Introduction

Rainfall patterns are expected to change dramatically in the near future; drought events will likely increase in frequency, the areal extent of arid zones will expand, and floral compositions will change, especially in drylands (IPCC 2007). Water is usually the most limiting resource for plant growth in arid regions, but nitrogen (N) is a co-limiting factor for net primary production (NPP) (Austin and Sala 2002; Eskelinen and Harrison 2015). When drought periods are protracted, N is a greater limiting resource for NPP than water (Ren et al. 2017). Studies of N cycling responses to rainfall pattern changes are essential for a better understanding of soil dynamics in an era of rapid climate change. Many studies have examined the mechanism of soil N dynamics responses to rainfall (Landesman and Dighton 2010; Chen et al. 2013, 2017; Ladwig et al. 2015; Bowles et al. 2018). However, most have focused on only single parts of the N transformation process or have failed to provide information on microbial N drivers. A comprehensive study that includes steps of soil N transformation and relevant microbes will contribute significantly to a deeper understanding of the mechanism underlying the responses of soil N dynamics to rainfall pattern change.

The series of N transformation processes in the soil can be divided into three steps. The first step is degradation, which produces small-molecule organic N from large-molecule organic N, such as the organic matter in soil. In the second step, the small-molecule organic N compounds are mineralized and converted to ammonium N. The third step is nitrification, in which ammonium N is converted to nitrate N. Thus, soil N transformation is a step-by-step process; substrate quantity and quality can limit downstream transformations (Updegraff et al. 1995; Norman and Barrett 2014). Changes in an upstream step can affect the subsequent downstream step by acting as a bottleneck, i.e., a step that delays the progress of subsequent steps. For example, in cold ecosystems, the mineralization step is limited by temperature and can act as a bottleneck (Rustad et al. 2001; Schimel et al. 2004). The nitrification step can also act as a bottleneck when the pH is low (Ste-Marie and Paré 1999). Thus, each step may react differently to changes in rainfall patterns. Our understanding of the total soil N dynamics process will be improved by dividing the chain of events into separate steps that can be examined in detail.

The responses of soil N dynamics to rainfall pattern changes are mediated by soil microbes (Chen et al. 2017; Bowles et al. 2018). Microbes also fall into three categories of step drivers. Fungi and prokaryotes are the main drivers of the degradation and mineralization steps, respectively (Moore et al. 2003). In the nitrification step, Ammonia-oxidizing bacteria and archaea are important because they perform the rate-limiting process. (Kowalchuk and Stephen 2001; Isobe et al. 2015). Nitrite-oxidizing bacteria have recently received increasing attention, although concentrations of the nitrite N concentration was much smaller than those of ammonium and nitrate N (Burns et al. 1996; Smith et al. 1997; Xu et al. 2012; Maharjan and Venterea 2013; Venterea et al. 2015). A focus on microbial composition improves understanding of the mechanism of soil N dynamics because the abilities and strategies for N transformation vary among microbial taxa. For example, among fungi, saprotrophs and mycorrhizal species decompose organic matter primarily to obtain carbon (C) and N, respectively. Mycorrhizal fungi receive C from their host plants (Lindahl et al. 2007; Smith and Read 2010). The soil bacteria may be categorized as copiotrophs, which readily decompose labile compounds, and oligotrophs, which decompose recalcitrant compounds (Fierer et al. 2007). These two bacterial groups also differ in their ability to produce extracellular enzymes, including N-acetylglucosaminidase and chitinase, which decompose monomers and polymers, respectively (Zimmerman et al. 2013). Ammonia-oxidizing bacteria and archaea also have different oxidization capabilities: nitrification rates tend to correlate with ammonia-oxidizing bacterial abundance, but not with archaeal abundance, likely because ammonia-oxidizing bacteria are competitive at high ammonium concentrations (Di et al. 2009; Chen et al. 2013; Banning et al. 2015), although not in acidic soil (Prosser and Nicol 2008; Isobe et al. 2015).

Many studies have examined the ways in which soil physicochemical properties influence the microbes driving soil N processes. The low soil moisture, low organic matter content, and high pH in drylands limit fungal and bacterial growth (Frey et al. 1999; Maestre et al. 2015) and change community compositions by reducing the abundances of copiotrophic and ammonia-oxidizing bacteria (Adair and Schwartz 2008; Xiong et al. 2012). However, under such stressful conditions, microbial groups with high drought tolerance or enhanced abilities in recalcitrant nutrient decomposition flourish, as observed in Actinobacteria and Verrucomicrobia (Bouskill et al. 2013; Cederlund et al. 2014; Iwaoka et al. 2018). Thus, a comprehensive view of soil N dynamics will emerge from studies that link information on each step of soil N transformation, their drivers, the soil microbial community, and soil physicochemical properties. With a such a broad view, we can better predict how rainfall pattern shifts will alter the soil conditions that regulate microbial communities and the chain of N transformation processes.

Natural aridity gradients provide appropriate candidate conditions for studies aiming to predict the effects of rainfall pattern changes on soil N dynamics. Along an aridity gradient, soil moisture content changes with other soil properties, such as organic matter content, as a function of rainfall history, unlike in shorter-term artificial rainfall experiments. Soil N availability, N transformation rate, and microbial abundance tend to increase with rainfall toward the wetter end of an aridity gradient (Burke et al. 1997; Austin and Sala 2002; Adair and Schwartz 2008). However, this is not always the case, as mineralization rate, nitrification rate and N availability have been reported not to change with rainfall (Smith et al. 2002; Feral et al. 2003; Meier and Leuschner 2014). Contradictions among studies have arisen because the underlying mechanisms are not understood. In our study, we examined three black locust (Robinia pseudoacacia) forests occurring along a natural aridity gradient on the Loess Plateau in northeastern China. Differences in vegetation, including variability in canopy trees, strongly influence soil N dynamics (Finzi et al. 1998; Phillips et al. 2013); we therefore compared the same vegetation along the aridity gradient.

We examined the process of soil N dynamics changes (mediated by shifts in the soil microbial community) along the aridity gradient. We aimed to use the data from this study to predict responses to rainfall pattern alterations in a time of global climate change. We studied all three steps of N transformation, the associated microbial communities, and soil physicochemical properties along the aridity gradient. Our fundamental hypothesis was that inhibition of a single step in the transformation process would cause a bottleneck in the soil N cycling process. Such bottlenecks would be key elements in the N limitation of forest productivity in arid ecosystems during future climate change.

Materials and methods

Study sites and soil sampling

We conducted the study in three black locust (Robinia pseudoacacia) forests spanning the southern to central sectors of the Loess Plateau. The three stands were located in mountainous areas of Shaanxi Province, China: (i) near Yongshou County (Huiping; 34°48N, 107°59E, 1395 m elevation), (ii) near Yan’an City (Mt. Gonglushan; 36°25N, 109°32E, 1320 m elevation), and (iii) near Ansai Country (Zhifanggou; 36°45N, 109°15E, 1175 m elevation), which we refer to hereafter as the Wet, Med (medium), and Dry sites, respectively. The vegetation type has been classified as a forest–steppe transitional zone; the mean annual rainfall decreases gradually from southeast to northwest (Tsunekawa 2014). The reference evaporation rate increases from south to north in our research region (Li et al. 2012). Annual rainfall and mean annual air temperature at the sites were as follows: 606 mm and 10.8 °C at the Wet site (Li et al. 2015), 514 mm and 10.2 °C at the Med site (Otsuki et al. 2005), and 449 mm and 8.8 °C at the Dry site (Qiu et al. 2012). The area has hot summers and cold winters; heavy precipitation occurs in summer. Although the precipitation gradient in this study, only 150 mm in mean annual precipitation, is smaller than those in other studies (Austin and Sala 2002; Feral et al. 2003; Adair and Schwartz 2008; Meier and Leuschner 2014), 150 mm is a critical difference in this area, the arid boundary of a forest. The rainfall amount at the Dry site was 74% of that at the Wet site; 30% exclusion of rainfall in semi-arid forests has been shown to reduce forest production (Ogaya and Penuelas 2007). Our gradient enables a comparison of the effects of precipitation under similar environmental conditions of mean annual temperature and soil type.

At all three sites, forest canopies were closed and occupied by > 90% black locust, with understory covers of shrubs and herbaceous species. Tree basal area, the amounts of organic layer and litter quality (C:N ratio) decreased along the aridity gradient (Table S1). In each forest, we established four plots (20 × 20 m), each > 30 m distant from all neighboring plots. All plots had similar slopes (flat land or gentle slopes near ridges) because topography considerably affects soil and N contents in this region (Tateno et al. 2017). Soil samples (0–10 cm deep) were taken from four locations in each plot. The sampling locations were ca. 10 m distant from one another. The four soil samples were combined into a single sample. Samples were taken in September 2015, and in April, June, and September 2016.

Measurement of soil physical and chemical properties

Soil water content was determined after samples were dried at 105 °C for > 3 days. Soil pH was measured with a pH meter (D-51; HORIBA, Kyoto, Japan) in a 2:5 soil/water suspension. After soil samples were ground, total C and N contents were measured using the dichromate oxidation and Kjeldahl methods, respectively. Soon after soil collection and incubation at 25 °C for > 6 days without freezing soil samples, we extracted dissolved C and N with 2 M KCl in a 1:10 soil/extractant ratio. The extracts were stored in a freezer until further analysis. The total dissolved organic C and N contents in the extracts were measured with a TOC-LCPH/CPN + TNM-L device (SHIMADZU, Kyoto, Japan) (Shimadzu 2013). We measured the amounts of ammonium and nitrate N in the extracts with a Bran + Luebbe AutoAnalyzer III device (BLTEC, Tokyo, Japan) using a colorimetric procedure. We also measured the amount of nitrite N in the extracts which was from soils collected on two occasions from the same sites. Soil dissolved organic N content, reportedly correlated with protein abundance that can be degraded by prokaryotes (Yu et al. 2002; Wang et al. 2013), was calculated as the difference between the concentrations of total dissolved N and inorganic N (i.e., ammonium + nitrate N). Net N mineralization and nitrification rates in the soil were calculated as the differences in inorganic N and nitrate N, respectively, before and after incubation. We report N content per kilogram of dry soil. Soil organic matter quality and dissolved organic matter quality were examined based on the total organic C:N and dissolved organic C:N ratios, respectively.

Soil DNA extraction and quantification by real-time quantitative PCR

DNA was extracted from 0.25-g soil samples using the MoBio Powersoil DNA Isolation Kit (MoBio, Carlsbad, CA, USA) according to the manufacturer’s instructions. The real-time quantitative polymerase chain reaction (qPCR) was run in a Light Cycler Nano thermal cycler device (Roche Diagnostics K.K., Mannheim, Germany) with the SYBR Green I intercalating dye. Fungal ITS, bacterial 16S rRNA, archaeal 16S rRNA, ammonia-oxidizing bacterial amoA, and ammonia-oxidizing archaeal amoA genes were targeted to estimate the population size of each soil microbial group. Fungal ITS abundances were determined using the universal primer sets ITS1F_KYO2/ITS2_KYO2 (Toju et al. 2012). The qPCR conditions for fungal ITS genes were as follows: initial denaturation at 95 °C for 10 min, followed by > 30 cycles at 95 °C for 1 min, 53 °C for 30 s, and 72 °C for 1 min (Fierer and Jackson 2005). The subsequent step in the analysis for 16S rRNA and amoA genes has been described in detail by Iwaoka et al. (2018).

Sequence analysis

We amplified the ITS and 16S rRNA genes in the DNA extracts using the primer sets ITS1F_KYO2/ITS2_KYO2 (Toju et al. 2012) and U519f (Suzuki and Giovannoni 1996)/U785r (Wang and Qian 2009), respectively, with the following reaction solution ratio: 44:1:1:4 Platinum PCR Super Mix High Fidelity (Thermo Fisher Scientific)/10 mM forward primer/10 mM reverse primer/tenfold-diluted DNA sample. The subsequent step in the analysis has been described in detail by Iwaoka et al. (2018). Sequencing was performed on an Ion Personal Genome Machine (PGM) (Thermo Fisher Scientific). The processing of the sequencing data for 16S rRNA genes was performed as described by Iwaoka et al. (2018). ITS sequences shorter than 160 bp in length were removed; ITSx was used to extract fungal ITS sequences (Bengtsson-palme et al. 2013). The remaining steps were the same as for 16S rRNA sequences. Sequence data were deposited in the Sequence Read Archive of the National Center for Biotechnology Information under accession number PRJDB7067.

In our analyses of the fungal and prokaryotic community structures, we standardized the read numbers to 10031 and 8726, respectively, using random pick-up based on the minimum read number. We searched for fungal trophic modes from the fungal taxonomy and predicted specific functions based on the 16S rRNA genes using the FUNGuild database and the Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt) pipeline, respectively (Langille et al. 2013; Nguyen et al. 2016). To perform PICRUSt searches, the sequences were re-picked at a 97% identity level with reference to the Greengenes database (version 13_05). We eliminated reads not corresponding to the reference database; the OTU table was normalized by gene copy number, and the metagenome functional profiles were predicted using PICRUSt (ver. 1.1.1) to generate a table of Kyoto Encyclopedia of Gene and Genomes (KEGG) orthologs (KOs). We selected the genes necessary for coding extracellular enzymes responsible for N compound degradation in soil, i.e., cellulase (EC 3.2.1.4), N-acetylglucosaminidase (EC 3.2.1.52), and chitinase (EC 3.2.1.14), based on the KOs (Isobe et al. 2018).

Statistical analyses

We used one-way analysis of variance (ANOVA) with the sampling date (September 2015/April 2016/June 2016/September 2016) as a random effect to test for significant differences among the sites in soil physicochemical properties, total and dissolved organic C, N contents and C:N ratios, inorganic N content and net transformation rate, microbial abundance, relative abundances of fungal and prokaryotic groups, and predicted gene abundances. We chose a significance level of P < 0.05 for all tests. Prior to ANOVA, we tested for normality and homogeneity of variance in the data using the Kolmogorov–Smirnov and median Levene’s tests, respectively. The car package in R software (version 3.1.2; R Development Core Team 2014) was used to perform Levene’s test (Fox et al. 2014). When the P values for these tests were < 0.05, data were square-root or reciprocally transformed prior to ANOVA analysis. Significant pairwise differences between sites were identified by Tukey–Kramer multiple comparison tests, or Steel–Dwass multiple comparison tests when the homogeneity of variance was violated. Correlations were tested using the Pearson procedure or by a generalized linear mixed model with site and sampling date as random effects.

Non-metric multidimensional scaling (NMDS) analysis of community structure dissimilarity based on the Bray–Curtis index was performed using the metaMDS function in the vegan package (Oksanen et al. 2016) of R software. The envfit function in the vegan package was used to identify significant correlations between (i) soil properties and the relative abundances of the fungal trophic groups, saprotrophic fungal orders, and the predicted abundances of genes for N-degrading enzymes and (ii) the NMDS values of points as vectors on the NMDS ordination plot. Permutational multivariate analysis of variance (PerMANOVA) was performed to test the significance of the effect of site on the microbial community, with sampling date as a random effect, using the Adonis function in the R vegan package.

Results

Soil physicochemical properties and organic matter quality and quantity

Soil water content was significantly higher at the Wet site than at the other two sites (Table 1). Soil pH also differed significantly among sites and was lowest at the Wet site. Soil total N content differed significantly among sites and was highest at the Wet site, followed by the Med site. Total C:N ratios did not differ significantly among sites. The dissolved organic C:N ratio differed significantly among sites, and was lowest at the Wet site. For soil water content, pH, and dissolved organic C:N ratio, the values for the Med site were more similar to those in the Dry site, but for total N content, the Med site was more similar to the Wet site.

Table 1 Soil physicochemical properties, organic matter quality and quantity and net N mineralization and nitrification rates
Full size table

Extractable N dynamics in soils

Dissolved organic N, ammonium N, nitrate N, and total inorganic N contents differed significantly among sites (Fig. 1). The Wet site had significantly higher dissolved organic N and ammonium N contents than did the Dry site. The Wet site had significantly higher nitrate N and total inorganic N contents than did the Med and Dry sites. The net N mineralization rate in the soil did not differ significantly among sites (Table 1). The net nitrification rate differed significantly among sites, and was highest at the Wet site. Nitrite N was generally undetectable, with a maximum of 0.008 ppm (< 0.1 mg kg−1).

Fig. 1
figure 1

Extractable N contents in the soils. Values are means ± SDs. Significant differences among sites, identified by one-way ANOVA with the sampling date as a random effect, are indicated in the upper right text panel. *P < 0.05, ***P < 0.001. Results of Tukey–Kramer or Steel–Dwass tests are shown inside the bars; different lowercase letters indicate significant pairwise differences between sites for each parameter

Full size image

Soil fungal communities

Fungal ITS gene abundance in the soil did not differ significantly among sites (Table S2). However, the PerMANOVA showed that soil fungal community structure was influenced significantly by site (F = 11.8, partial R2 = 0.34, P < 0.001, Table S3), and the fungal communities at the three sites along the aridity gradient were separated on axis 1 of the NMDS ordination (Fig. 2a).

Fig. 2
figure 2

NMDS based on a fungal and b prokaryotic community dissimilarities. Blue, green, and red symbols represent the Wet, Med (medium), and Dry sites, respectively. Circles, triangles, squares, and diamonds represent samples collected in September 2015, April 2016, June 2016, and September 2016, respectively. Only significant vectors are plotted on the ordination. Black vectors represent soil properties. Dark gray vectors represent relative abundances of trophic groups or predicted gene abundances. WC water content, TN total N content, TC total C content, C:N total C:N ratio; NO3 nitrate N content, DON dissolved organic N content, DOC:DON dissolved organic C:N ratio

Full size image

The relative abundances of saprotrophic and pathotrophic fungi differed significantly among sites (Fig. 3, Table S4). The relative abundance of saprotrophic fungi was highest and second highest at the Wet and Med sites, respectively. The relative abundance of pathotrophic fungi was highest at the Dry and Med sites. Saprotrophic fungal community composition also changed along the aridity gradient, with an apparent reduction in the relative abundance of the order Mortierellales as dryness increased (Fig. S1; Table S4). In our NMDS ordination plot, the vectors for the relative abundances of saprotrophic fungi and Mortierellales were more highly correlated with the vectors for soil total N content than with those for water content and pH, and they were also strongly negatively correlated with dissolved organic C:N ratio (Fig. 2a).

Fig. 3
figure 3

Relative abundances of fungal trophic groups based on FUNGuild database trophic modes. The values in the bar graphs are the averages. The FUNGuild algorithm frequently outputs multiple trophic modes for single OTUs (e.g., saprotroph–symbiotroph); in these cases, the OTUs were placed in the “Others” category

Full size image

Soil prokaryotic communities

The soil 16S rRNA gene abundances of bacteria and archaea did not differ significantly among sites (Table S2), but the PerMANOVA showed that the soil prokaryotic community structure differed significantly by site along the aridity gradient (F = 14.9, partial R2 = 0.36, P < 0.001, Table S3). The NMDS also identified clear differences in prokaryotic community structures among sites (Fig. 2b).

The predicted relative abundances of genes associated with N-acetylglucosaminidase and chitinase differed significantly among sites and were highest at the Wet and Med sites, respectively (Fig. 4). However, the predicted relative abundances of genes associated with cellulase did not differ among sites. The relative abundances of the dominant copiotrophic bacteria, Betaproteobacteria, Gammaproteobacteria, and Bacteroidetes and its classes (Cleveland et al. 2007; Fierer et al. 2007), were significantly different between sites (Fig. S2, Table S5) and were highest in the Wet site. In contrast, the relative abundance of the second-ranking Actinobacteria significantly differed among sites and was highest in the Med site. In our NMDS, the vectors of predicted abundances of genes associated with N-acetylglucosaminidase pointed in directions consistent with the vector for dissolved organic N content and inorganic N content (Fig. 2b).

Fig. 4
figure 4

Predicted abundances of genes associated with soil N decomposition (cellulase, chitinase, and N-acetylglucosaminidase) enumerated using the Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt) pipeline procedures. Values are means ± SDs. Significant differences among sites, identified by one-way ANOVA with the sampling date as a random effect, are indicated in the upper right corner (***P < 0.001). Different lowercase letters indicate significant pairwise differences between sites (Tukey–Kramer or Steel–Dwass test)

Full size image

Soil ammonia-oxidizing communities

Abundances of amoA genes in the soil differed significantly among sites for bacteria, but not for archaea (Fig. 5). The ammonia-oxidizing bacterial abundances were significantly positively and negatively correlated with soil moisture and pH, respectively (Table 2). Ammonia-oxidizing archaea abundance was not correlated with soil moisture, pH, or ammonium N content.

Fig. 5
figure 5

Logged values of soil bacterial and archaeal amoA gene abundances. Values are means ± SDs. Significantly different values among sites, identified by one-way ANOVA with the sampling date as a random effect, are indicated in the upper right sector of the figure (*P < 0.05, ***P < 0.001). Different lowercase letters above the bars indicate significant pairwise differences between sites (Tukey–Kramer or Steel–Dwass test)

Full size image
Table 2 Coefficients for relationships between ammonia-oxidizer abundances and soil properties, based on single correlation tests and a generalized linear mixed model with the site and sampling date as random effects (in parentheses)
Full size table

Discussion

Soil N dynamics changes along the aridity gradient

Each of the soil extractable N contents, i.e., dissolved organic N, ammonium N, and nitrate N, was reduced with increasing aridity (Fig. 1), which was contrary to our expectation that the content of a specific form of N would remain constant or increase with increasing aridity due to the bottleneck step. Furthermore, nitrite N did not accumulate in the study sites, as expected. The change in net nitrification rate was consistent with the decrease in nitrate N content (Table 1). The net N mineralization rate by laboratory incubation did not differ significantly among sites, although the ammonium N and total inorganic N contents decreased with increasing aridity. N mineralization may be activated after a rainfall event, as suggested in previous studies in which high microbial activity was observed after the re-wetting of dry soil (Franzluebbers et al. 1996; Mikha et al. 2005). Therefore, N mineralization activity in the field may have been higher after rainfall, and rainfall frequency increases from the southeast to northwest on the Loess Plateau (Wan et al. 2014); therefore, the Wet site likely had more frequent rainfall and higher N mineralization activity in the field. According to the pattern of soil extractable N contents, all N degradation steps were affected by the aridity gradient. Substrate accumulates when a bottleneck develops in the stepwise degradation process, but we did not detect accumulated forms of N, suggesting that changes maintained a balance, with no evidence of a continuous bottleneck along the aridity gradient.

Degradation of organic matter by the fungal community

Fungal abundance did not decrease with increasing aridity (Table S2), although dissolved organic N content which was primarily produced by fungal community decreased with increasing aridity. However, fungal community composition changed with the aridity level (Figs. 2a, 3). The relative abundances of saprotrophic fungi, especially that of the dominant order Mortierellales, decreased markedly with increasing aridity (Figs. 2a, 3, S1). As saprotrophs obtain C energy from decomposition and Mortierella-dominated communities have high C-degrading potential (e.g., phenol-oxidase production) (Phillips et al. 2014), the decreases in these groups should reduce the degradation rate and contribute to the increase in the C:N ratio of dissolved organic matter. Pathotrophs, which alternatively increased in abundance along the aridity gradient (Fig. 3), acquire energy from host cells (Nguyen et al. 2016); therefore, the group cannot contribute to decomposition.

Reductions in the abundances of saprotrophs such as Mortierellales with increasing aridity may be caused by reductions in the content of soil organic matter: we found a strong positive correlation between (i) saprotroph and Mortierellales relative abundances and (ii) soil total N contents (Fig. 2a). This finding is consistent with those of other studies, in which increases in saprotrophic fungi and taxa in Mortierellales with addition of organic matter have been observed (Hanson et al. 2008; Kao-Kniffin and Balser 2008; Chigineva et al. 2009), and is further supported by the apparent decrease in the amount of plant input and soil organic matter content with increasing aridity (Table S1). Moreover, fungal communities are reportedly resilient to changes in water content or pH (Rousk et al. 2010; Barnard et al. 2013; de Vries et al. 2018). Thus, the decrease in organic matter content along the aridity gradient likely reduced saprotrophic fungi, particularly Mortierellales, thereby decreasing the dissolved organic N content and increasing the dissolved organic C:N ratio. In contrast, it has been reported that soil moisture content determines fungal decomposition activity which may control soil organic matter content (Hawkes et al. 2011; Cregger et al. 2012). However, if fungal degradation decreased at our study sites due to soil dryness, then soil organic matter content would have increased, not decreased, with increasing aridity. Therefore, organic matter content was a limiting factor of fungal degrading ability on the aridity gradient in the current study.

N mineralization by the soil prokaryotic community

Prokaryotic abundance did not decrease with increasing aridity (Table S2), although ammonium N content, which was produced mainly by prokaryotic community, decreased with increasing aridity. However, community composition changed with aridity, which was also the case for the degradation step (Figs. 2b, S2). The Wet site had the highest N-acetylglucosaminidase production potential for labile compound decomposition, whereas the Med site had the highest chitinase production potential for recalcitrant compound decomposition (Fig. 4). Consistently, the relative abundances of copiotrophs such as the Betaproteobacteria, Gammaproteobacteria, and Bacteroidetes were highest in the Wet site (Fig. S2; Table S5). And that of Actinobacteria, which are strongly competitive in the utilization of recalcitrant compounds such as chitin, was highest at Med site (Bastian et al. 2009; Bell et al. 2013; Cretoiu et al. 2013). Active decomposition of labile compounds by abundant copiotrophs at the Wet site and decomposition of recalcitrant compounds by Actinobacteria at the Med site were key elements in the N mineralization step gradient along the aridity gradient.

The Wet site had the highest dissolved organic N content and the lowest dissolved organic C:N ratio; such organic matters would be suitable substrates for N-acetylglucosaminidase. The Med site would contain relatively more high-molecular-weight organic compounds such as chitin, the substrate of chitinase, based on that the Med site had relatively low dissolved organic N contents and a relatively high dissolved organic C:N ratio (similar to the Dry site), but relatively high total C and N contents (similar to the Wet site; Table 1). Increasing pH and decreasing moisture with increasing aridity may reduce N-acetylglucosaminidase production potential by limiting the growth of copiotrophs such as Betaproteobacteria (Lauber et al. 2009; Lennon et al. 2012; Armstrong et al. 2016). However, Actinobacteria are resilient to low moisture and alkaline soil (Lauber et al. 2009; Placella et al. 2012; Barnard et al. 2013). The increase in Actinobacteria in the Med site are difficult to explain without considering organic matter quality and quantity, because the Med site had medium moisture and pH. Our findings agree with previous observations that the response of bacterial communities to soil water conditions is mediated by organic matter type and availability (de Vries et al. 2012; Placella et al. 2012; Göransson et al. 2013).

Nitrification

The gradient in nitrification was likely related to the activities of the ammonia oxidizer communities. Ammonia-oxidizing bacterial abundance decreased along the aridity gradient, but the abundances of ammonia-oxidizing archaea did not differ among sites (Fig. 5), as observed in previous studies (Adair and Schwartz 2008; Chen et al. 2013). The frequently observed positive relationship between the net nitrification rate and ammonia-oxidizing bacterial abundance (Di et al. 2009; Chen et al. 2013; Banning et al. 2015) suggests that nitrification is driven mainly by ammonia-oxidizing bacteria at this study site. However, we did not test differences in archaeal versus bacterial ammonia oxidation activity through addition of octyne as previously described (Taylor et al. 2013).

Ammonia-oxidizing bacterial abundance was significantly positively correlated with soil moisture and negatively correlated with pH, but the abundance of these microbes was not related to ammonium N content (Table 2). The positive correlation between the abundance of this group of prokaryotes and moisture content was consistent with the findings of previous studies (Gleeson et al. 2010; Placella and Firestone 2013; Marcos et al. 2016), but the negative correlation with pH was not (Hu et al. 2013). Ammonium N content has been shown to determine ammonia-oxidizing bacterial abundance with a tenfold gap in the ammonium N content (Verhamme et al. 2011; Wang et al. 2011). Hence, the determining factor in ammonia-oxidizing bacterial abundance at this study site was probably soil moisture and not substrate ammonium N, unlike in upstream steps. Soil moisture obviously declined along the aridity gradient, thereby reducing ammonia-oxidizing bacterial abundance and creating the nitrification step gradient.

Predicting shifts in N transformation with changing rainfall pattern

In summary, our findings suggest that substrate quality and quantity strongly affect the degradation and N mineralization steps by changing fungal and prokaryotic community compositions. Although soil moisture or pH may contribute to changes in the fungal and prokaryotic community, it is difficult to explain the changes observed in these communities without substantial effects of their substrate quality and quantity. However, in the nitrification step, soil moisture appeared to be more important than substrate for the ammonia-oxidizing bacteria, although ammonia-oxidizing archaeal population was resilient to changes in soil moisture. Hence, as rainfall patterns change, N degradation and mineralization will proceed well unless the soil organic matter content changes, but the nitrification rate will slow as moisture decreases. Substrate and soil moisture reportedly determine N mineralization and nitrification rates (Saetre and Stark 2005; Ford et al. 2007; Marcos et al. 2016), but substrate has a greater effect on N mineralization than does moisture, which has a greater effect on nitrification (Yahdjian and Sala 2010; Hu et al. 2015), as demonstrated in this study.

Furthermore, the vulnerability of nitrification to soil moisture depletion resulting from climate change may control whole-forest N cycling. A decrease in N availability will limit N uptake, NPP and litter production (Norby et al. 2010; Ren et al. 2017; Tateno et al. 2017), consequently reducing soil organic matter content (Fig. 6). This reduction in soil organic matter content imposes substrate limitation on the fungal community, which will limit the whole N dynamics sequence through mineralization (Fig. 6). Thus, precipitation decreases may limit whole-forest N cycling, beginning with nitrification limitation, via the plant–soil feedback process. In this study, we observed no accumulation of specific N forms, perhaps because limitation in the nitrification step extended to the other steps; hence, the N cycling process was maintained in particular stable states, each suited to the precipitation at the different sites. Thus, we found balanced changes in extractable content of all N forms along the aridity gradient. The nitrification step is likely to become a temporary bottleneck step with increasing aridity. However, we expect the whole N cycling process to achieve a new stable state suited to the decreased precipitation pattern under future climate change.

Fig. 6
figure 6

Conceptual N cycling model based on results from this study. Black solid arrows and black thick dashed arrows refer to the effects observed in this study and the predicted effects, respectively. Aridity determines the nitrification step, which is controlled by soil moisture and ammonia oxidizers. The nitrification step would affect N uptake and net primary production (NPP), consequently controlling the degradation and mineralization steps by changing their substrates and drivers. Thin black dashed arrows indicate other possible effects that were not strongly supported in this study

Full size image