Data from: Rainfall events stimulate episodic associative nitrogen fixation in switchgrass

Description

Notes

Methods

Site description

This study was conducted at the W.K. Kellogg Biological Station (KBS) Long-Term Ecological Research (LTER) Interactions site in Hickory Corners, MI (42°24'11" N, −85°22'26" W) from May through October 2019. Mean annual temperature at KBS is 10.1°C and average annual precipitation includes 1005 mm of rain and 1300 mm of snow (Robertson and Hamilton 2015). Soils developed on glacial outwash with intermixed loess (Sprunger and Robertson 2018) and are classified as mixed, mesic Typic Hapludalfs of co-mingled Kalamazoo and Oshtemo series (Crum and Collins 1995).

The Interactions Experiment, which was originally planted with continuous corn, was established in 1985 to examine interactions between tillage history (historically tilled or never tilled), tillage treatment (conventional tillage or no tillage) and fertilizer treatment (with and without N). In 2005, the 16 historically tilled plots (27 x 40 m) were switched over to a corn-soybean-wheat rotation, while maintaining their tillage and fertilizer treatments assigned 20 years earlier. Finally, all 16 plots were planted with switchgrass (Cave-in-rock variety) in the spring of 2016. Switchgrass is harvested annually after senescence with all aboveground biomass greater than 10 cm in height removed from the plots.

Study design

Our study used four switchgrass plots that had not been tilled or fertilized for over 30 years. Each plot was subdivided into four smaller subplots, including an ambient subplot (5 x 2 m) for phenological measurements and three subplots (5 x 4 m) for manipulation of rainfall frequency. Subplots were assigned to their location (i.e., NW, NE, SW, or SE) within the four plots using a stratified random sampling approach such that each subplot type never occupied the same corner in multiple plots.

To examine potential phenological effects, we measured N₂ fixation throughout the switchgrass growing season in the four ambient (unmanipulated rainfall) subplots. Because Roley et al. (2018) found that root-associated N₂ fixation rates were higher after switchgrass senescence than earlier in the season, we sampled more frequently during senescence. We measured N₂ fixation on almost a monthly basis until senescence began in mid-August, at which point sampling events took place every two weeks. More specifically, 2019 sampling occurred as follows – 1) May 1: pre-emergence, 2) May 27: tiller stage, 3) June 18: stem elongation, 4) July 22: flowering, 5) August 5: seed-set, 6) August 19: senescence, 7) September 2: senescence, 8) September 16: senescence, 9) September 30: senescence, and 10) October 15: senescence (see (Sanderson 1992) for definition of stages). For analysis, tiller and stem elongation were considered part of the vegetative stage, while flowering and seed-set were considered part of the reproductive stage; all other time points were considered either dormancy (pre-emergence sampling) or senescence (all others).

To investigate the effects of rainfall magnitude and frequency, we induced wetting events after various drying intervals using rainfall exclusion shelters coupled with irrigation systems and then measured N₂ fixation. During mid-June to mid-July 2019, we replicated three rainfall frequency treatments in four switchgrass subplots for a total of 12 subplots. All treatments received the same amount of total rainfall (~ 60 mm) over a 20-day period, which corresponds to the site's average weekly rainfall, but they received it at different frequencies. The constant moisture (CM) treatment received 15 mm every five days for a total of four wetting events, while the moderate drying (MD) treatment received 30 mm every ten days for a total of two wetting events and the extended drying (ED) treatment received 60 mm on day 20 (Figure 1). To capture the response of N₂ fixation to a wetting event, we sampled before and after subplots were irrigated, usually within eight hours of each other, but long enough after irrigation for the water to move into soil (Blazewicz et al. 2014). We also measured N₂ fixation in a subset of the plots after a drying interval of 5, 10, 15, and 20 days. Finally, we continued sampling the plots after the final wetting event (i.e., Day 22, 25, 30) to determine the effect of rainfall frequency treatment as the soils dried down.

Rainout shelters and irrigation

Twelve of the 16 subplots (three in each plot) were covered with a rainfall exclusion shelter, and rainwater was collected from gutters attached to each shelter. The shelter was constructed with a galvanized steel frame (4.9 x 4.3 x 1.8 m) and a roof of clear, corrugated polycarbonate (AmeriLux LEXAN, Wisconsin, USA), with a layer of UV protectant that also permits transmittance of 90% photosynthetically active radiation (see Figure 1 in (Hess et al. 2020) for a similar set-up). Each shelter gutter was connected to an irrigation tank for a total of two tanks per subplot. Rainwater was applied using a Rule 360 GPH/12 V bilge pump (Xylem Inc., Rye Brook, NY, USA), and the total amount was estimated visually to the nearest 20 L using the tank's volumetric markings. We determined when to stop the pumps based on visual assessment of the tanks, but we failed to disconnect a bilge pump on one instance, accidentally doubling the water added to a single CM subplot on Day 5. Nevertheless, flow rates (87 ± 14 L h⁻¹) were generally consistent across irrigation set-ups. Before each wetting event, water samples were collected from each irrigation tank with a 60-mL syringe and filtered with a 25-mm glass fiber filter (Pall Type A/E, pore size of 1 µm) for later measurement of dissolved inorganic nitrogen.

Soil sampling

For each sampling event, we collected soil for N₂ fixation and physicochemical measurements. For N₂ fixation, we used a hammer corer to collect two intact soil cores (3.6 cm in diameter, 10 cm in depth) per subplot within 20 cm of a switchgrass crown and within 10 cm of each other to minimize plant damage and spatial heterogeneity. For physicochemical measurements, we collected multiple short soil cores (2 cm in diameter, 10 cm in depth) in the vicinity of the cores with a push probe for a total of about 200 g of soil.

N₂ fixation

To measure N₂-fixation in intact cores, we conducted a three-day ¹⁵N₂ incorporation assay by adapting commercial stainless steel liners used for retrieval of hammer core samples. We made the cores gas-tight for the assay by topping each end of the liner with a rubber stopper and an open Luer-Lok fitting, sealing it with a waterproof adhesive (Amazing Goop®, All-Purpose), and allowing the adhesive to dry overnight in the laboratory. To begin the assay, we either vacuumed the headspace of the soil cores and then added artificial headspace (for the first two phenology sampling events) or purged the cores with the appropriate artificial headspace (for the rest of the phenology and all rainfall sampling events) before filling the core headspace to 3 psi using a pressure gauge. We verified that neither N₂ fixation nor carbon mineralization rates varied due to vacuuming versus purging the cores. All other N₂ fixation procedures remained the same across sampling events.

The paired cores (1 control, 1 enriched) from each sampling period were used to determine the N₂ fixation rate for each subplot. All cores received an artificial headspace of 78% N₂, 12% O₂, and 10% H₂. Enriched cores received half of their N₂ as 99% ¹⁵N¹⁵N (Cambridge Isotope Laboratories, Tewksbury, MA, USA) resulting in a ¹⁵N atom percent of about 50%. While contamination of ¹⁵N₂ can artificially inflate N₂ fixation rates (Dabundo et al. 2014), the commercial source we chose exhibits very little contamination (Roley et al. 2018), which we independently verified for our batch. Because we wanted to examine the effects of wetting on N₂ fixation, we used H₂ instead of more typical energy sources such as a glucose solution (Gupta et al. 2014) or a carbon cocktail (Smercina et al. 2019a).  Such carbon sources are typically added to soil in aqueous solution, which would introduce a wetting artifact, rendering such additions problematic given our focus on understanding the effects of rainfall treatments on N₂ fixation rates. H₂ is readily consumed by soil microbes, diazotrophs can use H₂ as a substrate and electron donor in N₂ fixation (Robson and Postgate 1980), and it is added as a gas so that it will not alter soil moisture. We ensured that H₂ was not depleted during the assay by adding 2 mL of pure H₂ after each day of incubation except on the final day of the assay. Our methods testing showed that we were able to detect N₂ fixation rates in cores receiving an artificial atmosphere with 10% H₂, but not those with 5% H₂. In addition, we observed that saturating cores with a 4% glucose solution enhanced N₂ fixation rates up to four orders of magnitude, indicating that glucose addition results in substantial growth of diazotrophs, dramatically inflating estimates of N₂ fixation. We found that H₂ is an energy source that raises N₂ fixation rates to a detectable level without altering soil moisture content or causing substantial growth during the period of measurement.

To calculate N₂ fixation rates, we first determined the ¹⁵N concentration of the core's headspace and that of the switchgrass root tissue and surrounding soil. Core headspace was sampled on the second day of the assay by removing 5 mL and injecting it into a 3-mL exetainer, which was stored upside down in water until isotopic analysis. The headspace volume was immediately replaced with the appropriate artificial headspace after sampling to maintain positive pressure in the core. After three days had passed, the assay was terminated by opening the Luer-Lok fittings and immediately sieving the soil using a 4-mm mesh. Roots and soil were separated and weighed. Then we allowed the roots and soil to air-dry, weighed them again, and ground them to a fine powder using a combination of scissors, a mortar and pestle, and a coffee grinder. We then packed the finely powdered samples into tin capsules for isotopic analysis (δ¹⁵N, %C, %N). Solid samples were analyzed by the GLBRC Stable Isotope Facility at Michigan State University on a Vario ISOTOPE cube Elemental Analyzer (Elementar, Langenselbold, Germany) coupled to an Isoprime visION Isotope Ratio Mass Spectrometer (Elementar), while the gas samples were analyzed on a HP 5890 Gas Chromatograph (Agilent, Santa Clara, CA, USA) coupled to a isoprime Dual Inlet-Isotope Ratio Mass Spectrometer (Elementar).

            Rates of N₂ fixation were calculated on a root or soil basis, in μg N d⁻¹, as (AE_i  x TN_i) / (AE_atm x t),

where AE_i is the ¹⁵N atom % excess in the roots or soil of the enriched core relative to the paired control core, TN_i is the total amount of N in the roots or soil of the enriched core, AE_atm is the ¹⁵N atom % excess in the headspace of the enriched core relative to the control core, and t is incubation time (Warembourg 1993). We then divided each rate by the dry mass of the roots or soil in the enriched core, resulting in root or soil N₂ fixation rates expressed as μg N g⁻¹ d⁻¹. To convert to an areal estimate (mg N m⁻² d⁻¹), we added the root and soil N₂ fixation rates for a core (μg N d⁻¹), divided it by the area of the core's top surface area (i.e., 10.2 cm²), and converted units. These areal rates are likely an underestimate because they assume an active depth of 10 cm (i.e., the depth of our core), but N₂ fixation can occur deeper in the soil profile, sometimes even at high rates (e.g., Casselman et al., 1981; Kreibich and Kern, 2003). For those samples (~23% of roots and 44% of soil) where N₂ fixation could not be detected, we assigned root and areal rates of 0.0001 μg N g⁻¹ d⁻¹ and 0.0001 mg N m⁻² d⁻¹, respectively, which were lower than any rates observed (root: 0.00013 μg N g⁻¹ d⁻¹; areal: 0.00023 mg N m⁻² d⁻¹).

Soil physicochemical properties

Following collection, the small soil cores from a single subplot were composited, sieved (4-mm mesh), and divided for measurement of gravimetric soil moisture, C mineralization, and soil inorganic N concentrations, all in triplicate. While we quantified gravimetric soil moisture and C mineralization for all soil samples, we assessed inorganic N concentrations only for the ambient subplots. Ammonium (NH₄⁺-N) and nitrate (NO₃⁻-N) concentrations were evaluated using a 24-hr 1 M KCl extraction of triplicate soil subsamples (Robertson et al., 1999). We determined gravimetric soil moisture by weighing triplicate subsamples before and after they were completely air-dried. Then we measured short-term C mineralization following the re-wetting of the air-dried soils as an indicator of soil C availability (Robertson et al. 1999; Franzluebbers et al. 2000).

To measure C mineralization, we placed triplicate soil subsamples (10 g of air-dry soil) into separate 235-mL mason jars equipped with gas-sampling septa and re-wet them to 50% water-filled pore space (Franzluebbers et al. 2000). Then we sampled 5 mL of headspace from each jar at five intervals (0, 2, 4, 8, and 24 hrs), put the gas sample in a 3-mL exetainer, and replaced the headspace with laboratory air. We analyzed the CO₂ samples within three days using a LI-820 CO₂ Gas Analyzer (LI-COR Biosciences, Lincoln, NE, USA). We accounted for headspace dilution, performed a linear regression of CO₂ concentrations through time, and converted the slope to a standardized C mineralization rate using the ideal gas law. Because wetter soil samples can lose more CO₂ while air-drying, which can affect the interpretation of this assay, we corrected the C mineralization rates for soil moisture as described in Belanger et al. (2021).

Soil C:N ratio was determined during isotopic analysis of core samples (see N₂ fixation, above); total C and N content of each sample was determined via the elemental analyzer and the C:N subsequently calculated by dividing the total C content by the total N content.

Statistical analyses

To identify the factors that influenced N₂ fixation throughout the switchgrass growing season, we performed mixed effect model selection after Zuur et al. (2009) on N₂ fixation rates in the ambient subplots. We individually analyzed root fixation rates and areal fixation rates (i.e., soil and roots combined). We did not analyze soil N₂ fixation rates separately because soil comprised most of the cores' mass, and the areal N₂ fixation rates were highly dependent on soil N₂ fixation rates (r² > 0.99). We log₁₀-transformed both areal and root N₂ fixation rates to better meet assumptions of normality and heteroscedasticity. During model selection, we assessed whether a spatial structure, or subplot as a random factor, should be added to the model, based on comparisons between models with and without those factors. Additionally, we included date, air temperature, the previous day's precipitation, the phenological stage of switchgrass, and soil moisture as fixed factors in the model selection for both response variables. For the areal rates, we also considered soil chemical factors in model selection (i.e., soil moisture, C:N, inorganic N, and C mineralization). Results did not vary when we used ammonium and nitrate as separate factors in our statistical analyses, so we used total soil inorganic N as an indicator of soil N availability. Total soil inorganic N was highly correlated with ammonium concentrations (r = 0.99; P < 0.0001) because ammonium (3.6 ± 1.1 μg N g⁻¹ dry soil) tended to be more than 10x higher than nitrate (0.26 ± 0.17 μg N g⁻¹ dry soil).

To determine whether wetting increased N₂ fixation, we compared both areal and root N₂ fixation rates in paired samples before and after we induced a wetting event in the rainfall exclusion shelter subplots. Because the response variables violated the assumptions of a paired t-test, we instead conducted two Wilcoxon signed-rank tests (H_a: difference > 0) on N₂ fixation rates before and after wetting events, one for areal and one for root N₂ fixation rates. Then we calculated the nonparametric effect size for both areal and root N₂ fixation using Vargha and Delaney's  (Peng and Chen 2014). According to (Vargha and Delaney 2000),  values of 0.56, 0.64, and 0.71 correspond to the small (0.2), medium (0.5), and large (0.8) values of Cohen's d.  

To identify factors that influenced the wetting response, we performed mixed effect model selection for the response of areal N₂ fixation to wetting. Specifically, we used the difference in N₂ fixation (i.e., after wetting − before wetting) as the response variable; no transformation was necessary. We included the amount of water and inorganic N added in an irrigation event and their interaction because they were correlated, the change in soil moisture, the number of times that a subplot had been wetted, and initial soil characteristics that might shape its response to wetting (i.e., soil moisture, C:N, and C mineralization) as fixed factors. Finally, we considered both day and subplot as potential random factors. We removed two outliers whose wetting responses were more than 3x the interquartile range because while their presence in the analysis did not change the factors in the best model, their inclusion made the model violate assumptions of normality.

To determine the effect of rainfall frequency on soils as they dried down, we conducted mixed effect model selection on the areal N₂ fixation rates from all sampling events following the final wetting. We log₁₀-transformed the response variable to meet assumptions of normality and we included rainfall frequency treatment, soil moisture, soil C:N, and C mineralization as potential fixed factors in model selection. We also considered date and subplot as potential random factors.

For all mixed effect model selection exercises, we quantified the fit of the best model. Specifically, we performed a linear regression of the observed N₂ fixation rates versus those predicted by the best model and calculated an r² value. If a random factor was included in the best model, we also performed a linear regression of the observed N₂ fixation rates versus those predicted by the model without any fixed factors (i.e., only random factor and variance structure, if applicable) and calculated an r² value for this model. These two analyses allowed us to compare the influence of the fixed versus the random factors. We also verified that the best model did not violate assumptions of a generalized linear model (GLM) or generalized least squares (GLS) regression.