The response of soil microbial communities to variation in annual precipitation depends on soil nutritional status in an oligotrophic desert

Study site

The study was carried out in the Cuatro Ciénegas Basin (CCB; 26°45′-27°00′N and 101°48′-102°17′W) in central northern Mexico, within the Chihuahuan Desert. The CCB has an area of 150,000 km², with an elevation of 740 m.a.s.l. The climate is arid with an average annual temperature of 21 °C and 252 mm of annual rainfall, which is concentrated during the summer months (http://smn.cna.gob.mx/). However in the last 30 years the annual precipitation showed a high variability among years. In this study the annual precipitation was estimated as the amount of rain accumulated 9-months before the sampling month. The precipitation data were obtained from meteorological station 5044 “Cuatro Cienegas” located at 26°59′0″N and 101°04′0″W (http://smn.cna.gob.mx/). Annual precipitation and the average temperature of the sampling months varied strongly during the four studied years: the year 2011 was the wettest year (348 mm and 25 °C), 2012 was particularly dry and hot (89 mm and 28 °C) and was followed by two wet years (217 mm and 230 mm for 2013 and 2014, respectively) with lower temperatures (24.9 and 24.8 °C for 2013 and 2014, respectively).

Jurassic-era gypsum is the dominant parent material on the western side of the basin (McKee, Jones & Long, 1990). According to the WRB classification (2007), the predominant soil on the western side of the basin is Gypsisol. The main vegetation types are: (1) grassland (G), dominated by Sporobolus airoides (Torr.) Torr. and Allenrolfea occidentalis (S. Watson) Kuntze; (2) microphyll scrub, dominated by Jatropha dioica Cerv., Larrea tridentate (DC) Cov. and Fouqueria sp Kunth (Perroni, García-Oliva & Souza, 2014); and (3) rosetophylous scrub (RS) dominated by Dhasylirium cedrosanum Trel., and Yucca treculeana Carriére (González, 2012).

Sampling

Mean air temperature for the sampling month (September) and annual rainfall data in each studied year were obtained from the meteorological station “Rancho Pozas Azules” INIFAP. Soil collection was carried out in Churince on the west side of the CCB, where Gypsisol is the predominant soil type (Perroni et al., 2014). The samples were taken from two vegetation cover types, rosetophylous scrub (RS) and grassland (G), during February (2011) and September (rainy of 2012, 2013 and 2014). For each vegetation cover, we sampled seven sites located at a distance of 140 m apart, along a one km north-to-south transect. At each sampling site, a 4 × 4 m plot was demarcated and five soil samples were taken from the first 15 cm of soil depth within the plot, and mixed to produce one compound sample per site. A total of seven composite samples were therefore obtained from each vegetation cover in each sampling year. The soil samples were stored in black plastic bags at 4 °C until subsequent laboratory analysis.

Moisture and pH

Soil pH was measured in deionized water (soil/solution, 1:2 w:v) with a digital pH meter (Corning™). A subsample of 100 g was oven-dried at 75 °C to constant weight for soil moisture determination using the gravimetric method.

Biogeochemical analyses

Molecular analysis

Bacterial composition analysis was performed on the samples from the wettest year (2011). We extracted DNA from each soil sample using the methodology described in López-Lozano et al. (2013) and sent it to J. Craig Venter Institute (JCVI) in order to construct a 16S library using 454 ROCHE tag, 50,000 reads per site of 500 bp and primers 341F-926R. Sequences were trimmed and chimeras eliminated using JCVI protocols. Taxa were assigned using Blast via JCVI pipeline, these methods are detailed by Tanenbaum et al. (2010).

Ecoenzyme activity analyses

The activities of six ecoenzymes (extracellular enzymes) involved in the cleavage of organic molecules with C, N and P were measured: β-1,4-glucosidase (BG), cellobiohydrolase (CBH), β-1,4-N-acetylglucosaminidase (NAG), polyphenol oxidase (PPO), phosphomonoesterase (PME) and phosphodiesterase (PDE), using assay techniques reported by Tabatabai & Bremner (1969), Eivazi & Tabatabai (1977), Eivazi & Tabatabai (1988), Verchot & Borelli (2005) and Johannes & Majcherczyk (2000).

For all ecoenzymes, we used 2 g of fresh soil and 30 ml of modified universal buffer (MUB) at pH 9 for ecoenzyme extraction. Three replicates and two control samples (soil extract with no substrate, and pure MUB with substrate) were included per assay. All ecoenzyme assays were incubated at 40 °C: the BG and CBH for 2 h, NAG for 3 h, PPO for 2.5 h, PME and PDE 1.25 h. Following the incubation period, the tubes were centrifuged at 10,000 rpm for 2 min and 750 µl of supernatant was recovered.

For all ecoenzymes with substrates containing p-nitrophenol (pNP), we diluted the supernatant in 2 ml of deionized water with 75 µl of NaOH and measured the absorbance of pNP liberated at 410 nm on an Evolution 201 spectrophotometer (Thermo Scientific Inc.). For the PPO, we used 2,2′-Azinobis [3-ethylbenzothiazoline-6-sulfonic acid]-diammonium salt (ABTS) as a substrate. The resulting supernatant was measured directly at 410 nm. Ecoenzyme activities were expressed as nanomoles of pNP per gram of dry soil per hour (nmol pNP [g SDE]⁻¹ h⁻¹) for substrates containing p-nitrophenol (pNP) and O₂ formed per gram of dry soil per hour (nmolO₂ [g SDE]⁻¹ h⁻¹) for the PPO, respectively. Specific enzymatic activity was calculated using Eqs. (1)–(3) (Chavez-Vergara et al., 2014; Waldrop, Balser & Firestone, 2000): (1)${\displaystyle \mathrm{SEA}\mathrm{\mu }\mathrm{mol}∕\left(\mathrm{mg}{C}_{\mathrm{mic}}\mathrm{h}\right)=A∕\left({\mathrm{C}}_{\mathrm{mic}}\times 0.001\right)}$ (2)${\displaystyle \mathrm{SEA}\mathrm{\mu }\mathrm{mol}∕\left(\mathrm{mg}{N}_{\mathrm{mic}}\mathrm{h}\right)=B∕\left({\mathrm{N}}_{\mathrm{mic}}\times 0.001\right)}$ (3)${\displaystyle \mathrm{SEA}\mathrm{\mu }\mathrm{mol}∕\left(\mathrm{mg}{\mathrm{P}}_{\mathrm{mic}}\mathrm{h}\right)=C∕\left({\mathrm{P}}_{\mathrm{mic}}\times 0.001\right)}$ where A is the enzymatic activity of BG or CBH or PPO, B is the enzymatic activity of NAG and C is the enzymatic activity of PME or PDE.

Data analysis

Soil moisture, and pH

Regardless of vegetation cover, soil moisture was higher in 2013 and 2014 than in 2012; while the G soil had higher soil moisture than the RS soil, regardless of year (Tables 1 and 2). In the driest year (2012), soil pH was higher than in the wetter years (2013 and 2014), with an exception in the G soil in 2014 (Tables 1 and 2). Soil pH correlated with annual precipitation in both sites (R² =  − 0.85 and R² =  − 0.61 for RS and G, respectively), as well as soil moisture correlated with annual precipitation (R² = 0.76 and R² = 0.88 for RS and G, respectively).

Dissolved organic nutrients and available nutrients

For the two vegetation covers, the lowest values of DOC, DON and DOP were found in 2012 (Table 1). In this year, the RS and the G soils had similar DOC and DOP concentrations, while the G soil had a higher DON concentration than the scrub soil; moreover, the G soil had higher dissolved organic nutrient concentrations than in the RS soil in both 2013 and 2014 (Tables 1 and 2). Consequently, the DOC:DON ratio was lower in 2012 than in the other two years (2013 and 2014) and the RS soil had lower values than the G soils (3.5 and 4.9, respectively); the RS soil also had lower DOC:DOP ratios than the G soil (9 and 16, respectively).

The year trends of available NH${}_4^{+}$ concentration differed between the two vegetation cover types. Available NH${}_4^{+}$ concentration was similar over the three years in the RS soil, while G soil samples from 2012 and 2013 had the lowest and the highest NH${}_4^{+}$ concentrations, respectively (Tables 1 and 2). However, the G soil had higher values than the RS soil in the three studied years. In contrast, the NO${}_3^{-}$ concentration was higher in the samples collected in 2012 than those of the other two years, while the RS soil had higher NO${}_3^{-}$ concentration than the G soil only in the 2012 samples (Tables 1 and 2). The 2012 samples had lower available P concentration than in those collected in the other two years, and the G samples had 40% higher available P concentration than the RS samples, regardless of the sampling year (4.1 and 2.9 µg P g⁻¹, respectively).

Microbial nutrients and ecoenzymatic activities

The highest and the lowest values of Cmic and Nmic were found in 2014 and 2012, respectively, and Cmic values of the G soil samples were 39% higher than in the RS soil samples, regardless of sampling year (254 and 184 µg C g⁻¹, respectively). This was also the case with the Nmic and Pmic concentrations, with an exception in the 2012 samples (Tables 3 and 4). In contrast, Pmic concentrations presented no differences among years within the RS samples, while the 2012 samples had lower Pmic values than was the case in the other two years, within the G samples (Tables 3 and 4). The 2014 samples had lower Cmic:Nmic than the other two years regardless of vegetation cover type (2012 and 2013), while the lowest and the highest Cmic:Pmic and Nmic:Pmic ratios were found in 2012 and 2014, respectively (Tables 3 and 4). The RS soil samples had higher Cmic:Pmic and Nmic:Pmic ratios than in the G soil samples, with an exception in the 2012 samples (Tables 3 and 4).

Significant positive correlations were observed between precipitation and immobilized nutrients within the microbial biomass (Cmic, Nmic, Pmic), in both soils. Moreover, significant positive correlations were detected between precipitation and the Cmic:Nmic and Cmic:Pmic ratios in the RS soil and the Cmic:Pmic and Nmic:Pmic ratios in the G soil. The slopes of the regression with Cmic and Nmic were higher in the G soil and lower in the RS soil (Figs. 1, 2 and Table S1).

The specificenzymatic activity of BG under both vegetation cover types was lower in the wet (2014) than in the dry year (2012; Fig. 3A, Table 4), while that of CBH in the dry year was lower than in both wet years (2013 and 2014), in both vegetation covers (Fig. 3B). The specific enzymatic activity of the PPO in the scrub soil did not differ among years, while the dry year (2012) had lower values than the wet years (2013 and 2014) in the G soil (Fig. 3C and Table 4). Furthermore, the G soil had higher specific PPO enzimatic activity than the RS soil in the wet year (2014). In contrast, the wet year (2014) had the lowest NAG specific enzymatic activity under both vegetation cover types, and the RS soil had lower values only in the dry year (2012; Fig. 3D). The specific enzymatic activity ofPME and PDE was similar, and the lowest values of specific enzymatic activity were in the driest year (2012). In the two wet years, the RS soil presented higher specific activities than the G soil (Figs. 3E and 3F).

Soil bacterial composition

Even at 97% similarity, a very high diversity was found, encompassing all the known phyla of bacteria but a very low diversity and abundance of Archaea. A total of 46,898 sequences were obtained for the RS soil and 9,979 for the G soil, comprising 24 phyla. We observed a high number of unclassified bacteria; 26% for the RS soil and 20% for the G soil (Fig. 4). In the two vegetation cover types, the Proteobacteria was the most abundant bacterial phylum, accounting for 20% in the RS soil and 30% in the G soil. Similarly, Actinobacteria was the second most dominant phylum in the RS soil and in the G soil, with an abundance of 14% in both soils. Interestingly, the Cyanobacteria was the third most dominant phylum, with 13% of abundance both soils, suggesting the importance of the desert crust in both sites. Other important phyla observed were: Chloroflexi (10%), Bacteroidetes (5%), Plantomycetes (4%), Firmicutes (4%), Nitrospira (1% in the RS and 0.5% in the G soils) and Acidobacteria (6% in RS and 0.8% in G; Fig. 4).

Ecoenzymatic stoichiometry, homeostasis and threshold elemental ratios

In all of the model II regressions analyzed, there were no differences found in slopes between soils of the two vegetation cover types within sampling years (Figs. S1 and S2). To test the strength of stoichiometric homeostasis, we analyzed for associations between microbial biomass elemental ratios and those in the soil resources (Tapia-Torres et al., 2015a). In both soil vegetation cover types, the relationships between log C:N_R and log C:N_B, and between log C:P_R and log C:P_B did not differ from zero (p > 0.05), regardless of year (Figs. S1 and S2); indicating strong community-level elemental homeostasis in the soil of both sites.

Moreover, we used the parameters generated from the type II regressions using enzymatic data and microbial C:N:P stoichiometric values to estimate TER_C:N and TER_C:P values. The lowest TER_C:N values were observed in 2014 (wet year), but no differences were observed between 2012 and 2013, or even between vegetation cover types (RS and G) among study years (Fig. 5). The opposite was found for TER_C:P, where we obtained the lowest value in the dry year (2012), but only in the RS soil. For the dry year (2012), no differences were observed between vegetation cover types, while we observed lower TER_C:P values in the G soil than in the RS soil for the wet years (2013 and 2014; Fig. 5).

Reallocation of resources by the soil microbial community

Our first prediction, that the soil microbial community invests more energy in the production of ecoenzymes to acquire nutrients in sites of low resource availability, such as the RS soil, was confirmed. We observed that the RS soil showed a lower concentration of available P than the G soil in the three years studied and, consequently, the RS soil microbial community invested more energy in the acquisition of P (increased enzymatic activity of phosphomonoesterase and phosphodiesterase) than the G soil microbial community only during the two wet years. In contrast, the G soil had higher Cmic, Nmic and Pmic concentrations and lower enzymatic activity of phosphomonoesterase and phosphodiesterase than the RS soil during both wet years, which also supports our prediction (Fig. 1 and Table 3). These results suggest that the microbial community in the RS soil, with lower resource availability, must reduce growth as a result of: (1) the physiological cost associated with a low reallocation to P-rich ribosomal RNA, as suggested by the growth rate hypothesis (GRH) (Sinsabaugh & Follstad Shah, 2012; Sterner & Elser, 2002; Zechmeister-Boltenstern et al., 2015) and (2) the required investment of energy towards the acquisition of P in order to produce ecoenzymes (Evans & Wallenstein, 2012; Schimel, Balser & Wallenstein, 2007; Wallenstein & Hall, 2012). The microbial C:N:P ratio was greater in the RS soil (127:19:1) than in the G soil (63:10:1), suggesting that the microbial community in the former site is more P-constrained (Cleveland & Liptzin, 2007). The studied soils are characterized by low P availability and a high capacity for P occlusion within inorganic molecules, mainly by Ca-bound (Perroni et al., 2014). Therefore, the main source of available P is mineralization of organic P mediated by phosphatase activity (Waring, Weintraub & Sinsabaugh, 2014). Among organic P molecules, phosphodiester forms are the preferred substrate in P-limited ecosystems (Karl, 2014; Tapia-Torres et al., 2016), although phosphomonoester forms may also be an important source of available P in most soils (Turner, Mahieu & Condron, 2003). In our study sites, phosphodiesterase activity was almost ten times higher than that of phosphomonoesterase, mainly in the scrub soil, suggesting mineralization of phosphodiesters as the main source of soil available P. Several bacteria isolates from CCB soils prefer to grow in DNA as a P source, associated with phosphodiesterase activity (Tapia-Torres et al., 2016). We suggest that the main P source in sites with low nutrient availability, such as the RS soil, is recycling of the organic molecules that are the product of cellular lysis.

However, the G soil had higher enzymatic activity of polyphenol oxidase in the wet year 2014 than was the case in the RS soil. This result is consistent with other studies (Sinsabaugh, 2010; Sinsabaugh & Follstad Shah, 2011), which have reported that polyphenol oxidase activity does not present the same behavior as the β-1,4-glucosidase and other hydrolases that degrade labile C. Microbial community size begins to be limited by the availability of labile C, which produces a change in the microbial community composition towards microbial guilds with lower growth rates (low concentration of Cmic), but with the capacity to produce polyphenol oxidase to break down structurally complex molecules and obtain C (Moorhead & Sinsabaugh, 2006). This situation is comparable to the conditions of the G soil in the wet year 2014, where the microbial community was required to cleave lignin in order to maintain its growth rate.

Furthermore, the differences in soil nutrient dynamics between both sites can be strongly affected by soil microbial composition. While analyses of soil bacterial composition were only determined for 2011, this year presented the highest soil water availability and also showed higher concentrations of Cmic, Nmic, and Pmic than in the other studied years. Several studies (Nemergut et al., 2010; Philippot et al., 2009) have reported that heterotrophic decomposition depends on the relative abundance of specific taxa because different species process organic matter at different rates, even under similar soil conditions. The G soil had a higher proportion of Proteobacterias, Actinobacterias and Bacteroidetes than the RS soil, and some species of these taxa have the capacity to produce β-glucosidase (BG) (Moreno et al., 2013), cellobiohydrolase (CBH), poliphenoloxidase (PPO) (Kirk & Farrell, 1987), glucanases and glycosidases (Xie et al., 2007), which act to cleave C molecules.

In contrast, the scrub soil had higher proportion of Acidobaterias and Firmicutes, including species with the capacity for producing enzymes for P mineralization (Koch et al., 2008; Tan et al., 2013). Moreover, the acidobacterias of the RS soil could contribute to the release of unavailable P through organic acid release (Tan et al., 2013) and, together with Firmicutes, can mineralize P via the production of phosphatases, as has been observed in isolates of acidobacterias from substrates with low C concentrations (Koch et al., 2008; Tan et al., 2013). Chloroflexi was present in a higher proportion in the RS than in the G soil, but both soils had a similar proportion of Cyanobacteria suggesting that the amount of microbial desert crust is similar in both sites. Both phyla are facultative autotrophic bacteria (Smith, 1983) and therefore have the capacity to fix atmospheric C and to produce ecoenzymes for depolymerization and mineralization of C (Berg et al., 2010; Mitsui et al., 1986; Smith, 1983). The Cyanobacteria also have the capacity to fix atmospheric N. Fixation in the microbial biomass of C and N by these taxa could represent an important input of both nutrients to the soil (Mitsui et al., 1986; Smith, 1983). Wallenstein & Hall (2012) proposed that sites limited by nutrients are more vulnerable to rainfall variability, because the microbial community must invest energy in nutrient acquisition, thus reducing its capacity for adaptation required by fluctuation in water availability. We proposed that sites with low resources availability, such as the RS soil, could be thus more vulnerable to annual precipitation variability.

Resilience in the face of precipitation changes

Our second prediction, that the microbial community will be more vulnerable to variability in precipitation in the site with lower soil resources (RS), was not confirmed because the soil community was resilient to soil P coinstrains by ecoenzyme upregulation during times of adequate moisture. In both vegetation cover types, nutrient availability increased with increased precipitation. The correlation between precipitation and the Cmic, Nmic and Pmic, indicate that a higher amount of rainfall favored the microbial immobilization of these nutrients under both vegetation cover types. Nevertheless, compared to the RS soil, the G soil showed steeper slopes in regressions between the precipitation and the concentrations of Nmic and ratios of Cmic:Pmic and Nmic:Pmic (Table S1), suggesting that the microbial community of the grassland soil has the ability to immobilize more N within its microbial biomass and more rapidly than the microbial community of the RS soil. Positive correlations between Cmic and rainfall have been reported for an oak forest (Baldrian et al., 2010) and a semiarid grassland (Zhou et al., 2013), but a correlation between precipitation with Nmic and Pmic concentrations has not hitherto been reported for natural ecosystems.

Furthermore, in the soil community homeostasis analyses, the relationships between log C:N_R and log C:N_B, and between log C:P_R and log C:P_B in the G and the RS soils had slopes that did not differ significantly from zero (Figs. S1 and S2), suggesting that the soil microbial communities adjust physiologically (Sinsabaugh & Follstad Shah, 2012) to processing low N and P resources in order to cope with the nutrient limitation, particularly in dry years. Our data also suggest that these physiological adjustments occurred differently in the soil microbial communities of the two vegetation covers and was related to both precipitation quantity and nutrient availability.

Our results show how values of TERC:N and TER_C:P may shift with respect to variation in annual rainfall and different vegetation cover. The estimated TER_C:N was lower in the wet year for both sites, indicating greater sensitivity to N limitation due to the rapid growth of the microbial community produced by the water availability. For TER_C:P, we observed site-specific differences. The TER_C:P was higher in the RS soil than in the G soil for 2013 and 2014, indicating a greater sensitivity of the microbial community to P limitation in the G soil. However, in order to determine the nutritional limitations of the microbial community, we also compared the estimated TER values and the C:N or C:P ratios of the organic matter. If the C:N or C:P ratio of the organic matter being consumed is greater than the TER for that element, this would suggest nutrient limitation (Sterner & Elser, 2002). We observed P limitation in both soils, regardless of year (C:P > TER_C:P; p < 0.05) and N limitation in the G soil in the wet year (C:N = 11.3 and TER_C:N = 6; p = 0.002). Our results for the dry year (2012) showed that the ecoenzymatic activities associated with C and P acquisition were lowest in the RS and G soils. Values for TERC:N and TER_C:P were similar between the RS and G soils, suggesting that both sites may be vulnerable to drought. However, with the increase of the annual precipitation (years 2013 and 2014), the G soil microbial community requires more P and N to meet its metabolic demands and it makes metabolic adjustments in order to maintain its growth which makes it more susceptible or sensitive to resource limitation. Similarly, increased ecoenzyme activities associated with P acquisition and elevated TER_C:P values when the water is not limiting (2013 and 2014) suggest that the RS soil microbial community is well adapted to acquire P resources via ecoenzyme upregulation post drought.

We suggested that, under the scenario proposed by Global Climate Change models for desert ecosystems that predict reduced annual precipitation and increased rainfall variability, the microbial community from both sites could be vulnerable to drought events, but the RS soil microbial communities can make adjustments in order to obtain nutrients in wet years, suggesting that this community is resilient post drought.