A pot experiment simulating crop rotation with Chinese milk vetch (CMV, Astragalus sinicus L. “Xiangzi No.1”) and rice (Oryza sativa L. “Y liangyou 2108”) was conducted from October 2018 to October 2019 in the greenhouse of the Chongming experimental station of the Yunnan Academy of Agricultural Sciences (25° 21′ 22″ N, 103° 01′ 48″ E). The soil was collected from the Ap horizon (0–20 cm) in a nearby rice field. The paddy soil was classified as an Anthrosols according to the FAO soil taxonomy. The region has a tropical monsoon climate with an average annual temperature of 15°C and precipitation of 1,035 mm. After the removal of visible plant tissue, the soil was air-dried, sieved through a 2 mm mesh, and fully mixed. Mixed soil was placed in plastic pots, and each pot contained 3.0 kg of soil. Soil properties are shown in Supplementary Table S1.

The experiment had three fertilization treatments: (1) NPK: chemical fertilizer N (N at 0.10 g·kg⁻¹ dry soil), (2) GM: chemical fertilizer N and green manure (N at 0.08 g·kg⁻¹ dry soil + CMV fresh green manure at 4.06 g·kg⁻¹ dry soil), and (3) GMC: chemical fertilizer N, green manure and biochar (N at 0.08 g·kg⁻¹ dry soil + CMV fresh green manure at 4.06 g·kg⁻¹ dry soil + biochar at 20 g·kg⁻¹ dry soil). In addition, P₂O₅ (0.05 g·kg⁻¹) and K₂O (0.06 g·kg⁻¹) were added in all treatments. The fertilizers used were urea (N 46%), calcium superphosphate (P₂O₅ 12%), and potassium chloride (K₂O 60%). The properties of green manure and biochar are provided in Supplementary Table S2. Each treatment had eight replicates, with each replicate represented by one plastic pot. Chinese milk vetch was sown in October 2018, and in April 2019, and CMV at the blooming stage, approximately 1 month before the middle rice transplantation, was incorporated into the soil as green manure in GM and GMC treatments. Biochar in GMC and chemical fertilizer in all three treatments were incorporated into the soil at the same time as the CMV, with the water layer kept at 2–3 cm. The soil was drained for a week at the tillering stage and for 2 weeks before harvest. In August 2019, at approximately 90 d after middle rice transplantation, five soil cores were randomly sampled from each pot and then fully mixed to serve as one composite sample. Fresh samples were placed on ice and immediately transported to the laboratory. Soil samples were divided into three parts for soil physicochemical analysis, organic C mineralization, and DNA extraction, respectively.

Soil organic C mineralization was measured in microcosm incubation experiments using an alkali absorption method according to Jiang et al. (2018). Briefly, 30.0 g of each air-dried soil sample was adjusted to 65% moisture content of field capacity and then after 3 d of pre-incubation, incubated in 500 mL jars for 30 d at 25°C. Soil samples were replicated six times, and six control jars (pure silica sand without soil) were used to measure background CO₂ concentration. In each jar, a glass vial containing 0.5 mol·L⁻¹ NaOH was placed to trap CO₂ emitted from the soil. After incubating for 1, 3, 5, 7, 10, 15, 20, 25, and 30 d, the amount of CO₂ trapped in the alkali solution was measured by titration. The remaining alkali was titrated to pH 7 with 0.4 mol·L⁻¹ HCl after precipitating carbonate with 20 mL of 1 mol·L⁻¹ BaCl₂ solution. After each sampling, incubation jars remained open for 1 h for gas exchange and to reach ambient O₂ level, and glass vials were refilled with new NaOH solutions. Incubation jars containing wetted soil samples were weighed before incubation, and distilled water was added when needed to maintain soil moisture. Finally, mineralization rates were calculated on the basis of daily CO₂ emissions. Cumulative C mineralization (Cum) was the sum of the mineralization in each stage of the incubation (mg CO₂-C per kg soil).

Soil DOC was extracted with Milli-Q water (1:10 w/v) by mixing on a reciprocating shaker (ZWYR-4912, Zhicheng, China) for 2 h at 180 rpm at 25°C (Li et al., 2020). Supernatants were collected as DOC solutions after centrifugation at 8,000 rpm for 10 min and filtration through a 0.45 μm (Fisher) microporous filter membrane. Dissolved organic C concentration was measured on a total organic carbon analyzer (multi N/C3100, Analytik Jena, Germany). Excitation-emission fluorescence spectra (EEMs) were determined using a photometer (F7000 FL, Hitachi, Japan). To obtain EEMs, excitation (Ex) wavelengths were set to increase gradually from 200 nm to 450 nm in increments of 5 nm, and the emission (Em) was measured at each excitation wavelength from 250 nm to 600 nm with 1 nm increments. Blank EEMs were obtained using Milli-Q water in the same way. To minimize instrument-specific effects on EEMs and eliminate scatters, first, the EEMs of samples and blank were normalized to the Raman peak area (Ex = 350 nm) to eliminate the fluorescence intensity differences caused by the fluorescence meter. Rahman normalized blank EEMs were then subtracted from sample EEMs to remove Rayleigh scatters and obtain corrected EEMs.

Parallel factor analysis (PARAFAC) was used to resolve the EEMs into different components using the package “StaRdom” in R (Pucher et al., 2019; Liu J. Z. et al., 2021). Non-negative constraints were applied to the parameters to allow only realistic results on the chemical composition. A three-component model was validated with the split-half analysis program of Murphy et al. (2013). The model explained 99.5% of the variance in the dataset. The Fmax values were the fluorescence of each component in each sample at their respective excitation and emission maximums and were used to quantify the component. For quantitative comparisons with published data, the components identified as constituents were searched using the OpenFluor online database of automated fluorescence spectra of organic compounds in the environment (http://www.OpenFluor.org). OpenFluor defines a similar spectrum to that having a Tucker congruence (TCC) greater than 0.95 on both excitation and emission spectra, and possible sources of the identified components were obtained based on the OpenFluor search results.

Biological index (BIX) is used to evaluate the proportion of microbially derived carbon in DOC (Wang et al., 2015), and the humification index (HIX) represents the degree of DOM humification (Wang et al., 2013). These indexes are calculated by the following equations:

where the F_em/ex represents the fluorescence intensity at a specific emission wavelength (em) or an emission wavelength (em) range which excited at a specific excitation wavelength (ex).

Soil DNA was extracted from 0.5 g samples using a FastDNA Spin Kit (MP Biomedicals, Santa Ana, CA, United States) following the manufacturer’s instructions. After PCR amplifications by the bacterial universal primers 341F (5′-CCTAYGGGRBGCASCAG-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′) (Gardes and Bruns, 1993) and the fungal universal primers ITS1F (5′-CTTGGTCATTT AGAGGAAGTAA-3′) and ITS2R (5′-GCTGCGTTCTTCA TCGATGC-3′) (Lee et al., 2012). Samples were sent to Shanghai Biozeron Co., Ltd. (Shanghai, China) for sequencing via MiSeq (Illumina). After sequencing, raw data were trimmed with the program Trimmomatic, including quality trimming and chimera removal, to obtain paired-end clean reads. Then, clean paired-end reads were aligned by FLASH (V1.2.11) and filtered to remove barcodes and primers with Mothur (V1.42.1). Finally, operational taxonomic units (OTUs) were clustered by the uparse method via Usearch at 97% identity. The SILVA database (v138.1) was used to align bacterial 16S data (Pruesse et al., 2007), and the UNITE database (v8.2) was used to align fungal ITS data (Koljalg et al., 2013). In all samples, 9,087 and 2,264 OTUs were identified from16S rRNA and ITS data, respectively. Sequencing data were deposited in the Sequence Read Archive (SRA) in NCBI under Bioproject PRJNA933112.

Analysis of variance (ANOVA) and Tukey tests were used to determine differences in soil properties and C mineralization among different treatments using R (v 4.1.3, R Development Core Team). All data are presented as the mean ± standard error. Principal Coordinates Analysis (PCoA) and Adonis test were used to explore the changes in bacterial and fungal community structures among different treatments via the cmdscale function in the vegan package.

Package “Hmisc” was used to construct co-occurrence networks based on a Spearman correlation matrix. The OTUs were filtered by their presence in more than half of the samples, and 2,895 and 333 OTUs were included in bacterial and fungal network analyses, respectively. Then all Spearman correlations between OTUs were calculated, and p-values were adjusted by the Benjamini and Hochberg false discovery rate (FDR) test, with adjusted p-values having a 0.01 cutoff. The focus was only on microbial taxa that were highly correlated with one another because that indicated that they had strong coexistence within a community (Herren and Mcmahon, 2018). Then, Gephi 0.9.3 was used to visualize the final networks, calculate network topological characteristics and construct modules. Mantel test in the package “vegan” in R (v4.1.3) was used to calculate the Pearson correlation coefficients between module eigengenes and soil properties and cumulative C mineralization (Cum).

Random forest modeling was used to assess the importance of predictors of C mineralization using the package randomForest in R (v4.1.3). The significance of the model and predictor importance were determined using the packages “A3” and “rfPermute,” respectively, modified according to Jiao et al. (2018). Potential impact factors were then classified into two categories: DOC characteristics and microbial community structures. Canonical correlation analysis (CCA)-based variance partitioning analysis (VPA) was used to determine the relative importance of the two categories on C mineralization using the “vegan” package in R (v4.1.3).

Organic amendments significantly affected the soil nutrient contents (Table 1). The highest SOC, total N (TN), and total P (TP) contents were in GMC. Whereas the contents of most soil nutrients were not significantly different between GM and NPK. In addition, organic amendments also significantly increased Cum, with GMC (3,434 mg kg⁻¹) > GM (2,934 mg kg⁻¹) > NPK (2,592 mg kg⁻¹) (p < 0.05; Figure 1A). Similarly, GMC had the highest DOC concentrations, followed by that in GM and NPK (p < 0.05; Figure 1B). Moreover, the aboveground biomass fractions of stem, leaf, and grain, were the highest in GMC and the lowest in NPK, whereas there were no significant differences in root biomass among treatments (Supplementary Table S3).

The fluorescence EEMs spectra of the solutions extracted from the three treatments were resolved into three components by the PARAFAC analysis (Figures 2A–F). According to fluorescence spectra and published data, the three major components were microbial humic acid-like substances (C1), and terrestrial humic acid-like substances (C2 and C3). Excitation (Ex) and emission (Em) peaks and characteristics of each component are summarized in Table 2. In addition, the relative abundance of C1 was significantly higher in GMC than in the other treatments, whereas C3 showed the opposite trend (Figure 2I). The biological index (BIX) was significantly higher in GMC than in the other treatments, whereas there were no significant differences in the humus index (HIX) among the treatments (Figures 2G,H).

Bacterial communities were dominated by Gammaproteobacteria (13.4% on average), Alphaproteobacteria (10.6% on average), and Subgroup 6 of Acidobacteria (9.3% on average) (Figure 3A). Principal co-ordinates analysis (PCoA) indicated the composition of bacterial communities was altered by co-incorporation of green manure and biochar (R² = 0.189; p < 0.01; Figure 3C). When considering the relative bacterial abundances, the relative abundances of Gammaproteobacteria, Alphaproteobacteria, Deltaproteobacteria, and Bacteroidia increased by co-incorporation of green manure and biochar, whereas those of Subgroup 6 of Acidobacteria, Gemmatimonadetes, and Anaerolineae decreased (p < 0.05; Figure 3A). In addition, the fungal communities were dominated by Botryotrichum (26.5% on average) and Cladorrhinum (24.8% on average) (Figure 3B). Similar to the bacterial community, the PCoA of fungal community compositions revealed significant (R² = 0.197, p < 0.01) separation among NPK, GM, and GMC treatments (Figure 3D). The relative abundance of Cladorrhinum increased with organic amendments (GM and GMC), whereas that of Botryotrichum decreased with organic amendments, with abundance in the order NPK > GM > GMC (p < 0.05; Figure 3B).

Co-occurrence networks were constructed to identify differences in co-occurrence patterns of soil bacterial (Figure 4A) and fungal (Figure 4B) communities in different treatments. For a meaningful comparison of network topological properties in individual treatments, subnetworks were generated from bacterial and fungal communities. Organic amendments significantly changed the topological properties of bacterial and fungal networks (Table 3). In bacterial networks, green manure application significantly increased the proportion of positive linear relationships and the network complexity, expressed as the average clustering coefficient (ACC) (Table 3). In fungal networks, positive interactions between species dominated the whole network. In addition, organic amendments significantly increased the network complexity (expressed as ACC), with complexity in the order GMC > GM > NPK. Bacterial and fungal networks were clustered into modules to identify significant module-environmental parameter relations. According to Mantel tests, most modules in bacterial networks were positively correlated with environmental parameters. Modules 1, 2, and 3 were positively correlated with Cum (cumulative C mineralization) and C1 of DOC (Figure 5). In fungal network modules, except for modules 1 and 3, there were few correlations between other modules and environmental parameters.

To determine the dominant factors affecting C mineralization, correlation, random forests, and CCA based-VPA were applied. Firstly, the correlation analysis showed that rice biomass was positively correlated with Cum, DOC concentrations, and relative abundance of C1 and that Cum was positively correlated with TN, DOC concentrations, the relative abundance of C1(Component 1), and ACC (average clustering coefficients) of fungal networks (Figure 6A). In addition, the ACC of a fungal network was positively correlated with the relative abundance of C1. Random forests were constructed to evaluate the importance of individual factors on Cum (Figure 6B). Among all the factors, ACC of fungal networks and relative abundance of C1 were the two most important predictors of C mineralization, followed by TN, DOC concentrations, and indexes of DOC fluorescence spectra (p < 0.01). Then, the important predictors were classified into two categories: DOC characteristics (represented by DOC C1 and DOC concentrations) and microbial networks (represented by ACC of bacterial and fungal networks). Further, VPA was used to evaluate the relative importance of DOC characteristics, microbial networks and their interaction (Figures 6C,D). The interaction effect of the two categories explained approximately over 60% of the variation in C mineralization. In addition to the interaction effect, DOC characteristics (15.9%) explained a larger proportion of the variation in C mineralization than that of the microbial network (5.3%) (Figures 6C,D). Overall, it was found that ACC of the fungal network alone explained 90.6% of the variation in C mineralization due to microbial community, and DOC concentrations and C1 of DOC accounted for 50.2% and 43.5%, respectively, of the variation in C mineralization due to DOC.

As hypothesized, compared to other treatments, the co-incorporation of green manure and biochar had greater effects on DOC concentrations and the proportion of microbially derived C, represented by microbial humic acid-like substances (C1) and BIX (Figures 1B, 2I). The increased DOC in GMC was likely not directly from the soluble C fraction in biochar because the biochar DOC concentrations was very low (Supplementary Table S2), and soil samples were collected 90 d after the biochar application. Instead, it was highly likely that biochar application increased microbial activity and biomass by providing habitats and increasing soil porosity (Chen et al., 2019). An increase in microbial biomass and activity can stimulate the degradation of large biopolymers into small soluble molecules and therefore increase the DOC concentrations. In addition, an increase in microbial activity can also increase available nutrient contents and promote rice growth (Supplementary Table S3), consequently leading to an increase in the abundance of rhizodepositions (Eisenhauer et al., 2017). Rhizodeposits can be further adsorbed by biochar and reduced losses from leaching (Joseph et al., 2021). The correlation analysis confirmed that DOC concentrations was positively correlated with rice biomass (Figure 6A).

Compared with the other treatments, the co-incorporation of green manure and biochar also increased the relative abundance of the microbial humic acid-like component (C1). In addition, the BIX was increased from 0.76 to 0.87 with the co-incorporation of green manure and biochar (Figure 2G). The BIX represents the autochthonous biological activity of DOC. A high BIX value (>0.8) indicates autochthonous origins, i.e., microbially derived materials, whereas a low BIX value (< 0.8) indicates allochthonous origins, i.e., exogenous organic materials (Catalan et al., 2014). The increase in microbial humic acid-like components and BIX suggested the proportion of microbial-derived DOC increased (Wang et al., 2015; Zhang et al., 2017). The observed increases may be attributed to the fact that biochar enhances microbial biomass and activities by promoting oxygen availability, particularly in nutrient-rich soils (Zhang et al., 2018 Pokharel et al., 2020). Therefore, in GMC, it was highly likely that increases in exocellular enzyme activity by biochar and green manure promoted the degradation of plant residues, such as green manure and root exudates, and decrease the abundance of exogenous organic materials. Meanwhile with sufficient labile C from green manure, increases in microbial biomass and assimilative activity by biochar could lead to the accumulation of microbial necromass, i.e., microbially derived carbon, and result in a shift in DOC composition toward autochthonous origins.

The co-incorporation of green manure and biochar significantly shifted the composition of microbial communities and changed network topological properties (Figures 3, 4). Firstly, co-incorporation of green manure and biochar primarily promoted the relative abundance of fast-growing copiotrophic gram-negative bacteria, such as Alphaproteobacteria and Deltaproteobacteria, because of the sufficient liable C sources from green manure application (Almagro et al., 2021). Meanwhile, the relative abundances of some slow-growing, oligotrophic gram-positive bacteria also increased in GMC, such as Bacteroidia and Actinobacteria, likely because of the prevalence of large organic compounds from biochar application (Cui et al., 2020). In addition, although both belonged to saprophytic fungi, Cladorrhinum increased, whereas that of Botryotrichum decreased under the co-incorporation of green manure and biochar treatments (Figure 3B). Botryotrichum can produce some secondary metabolites that are unfavorable for plant growth (Elkhateeb and Daba, 2022). Therefore, the decrease in abundance Botryotrichum suggested that the co-incorporation of green manure and biochar could potentially decrease crop morbidity and increase soil health.

Whereas the increase in fungal network complexity (expressed by the ACC) with co-incorporation of green manure and biochar provided partial support for the second hypothesis of the study, the lowest bacterial network complexity with co-incorporation of green manure and biochar did not support the hypothesis. The decrease in bacterial network complexity could be related to the co-incorporation of green manure and biochar increasing the abundance of both copiotrophs and oligotrophs, as discussed above. The corticotrophs likely used the labile C from green manure, and the oligotrophs likely mineralized soil recalcitrant C from biochar (Ling et al., 2022). Therefore, the cooperation between species decreased (Table 3). To explain the increase in fungal network complexity, one possible reason is that biochar can increase the fungi-to-bacteria ratio, as well as fungal activity (Wang X. C. et al., 2022). In addition, fungal hyphae can more easily inhabit the relatively large nutrient-rich pores of biochar than bacteria (Joseph et al., 2021). With increases in the density of fungi colonizing large pores of biochar, interactions between fungi likely increased, resulting in an increase in network complexity.

Comprehending the results of random forests and VPA, this study suggested that the interaction between DOC characteristics and microbial network, particularly fungal network complexity, was the dominant factors regulating C mineralization. The conclusion is consistent with the third hypothesis of the study. Dissolved organic carbon (DOC) is the most active component in SOC and provides a soluble organic substrate for heterotrophic microorganisms, and thus concentrations and components of DOC are essential factors affecting microbial community composition and function and vice versa (Zheng et al., 2021; Dou et al., 2022; Hu et al., 2022). The interaction between DOC components and soil microbial structure is the core of the C cycle. Microbial community composition (Schimel and Schaeffer, 2012) and specifically microbial network characteristics (Banerjee et al., 2016; Morrien et al., 2017) are increasingly considered to regulate soil C mineralization. In this study, increases in DOC concentrations and microbially derived compounds accounted for 18.3% of the variance in C mineralization. Meanwhile, the interaction between DOC characteristics and increased fungal network complexity accounted for over 60% of the variance in C mineralization (Figures 6C,D). Compare with less connected networks, highly connected networks may contribute to efficient C utilization (Morrien et al., 2017). The results of this study indicated that fungi were more active in C cycling with a potentially more connected network, especially when biochar-induced saprophytic fungi, such as Cladorrhinum, mineralize recalcitrant C (Ling et al., 2022). The results highlighted the importance of the interaction between DOC characteristics and fungal network complexity on C mineralization.