Introduction

Soil contains approximately 1500 Pg of organic carbon (C) within a depth of 1 m [1]. Soil is thus an important sink for C and could play a vital role in our quest towards achieving a C neutrality [2, 3]. According to the “4 per 1000” initiative, annual increase in soil CO2 content can be neutralized by increasing the C content in the first 30–40 cm of soil by 0.4% per year, that is, at C sequestration rates of 400–600 kg C ha-2 yr-1 [4]. Therefore, enhancing soil organic carbon (SOC) sequestration has considerable potential to contribute to C neutrality. Above-ground plant C inputs have been suggested to be the dominant factor controlling SOC sequestration, particularly in agroecosystems with appropriate management [5, 6]. Therefore, estimating the capacity and stability of C sequestration induced by above-ground plant C input, under a carbon-neutral scenario, would be important.

Soil organic carbon sequestration pathways and soil matrix protection processes are related to the molecular complexity of SOC [7,8,9,10]. Thus, understanding the chemical complexity of above-ground plant C input, following decomposition, is crucial for estimating soil carbon sequestration potential [7, 11, 12]. There are conflicting views on the chemical fate of residual carbon [13, 14]. The traditional view is that plant residues with different chemical compositions eventually converge to an array of common compounds that are more resistant to decay, despite different environmental conditions [12, 15,16,17,18]. However, various studies have demonstrated that the chemistry of plant residues diverges under the combined effect of initial litter quality and decomposer community composition [14, 19, 20]. A recent study challenged the views above through hypothesizing that microorganisms control the dynamics of residue chemistry via dual microbial metabolism, that is, a great relative contribution of microbial anabolism supports the convergence of residue chemistry, whereas residue chemistry diverges when microbial catabolism is dominant, leading to increasing molecular modifications through extracellular enzymatic degradation [13]. However, this theory lacks empirical evidence from large-scale long-term field observations and from laboratory studies.

Plant residue decomposition is a microbially controlled “broad process” that involves a wide range of microorganisms [21, 22]. According to the dual microbial metabolism theory, the fate of residue chemistry largely relies on the interactions within microbial communities [23,24,25,26,27]. Because of functional redundancies in microbial communities, the relative importance of individual species differs. Microorganisms that are important for regulating soil functions are recognized as keystone taxa [28, 29]. Elucidating how keystone taxa drive plant residue decomposition over space, time, and environmental gradients remains a core focus in ecological research. However, only a limited number of studies have investigated keystone taxa during residue decomposition process [30, 31]. Moreover, to the best of our knowledge, no study has yet investigated the role of keystone taxa in regulating residue chemical complexity via their impact on extracellular enzyme production or microbial assimilation.

Here, we investigated the succession of chemical complexity of plant residues and the roles of keystone taxa in regulating this process by influencing extracellular enzyme production and microbial assimilation based on a large-scale long-term field experiment. We revealed the divergence-convergence trajectory of residue chemical composition using nuclear magnetic resonance (NMR) and gas chromatography-mass spectrometry. We defined keystone taxa by constructing co-occurrence networks based on amplicon sequencing. Structural equation modeling (SEM) was employed to determine the role of keystone taxa on controlling extracellular enzyme production, microbial assimilation, and therefore the succession of residue chemical composition. Subsequently, metagenomic sequencing was performed to verify the function of keystone taxa on extracellular enzyme production. The results demonstrated that keystone taxa regulate the divergent–convergent trajectory in residue chemistry.

Materials and Methods

Experimental design

Initiated in May 2008, the litter decomposition experiment was based on a large-scale soil transplantation experiment set up in 2005, involving three agroecological experimental stations along a temperature and precipitation gradient from a cold temperate region (Hailun, 47°26’N, 126°38’E) to warm temperate (Fengqiu, 35°00’N, 114°24’E) and mid-subtropical regions (Yingtan, 28°15’N, 116°55’E) (Fig. S1). This design allowed us to evaluate our results at a large scale by including environmental gradients that result in differing microbial community compositions and functions. Soil transplantations were performed with three typical soils from the local area surrounding each experimental station (Table S1), namely neutral black soil (Phaeozem) in Hailun, alkaline Chao soil (Cambisol) in Fengqiu, and acidic red soil (Acrisol) in Yingtan. A detailed description of soil transplantation experiments has been included in the Supplementary materials.

In the soil transplant experimental plots, litter bags containing maize straw and wheat straw were buried in the soils at an average depth of 12 cm. The litter bags were made of double-layered nylon mesh and were 15 × 10 cm in dimension with a 0.074 mm mesh. Each bag was filled with 100 g of maize straw or wheat straw. All the straw material was cut into 5 cm length and oven-dried at 100 °C. In total, 378 bags (189 of maize and 189 of wheat) were included in this study. Three nylon mesh bags that served as replicates were removed at 1 month, 3 months, 0.5 year, 1 year, 2 years, 3 years, and 9 years after burial. The samples were stored frozen (at -80 °C) until analysis. The residue samples were weighed, and their dry weights were calculated according to the water contents. Soil samples from the experimental plots were collected after 9 years of decomposition and stored frozen (at -80 °C) until further analysis.

NMR analysis

After drying, straw samples and soil samples were subjected to 13C NMR analysis [32]. The straw samples were grounded to powder, and the soil samples were pretreated with 2% hydrofluoric acid to remove paramagnetic material [33]. The Solid-state Cross Polarization Magic Angle Spinning (CP-MAS) NMR spectra were acquired using the parameters as previously described [12]. The overall signal to noise ratio of the spectra ranged from approximately 20 to approximately 100, suggesting a good quality of identification [34]. The spectra were analyzed within six chemical shift regions [12]: alkyl-C (0–45 ppm) is from plant polymers, fatty acid, and lipids; O/N-alkyl-C (45–94 ppm) and di-O-alkyl-C (94–112 ppm) are from proteins or peptides, methoxyl-C, cellulose, hemicellulose, and carbohydrates; Aromatic C-C + /H (112–143 ppm) and Aromatic C-O (143–162 ppm) are from aromatic carbons in lignin and/or olefinic carbons in lipids; carbonyl/amide-C (162–220 ppm) is from esters, carboxyl groups, and/or amide carbonyls. The decomposition degree was represented by O-alkyl/alkyl ratio, which is an important decomposition index. Alkyl refers to the spectral region at 0–45 ppm, and O-alkyl refers to the region at 45–112 ppm [12].

Pyrolysis-gas chromatography and mass spectrometry (py-GCMS)

The independent method, py-GCMS, with high resolution was used to determine the molecular chemistry of straw residues and for cross validation [14]. Samples were analyzed by Analytical & Testing Center of Huazhong University of Science and Technology. Samples were pyrolyzed at 600 °C for 20 s using CDS Pyropobe 6150 pyrolyzer (CDS Analytical, Inc., Oxford, PA, USA). Subsequently, compounds were transferred onto a gas chromatograph - mass spectrometer (GC-MS, Agilent 7890 A/5975 C, Agilent Technologies, Inc., Santa Clara, CA, USA) following the protocol described previously [20]. The abundance of each compound was determined according to the total ion signal from all detected peaks and was reported as percentage. Approximately 100 compounds were detected from each sample, and in total 3228 compounds were detected. Among these, 425 compounds either with relative abundance > 0.5% and/or observed in more than 10 samples were identified [35], and classified into ten groups (Fig. S4). The relative abundance of these identified compounds accounted for more than 80% of all detected peaks. Among the 425 compounds, 90 of the most abundant compounds belonged to eight out of the ten groups (Table S5).

Extracellular enzyme activity (EEA) assays

Four hydrolytic enzymes (β-1,4-glucosidase (βG), Cellobiohydrolase (CBH), β-1,4-xylosidase (βX), and β-N-acetylglucosaminidase (NAG)) and two oxidative enzymes (phenol oxidases (PhOX) and peroxidases (Perox)) were analyzed after decomposed for 0.5, 2, and 9 years. The analysis of the four hydrolytic enzymes was performed using 96-well microplates according the method previously described [36]. The activity of phenol oxidases and peroxidases was measured following the method adapted from a previous publication [37]. A detailed description of EEA assays has been included in the Supplementary materials. The EEA data were processed by Z-score transformation.

Amino sugar

Amino sugars were extracted from straw samples (decomposed for 0.5, 2, and 9 years) and soil samples (collected in year 9) according to the method previously described [38]. A detailed description of amino sugar assays has been included in the Supplementary materials. Total soil amino sugars were the sum of the individual ones (glucosamine (GluN), galactosamine (GalN), and muramic acid (MurA)). The ratio of amino sugar-C to TOC was to estimate the relative contribution of microbial residues in decomposed straw residues.

DNA extraction and MiSeq sequencing

Genomic DNA was extracted from straw residue samples (0.05–0.25 g, after decomposed for 0.5, 2, and 9 years) and soil samples (0.5 g, collected in 0.5 year and 9 years), using the MO-BIO Power-Soil DNA kit following the manufacturer’s standard operation protocol. Samples were sent to Shanghai Majorbio Bio-Pharm Biotechnology Co., Ltd. (Shanghai, China) for sequencing by MiSeq (Illumina), after amplified by bacterial primers 338 F/806 R [39] and fungal primers ITS1-1737F/ITS2-2043R [40]. After sequencing, FLASH and Trimmomatic program was used to screen and trim the raw sequences, including quality trimming, chimera detection, and removing. The SILVA database (Silva.seed_v138) was used to align the sequences for 16S rRNA gene data [41], and the UNITE database (version 6.0) was used to align ITS gene data [42]. Then, taxonomy was assigned by the Ribosomal Database Project, and operational taxonomic units (OTUs) were clustered at 3% divergence (97% similarity) and classified. The sequencing results were deposited in Sequence Read Archive (SRA) in NCBI under SRA accession: SRP312985.

Co-occurrence network analysis

Co-occurrence networks were used to visualize the associations among microbial communities. The analysis was performed by Co-occurrence Network inference (CoNet) in Cytoscape 3.6.1 (Cytoscape core develop team, 2017) (see the details in Supplementary materials) [43]. To identify the keystone taxa of the co-occurrence network, value importance in projection (VIP) by partial least squares (PLS) regression was used as a predictor to estimate the importance of OTUs occurred in the network to EEA and amino sugar content. The R package mixOmics was used to calculate the PLS of high throughput omics data. The predictors were ranked according to their VIP. Higher VIP values indicate higher contribution in PLS regression. The ones with VIP > 1 were recognized as relevant predictors [44]. Then the OTUs with a high degree in the network (the top 5 within each dominant module) and high VIP values (VIP > 1) were recognized as keystone species.

Metagenomic sequencing

To address the role of keystone taxa in the EEA responsible for straw component decomposition, metagenomic sequencing and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway annotation were performed on straw samples buried in red soil (Acrisol) under subtropical climate conditions after decomposition for 0.5 year, 2 years, and 9 years. As the subtropical region has the highest annual mean temperature among the three sites, it had the fastest decomposition rate and was more likely to reach at stable stage of decomposition. A detailed description of metagenomic sequencing has been included in the Supplementary materials. Sequencing results were deposited in the Sequence Read Archive of NCBI under SRA accession: SRP417783.

Data analysis

The decomposition pattern was described by the double exponential equation [45, 46] as y = CR exp(-KRt)+CS exp(-KSt), where CR is the fraction decomposed during the rapid decomposition stage (0 ≤ CR ≤ 100%), with a rapid decay rate KR (year-1), and CS is the fraction decomposed during the slow decomposition stage (0 ≤ CS ≤ 100% and CR + CS = 100%), with a slow decay rate KS (year-1). The distribution patterns of residue chemical composition in wheat and maize straw samples under different climatic conditions across decomposition times were determined by principal coordinate analysis (PCoA) using the package vegan in R [47]. The β-diversity of residue chemical composition between any two samples was determined by the Bray-Curtis distance, which is a taxon-based metric of differences in community composition. A detailed description of the calculation of divergence index has been included in the Supplementary materials.

One-way analysis of variance was performed to assess the effect of decomposition time on EEA, amino sugar content, and PLFA content using Tukey’s test in R (version 4.1.0, R Development Core Team). A pearson’s correlation test was used to analyze the correlation between O-alkyl/alkyl, EEA, and amino sugar content using R (4.1.0, R Development Core Team). Structural equation modeling was used to assess how keystone taxa alter the chemical composition of residues via EEA and amino sugar contents (see details in Supplementary materials) [48]. Random forests were used to quantitatively assess the importance of keystone taxa in individual extracellular enzyme activities (see details in Supplementary materials).

Results

Divergence-to-convergence shift in residue chemistry during straw decomposition

The decomposition of straw followed comparable trajectories in all the three climate zones and soils (Fig. S2). The decomposition trajectory could be effectively described with a double exponential model [45] (Fig. 1A) and was categorized into a rapid decomposition stage (0–1 year) and a slow decomposition stage (1–9 years) (Table S2). Generally, during 0–1 year, wheat straw decomposed faster than maize straw. However, more wheat mass (approximately 16.0%) remained than maize mass (approximately 10.7%) after 9 years.

Fig. 1: Decomposition patterns of wheat and maize straw residues represented by mass remaining and O-alkyl/alkyl ratio.
figure 1

A Changes in the mass remaining with decomposition time. The y axis of mass remaining is presented in a logarithmic scale. The yellow symbols represent maize straw, and the green symbols represent wheat straw. Double exponential models were fitted on the mass remaining of wheat straw and maize straw: \(y_{maize} = 52.5e^{ - 5.85t} + 47.5e^{ - 0.23t}({{{{{{{\mathrm{R}}}}}}}}^2 = 0.89)\) \(y_{wheat} = 70.9e^{ - 4.68t} + 30.91e^{ - 0.075t}({{{{{{{\mathrm{R}}}}}}}}^2 = 0.95)\). B Changes in the O-alkyl/alkyl ratio with decomposition time. The gray dashed line indicates the average O-alkyl/alkyl ratios of soils from the field plots.

Full size image

The changes in C functional groups in the straw residues undergoing decomposition were distinguished using NMR spectroscopy. The changes in O/N-alkyl-C and alkyl-C contributed to approximately 60% of the differences across treatments, according to SIMPER analysis. The O-alkyl/alkyl ratio decreased with decomposition and became similar in the soil with both maize and wheat straw residues after 9 years (Fig. 1B and Fig. S3), suggesting similar residue chemistry and SOC compositions. Moreover, according to the analysis of molecular compositions using py-GCMS, during decomposition, the relative abundances of polysaccharide-derived compounds and short-chain fatty acids (C8–C19) declined in both straw residues during the first 3–6 months, and then increased thereafter (Fig. S4). The relative abundances of aromatics, alkyl compounds, long-chain fatty acids (C20–C30), lignin-derived compounds, and phenols showed the opposite trend, as they increased during the first 1–2 years of the study and decreased thereafter.

The changes in residue chemistry complexity across the decomposition period were visualized using PCoA (Fig. 2A). Along the decomposition continuum, for both maize and wheat straw residues, the C functional groups and molecular composition of the residue showed a divergent pattern before 3 years and then a convergent pattern after 9 years. This divergence-convergence trajectory was further supported by the divergence index measured by the Bray-Curtis distance (Fig. 2B). Specifically, the divergence caused by the initial litter quality drastically increased from 0 to 1 year according to NMR, or from 0 to 0.5 year according to py-GCMS, and then decreased thereafter; meanwhile, the greatest dissimilarities caused by climatic conditions appeared at approximately 2–3 years, according to NMR, and at the beginning of the experiment (1–3 months), according to py-GCMS. Moreover, in our study, the divergence-convergence shift was commonly observed under all three climatic conditions and in all three soil types (Fig. S5). As expected, convergence of residue chemical composition occurred more rapidly in warm temperate and subtropical zones (approximately 3 years) than in the cold temperate zone (approximately 9 years).

Fig. 2: Principal coordinate analyses and divergence index based on Bray-Curtis distances.
figure 2

A Principal coordinate analyses show that the straw residue chemical composition diverged during the 0.5–3 year period and then converged during the 3–9 year period for both maize straw and wheat straws as analyzed by NMR and py-GCMS. The squares represent samples from the cold temperate (cold temp.) region, the circles represent samples from the warm temperate (warm temp.) region, the upward-pointing triangles represent samples from subtropical regions, and the downward-pointing triangles represent the samples at time 0, i.e., the original straw samples. The color gradient represents the chemical composition succession of the straw residue from 0 year (light green) to 9 years (dark green). B The divergence index shows that the differences among samples first increased and then decreased with the decomposition time. This pattern was similar for the divergence index in general and was caused by the straw types and climatic conditions. The black symbols show the divergence of the samples based on NMR analysis; the grey symbols show the divergence of the samples based on py-GCMS.

Full size image

Changes in EEA and microbial necromass abundance with decomposition

With decomposition, for both straw residues, EEA (expressed by the Z-score) markedly decreased from 0.5 to 2 years but did not change significantly thereafter (Fig. 3A). In contrast, the microbial necromass abundance, that is, the amino sugar-C/TOC ratio (ASC/TOC), increased with decomposition time (Fig. 3B). More than 80% of the microbial necromass was generated via fungal deposition. In addition, a decrease in EEA (Fig. S6), and an increase in ASC/TOC (Fig. S7) were observed under different climatic conditions and in different soils. Climatic conditions had a significant impact on the changes in EEA and ASC/TOC; a warm climate accelerated the decrease in EEA and increase in ASC/TOC.

Fig. 3: Extracellular enzyme activity and microbial necromass abundance in maize straw and wheat straw residues after decomposition for 0.5 year, 2 years, and 9 years.
figure 3

A The extracellular enzyme activities (represented by the Z-score transformation) decrease with decomposition. B The microbial necromass abundance (represented by the total amino sugar-C/total organic C (ASC/TOC)) increases with decomposition. Tukey’s test was conducted on extracellular enzyme activity and ASC/TOC for maize and wheat straw separately, with p < 0.05.

Full size image

The EEA and ASC/TOC were negatively correlated (Fig. S8). In addition, there was a positive power relationship between hydrolytic enzyme activity and O-alkyl/alkyl, and a negative power relationship between ASC/TOC and O-alkyl/alkyl. A negative correlation between ASC/TOC and O-alkyl/alkyl was observed in most climate zones and soils (Fig. S9). When the O-alkyl/alkyl ratio stabilized at a mean value below 4 during 3–9 years, the ASC/TOC ratio was found to be greater than 30 mg g−1.

Changes in microbial community composition and keystone species with decomposition

Bacterial and fungal communities in straw residue samples were analyzed at 0.5, 2, and 9 years of decomposition by pyrosequencing. Across the three decomposition periods, the fungal community was dominated by Sordarimycetes (58.9%) and Eurotiomycetes (18.9%), whereas the bacterial community was dominated by Alphaproteobacteria (27.9%) and Actinobacteria (13.2%) (Fig. S11 and Fig. S12). Unlike residue chemistry, the divergence of both bacterial and fungal communities caused by climate and soil type significantly increased with time (Table S6, Fig. S13). In particular, the bacterial community in the residues showed a clear tendency to shift toward the community composition of bulk soil.

To visualize the changes in the co-occurrence patterns of microbial communities with decomposition, bacterial and fungal networks were constructed based on the community successions during the 0.5–2 year and 2–9 year periods, and clustered into modules thereafter (Fig. 4). The dominant modules of the bacterial network primarily belonged to Alphaproteobacteria (0.5–2 years and 2–9 years) and Actinobacteria (2–9 year) (Fig. S14). In the fungal network, all nodes within the dominant modules belonged to the Chaetomium genus, which was also the most abundant fungus across the 9 years of decomposition. According to the Mantel test, all dominant modules in the bacterial network were positively correlated with EEA and amino sugars during the 0.5–2 year and 2–9 year periods (Fig. S15). However, in the fungal network, only during the 2–9 year period, the dominant modules were significantly correlated with amino sugars (Fig. S15). Furthermore, the nodes with high degrees in the network and high contribution to the EEA or amino sugar content (VIP > 1) were identified as keystone taxa (Table S8). Specifically, for the bacterial networks, the keystone taxa belonged to Rhizobiales and Caulobacterales orders of Alphaproteobacteria during 0.5–2 years, and belonged to Rhizobiales order of Alphaproteobacteria and Micrococcales order of Actinobacteria during 2–9 years. For fungal networks, all keystone taxa belonged to Chaetomium genus. In addition, most of the keystone taxa were also found in the sub-networks constructed under individual climate zones (Fig. S16, Fig. S17).

Fig. 4: The co-occurrence networks of bacterial and fungal communities in straw residues during different periods of decomposition.
figure 4

A Bacterial network during 0.5–2 years. B Bacterial network during 2–9 years. C Fungal network during 0.5–2 years. D Fungal network during 2–9 years. A connection indicates a significant (p < 0.005) correlation between two OTUs. The nodes in the co-occurrence networks are colored by modules. The modules are named from 1 to 5 based on their relative dominance, and modules 1 to 5 are sequentially colored by blue, light orange, red, purple and dark yellow. The size of each node is proportional to its number of connections (i.e., degree), and the thickness of each connection between two nodes (i.e., edge) is proportional to the value of the Spearman’s correlation coefficient. A red edge indicates a positive interaction, while a blue edge indicates a negative interaction. The nodes are labeled with the keystone species associated with EEA or ASC/TOC; detailed information is provided in Table S8.

Full size image

Factors controlling the divergence-convergence trajectory of residue chemistry during decomposition

Structural equation models were fitted to infer the relative contribution of EEA and microbial necromass abundance to the divergence of residue chemical complexity and to evaluate the direct and indirect effects of the keystone taxa-guided network modules and environmental factors on EEA and microbial necromass abundance during 0.5–2 years and 2–9 years. Based on SEM, during the 0.5–2 year period, EEA was significantly correlated with the divergence of residue chemical complexity and had a direct and dominant effect, explaining over 40% of the variation (Fig. 5A, B, Table S9). During the 2–9 year period, the increased amino sugar content was the direct and dominant factor causing the decrease in divergence of residue chemical complexity, explaining approximately 40% of the variation (Fig. 5C, D, Table S9). In both the 0.5–2 year and 2–9 year periods, the keystone taxa guided bacterial network modules were the dominant controlling factor for changes in EEA (explaining 44.2% and 38.6% of the variation, respectively). The amino sugar content was correlated with the keystone taxa-guided bacterial (0.5–2 years and 2–9 years) and fungal (2–9 years) network modules. In addition, the correlation between fungal modules and amino sugar content was coupled with an increase in the relative abundance of Chaetomium (from approximately 32% to approximately 44%), thereby suggesting the importance of Chaetomium in amino sugar accumulation.

Fig. 5: The effects of abiotic and biotic factors on the divergence of residue chemical composition as estimated using the structural equation model.
figure 5

A Factors impact divergence of residue chemical composition based on NMR during 0.5–2 years. B Factors impact divergence of residue chemical composition based on py-GCMS during 0.5–2 years. C Factors impact divergence of residue chemical composition based on NMR during 2–9 years. D Factors impact divergence of residue chemical composition based on py-GCMS during 2–9 years. The arrow width indicates the strength of the significant standardized path coefficients (p < 0.05). The red lines indicate positive relationships, while the blue lines indicate negative relationships. The bacterial and fungal networks are represented by the keystone taxa-guided module eigenvectors. Temp/Prec. indicates the ratio of the average annual temperature to the average annual precipitation and is a proxy for the impact of the climatic conditions; the soil pH is a proxy for the impact of the soil properties.

Full size image

Addressing the role of bacterial keystone taxa in regulating EEA

To address the role of keystone taxa in affecting EEA, relative abundance of the functional gene groups responsible for six extracellular enzymes (Table S10) in the dominant microbial classes at 0.5, 2, and 9 years in the subtropical region was evaluated via KEGG pathway annotation based on metagenome sequencing. The two dominant classes responsible for these functional genes were Alphaproteobacteria and Actinobacteria (Fig. 6A–F). The orders with keystone taxa were also dominant in Alphaproteobacteria and Actinobacteria. The keystone order Rhizobiales accounted for approximately 20%–75% of the relative abundance of the functional genes in Alphaproteobacteria. Similarly, the keystone order Micrococcales accounted for approximately 10%–80% of the relative abundance of the functional genes in Actinobacteria.

Fig. 6: The importance of keystone taxa in extracellular enzyme activity.
figure 6

A-F The left column showed the relative abundance of the functional gene groups responsible for extracellular enzyme activities of PhOX (A), Perox (B), βG (C), βX (D), NAG (E), and CHB (F) in the dominant microbial classes at 0.5, 2, and 9 years in the subtropical region evaluated via KEGG pathway annotation based on metagenome sequencing. G-L The right column showed the relative importance of the keystone orders in the changes of each extracellular enzyme activity evaluated via Random forests, PhOX (G), Perox (H), βG (I), βX (J), NAG (K), and CHB (L).

Full size image

Random forests were used to evaluate the relative importance of the keystone orders in each EEA (Fig. 6G–L). The relative abundance of functional genes distributed in the keystone order Rhizobiales served as the main universal determinants of the activities of βG, βX, NAG, and PhOx. Meanwhile, the relative abundance of functional genes distributed in the keystone orders Micrococcales and Caulobacterales served as one of the main determinants of CBH and Perox activities, respectively. In addition, the relative abundance of functional genes distributed in the three keystone orders was positively correlated with enzyme activity, particularly in Rhizobiales (Table S11). Although the relative abundance of functional genes distributed in the non-keystone orders, that is, unclassified Actinobacteria, also showed a significant impact on the activities of PhOx, CBH, NAG and βG, functional gene abundance in the unclassified Actinobacteria was negatively correlated with the enzyme activities.

Discussion

The residue chemical composition showed a divergent (0.5–3 year) and then a convergent (3–9 year) trajectory along the decomposition continuum across different climate zones and soils. This result was confirmed by summarizing the findings of 13 previous studies (Table S12) on residue decomposition in different ecosystems, which reported that during short-term decomposition (<3 years), residue chemical complexity showed a divergent pattern, whereas during long-term decomposition, a convergent pattern was common. However, very few studies have focused on the microbial mechanisms underlying the divergence-convergence trajectory. Here, we aimed to explore this through incorporating the concept of keystone taxa in extracellular enzyme degradation and microbial anabolism.

Bacterial keystone taxa-mediated EEA is responsible for divergent residue chemistry

During the early stage of decomposition (0–2 years in this study), we observed a divergent pattern of residue chemistry with a high EEA (Fig. 3A). Moreover, SEM analysis confirmed that, from 0.5 to 2 years, EEA was the dominant factor regulating the residue chemical divergence (Fig. 5A, B). During the early stage of decomposition, EEA is high because of the large amount of labile carbon sources [49], and a large number of straw biopolymers are degraded and modified by extracellular enzymes into plant-derived molecules that are not readily assimilated by microorganisms [13]; the increased relative abundance of polymers was associated with degradative lignin products (Fig. S4) and resulted in a divergence in residue chemical composition [15, 50]. Moreover, a low microbial assimilation level, that is, low ASC/TOC (Fig. 3B), indicated a less contribution of microbial anabolism. This corresponded with the microbial dual control theory, according to which, a high contribution from extracellular enzyme degradation and low contribution from microbial anabolism resulted in the observed divergent residual chemical complexity pattern [13].

The EEA was primarily controlled by keystone taxa-guided modules in the bacterial network (Figs. 4 and 5). Rhizobiales served as keystone taxa in all three climate zones (Fig. S16). The role of keystone taxa in the regulation of EEA was confirmed using the KEGG pathway annotation. Specifically, the keystone order Rhizobiales dominated Alphaproteobacteria which was the most dominant class where the functional genes responsible for extracellular enzymes were distributed (Fig. 6). More importantly, Rhizobiales was the most important predictor of βG, βX, NAG, and PhOx activities. Consistent with previous studies, we found that Alphaproteobacteria, particularly Rhizobiales, was the keystone taxon responsible for straw decomposition [51]. The keystone taxa, such as Brevundimonas and Rhizobium cellulosilyticum of the Rhizobiales order (Table S8), have been proven to be producers of xylanase and other hydrolytic enzymes [52]. In addition, during straw decomposition, with high C/N, Rhizobiales could facilitate decomposition by nitrogen fixation [53]. As discussed earlier, decomposition is a microorganism-mediated “broad process” that might be affected by a keystone guild consisting of multiple keystone taxa instead of individual species [21, 29]. In our study, the bacteria detected in keystone taxa-guide modules, which correlated with EEA, primarily belonged to Alphaproteobacteria (Fig. S15). In addition, over 85% of the correlations in these modules was positive, and the correlations associated with keystone taxa were also mostly positive. This suggested that keystone taxa are likely to cooperate with other microorganisms in their modules to produce lignocellulose-degrading enzymes, rather than competing for resources and therefore facilitate C mineralization [54]. In summary, the keystone taxa and their associated microbial communities in the bacterial network are the dominant factors controlling EEA, thereby, controlling the divergence of residue chemistry.

Chaetomium is the main contributor in the convergent residue chemistry

After 9 years of decomposition, convergence of residue chemical complexity was observed when the contribution of microbial assimilation became dominant (Fig. 5C, D). During this period, with limited labile carbon sources, the EEA decreased, whereas microbial necromass accumulated over time (Fig. 3). Moreover, when the amino sugar-C/TOC ratio reached 30 mg g-1, the O-alkyl/alkyl ratio became stable (Fig. S8A). The results revealed that when amino sugar-C/TOC was higher than 30 mg g-1 (>9 years), microbial anabolism became the dominant decomposition process, and the residue chemical complexity converged. Amino sugar-C/TOC is a representative indicator of the contribution of microbial necromass to SOM; however, the actual proportion of microbial necromass deposition in residues still remains unknown because of technical limitations. According to a previous study [55], the microbial-derived carbon content can be estimated by multiplying fungal GlcN by 9 and then adding the product to bacterial MurN multiplied by 45. This calculation resulted in the proportions of microbial-derived carbon to SOC being approximately 10%, 19%, and 32% at 0.5, 2, and 9 years, respectively. Therefore, when the contribution of microbial necromass was over 30%, the convergence of residue chemistry was observed. This corresponds with the microbial dual control theory that when the contribution of microbial assimilation increased, chemical complexity of residue showed a convergent pattern.

Microbial anabolism is primarily reflected by the accumulation of microbial necromass. In this study, fungi were responsible for over 80% of the microbial necromass deposition. Similarly, compared to the densities of bacteria, higher densities of fungal mycelia and spores were found in the residues under the scanning electron microscope (Fig. S15). This suggests that microbial assimilation may be driven by fungi. Moreover, as the relative abundance of Chaetomium among the fungi was approximately 35.3%, Chaetomium could be the major contributor to microbial anabolism. In addition, the keystone taxa and most nodes in the network belonged to the Chaetomium genus, which is an important decomposer of straw [56]. Therefore, Chaetomium was the main contributor of microbial-derived C during litter decomposition. We also found that keystone taxa-guided modules in the bacterial network had a significant impact on both EEA and microbial necromass over 2–9 years. KEGG pathway annotation suggested that the keystone taxa in the bacterial network, Rhizobiales and Micrococcales, are responsible for most exocellular enzyme production, which aligns with previous findings [52, 57]. It is thus highly likely that during the slow decomposition stage, the bacterial network facilitated the production of extracellular enzymes that decomposed large polymers into molecules that were readily taken up by microbes. Both the bacteria and fungi were subjected to microbial assimilation. As bacteria contributed to less than 20% of the microbial necromass deposition, fungi, particularly Chaetomium, were in dominance. Taken together, the results suggest that over 2–9 years, microbial anabolism was primarily performed by Chaetomium, which then caused the convergence of residue chemistry.

Implications for carbon sequestration and stabilization

To achieve C neutrality and mitigate climate change, there should be a more effective C sequestration across agricultural systems [5], particularly for the stabilization of above-ground plant C inputs in soils. Residue chemistry is a key factor in determining the capacity, quality, and pathway of C sequestration [10]. Our study suggested that although residue chemistry diverged after the rapid decomposition stage, it converged after the straw decomposed over nearly a decade in all three climate zones. Notably, the convergence of residue chemistry is more commonly observed in aquatic ecosystems [58, 59] than in terrestrial ecosystems [60, 61]. Physicochemical protection from minerals and aggregation can extend the residence time to decades. The litter bag method minimized protection, resulting in a convergence pattern developed within a decade because of microbial anabolism. Convergent residue chemistry indicated that over a long time-scale, regardless of different straw types and climates, plant residues share a universal impact on C sequestration. This deduction is consistent with a previous study, which showed that the initial quality does not control the long-term decomposition of SOM [8].

We found that bacterial keystone taxa mediated extracellular enzyme degradation and were responsible for divergent residue chemistry during 0.5–3 years of decomposition. Chaetomium was the main contributor in microbial assimilation and was the dominant control of residue chemistry convergence after 9 years of decomposition. Overall, we proposed a mechanism for microbial control of the long-term succession of residual carbon chemistry across different climates. Further research should explore the role of keystone taxa in the formation of stable C pools under the interaction between microbial residues and the mineral matrix. In addition, future research should enable the assembly of microbial networks to enhance the function of keystone taxa in regulating the changes in residue chemistry and thus SOC sequestration. In summary, our study provided a fundamental understanding of residue decomposition, the origin and consequences of residue chemistry, and SOC sequestration.