Soils were collected in October 2015 from long-term crop rotation experiments at USDA-ARS sites in Akron, Colorado (CO); Beltsville, Maryland (MD); and Brookings, South Dakota (SD) and at the W.K. Kellogg Biological Station (KBS) Long-Term Ecological Research Site (LTER), Michigan (MI). All sites maintain field experiments that include treatments with different rotation lengths. Composite topsoil samples from within the first 10 cm were collected from three (Colorado) to four (Maryland, South Dakota, Michigan) field plot replicates in simple (two crops in rotation) and complex rotational treatments (three or more) each. Soils from Maryland, South Dakota, and Michigan were sampled under corn and those from Colorado were sampled under wheat. Site descriptions can be found in Table 1 and in Cavigelli et al. (2008), Lehman et al. (2017), Tiemann et al. (2015), and White et al. (2019).

After sampling, soils were sieved and shipped on ice to the University of New Hampshire and refrigerated at 5∘ for less than 1 week. Approximately 30 g soil from each replicate plot (three from Colorado and four from each of the other sites) were weighed into 100 mL plastic cups resulting in a total of 27 cups per replicate from Colorado and Maryland and six for South Dakota and Michigan. Soils in the microcosms were adjusted to 50 % water holding capacity (WHC). One set of cups was covered with Parafilm and kept at constant water content by replacing evaporated water once a week and after every CO2 measurement over the course of 165 d. One set of cups was subjected to drought and another to flooding (Fig. 1). All soil microcosms were kept at a constant temperature of 25∘. Microcosms for the drought treatment were allowed to gradually dry out over the course of 3 d, kept at peak drought for 4 d, and slowly brought back to 50 % WHC by adding one third of the evaporated water every day for 3 d to avoid even short time flooding effects. Microcosms for the flooding treatment were gradually brought to 100 % WHC but not higher to avoid submerging the soils in water over the course of 3 d, then kept at 100 % WHC for 4 d, and then kept open to dry back to 50 % WHC again within 3 d. Drought and wetting were repeated after 2 weeks of the soils being held at constant WHC. Soils were subjected to a total of five stress cycles during the first 125 d of the total 165 d incubation period. Soils from all sites, rotations, and water treatments were set up twice. To determine long-term recovery, one set was subjected to only one stress cycle and was kept at 50 % WHC for 6 weeks after the stress. The second set was subjected to a total of five stress cycles. For each combination of rotation complexity, water treatment, and harvest we had three replicates from Colorado and four from Maryland, South Dakota and Michigan.

To determine soil carbon and nitrogen pools and microbial enzyme activities in soils from Colorado and Maryland with the lowest and highest MAP, respectively, we set up 10 sets of each of the control, drought, and flooding treatments for each of the sites. One set from each site was destructively harvested before, at the peak of, right after, 2 weeks after, and 6 weeks after the first and last stress cycle (Fig. 1).

Samples for total carbon and nitrogen analysis were dried at 60∘ for 24 h and finely ground in a ball mill before subsamples were packed in tin capsules and total carbon measured on an elemental analyzer (Costech Instruments ECS 4010) (Paul et al., 2001). Total carbon content as well as water holding capacity were determined for all soils prior to the incubation experiment. Water holding capacity was measured by determining soil water content after saturating the soils with water in a funnel with filter and letting the excess water leach gravimetrically for 2 d while preventing evaporation by covering the funnels with Parafilm (Paul et al., 2001). Water content and pH were determined for all soils before the start of the incubation and during the 10 destructive samplings for Colorado and Maryland soils. Water content was determined gravimetrically in sample aliquots that were dried in a forced draft oven at 60∘ for 24 h. Soil pH was determined in a 1 : 5 soil to water mixture using a Mettler Toledo SevenEasy pH Meter 20.

CO2 production was measured daily for the first week, twice a week after that, and just before the destructive harvests. For respiration measurements, cups were temporarily closed airtight with lids fitted with rubber septa that served as a sampling ports. We took 3 mL of the headspace using a syringe immediately after closing the sample and after 30 min to 2 h, depending on the incubation duration within the experiment and the carbon content of the samples; longer incubation times were used at the end of the experiment and for soils with lower total carbon contents. The gas samples were immediately injected into an infrared gas analyzer (LI-COR LI 820) to measure CO2 concentration. Rates of CO2 production were calculated from the increase of CO2 concentration in the headspace of the jar over time, accounting for the jar and syringe volume and temperature and assuming a linear increase between the two sampling time points. Cumulative respiration was calculated by using respiration rates measured at a certain time point and multiplying that flux by the number of days to the next respiration measurement and summing all resulting CO2 emissions (Grandy and Robertson, 2007). To compare cumulative respiration among sites, values are expressed as µg CO2-C per gram soil carbon.

For the determination of N2O production, which was measured at every destructive harvest, cups were inserted in pint-sized mason jars and sealed airtight with a lid fitted with a rubber septum. Right after closing the jars, a headspace sample of 30 mL was taken with a syringe and needle and transferred into pre-evacuated Exetainers. The jars then remained closed for 24 h before a second sample was taken and transferred to Exetainers. N2O concentration in the Exetainers was determined using a Shimadzu GC-2014 equipped with an ECD detector. N2O flux was calculated as the difference in N2O concentration between samples collected right after sealing and after 24 h divided by the time of incubation and the amount of dry soil in the cup and accounting for jar and syringe volume and temperature.

Extractable organic carbon and total extractable nitrogen were measured in 1 M KCl extracts (15 mL) from approximately 2 g of soil using a TOC-L CPH/CPN analyzer (Shimadzu). Ammonium and nitrate concentrations were measured in the same extracts by colorimetric assays as described by Hood-Nowotny et al. (2010). Microbial biomass carbon was determined using chloroform-fumigation extraction (Brookes et al., 1985; Vance et al., 1987). Two grams of fresh soil were fumigated in a desiccator under chloroform atmosphere for 24 h in the dark and then extracted with 1 M KCl. Extracts of fumigated samples were measured on the TOC-L CPH/CPN analyzer and microbial carbon was calculated as the difference in EOC between the fumigated and the fresh soil extracts. Microbial carbon is presented without the use of a correction factor for extraction efficiency. EOC, TEN, NH4, NO3, and MBC were determined at every destructive harvest.

Potential extracellular enzyme activities were measured, with adaptations, as described in Schnecker et al. (2015). In short, 2 g of soil were suspended and homogenized in 100 mL of 100 mM sodium acetate buffer at pH 5.5. For each sample and each enzyme, five wells of a black microtiter plate were filled with 200 µL of the soil slurry. The respective wells were amended substrates labeled with MUF (4-methylumbelliferyl): β-D-glucopyranoside for β-glucosidase (BG), β-D-cellobioside for cellobiohydrolase (CBH), and N-acetyl-β-D-glucosaminide for N-acetyl-glucosaminidase (NAG). L-Leucine-7-amido-4-methyl coumarin was used as substrate for leucine-amino-peptidase (LAP). Plates for the assays of BG, CBH, NAG, and LAP were incubated for 140 min. Afterwards, activity was measured fluorimetrically (excitation 365 nm and emission 450 nm). Phenoloxidase (POX) and peroxidase (PEX) activities were measured using L-3,4-dihydroxyphenylalanine (DOPA) as a substrate in a photometric assay. Three 1 mL samples of the original soil slurry was mixed with 1 mL of a 20 mM DOPA solution. After shaking and centrifuging, two wells of each transparent microtiter plate were filled with 250 µL of the supernatant. One of these wells additionally received 10 µL H2O2 (0.3 %) for determination of peroxidase activity. Plates for oxidative enzyme activities were measured photometrically (absorbance 450 nm) at the beginning and after incubation for 20 h. PEX activity was calculated as the difference in the increase in color during the incubation time between the wells with and without H2O2 addition. All other enzyme activities were calculated as the increase in color or fluorescence during the incubation time. Potential enzyme activities were determined at every destructive harvest.

To quantify labile soil carbon we used the permanganate oxidizable carbon (POXC) method (Weil et al., 2003) as described in Culman et al. (2012). In short, 2.5 g of air-dried soil were mixed with 18 mL of deionized water and 2 mL of 0.2 M KMnO4 stock solution and shaken for 2 min at 240 oscillations per minute on an oscillating shaker. Tubes were removed from the shaker and allowed to settle for 10 min. After 10 min, 0.5 mL of the supernatant was mixed with 49.5 mL of deionized water. An aliquot (200 µL) of each sample was loaded into a 96-well plate containing a set of internal standards, a soil standard, and a solution standard (laboratory reference samples). Sample absorbance was read with a spectrometer at 550 nm. POXC was determined at every destructive harvest for Colorado and Maryland soils.

To evaluate the effect of the specific stress treatments, we calculated response ratios of all variables measured during destructive harvests as the values for the treated samples divided by the values for the control samples under constant water conditions. To evaluate differences between the control and treatment samples, we performed two-sample comparison tests (t-test, Welch's t-test or Mann–Whitney U test as appropriate for each variable's normality and homogeneity of variance). We further used all data measured at destructive harvests, with the exception of water content, individually for each harvest date and including all water treatments at peak stress and right after the stress during the first and last stress cycles (days 6, 14, 120, and 127) in principal components analysis (PCA). We used one-way ANOVA and Tukey HSD as a post hoc test on the first two axes of the PCAs to evaluate differences among water treatments, crop rotation complexity treatments, and their interaction. Before analysis, data were log-transformed or rank-normalized to meet the assumptions for ANOVA. Differences and correlations were assumed to be significant at p<0.05. Statistics were performed in R 3.3.2 (R Development Core Team, 2013).

The sites used in this study represent a gradient in MAP from an arid system in Colorado (MAP 421 mm) to a site with relatively high MAP in Maryland (1192 mm), with South Dakota (580 mm) and Michigan (892 mm) providing intermediate MAP. Soil organic carbon (SOC) content varied greatly between sites, being highest in South Dakota, followed by Maryland, Colorado, and Michigan (Table 1). Significant differences in SOC content between rotation lengths could only be found in Maryland where soils from the complex rotation had an average of 1.3 % OC and soils from the simple rotation length had 1.0 % OC; notably, this was also the only site that included a perennial crop in the complex rotation.

Respiration decreased significantly in response to drought in soils from all sites and crop rotation complexities and returned to control levels as soon as microcosms were returned to 50 % WHC after the first (day 1–15) and fifth (day 113–127) stress cycle (Fig. 2). In some cases (Maryland first and fifth stress, South Dakota fifth stress, and Michigan fifth stress) respiration in rewetted microcosms exceeded respiration in control microcosms. Soils from simple and complex rotations did not differ in the response to drought, but in some cases differed in their recovery from the stress, most notably in soils from Maryland and South Dakota where soils from simple rotations showed lower respiration rates. In general, soils experienced slight but mostly not significant increases in respiration in response to flooding. In the Colorado simple rotation, the Maryland simple rotation, the Michigan complex rotation, and both South Dakota rotations, flooding significantly increased respiration only on day 4 and thereafter was indistinguishable from the control until the end of the first stress. Soils from the complex rotation in South Dakota further showed an increase in respiration during the recovery from flooding after the fifth stress.

We measured cumulative respiration to estimate soil carbon loss. Respiration was highest in soils from Colorado, followed by those from Maryland, Michigan, and South Dakota (Fig. 3). During the first stress cycle, soils from all sites lost significantly less CO2 under drought compared to control and flooding treatments (Fig. 3b), but CO2 did not differ between flooded and control soils at any site for a given rotation treatment. During the fifth stress cycle, CO2 losses were significantly lower with drought compared to control and flooding in soils from Colorado and South Dakota under both rotation regimes, and drought-stressed soils from the Maryland complex rotation lost significantly less CO2 than flooded microcosms. Total loss of carbon as CO2 (calculated per gram of SOC) over the entire incubation period tended to be lower in microcosms experiencing repeated drought compared to the control and flooded microcosms (Fig. 3a). However, drought-stressed soils under complex rotation in Colorado lost significantly less carbon as CO2 than the control and flooded soils, and soils from simple rotations in South Dakota lost more carbon when flooded compared to drought; control soils were not significantly different with either stress treatment.

Flooding and drought caused significant changes in soil nitrogen and carbon pools, microbial biomass, microbial enzyme activities, and nitrous oxide production. Except for experimentally manipulated water content and a decrease in respiration during drought, no variables changed consistently and significantly among soils from Colorado and Maryland (Tables 3 and 4). In general, drought tended to decrease measured parameters, while flooding increased them. Microbial biomass carbon increased during the first flooding event in soils from complex rotations in Maryland and strongly decreased during the fifth drought cycle in all Maryland soils. Microbial biomass in soils from Colorado was not affected by the treatments. In Colorado, all enzyme activities decreased with drought while NAG alone increased with flooding. This was not the case in Maryland soils, where enzyme activities remained constant through water stress or changed only after the stress was over. Only a few parameters changed similarly during the first and fifth stress periods. LAP decreased during the first and fifth drought stresses in the simple Colorado rotation treatment. Production of N2O increased in response to the first and fifth flooding in the Maryland complex rotation soil; in this rotation treatment NAG decreased after both the first and fifth drought. All other parameters affected by the stress treatment changed only during either the first or fifth stress in single site-rotation combinations at peak stress or following the stress.

Principal component analyses caused samples from flooding and drought treatment to cluster in to separate groups (PCAs; Fig. 4). Samples clustered similarly and treatment differences on individual PCA axes were significant (Table 2) during the first (Fig. 4) and fifth stress (Fig. 5). In both cases these differences were no longer significant after the end of drought and flooding, when microcosms returned to 50 % WHC (Figs. 4 and 5; Table 2). In the case of the Maryland samples, significant differences could be found between simple and complex rotations during and after the first and fifth stress periods. Differences between rotations in Colorado were only found after the fifth stress period.

All soils had significantly decreased respiration rates in response to drought during the first stress cycle with the strongest relative decreases at the Colorado site. Compared to drought, flooding had a smaller effect on microbial respiration. In general, flooding slightly increased respiration rates, but this was only significant during the first stress cycle for 1 to 2 d and was not consistent across rotation treatments. Flooding may increase respiration rates in part due to increased connectivity and availability of previously untapped DOC sources to microorganisms (Schimel, 2018 and therein). Cumulative carbon losses were significantly different among sites. Surprisingly, despite having the highest SOC contents, soils from South Dakota lost the least carbon as CO2 over the whole incubation period. Differences in cumulative respiration between sites could be related to differences in mineralogy and soil texture (Saidy et al., 2015; Schmidt et al., 2011), microbial community composition (Babin et al., 2013), or the chemistry of plant and fertilizer inputs (McDaniel and Grandy, 2016). Our data hint at physical soil properties as potentially relevant factors for SOC stabilization and susceptibility to water stress as our four sites not only represent a precipitation gradient but also a gradient in texture, which is reflected to some extent in SOC contents but less so in the observed cumulative carbon loss.

Aside from CO2 production during drought, no factor investigated in the two soils from climate extremes (Colorado and Maryland) changed consistently in response to drought and flooding (Table 3). Microbial biomass was affected during the first flooding cycle and fifth drought cycle in Maryland soils but not in soils from Colorado. In Colorado, for soils of both rotation lengths, LAP decreased with drought, which was not the case in soils from Maryland. All flooded Colorado soils produced N2O, while this was only the case in the complex rotation in Maryland. Beside these site-specific effects, we could not find drought or flooding effects that occurred in either rotation treatment at both sites. Extracellular enzyme activities in particular remained relatively unaffected by the applied stresses. A reason for this might be the stabilization of enzymes on soil minerals, which might protect them against drought and flooding (Allison and Jastrow, 2006; Kramer et al., 2013). This stabilization might also explain why our findings are in contrast to results from temperature stress experiments in plant litter (Mooshammer et al., 2017).

However, when soil samples were ordinated in response parameter space in a PCA, a clear stress treatment effect emerged in Colorado and Maryland soils irrespective of the rotation treatment (Tables 2, 3, and 4). Maryland but not Colorado soils showed an additional rotation effect: samples from simple and complex rotations at this site separated in the PCA, both during and after drought and flooding events. This was most likely related to higher SOC content and co-varying soil properties in the complex crop rotation soils, which we found only in Maryland. The experimental field in Maryland had the most complex crop rotation (four crops) and was the only site where the complex rotation included a perennial crop and fertilization with poultry litter, both effective methods for increasing soil carbon stocks and soil health (Ashworth et al., 2018; King and Blesh, 2018). The introduction of cover crops might be more effective than increasing cropping diversity alone (McDaniel et al., 2014).

One reason for the apparently mild reaction to drought and flooding in this experiment might be the duration of the stress we applied. Stress slowly applied over several days and lasting less than 2 weeks in total might be similar to conditions that microorganisms in the investigated soils experience frequently in the field. Stress effects have indeed been found to strongly vary with the duration of stress as well as its intensity (Tiemann and Billings, 2011, 2012).

In our experiment, only soils from Maryland showed a stress-induced increase in respiration during recovery compared to unstressed control soils; specifically, respiration increased in soils from the complex Maryland rotation in the first day following stress and in the simple rotation after 3 d. The lack of a Birch effect in all other soils might be because we rewetted the soil in small increments over the course of 3 d rather than a flush rewetting, which often results in a large increase in respiration (Birch, 1958; Göransson et al., 2013). It might also be the case that the duration of desiccation was too short to lead to a pronounced CO2 pulse at rewetting (Unger et al., 2010).

Like respiration rates, most other measured parameters returned to control levels following stress; this was also apparent in the PCA where dried and flooded soils were indistinguishable from control soils at a constant water content of 50 % WHC. This indicates that while all soils were significantly affected by the stress treatment, they are highly resilient and recovered quickly from stress. This is in accordance with Kaurin et al. (2018) who found that microbial communities in agricultural soils recovered after rewetting even after severe and prolonged drought periods and Barnard et al. (2013) who found similarly quick recovery of the soil microbial community after rewetting of dry grassland soils.

As during the first stress period, respiration rates declined in soils from all sites and rotations during the fifth drought cycle. In contrast, respiration was unaffected by flooding, with the exception of a slight increase in the Michigan complex rotation 5 d after flooding. After the end of the fifth drought cycle, in contrast to the first cycle, we found clear increases in respiration when soils from Maryland, South Dakota, and Michigan returned to 50 % WHC. Both observations might be related to the absence of plant inputs in our incubation experiment. Plants have been shown to strongly respond to drought (Fuchslueger et al., 2014; Kaisermann et al., 2017), but will provide some amount of carbon even if rhizodeposition is reduced under drought (Canarini and Dijkstra, 2015). During the vegetation period, such a continuous supply of root exudates might prevent an increase in respiration at the end of a stress event when connectivity between microbes and substrate is re-established, such as our observation after the fifth but not the first stress. Management practices that extend the vegetation period and minimize fallow periods might help maintain a constant supply of DOC to soil microorganisms and thereby buffer their response to drought and flooding.

We also found site-specific differences between the first and the fifth stress cycles. In soils from Colorado from both rotation regimes, NAG was significantly reduced in the drought treatment compared to the control. In contrast, drought reduced total extractable nitrogen in Maryland soils while flooding released N2O in both Maryland rotations. In Colorado soils no N2O production could be detected by the fifth stress cycle. This might have been caused by a depletion of the dissolved substrate for N2O production or could be related to plant-induced differences in microbial community composition (Hammerl et al., 2019) that had faded after 160 d of incubation.

The lack of microbial adaptation to the re-occurring stress in all except the Colorado soils might be interpreted as an already existing adaptation to conditions mimicked in our experiment or could again be ascribed to the modest stress events of our lab experiment compared to the larger environmental fluctuations these soils experienced in situ.

In this study we found that drought – and, to a lesser extent, short-term flooding – significantly affected respiration rates at all sites and at some sites had additional effects on some microbial enzyme activities, soil carbon and nitrogen pools, and nitrous oxide emissions. Furthermore, an increase in crop rotational diversity did not lead to generally different responses of soils to short-term drought and flooding, even at the one site where increased crop rotation complexity also caused an increase in SOC. While soil function reacted significantly during stress events, all soils recovered quickly and returned to control levels once the stress ended. This indicates that soil microbial processes in these agricultural soils collected from variable climate regions within the United States are highly resilient to short-term drought and flooding. Future experiments should also include plants as they are an important component of agroecosystems in the field and could strongly influence DOC and DON availability. Our laboratory study focused on the soil–microbe system and showed that, at least in the absence of plants, microbial functions and activities are highly resilient to drought and flooding and recover quickly from stress.