Introduction

The global rate of anthropogenic nitrogen (N) addition (210 Tg N a−1) is currently greater than the supply by natural processes (203 Tg N a−1; Fowler et al. 2013) as human activities have disturbed the global N cycle (Gruber and Galloway 2008). Chronic N deposition has several effects on net primary productivity, biodiversity, the global carbon cycle, and climate (Erisman et al. 2013). The effects of increased N deposition on global biodiversity, especially on soil microorganisms in forest ecosystems (Wang et al. 2018a), have received increased attention. This is because soil microorganisms, which represent a high proportion of terrestrial biodiversity, affect major ecosystem processes, including nutrient cycles (Canfield et al. 2010), plant growth (van der Heijden et al. 2008), and soil structure (Rillig and Mummey 2006).

Although numerous studies have examined the responses of soil microorganisms to long-term N addition, there remain unresolved issues. The effects of increased N on soil microbial diversity vary among forest ecosystems, with some studies showing that added N can have positive (Haas et al. 2018) or negligible (Contosta et al. 2015) effects. In contrast, N deposition has resulted in dramatic decreases in microbial diversity in cold temperate forests (Yan et al. 2018), in bamboo plantations (Li et al. 2016), and in boreal forests (Allison et al. 2007). The effects of added N on microbial communities also vary, with some studies showing that these communities are altered (Zeng et al. 2016) or remain unaltered (Li et al. 2019). Most studies of N addition have been conducted in boreal and temperate forest ecosystems (WallisDeVries and Bobbink 2017). In contrast, tropical and subtropical forests are often referred to as N-rich and phosphorus (P)-poor (Walker and Syers 1976; Vitousek and Farrington 1997). N and P are necessary for the development and reproduction of plants. These two nutrients are generally in a dynamic equilibrium in natural environments (Elser et al. 2007), but increases in one can induce a limitation in the other (Elser et al. 2007). Some studies have shown that adding N has little effect on plant productivity in N-saturated forests (Harrington et al. 2001; Cusack et al. 2011), but can alter soil C sequestration (Zak et al. 2017; Lu et al. 2021b). Plant productivity might be nutrient limited, while soil microbes always are carbon limited in natural ecosystems (Soong et al. 2020). Tropical and subtropical forests thus warrant increased research attention.

Bacteria and fungi coexist in the natural environment, and there is substantial nutrient and energy exchange between them (Benoit et al. 2015). The ecological functions of microorganisms can be performed via the formation of an effective interaction co-occurrence network rather than by a single species (Pande and Kost 2017). In these networks, the interactions between microbes are diverse and include both cooperative and competitive relationships (Faust and Raes 2012). Yuan et al. (2021) reported that climate warming increases the complexity of soil microbial networks. More intense microbial interactions after N addition have been reported by Li et al. (2021). These findings suggest that soil microbial interactions likely play a key role in forest management.

In this study, the effects of seven years of N addition on soil bacteria and fungi in a subtropical evergreen broad-leaved old-growth forest with rich in N but poor in P were reported, using metagenomic sequencing. The objectives were: (1) to clarify the effects of N and P addition on the diversity and community composition of soil microbes; (2) to identify the bacterial and fungal taxa sensitive to nutrient addition; and (3) to clarify the effects of adding N on microbial co-occurrence networks. The same data as that in Dai et al. (2021) were used, but information on soil fungi was included, which are believed to be potentially more sensitive to N addition than bacteria (Wang et al. 2021). Further investigation off the responses of microorganisms to N addition at the species level was carried out using identifying biomarkers and creating bacterial and fungal co-occurrence networks, rather than only analyzing microbial community diversity.

Materials and methods

Study site description and experimental design

An N deposition experiment was conducted in a 200-year-old, evergreen broad-leaved forest on Xianyu Mountain (30° 01′47″ N, 117° 21′23″ E), Shitai County, Anhui Province, eastern China. The site has a humid subtropical monsoon climate, with an annual average temperature of 16 °C and precipitation of 1,626 mm (Dai et al. 2021). The dominant species in the stands include Castanopsis eyrei (Champ. ex Benth.) Hutch. and Cyclobalanopsis glauca (Thunberg) Oersted; the main associated species are Symplocos spp. and Pinus massoniana Lamb.; the main understory shrubs are Camellia cuspidata (Kochs) Wright ex Gard., Loropetalum chinense (R. Br.) Oliver, Tricyrtis macropoda Miq., and various species of ferns. The soil is a ultisol or red clay soil, according to the United States Department of Agriculture (USDA) Soil Taxonomy classification and has developed from a regolith of granites; the pH is acidic.

The field experiment was initiated in June 2011 and consisted of four treatments: control (CK, neither N nor P added), low N (LN, 50 kg N ha−1 a−1), high N (HN, 100 kg N ha−1 a−1), and combined N and P additions (HNP, 100 kg N ha−1 a−1 + 50 kg P ha−1 a−1). The experiment was conducted in a randomized block design. There were three replicates 20 m × 110 m blocks separated by a 10-m buffer, each containing four treatments and a total of 12 experimental plots (20 m × 20 m) separated by 10 m. From 2011 to 2018, NH4NO3 and Ca(H2PO4)2 solutions dissolved in 20 L of water to simulate wet deposition, were applied bimonthly using a backpack sprayer. Each control plot received equivalent amounts (20 L) of water corresponding to an increase of 0.3 mm a−1 precipitation.

Soil sampling and physicochemical analysis

After seven years of nutrient addition, soil samples were collected in July 2018. Within each plot, samples of the upper mineral soil (0–10 cm) were collected according to the five-point method using a steel auger after removing the litter layer and were evenly combined. Twelve samples were collected and transported to the laboratory and passed through a 2-mm steel sieve to remove stones and roots. One portion of the samples was kept at 4 °C for chemical and physical properties analysis; another was kept at − 80 °C for DNA sequencing.

The following physicochemical characteristics were quantified: pH, water content (SWC), ammonium (NH4+ − N), nitrate (NO3− − N), organic carbon (SOC), total N (TN), total phosphorus (TP), dissolved organic carbon (DOC), dissolved organic N (DON), microbial biomass carbon (MBC), and N (MBN). pH was measured using a pH meter in 2.5:1 water/soil suspensions. SWC was determined by oven-drying fresh soil to a constant mass at 105 °C. Available N (NH4+ − N and NO3− − N) were extracted from fresh soil samples (30 g) with 50 mL of 1 M KCl solution and measured with a flow injection autoanalyzer (FIAStar 5000, FOSS, Sweden). A 1-g air-dried sample was digested in a 10 mL H2SO4 − HClO4 (3:1) solution, and TP in the extract was measured by a flow injection autoanalyzer based on molybdenum blue chemistry. SOC and TN were determined using a CHN autoanalyzer (EA 3000, EuroVector, Italy). The chloroform-fumigation-extraction method was used to calculate MBC and MBN (Vance et al. 1987). Fresh subsamples (30 g) were extracted using a 100-mL 0.5 M K2SO4 solution, shaken for 30 min, and filtered. The subsamples were fumigated with ethanol-free chloroform for 24 h at 25 °C in a vacuum desiccator; the other subsamples were treated without fumigation. DOC and DON were quantified using a TOC analyzer (Multi N/C 3100, Jena Analytik AG, Germany). δMBC (mg kg−1) was estimated using the following:

$$\delta_{{{\text{MBC}}}} = \delta_{{{\text{CFE}}}} /0.{45}$$
(1)

where δCFE refers to the difference in DOC content between fumigated and non-fumigated extracts. δMBN (mg kg−1) was estimated using the following formula:

$$\delta_{{{\text{MBN}}}} = \delta_{{{\text{NFE}}}} / \, 0.{54}$$
(2)

where δNFE is the difference in DON content between fumigated and non-fumigated extracts (Wu et al. 1990).

DNA extraction, library construction, sequencing, assembly, and annotation

Metagenomic sequencing was used to analyze soil microbial communities which precludes the need to cultivate species (Chen and Pachter 2005), and overcomes limitations of amplicon sequencing (Sharpton 2014). Metagenomic sequencing permits nutrient-susceptible biomarkers to be precisely identified and to predict the ecological effects of N deposition. In this study, the taxonomic information from our metagenomes were used for analysis.

An E.Z.N.A. soil DNA Kit (D5625-02, Omega, Inc. USA) was used to extract genomic DNA from different samples. DNA was extracted using 50 μL of elution buffer, and stored at − 80 °C until further analysis. The extracted DNA samples were submitted to LC-Bio Technology Co. Ltd. (Hangzhou, China) for sequencing.

A DNA library was constructed using the TruSeq Nano DNA LT Library Preparation Kit (FC-121-4001). DNA was fragmented for 30 min at 37 °C with dsDNA fragmentase (NEB, M0348S). Fill-in reactions and exonuclease activity produced blunt-end DNA fragments. An A-base was added to the blunt ends to facilitate ligation to the indexed adapters. The DNA fragments with ligated adapters were amplified by PCR via thermal-cycling conditions: initial denaturation at 95 °C for 3 min; eight cycles of denaturation at 98 °C for 15 s, annealing at 60 °C for 15 s, and extension at 72 °C for 30 s; and a final extension at 72 °C for 5 min. Amplification of DNA fragments was performed with universal PCR primers (forward primer: 5′-AATGATACGGCGACCACCGAGATCTACAC [i5] TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGCCTACGGGNGGCWGCAG-3′ and reverse primer: 5′-CAAGCAGAAGACGGCATACGAGAT [i7] GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGACTAC-HVGGGTATCTAATCC-3′). After amplification, purification, quantification, and quality control, the DNA library was sequenced on an Illumina-Hiseq 4000-PE 150 platform (Illumina Inc., San Diego, CA, USA).

Raw sequencing reads were preprocessed to obtain clean reads before subsequent data analysis. Joint reads were removed using Cutadapt version 1.9 and low-quality reads truncated using fqtrim version 0.94 with a window size of 6 bp. The samples were assembled by SPAdes version 3.7.0 to generate a FASTQ file, and each sequence was referred to as a contig; these were used to predict coding regions (CDs) with MetaGeneMark version 3.26. CDs were clustered by CD-HIT version 4.6.1 to obtain unigenes. Abundance information was determined using the transcripts per million method based on the number of aligned clean reads against a database of gene sequences using Bowtie2 version 2.2.0 (Langmead and Salzberg 2012). DIAMOND version 0.7.12 with an e-value threshold of 1 × 10−5 was used to align the unigenes against the NCBI-nr version July 2016 database of non-redundant protein reference sequences to identify species. Based on taxonomic classification from the NCBI taxonomy database, specific species annotation information at each classification level was obtained using the least common ancestor algorithm (Huson et al. 2011).

Statistical analysis

Hill numbers, which are a unified family of diversity indices, were used to quantify soil bacterial and fungal diversity:

$${}^{q}D = \left( {\sum\limits_{i = 1}^{S} {p_{i}^{q} } } \right)^{1/(1 - q)}$$
(3)

where qD denotes the Hill number, the exponent and superscript q is the order of the diversity, S the number of species, and pi the relative abundance of species i in the samples (Jost 2006). When q = 0, 0D is identical to the species richness. When q = 2, 2D is equal to the inverse Simpson index and can be interpreted as the effective number of dominant species (Tuomisto 2011; Daly et al. 2018). This function is undefined for q = 1, but 1D can be calculated by modifying q to 0.99999 (Alberdi and Gilbert 2019). 1D corresponding to the limit of the unity is the analog of the Shannon index and can be considered as the effective number of typical species. Additionally, −1D (q =  − 1) and 3D (q = 3) represent rare and common species, respectively. The comparisons of differences among groups are examined by multiplicity-adjusted P-values to avoid increasing the rate of false positives using the “simboot” package in R software (Pallmann et al. 2012). Linear mixed models (LMM) were used to test the effect of treatments on soil properties. The treatments were included as fixed factors, and a block was included as a random factor using the function “lmer” in “lme4” the package in R (Bates et al. 2015; Piepho 2018; Piepho and Edmondson 2018). The residuals were examined by the Shapiro test to check assumptions of a one-way analysis of variation (ANOVA) (Kozak and Piepho 2018). Post-hoc tests between treatments were performed as all-vs-all comparisons using the package “emmeans” in R (Lenth et al. 2023), and statistical differences were deemed significant at P < 0.05. All data are presented as mean ± standard deviation.

Non-metric multidimensional scaling (NMDS) is a rank–order–mapping method that makes no assumptions about the underlying data structure. The rank–order relationships between ordination distances and original distances are iteratively improved using an algorithm of successive approximations to obtain an optimized map of n individuals in a low-dimensional ordination space. In most cases, microbial data are nonlinear, and NMDS is the appropriate model for these data (Kenkel et al. 2002; Ramette 2007). In this study, NMDS based on the Bray–Curtis distance at the species level was used to investigate dissimilarity in the composition of bacteria and fungi among treatments. Permutational multivariate analysis of variance (i.e., PERMANOVA, 999 permutations) was conducted to evaluate statistical significance using the function “adonis” in the R package “vegan” (Oksanen et al. 2022).

Linear discriminant analysis (LDA) effect size (LEfSe) is an algorithm for high-dimensional biomarker discovery and explanation that can be used in metagenomic analyses. Because the relative abundance in metagenomic data frequently violate the main assumptions of typical parametric tests, the non-parametric tests used by LEfSe are more reliable (Segata et al. 2011). In our study, the LEfSe algorithm was used (https://huttenhower.sph.harvard.edu/galaxy) to identify potential biomarkers based on various taxonomic levels (from supergroup to species) in soil microbes. Specifically, we used the non-parametric factorial Kruskal–Wallis rank-sum test to detect significant differential abundance, and LEfSe then estimated the effect size of each differentially abundant taxon and determined the taxa that were significantly affected by N addition. The factorial Kruskal–Wallis test was conducted using an alpha value of 0.05, and the threshold of the logarithmic LDA score was 2.0.

Four co-occurrence networks were created for the treatments by integrating information on the bacterial and fungal species in each group. Those with average relative abundances > 0.1% were retained from each group. Pearson correlation analysis between microbes was conducted at the species level using the package “psych” in R (Revelle 2022). Pearson correlation coefficients greater than 0.6 were considered statistically robust. Gephi version 0.9.2 was used to build the co-occurrence networks and determine their topological properties. The keystone microbes were recognized based on their within-module connectivity (Zi) and among-module connectivity (Pi) values. We defined the nodes as network hubs (highly connected nodes within the entire network, Zi ≥ 2.0 and Pi ≥ 0.6), module hubs (highly connected nodes within modules, Zi ≥ 2.0 and Pi < 0.6), connectors (nodes that connect modules, Zi < 2.0 and Pi ≥ 0.6), or peripherals (nodes connect in modules with few connections to the outside world, Zi < 2.0 and Pi < 0.6).

Results

Changes in soil physicochemical properties

The pH was significantly lower (P < 0.01) and DOC higher (P < 0.01) in the nutrient addition treatments than in the controls (Table 1). However, there were no significant differences in SWC, SOC, TN, TP, DON, NH4+ − N, NO3− − N, MBC, MBN, and SOC/TN among treatments (Table 1).

Table 1 Effects of nutrient addition on soil physicochemical properties with linear-mixed effects
Full size table

Soil microorganisms

Bacterial community

High-quality sequences were clustered into 9230 bacterial species. The shapes of the soil bacterial Hill number curves for the different treatments were similar (Fig. 1a). The multiplicity-adjusted P-values from q = − 1 to q = 3 were not significant, and some of them were far from the significance threshold (P > 0.05; Table S1).

Fig. 1
figure 1

Soil microbial diversity profiles based on abundance of each species; x-axis is the order q and y-axis the Hill numbers. a Bacterial diversity; b fungal diversity

Full size image

The most abundant bacterial phyla across all samples were Proteobacteria (an average of 51.5% of all unigenes), followed by Acidobacteria (26.0%) and Actinobacteria (4.1%), which together comprised 81.5% of bacterial unigenes (Table S2). With the exception of Gemmatimonadetes, nutrient addition had no significant effect (P > 0.05) on bacterial phyla (Table S2). The addition of nutrients had similar effects on Hill numbers; the dissimilarity of the bacterial community at the species level did not significantly differ between the treatments and control according to PERMANOVA (P = 0.36; Fig. 2a).

Fig. 2
figure 2

NMDS ordination plot showing the dissimilarity of the composition of microbial species between different treatments based on Bray–Curtis (a, stress = 0.06; b, stress = 0.12) distance matrices; a dissimilarity of the bacterial community; b dissimilarity of the fungal community

Full size image

LEfSe analysis was conducted to identify the taxa affected by the addition of nutirents; nine biomarkers belonging to two phyla were affected (P < 0.05, LDA > 2.0; Fig. 3a). Eight taxa of the phylum Gemmatimonadetes were enriched in the low N group, including class Gemmatimonadetes, the orders Gemmatimonadales and Gemmatimonadetes_noname, the family Gemmatimonadetes_noname, the genus Gemmatimonadetes_noname, and the species Gemmatimonas_sp._AP64 and Gemmatimonadetes_bacterium_KBS708. The species Acinetobacter_sp._CIP_64.2 in the phylum Proteobacteria was sensitive to combined N and P addition (Fig. 3a, Table S3).

Fig. 3
figure 3

LEfSe cladograms showing the soil microbial biomarkers sensitive to nutrient addition; a bacterial biomarkers; b fungal biomarkers. The cladograms show the supergroup, phylum, class, order, family, genus, and species levels from inside to outside; the nodes indicate taxa belonging to the taxonomic level. Taxa with significantly greater relative abundance in one group are shown by different colored nodes; taxa with no significant changes in the treatments are represented by yellow nodes

Full size image

Fungal community

A total of 355 fungal species were identified from the samples. Soil fungal Hill numbers at all values of q were lower in the controls than in the nutrient addition treatments (Fig. 1b). The addition of nutrients did not have any significant effect on Hill numbers (Table S1). The most abundant fungal phyla were Ascomycota (average of 67.2% of the unigenes) and Basidiomycota (23.4%), together comprising 90.6% of the fungal unigenes detected (Table S4). In addition, adding nutrients did not result in significant shifts in the dissimilarity of the fungal community (P = 0.69; Fig. 2b).

Seven nutrient-susceptible fungi of two phyla were identified by LEfSe analysis (P < 0.05, LDA > 2.0; Fig. 3b). The sensitive fungal taxa were in the order Dothideales, the family Aureobasidiaceae, and the genus Aureobasidium in the HN group; the genera Chaetomium and Nosema in the LN group; and the genera Brettanomyces and Naumovozyma in the controls (Fig. 3b, Table S5).

Co-occurrence of bacteria and fungi by network analysis

To identify the responses of microbe interactions to nutrient additions, four co-occurrence networks were created for each treatment (CK, LN, HN, HNP) using bacterial and fungal species with relative abundances (> 0.1%) in each group (Fig. 4). The average clustering coefficient of the networks were 0.016, 0.002, and 0.008 higher in the LN, HN, and HNP group than in the controls, respectively (Fig. 4, Table 2). The low N treatment had the highest number of edges (1,861; 37.8%), the highest average edges per node value (7.85; 34.4%), and the lowest average path length value (11.469; 4.8% lower compared with the controls; Fig. 4, Table 2). In the four networks, the percentage of inter-kingdom edges was higher than the percentage of edges within fungal nodes, which was higher than the percentage of edges within bacterial nodes (Fig. 4, Table 2). The low and high N addition treatments increased the percentage of positive edges to 63.3% and 62.1%, respectively, and the combined N and P treatment decreased the percentage of positive edges to 51.6% (Fig. 4, Table 2).

Fig. 4
figure 4

The networks reflecting the effects of nutrient addition on the co-occurrence patterns between bacteria and fungi at the species level. The size of nodes is proportional to the relative abundance of species, with blue nodes representing bacteria and purple nodes representing fungi. Red links indicate positive correlations between two nodes, and blue links negative correlations

Full size image
Table 2 Topological indices of bacterial and fungal networks in the four treatments: control (CK), low N (LN), high N (HN), and combined N and P (HNP)
Full size table

To explore the topological roles of microbial taxa in the networks, four categories were established according to their Zi and Pi values: peripherals, module hubs, network hubs, and connectors. Most nodes were peripherals in each network (Fig. 5). Three module hubs were discovered in the networks, which originated from various groups: Mycosphaerella pini within Ascomycota in LN, Melanopsichium pennsylvanicum within Basidiomycota in HN, and Zymoseptoria tritici within Ascomycota in HNP, and were considered potential keystone taxa because of their key roles in the networks (Fig. 5). The relative abundances of these module hubs were relatively low, 0.3%, 0.4% and 0.6% in LN, HN, and HNP, respectively. No network hubs or connectors were detected.

Fig. 5
figure 5

Distribution of bacterial and fungal species based on their network roles which was used to identify keystone species within networks. Network nodes were categorized as peripherals, module hubs, network hubs, or connectors. x − Pi represents among-module connectivity, and the y − Zi represents within-module connectivity. Network hubs: Zi ≥ 2.0, Pi ≥ 0.6; module hubs: Zi ≥ 2.0, Pi < 0.6; connectors: Zi < 2.0, Pi ≥ 0.6; peripherals: Zi < 2.0, Pi < 0.6

Full size image

Discussion

Responses of soil physicochemical properties to nutrient addition

The pH at a depth of 10 cm in the nutrient treatments decreased substantially after seven years relative to the controls, indicating that long-term addition of nutrients will lead to soil acidification. This finding is consistent with some studies showing that N deposition lowers the pH in N-rich tropical (Wu et al. 2019) and subtropical soils (Lu et al. 2021a). N enrichment contributes to reductions in pH as a result of the following processes: (1) release of H+ when NH4+ transforms into NO3− via nitrification; (2) release of H+ when plants absorb NH4+; (3) loss of base cations resulting from the exchange of NH4+; and (4) leaching of NO3− (Chen et al. 2016). In the present study, the addition of N significantly increased the quantity of dissolved organic carbon (DOC) in the upper 10-cm layer. Similar to our results, Fang et al. (2014) found that DOC in surface layers were strongly increased by the addition of N. This may be because added N promotes plant growth, leading to increased organic matter inputs to soil through litterfall and root exudates that favor the accumulation of DOC (Portillo-Estrada et al. 2013). In addition, a lower pH may reduce the capacity of soils to stabilize SOC as a result of reductions in the (negative) surface charge of minerals (Fissore et al. 2008), causing more organic carbon to be converted into DOC. The levels of soil NH4+ − N and NO3− − N were not affected by the addition of nutrients in our experiment, possibly because NH4NO3 is absorbed by plants and microbes or lost through denitrification, volatilization, and/or leaching.

Response of microbial communities to nutrient addition

In this study, the predominant bacterial phyla were Proteobacteria, Acidobacteria, and Actinobacteria, and the major fungal phyla were Ascomycota and Basidiomycota. These phyla are widely distributed in acidic forest soils (Wang et al. 2018b). The diversity and community composition of these microbes were not affected by the addition of nutrients. In other regions, such as temperate forests in northeastern China, N input reduced soil bacterial richness and altered microbial community structure (Yan et al. 2018). One possible explanation for these contrasting findings might be that the responses of microorganisms to N deposition in N-rich subtropical forest soils differed from those in N-deficient soils. Compared with previous experiments carried out on N-rich soils, such as N addition (150 kg N ha−1 a−1) in the Dinghushan Nature Reserve (Wang et al. 2018b), the amounts of added N (100 kg N ha−1 a−1 for the highest level) applied in our study might not be enough to cause significant changes in soil microbe communities. Some studies reported differences in microbial composition between primary and secondary forests (Li et al. 2019) and between old-growth and young forests (Dai et al. 2021), and this might result in a greater resilience of microbes in old-growth forests to nutrient additions. Furthermore, seasonal and interannual differences in microbial communities were greater than the effects of added N for years to decades (Gutknecht et al. 2012; Contosta et al. 2015), indicating that microbial communities might have adapted to their historical environment, and long-term experiments over decades might be required to observe shifts in microbes.

Microbes sensitive to nutrient addition

Nine bacterial biomarkers in the phyla Gemmatimonadetes and Proteobacteria were sensitive to nutrient addition according to LEfSe analysis. However, no changes in overall microbial community composition were detected. Bacteria of the phylum Gemmatimonadetes are frequently detected in soils, with their relative abundance ranging from 0.2 to 6.5% (DeBruyn et al. 2011), which is consistent with the abundance of Gemmatimonadetes in our study (0.5%). Members of Gemmatimonadetes have been described as copiotrophic organisms, ones living in nutrient-rich environments (Li et al. 2022), and they were enriched by the addition of N in our study. Some studies have shown that taxa of the phylum Gemmatimonadetes can be antagonistic against plant pathogensand have thus been suggested to be beneficial bacteria (Yin et al. 2013; Fernández-González et al. 2020; Liu et al. 2020). Acinetobacter_sp._CIP_64.2 in the class Gammaproteobacteria in the phylum Proteobacteria was enriched in the HNP treatment. Gammaproteobacteria comprise N2-fixing rhizobia, as well as fast-growing copiotrophic bacteria (Leewis et al. 2016).

Seven nutrient-sensitive fungal biomarkers were detected in this study. The rare genera Brettanomyces and Naumovozyma within the Saccharomycetales family in the phylum Ascomycota were relatively enriched in the controls. Yang et al. (2015) found that the relative abundance of Naumovozyma were negatively related to the number of bacterial-feeding nematodes, which have been reported to stimulate N mineralization and promote plant growth (Ferris and Matute 2003; Irshad et al. 2011). Therefore, we speculate that in soils without the addition of nutrients, the growth of bacterivorous nematodes might be suppressed by promoting the enrichment of Naumovozyma fungi. The relative abundances of the genus Chaetomium of the phylum Ascomycota, normally found in soil, increased under low N addition. Several studies have confirmed the antagonistic activities of Chaetomium to pathogenic microorganisms, such as their ability to resist Venturia inaequalis, Verticillium dahliae, Puccinia recondite, Pythium ultimum, and Fusarium nivale (Aggarwal et al. 2004; Fernández-González et al. 2020). Members of the genus Nosema of the phylum Microsporidia, which are widespread in the environment, were abundant in the low N treatment, but the effects of increases in Nosema abundance in forest ecosystems remain unclear. The relative abundances of several N-susceptible taxa in the order Dothideales, the family Aureobasidiaceae, and the genus Aureobasidium were markedly higher under high N additions than in the other treatments. Among the fungi identified, the order Dothideales contained only one family Aureobasidiaceae, one genus Aureobasidium, and four species (A. melanogenum, A. pullulans, A. namibiae, and A. subglaciale), which have been reported to be sources of important metabolites, including pullulan, polymalic acid, and extracellular polysaccharides. They are thus thought to be key for enhancing resistance to pathogenic microorganisms (Di Francesco et al. 2020).

Our findings suggest that N addition can promote the accumulation of copiotrophic bacteria. Several potentially beneficial microorganisms were enriched by nutrient additions, which might promote the development of resistance of soil to disease and resilience to environmental changes. These biomarkers sensitive to nutrient addition however, are low-abundance taxa and long-term experiments are required to determine whether they have an effect on microbial communities.

Response of microbial co-occurrence networks to nutrient addition

Most linkages between species were positive in our four networks, which is consistent with the results of previous studies in natural ecosystems (Shi et al. 2019), as many microbes show high niche overlap and engage in diverse interactions, including cross-feeding, syntrophic interactions, mutualistic interactions, commensalism, co-aggregation, and co-colonization (Faust and Raes 2012). In this study, low N and high N additions increased the percentage of positive edges in the network, and the combined N and P addition increased the percentage of negative edges compared with the control. In general, positive connections (positive edges between nodes) predict cooperative relationships, including cross-feeding, co-aggregation, and co-colonization, and reflect common environmental requirements and shared dispersal barriers (Berry and Widder 2014), whereas negative connections mostly indicate competition for limited resources, different environmental niches, and spatial isolation. The addition of N increased the number of mutualistic microorganisms because it weakened competitive advantage and promoted the development of various trophic levels. In contrast, N and P addition increased competition among microorganisms. We also found that inter-kingdom (between bacteria and fungi) species created more associations than intra-kingdom (within bacteria or fungi) species, indicating that inter-kingdom species constructed mutualistic communities more easily in this study.

More links within fungal nodes than within bacterial nodes were identified in our study, and the three module hubs were all fungi. This might indicate that fungi with hyphae could communicate more effectively within a module, whereas the smaller size and shorter generation times of bacteria resulted in decreased levels of communication (Klein et al. 2016). The mycelia developed by members of Ascomycota and Basidiomycota, pervasive in soils, might explain why the three species from the two phyla were identified as module hubs.

Previous work generally focused on high-abundance microorganisms, but whether these taxa play the most important roles in microbial communities remains unclear. A high number of high- and low-abundant taxa were identified via metagenomic sequencing. Three module hubs were detected, identified as keystone taxa within the networks in the nutrient addition treatments, and their relative abundance ranged from 0.3 to 0.6%, indicating that less-abundant taxa might be more sensitive to the addition of nutrients. Similarly, Coyte et al. (2015) found that rare microbes were keystone species in networks. The number of such species in a network is known to be positively related to network stability. The presence of the three key taxa indicated that the nutrient addition treatment increased the stability of the network.

High average clustering coefficient values, links, links per node, and the short average path lengths of the low nitrogen network indicated that interactions among microorganisms were more complex than in the other treatments. Complexity often results in stability in ecological systems (MacArthur 1955), and macroecological research has shown that network stability increased with complexity (Montoya et al. 2006). A number of studies have indicated that complex soil microbial networks could reduce the incidence of disease (Rybakova et al. 2017; Yang et al. 2017; Fernández-González et al. 2020). Therefore, we predict that low N addition could increase the complexity of microbial networks, which might promote resistance to environmental changes and inhibit soil-borne diseases.

Although our findings enhance understanding of the responses of soil microbes to N deposition in subtropical forests, seven years of the experiment is insufficient for detecting the long-term effects of N deposition on microbes; experiments conducted over longer periods are needed to clarify the responses of microbes to N deposition.

Conclusions

Metagenomic sequencing provided insights into the responses of soil bacteria and fungi to nutrient enrichment. Long-term N addition enriched copiotrophic bacteria and increased the abundances of several potentially beneficial microbes, which might increase their resistance to diseases. Low N addition increased the complexity of microbial networks. Microorganisms in subtropical old-growth forest soils are adapted to changes induced by adding nitrogen, which could explain why there were no notable differences in the diversity and community composition of soil microbes after nutrient addition at our site. Additional studies are needed to validate our hypotheses.