Environmental Metagenomics: Characterizing Microbial Communities in Soil, Water, and Sediment Samples

A teaspoon of soil holds more microbial life than there are people on Earth. Thousands of bacterial and archaeal species, most of which have never been grown in a laboratory, carry out the biogeochemical reactions that sustain every ecosystem on the planet. They fix nitrogen, mineralize carbon, oxidize methane, and detoxify pollutants. Until recently, studying these communities meant either culturing them — which captures perhaps 1% of species — or amplifying marker genes like 16S rRNA — the target of 16S/18S/ITS Amplicon Sequencing — which tells you who is there but not what they are doing.

Shotgun metagenomic sequencing changed this. By sequencing all the DNA in an environmental sample, you can reconstruct near-complete genomes from uncultured organisms, quantify the abundance of every gene in a metabolic pathway, and track how microbial communities shift in response to drought, pollution, or remediation. An environmental metagenomics project is not fundamentally different from a human gut microbiome project in its bioinformatic pipeline, but the sample-level challenges — humic acid-laden soil, low-biomass groundwater, physically refractory sediment — demand a different set of upstream decisions.

This guide walks through those decisions. It covers sample collection and preservation for soil, water, and sediment; DNA extraction strategies for inhibitor-rich matrices; assembly and binning workflows that recover high-quality metagenome-assembled genomes; and functional annotation that turns gene catalogs into biogeochemical insight. It also addresses two applications — bioremediation monitoring and biosurveillance — where environmental metagenomics is moving from academic research into regulatory and industrial practice.

Figure 1: Environmental metagenomics workflow — from field sampling to functional insight across soil, water, and sediment matrices.

Sample Collection and Preservation: Getting the Matrix Right

The most expensive mistake in environmental metagenomics happens before a sequencer is turned on. A sample collected without metadata, or preserved in a way that lets the community shift during transport, produces data that no amount of bioinformatic sophistication can rescue.

Soil is the most complex matrix. A standard approach uses a 2.5-cm-diameter corer to a depth of 10 to 15 cm for surface soils, with deeper intervals for subsurface studies. Sieve to 2 mm in the field to remove roots and stones, then immediately freeze on dry ice or liquid nitrogen. If freezing is not possible, DNA/RNA Shield or lifeGuard Soil Preservation Solution provides room-temperature stability for up to several days, though flash freezing remains the gold standard. For each sampling point, record GPS coordinates, soil type, pH, total organic carbon, moisture content, and temperature. These are not optional extras. A 2022 analysis of topsoil metagenomes across 189 sites worldwide showed that soil pH alone explains more variation in microbial community composition than any other single variable, and that failing to adjust for pH in a differential abundance analysis generates spurious associations between taxa and treatment conditions (1). This principle of covariate-aware analysis extends across microbiome domains — a population-scale study of over 8,000 individuals demonstrated that environmental and host factors together account for substantial gut microbiome variation, reinforcing that comprehensive metadata collection and covariate adjustment are essential methodological practices regardless of the sample type under study (10).

Water samples require filtration. Pass 1 to 10 liters through a 0.22-μm polyethersulfone or polycarbonate membrane, depending on the turbidity of the source. For oligotrophic groundwater or open-ocean samples, larger volumes are necessary to capture sufficient biomass. Record temperature, pH, conductivity, dissolved oxygen, and turbidity at the time of collection. Freeze the filter immediately. For samples with high suspended sediment, pre-filter through a larger pore size to prevent the 0.22-μm membrane from clogging, but be aware that the pre-filter retains particle-associated microbes and the final filtrate represents only the free-living fraction. Decide deliberately which fraction matters for your question.

Sediment sampling uses a grab sampler or corer, depending on the study design. For cores, section at defined depth intervals — typically 1 to 2 cm for fine-scale redox gradient studies, or broader intervals for reconnaissance work. Record redox potential if the study concerns biogeochemical cycling across oxic-anoxic transitions. Sediment samples are typically high in humic substances and require specific extraction adjustments.

Figure 2: Comparative illustration showing soil coring, water filtration through 0.22-μm membrane, and sediment grab sampling methods, with key metadata parameters for each matrix.

DNA Extraction: Fighting Humic Acids and Winning

If the sample collection step is where environmental metagenomics projects succeed or fail silently, the DNA extraction step is where they succeed or fail loudly — with low yields, failed library preparations, and sequencing runs dominated by inhibitor carryover.

Humic acids are the primary antagonist. These complex organic molecules, abundant in soil and sediment, co-extract with DNA and inhibit downstream enzymatic reactions including the tagmentation and amplification steps in Illumina library preparation. A DNA extract that looks clean by NanoDrop — with a respectable 260/280 ratio above 1.8 — can still fail library preparation because humic substances absorb at 230 nm, depressing the 260/230 ratio. A 260/230 ratio below 1.5 is a warning sign; below 1.0, the extract nearly always requires cleanup.

Several strategies reduce humic acid carryover. Commercial kits designed for soil — the DNeasy PowerSoil Pro Kit and the FastDNA Spin Kit for Soil are the most widely adopted — include humic acid-binding steps. For particularly inhibitor-rich samples, including peat, compost, and clay-rich sediment, supplement with a post-extraction cleanup using CTAB, PVPP spin columns, or a commercial inhibitor removal kit. A 2025 comparison of extraction protocols across eight soil types found that bead-beating with 0.1-mm or mixed-size beads improved DNA yield from Gram-positive bacteria and archaea by roughly 2- to 4-fold compared to enzymatic lysis alone, and that the addition of a CTAB cleanup step reduced the inhibitor load by approximately 70% in high-organic-matter soils (2).

Lysis efficiency across the tree of life is a separate concern. Gram-positive bacteria, with their thick peptidoglycan walls, require aggressive mechanical lysis. Archaea, depending on the species, range from easily lysed to nearly refractory. Fungal spores and thick-walled protozoan cysts resist gentle extraction. A bead-beating step with 0.1-mm silica-zirconia beads for at least 6 minutes, preferably with a FastPrep or similar high-energy homogenizer, is the pragmatic compromise — it recovers most bacterial and archaeal DNA while shearing it to a size range compatible with short-read library preparation. For long-read metagenomics on the Oxford Nanopore or PacBio platforms — an application CD Genomics supports through its Long-Read Metagenomic Sequencing service — gentler extraction with wide-bore pipette tips and minimal vortexing preserves high-molecular-weight DNA, at the cost of some lysis efficiency for hard-to-lyse taxa.

Figure 3: A four-panel comparison showing humic acid contamination effects — clean vs. brownish extract, 260/230 ratio warning zones, CTAB cleanup workflow, and bead-beating lysis efficiency across microbial groups.

From Reads to Genomes: Assembly, Binning, and MAG Quality

Once sequencing data pass quality control and host-associated reads are removed — relevant for rhizosphere samples where plant DNA can dominate — the central computational challenge is reconstructing genomes from a mixed community without the benefit of a reference.

The first decision is assembly strategy. Co-assembly pools reads from all samples in a study into a single assembly, maximizing depth for rare species at the cost of mixing strain-level variants across samples. Per-sample assembly preserves sample-specific genomic features, including strain-level differences that co-assembly collapses. For a study comparing microbial communities across a contamination gradient, the recommended approach is per-sample assembly followed by aggregation and dereplication. MetaSPAdes remains the most widely used metagenomic assembler for short-read data; MEGAHIT is an alternative with lower memory requirements and comparable performance for less complex communities (3).

Binning groups assembled contigs into draft genomes based on tetranucleotide frequency and coverage patterns across samples. MetaBAT 2, CONCOCT, and VAMB are the most common tools, and using at least two and selecting bins supported by both improves precision. Semi-supervised binning tools like SemiBin, which incorporate taxonomic information from marker genes, represent a 2023-2025 advance that improves bin completeness and reduces contamination for difficult-to-resolve lineages.

Binning produces a collection of metagenome-assembled genomes — MAGs. Not all MAGs are created equal, and the field now applies standardized quality thresholds. CheckM2 estimates genome completeness and contamination using lineage-specific marker genes. A MAG with greater than 90% completeness and less than 5% contamination is classified as high quality; a MAG with greater than 50% completeness and less than 10% contamination is medium quality. For studies aiming to describe novel phyla or classes, only high-quality MAGs should carry the taxonomic claim (4).

Dereplication removes redundant genomes from the set using dRep, which clusters MAGs at a specified average nucleotide identity — typically 95% or 99% for species-level clustering. A dereplicated set of MAGs represents the non-redundant genomic diversity captured by the study and is the basis for downstream functional annotation and comparative genomics. Taxonomic classification of MAGs using tools such as GTDB-tk places each genome on a standardized reference tree. Database and parameter choices strongly influence classification performance, and consistent tool-database combinations should be applied across all samples in a study to ensure comparability (9).

CD Genomics' Metagenomic Shotgun Sequencing service supports assembly and binning for environmental samples, with MAG quality reported using CheckM2 and GTDB-tk for taxonomic classification.

Figure 4: Assembly and binning workflow diagram showing co-assembly vs. per-sample assembly, MetaBAT 2 / CONCOCT / SemiBin binning, CheckM2 quality assessment, and dRep dereplication to non-redundant MAG set.

Functional Annotation: What the Community Can Do

A MAG catalog tells you which organisms are present. Functional annotation tells you what those organisms are capable of, and it is here that environmental metagenomics directly addresses ecological questions.

The standard annotation pipeline maps predicted protein-coding genes against reference databases of protein families, metabolic pathways, and functional modules. eggNOG-mapper, run against the eggNOG database of orthologous groups, provides broad functional classification covering roughly 90% of prokaryotic protein families. DRAM — the Distilled and Refined Annotation of Metabolism tool — is purpose-built for metagenome-derived genomes and distills the output of multiple annotation databases into a metabolism-focused summary that highlights which pathways a genome encodes and, critically, which key steps are missing. A genome that carries all eight enzymes of the denitrification pathway has a different ecological interpretation than one that stops at nitrite reduction.

For environmental studies, the most informative functional categories are those linked to biogeochemical cycling. These are not generic KEGG level-2 pathways but specific, enzymatically defined processes:

Nitrogen cycling genes are the most extensively cataloged and functionally validated. Denitrification — the stepwise reduction of nitrate to N₂ gas — is tracked through four key genes: narG/napA for nitrate reduction, nirK or nirS for nitrite reduction, norB for nitric oxide reduction, and nosZ for nitrous oxide reduction. A community with abundant nirK and nirS genes but sparse nosZ is a potential N₂O emission hotspot. Nitrification — the oxidation of ammonia to nitrite and then nitrate — is tracked through amoA, which encodes a subunit of ammonia monooxygenase found in both bacterial and archaeal nitrifiers. A 2024 quantitative review of meta-analyses confirmed that ammonia-oxidizing archaea (AOA) and bacteria (AOB) amoA genes respond predictably to nitrogen inputs and environmental variables, making them robust functional markers for nitrification capacity in metagenomic studies (5). Nitrogen fixation is tracked through nifH, encoding the nitrogenase reductase subunit, which is the most widely used functional marker gene in microbial ecology.

Carbon cycling is tracked through carbohydrate-active enzyme families. CAZy classifies enzymes into glycoside hydrolases, glycosyltransferases, polysaccharide lyases, carbohydrate esterases, and auxiliary activities — each of which targets specific bonds in specific polysaccharides. A soil community with abundant GH6 and GH7 cellulases has a different carbon-degradation profile from one dominated by GH11 xylanases. For methane cycling, the pmoA gene encodes particulate methane monooxygenase for aerobic methanotrophy, while mcrA encodes the methyl-coenzyme M reductase that catalyzes the final step of methanogenesis — the two genes that define the methane cycle.

Sulfur cycling includes dsrA and dsrB for dissimilatory sulfate reduction, soxB for sulfur oxidation, and the diverse family of sulfatases for organic sulfur mineralization. Phosphorus cycling is tracked through phosphatase-encoding genes, particularly phoD and phoX for alkaline phosphatases, which are expressed under phosphate limitation.

The deliverable is a gene abundance table stratified by sample and condition, combined with a pathway presence-absence matrix for each MAG. This makes it possible to state not just which organism carries which pathway but whether that pathway is differentially abundant across experimental conditions — the functional equivalent of differential abundance testing in taxonomic profiling.

CD Genomics' Metagenomic Shotgun Sequencing service includes functional annotation against eggNOG, CAZy, and KEGG, with custom biogeochemical gene catalog reporting available for nitrogen, carbon, sulfur, and phosphorus cycling pathways.

Figure 5: Functional annotation pipeline — from predicted genes through eggNOG-mapper and DRAM to biogeochemical gene catalogs, with a nitrogen cycling example showing the eight enzymes of the denitrification pathway.

Nitrogen Cycle Reconstruction: A Worked Example

Consider a study of nitrate-contaminated groundwater in an agricultural watershed. The research question is whether the indigenous microbial community has the genetic capacity for complete denitrification — that is, whether it can reduce nitrate to harmless N₂ gas rather than stopping at nitrite or N₂O.

The experiment collects groundwater samples from ten wells along a contamination gradient. DNA is extracted, libraries are prepared, and 20 million read pairs per sample are sequenced. After quality control and assembly, MetaBAT 2 produces 340 MAGs, of which 68 pass the high-quality threshold in CheckM2. Functional annotation with DRAM and targeted screening against the NCBI nitrogen cycling gene catalog reveals that 23 of these MAGs carry nirK or nirS, indicating the capacity for nitrite reduction, and 14 carry nosZ, indicating the capacity for N₂O reduction to N₂. The nosZ-carrying MAGs are enriched in the low-nitrate wells and depleted in the high-nitrate wells, suggesting that nitrate loading suppresses the final denitrification step — a hypothesis testable with follow-up quantification of N₂O flux from the same wells.

This pattern — abundant upstream denitrification genes with sparse nosZ at high nitrate concentrations — has been observed in agricultural soils and river networks and represents one of the most actionable findings that environmental metagenomics can deliver (6).

Figure 6: A nitrogen cycle reconstruction diagram showing a groundwater contamination gradient, MAG distribution across wells, and key denitrification gene abundance heatmap (narG, nirK, nirS, norB, nosZ).

Bioremediation and Biosurveillance

Environmental metagenomics is transitioning from academic tool to regulatory and industrial asset in two domains: bioremediation monitoring and biosurveillance.

In bioremediation, the question is whether a microbial community has the genetic machinery to degrade a specific pollutant, and whether that machinery becomes active under treatment conditions. Hydrocarbon-contaminated sites — from diesel spills to crude oil pipeline releases — are the classic use case. Shotgun metagenomics identifies the abundance and taxonomic affiliation of genes encoding alkane hydroxylases, ring-cleaving dioxygenases, and other xenobiotic-degrading enzymes without requiring cultivation of the organisms that carry them. A 2024 metagenomic analysis of tungsten tailings under phytoremediation with ryegrass and clean soil amendment found that nitrogen-fixing genera including Bradyrhizobium increased significantly with planting and that metabolic pathways dominated functional gene profiles at over 71% relative abundance, demonstrating how metagenomic functional annotation can track the genetic basis of remediation processes and identify microbial taxa that drive them (7).

In biosurveillance, the question is whether a sample contains pathogens, antibiotic resistance genes, or virulence factors. Shotgun metagenomics of water treatment plant influent and effluent, agricultural runoff, or aquaculture pond sediment provides an untargeted surveillance capability that PCR-based assays cannot match because PCR only finds what you design primers for. Antibiotic resistance genes are identified by mapping reads against the CARD database — the foundation of dedicated Antibiotic Resistance Gene Analysis — and virulence factors against VFDB. The limitation, as with all DNA-based surveillance, is that gene presence does not confirm gene expression — a metagenomic detection of a beta-lactamase gene means the genetic potential exists, not that it is actively conferring resistance at the time of sampling (8).

For long-term monitoring studies, absolute abundance quantification — in which a known quantity of internal standard DNA is spiked into each sample before sequencing — converts relative abundance data into copies per gram or copies per liter. This removes the compositional bias inherent in relative abundance comparisons and enables direct comparison across timepoints and sites. CD Genomics' Absolute Metagenomic Sequencing Service provides this capability for monitoring and surveillance applications.

Figure 7: A dual-panel illustration comparing bioremediation monitoring (contamination gradient → functional gene abundance → degradation pathway completeness) and biosurveillance (sample collection → ARG/VF mapping → risk categorization).

For a broader overview of metagenomic sequencing approaches including clinical microbiome studies, viromics, and multi-omics integration, see our guide on Metagenomic Sequencing Services — Overview. For studies examining the viral fraction of environmental samples — including phage-host dynamics and viral community ecology — our Viral Metagenomic Sequencing service extends the analysis to the virome.

How CD Genomics Delivers Your Environmental Metagenomic Project

A well-executed environmental metagenomics project begins at sample collection and ends with an analyzed data package that answers a specific ecological or engineering question. The process follows a defined pipeline.

Samples arrive at the laboratory with chain-of-custody documentation and are accessioned with metadata verification. DNA is extracted using a matrix-appropriate protocol — PowerSoil Pro for soil and sediment, a filtration-optimized protocol for water samples — with quality assessed by Qubit fluorometry for concentration, NanoDrop spectrophotometry for purity, and agarose gel electrophoresis for integrity. For inhibitor-rich samples, a post-extraction cleanup step is applied and re-assayed before library preparation.

Libraries are prepared using the NEBNext Ultra II FS kit with fragmentation optimized for the target insert size, barcoded for multiplexing, and pooled for sequencing on an Illumina NovaSeq platform. Sequencing depth is typically 20 million read pairs per sample for functional profiling, or 5 to 10 million read pairs for taxonomic profiling — adjusted upward for complex soil communities with high species richness.

The Metagenomic Shotgun Sequencing bioinformatic pipeline includes quality trimming with fastp, host read removal against the appropriate reference genome for plant-associated samples, assembly with metaSPAdes or MEGAHIT, binning with MetaBAT 2 and CONCOCT, MAG quality assessment with CheckM2, taxonomic classification with GTDB-tk, and functional annotation with eggNOG-mapper, DRAM, CAZy, and custom biogeochemical gene catalogs. For nitrogen cycling studies, targeted gene screening against NCBI-curated nifH, nirK, nirS, norB, nosZ, amoA, and narG databases is included.

Deliverables include raw FASTQ files, quality control reports, assembled contigs, dereplicated MAGs with CheckM2 quality statistics, taxonomic abundance tables, functional gene and pathway abundance tables, differential abundance testing with covariate adjustment, and a comprehensive report with publication-ready figures. Turnaround for a 20-sample environmental metagenomics project is approximately six to eight weeks from sample receipt to analyzed data delivery.

For projects that require isolate-level genome sequencing of cultured organisms identified in the metagenomic survey, CD Genomics' Microbial Whole Genome Sequencing service provides genome sequencing for strains of interest. For studies that ask not just what genes are present but which ones are actively expressed under specific environmental conditions, our Metatranscriptomic Sequencing service adds the gene expression dimension. For projects that integrate metagenomic discovery with metabolomics, metatranscriptomics, or metaproteomics to build a systems-level understanding of environmental microbial communities, CD Genomics' Multi-Omics Service provides unified multi-platform data generation and integrated bioinformatic analysis.

FAQ

How many samples do I need for a robust environmental metagenomics study?

For comparative studies across sites or conditions, three to five biological replicates per group is the minimum. Environmental samples are inherently heterogeneous — one soil core from a field is not a replicate of the next. Include field blanks and extraction blanks to track contamination. For longitudinal monitoring, sample at regular intervals that capture the relevant temporal dynamics — monthly for seasonal studies, more frequently for event-driven sampling. Statistical power depends on within-group variability, which is often higher in environmental samples than in host-associated samples.

What is the difference between co-assembly and per-sample assembly?

Co-assembly combines reads from all samples into one assembly, maximizing depth for rare organisms but collapsing strain-level differences between samples. Per-sample assembly preserves sample-specific genomic features, including strains, but may recover fewer rare genomes from individual samples. For most environmental comparison studies, per-sample assembly followed by dereplication is recommended.

How do I know if my MAGs are real or assembly artifacts?

CheckM2 provides completeness and contamination estimates using lineage-specific marker genes. Apply the community standard: greater than 90% completeness and less than 5% contamination for high-quality MAGs. Additionally, check for chimerism through GC content and coverage outliers. Bins with anomalous tetranucleotide signatures or coverage patterns relative to other bins from the same sample should be flagged for manual inspection. A MAG that passes CheckM2 but has a 16S rRNA gene from a completely different phylum is likely chimeric.

Can environmental metagenomics replace 16S amplicon sequencing?

In many applications, yes — but 16S/18S/ITS Amplicon Sequencing remains more cost-effective for large-scale surveys where taxonomic profiling is the only endpoint. Shotgun metagenomics provides species-level and strain-level resolution plus functional gene content that 16S cannot deliver. Shallow shotgun metagenomics (3 to 5 million reads per sample) is an emerging middle ground, providing species-level taxonomy at approaching-16S cost for environmental samples with moderate microbial diversity.

How do I handle plant DNA contamination in rhizosphere samples?

Rhizosphere samples — soil adherent to plant roots — can contain 20 to 50% plant DNA. Align reads against the host plant reference genome using Bowtie 2 or BWA, and discard mapped reads before assembly. For non-model plants without a reference genome, computational depletion using a database of plant chloroplast and mitochondrial genomes can reduce the plant fraction, though less efficiently than a complete reference.

What metadata should I absolutely collect for each sample?

GPS coordinates, date and time of collection, matrix type, pH, temperature, and a description of recent environmental history (rainfall within 48 hours, flooding, recent fertilizer or pesticide application). For soil, add total organic carbon and moisture. For water, add conductivity, dissolved oxygen, and turbidity. For sediment, add redox potential and grain size class. These variables are required for the MIxS environmental packages and are the minimum set needed for differential abundance analysis with covariate adjustment.

Can I combine short-read and long-read metagenomics for environmental samples?

Yes. This hybrid approach is increasingly standard for complex environmental samples. Short reads provide high-accuracy taxonomic and functional profiling. Long reads (Oxford Nanopore or PacBio) resolve repetitive regions and mobile genetic elements, dramatically improving MAG contiguity. A common strategy is to sequence the same samples on both platforms, assemble with a hybrid assembler or use long reads to scaffold short-read assemblies, and bin from the combined dataset.

References:

1. Bahram M, Espenberg M, Pärn J, et al. Structure and function of the soil microbiome underlying N2O emissions from global wetlands. Nature Communications. 2022;13:1430. doi:10.1038/s41467-022-29161-3 (CC BY 4.0)： https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8927123/
2. Nurk S, Meleshko D, Korobeynikov A, Pevzner PA. metaSPAdes: a new versatile metagenomic assembler. Genome Research. 2017;27(5):824-834. doi:10.1101/gr.213959.116 (CC BY 4.0)： https://doi.org/10.1101/gr.213959.116
3. Chklovski A, Parks DH, Woodcroft BJ, Tyson GW. CheckM2: a rapid, scalable and accurate tool for assessing microbial genome quality using machine learning. Nature Methods. 2023;20(8):1203-1212. doi:10.1038/s41592-023-01940-w：https://www.biorxiv.org/content/10.1101/2022.07.11.499243v1
4. Hui D, Ray A, Kasrija L, Christian J. Impacts of climate change and agricultural practices on nitrogen processes, genes, and soil nitrous oxide emissions: a quantitative review of meta-analyses. Agriculture. 2024;14(2):240. doi:10.3390/agriculture14020240 (CC BY 4.0)：https://doi.org/10.3390/agriculture14020240
5. Wei Y, Xiao J, He J, et al. An integrated global resource of wetland microbiomes linking environmental metadata, community profiles, and genome-resolved metabolic traits. Scientific Data. 2026;13:284. doi:10.1038/s41597-026-07581-w (CC BY 4.0)：  https://pubmed.ncbi.nlm.nih.gov/42236734/
6. Zheng X, Li Q, Peng Y, Wang Z, Chen M. Phytoremediation of tungsten tailings under conditions of adding clean soil: microbiological research by metagenomic analysis. Sustainability. 2024;16(13):5715. doi:10.3390/su16135715 (CC BY 4.0)：https://doi.org/10.3390/su16135715
7. Alcock BP, Huynh W, Chalil R, et al. CARD 2023: expanded curation, support for machine learning, and resistome prediction at the Comprehensive Antibiotic Resistance Database. Nucleic Acids Research. 2023;51(D1):D690-D699. doi:10.1093/nar/gkac920 (CC BY 4.0)：https://doi.org/10.1093/nar/gkac920
8. Wright RJ, Comeau AM, Langille MGI. From defaults to databases: parameter and database choice dramatically impact the performance of metagenomic taxonomic classification tools. Microbial Genomics. 2023;9(3):000949. doi:10.1099/mgen.0.000949 (CC BY 4.0)：https://doi.org/10.1099/mgen.0.000949
9. Gacesa R, Kurilshikov A, Vich Vila A, et al. Environmental factors shaping the gut microbiome in a Dutch population. Nature. 2022;604:732-739. doi:10.1038/s41586-022-04567-7 (CC BY 4.0)：https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9048813/

For research purposes only, not intended for clinical diagnosis, treatment, or individual health assessments.