicate peeper observations of Fe , SO , and CH (symbols) on 14 June (black) and 15 August (red) 2016. No observations were available for FeOH and OrgC. For FeOH , simulations are shown for the net change relative to arbitrary initial conditions; results are in molar (M) units to facilitate easier comparison against aqueous phase concentrations. Gray dotted lines show daily model results at all times (14 June to 30 September 2016); gray squares at the top of the profile show all surface water (sw) measurements in June and August in both sampling locations. Note that observed August geochemical wetland profiles were likely mislocated in channel sediments, and as a result, no effort was made to calibrate August wetland simulations to those observations.

4.2.2 Calibrated Simulations of Reaction Mass Balance

4.2.2.1 Channel Results

The calibrated channel model produced reaction mass balance results in Figure 5 (top) indicating that Fe reduction (negative Fe(OH) mass change), SR (negative SO mass change), and methanogenesis (positive CH mass change) all actively occurred in response to organic carbon oxidation (negative OrgC mass change) in the channel sediment profile from June to August 2016. Simulations of distinct, yet overall complementary, spatial distributions of these three anoxic redox processes provided insights into linked Fe-S-CH interactions. Fe reduction occurred most ubiquitously in the top 65 cm of the sediment profile where organic carbon degradation was simulated, because of the availability of ferric oxyhydroxide minerals throughout the upper sediment column. Net SO reduction also occurred in the top 65 cm, but only near the sediment-water interface, where rates are highest, as well as toward the bottom of the 65 cm interval with organic carbon degradation. This can be attributed to the downward diffusion of overlying surface water and upward advection of deeper porewater, both of which transported SO (3 mM concentration at sediment-water interface and 0.3 mM in the deep porewater/groundwater) into the simulated profile domain. It is noteworthy that methanogenesis was simulated only at about 10 to 35 cm depth, where there was no SO reduction. Above this depth, the model showed CH to be consumed (negative CH mass change) in the upper 5 to 8 cm, where there was the greatest Fe reduction and SO reduction.

Figure 5

Simulated reaction rate profiles in the channel (top) and wetland (bottom) below the sediment-water interface ( ). Results are net averages over 14 June to 30 September 2016, with positive values indicating net inputs and negative values indicating net outputs due to redox and/or mineral phase change reactions. Gross reaction input and output results (not shown) indicate that, at any particular depth, there is generally only reaction input or output over the entire 2.5 months and not both. The dotted blue line shows net 0 rate as a reference. Negative organic carbon (OrgC) reaction rate indicates degradation (oxidation); negative sediment ferric oxide (Fe(OH) ) reaction rate indicates reductive iron (oxy)hydroxide dissolution; negative sulfate (SO ) reaction rate indicates SO reduction; positive sediment iron-sulfide (FeS) reaction rate indicates iron-sulfide precipitation; positive dissolved methane (CH ) reaction rate indicates methanogenesis; and negative dissolved methane (CH ) flux indicates methane oxidation.

Despite ongoing Fe and SO reduction, concentrations of both aqueous Fe and SO remained relatively stable in the channel over the summer (compared to CH ) because nearly all of the sulfide produced immediately precipitated with Fe as FeS (Figure 5), which, as noted earlier, had been widely detected in Second Creek sediments (Ng et al., 2017; Myrbo et al., 2017). In fact, a comparison of channel Fe(OH) and SO profiles in Figure 5 suggested that aqueous Fe concentrations were elevated within approximately the top 10 to 30 cm (Figure 4) in part because SO reduction—and hence FeS precipitation—was lowest around there (10 to 30 cm depth), in the gap between downward diffusion and upward advection of SO .

Taken together, in order to match observations, the model simulated net methanogenesis only where it could not be thermodynamically outcompeted by SO reduction (10 to 30 cm depth), and it simulated oxidation of the upward advecting CH where downward diffusing SO becomes reduced and where Fe reduction was greatest (5 to 8 cm depth). This suggested that AOM could be occurring coupled to SO and Fe reduction in the upper profile. Given the slightly higher simulated mass reaction rate for SR compared to Fe reduction, compounded by the much higher electron equivalence of SO reduction (8 for SO to sulfide vs. 1 for Fe to Fe ), it appears that SO is the more important terminal electron acceptor for AOM. Further, in the model, both transport of SO and AOM together drove Fe-S-CH processes. Downward diffusion of surface water provided high SO concentrations, while upward advecting CH provided an additional electron donor, both of which enhanced SO reduction rates near the sediment-water interface, triggering greater FeS precipitation and Fe(OH) reduction through partial equilibrium.

Over the entire channel profile, on an electron equivalence basis, the model predicts SO reduction to be the most important TEAP in the channel profile (from June to August 2016: SO reduction = 19.2 neq/cm /day, Fe reduction = 4.1 neq/cm /day, and methanogenesis = 1.1 neq/cm /day), consistent with the TEAP ordering in summer 2015 simulations (Ng et al., 2017). This dominance of SO reduction over Fe reduction, despite being lower on the thermodynamic ladder, suggests the possibility of fast “cryptic” sulfur cycling, which is further assessed using microbial observations in section 4.3. This upending of the thermodynamic ladder is possible within a partial equilibrium approach through the calibration of effective model parameters, such as the equilibrium dissolution parameter for Fe(OH) . As described in Ng et al. (2017), this calibrated value may represent the net effect of a variety of ferric oxides, ferric oxyhydroxides, and mixed Fe(II)/Fe(III) compounds, and it serves as an approximation for actual kinetic controls. While detailed mechanisms are concealed in this partial equilibrium approach with effective parameters, it provides a tractable number of parameters to calibrate to bulk aqueous observations in the absence of more specific constraints.

4.2.2.2 Wetland Results

The wetland model simulation over June to August 2016 showed overall more reducing conditions compared to the channel, due to a combination of both higher total organic carbon mass and lower flux magnitude, but it exhibited many similar features as the channel. Similar to channel simulations (Figure 5, top), wetland simulations (Figure 5, bottom) demonstrated vertically distributed Fe reduction, and the greatest Fe and SO reduction rates co-occurred with CH loss just below the sediment-water interface. CH production aligned with the depth interval with the least SO reduction, and SO reduction dominated over Fe reduction in terms of electron equivalence (from June to August 2016: SO reduction = 24.8 neq/cm /day, Fe reduction = 4.2 neq/cm /day, and methanogenesis = 1.6 neq/cm /day). Key distinctions from the channel were the greater rates of SO reduction and CH loss near the top of the profile in the wetlands compared to in the channel, and in relation, less SO reduction at depth. These differences appeared to result from the weaker upward flux in the wetland compared to the channel, which allowed greater downward diffusion of high SO concentration surface water at the top, as well as less upward influx of SO from the deeper porewater at the bottom of the profile. This suggested that even with prevailing upward flux conditions in both locations, differences in magnitude could result in distinct spatial distributions of geochemical processes that may include AOM and S processes.

4.3 Microbial Evidence for Hydrobiogeochemical Processes

4.3.1 Microbial Evidence for the Distribution of TEAPs

Taxonomic profiling and metabolic inference of the 16S rRNA gene sequencing data were used to both corroborate simulation results and offer mechanistic insights into potential processes folded into the calibrated reactive transport model. The high relative abundance of OTUs belonging to families containing known SO -reducing bacteria (SRB) compared to families with iron-reducing bacteria at all depths in the channel and wetland sediments (Figure 6) supported model findings that SO reduction served as the dominant anaerobic TEAP. OTUs belonging to eight SRB-containing families were present in the microbial communities at Second Creek: Desulfoarculaceae, Desulfobacteraceae, Desulfobulbaceae, Desulfovibrionaceae, Syntrophaceae, Syntrophobacteraceae, and Peptococcaceae (see review by Pester, 2012). Across all depths in the channel, these OTUs comprised 1–9% of all V4 sequences in both June and August; in the wetland, these OTUs comprised 4–7% of the sequences in June and 4–11% of the sequences in August.

Figure 6

Relative abundance of OTUs belonging to known bacterial sulfate reducing families (blue shades), bacterial iron reducing families (red/orange shades), archaeal methanogen orders (gray shades), and bacterial methanotroph families (yellow shades) based on taxonomic profiling of 16S rRNA gene (V4 region) sequence data of subsurface sediments. Results are shown for duplicate profiles in the wetland (left) and channel (right) below the sediment-water interface ( ) during June (top) and August (bottom) 2016 sampling periods. Other known sulfate-reducing taxonomic families absent from our data are not listed in the legend.

In comparison, the relative abundance of OTUs belonging to the bacterial families Geobacteraceae and Shewanellaceae, which contain many of the most common Fe reducing taxa in circumneutral terrestrial and subsurface environments (Lovley et al., 2004; Weber et al., 2006), only ranged from 0.06–0.7% in the channel in both June and August, 0.06–1% in the wetland in June, and 0.16–1.7% in the wetland in August. Nitrate reducing bacteria (and archaea) are distributed across distantly related phyla, and consequently, taxonomic profiling of these groups was not conducted at Second Creek. However, inferred dissimilatory nitrate reducing pathways were identified via paprica analysis at relative abundances substantially lower than sulfur-related pathways (0.03–0.05% vs. 0.6–2% across all depths, sampling dates, and locations) (compare Figure S5 in the supporting information and Figure 7). Low porewater concentrations of nitrate (<0.1 mM) in both the channel and wetland further suggested that nitrate reduction likely does not act as a major TEAP at Second Creek.

Figure 7

Abundance of sulfur-related bacterial pathways predicted by paprica via phylogenetic placement of bacterial 16S rRNA gene sequences in a reference genome database. Pathways shown include those for dissimilatory sulfate reduction (blue shades), oxidation of sulfide and S intermediates (red/yellow shades), reduction of S intermediates (green shades), and disproportionation (purple). Results are shown for duplicate profiles in the wetland (left) and channel (right) below the sediment-water interface ( ) during June (top) and August (bottom) 2016 sampling periods. Pathways absent from our samples are not listed in the legend.

Microbiome analyses showed that widespread simulations and observations of porewater CH concentrations in summer 2016 were likely biotically produced by methanogenic archaea. Abiotic methanogenesis typically requires high heat, high pressure, ultraviolet radiation, or strong oxidizing conditions (Althoff et al., 2014; Keppler et al., 2009) that would not be expected at our sampling depths in the sediments. Meanwhile, archaeal OTUs occurred throughout all sediment profiles at relative abundances that reach up to 25% of the sequencing data (Figure 8), and taxonomic profiling revealed that OTUs included four of the seven known orders of archaeal methanogens (Borrel et al., 2013): Methanobacteriales, Methanocellales, Methanomicrobiales, and Methanosarcinales. Below the top few centimeters of the sediment profiles, these archaeal methanogen OTUs reached similar abundances as OTUs classified as SRB-containing taxa at several depths and were substantially greater in abundance than the potential iron reducer OTUs in nearly all samples. Predicted pathways showed that these archaea could include hydrogenotrophic methanogens (methanogens with H /CO coupled redox reactions) (Figure 9). The two other methanogenic pathways (acetoclastic methanogenesis and methylotrophic methanogenesis) were not included in the paprica analysis, but this does not preclude their presence. For example, several species of Methanosarcinales are metabolically diverse and can use all three pathways, while other species are obligately acetoclastic (Zhu et al., 2012).

Figure 8

Relative microbial OTU abundance of archaea and bacteria determined by 16S rRNA gene (V4 region) sequence analysis. Results are shown for duplicate profiles in the wetland (left) and channel (right) below the sediment-water interface ( ) during June (top) and August (bottom) 2016 sampling periods.

Figure 9

Abundance of sulfur- and methane-related archaeal pathways predicted by paprica via phylogenetic placement of archaeal 16S rRNA gene sequences in a reference genome database. Pathways shown include those for dissimilatory sulfate reduction (blue shades), oxidation of sulfide and S intermediates (red/yellow shades), reduction of S intermediates (green shades), disproportionation (purple), and hydrogenotrophic methanogenesis (gray). Results are shown for duplicate profiles in the wetland (left) and channel (right) below the sediment-water interface ( ) during June (top) and August (bottom) 2016 sampling periods. Pathways absent from our samples are not listed in the legend.

4.3.2 Microbial Evidence for Cryptic Sulfur Cycling

The dominance of SO reduction over other TEAPs in simulations could be indicative of “cryptic” sulfur cycling, but this could not be determined with the model alone, which represents only net reactions through its partial equilibrium formulation. This prompted us to look within the microbial community for evidence of cryptic sulfur cycling. Like in our study, Berg et al. (2019) found putative Fe reducing bacteria to be less abundant than SO reducing bacteria in a low-SO lake, which they interpreted to be support for an active “cryptic” S cycle. We further searched for potential pathways to complete S redox cycling at Second Creek, including sulfide reoxidation coupled with Fe reduction and the disproportionation, reduction, and oxidation of sulfur intermediates. Proposed as a key step in S cycling in anoxic freshwater sediments, sulfide has been found to reoxidize abiotically to S intermediates coupled to the reduction of ferric oxides (Elsgaard & Jørgensen, 1992; Flynn et al., 2014; Hansel et al., 2015 Holmkvist et al., 2011; Jorgensen, 1990; Lohmayer et al., 2014). Although Geobacteraceae and Shewanellaceae present at Second Creek may have mediated dissimilatory Fe reduction, high concentrations of porewater Fe together with low relative abundances of these dissimilatory Fe reducers throughout the sediment profiles suggested that some or all of the Fe(III) may have been reduced abiotically through coupling with sulfide (re)oxidation. In fact, Geobacteraceae and Shewanellaceae at our site could even be utilizing electron acceptors other than Fe(III) (e.g., Mn(IV), S(0), and other metals by Geobacteraceae; Lovley et al., 2004; and Mn(IV), S(0), sulfite, nitrate, thiosulfate, and other oxidized metal(loids) by Shewanellaceae; Nealson & Scott, 2006). A coupled Fe and S cycle is consistent with model simulations, which showed higher Fe reduction rates—calibrated to observed increases in porewater Fe concentrations—at the same depths that SO reduction rates are greatest. Co-occurrence of SO and Fe reduction may be driven by the sink to iron sulfide precipitation, and/or it could arise from coupling of abiotic Fe reduction to sulfide (re)oxidation, which subsequently refuels SO reduction.

Although anaerobic sulfide oxidation coupled to Fe reduction occurs abiotically (Canfield, 1989; Dos Santos Afonso & Stumm, 1992; Poulton, 2003; Yao & Millero, 1996), predicted bacterial pathways include sulfide oxidation (Figure 7) in the upper wetland profiles and throughout the channel profiles at low relative abundances compared to other S-related pathways. The main sulfide oxidation pathway predicted (sulfide oxidation II) could be used by some anoxygenic phototrophic bacteria for anaerobic sulfide oxidation (Friedrich et al., 2001), and/or they could indicate the presence of aerobic S organisms (Lü et al., 2017) no longer active under the anoxic sampling conditions of summer 2016.

As further support for anoxic S cycling at Second Creek, both taxonomic profiling and pathways predicted via paprica indicate potential activities involving S intermediates, which are products of abiotic sulfide oxidation coupled to Fe reduction. Genera containing members known to mediate S-intermediate disproportionation, Desulfobulbus (Widdel & Pfennig, 1982) and Desulfococcus (Milucka et al., 2012), were present in all taxonomic profiles, primarily in the top 20 cm of most profiles (Figure 10). Sulfurospirillum, a genus containing some species capable of reducing zero-valent sulfur but not SO (Straub & Schink, 2004), was also found in a few depths of a few cores, though only at low relative abundance (7 10 –2.5 10 %) (Figure 10). Occurring more prevalently, sulfur-cycling bacterial pathways inferred via paprica included various processes utilizing S intermediates that together exceeded the predicted relative abundance of SO reduction in all samples (Figure 7).

Figure 10

Relative abundance of bacterial OTUs belonging to known genera involved in sulfur disproportionation (brown shades), sulfate reduction (green shades), and reduction of sulfur intermediates but not sulfate (blue) based on 16S rRNA gene sequence analysis. Results are shown for duplicate profiles in the wetland (left) and channel (right) below the sediment-water interface ( ) during June (top) and August (bottom) 2016 sampling periods. Microbial groups absent from our samples are not listed in the legend.

Potential S-intermediate utilizing processes could be serving to complete full redox cycles between SO and sulfide, and they could also be contributing to subcycles embedded within the full cycle (Findlay & Kamyshny, 2017). For example, predicted sulfur reduction pathways could be re-reducing S intermediates formed by abiotic sulfide oxidation coupled to Fe reduction, possibly serving as an electron shuttle to facilitate further Fe reduction (Lohmayer et al., 2014). Additionally, predicted sulfite oxidation could be occurring before the full reduction of SO to sulfide, possibly helping to sustain SO reduction when sulfite to sulfide reduction becomes a bottleneck (although this is mostly expected in low-organic matter sediments) (Antler et al., 2013). All known dissimilatory SO reduction pathways produce sulfite as an intermediate, but they differ in how sulfite is reduced to sulfide. Although a direct six-electron reduction of sulfite to sulfide is possible, the SR V pathway (MetaCyc Pathway: dissimilatory SR II to thiosulfate), the dominant dissimilatory SO reduction pathway predicted in our samples, uses a complex of dissimilatory sulfite reductase enzymes, DsrAB and DsrC, to first produce thiosulfate through an initial loss of two electrons (Crane & Getzoff, 1996; Santos et al., 2015). The high relative abundance of predicted thiosulfate disproportionation pathways in the bacterial and archaeal communities in our samples could be utilizing thiosulfate produced during this SO reduction pathway as well as by abiotic oxidation of sulfide. Disproportionation produces both sulfide and SO to further drive both oxidative and reductive S pathways coupled to Fe and carbon. Providing insight into the relative roles of the potential S intermediate reactions, freshwater incubation experiments by Findlay and Kamyshny (2017) showed S consumption rates to be greater than that of sulfite and thiosulfate. Together, our S-related microbial results suggest that the microorganisms present in the wetland and channel sediments could facilitate complex cryptic sulfur cycles that comprise diverse redox pathways and support SO reduction as the dominant apparent TEAP in this environment. This microbial evidence for cryptic sulfur cycling as the primary driver of SR in sediments helps justify our use of a simplified partial equilibrium model that omits explicit kinetic competition among TEAPs.

Reoxidation of sulfide to SO , mediated by abiotic coupling with Fe reduction and enzymatically driven sulfur intermediate oxidation or disproportionation pathways, was not explicitly represented in the model. This could explain the undersimulation of SO in the middle portion of the sediment profiles ( 10 to 30 cm depth in the channel and 10 to 50 cm depth in wetland). While SO fully depletes before methanogenesis begins in the model, observations show low concentrations at these depths that could have been produced through reoxidation processes, and/or they could have resulted from unrepresented kinetic factors limiting complete SO reduction.

4.3.3 Microbial Evidence for AOM

A challenge to corroborating model predictions of AOM in the upper 10 cm of the sediment profiles with microbial analysis is the lack of pure cultures of organisms capable of mediating AOM and persisting uncertainties in AOM pathways. However, a metabolically active community of anaerobic methanotrophic (ANME) archaea has been identified in terrestrial and freshwater subsurface environments (Takeuchi et al., 2011; Timmers et al., 2016; Weber et al., 2016, 2017), and recent studies have linked AOM to SO reduction for specific taxonomic groups of ANME and other AOM-associated archaea in freshwater subsurface environments (Milucka et al., 2012; Timmers et al., 2016, 2017; Weber et al., 2017). In our samples, it is possible that the archaeal community includes consortia of ANME, but unfortunately, taxonomic profiling of the OTUs could not resolve the ANME-specific clades with the short-read DNA sequences, and paprica analysis is limited to known pathways in pure culture isolates. We do, however, have indirect microbial evidence that is consistent with possible AOM coupled to SO reduction. Desulfococcus, a known sulfur disproportionator and SO reducer, was measured in low abundance at several depths at our site, and it belongs to the Desulfosarcina/Desulfococcus clade that has been found to reduce SO in association with ANME in marine sediments (Milucka et al., 2012; Orphan et al., 2001). Additionally, nitrite-dependent AOM was inferred via paprica analysis to be present at low relative abundance at many depths (Figure S5). Although low nitrate and nitrite concentrations in the Second Creek system likely make this an unimportant part of the CH budget, it points to the possibility of anaerobic methane oxidation in the sediments.

Bacterial community analysis did identify OTUs belonging to families with known aerobic methanotrophs, Methylococcaceae and Methylocystaceae (Hanson & Hanson, 1996), at depths in both the channel and wetland profiles where observations of dissolved Fe and CH indicate anaerobic conditions (Figure 6). Methylococcaceae has been found in anaerobic aquatic settings, where photosynthesis produces small amounts of oxygen immediately used by the microbes (Milucka et al., 2015; Oswald et al., 2015). Plant roots could create aerobic microenvironments that support aerobic methanotrophs. However, this was not likely to be the main explanation at Second Creek, where the potential aerobic methanotrophs extended deeper into the channel profile (down to at least 28 cm depth) compared to in the wetland (mostly within the top 10 cm). While aquatic plants are common in both locations, they are densest in the wetland, and roots were seldom noted below about 5 to 15 cm depth in any of the sediment profiles. Alternatively, the presence of these taxa may have resulted from remnant DNA or inactive microbes. Methanotrophs have previously been demonstrated to survive extended anoxic conditions and become metabolically active again when oxygen is present (Takeda, 1988). The previous summer (2015), a 2-month flood event caused downward flux of oxidizing surface water (with elevated SO concentrations) to at least 25-cm depth in the channel sediments and to at least 5- to 10-cm depth in the wetland sediments (Ng et al., 2017). These depths correspond closely with the observed depths of the aerobic methanotrophic families in the channel and wetland, providing evidence for aerobic methanotrophy under the flooded conditions of the previous summer (2015).

4.3.4 Spatiotemporal Microbial Variability

Overall, the wetland profiles exhibited more spatially heterogeneous microbial distributions than the channel profiles. We infer this to be in response to lower hyporheic flux magnitudes measured in the wetland, which in the summer of 2016 allowed greater downward diffusion of high-SO surface water near the sediment-water interface while limiting the upwelling of groundwater with moderate SO concentrations at the profile bottom. Consistent with the more sharply contrasting influxes of SO at the top and bottom of the wetland profile, the wetland showed a greater attenuation with depth of the relative abundance of taxa capable of sulfur disproportionation and predicted thiosulfate disproportionation bacterial pathways. Also, the relative abundance of archaeal OTUs increased more strongly with depth in the wetland than channel (Figure 8). As archaea are considered to contribute substantially to methanogenesis (Bridgham et al., 2013; Thauer et al., 2008), this is consistent with deeper and greater CH concentrations observed in the wetland (Figure 4), where more reducing conditions were supported by higher amounts of organic matter inferred through model calibration.

In contrast to the spatial trends, we observed little temporal variability in the microbial community composition of the targeted functional groups, with no systematic changes between June and August profiles collected from the same area (channel or wetland). In addition to relatively static community composition over a single summer season, the presence of aerobic methanotrophs, most notably throughout the higher-flux channel profiles, suggested that the composition could remain relatively stable for over a year, complicating interpretations of microbial community data relative to concurrent environmental conditions. The minimal temporal variability suggested that the observed microbial community included substantial components of both active and dormant organisms. Dormant organisms may be able to relatively quickly (re)activate when specific allowing conditions return. This suggested that the sediment may harbor a microbial repository that can facilitate fast responses to biogeochemical changes under transient hyporheic conditions.

4.3.5 Uncertainties in Functional or Metabolic Predictions

Assigning potential functional or metabolic activity based on either taxonomic profiling and manual curation or inference-type microbial analysis tools such as paprica with short-read microbiome data sets each had its strengths and weaknesses. Use of taxonomy works well when clades of microorganisms have a known single metabolism (e.g., methanogens), but difficulties arise when taxa are metabolically diverse, such as Shewanella, and when metabolisms are widespread among numerous phylogenetic groups, such as for nitrate reduction. Further, manual curation relies on establishing a database of known metabolisms, and some newer or lesser known taxa may be overlooked. Tools such as paprica provide a convenient alternative to the time-intensive manual-curation methods, and they allow for exploring potential functional capabilities within much larger community data sets (Bowman et al., 2017). However, these approaches have uncertainties related to inherent limitations of phylogenetic placement in genomic databases, including low representation of key enzymes in the genomic databases and/or the inability to assign DNA sequences to such enzymes during metabolic inference (Bowman et al., 2017). Both methods should be interpreted with caution because of differences that may arise between potential function and expressed function in a specific community.

These uncertainties likely contributed to the contradictions found between predicted potential metabolic pathways and other microbial patterns or modeled and observed geochemical results. Incomplete archaeal databases in paprica (only 220 genomes are included) may explain a confounding decrease in archaeal hydrogenotrophic methanogenesis pathway abundance with depth in the wetland compared to overall greater CH concentrations measured and simulated. The archaeal reference genome database does not include acetotrophic and methylotrophic methanogenesis pathways that are known to be possible in some observed methanogenic orders, and these may produce most of the CH measured at greater depths in the wetland. Similarly, gaps in archaeal inferences may explain the lack of depth dependency in sulfur disproportionating archaeal pathways, in contrast to decreasing relative abundance of sulfur disproportionating bacterial pathways that align with decreasing SO concentrations. There are uncertainties in bacterial as well as archaeal pathway inferences. For example, the relative abundance of predicted SO reduction and intermediary sulfur oxidation bacterial pathways increased with depth in the wetland, a trend not seen in SO -reducing taxa abundances nor in observed SO concentrations. Despite the uncertainties related to each approach, however, the trends in geochemistry largely correlate with changes in microbial community composition and inferred function in this study.

4.4 Simulations of Hyporheic Flux-Driven Sulfate and Methane Interactions

Despite minimal temporal variability observed in the microbial community composition, the spatial distribution in microbial composition differed between channel and wetland conditions in a manner consistent with prevailing flux and geochemical conditions at each location. These consistent observations point to the importance of dynamic hyporheic exchanges and raises the question of whether fluctuations in upward flux, such as those that occurred in the summer of 2016, can impact inferred “cryptic” S cycling and AOM, thus driving variable SO and CH concentrations throughout the sediments.

The influence of fluctuating hyporheic flux in determining sulfur's role in CH -cycling was tested by comparing 2016 simulations that incorporated daily varying flux (estimated flux time series from the temperature probe data, rescaled to match the calibrated time-average flux) against baseline simulations produced with constant time-average flux (Figure 11, left). In both simulations, the greatest temporal changes in SO and CH concentrations occurred during the first month (channel) to month and a half (wetland), due to the response to increasingly reducing conditions in the summer. Following this adjustment period, the difference in concentrations between the two simulations represented the impact of varying flux. In the channel, variable flux drove a change of up to 0.07 mM in CH concentrations from the baseline—this corresponds to up to a 40% change near the top of the sediment profile and up to a 100% change at the bottom of the actively biodegrading profile. In the wetland, variable flux similarly resulted in up to a 0.07 mM change in CH concentrations, but because of the overall higher concentrations there, this represents a subtler 15% change at the top of the profile, while the bottom of the profile showed higher sensitivity with up to about 85% change.

Figure 11

(left) Simulated concentration time series using time-varying hyporheic flux (light blue) for the channel (a, c) and wetland (e, g) are shown in solid lines, compared to dashed lines showing simulated concentration time series using constant hyporheic flux (average of time-varying). (right) For simulations with time-varying hyporheic flux, cross-correlation of simulated concentrations lagging hyporheic flux for the channel (b, d) and wetland (f, h). The dotted lines in the cross-correlation plots show 0 cross-correlation for reference. “Top” profile simulations include sulfate concentrations at 3.7 cm depth below the sediment-water interface and dissolved CH concentrations at 5.25 cm depth (a, e). “Bottom” profile simulations include sulfate concentrations at 50-cm (70-cm) depth and 30-cm (55-cm) depth in the channel (wetland) (c, g). CH and SO are shown at slightly offset depths for each part of the profile, because sulfate and CH cannot thermodynamically co-occur in the partial equilibrium implementation of the model. The “bottom” of the degrading profile is set deeper in the wetland model domain compared to in the channel model domain due to more reducing conditions observed in the wetland.

Depth-dependent cross correlations between flux and the concentrations in both locations in Figure 11 (right) provide insights into the processes driving the differences in concentrations and highlight the vertical spatial heterogeneity within the hyporheic zone. Near the bottom of the active biodegradation zone in both the channel and wetland, SO concentrations are positively correlated over time with flux with about a 3- to 4-day lag. This adds more rigorous support for the above inference (based on geochemical profile results) that moderate SO concentrations in deeper porewater advected upward with variable hyporheic flow to replenish areas where SO had been reduced previously. The time series analysis further demonstrated that the transport time was on the order of a few days. Toward the top of the profile by the sediment-water interface, the correlation between flux and SO was in contrast negative, which confirmed that in near-surface sediment, times of lower-magnitude upward flux allowed greater downward diffusion of high surface water SO . This effect at the top of the profile is immediate—usually with no or little lag time, likely because the downward diffusing SO concentrations were high compared to in deeper porewaters. Interestingly, at greater lag times (e.g., approaching 7 days), the top of the profile preserved some of the relationship with flux that formed at the bottom of the profile (positive correlation between SO concentration and flux - opposite of the negative correlation at short lag times), as that porewater moved up the profile. This convolution of different flux influences over space and time on SO concentrations complicates predictions of solute concentrations near the sediment-water interface based only on flux at a particular moment. These dynamic flux-controlled results point to transport as a major driver of sulfate processes in the sediment, which helps support our simplified representation of geochemical reaction kinetics in the partial equilibrium model.

At both the top and bottom of the profiles, SO concentrations were negatively correlated with CH concentrations, but as a result of different processes at either end. Reaction mass balance model results in Figure 5 suggested that at the bottom, SO reduction and methanogenesis compete as TEAPs, while near the top, the model showed SO reduction occurring alongside CH oxidation. The microbial data were consistent with the model, with the lower profile showing plentiful methanogens and SO -reducing bacteria and the upper profile showing lower abundance of methanogens and greater abundance of potential sulfur disproportionators—an indicator of potential “cryptic” S cycling that may include S-intermediate pathways and SO -reduction coupled to AOM.

Overall, the transient flux simulations suggested that times of overall greater upward flux may have attenuated CH production at depth but could have facilitated the transport of CH to the surface where it may be outgassed. Conversely, times of weaker upward flux may have experienced the greatest rates of CH production, though they may also have had greater CH consumption near the surface coupled to SO reduction. Collectively, the model simulations and microbial observations suggest that time-varying hyporheic fluxes caused alternating times of CH production deeper in the sediment versus CH transport toward the surface, resulting in hot moments of CH emissions that may occur at a spatial and temporal offset from hot moments of methanogenesis.

5 Summary and Conclusion

Using a combination of reactive transport modeling and microbiome analysis at a SO -impacted wetland-stream system in northeast Minnesota, we have determined that dynamic hyporheic flow allowed spatiotemporally variable influxes of SO into channel and wetland sediments, which controlled a coupled Fe-S-CH network of reactions that likely involved “cryptic” S cycling and AOM. Similar to the previous summer (2015) modeling results at the site by Ng et al. (2017), SO reduction rates simulated in the summer of 2016 appeared to dominate over thermodynamically favorable Fe reduction. In this current work, we found a relatively high abundance of SO reducing bacteria that was consistent with these model results, as well as further evidence that modeled net SO reduction rates may be influenced by “cryptic” S cycling. Relatively low abundance of known Fe-reducing bacteria raised the possibility of abiotic Fe reduction driving sulfide reoxidation to S intermediates, and the widespread abundance of taxa and predicted pathways related to disproportionation, reduction, and oxidation of sulfur intermediates suggested diverse potential redox mechanisms that could complete S redox cycles. Although contributing a smaller portion of the redox electron budget than SO or Fe reduction, model simulations and microbial observations suggested that methanogenesis occurred ubiquitously in both the wetland and channel sediments throughout the summer of 2016. Calibrated simulations showed that CH concentrations were limited in the lower profile due to TEAP competition with SO reduction, and they attenuated in the upper profile due to AOM mostly coupled to SO reduction. Archaea, present throughout all profiles at abundances of up to 25% of the microbial community, could have been responsible for carrying out AOM at the site, although targeted ANME (consortia of anaerobic methanotrophic archaea) analysis is needed to resolve this.

Microbial community composition appeared overall stable throughout the summer of 2016, but spatial differences and model simulations indicated that dynamic hyporheic fluxes both within a season and across years influenced biogeochemical activity. Higher-magnitude hyporheic flux in the channel compared to the wetland attenuated the magnitude of vertical geochemical gradients, allowing more vertically homogeneous concentrations in channel sediments and correspondingly more broadly distributed microbial communities. Stronger spatial patterns emerged in wetland sediments. Under typical upwelling conditions, weaker upward advection in the wetland permitted greater downward diffusion of high surface water SO concentrations at the top of the profile and more limited upwelling of moderate groundwater SO concentrations at the bottom of the profile, creating a sharper SO front between the top and bottom compared to in the channel. This led to generally greater attenuation with depth of S-related taxa and predicted microbial pathways. Aerobic methanotroph taxa found throughout most of the channel profile under observed anoxic conditions were hypothesized to be remnants from an unusual flood event in the previous year, suggesting that the sediment may harbor a long-term (multiyear) microbial repository that facilitates fast responses to biogeochemical changes under transient hyporheic conditions. Within a season with overall upwelling conditions, the model showed that natural fluctuations in flux magnitude could cause up to 40% changes in CH concentrations in the upper portion of the sediment and up to 100% changes deeper in the profile. Further, because of the distinct interactions between CH and SO in different portions of the sediment profile, fluctuating fluxes may cause hot spots and hot-moments of CH transport toward the sediment-water interface that occur at a spatiotemporal offset from hot spots and moments of CH production.

Together, the partial equilibrium reactive transport model and microbiome survey results in this study present quantitative and qualitative evidence for a hyporheic flux-driven S cycle involving “cryptic” reactions and coupling to Fe processes and AOM. These findings now prompt more detailed measurements and analyses to constrain actual biotic and abiotic geochemical pathways and their respective kinetic properties. Speciation measurements can be used to probe the Fe phases and the various intermediate valence sulfur forms implicated in the microbiome analysis, and metagenomic and metatranscriptomic analysis can be applied to interrogate actual microbial activity. Higher-resolution geochemical data will be needed to characterize dynamic changes for each reaction in the network, including isotopic measurements to constrain gross (not just net) cycling rates. Such data will make it possible to move to a more mechanistic reactive transport model, built around an explicit representation of Fe-S-CH reaction network with data-driven rate laws, in order to more accurately quantify CH and other elemental budgets under changing environmental conditions. Our current study has demonstrated the potential importance of these processes and thus provides clear motivation for further work in both SO -impacted and unimpacted wetlands to develop a more mechanistic and predictive understanding of linked biotic-abiotic Fe-S-CH processes under hydrologically dynamic conditions.

Acknowledgments

To support this work, G.-H. C. Ng received funds from the Department of Earth Sciences, and C. Santelli received funds from MnDRIVE Environment, both at the University of Minnesota, Twin Cities. The authors would like to acknowledge Patrick O'Hara (University of Minnesota-Twin Cities) for graphical assistance in creating Figures 1 and 2. Joshua Torgeson and Liz Roepke from University of Minnesota, Twin Cities, and Daniel Fraser and Sophie LaFond from University of Minnesota, Duluth, provided field assistance. Chad Sandell (University of Minnesota, Twin Cities) helped design and construct temperature probes for determining hyporheic flux. We thank two anonymous reviewers and the associate editor for their helpful comments. Hydrologic and geochemical data are publically available in the EDI (Environmental Data Initiative) Data Repository (https://doi.org/10.6073/pasta/611c867faf4da200141246c8e9c494c5). Microbial metadata and raw DNA sequences are publically available in the NCBI Sequence Read Archive under BioProject PRJNA530072 and Biosamples SAMN11292835 to SAMN11292879.