Stratification-driven divergence between taxonomic and functional diversity in a deep lake microbiome

Abstract

Thermal stratification drivers of microbial community organization and functional potential in deep lakes, yet comparative analyses of epilimnetic and hypolimnetic microbiome dynamics remain limited. In this study, we combined 16S rRNA gene sequencing with functional microarray (GeoChip 5.0) to investigate stratification-induced shifts in microbial community composition and functional structure in Lake Fuxian, a deep monomictic plateau lake in Yunnan Province, Southwest China. Our analyses revealed a partial decoupling between taxonomic and functional diversity across water layers: the oxygen-depleted hypolimnion harbored higher bacterial taxonomic richness and distinct taxa (Nitrospirae, Parcubacteria, and Thaumarchaeota), whereas the epilimnion exhibited greater functional gene richness with lower beta diversity, indicating enhanced metabolic flexibility. Molecular ecological network analysis uncovered contrasting interaction patterns, with hypolimnetic communities exhibiting greater complexity and modularity. Notably, the Chloroflexi-associated amyA gene emerged as a module hub in hypolimnetic functional molecular ecological networks while distinct connector taxa characterized both epilimnetic and hypolimnetic species molecular ecological networks. Multivariate analyses identified dissolved oxygen and nutrient availability as key environmental drivers of vertical microbial stratification. These findings elucidate microbial adaptation to stratified conditions and underscore the distinct roles of epilimnetic and hypolimnetic communities in biogeochemical cycling in deep lakes experiencing climate-mediated thermal regime shifts.

In a deep lake, we found that oxygen-rich surface and oxygen-poor deep waters host vastly different microbial communities, each with unique structures and functions that drive carbon and nutrient cycling.

Introduction

Vertical thermal stratification, driven by seasonal temperature changes, is a key feature of deep lakes. In warm summer, stable thermal stratification impedes vertical mixing between the epilimnion and hypolimnion, leading to contrasting thermal and redox conditions above and below the thermocline (Wang et al. 2017, Liu et al. 2024). The formation of this phenomenon significantly influences the exchange of oxygen and nutrients between the epilimnion and hypolimnion, thereby affecting the metabolic processes of aquatic organisms (Zhang et al. 2015b, Yu et al. 2014). Moreover, thermal stratification profoundly impacts water chemistry and material cycling, particularly through its effects on microbial activities (Lawson and Anderson 2007, Moreira et al. 2011). As microorganisms are pivotal in the biogeochemical cycle and microbial food webs of aquatic ecosystems (Cotner and Biddanda 2002, Comeau et al. 2012, Sylvan et al. 2012), revealing the effects of vertical stratification of water bodies on the taxonomic and functional gene diversity of microbial communities is the key to further understanding the material cycle and energy flow of deep-water lake ecosystems.

Despite growing recognition of microbial stratification in aquatic ecosystems, the effects of thermal stratification on both taxonomic and functional diversity in deep-water lakes remain poorly understood, particularly regarding the underlying mechanisms (Lv et al. 2024a, Yu et al. 2014, Yue et al. 2023, 2024b). Thermal stratification profoundly influences vertical physicochemical gradients, including pH, dissolved oxygen, conductivity, chlorophyll a, and nutrient distributions (Becker et al. 2010, Ryabov et al. 2010, Yu et al. 2010), which in turn shape microbial community structure throughout the water column (Lv et al. 2024a, Wu et al. 2024a, Yue et al. 2023, Wei et al. 2024, 2024b). Metagenomic studies on meromictic lakes have further revealed that such stratification also drives vertical niche partitioning of microbial metabolisms, with enhanced synergy between sulfate reducers and fermenters in deep anoxic layers (Cabello-Yeves et al. 2023). Studies in Lake Biwa, for instance, revealed pronounced beta diversity differences between the epilimnion and hypolimnion during stratification, while alpha diversity remained largely conserved, suggesting that community reorganization was driven by factors beyond simple environmental filtering (Okazaki and Nakano 2016). This notion of complex niche partitioning is further exemplified by a study on deep oligotrophic Lake Redon, which documented not only higher hypolimnetic diversity (partly driven by low dissolved oxygen) but also the presence of previously unrecognized metabolic pathways (e.g. carbon monoxide oxidation) (Llorens-Marès et al. 2020). These cases illustrate that the reorganization of microbial communities extends beyond taxonomy, introducing substantial complexity at the functional level. Here, microbial taxonomic shifts may (Comte et al. 2013, Yang et al. 2015, Galand et al. 2018) or may not (Louca et al. 2016, 2018) correspond to functional gene structure changes, depending on the degree of functional redundancy, the phylogenetic conservation of the functions in question (Amend et al. 2016), and the rate of horizontal gene transfer (Stepanauskas et al. 2025). Such apparent contradictions highlight the need to disentangle taxonomic versus functional responses to thermal stratification. The distribution is particularly crucial given that functionally equivalent microbial groups often show distinct environmental distributions (Allison and Martiny 2008, Comte et al. 2013, Gonzalez et al. 2015), with dispersal limitations and environmental adaptation potentially decoupling genetic potential from expressed traits (Martinez-Garcia et al. 2012, Walsh et al. 2015).

Lake Fuxian, a deep (maximum depth 155 m) monomictic plateau lake in Yunnan Province, China, serves as an exemplary system for investigating microbial stratification dynamics due to its predictable annual mixing regime (Zhou et al. 2022, Xie et al. 2024). The lake undergoes complete vertical mixing during January–February and remains strongly stratified from March to December, creating distinct physicochemical gradients that shape microbial habitats. In this study, we employed an integrated molecular approach combining Illumina MiSeq sequencing of the 16S rRNA gene with GeoChip 5.0 functional microarray analysis to simultaneously resolve bacterioplankton community composition (BCC) and microbial functional genes (MFG) structure across stratification layers. Unlike the untargeted nature of metagenomic sequencing, the GeoChip 5.0 microarray provides a high-throughput, semiquantitative platform for profiling MFG, detecting over 57 000 probes that target key biogeochemical cycling processes and other ecological functions (Shi et al. 2019, Lu and Su 2022). By coupling these datasets with molecular ecological network analysis, we: (1) characterized stratification-driven patterns in taxonomic versus functional diversity of microbial communities, (2) identified dissolved oxygen and nutrients as key drivers of vertical zonation in bacterioplankton, and (3) observed conserved functional potential despite taxonomic turnover across strata, coinciding with modular network architectures. This work establishes a new framework for understanding microbial adaptation to lacustrine stratification while providing methodological benchmarks for predicting ecosystem responses to climate-mediated changes in mixing regimes.

Materials and methods

Study site, sample collection, and measurements of environmental parameters

Lake Fuxian, situated in Yuxi City of Yunnan Province, extends 31.5 km from north to south and 6.7 km from east to west, covering 212 km² with an average depth of 89.7 m and a maximum depth of 155 m (Chen et al. 2019, Miao et al. 2022). In January and February every year, the water body in Lake Fuxian was vertically mixed. From March to December, the thermocline experienced a process of formation, development, and disappearance. In summer, there was a thermocline at a depth of 15–40 m in the north of Lake Fuxian, and the temperature dropped from 23°C to 15°C. There were obvious differences in physical and chemical environmental factors between the epilimnion (0–10 m) and hypolimnion (41–155 m) (Zhou et al. 2022). To study the BCC and MFG structure in Lake Fuxian, we collected water samples from the epilimnion (0.5 and 10 m depths) and the hypolimnion (100 and 130 m depths) at the site of S1 (24° 35′ 22.80″ N, 102° 54′ 59.10″ E) in the northern lake area of Lake Fuxian with Multi Water Sampler (Slimline, HYDRO-BIOS, Germany) in August 2013 (Fig. S1). Three parallel water samples were collected from each water layer (0.5, 10, 100, and 130 m), stored in 10-l plastic buckets, and transported to the laboratory in an insulated container with ice for further processing. In addition, the water temperature (T), pH, oxidation–reduction potential (ORP), conductivity (Cond), dissolved oxygen (DO), and turbidity (NTU) of different water layers were measured in situ using a Multi-Parameter Water Quality Sonde (YSI 6600, Yellow Springs, USA).

In the laboratory, part of each water sample was used for chemical analyses. Concentrations of total nitrogen (TN), dissolved total nitrogen, ammonium (NH₄⁺-N), nitrate (NO₃⁻-N), total phosphorus (TP), dissolved total phosphorus (DTP), orthophosphate (PO₄³⁻-P), dissolved organic carbon (DOC), and chlorophyll a (Chl a) were determined following standard protocols (Greenberg et al. 1992). For microbial community and functional gene analyses, 1 l of water was filtered through 0.2 μm polycarbonate membranes (Isopore™ Membrane Filters, Merck Millipore, USA). Filters were immediately flash-frozen in liquid nitrogen and stored at −80°C until further processing.

DNA extraction, high-throughput sequencing and GeoChip analysis

Total microbial DNA was extracted using a phenol–chloroform protocol wherein filters were cut into pieces and subjected to an extended 24-h lysis incubation, followed by ethanol precipitation, and then amplified via polymerase chain reaction (PCR) (Li et al. 2011). Amplification was performed on a Gene Amp PCR-System® 9700 (Applied Biosystems, Foster City, CA, USA). Each 25 μl reaction mixture contained 0.4 μmol·l⁻¹ of forward primer 515F (5′-GTGCCAGCMGCCGCGG-3′) and reverse primer 806R (5′-GGACTACHVGGGTWTCTAAT-3′) (Li et al. 2017a), 10 ng DNA template, 2.5 μl 10X AccuPrime™ PCR buffer Ⅱ, and 0.5 units of AccuPrime™ Taq Hi-Fi DNA polymerase (Invitrogen, Carlsbad, CA, USA). PCR cycling conditions were as follows: initial denaturation at 94°C for 1 min; denaturation at 94°C for 20 s, annealing at 53°C for 25 s, and extension at 68°C for 45 s, repeated for 30 cycles; followed by a final extension at 68°C for 10 min. To minimize stochastic errors, each sample was amplified in triplicate. The PCR products were pooled, examined by agarose gel electrophoresis, purified using a QIAquick Gel Extraction Kit (Qiagen, Germantown, MD, USA) and quantified with PicoGreen (BMG Labtech, Jena, Germany). After addition of PhiX (Illumina, San Diego, CA, USA), sequencing was conducted on the Illumina MiSeq platform (Li et al. 2017a).

Functional genes were profiled using GeoChip 5.0, which includes over 57 000 probes representing more than 144 000 sequences from 393 gene families related to carbon, nitrogen, phosphorus, and sulfur cycling, etc (Wang et al. 2014). For each sample, 500 ng of purified DNA was labeled with Cy-3 dye (GE Healthcare, CA, USA) using random priming as described previously (Wu et al. 2006a, b, Van Nostrand et al. 2009). Labeled DNA was purified with a QIAquick kit (Qiagen, CA, USA), dried in a SpeedVac (Thermo Savant, NY, USA) for 45 min, and then resuspended in 42 μl hybridization solution containing 1 × HI-RPM hybridization buffer, 1 × CGH blocking agent, 0.05 µg·µl⁻¹ Cot⁻¹ DNA, 10 pM universal standard, and 10% formamide. The mixture was denatured at 95°C for 3 min and held at 37°C prior to hybridization. Hybridization was performed at 67°C using an Agilent hybridization station (Agilent Technologies Inc., Santa Clara, CA, USA). Hybridized arrays were scanned on a NimbleGen MS200 scanner (Roche NimbleGen, Inc., Madison, WI, USA), and signal data were extracted with Agilent Feature Extraction 11.5 software.

Data preprocessing

For sequencing data, raw reads were first quality-filtered and truncated to 140 bp using moira (Puente-Sánchez et al. 2016), with sequences discarded if the probability of multiple errors exceeded 0.005. The resulting high-quality reads were then processed with DADA2 (Callahan et al. 2016) for the inference of amplicon sequence variants (ASVs). The DADA2 workflow was executed with its standard default parameters and included denoising, merging of paired-end reads, and removal of chimeric sequences. Taxonomic assignment was performed at the genus level using the SILVA database (release 138). Following taxonomic assignment, all ASVs classified as “Chloroplast” (order) or “Mitochondria” (family) were identified and removed from the dataset to obtain a pure bacterial community. The final bacterial ASV table was rarefied to 19 969 reads per sample, yielding a total of 816 ASVs for downstream analysis. To improve visualization and reduce disparities among taxa or gene categories spanning several orders of magnitude, richness values of both ASVs and functional genes were log-transformed using log₁₀(richness + 1) prior to plotting.

GeoChip data were processed following the established microarray pipeline (Yang et al. 2014, Yan et al. 2015). Briefly: (i) spots with signal-to-noise ratio ≥ 2 were retained; (ii) only probes present in >50% of epilimnetic or hypolimnetic samples (≥3 of six replicates per layer, with prevalence calculated separately for each layer) were retained; (iii) signal intensities were ln(x+1)-transformed; and (iv) normalized by dividing each probe signal by the mean of positive probes in the sample (Chan et al. 2013). The obtained microarray data matrix (normalized signal intensity) was considered as “species” abundance for MFG richness and composition analysis (Zhou et al. 2010) and subsequently used to assess microbial metabolic potentials (Ren et al. 2018). Finally, (v) the difference in normalized signal intensity of a specific gene family was calculated as: (Σ_hypolimnion/Σ_epilimnion) − 1, where the Σ_epilimnion and Σ_hypolimnion were the total normalized signal intensity of a specific gene family in the epilimnion and hypolimnion, respectively (Li et al. 2021).

Molecular ecological network construction and node role identification

To elucidate the interaction patterns and identify keystone taxa and functional genes within the stratified water column, we constructed molecular ecological networks at both taxonomic and functional gene levels. Two types of molecular ecological networks were generated: (i) species-level MENs (sMENs) based on ASVs, and (ii) functional MENs (fMENs) derived from functional genes, with separate networks for epilimnion and hypolimnion layers. To simplify data while retaining ecological relevance, we restricted the analysis to functional genes associated with major biogeochemical cycles of carbon, nitrogen, phosphorus, and sulfur. Networks were constructed using the Random Matrix Theory (RMT)-based Molecular Ecological Network Analysis Pipeline (http://ieg2.ou.edu/MENA/) (Zhou et al. 2010, Deng et al. 2012). Briefly, the sMEN or the fMEN were constructed based on the Pearson correlation matrix calculated by the correlation coefficient (r-value) between each of the two detected ASVs or genes (Deng et al. 2012). A series of thresholds was then applied to the matrix and the similarity threshold (st) was kept or matrix eigenvalues were calculated using an RMT-based approach (Deng et al. 2016). During these calculations, the most suitable st was selected to obtain the Poisson distribution of the calculated eigenvalues, which indicates the nonrandom properties of a complex system (Deng et al. 2016). Networks were partitioned into modules using fast greedy modularity optimization. For each empirical network, 100 random networks with identical node and edge numbers were generated, and topological differences were assessed by Z tests. Network visualization was performed with Cytoscape 3.6.1 (Shannon et al. 2003) and Gephi 0.9.2-beta.

The topological role of each node was evaluated using within-module connectivity (Zi) and among-module connectivity (Pi) (Guimerà and Amaral 2005). Nodes were classified into four groups: network hubs (Zi > 2.5, Pi > 0.62), connectors (Zi < 2.5, Pi > 0.62), module hubs (Zi > 2.5, Pi < 0.62), and peripherals (Zi < 2.5, Pi < 0.62). The first two categories represent key nodes in network structure (Zhou et al. 2010).

Statistical analyses

All analyses were carried out in R v4.1.3. Differences in environmental parameters, alpha diversity (taxonomic and functional gene richness), functional gene abundance, and normalized signal intensity between layers were tested using Student’s t-test (for normally distributed, homoscedastic data) or Wilcoxon rank-sum test (otherwise). The richness of the bacterioplankton community and functional genes were analysed using the Vegan package of R software (http://www.r-project.org). Vegan 2.6–2 package was used to analyse BCC and MFG structure by principal coordinates analysis (PCoA). Nonparametric multivariate tests, including permutational multivariate analysis of variance (PERMANOVA) with 999 permutations, analysis of similarity (ANOSIM), and multiple response permutation procedure (MRPP), using distance matrices were performed to determine the differences in BCC and MFG structue. The assessment of bacterial community composition and functional gene differences between the epilimnion and hypolimnion, based on the Bray–Curtis dissimilarity index, is conducted using the ANOSIM test.

Transformation-based redundancy analysis (tb-RDA) was applied to explore relationships between environmental factors and microbial communities, as detrended correspondence analysis indicated axis lengths <2 for both ASV and gene data. All environmental variables except water temperature, pH, and ORP were log(x+1)-transformed prior to analysis. Tb-RDA was performed using the vegan 2.6–2 package in R.

Results

Microbial taxonomic and functional gene richness

A total of 816 bacterial ASVs were obtained after quality filtering, removal of chloroplast sequences, and rarefaction of the entire dataset to 19 969 reads per sample. ASV richness in the epilimnion was significantly lower than that in the hypolimnion (P < .05; Fig. 1A). This pattern was consistently observed across all detected phyla, including Proteobacteria, Actinobacteria, Bacteroidetes, Planctomycetes, Verrucomicrobia, Cyanobacteria, Firmicutes, Acidobacteria, Armatimonadetes, Deinococcus-Thermus, Chlamydiae, Chloroflexi, Gemmatimonadetes, Nitrospirae, Parcubacteria, Thaumarchaeota, BRC1, and Rhodothermaeota, with significantly fewer ASVs detected in the epilimnion than in the hypolimnion for each phylum (all P < .05; Fig. 1A).

Figure 1.

Log-transformed richness and distribution (shared or unique) of the bacterial community and functional genes identified from the epilimnion and hypolimnion in Lake Fuxian. (A) ASV richness (log₁₀-transformed); (B) MFG richness (log₁₀-transformed); (C) distribution of detected ASVs; and (D) distribution of detected genes. Significant differences between the epilimnion and hypolimnion are indicated by * (P < .05).

A total of 24 979 functional genes were detected in the 12 samples, of which 21 971 genes were derived from bacteria, 597 genes were from archaea, 1883 genes were from fungi, and the remaining genetic sources were not identified. Furthermore, in both the whole and all of the detected gene categories (e.g. carbon, nitrogen, phosphorus, and sulfur cycling, organic remediation, metal homeostasis, virulence, and secondary metabolism), the gene richness of the epilimnion was higher than that of the hypolimnion (all P < .05; Fig. 1B).

Spearman’s correlation analysis showed that the bacterial taxonomic richness showed a positive correlation with NO₃⁻-N and DOC, and a negative correlation with T, DO, Cond, NTU, TN, TP, and PO₄³⁻-P (Spearman’s correlation; all P < .05; Table S2). The functional gene richness was positively correlated with T, DO, Cond, Chl a, NTU, TN, TP, DTP, and PO₄³⁻-P, while negatively correlated with NO₃⁻-N and DOC (all P < .05; Table S2).

The overlap of ASV and functional genes between the two water layers was calculated. The epilimnion harbored a lower proportion of unique ASVs compared to the hypolimnion, both overall and within most phyla (Fig. 1C). Specifically, 42.16% of total ASVs were shared between layers, while 13.85% were unique to the epilimnion and 44.00% to the hypolimnion. Notably, Nitrospirae, Parcubacteria, Thaumarchaeota, BRC1, and Rhodothermaeota were exclusively detected in the hypolimnion (Fig. 1C). In contrast, 69.94% of functional genes were shared between layers, with 29.21% unique to the epilimnion and only 0.85% unique to the hypolimnion (Fig. 1D).

Microbial taxonomic and functional gene structure

PCoA revealed distinct clustering of BCC and MFG structure between the epilimnion and hypolimnion along the PCoA axis1 and PCoA axis2 (Fig. 2A and B). Consistently, three different nonparametric multivariate statistical tests also demonstrated that BCC and MFG structure differed significantly between the epilimnion and the hypolimnion (PERMANOVA, ANOSIM, and MRPP, all P < .01; Table S4). Bray–Curtis dissimilarity analysis indicated greater variability in BCC within the epilimnion but lower variability in MFG structure compared to the hypolimnion (P < .01; Fig. 2C and D). This suggests that the dynamic and heterogeneous conditions in the epilimnion lead to rapid taxonomic turnover with relatively conserved functional roles. In contrast, the environmental stability of the hypolimnion may support taxonomically stable communities that are functionally more differentiated across depths, possibly due to niche partitioning and specialized metabolic processes in this low-energy environment.

Figure 2.

PCoA and Bray–Curtis dissimilarity of the bacterial community and functional genes. (A) PCoA of ASVs; (B) PCoA of functional genes; (C) Bray–Curtis dissimilarity of taxonomy; and (D) Bray–Curtis dissimilarity of functional genes. Significant differences between the epilimnion and hypolimnion are indicated by ** (ANOSIM, P < .01). The values on axis 1 and 2 in panels (A and B) are percentages of total variations that can be attributed to the corresponding axis.

At the phylum level, 18 bacterial phyla were identified in both water layers. Proteobacteria (27.24%−46.54%) and Actinobacteria (27.02%−41.97%) dominated the bacterioplankton community, collectively accounting for about 70% of total relative abundance (Fig. 3). Proteobacteria and Actinobacteria were the main bacterial groups in Lake Fuxian, accounting for about 70% of the total relative abundance of bacteria. There were significant differences in the relative abundance of Actinobacteria, Bacteroidetes, Chloroflexi, and Acidobacteria between epilimnion and hypolimnion (Wilcoxon; P < .05). Actinobacteria (12.71% higher than the hypolimnion) and Bacteroidetes (2.06% higher than the hypolimnion) were more abundant in the epilimnion, while Chloroflexi (2.17% higher than the epilimnion) and Acidobacteria (4.73% higher than the epilimnion) were more abundant in the hypolimnion. In addition, Thaumarchaeota, Nitrospirae, Parcubacteria, BRC1, and Rhodothermaeota were unique in the hypolimnion (Fig. 1C).

Figure 3.

Mean relative abundance of bacteria at the phylum (A) and genus (B) levels in the epilimnion (n = 6) and hypolimnion (n = 6).

To elucidate the vertical distribution of microbial functional potential in Lake Fuxian, we compared the relative normalized signal intensities of genes involved in major biogeochemical cycles between the epilimnion and hypolimnion. The metabolic potentials of most gene categories, including carbon, nitrogen, and sulfur cycling, as well as organic remediation, virulence, and secondary metabolism, differed significantly between the two layers (Fig. 4). In contrast, functional genes associated with phosphorus cycling and metal homeostasis showed no significant differences (Fig. 4A).

Figure 4.

Differences in the normalized intensities of certain biogeochemical cycling processes between the epilimnion and hypolimnion. The y-axis represents the relative difference in normalized signal intensity between the hypolimnion and epilimnion, calculated as (Σ_hypolimnion/Σ_epilimnion) – 1, where positive values indicate higher intensity in the hypolimnion and negative values indicate higher intensity in the epilimnion. (A) Each category of total functional gene; (B) each subcategory of carbon cycling; (C) each subcategory of nitrogen cycling; (D) each subcategory of phosphorus cycling; (E) each subcategory of sulfur cycling; and (F) each subcategory of organic remediation. Significant differences between the epilimnion and hypolimnion are indicated by * (P < .05).

Among all evaluated functional pathways, the normalized signal intensities of most subcategories (79 out of 81) in the hypolimnion were significantly lower than those in the epilimnion (P < .05; Fig. S2) except for the nonsignificant reduction in genes associated with chlorinated solvents—heterocyclic aromatics and silicon biosynthesis. For example, the normalized signal intensities of functional genes related to lipid degradation, methane oxidation, methanogenesis, the 3-hydroxypropionate/4-hydroxybutyrate cycle, silicon transport, rhodopsin metabolism, and virulence degradation were reduced by >40% in the hypolimnion compared with the epilimnion. In addition, functional genes related to hyaluronic acid degradation, heparin degradation, the reductive acetyl-CoA pathway, dissimilatory nitrogen reduction, nitrogen fixation, and organic remediation of halogenated compounds were reduced by >35% in the hypolimnion compared with the epilimnion (P < .05 in all cases; Fig. S2).

The sMENs and fMENs in the epilimnion and hypolimnion

The sMENs analysis revealed distinct network structures between lake layers, with the epilimnion network comprising 130 nodes and 425 edges (48% negative, 52% positive), compared to 215 nodes and 820 links (48% negative, 52% positive) in the hypolimnion network (Fig. 5A and B; Table S5). Hypolimnion networks displayed higher modularity, connectivity, and harmonic geodesic distances (P < .001 vs. random networks), suggesting stronger species interactions and diversity compared to the epilimnion (Table S5). A similar trend was observed for fMENs: the hypolimnion network contained 708 nodes and 576 links (17% negative, 83% positive), surpassing the epilimnion network (620 nodes, 398 links; 38% negative, 62% positive) in connectivity, clustering coefficient, and transitivity (Fig. 5C and D; Table S5). Both sMENs and fMENs showed nonrandom topology (Z-test; P < .001), with correlation values of 0.97 (sMENs) and 0.99 (fMENs) (Table S5).

Figure 5.

Differences in network interactions of these ASVs (A and B) and genes (C and D) between the epilimnion and hypolimnion. A connection indicates a strong (Spearman’s r > 0.8 or r < −0.8) and significant (P < .01) correlation. A blue line indicates a negative interaction between two nodes, and a red line indicates a positive interaction. The size of each node is proportional to the number of connections (i.e. degree).

Zi-Pi plot analysis revealed distinct topological patterns between the two water layers. The hypolimnion sMENs exhibited a higher abundance of connectors (e.g. Cyanobacteria, Actinobacteria) compared to the epilimnion sMENs (dominated by Bacteroidetes and Proteobacteria), indicating stronger interspecies connectivity in the hypolimnion (Fig. S3; Table S6). In contrast, fMENs displayed a unique module hub in the hypolimnion—the amyA starch degradation gene—which underscored its pivotal role in carbon cycling and refractory organic matter degradation (Table S7). Both fMENs lacked connectors, and aside from this hub, node topological roles were consistent between layers (Fig. S3).

Environment drivers of microbial community variation

Redundancy analysis (RDA) identified DO, PO₄³⁻-P, and NH₄⁺-N as the major environmental drivers shaping BCC and MFG structure across water layers. The first two axes explained 17.25% of the observed variation in BCC (RDA1: 9.75%, RDA2: 7.50%) and 66.75% of the variation in MFG structure (RDA1: 62.72%, RDA2: 4.03%), respectively (for all canonical axes P < .05; Fig. 6A and B).

Figure 6.

Transformation-based redundancy analysis (tb-RDA) of relationships between the (A) bacterial community composition or (B) MFG structure and the environmental variables. Transformation-based correlation analysis showing the relationships between environmental variables and (C) bacterial community composition or (D) MFG structure in the epilimnion (F0.5_10 m) and hypolimnion (F100_130 m). The heatmaps integrate Mantel’s r and Spearman’s correlation coefficients. Color gradients represent the strength and direction of the correlations, with warm tones indicating positive and cool tones indicating negative associations.

Distinct physicochemical stratification characterized Lake Fuxian (Table S1). The epilimnion exhibited significantly higher temperature, DO, Chl a and PO₄³⁻-P (P < .01). In contrast, the hypolimnion showed significantly NO₃⁻-N and DOC (P < .01). Notably, NH₄⁺-N identified as a key microbial driver by RDA, showed no significant difference between layers (P > .05). This pronounced vertical gradient establishes distinct niches for microbial communities.

Mantel tests were performed to quantify the relationships between environmental variables and the vertical distribution of microbial communities (Fig. 6C and D). Taxonomic composition (ASVs) exhibited strong positive correlations with nearly all measured parameters in both the epilimnion and hypolimnion (Fig. 6C). In the epilimnion, TP (r = 0.863, P = .001) and NO₃⁻-N (r = 0.800, P = .001) were the strongest predictors, followed by T, DO, Cond, Chl a, TN, PO₄³⁻-P, and DOC (all r ≥ 0.64, P ≤ .001). In the hypolimnion, T (r = 0.900, P = .001) and DO (r = 0.903, P = .003) showed the highest correlations, with TP, NO₃⁻-N, Chl a, Cond, NTU, DTP, PO₄³⁻-P, TN, DOC, pH, ORP, and NH₄⁺-N also remaining significant (r = 0.27–0.85, P ≤ .046). Thus, overall community structure covaried with a wide range of physicochemical factors across depths. Functional gene composition, however, was associated with a much narrower set of variables (Fig. 6D). In the epilimnion, only T (r = 0.751, P = .002), DO (r = 0.750, P = .001), NO₃⁻-N (r = 0.683, P = .004), Chl a (r = 0.685, P = .001), TP (r = 0.549, P = .004), Cond (r = 0.563, P = .001), NTU (r = 0.437, P = .006), TN (r = 0.291, P = .047), and DOC (r = 0.373, P = .004) were significantly correlated with functional profiles. In contrast, within the hypolimnion, only T (r = 0.322, P = .034), DO (r = 0.316, P = .039), Chl a (r = 0.382, P = .014), and NH₄⁺-N (r = 0.267, P = .046) remained significantly correlated; all other variables were nonsignificant (P > .05).

Discussion

Stratification-driven decoupling of microbial taxonomic and functional gene richness in Lake Fuxian

Traditionally, the hypolimnia of deep lakes have been perceived as inhospitable for diverse microbial communities due to the combination of low temperatures, oxygen depletion, and limited light penetration (Wetzel and Gopal 2001, Cabello-Yeves et al. 2020). Contrary to this paradigm, our present study revealed significantly greater bacterial taxonomic richness in the hypolimnion compared to the epilimnion of Lake Fuxian—a phenomenon consistently observed across various aquatic systems including stratified marine systems (Ravenschlag et al. 1999), saline lakes (Humayoun et al. 2003, Dirnitriu et al. 2008, Zhang et al. 2020), and reservoirs (Zhang et al. 2015a). This persistent pattern strongly suggests that physical stratification creates distinct ecological niches that foster microbial diversity (De Wever et al. 2005, Meuser et al. 2013). Several factors likely contribute to this enhanced diversity: (1) elevated hypolimnetic nutrient concentrations (particularly NH₄⁺-N and PO₄³⁻-P) in the hypolimnion may promote richness by providing metabolic substrates (Li et al. 2017b, Sun et al. 2020); (2) sediment–water interactions introduce redox-sensitive compounds and refractory organic matter that further enrich nutrients and sustain diversity (Perkins et al. 2014); and (3) low-oxygen conditions that support alternative energy pathways, thereby reducing competitive exclusion and promoting niche specialization (Humayoun et al. 2003).

In contrast, the epilimnion exhibited significantly higher MFG richness in Lake Fuxian. This disparity pattern correlates strongly with greater phytoplankton biomass in the epilimnion, evidenced by elevated Chl a (Table S1) and consistent with the depth-dependent decline in phytoplankton diversity during stratification (Goldman et al. 1996). The limited phytoplankton diversity and biomass in the hypolimnion likely restrict the variety of labile DOC, consequently constraining metabolic niches and ultimately reducing functional gene richness (Wu et al. 2024b).

These contrasting vertical trends indicate a partial decoupling between microbial taxonomic diversity and metabolic potential. While the hypolimnion harbored greater bacterial diversity, its functional gene richness was constrained—likely reflecting evolutionary adaptation to stable, low-energy conditions (Dlugosch et al. 2022). Conversely, epilimnetic taxa exhibited broader metabolic capabilities (Linz et al. 2017), which indicates functional redundancy. In organic matter-rich environments, multiple taxa often perform overlapping ecological roles, thereby maintaining high functional potential despite lower taxonomic diversity (Comte et al. 2013, Louca et al. 2018). This is particularly evident in dominant epilimnion groups such as Actinobacteria and Bacteroidetes, which have evolved specialized mechanisms to efficiently utilize labile carbon sources and degrade diverse organic substrates (Pinhassi et al. 2004, Giovannoni et al. 2005, Salcher et al. 2011).

Vertical divergence in microbial taxonomic and functional gene structure under stratification in Lake Fuxian

Our study revealed pronounced stratification-induced divergence in both BCC and MFG structure between the epilimnion and hypolimnion of Lake Fuxian. Notably, while BCC exhibited higher β-diversity in the epilimnion (Fig. 2C), potentially driven by greater environmental variability and episodic mixing events that enhance stochastic colonization processes (Donohue et al. 2013, Dini-Andreote et al. 2015), the MFG structure conversely showed elevated β-diversity in the hypolimnion (Fig. 2D), reflecting increased spatial heterogeneity in metabolic potential shaped by persistent chemical gradients and fine-scale niche partitioning (Zhou et al. 2022, Silverstein et al. 2024). This vertical niche partitioning was further evidenced by the distinct distribution of key microbial phyla (Fig. 3A), with Actinobacteria and Bacteroidetes were enriched in the lake’s epilimnion where they utilize light-driven energy production and degrade phytoplankton-derived substrates respectively (Pinhassi et al. 2004, Giovannoni et al. 2005, Mary et al. 2006, Sharma et al. 2009), whereas Chloroflexi and Acidobacteria thrived in the hypolimnion’s dark, low-oxygen conditions specialized in refractory organic matter degradation (Okazaki et al. 2013, Okazaki and Nakano 2016). Additionally, the hypolimnion-enriched Nitrospirae, Parcubacteria, Thaumarchaeota, BRC1, and Rhodothermaeota highlighted functional specialization in nitrification under oxygen-limited conditions (Bel’kova et al. 2003, Tada and Grossart 2014, Inceoglu et al. 2015, Mukherjee et al. 2016), collectively demonstrating strong environmental filtering and vertical niche partitioning in this system.

Environmental drivers analysis through tb-RDA identified DO, PO₄³⁻-P, and NH₄⁺-N as primary factors shaping both BCC and MFG structure in Lake Fuxian (Fig. 6), consistent with global patterns of redox-controlled microbial stratification (De Wever et al. 2005, Liu et al. 2016, Zhou et al. 2022). Although PO₄³⁻-P concentrations were higher in the epilimnion (Table S1), its stronger correlation with functional genes in the hypolimnion suggested enhanced microbial phosphorus sensitivity under energy-limited conditions (Granéli et al. 2004, Vrede 2005), a pattern reinforced by the thermocline’s barrier effect on nutrient redistribution (Selmeczy et al. 2016).

Correspondingly, the hypolimnion showed generally reduced signal intensities across most functional gene categories (Fig. 4), indicating metabolic constraints from both substrate limitation and adverse redox conditions (Jiao et al. 2010, Berggren et al. 2023), particularly affecting energetically costly processes like methane oxidation (Apple and del Giorgio 2007, Wang et al. 2019, Havas et al. 2023, Foucault et al. 2025). However, genes involved in phosphorus cycling and metal homeostasis maintained stable signals across depths (Harke et al. 2016), underscoring their fundamental biochemical roles regardless of environmental stratification.

Although NH₄⁺-N concentrations did not differ significantly between layers, its strong explanatory power in RDA suggests that micro-scale turnover and bioavailability, rather than bulk concentration, shape microbial responses. Ammonium regeneration within particle aggregates and redox interfaces can generate steep chemical gradients that selectively stimulate nitrifiers and heterotrophs with high NH₄⁺ affinity (Kuypers et al. 2018). Likewise, variation in DO and PO₄³⁻-P influences microbial metabolism through metabolic trade-offs between aerobic respiration, denitrification, and phosphorus limitation (Jørgensen et al. 2019). These interactions reveal that environmental drivers affect microbial communities primarily via biogeochemical coupling mechanisms rather than through simple concentration differences.

Hypolimnetic microbial networks exhibit greater topological complexity and functional differentiation in Lake Fuxian

The hypolimnion of Lake Fuxian exhibited significantly greater topological complexity in its microbial networks compared to the epilimnion, with both sMEN and fMEN showing increased nodes, links, and higher average degrees (Fig. 5; Table S5). These structural characteristics suggest enhanced microbial interdependence under the stable physicochemical conditions of the hypolimnion. (Zhou et al. 2010). Notably, the hypolimnetic fMEN displayed elevated clustering coefficients and transitivity, reflecting tighter functional connectivity that likely promotes localized metabolic cooperation and system stability (Peura et al. 2015, Deutschmann et al. 2023). While the higher module count of sMEN indicated finer niche partitioning, the elevated modularity of fMEN revealed stronger compartmentalization of biogeochemical processes (Okazaki and Nakano 2016, Shi et al. 2023), mirroring the redox-stratified environment.

The taxonomic composition of network connectors showed striking divergence between water layers, corresponding to their distinct environmental regimes. In the sMEN, hypolimnetic connectors not only were more numerous but also exhibited higher average connectivity. Epilimnetic networks were dominated by Bacteroidetes as key connectors, consistent with their established role in degrading phytoplankton-derived polysaccharides (Unfried et al. 2018) and matching our observed elevated Chl a concentrations (1.11 μg·l⁻¹; Table S1) that indicate active processing of labile dissolved organic carbon (McKee et al. 2021). In contrast, the hypolimnetic sMEN featured Cyanobacteria as the most connected taxa (degree = 19), reflecting their evolutionary adaptations to cold, low-light conditions through specialized pigment systems and dual metabolic capacity for both oxygenic photosynthesis and nitrogen fixation (Forchhammer and Selim 2020, Nwosu et al. 2021). Acidobacteria further exemplified hypolimnetic specialization, with their connector status underscoring exceptional metabolic plasticity under oxygen-limited conditions (Trojan et al. 2024). Functional network analysis uncovered even more pronounced stratification, with the Chloroflexi-associated amyA gene emerging as a distinct module hub in the hypolimnetic fMEN. This starch-degrading enzyme occupied a central network position within a densely interconnected module, coinciding with significant enrichment of Chloroflexi in the hypolimnion. This pattern provides clear evidence of their niche specialization in oxygen-depleted zones (Tran et al. 2021). The ecological relevance of this pattern is reinforced by the documented versatility of Chloroflexi in redox-sensitive processes, such as sulfur oxidation and nitrate reduction (Mehrshad et al. 2018), positioning the amyA gene as a critical coordinator of biogeochemical cycling in stratified lake ecosystems.

Collectively, the greater number of nodes, links, and higher average connectivity in the hypolimnetic sMEN and fMEN (Table S5) indicate intensified intertaxa and intergene associations under stable, oxygen-depleted conditions. Such dense connectivity suggests that hypolimnetic microorganisms rely on metabolic interdependence to sustain energy flow when electron acceptors and labile substrates are scarce. In this context, the Chloroflexi-associated amyA gene and connector taxa such as Cyanobacteria and Acidobacteria likely act as metabolic linkers, coupling carbon and nitrogen transformations across modules (Faust and Raes 2012, Ma et al. 2020). These nodes help to maintain biogeochemical continuity within a redox-stratified system by bridging otherwise weakly connected modules (Peura et al. 2015, Deutschmann et al. 2023). This network architecture reflects a functional strategy emphasizing cooperation and resource sharing, thereby enhancing system stability and resilience in the hypolimnion.

Conclusion

This study reveals that thermal stratification in Lake Fuxian creates distinct ecological niches for microbial communities, with pronounced differences between epilimnetic and hypolimnetic assemblages. Using an integrated molecular approach, we observed: (1) higher microbial taxonomic richness but lower functional gene richness in the hypolimnion, dominated by specialized taxa (Nitrospirae, Parcubacteria, and Thaumarchaeota); (2) greater functional potential but reduced β-diversity in the epilimnion, suggesting metabolic flexibility; and (3) more complex, modular interaction networks in the hypolimnion, featuring Chloroflexi’s amyA as a functional hub. These patterns were primarily driven by dissolved oxygen and nutrient gradients, highlighting how redox conditions shape microbial ecology in stratified lakes. Our findings provide new insights into microbial adaptation to vertical environmental gradients, with implications for understanding biogeochemical cycling in freshwater ecosystems under climate change. Future studies should examine whether these patterns persist across different lake types and stratification regimes.

Author contributions

Jianing Ding (Conceptualization, Formal Analysis, Methodology, Visualization, Writing – original draft), Chunyan Yu (Conceptualization, Formal Analysis, Methodology, Visualization), Jiawei Gao (Formal Analysis, Methodology), Wenlei Luo (Conceptualization, Methodology, Resources), Yuanyuan Yang (Investigation, Methodology), Huabing Li (Conceptualization, Methodology, Supervision, Writing – review & editing), Qinglong L Wu (Conceptualization, Methodology, Supervision, Writing – review & editing).

Conflict of interest

The authors declare no conflict of interest.

Funding

This work was supported through the National Science Foundation of China (grant numbers 32171535 and 32471626) and the Science and Technology Planning Project of NIGLAS (grant number NIGLAS2022GS02).

Data availability

The GeoChip 5.0 data has been deposited in the National Omics Data Encyclopedia (NODE) Sequence Read Archive (https://www.biosino.org/node/) under the accession number OEP00006546.