Impacts of Wetland Plants on Microbial Community and Methane Metabolisms


 Aims

Microbial activity in the soil of wetlands is responsible for the emission of more methane to the atmosphere than all other natural sources combined. This microbial activity is heavily impacted by plant roots, which influence the microbial community by exuding organic compounds and by leaking oxygen into an otherwise anoxic environment. This study compared the microbial communities of planted and unplanted wetland soil from an Alaskan bog to elucidate how plant growth influences populations and metabolisms of methanogens and methanotrophs.
Methods

A common boreal wetland sedge, Carex aquatilis, was grown in the laboratory and DNA samples were sequenced from the rhizosphere, unplanted bulk soil, and a simulated rhizosphere with oxygen input but no organic carbon.
Results

The abundance of both methanogens and methanotrophs were positively correlated with methane emissions. Among the methanotrophs, both aerobic and anaerobic methane oxidizing microbes were more common in the rhizosphere of mature plants than in unplanted soil, while facultative methanotrophs capable of utilizing either methane or other molecules became relatively less common.
Conclusions

These trends indicate that roots create an environment which favors highly specialized microbial metabolisms over generalist approaches. One aspect of this specialized microbiome is the presence of both aerobic and anaerobic metabolisms, which indicates that oxygen is present but is a limiting resource controlling competition.


Introduction
Microbial activity in the soil of wetlands is responsible for the emission of more methane (CH 4 ) to the atmosphere than all other natural sources combined (Ciais et al. 2013). This ux is in uenced by many factors, but in all cases, the generation of CH 4 (methanogenesis) and any oxidation of CH 4 (methanotrophy), which may attenuate emissions, are microbially mediated. Therefore, when factors like temperature are cited as in uencing wetland CH 4 emissions (e.g., Hargreaves and Fowler 1998) they do so by impacting the microbial community either directly (e.g., microbial metabolic rates increase at warmer temperatures), or indirectly by altering other environmental factors, such as plants, which in turn affect the microbial ecosystem (Gill et al. 2017).
The microbial ecosystem inhabiting wetland soils is comprised of a complex mixture of bacteria and archaea that respond to a host of environmental variables. Community composition can vary greatly based on depth in the soil column (Lipson et al. 2013; Bai et al. 2018), geographic setting of the wetland (Grodnitskaya et al. 2018), and types of plants growing in the wetland (Robroek et al. 2015). The majority of microbial species present in wetland soil samples, as in most environments, are uncultured (Ivanova et al. 2016).
Plants impact the wetland microbial community through two primary modes. First, plants exude carbon compounds from their roots which may be more biodegradable than the other soil carbon (Bais et al. 2006; Girkin et al. 2018). These root exudates can stimulate microbial activity and CH 4  The second effect that wetland plants have on the microbial environment is leakage of oxygen into the soil from aerenchyma in their roots (Fritz et al. 2011). This oxygen can be used for methanotrophy (Fritz et al. 2011), but other aerobic metabolisms will compete for the limited oxygen supply (Lenzewski et al. 2018). Even when oxygen is used quickly enough that it does not accumulate in the soil (Waldo et al. 2019;Turner et al. 2020), it can in uence microbial communities by facilitating the recycling of alternate electron acceptors (Keiluweit et al. 2016), or by creating mixed-redox environments where carbon compounds are partially respired aerobically and partially anaerobically (Chanton et al. 2008). This variety of uses can lead to intense competition for oxygen in the rhizosphere. As with root exudation, oxygen transport changes over time as plants grow throughout the season, and different species of plants allow for varying amounts of oxygen transport (Schimel 1995). The balance between the dynamic effects of root exudation and oxygen transport will control what types of microbial CH 4 metabolisms are favored.
In addition to the traditional model of aerobic obligate methanotrophs, the rhizosphere also supports two other methanotrophic metabolisms. Once considered insigni cant in wetlands (Conrad 2009), recent work has shown that anaerobic oxidation of CH 4 (AOM) is common in freshwater wetlands (Segarra et al. 2015). Though it may be common, AOM is performed by a limited number of microbes, primarily the ANME2d anaerobic archaea (Haroon et al. 2013) and bacteria of the NC10 phylum (He et al. 2016). To avoid the use of oxygen, AOM relies on alternative terminal electron acceptors (TEAs). In freshwater bogs, rain is the primary source of water and nutrients; groundwater is not available to transport TEAs into the wetland. The continued availability of non-oxygen TEAs without transport into bogs can be explained by recycling and regeneration of the TEAs within the wetland (Keller and Bridgham 2007). This recycling requires an ultimate electron sink that is used to regenerate the TEAs used by anaerobic methanotrophs.
Plants can supply that electron sink by leaking oxygen from their roots which is used to generate a variety of TEAs in the relatively oxidized rhizosphere (Keiluweit et al. 2016).
The second non-traditional methanotrophic metabolism within the rhizosphere is facultative methanotrophy. Most methanotrophs are only capable of using single-carbon compounds (Conrad 2009 Leng et al. 2015). These facultative methanotrophs are widely distributed in the environment, but are especially prevalent in acidic soils, including peatlands (Rahman et al. 2011). Because the rhizosphere is a dynamic soil zone where the balance of microbial activity, root exudation, and oxygen availability may change over time, the ability to use different carbon sources for energy could be a competitive advantage.
Plants have great potential to in uence the environment for microbes, including both methanogens and methanotrophs. By doing so, plants impact the amount of CH 4 , a potent greenhouse gas, which is emitted from wetlands. However, plant effects are not uniform and can either increase ( Joabsson et al. 1999). Determining metabolisms fostered by the presence of roots can be used to build a mechanistic understanding of why some plant species increase while other decrease CH 4 emissions. In this study, we focused on Carex aquatilis, a common wetland sedge shown to increase methane emissions (Schimel 1995;Waldo et al. 2019). We compared the microbial communities of planted and unplanted wetland soil to elucidate how Carex growth in uenced populations of methanogens and methanotrophs, with special focus on the different forms of methanotrophy.

Experimental Setup
This investigation used samples collected during a previous study, Waldo et al. (2019), which described the experimental setup in detail. Brie y, Carex aquatilis, a common boreal wetland sedge, were grown for 10 weeks in rhizoboxes (48cm tall, 20cm wide, 5cm thick) lled with peat collected from a thermokarst bog in central Alaska. There were also two unplanted box types: control boxes with peat alone, and simulated plants that utilized silicone tubes to transport gases, thus simulating gaseous exchange without the biochemical effects of roots. There were 6 planted boxes, 2 control boxes, and 2 simulated plant boxes analyzed. Optical oxygen sensors (optodes) measured oxygen concentration around the roots of plants and around the simulated plant roots (Larsen et al. 2011). Methane emissions were monitored throughout the experiment by placing a clear uxing hood over each box and measuring the rate of CH 4 concentration increase in the headspace. During weeks 5 and 10 of the experiment, 4 plants were exposed to 13 CO 2 by placing a hood on each rhizobox and injecting 99 atom% 13 CO 2 into the headspace over a period of ve consecutive days. This 13 CO 2 was photosynthesized and isotopically labeled the plants. Following labeling, root and soil samples were collected under nitrogen. Plants were destructively sampled in both weeks 5 and 10; the unplanted control boxes and simulated plants were only sampled in week 10, at the end of the experiment. Samples collected for chemical analysis were documented in Waldo et al. (2021), and samples collected for DNA analysis and microbe counts are described below.
Soil samples were collected at depths of approximately 5 cm, 20 cm, and 35 cm. All samples were collected inside a gasbag lled with high-purity nitrogen. At each depth, samples were taken from three sites, one in the center and one 6 cm from either edge of the box. At each sample site separate samples were taken for uorescence microscopy and DNA sequencing. In planted boxes, roots and associated rhizosphere soil were collected. In control and simulated plant boxes, soil was collected.

Fluorescence Microscopy
Fluorescence microscopy was used to enumerate the microbes in samples from the rhizosphere and unplanted soil, but not in samples from boxes with simulated plants due to nite access to instrumentation. For planted boxes, root sections were cut from each sampling location. Root sections were sonicated in 4% paraformaldehyde (PFA). Soil dislodged from root samples was classi ed as rhizosphere soil (White et al. 2015), and was recovered by centrifugation (20 minutes at 15,000 g). The sample was then stored in a 50/50 mix of 70% ethanol and 1X phosphate buffered solution (PBS, Fisher Scienti c). For unplanted boxes, the protocol was the same, except the sample was not sonicated or centrifuged during PFA incubation. All samples were then stored at -20 C before being shipped on dry ice to the Environmental Molecular Sciences Laboratory (EMSL) where they were stored at -80 C until analysis.
For microbe counting, the samples were thawed and either the entire rhizosphere pellet (for plant samples) was used, or an aliquot of bulk soil (for control box samples) was taken that had similar volume to that of a typical rhizosphere pellet. To the soil sample, 0.3 to 0.4 g of sterile garnet beads were added with enough water to bring the total volume up to 1.5 mL. This mixture was then vortexed for 45 seconds. In a fresh tube, 98 µL of the mixture was combined with 2 µL of a 100X Vybrant Green DNA stain. One µL of the stained cell suspension was placed onto a slide and imaged with a 40X NA1.1 water immersion objective lens on a Zeiss LSM 710 inverted confocal uorescence microscope exciting the dye with a 488 nm laser and measuring uorescence in the 497-590 nm band. To count the microbes, the images were uploaded into ImageJ (Abramoff et al. 2004;Collins 2007) and the 3D Objects Counter function was used to classify uorescent objects between 0.5 µm 3 and 3.2 µm 3 as microbes. The combined mass of water and soil in each tube was measured, then the soil was dried overnight in an oven. These measurements were used to calculate the dry mass of soil per volume of water. Massnormalized cell density was calculated by dividing the total cell count by the mass of solids in the droplet which was imaged. Any sample which had less than 0.5 mg of soil in the 98 µL aliquot was excluded from analysis.

DNA Sequencing
For DNA sequencing, approximately 1 mL of soil was collected from each sample site for all three treatment types. DNA was extracted from the peat using a MoBio PowerSoil kit, with modi cations made to optimize the kit for extractions from peat soils (See Online Resource 1). A DNA quality check was conducted according to the Department of Energy Joint Genome Institute (JGI) "iTag Sample Ampli cation QC SOP" v. 1.3 (Online Resource 2). Brie y, an aliquot of the DNA was ampli ed using PCR; the PCR product was visualized on an agarose gel compared to size standards. DNA was stored at -20 C until transport to JGI for analysis. The DNA samples were shipped to JGI on dry ice. Once there, the samples were processed to produce one of two sequencing products: iTags or metagenomes.
The iTags classi ed microbes to the genus level using the V4 region of 16S rRNA sequences, using primers designed to amplify both bacteria and archaea (FW (515F): GTGCCAGCMGCCGCGGTAA, RV (805R): GGACTACHVGGGTWTCTAAT) (Rivers 2016). Sequencing and classi cation was done using an Illumina MiSeq instrument and the iTagger software (Tremblay et al. 2015). The methods summary produced by JGI is available as Online Resource 3.
The metagenomes were sequenced on an Illumina NovaSec instrument. The reads were trimmed and screened using the BBTools software (Bushnell 2015) and read corrected using BFC version R181 (Li 2015). The corrected reads were assembled and mapped using SPAdes assembler 3.

Statistical Analysis
All tests to determine whether multiple groups of data were or were not from the same distribution were done rst using a mixed-effects model (" tlme" in MATLAB R2018b) in which the box was a random variable, and the box type was the test variable. The mixed-effects model used only returns whether a difference between groups exists, not which groups are different. When a signi cant difference existed in the data, the Kruskal-Wallis test was used to determine between which groups the difference existed, performed using the "kruskalwallis" function in MATLAB (R2018b). All tests for relationships or trends within a dataset were done using a Spearman Rank Correlation Coe cient with the "corr" function in MATLAB (R2018b). The Spearman Rank Correlation returns both a p-value, indicating statistical signi cance, and ρ, indicating direction and strength of monotonic correlation.

Sequence Data Analysis
The iTag data was analyzed for the frequency of methanogens and methanotrophs. For methanogens, the classes Methanobacteria and Methanomicrobia were included. For obligate methanotrophs, all members of the family Methylocystaceae, as well as the entire order Methylococcales were included. The iTag data did not include su cient detail to differentiate facultative methanotrophs of the genera Methylocapsa and Methylocella from other members of their family, and so metagenomic data was used for facultative methanotroph analysis. The genus Methylocystis was also counted as facultative methanotrophs. Similarly, the iTag data did not identify any taxa that are documented to perform AOM, so the metagenomic data were used to isolate the candidate genus Candidatus Methanoperedens, which contains ANME2d anaerobic methanotrophs (Haroon et al. 2013). Bacteria of the NC10 phylum also perform AOM but were not identi ed in the metagenomic phylogeny through IMG. Instead, NC10 presence was determined through a BLAST search for sequences from the GenBank database of the National Speci c gene sets found in the metagenomes were used to assess functional differences in microbial populations. To determine whether samples had microbes with aerobic or anaerobic metabolisms present, the number of genes involved in glycolysis (a process which occurs in both anaerobic metabolism) was compared to genes involved in the Krebs cycle (aerobic metabolism). Because glycolysis is also used by aerobes, the glycolysis to Krebs ratio is not equal to the ratio of anaerobes to aerobes. However, there will be a qualitative correlation between the two ratios. For the Krebs cycle, only those genes involved in the rst oxidation were used because that limited the number of genes involved and focused the results. To compare methanotrophic metabolisms, methane monooxygenase (MMO) genes were compared. In addition to number of genes, a principle components analysis (PCA) was performed on the MMO gene sets to determine if different types of MMO were used in different samples.
PCA was performed in MATLAB (2018b) using the "pca" function and default settings.
The gene sets were identi ed through the KEGG Orthology (Kanehisa and Goto 2000; Kanehisa et al. 2016). The gene sets used for the Krebs Cycle are presented in Table 1, gene sets for glycolysis are in Table 2, and gene sets used for MMO are in Table 3.

Fluorescence Microscopy
Ten weeks after the start of the experiment, rhizosphere soil samples had a signi cantly (p < 0.05) higher concentration of microbes than did the unplanted control box samples (Fig. 1A). The rhizosphere soil collected during week 5 of the experiment did not have a signi cantly different number of microbes from rhizosphere soil collected in week 10 or from the control box soil. The comparison of the three groups indicates that roots encouraged microbial growth, but that it took time for the increased growth to take effect. However, there was not a statistically signi cant correlation between microbe count and CH 4 ux ( Fig. 1B,  The median percentage of microbes that were methanogens in samples from each box was positively correlated with CH 4 emissions ( Fig. 2A, p < 0.05, ρ = 0.69) as was the percentage of microbes that were methanotrophs, when excluding simulated plants (Fig. 2B, p < 0.01, ρ = 0.78). Simulated plants were excluded from the correlation test of methanotrophs because in the other three box types (planted boxes from weeks 5 and 10 and control boxes) the oxygen concentrations were low, but in simulated boxes, the oxygen concentrations were higher (Waldo et al. 2019) so the microbes faced a fundamentally different environment. Correlating methanotrophs with CH 4 emissions acts as a proxy for correlating methanotrophs with CH 4 availability in the rhizosphere.
When microbe count data was used with the percentages to nd the total number of each type of microbe, there was a positive correlation between CH 4 ux and methanotroph count (p < 0.01, ρ = 0.87, Fig. 2D), but the correlation with methanogen count was on the edge of signi cance (p = 0.07, ρ = .65, Fig. 2C). The number of methanogens and methanotrophs were also signi cantly correlated with each other (p < 0.05, ρ = .31). Microbe count data was not available for all samples that were sequenced, so the number of replicates was smaller in the count analysis, and no microbe counts were conducted on samples from simulated plant boxes.

Metagenomes
The metagenomic data were used to identify functional genes and taxa which could not be identi ed in the iTag data. Facultative methanotrophs comprised less than 1% of all samples (Fig. 3A). In contrast to the obligate methanotrophs (Fig. 2), there was no statistically signi cant (p > 0.05) correlation between the ux of CH 4 in the nal week before harvest and either the percentage of facultative methanotrophs ( Fig. 3A, with simulated boxes ρ=-0.16 or excluding simulated boxes ρ = 0.10) or the number of facultative methanotrophs (Fig. 3B, ρ = 0.62). However, the percentage of microbes that were facultative methanotrophs in simulated plant boxes was greater (p < 0.05 by mixed-effects model and Kruskal-Wallis) than the other box types (Fig. 3A), as was observed in obligate methanotrophs (Fig. 2B, p < 0.01).
The ratio of obligate to facultative methanotrophs was signi cantly larger (p < 0.05) in rhizosphere samples from week 10 than in simulated boxes, while the other two treatment types (control boxes and rhizosphere samples from week 5) had intermediate ratios that were not signi cantly different (p > 0.05) from the ratios in any other treatment (Fig. 4A). There was no signi cant correlation (p > 0.05) between the ratio of obligate to facultative methanotrophs and the ux of CH 4 in the nal week before harvest (Fig. 4B).
The ratio of ANME2d archaea, which are capable of AOM, to total obligate methanotrophs was signi cantly larger (p < 0.05) in rhizosphere soil from week 10 than in the simulated plant boxes, with rhizosphere soil from week 5 and control boxes having an intermediate ratio (Fig. 5A) -as was seen with the ratio of facultative to obligate methanotrophs. There was no signi cant relationship (p > 0.05, ρ = .16) between the ratio of ANME2d to total methanotrophs and the ux of CH 4 in the nal week before harvest (Fig. 5B). The BLAST searches did not return any matches for the NC10 pmoA genes and the NC10 16s sequences returned did not display any statistically signi cant relationships with other relevant data (data not shown). The lack of pmoA gene detections, even at low match quality, indicates that the 16s sequences may not be derived from NC10 bacteria. For this reason, the NC10 BLAST results were omitted from further analyses and all discussion of AOM are related to the ANME2d results.
The ratio of genes involved in glycolysis to those involved in the Krebs Cycle was positively correlated with CH 4 emissions (p < 0.05, ρ = .72, Fig. 6 The positive relationship between obligate methanotrophs and CH 4 emissions (Fig. 2B&D) tells us more about the system. Obligate methanotrophs rely on both CH 4 and TEAs to function. Assuming CH 4 emissions are a good proxy for CH 4 availability, the positive correlation indicates that the obligate methanotroph population responded directly to methane availability. The second resource that methanotrophs need, TEAs, are harder to directly measure, but this study has two lines of evidence that they were a limiting factor in the rhizosphere. First, optical oxygen measurements from the experiment from which these sample were obtained (Waldo et al. 2019) indicated that soil within planted boxes at both time points and within control boxes lacked standing pools of oxygen (Waldo et al. 2019). Second, the ratio of glycolysis to Krebs Cycle genes from the metagenomic data (Fig. 6) indicate the boxes producing the most CH 4 also had a potentially greater prevalence of anaerobes in the microbial community. These data cannot be used de nitively because glycolysis is used by both aerobic and anaerobic metabolisms, and anaerobic metabolisms exist that do not use it. However, lacking a more direct measurement of total aerobic versus anaerobic activity it can be used to qualitatively rank samples by relative abundance of anaerobic activity. The increased ratio of glycolysis to Krebs Cycle genes in boxes with high CH 4 emissions implies that when CH 4  the extra oxygen allowed methanotrophs to thrive at lower CH 4 concentrations or allowed them to oxidize a higher portion of the CH 4 produced. The methane monooxygenase (MMO) analysis showed that the microbes in simulated plant boxes were using more particulate methane oxygenase (PMO) while the rhizosphere microbes in planted boxes were using a soluble MMO to conduct methanotrophy. The reason why PMO would be preferable to MMO in a setting with more oxygen and no root exudates is not immediately clear, but it is further evidence that simulated and real plants had important differences in the environment they created for methanotrophs.
The apparent success of methanotrophs in low-oxygen environments has two potential explanations. Facultative methanotrophs, however, displayed a different pattern of abundance than the obligate methanotrophs (Figs. 2-4). Both types of methanotrophs were most common in the simulated-plant boxes where oxygen was most abundant. However, while obligate methanotrophs had a signi cant positive (p < 0.05) relationship with CH 4 emissions whether measured by percentage of genes (Fig. 2B) or by number of microbes (Fig. 2D) in the planted and control boxes, the facultative methanotrophs' correlation was not signi cant (p > 0.05) by either percentage or number of cells (Fig. 3). The ability of obligate methanotrophs to increase in abundance with CH 4 availability while facultative methanotrophs cannot implies that in low-oxygen environments with high rates of metabolic activity, such as the rhizosphere examined in this study (Waldo et al. 2019), obligate methanotrophs were able to out-compete facultative methanotrophs. This outcome is reinforced by the observation that the obligate to facultative ratio was signi cantly higher in the rhizosphere from the end of the experiment than it was in unplanted control soil (Fig. 4A). Because CH 4 is generated in anoxic environments, obligate methanotrophs could gain a strong advantage over their facultative competitors if they are able to conduct methanotrophy when oxygen concentrations are limiting. It has been hypothesized that obligate methanotrophs exist because their extreme specialization gives them a competitive advantage over more generalist microbes

Conclusions
The most direct measures of the methanogenic potential of the microbial community behaved as expected: methanogens were positively correlated with CH 4 emissions (Fig. 2A&C) and were most common in the rhizosphere, genetic indicators of oxygen limitation were highest in the boxes with the highest CH 4 emissions (Fig. 6), and microbial populations were largest in number when the most root exudates were available (Fig. 1, Waldo et al. 2019).
Obligate and facultative methanotrophs responded unevenly to the experimental conditions, demonstrating differing metabolic strategies. Both types of methanotrophy were most abundant around the simulated plants where oxygen was abundant; however, in the rhizosphere and control box soil where oxygen was limited, obligate methanotroph abundance was correlated with CH 4 availability (Fig. 2B&D), while facultative methanotroph abundance was not (Fig. 6). This nding implies that in low-oxygen, high CH 4 environments, the highly specialized obligate methanotrophs were able to out-compete the more generalist facultative methanotrophs through either an increased a nity for oxygen or a greater ability to perform AOM.
The net effect of these various impacts is that the Carex plants studied here greatly increased methanogen abundance, and therefore likely methanogenesis, but also increased methanotroph abundance, and likely methanotrophy. The rhizosphere became a region of intense competition for oxygen, implying that in the rhizosphere of a plant species with a higher rate of oxygen transport through aerenchyma the methanotroph abundance, and likely related rate of methanotrophy, could increase correspondingly.  Methane emissions the week prior to harvest compared to soil microbe counts. Each datapoint is one box, error bars are upper and lower quartiles determined from multiple samples measured from each box.

Figure 2
CH4 ux versus (A) methanogen relative abundance, which had a correlation (p < 0.05, ρ=0.69), (B) methanotroph relative abundance, which had a correlation (p<.01) when excluding simulated plant boxes, (C) microbe count of methanogens, which had a correlation on the edge of signi cance (p = 0.066, ρ=.65), and (D) microbe count of methanotrophs, which had a correlation (p<0.01). Each datapoint is one box median, error bars are upper and lower quartiles determined from multiple samples measured from each box. Each ux value is calculated from 1-7 uxes (mean 4.2). Each percentage value is calculated from 3-7 samples (mean 4.8) and each count is calculated from a combination of that sample's percentage and 1-3 total microbe counts (mean 1.8).