Ecological networks of dissolved organic matter and microorganisms under global change

Dissolved organic matter (DOM) constitutes a major carbon pool in aquatic ecosystems and is tightly coupled to microbial metabolism. Microbial communities both degrade complex DOM into smaller molecules and synthesize more refractory compounds, giving rise to molecular traits (e.g., molecular weight, stoichiometry, oxidation state, bioavailability) that influence DOM persistence. DOM simultaneously shapes microbial diversity and function. Despite advances in ultrahigh-resolution mass spectrometry and high-throughput sequencing, quantifying DOM–microbe associations in nature and disentangling the effects of global change drivers (temperature increase and nutrient enrichment) remains challenging. The authors introduce the Energy–Diversity–Trait integrative Analysis (EDTIA) framework to quantify specialization in bipartite DOM–microbe networks and assess how distal drivers (temperature, nutrients) affect associations via proximal drivers: energy supply, diversity (chemodiversity and biodiversity), and molecular traits. They pose three questions: (1) how DOM molecular composition and its association with microbial biodiversity vary along temperature and nutrient gradients at the compositional level; (2) how specialization between DOM and microbes inferred from interaction networks varies along these gradients at the molecular level; and (3) how temperature and nutrient enrichment influence specialization through energy, diversity, and traits. Field microcosm experiments along subtropical (China) and subarctic (Norway) elevation gradients with controlled initial DOM composition but locally colonized microbiota were used to address these aims.

Background literature indicates: DOM is a complex mixture whose composition and traits (e.g., H/C, O/C, NOSC, aromaticity) influence degradation and persistence. Resource diversity can promote microbial specialization, while microbial diversity can expand metabolic pathways for DOM transformation. Ultrahigh-resolution FT-ICR MS enables molecular-level characterization beyond bulk optical metrics. Prior co-occurrence and network approaches in aquatic and terrestrial systems have begun linking DOM molecules to microbes, but integrating the roles of energy supply, diversity, and molecular traits under global change remained unresolved. This study builds on concepts of bipartite ecological networks and specialization metrics (H2') to partition production vs decomposition processes in DOM–microbe associations.

Comparative field microcosm experiments were established on Laojun Mountain (China; subtropical) and Balggesvarri Mountain (Norway; subarctic) in 2013. At each of five elevations per mountain (China: 3822, 3505, 2915, 2580, 2286 m; Norway: 750, 550, 350, 170, 20 m), 30 microcosms (1.5 L bottles) were set with 15 g sterilized Taihu Lake sediment and 1.2 L sterilized artificial lake water. Ten nutrient levels of nitrate (KNO3; 0–36.00 mg N L−1: 0, 0.45, 1.80, 4.05, 7.65, 11.25, 15.75, 21.60, 28.80, 36.00 mg N L−1) were imposed; KH2PO4 was added to keep initial N:P at 14.93 (similar to Taihu Lake 2007). Each nutrient level had three replicates, totaling 300 microcosms (2 regions × 5 elevations × 10 levels × 3 reps). Microcosms incubated in situ for one month, open to airborne colonization; bottles were partially buried to reduce UV; water temperature and pH measured late in the experiment. Samples of water and sediments were collected for physicochemical analyses, microbial community profiling, and DOM characterization. Physicochemical and microbial data: Sediment TN, TP, dissolved nutrients (NO3−, NO2−, NH4+, PO43−), TOC, DOC, chlorophyll a (Chl a); overlying water nutrients and pH were measured. Sediment bacterial communities were profiled by 16S rRNA gene amplicon sequencing (QIIME v1.9; OTUs at 97%; rarefied to 20,000 reads/sample). Alpha diversity (OTU richness) and beta diversity (Bray–Curtis; NMDS axes) were computed. DOM molecular characterization: Sediment DOM was extracted (SPE) and analyzed by ultrahigh-resolution ESI FT-ICR MS (15 T solariX XR, negative mode, m/z 150–1200; resolving power 750,000 at m/z 400). Peaks were picked (S/N > 7), internally calibrated, and elemental formulae (C, H, O, N, S, P) assigned via Formularity with strict constraints (mass error <1 ppm; chemical rules). Across 300 samples, 19,538 molecular formulae were assigned. Molecules were categorized by van Krevelen classes (lipids, proteins, amino sugars, carbohydrates, unsaturated hydrocarbons, lignin, tannin, condensed aromatics) and by 12 elemental combinations. Relative abundances were normalized intensities. DOM features: Chemodiversity quantified as alpha diversity (molecular richness) and beta diversity (Bray–Curtis; NMDS). Overall composition visualized with detrended correspondence analysis (DCA). Sixteen molecular traits were computed per formula (mass, C count, AI, DBE, DBE-O, DBE-AI, GFE, Kendrick defect, NOSC, O/C, H/C, N/C, P/C, S/C, Ymet) and summarized as weighted means per sample. Molecules were also clustered into 10 groups via hierarchical clustering on traits. DOM–bacteria associations at compositional level: Pearson correlation between DOM and bacterial alpha diversity; Mantel correlation between beta diversities; Procrustes analysis between community ordinations to assess congruence and residuals along gradients. Molecular-level associations and networks: Spearman correlations between each DOM molecule and each bacterial OTU (or genus); for each molecule, difference between mean positive and mean absolute negative ρ values was mapped against H/C and O/C. Bipartite networks were inferred using SparCC on filtered matrices (molecules and genera present in >50% of samples): China (1340 molecules, 75 genera), Norway (1246 molecules, 49 genera). Edges retained for |ρ| ≥ 0.30, with separate negative (ρ < −0.30) and positive (ρ ≥ 0.30) networks. For each microcosm, negative and positive sub-networks were assembled based on local presence. Network specialization was quantified by H2' at network level and d' for individual molecules and genera; indices were standardized (z-scores) against null models (swap.web; 100 randomizations). Network visualization used circlize; analyses used bipartite R package. Statistical analyses: Relationships with nutrient enrichment and elevation were tested via linear models (one-sided F-statistics). Breakpoints in composition along nutrient gradients were identified by piecewise linear regression (segmented) and by gradient forest (standardized density of splits). Variation partitioning (VPA) attributed variance in DOM features to environments (climate, human impacts, contemporary nutrients), energy supply, and biodiversity. Random forests (2000 trees) assessed relative importance of environment, energy, biodiversity, chemodiversity, and traits for H2'. Structural equation models (SEM; lavaan) with composite variables tested direct and indirect effects of distal drivers (temperature, nutrient enrichment, contemporary nutrients) via proximal drivers (energy, biodiversity, chemodiversity, traits) on H2', selecting best-fit models by AIC, χ2 (P > 0.05), CFI > 0.95, SRMR < 0.05. Prediction application: Using SEM parameters from China, decadal (2007–2018) predictions of network specialization across 32 sites in Taihu Lake were made from annual means of water temperature and total nitrogen, propagating indirect effects through predicted contemporary nutrients, energy, biodiversity, chemodiversity, and traits.

• DOM compositional responses: Nutrient enrichment generally increased DOM molecular richness across elevations in both regions. Gradient forest and piecewise regressions identified abrupt compositional changes between ~1.80 and 4.05 mg N L−1. Weighted mean H/C declined with nutrient addition in China (to <1.5), indicating less bioavailable DOM, but remained ≥1.5 across nutrient levels in Norway.
• DOM–bacteria congruence: DOM composition was significantly predicted by bacterial community composition (Procrustes M2 = 0.701, P ≤ 0.001). Compositional congruence (Mantel r between beta diversities) strengthened with nutrient enrichment, especially in Norway at low nutrient levels (<1.80 mg N L−1).
• Molecular-level correlations: More labile molecules (H/C ≥ 1.5) tended to have negative correlations with OTUs (indicative of degradation), while more recalcitrant molecules (H/C < 1.5) showed more positive correlations (indicative of production), particularly in Norway. Differences between mean positive and mean absolute negative ρ (Δρ) mapped onto H/C–O/C space separated degradation reactants (H/C 1.5–2.0, O/C 0.4–1.0) from production products (H/C 1.0–1.5, O/C 0–0.5).
• Bipartite networks scale: China had 6916 negative and 8409 positive interactions; Norway had 1313 negative and 2888 positive interactions (SparCC |ρ| ≥ 0.30). In China, nutrient enrichment increased both number and strength of negative interactions (higher fraction of strong negative SparCC ρ; Fig. S11).
• Dependence on molecular traits: Molecule clusters by traits (10 groups) associated differently with negative vs positive networks. Trait similarity among molecules predicted similarity in their microbial correlations (positive Mantel relationships between trait distance and SparCC distance; P ≤ 0.001). Traits explained interaction strength more strongly in negative than positive networks for all molecules and most classes.
• Specialization: Network specialization (H2') was higher in negative than positive networks in both regions (China t = 2.11, P = 0.04; Norway t = 23.57, P ≤ 0.001). Mean H2' (both negative and positive) was higher in Norway than China (negative t = −10.19, P ≤ 0.001; positive t = −6.56, P ≤ 0.001), indicating more specialized decomposition and production in subarctic conditions.
• Nutrient effects on specialization: In China, nutrient enrichment decreased H2' of negative networks (more generalized decomposition; notably for recalcitrant components like lignin and CHNO) and increased H2' of positive networks, especially at lower elevations (warmer sites). This suggests greater DOM vulnerability to decomposition and reduced production under enrichment in the subtropics.
• Drivers of specialization (EDTIA): For negative networks, H2' correlated most with DOM molecular composition (r = 0.77, P ≤ 0.001), molecular richness (r = −0.76, P ≤ 0.001), and molecular N/P ratio (r = 0.76, P ≤ 0.001). Random forests ranked chemodiversity highest, followed by molecular traits; for positive networks, chemodiversity, biodiversity, environmental variables, and energy were similarly important. SEM showed temperature and nutrient enrichment influenced negative-network H2' mainly indirectly via energy supply and molecular traits (dominant direct effects of traits on H2' with standardized effect ≈ 0.57, P < 0.001). For positive networks, both drivers had large total effects mediated by biodiversity, chemodiversity, and traits (e.g., total mean effects of climate change: China 0.51 vs Norway −0.40; human impacts: China 0.44 vs Norway 0.62).
• Taihu Lake predictions (2007–2018): With mean TN reduction of 1.24 ± 1.41 mg L−1 and mean temperature decrease of 0.20 ± 0.87 °C, predicted specialization shifted towards more specialized decomposition and more generalized production. Estimated changes in H2' relative to 2007: negative networks +0.65 ± 0.58; positive networks −0.65 ± 0.46, with largest changes in the most eutrophic northern/western regions.

The study demonstrates that DOM–microbe interactions can be partitioned into decomposition (negative correlations) and production (positive correlations) processes whose specialization is differentially modulated by global change drivers. Negative (decomposition) networks are more specialized than positive (production) networks, especially in colder, subarctic conditions, implying reduced DOM vulnerability to decomposition where specialized consumers are required. Nutrient enrichment in the subtropics reduces specialization of negative networks (generalizing degradation across more taxa) and increases specialization of positive networks, together suggesting heightened DOM vulnerability to decomposition and constrained production under enrichment and warmth. Molecular traits strongly govern negative-network specialization, highlighting that DOM recalcitrance traits (e.g., low H/C, higher aromaticity/DBE, elemental ratios) modulate which microbes can decompose molecules, and how this control shifts with energy supply and nutrients. The EDTIA framework effectively separates direct and indirect effects of temperature and nutrients via energy, diversity, and traits, providing mechanistic insight and enabling predictive upscaling (as shown for Taihu Lake). These findings address the posed questions by linking compositional congruence and network specialization to environmental gradients and by attributing observed patterns to specific proximal drivers.

This work introduces and applies the Energy–Diversity–Trait integrative Analysis (EDTIA) framework to quantify and explain DOM–microbe associations under global change. Across subtropical and subarctic climates, decomposition-related (negative) networks are more specialized than production-related (positive) networks, with specialization controlled predominantly by DOM molecular traits and chemodiversity. Nutrient enrichment (especially in warmer subtropical settings) generalizes negative interactions and specializes positive ones, increasing DOM vulnerability to microbial decomposition. Structural equation modeling shows temperature and nutrients act mainly through energy supply and molecular traits (negative networks) and through biodiversity, chemodiversity, and traits (positive networks). Applying EDTIA to Taihu Lake suggests management-driven oligotrophication increased the specialization of decomposition and decreased specialization of production, potentially enhancing carbon storage. Future research should incorporate microbial functional traits (e.g., via metagenomics/transcriptomics) and translate EDTIA outputs into process-based ecosystem models to improve predictions of carbon cycling under multiple global change drivers.

• Mechanistic attribution: While correlations and network inference suggest decomposition vs production roles, definitive causal mechanisms (e.g., specific enzymatic pathways) are not resolved; FT-ICR MS provides formulae, not explicit molecular structures.
• Experimental context: Field microcosms, although standardized for initial DOM, differ from natural sediments in species pools and environmental heterogeneity; open colonization and short incubations may not capture longer-term dynamics.
• Prediction uncertainties: Taihu Lake projections rely on SEM parameters from microcosms, lack full spatiotemporal environmental detail (e.g., varying N:P), and may be affected by unmeasured local factors.
• Analytical constraints: Intensity-based relative abundances may not reflect absolute concentrations; SparCC infers associations under compositional data assumptions and thresholding may exclude weak but real interactions.
• Biome differences: Indirect effects via microbial composition and activity likely vary across biomes; generalization to other systems requires further validation.