Figure 2: Falling Creek Reservoir (FCR; 37.30ºN, 79.84ºW) and Beaverdam Reservoir (BVR; 37.31ºN, 79.81ºW) are eutrophic drinking water reservoirs located in southwest Virginia, USA (data from Carey, Lewis, Howard et al. 2022, following Woelmer et al. 2022).
2.2 Whole-ecosystem oxygenation experiments
In 2012, FCR was equipped with a side-stream supersaturation hypolimnetic oxygenation (HOx) system to improve water quality in the reservoir (Gerling et al., 2014). This type of HOx system functions by withdrawing water from the bottom of the reservoir, adding concentrated, pressurized oxygen gas (95% purity) to supersaturate the water with dissolved oxygen (DO), and then returning the oxygenated water at the same depth and temperature. Previous work in FCR has shown that the HOx system effectively increases DO concentrations throughout the hypolimnion without altering temperature or decreasing thermal stability (see Gerling et al., 2014). From 2013–2019, the HOx system in FCR was operated at variable rates, maintaining an oxygenated hypolimnion for at least part of the summer (Carey, Thomas, et al., 2022). Conversely, oxygenation was reduced in 2020 and 2021, resulting in median hypolimnetic DO concentrations < 1 mg/L throughout the summer stratified period. To assess the effects of multiannual changes in oxygen availability on OC and Fe-OC in sediment, we compared sediment core and sedimentation trap data from summer 2019 (which had a history of high-oxygen conditions during the preceding six years) to summer 2021 (which followed a summer of hypoxic conditions in 2020; Figure S1). Sediment data were not collected in 2020 due to the Covid-19 pandemic.
To assess how short-term changes in hypolimnetic DO concentrations impact Fe-OC on a whole-ecosystem scale, we operated the HOx in FCR on a variable schedule throughout the summer of 2019 (Carey, Thomas, et al., 2022). Oxygen was added in approximately two-week intervals at a rate of 25 kg O2 day-1 to the whole hypolimnion. Between oxygenation periods, we allowed the hypolimnion to become hypoxic over periods of at least two weeks without oxygenation. Because hypolimnetic volume varied throughout the summer (generally decreasing throughout the summer as the thermocline deepened), the mean concentration of oxygen added to the whole hypolimnion throughout an oxygenation period in 2019 ranged from 0.80 mg L-1 day-1 to 0.90 mg L-1 day-1.
BVR does not have a HOx system and experiences seasonal hypoxia from May through November (Hounshell et al., 2021). Consequently, BVR serves as a reference ecosystem to analyze the effects of oxygenation in FCR.
We monitored DO concentrations throughout the full water column approximately two times per week in FCR and one time per week in BVR (Carey, Lewis, McClure, et al., 2022). High-resolution (~1 cm) depth profiles were taken using a conductivity, temperature, and depth profiler (CTD; Sea-Bird, Bellevue, Washington, USA) equipped with a DO sensor (SBE 43; Carey, Lewis, McClure, et al., 2022) from the reservoir’s surface to the sediments. We also measured dissolved oxygen using a YSI ProODO DO probe when the CTD was not available due to maintenance (YSI Inc. Yellow Springs, Ohio, USA; Carey, Wander, McClure, et al., 2022). YSI measurements were taken at discrete 1 m depth intervals. For a comparison of YSI and CTD measurements, see Carey et al. (2022).
2.2.2 Hypolimnetic Fe and DOC We collected water samples for DOC and Fe analysis at the deepest site in each reservoir with a 4-L Van Dorn sampler (Wildlife Supply Company, Yulee, FL, USA). Samples were collected once per week at seven depths in FCR (0.1, 1.6, 3.8, 5.0, 6.2, 8.0, and 9.0 m), which corresponded to the reservoir’s extraction depths, and five depths in BVR (0.1, 3.0, 6.0, 9.0, and 11.0 m). In 2019, we conducted a limited amount of additional sampling in FCR on a second day each week, and these measurements included DOC from 0.1, 1.6, 5.0, and 9.0 m depths.
We analyzed DOC by filtering water samples through a 0.7-µm glass fiber filter into an acid-washed bottle, which was rinsed with the filtered water three times before sample collection. The filtered samples were frozen for less than six months before analysis on an OC analyzer (Elementar Vario TOC cube, following APHA standard method 5310B; American Public Health Association, 2018b).
We collected both total and dissolved (filtered through 0.45-µm filters) samples for Fe. Samples were preserved in the field using trace metal grade nitric acid and analyzed using ICP-MS (Thermo Electron X-Series, Waltham, MA, USA) following APHA Standard Method 3125-B (American Public Health Association, 2018a; Krueger et al., 2020; Munger et al., 2019; Schreiber et al., 2022).
We analyzed the concentration of Fe-OC in surficial sediments from both FCR and BVR on multiple dates throughout the summer stratified periods of 2019 and 2021. In 2019, sediment cores in FCR were collected immediately before the HOx system was turned on or off, resulting in the most oxic or hypoxic conditions during that SSS activation or deactivation interval, respectively. Sediment cores at BVR were taken once in the middle of summer and once approximately two weeks before fall turnover in 2019. In 2021, sediment core samples were taken from both reservoirs on the same dates, approximately once per month. Additional sediment core samples were collected in March 2021, when both reservoirs were unstratified and had oxic hypolimnia.
On each sampling date, we collected four replicate hypolimnetic sediment cores using a K-B gravity sediment corer (Wildlife Supply Company, Yulee, FL, USA). Cores were collected in the deepest part of each reservoir, approximately 20 m from where water samples were taken. In 2019, each core was capped and kept on ice while transported back to the lab, where the top 1 centimeter of sediment from each core was immediately extruded, collected, and frozen in scintillation vials for future analysis. In 2021, cores were extruded in the field, and the samples were kept on ice while being transported back to the lab.
To determine the amount of Fe-OC and total OC in samples of material settling from the water column (i.e., not estimate deposition rates), we deployed 19-L buckets approximately 1 m above the sediments at the deepest point of each reservoir (8 m at FCR and 10 m at BVR). These sediment traps were deployed from June–December 2021 and sampled every two weeks by slowly bringing the bucket to the surface, decanting and discarding water from the bucket, collecting up to 5 L of the remaining water and particulate matter, and transporting this material back to the lab on ice. Upon arriving at the lab, we allowed the particulates to settle for approximately 5 minutes before decanting and discarding as much water as possible and filling four 50-mL centrifuge tubes with the remaining material. The samples were centrifuged for 10 minutes at 3100 rpm, then combined into one vial and frozen for later analysis. No sediment traps were deployed for Fe-OC analysis in 2019.
2.3 Microcosm incubations
To isolate the effects of oxygen from other interacting factors that affect Fe and OC on a whole-ecosystem scale, we conducted six-week microcosm incubations using hypolimnetic sediment and water from FCR. Incubations were conducted in 177-mL glass jars with two-piece gasket-sealed lids (Verones brand; Figure S2), after extensive pilot testing revealed that these jars were highly effective at maintaining hypoxic conditions when sealed (DO concentrations < 0.5 mg/L in this experiment) and oxic conditions when uncapped. We started the experiment with 102 microcosms split evenly into oxic (uncapped) and hypoxic (capped) treatments. After two weeks (similar to the 2019 whole-ecosystem HOx manipulation), we switched the treatment of approximately half of the remaining microcosms, generating two additional oxygen regimes: hypoxic-to-oxic and oxic-to-hypoxic. Starting on week two, there were consequently a total of four oxygen regimes: hypoxic, oxic, hypoxic-to-oxic, and oxic-to-hypoxic.
To set up the experiment, we collected sediment and water from the deepest site in FCR on 30 June 2021, when the hypolimnetic DO concentrations were < 0.5 mg/L. Water was collected from 9 m depth using a Van Dorn sampler, and sediment was collected from the same location using an Ekman sampler. Samples were transported on ice back to the lab, then homogenized by stirring and shaking. We used a syringe to add the sediment slurry (20 mL) to each jar, then slowly added 150 mL of hypolimnetic water, making an effort to minimize sediment disturbance. We stored the microcosms in an unlit incubation chamber at 15 ºC for the duration of the experiment, which corresponded to warm, end-of-summer conditions in the hypolimnion of FCR (Carey, Lewis, McClure, et al., 2022).
Microcosms were sampled destructively for DO, total and dissolved Fe, total and dissolved OC, pH, sediment OC, and sediment Fe-OC. For the continuous oxic and hypoxic treatments, we sampled 3–6 replicates two times per week for four weeks (6 replicates: days 2, 6, 9, 13; 3 replicates: days 16, 20, 23). We added additional sampling for the hypoxic-to-oxic and oxic-to-hypoxic treatments: these treatments were sampled for the first three days after switching the oxygen regime (days 14, 15, 16), twice the following week (days 20, 23), and one more time a total of four weeks from when treatments were switched (day 34), with three replicates analyzed per sampling event. All microcosms under a hypoxic treatment were sampled in an anaerobic chamber which maintained mean ambient oxygen conditions <200 ppm (Coy Laboratory, Grass Lake, MI, USA) to reduce oxygen exposure during sampling.
To begin sampling a microcosm, DO was measured using a YSI DO probe. While measuring DO, we used the probe to gently swirl the water in the microcosm, homogenizing the water sample while minimizing sediment disturbance. Next, we used an acid-washed syringe to withdraw 30 mL of water for total OC (TOC), 13 mL for total Fe, 30 mL of water for DOC, and 13 mL for dissolved Fe analyses. DOC samples were filtered through a 0.7-µm glass fiber filter, and dissolved Fe samples were filtered through 0.45-µm filters. After taking samples for Fe and DOC, we withdrew as much water as possible without disturbing the sediment and measured pH from this sample in a separate container using an Ohaus Starter 300 pH probe (Parsippany, NJ, USA). Finally, we swirled the sediment with remaining water (approximately 1–5 mL) and poured this mixture into a 20 mL glass EPA vial, which we then froze for Fe-OC analysis. Hypoxic microcosms were stored in the anaerobic chamber for approximately two hours before analysis to ensure oxygen concentrations in the chamber were sufficiently low before opening the jars. Oxic microcosms were sampled immediately after removal from the incubator.
All microcosm samples were analyzed following standard methods. We stored TOC and DOC samples in bottles that had been acid-washed and rinsed three times with the sample water. All DOC and TOC samples were frozen for <6 months prior to analysis on an OC analyzer (Elementar Vario TOC cube, following Standard Method 5310B; American Public Health Association, 2018b) Fe samples were preserved using trace metal grade nitric acid and analyzed using the ferrozine method (Gibbs, 1979). We also analyzed Fe samples from days 16 and 23 using inductively coupled plasma mass spectrometry (ICP-MS). All microcosm data are published with complete metadata in the Environmental Data Initiative repository (Lewis et al., 2022).
We analyzed the amount of Fe-OC in both the whole-ecosystem and microcosm sediment samples using the citrate bicarbonate dithionite (CBD) method (Figure S3). This method was first described for marine systems by Lalonde et al. (2012) and has since been adapted for freshwater lakes by Peter and Sobek (2018). It is important to note that our measurement of Fe-OC as the percentage of OC that is extractable using the CBD method is an operational definition (Fisher et al., 2021). CBD extractions have documented inefficiencies when extracting crystalline hematite (Thompson et al. 2019; Adhikiri & Yang, 2015) and carboxyl-rich compounds (Fisher et al. 2020). While Fe is the primary reducible metal that associates with OC, other metals, including aluminum (Al) and calcium (Ca), may also release OC during CBD extractions. However, previous work in soils found that CBD-extracted aluminum was approximately an order of magnitude lower than CBD-extracted Fe, and therefore quantitatively much less important (Sondheim and Standish, 1983). Moreover, we found that Fe was present in much (≥ 5 times) higher quantities than Al and Ca in water samples across all of our sediment incubation treatments (Lewis et al., 2022), further justifying our use of the operational term Fe-OC. We used the CBD method to enable comparisons both between oxygen treatments and with other published work that used the same general approach (e.g., Lalonde et al., 2012; Peter & Sobek, 2018).
Following the CBD method, each sediment sample was freeze-dried and divided into three treatments: initial, reduction, and control (Figure S3). “Initial” samples received no treatment and were used to measure the OC content of the sediment. “Reduction” samples were treated with a metal-complexing agent (trisodium citrate) and reducing agent (sodium dithionite) in a buffered solution (sodium bicarbonate) to measure how much Fe and OC were released as a result of Fe reduction. Control samples were used to account for the release of OC in the reduction treatment that resulted from processes other than Fe reduction. They were treated with the same buffer (sodium bicarbonate) and sodium chloride in the same ionic strength as the trisodium citrate and sodium dithionite of the reduction treatment.
For both the control and reduction treatments, we measured 100 mg of homogenized, freeze-dried sediment into 15-mL polypropylene centrifuge tubes (Falcon Blue, Corning Inc., Corning, NY, USA). We then added 6 mL of either control or reduction buffer solution (0.11 M sodium bicarbonate) to each tube. The reduction buffer contained 0.27 M trisodium citrate, while the control buffer contained 1.6 M sodium chloride. After heating samples to 80ºC in an oven, 0.1 g sodium dithionite was added to the reduction samples and 0.088 g sodium chloride was added to control samples. Samples were kept at 80ºC for an additional 15 min, then centrifuged for 10 min at 3100 RPM. The supernatant was discarded. This extraction process was repeated two more times for both treatments, resuspending the sediment pellet each time by vortexing with buffer solution (Peter and Sobek, 2018).
Following the extraction step, samples were rinsed three times using OC- and Fe-free artificial lake water. Artificial lake water was prepared by diluting Artificial Hard Water from Marking and Dawson (1973) to 12.5% with Type I reagent grade water. We added 3 mL of artificial lake water to each tube and resuspended the sediment pellet using a vortex. Samples were then centrifuged for 10 min at 3100 RPM, and the supernatant was discarded.
After extraction and rinsing, all sediment samples (including those in the initial treatment) were dried and acid-fumigated for 48 hours to remove remaining citrate and bicarbonate (Harris et al., 2001). Samples were then run on a CN analyzer (Elementar VarioMax, Ronkonkoma, NY, USA) to determine the amount of OC per unit mass of sediment. In these calculations, we adjusted sediment mass to account for Fe loss during control and reduction treatments (Lewis et al., 2022; Lewis, Schreiber et al., 2022; Peter and Sobek, 2018; Text S1). The amount of OC removed with Fe reduction (CBD-extractable OC) was calculated as the difference between the OC content of the control and reduction samples and expressed as a percentage of the initial OC content of the sediment.
All analyses were performed in R (version 4.0.3; R core team 2020) using packages tidyverse (Wickham et al., 2019), lubridate (Grolemund & Wickham, 2011), ggpubr (Kassambara, 2020), egg (Auguie, 2019), rstatix (Kassambara, 2021), akima (Akima et al., 2022), colorRamps (Keitt, 2022), rLakeAnalyzer (Winslow et al., 2019), and tseries (Trapletti et al., 2022). All novel analysis code is archived as a Zenodo repository (Lewis, 2022).
2.5.1 Sediment Fe-OC characterization We calculated summary statistics to describe iron-bound organic carbon and total organic carbon in surficial sediment (2019 and 2021) and settling particulate material (2021 only) across both reservoirs. We then pooled data from both reservoirs to analyze the difference between settling material and surficial sediments using Welch’s t-tests. Because data were unavailable for settling material in 2019, the comparison of settling material to surficial sediment was limited to 2021 data only.
2.5.2 Whole-ecosystem experiments: short-term responses We used Welch’s t-tests to assess whether sediment properties differed between the two-week periods of HOx activation compared to HOx deactivation during summer 2019 in FCR. Sediment time series did not exhibit significant temporal autocorrelation, justifying this approach (Lewis, Schreiber, et al., 2022).
To qualitatively assess whether oxygenation experiments led to differences in water column chemistry, we overlayed plots of DOC and Fe from the deepest sampling depth in each reservoir with dissolved oxygen at the same depths throughout the summer stratified period of 2019 .
2.5.3 Whole-ecosystem experiments: interannual differences We assessed whether there were significant differences in sediment properties among the four reservoir-years—BVR in 2019 (hypoxic), BVR in 2021 (hypoxic), FCR in 2019 (oxic) and FCR in 2021 (hypoxic). First, we used Levene tests to assess homogeneity of variance among reservoir-years (Table S1). While Fe-OC (both per unit sediment and as a percentage of sediment OC) met the ANOVA assumption of homogeneous variance, total sediment OC did not. Consequently, we used one-way ANOVAs and Tukey post hoc tests for Fe-OC metrics, but used Welch one-way ANOVAs and Games-Howell post-hoc tests, both of which account for unequal variances, for sediment OC (Tables S2 and S3).
2.5.4 Microcosm incubations We used one-way ANOVAs and Tukey post-hoc tests to assess whether sediment properties differed between microcosm treatments, after testing for homogeneity of variance using Levene tests (Table S4). For this analysis, we used data from days 20 and 23 (pooled together because replicates were sampled destructively), as these were the final days when data were available for all treatments.
Equilibrium speciation-solubility calculations were conducted for day 23 of the microcosm experiments using the Spece8 module of Geochemists’ Workbench (GWB; Aquatic Solutions LLC, Champaign, IL, USA) and the wateq4f thermodynamic database (Ball & Nordstrom, 1991). The goal of the calculations was to assess the predicted speciation of Fe in the presence of OC under the environmental conditions of each microcosm treatment (following Oyewumi & Schreiber, 2017). Environmental conditions considered in this analysis included pH, DO, temperature, DOC, major cations (Ca, Na, K), Fe, and major anions (Cl, SO
4; bicarbonate was not measured so we calculated bicarbonate concentrations via charge balance). Data for major cations and anions were drawn from ICP-MS analyses (described in 2.3.1 Microcosm sampling). We assumed that DOC consisted primarily of humic acids (operationally defined within the wateq4f database) for the calculations.
3.1 Fe-OC comprised a substantial portion of sediment OC and a smaller proportion of OC in settling particulate matter
A substantial proportion of sediment OC was associated with Fe in both FCR and BVR. In FCR (averaged across 2019 and 2021), one gram of surficial sediment contained a mean of 481 µmol Fe-OC (±138, 1 SD), 31±8% of the total sediment OC pool (n=30). BVR had slightly lower Fe-OC than FCR on average, and one gram of surficial sediment contained a mean of 418±121 µmol Fe-OC, 24±7% of the total sediment OC pool (n=20). Total OC comprised 9±3% of sediment mass in FCR and 10±1% of sediment mass in BVR.
Levels of Fe-OC, both as a fraction of sediment mass and as a fraction of total sediment OC, were significantly higher in sediment core samples than in settling material collected in hypolimnetic traps (Figure 3). In 2021, averaged across both reservoirs, one gram of the hypolimnetic surficial sediments contained a mean of 443±133 µmol of Fe-OC (n=28), 69% higher than settling material collected in the traps, which contained a mean of 262±143 µmol Fe-OC (n=17; t32=-4.24, p<0.001; Figure 3a). A mean of 24±6% of the total sediment OC pool was bound to Fe in sediments (n=28), while only 9±4% of sediment OC was bound to Fe in settling material (n=17; t43=-10.44, p<0.001; Figure 3c). Total OC was 60% higher in settling material (µ=16.5±3.3) than in surficial sediments (µ=10.3±1.6; t20=7.33, p<0.001; Figure 3b).