On historic masonry and plaster, moisture-driven salt crystallization cycles impose mechanical stress and create niches for halophilic and halotolerant microbial communities. Halotolerant/halophilic microorganisms isolated from these man-made heritage environments are usually not as well characterized as those isolated from natural environments and represent a genetic and biotechnological potential that has not been thoroughly studied to date.
This study provides genome-resolved insights into five selected halophilic and halotolerant microorganisms isolated from two salt-weathered heritage sites in Austria: the St. Virgil Chapel beneath St. Stephen’s Cathedral (13th century) and the Charterhouse Mauerbach (14th century). Isolates displaying coloration under 10–20 % NaCl were sequenced using Oxford Nanopore long-read technology, yielding complete genomes and plasmids. Functional annotation focused on metabolic, osmoregulatory and pigment biosynthesis pathways to elucidate adaptive strategies underpinning their persistence under extreme salinity. Comparative genomic analyses revealed variations between the isolated strains with their respective references, as well as species-specific traits that have not been described in detail before and confirmed the presence of robust carotenoid pathways including bacterioruberin synthesis in Halococcus, mixed C40/C50 carotenoids in Nesterenkonia, and C30/C40 carotenoids in Halobacillus. Furthermore, two isolates, Marinobacter sp. 119-V2 and Modicisalibacter sp. 110-V3, likely represent novel taxa, indicating that salt-weathered heritage sites represent a specific environmental niche that selects for specific microbial colonizers. By integrating cultivation, phenotypic characterization, and genome-resolved analysis, this work advances beyond descriptive community surveys toward mechanistic understanding of microbial functions relevant to conservation science. Genome-informed insights into pigment biosynthesis and osmotic stress response help predict microbial behavior under environmental or treatment-induced salinity fluctuations, supporting the development of targeted, scientifically grounded preservation approaches for salt-affected cultural heritage.

The human gut microbiome represents a vast untapped reservoir of microbial functional diversity, including novel enzymatic functionality. Such functions can be screened by leveraging the large quantities of existing gut metagenome sequencing data to characterize enzyme diversity. We applied this approach to mine metagenomes for laccase enzymes relevant for microbial biotransformation of fluorinated compounds with the aim of understanding potential effects of these biotransformations on human health. Through a multi-study analysis of 1578 human gut metagenomes with the MOSHPIT metagenomics distribution, we found that laccase-coding genes are widely distributed in the human gut microbiome. We identified a significant association between both the abundance of laccase-coding genes and the phylogeny of laccase amino acid sequences with the degree of urbanization along a gut microbiome gradient from hunter-gatherer societies to highly industrialized urban populations. In an experimental follow-up, we heterologously expressed and tested gut bacterial laccases with a chemical mediator system as well as fluorinated pharmaceuticals and agrochemicals to validate the predicted activity, demonstrating successful depletion of cyflumetofen, fluazinam, and bisphenol AF by a laccase. By linking global gut metagenomes with activity assays, this work demonstrates the potential of mining public metagenome databases for novel enzymes with biotechnological and therapeutic applications.

Deciphering microbial ecology requires linking community composition to genome-resolved functional potential across diverse environmental contexts. Yet, limited availability of genomes from underexplored environments and high cost of environmental genome recovery often impede researchers from contextualising taxonomic profiles with the functional potential of detected taxa.

To bridge ecological distribution and genomic context, we have previously introduced the mOTUs database: a large-scale genome collection comprising 124,295 species, represented by 3.7 million genomes, 75% of which are systematically reconstructed metagenome-assembled genomes (MAGs) from over 50 habitats.

Here we present mOTUs 4.1, which adds 150,815 MAGs from underexplored environments including subsurface systems, wetlands, and hot springs. The updated web interface supports interactive sample and taxon abundance exploration on a map and introduces a new hierarchical environmental ontology, enabling customizable sample grouping and meta-analysis. Furthermore, each genome is now enriched with functional annotations from KEGG, Pfam, and eggNOG. The accompanying mOTUs command-line tool enables shotgun metagenome profiling, classification of user genomes into species-level units, and selective retrieval of genomes based on identifiers, taxonomic labels, and functional annotations, enabling their integration into custom pipelines.

Taken together, mOTUs 4.1 provides a comprehensive foundation for advancing microbial ecology by enabling the systematic integration of taxonomic composition, functional potential, and environmental distribution within a unified analytical framework.

The analysis and comparison of genomes can be complex, as it requires integrating and visualizing the results from various tools specialized for tasks such as orthology inference, phylogenetic inference, and structural and functional annotation. To simplify this process, we have developed zDB, an easy-to-use comparative genomics software integrating an analysis pipeline and a graphical user interface (GUI).

Starting from annotated Genbank files, zDB identifies orthologs and infers a phylogeny for each orthogroup as well as a species phylogeny constructed from shared single-copy orthologs. Pfam, COG, KEGG, antimicrobial resistance and virulence factor annotations can also be added to the genes, while homologs can be identified in the Swissprot and RefSeq databases. Finally, genomic islands can be predicted and their conservation evaluated across genomes.

This wealth of information can then be analysed in the GUI, which allows to search for specific genes or annotations, to perform BLAST queries, to compare genomic regions or whole genomes, to display the presence/absence of orthologs and annotations in subsets of genomes in various ways (heatmaps, presence/absence tables, Venn diagrams, phylogenies, etc.), and to compare metabolic capacities between genomes.

zDB is open-source (https://github.com/metagenlab/zDB), easy to install with conda and provides a command-line interface allowing to easily download the necessary reference databases, run the analysis pipeline and start the GUI. Overall zDB is easy to use while offering a wide range of features, making it useful both for bioinformaticians and researchers more accustomed to laboratory research. A demo of the GUI is available at https://zdb.metagenlab.ch/.

Chemical pollution is recognized as a driver of biodiversity loss. Gene - environment (GxE) interactions for humans are increasingly well-characterized using predictive approaches, but comparable methods remain underdeveloped for non-human species. Here, we demonstrate how a computational framework previously developed for human health characterization of GxE interactions in chemical-induced diseases can be restructured and expanded to predict GxE interactions in ecologically-relevant species using Mus musculus as a proof-of-concept. We integrated publicly available datasets (e.g., Chemical - Gene interactions from the Comparative Toxicogenomics Database, pathway data from REACTOME) with two sets of mouse genetic variants: one dataset from the literature for wild mice (>50,000 variants, representing wild variability, but for a subset of genome) and one dataset from lab mice (>15,000,000 variants, covering the whole genome, but not necessarily variability in wild populations). Through this, we built predictions of Chemical - Pathway - Gene - Variant - Phenotype associations in mice that may describe the toxicity mechanisms underlying GxE interactions for chemical pollution. By analyzing these novel linkages for three sets of common environmental contaminants (pesticides, cosmetics, and pharmaceuticals), we predict pathways and genes implicated across different chemicals and highlight genes and pathways with the highest variability, suggesting that these may represent important mechanisms underlying susceptibility or adaptation to chemical pollution in the wild. The genes and pathways we highlight can serve as a starting point to characterize GxE interactions in wild mouse populations and our framework can be expanded to other ecologically-relevant species to characterize chemical impacts on biodiversity.

Production and access to genomic data is critical for biological research and biodiversity management. However, the rapid growth of genomics data poses challenges for data management, including fragmentation, limited accessibility and insufficient metadata integration. These issues hinder reproducibility, reuse and compliance with Open Sciences and FAIR (Findable, Accessible, Interoperable, Reusable) principles, particularly in distributed and collaborative research environments involving multiple teams and stakeholders.

Here, we introduce two complementary technical solutions developed by Science for Life Laboratory (SciLifeLab), Sweden's national research infrastructure for life sciences, to address these challenges. First, the Swedish Reference Genome Portal provides an open-access, user-friendly platform for visualizing and disseminating genomes and genomic annotations generated by researchers affiliated with Swedish institutions. By lowering barriers to genomic data access and exploration, the portal enhances discoverability and long-term usability of genomic resources, thereby supporting open science and collaboration.

Second, DivBase (Genomic diversity database) addresses the management of large population genomics datasets and their associated metadata. In collaborative environments, such datasets are often dispersed and poorly linked across teams, increasing risks of data loss and limiting reuse. DivBase serves as a working repository where VCF files and sample metadata are securely stored, shared and queriable among collaborators, until final deposition in a discipline-specific repository. It provides a centralized system with versioning, metadata integration and query capabilities, facilitating collaborative analysis of genetic diversity and downstream submission to long-term archives.

Together, these solutions exemplify how national research infrastructures can operationalize FAIR principles and enable more efficient, transparent, and reproducible genomics research.

Identifying unknown biological stains at crime scenes is critical in forensic science, as body fluid origin significantly influences legal outcomes. While microbial profiles are promising biomarkers, existing machine learning models often suffer from study-specific biases. To address this, we developed an extreme gradient boosting classifier (XGBC) using the Microbiome Forensics Database (MFDB). We filtered MFDB to retain sites represented by more than three distinct studies and discarded OTUs (97% clustering) singletons. The training set comprised ~125k profiles across ten body sites: feces, saliva, vagina, skin, blood, urine, breast milk, nostril swabs, amniotic liquid, and semen.
The model utilizes ~8k features based on relative abundances across all taxonomic ranks. Using an 80% training and 20% test split, the model yielded an overall F1-score of 0.97. To ensure robustness and prevent study data leakage, we performed stratified k-fold (k=3) cross-validation using study IDs as groups. While performance decreased for several sites, predictions for feces, saliva, and vagina remained highly accurate (average F1-score 0.93; SD=0.01). This was further confirmed on an independent validation set of 660 samples (feces, saliva, vagina, urine) sequenced internally at CHUV, where the model achieved F1-scores of 0.85 (feces), 0.99 (saliva), and 0.92 (vagina).
Our results demonstrate that large-scale machine learning can confidently identify saliva, feces, and vaginal samples across studies with diverse amplicon target regions and sequencing technologies. This model provides a robust tool for forensic investigators, though further research is required to validate performance on body site mixtures.

Respiratory Syncytial Virus (RSV) is responsible for considerable global health burden. The recent introduction of novel immunoprophylactic interventions highlights the need for population-level RSV genomic surveillance to monitor circulating lineages and detect mutations that may compromise intervention effectiveness.
In this study, we implemented wastewater-based RSV surveillance by integrating both pan- and subtype-specific digital PCR and amplicon-based sequencing within Switzerland's national wastewater monitoring program. We sequenced 328 wastewater samples, covering the RSV peak season from November 2024 to May 2025.
Viral loads from digital PCR showed co-circulation of RSV-A and B across all locations, with site-specific differences in subtype dominance. Subtype-specific effective reproduction numbers estimated from wastewater viral loads via Bayesian generative model implemented in EpiSewer indicated similar epidemiological trajectories for both subtypes.
Genomic sequencing data were processed using the viral NGS computational analysis software V-pipe. Lineage abundances in wastewater samples were reconstructed based on signature mutation frequencies using Lollipop, a deconvolution tool leveraging time-series information via a kernel smoothing approach. We found that for RSV-A, the A.D.1 and A.D.3 lineages predominated, while B.D.E.1 was dominant RSV-B lineage. We quantified genetic diversity using mutation richness and nucleotide diversity metrics and assessed differences across subtypes and genes using generalized linear mixed effects models. RSV-A exhibited higher genetic diversity than RSV-B, with greater diversity in G compared to the F gene, mirroring diversity patterns observed in clinical data.
Our findings demonstrate the potential of wastewater-based RSV surveillance for continuous tracking of viral dynamics in the context of widespread introduction of new immunoprophylaxis products.

Over the past two decades, metagenomic sequencing has enabled the exploration of vast and novel regions of the protein universe, revealing an immense diversity of sequences from a wide range of environments. However, most proteins collected from metagenomic experiments have low to no similarity to proteins with experimentally characterized function;
Even the most extensively characterized biome, the human gut microbiome (HGM), resists full-scale functional annotation. HGM-derived sequence catalogs have identified, with homology methods, over 170 million putative proteins, but nearly 40% of these remain functionally uncharacterized, representing a large reservoir of biological ""dark matter"". Current functional annotation methods are even less effective for other biomes.
Compared to structural alignment and deep learning, genome-context-based methods offer a complementary approach to curating the functional annotation of metagenomic data. Nevertheless, there is a lack of tools to gather and organise these data on a large scale in a customisable way.
Here, we introduce metaGCsnap, a tool that leverages comparative genomics by analyzing large repositories of isolated and metagenomic assemblies, allowing to elucidate the functions of uncharacterized proteins. Unlike current methods, metaGCsnap's native integration with MGnify enables it to capture protein occurrence by linking sequences with environmental samples, paving the way for the investigation of sequence diversity and environmental coupling.
By enabling the systematic exploration of uncharted protein space, this study aims to accelerate functional discovery by amplifying scientist capabilities to interrogate their genomic data.

Rivers are dynamic ecosystems that contain diverse biological material, including viruses, yet the dynamics of river viral communities remain poorly understood. Given the complexity of aquatic metagenomic datasets, efficient computational approaches are required to resolve these gaps. Here, we combined high-throughput sequencing with a reproducible bioinformatic pipeline to characterize viral communities in a river in central Slovenia influenced by wastewater effluent inputs.
We collected samples of treated effluent from a wastewater treatment plant for 7 months across 2 years. Concurrently, we sampled the effluent-receiving river at three locations relative to the effluent discharge: upstream, immediately downstream and several kilometres downstream. Following the concentration and sequencing, viral communities of 55 samples were analysed through a fully automated custom pipeline implemented in Snakemake.
The pipeline consists of three modules: (1) initial processing of sequencing data and taxonomic classification of assembled sequences, (2) cross-sample clustering of viral contigs into viral operational taxonomic units (vOTUs) using a network-based clustering algorithm, and (3) per-sample quantification of viral abundance. The developed clustering strategy represents an innovative approach, leveraging sequence overlaps across samples to improve viral genome reconstruction.
Using this workflow, we identified 360.509 unique vOTUs, representing known and putative novel viral species. We observed clear spatial structuring of river water viral communities linked to wastewater effluent inputs. We further detected seasonal patterns in viral community dynamics and showed that episodic environmental perturbations can temporarily influence river viral communities. Our findings show that river viral communities are highly dynamic and influenced by various environmental and anthropogenic factors.

Identifying the animal host of Salmonella Typhimurium (STm), a zoonotic pathogen of significant public health concern, is a key challenge in outbreak investigations. Machine learning approaches have been used to successfully predict animal hosts of STm. However, these models provide no formal guarantees of their predictive performance and have previously shown degraded performance on phylogenetically distinct isolates, leading to silently overconfident model predictions. Here, we use conformal prediction to quantify uncertainty and provide statistical guarantees for model predictions of the animal host of STm isolates. Using a dataset of 1010 non-clonal isolates, we found that a random forest model trained on accessory genes enabled accurate prediction of swine, finch, duck, and pigeon STm isolates (F1-score > 0.75); lower F1-scores were observed for cattle, chicken, dog, horse and sheep isolates. Conformal prediction achieved an empirical coverage rate of 97.2%, exceeding the 95% nominal coverage. The host-conditional empirical coverage rate exceeded the nominal 95% coverage except for sheep (66%), likely reflecting low sample size or genomic overlap between animal hosts. Conformal prediction sets were consistent with STm epidemiology; for example, dogs, whose isolates are typically of a broad host range variant of STm, were included within most prediction sets. We show that conformal prediction enables uncertainty quantification and provides robust statistical guarantees for model predictions of the animal hosts of STm isolates. Conformal prediction could enable public health agencies to better utilise machine learning models to identify the animal host of STm isolates in routine surveillance and outbreak investigations.

The human gut microbiome plays a central role in host homeostasis, but its contribution to cancer remains under-characterized at fine taxonomic resolution. While previous studies have focused primarily on species-level associations, few studies are now investigating untargeted strain signals in the gut microbiome of patients with colorectal cancer.
Here, we performed an untargeted strain-level meta-analysis of metagenomic sequencing data from seven different datasets of fecal samples of patients with haematological cancer (different types of Lymphoma and Leukemia) and matched healthy controls. We used a combination of marker-based reference approach and assembly-based approach (i.e., to generate metagenome assembled genomes (MAGs)) to profile microbial communities at the species, strain, and functional gene levels.
We generated a catalogue of 14,647 medium to high quality cancer associated MAGs. Significant reductions in alpha-diversity and distinct beta-diversity were observed in both cancer types compared to control. Several pathobionts, especially Enterocloster boltae and Clostridium innocuum were strongly and consistently enriched in both Leukemia and Lymphoma, while Escherichia coli was only enriched in Lymphoma. Certain species that were not differentially abundant at the species level displayed strong cancer-specific strain-level signals, especially abundant members of the gut microbiome such as Bacteroides spp., possibly related to a strain-level adaptation to cancer-related dysbiosis.
Our findings highlight the importance of moving beyond species-level resolution to detect strain-specific signals that would otherwise remain hidden and underscores the potential for strain-resolution to inform cancer diagnostics and therapeutic strategies.

Changes in gene content through gain and loss play a key role in the adaptation and diversification of species. Accordingly, our ability to detect and accurately document the history of these changes is important for our understanding the evolutionary trajectories of life on Earth. Here we present GLADE, a tool that accurately reconstructs gene gains, losses, and duplications for a set of species under consideration and uses this information to infer ancestral gene contents for every speciation event in the species tree. GLADE requires as input only a standard OrthoFinder results directory, and outputs the full evolutionary history of every orthogroup, including branch-specific changes and reconstructed ancestral genomes. We benchmark GLADE using both real and simulated data and show that GLADE accurately identifies orthogroup gains, losses, and duplications, and reconstructs ancestral orthogroup sizes with higher precision and overall accuracy than any competitor method. To illustrate the utility of the method, we apply GLADE to a dataset of 78 mammalian genomes and uncover repeated contractions in orthogroups associated with tooth formation on branches leading to ant- and termite-eating mammals - revealing convergent genomic signatures underlying this dietary specialization. GLADE and accompanying documentation and tutorials are freely available at https://github.com/lauriebelch/GLADE/.

Plant-associated microbes play a crucial role in plant growth, enhancing nutrient uptake and disease resistance. However, modern crop varieties, bred for high-input agriculture, often show reduced root colonization by beneficial soil microbes. The international research consortium Wheat Alliance seeks to identify genetic factors in wheat that influence the recruitment of these beneficial microbes. With up to 1100 wheat-related genotypes available, we screen plant growth under nitrogen- and phosphate-limited conditions. Our approach integrates high-throughput phenotyping of shoot traits with root metagenome sequencing. Initial screening of 11 wheat genotypes revealed clearly distinguishable bacterial root communities (PERMANOVA: R2 = 0.44, adjusted p = 0.001). LASSO with five-fold cross-validation identified 17 bacterial genera that robustly discriminate wheat genotype, achieving a mean classification accuracy of 84.5 ± 3.9% (mean ± SE across folds). These are promising taxa to trace in larger screenings investigating plant growth-associated microbiota. One-hundred wheat genotypes have been screened in greenhouses, and root samples are currently being processed for sequencing. The plant microbiome and shoot phenotyping data will be used to train machine learning models, enabling us to uncover complex interactions between plant traits, microbial composition, and nutrient acquisition. A genome-wide association study (GWAS) will then pinpoint genetic drivers shaping the root microbiome, providing new avenues for microbiome-based breeding and more sustainable wheat cultivation.

Metagenomic sequencing enables cultivation-free analysis of microbial communities by capturing the total DNA content of environmental samples. In wastewater, this approach provides a powerful tool for monitoring pathogen spread and antimicrobial resistance genes relevant to public health. Here, we compared metagenomic data with genomes from cultured isolates of carbapenemase-producing Klebsiella pneumoniae (CPKP), a pathogen of global health concern according to the WHO.

A total of 12 metagenomic samples from two wastewater treatment plants in Germany were sequenced with Illumina and 56 CPKP isolates from the same samples with Illumina and Oxford Nanopore Technologies (ONT). Hybrid assemblies of the isolate samples yielded high-quality genome reconstructions, which were then analyzed alongside metagenomes using comprehensive antimicrobial resistance gene screening methods.

Genomic analysis of the isolates revealed diverse sequence types and resistance patterns. Although ONT data improved assembly contiguity, it failed to capture several small plasmids and antimicrobial resistance genes, indicating that a combined sequencing approach might be advantageous for a comprehensive genomic characterization. In contrast, metagenomic sequencing results revealed only 0.03% CPKP reads, which were insufficient for genome recovery. In addition, approximately 50% of metagenomic reads remained unclassified, which indicates high dataset complexity. While cultivation-based sequencing enabled detailed genomic profiling, metagenomics provided broader insights into microbial community composition and the resistome, including resident environmental species.

This study highlights the complementary strengths and limitations of cultivation-based and metagenomic approaches for pathogen and antimicrobial resistance gene surveillance in wastewater.

Latitudinal diversity gradients (LDGs), typically declining from equator to poles, are a pervasive macroecological pattern, yet their generality and drivers in the ocean microbiome remain widely unresolved. We integrated global-scale metagenomic data with habitat modeling to study marine microbial LDGs across seasons and depths. Surface mixed layer microbiomes exhibited diversity peaks at (sub)tropical latitudes and a poleward decline, whereas mesopelagic communities (200–1,000 m) showed no latitudinal diversity structuring. Taxonomic resolution revealed that the mixed layer LDG was underpinned by Alphaproteobacteria and Cyanobacteriia, while other taxa exhibited distinct or contrasting LDGs. Diversity structuring also varied by seasons and regions, governed by temperature and nutrient availability. Together, these findings highlight that within the ocean microbiome, LDGs are not universal, but lineage-specific ecological strategies and responses to environmental gradients. Our study provides fundamental insights into the structuring of ocean microbiome diversity and lays the foundation for predicting responses to environmental change.