The soil used in the experiments was collected from barren patches of an unmarked plot at the long-term ecological research (LTER) station of Avdat (30°47' N, 34°46' E; 600–700 m elevation) during the summer of 2012. The mean annual rainfall in the region is ~90–130 mm and rain occurs only during the winter months (UNESCO, 1979). Eight randomly selected subsamples were taken from the top 5 cm of the bulk soil (using an ethanol-cleaned scoop), after removing the crust. The collected soil was placed in sterile bags (Whirl-Pack, Nasco, Fort Atkinson, WI), transported to the laboratory and kept at 4°C until its homogenization, within less than 24 h, by passing through an autoclaved 2-mm grid size sieve. Subsamples were taken for physicochemical analyses and a total of 3.3 kg of homogenized soil was packed in each of three replicate columns (Figure 1).

A full factorial design was used to evaluate the soil microbial community response to hydration–desiccation cycles, resulting in 72 samples (3 replicate mesocosms × 4 treatments × 6 sampling times). The mesocosms (Figure 1) consisted of three soil-packed polyvinyl chloride (PVC) cylinders (30 cm high, 10 cm in diameter) fitted at the bottom with a 4- to 8-μm pore size ceramic filter (Ace Glass, Vineland, NJ, USA). An outlet tube at the base of the column led to a continuously operating vacuum pump to simulate natural capillary forces in the soil. Ten sampling ports were located around the top 12 cm of the column; the ports were fitted with Suba-Seal rubber septa (Sigma-Aldrich, St. Louis, MO, USA).

Rain events were mimicked with a shower-like rain simulator (Figure 1) consisting of a PVC disk (10 cm in diameter) equipped with 21 syringe needles (0.4 × 13 mm) dripping double-distilled water (DDW) onto the soil surface. The simulated rain was dispersed at a rate of 1.5 ml min⁻¹ by means of a peristaltic pump (Gilson Minipuls 3, Middleton, WI, USA). The applied volume of water was 80 or 400 ml, corresponding to light and heavy rain, respectively. Following the single rain event, the columns were kept in the dark for 21 days under constant temperature of 25 or 36°C, or under a 12 h diurnal cycle of 36 and 10°C (Table 1).

The soil columns were sampled 0.5, 1.5, and 3 days after the rain event, and then once a week for up to 3 weeks. The samples were removed with a sterile spatula through the column side ports (Figure 1). A total of 45–60 g soil was removed from each column and divided, resulting in triplicate samples for each sampling time. Samples from the soil homogenate prior to packing into the columns were considered the baseline prior to treatment (time 0). For molecular analyses, 20–30 g of soil from each time point was suspended in RNAlater (Zoetendal et al., 2006) at a 1:1 ratio (w/v), for better stabilization of the cellular RNA, and stored at −80°C.

Soil physicochemical properties were determined according to standard methods (Smith and Doran, 1996; Doran et al., 1996a,b). Soil water content was determined by gravimetry. Electrical conductivity (EC) and pH were measured with EC electrodes and pH meter, respectively, in saturated soil extract. Nitrite-N was determined by colorimetric method (HCl, sulfanilamide and ethylenediamine dihydrochloride mix) using an Infinite M200 spectrophotometer (Tecan, Männedorf, Switzerland). Nitrate-N was measured by the second-derivative method (Si et al., 2007) using a BioMate 5 spectrophotometer (Thermo Scientific, Waltham, MA, USA).

Total nucleic acids were extracted as previously described (Angel, 2012). We followed bacterial RNA, which has been proven to be a better indicator of active community variations than DNA (Blazewicz et al., 2013). The obtained RNA was purified using an RNA purification kit (Epicenter, Madison, WI, USA) according the manufacturer's instructions. The RNA was reverse-transcribed to cDNA using ImProm-II™ Reverse Transcriptase (Promega, Madison, WI, USA) in the presence of Recombinant RNasin Ribonuclease Inhibitor (Promega) following the manufacturer's protocol. The resulting cDNA was purified by PCR purification kit (Bioneer, Daejeon, Republic of Korea) and quantified spectrophotometrically with Nanodrop (Thermo Scientific).

The abundance of total bacteria, Actinobacteria and Firmicutes populations was quantified using group-specific quantitative (q) PCR assays targeting the 16S ribosomal (r) RNA units (Supplementary Table 1). All qPCRs were performed in an iCycler thermocycler equipped with a MyiQ detection system (Bio-Rad, Munich, Germany), and the data were processed using Bio-Rad CFX Manager 3.0 software. The quantification assays were based on SYBR Green I quantification (Thermo Scientific).

For all assays, standards containing known copy numbers of the target gene were used. Genomic DNA of Streptomyces griseus was used to quantify total bacterial and Actinobacteria. Cloned fragment of the 16S rRNA-encoding genes obtained from Bacillus subtilis was used to quantify the Firmicutes abundance. To evaluate total bacteria each qPCR contained 10 μl SYBR Absolute Blue qPCR Rox Mix (Thermo Scientific), 1 μl of 400 nM of each primer (Metabion, Munich, Germany), 5 μl template cDNA and 3 μl molecular-grade water (HyLab, Rehovot, Israel). To quantify the Firmicutes population, each qPCR contained 12.5 μl SYBR Absolute Blue qPCR Rox Mix, 0.5 μl of 250 nM of each primer, 5 μl template, and 6.5 μl molecular-grade water. Each reaction was repeated at least twice.

Bacterial abundance was estimated under the following conditions: 95°C for 15 min, followed by 35 cycles of 95°C for 10 s, 60°C for 15 s and 72°C for 30 s for extension. Conditions for estimating Actinobacteria abundance were: 95°C for 15 min, followed by 35 cycles of 95°C for 45 s, 63°C for 45 s and 72°C for 45 s. Firmicutes abundance estimation conditions were: 95°C for 15 min, followed by 35 cycles of 95°C for 10 s, 65°C for 15 s and 72°C for 45 s.

The effect of changing soil physicochemical parameters during each of the experiments was studied with linear mixed models using the R package lme4 v1.1-12 (Bates et al., 2015). The different experimental conditions were modeled together, each with a random slope to account for a different scale of community response to each condition. P-values were estimated using the Satterthwaite approximation for degrees of freedom. More information about the statistical analyses can be found in the Supplementary Information: qPCR model.

Amplicons of the soil bacterial 16S rRNA regions V3 and V4 were sequenced at the Core Facility of the University of Illinois (http://www.rrc.uic.edu/). The generated reads were quality-filtered and checked for chimeric sequences using the usearch method (Edgar, 2010). Sequences were aligned with the PyNAST tool (Caporaso et al., 2010a) and clustered into operational taxonomic units at 90% similarity between 16S rRNA using an open reference picking pipeline provided by the Qiime package (Caporaso et al., 2010b). Operational taxonomic units were picked and analyzed using Silva 111 (Quast et al., 2013) and sequences that were not present in the database were clustered de novo. The number of sequences in each sample was randomly subset (rarefied) to 9,000 sequences per sample using the phyloseq package (McMurdie and Holmes, 2013; Oksanen et al., 2016). Operational taxonomic units appearing more than 500 times in the data set were selected for a community structure stack plot and averaged per putative cluster. Both total counts were then normalized to 1.

Evenness and richness of the soil bacterial community were studied during each experiment. We chose species count as representative of population richness and Pielou's measure of species evenness as an estimator of population shape (Pielou, 1966).

The effect of the measured soil physicochemical parameters on community richness and evenness was examined. The overall effect was evaluated using a linear mixed model with random effects from the R package lme4 v1.1-12 (Bates et al., 2015). A random slope was introduced for the different experimental conditions. Details of the analysis can be found in the Supplementary Information: Evenness model, qPCR model, Richness model and Canonical correspondence analysis (CCA) model. P-values were estimated using the Satterthwaite approximation for degrees of freedom.

The initial water content in the soil was very low, 0.97% ± 0.05 (SEM). It increased, following the rain simulations, to 16.72% ± 0.5 after the heavy rain (50 mm), corresponding to field saturation of loess soil (Muñoz-Castelblanco et al., 2012), and 4.89% ± 0.98 after the light rain (10 mm) (Figure 2A).

We note that during the initial stages of the experiments (0.5, 1.5, and 3 days after the rain event), the soil water content was similar when heavy rain was applied (Figure 2B), despite the different temperatures. At later stages (1, 2, and 3 weeks after hydration), the higher temperature and diurnal cycle conditions increased desiccation of the soil mesocosm.

The salinity (EC) of the soil dropped in the early stages of hydration and increased at the later stages of desiccation, as the capillary rise transferred salts back to the sampled level of the mesocosms. This effect was less pronounced when heavy rain and lower temperature were applied, probably because the soil did not fully desiccate (Supplementary Table 2). Under all conditions tested, soil pH was stable at around 8.5.

The concentration of ammonium-N increased after both heavy and light rain events when the temperature was lower (Supplementary Table 2). The elevated concentration of ammonium-N under higher temperature and diurnal cycles might have been due to external contamination of the samples; however, during the experiments, the concentrations of ammonium equilibrated under both conditions (Supplementary Table 2).

The concentrations of nitrite- and nitrate-N did not change markedly during the experiments. Nevertheless, under higher temperature and diurnal cycles, elevated concentrations were found in the initial measurements, possibly due to external contamination. Nitrate-N concentration decreased during the early stages of the experiments, returning to its original value in the later stages of each experiment (Supplementary Table 2).

The temporal changes in the abundance of total bacteria, Actinobacteria and Firmicutes, as assessed by amplification and detection of the 16S rRNA by qPCR, are illustrated in Figure 3. The results suggest that the quantity of the bacterial 16S ribosomes were strongly dependent on treatment. The ribosomal abundance of total soil bacteria, Actinobacteria and Firmicutes decreased after heavy rain at the lower temperature, and then remained low throughout the experiment. No clear trend could be observed under the other three sets of conditions, except for the abundance of Actinobacteria, which increased over time under light rain at the lower temperature.

According to the linear mixed model, the different temperature conditions had the strongest effect on each of the observed groups (Table 2). The soil concentration of ammonium-N had a significant effect on the abundance of total bacteria (t = 3.615, p = 0.0006) as well as on that of Actinobacteria (t = 2.101, p = 0.039). The soil water content had a significant effect on total bacterial population (t = −2.137, p = 0.036).

We observed a drop in both population richness and evenness (Figure 4) under heavy rain and lower ambient temperature, with neither of these two measures recovering for the duration of the experiment. In contrast, no marked differences were detected in richness or evenness under the other applied treatments (Figure 4), i.e., light rain, higher temperature or temperature diurnal cycles. The effect of soil chemical parameters on bacterial richness and evenness was analyzed by a linear mixed model with random effects (Table 3). The results suggest that sampling time, nitrite-N, EC, and water content had significant effects on both bacterial richness and evenness (Table 3).

To further explore which members of the bacterial community changed during the experiments, we analyzed its composition and plotted the data using non-metric multidimensional scaling (NMDS), applying the Bray–Curtis distance measure (Figure 5). The final stress of the dimensionality reduction is equal to 0.083, suggesting that the plot well represents the data set. Further analysis using ANOSIM from the Vegan package (v2.4-1) in R (Oksanen et al., 2016) revealed that the community composition changed significantly during the experiment (R = 0.3797, p < 2.10⁻⁵, 50,000 permutations) (Supplementary Figure 1). These changes were probably due to the community shift upon application of heavy rain and lower temperature. However, when the soil was treated with light rain and lower ambient temperature, heavy rain and higher temperature, or heavy rain and a temperature diurnal cycle, no marked shifts in community composition were detected (Figure 5, Supplementary Figure 2). The phyla Actinobacteria, Bacteriodetes and Firmicutes showed the most significant differences between the “unchanged” cluster and the heavy rain and lower temperature cluster. Actinobacteria decreased from 39.0 to 11.3% under heavy rain and lower temperature, whereas Bacteroidetes increased from 3.3 to 16.4% and Firmicutes from 4.1 to 35.9% under these conditions. Details of community composition are presented in Supplementary Table 3.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.