The metabolism of early microbes shaped our planet's chemistry and enabled more life to evolve. However, it remains uncertain when many forms of metabolism arose and which organisms carried them out. Here we combine machine learning, metabolic networks, and phylogenomics of microbes to date 14 forms of metabolism and 6 other traits. We find that metabolism evolved in three waves, starting >3.4 billion years ago (Ga) with fermentation and sulfur compound respiration. A second wave began ~3.2 Ga with aerobic respiration and growth, while a third began ~1.4 Ga with further specialization. We compared these predictions with independent dates from the geochemical record, and agreement with conservative dates was good (CCC = 0.81 to 0.89). Several forms of metabolism are invisible in the geochemical record, and our approach dates these forms, too. Our results reconcile different lines of evidence for microbial evolution and provide a fuller picture of early Earth.

Retrieving an enzyme for a target reaction provides a starting point for evolution, but predicting how to improve it remains difficult. This requires connecting evolutionary information with the structural and chemical context of catalysis. Here we present AlphaEnzyme, an enzyme foundation model built on a structure-grounded representation of an enzyme paired with its reaction, from which catalytic compatibility, kinetics and mutational response are read out together. On a time-separated benchmark, AlphaEnzyme achieves a top-1 catalyst retrieval success rate of 30.1%, against 17.2% for the strongest baseline. Michaelis constant predictions achieve the highest rank correlations among the compared methods, reaching 0.64 for enzymes sharing at most 40% identity with the kinetic training set. The model also enriches high-fitness TrpB variants at a budget of 16 candidates without target-specific fitness training. In prospective transaminase engineering, several prioritized variants exceed the parent, with the best design yielding a 30.4-fold increase in product signal in a lysate-based assay while retaining (R)-selectivity. Together, AlphaEnzyme provides a common molecular basis for selecting starting enzymes and prioritizing sequence changes for their improvement.

CRISPR-Cas9 orthologs remain functionally underexplored despite the rapidly growing scale of publicly available metagenomic sequencing data. Here, we applied a multi-stage computational discovery pipeline to two independent sources of metagenomic sequence data, a multi-biome collection of 660,185 metagenome-assembled genomes spanning 17 public catalogs, and raw sequencing reads from human microbiome studies assembled de novo, to identify novel Cas9 orthologs. Candidates were filtered for guide RNA architectural plausibility, sequence novelty, conservation of catalytic residues, and structural integrity, yielding 14 candidates spanning both data sources for experimental characterization. To determine PAM preference at scale, we developed a cost-effective Nanopore long-read-based adaptation of the high-throughput PAM determination assay (NanoPAMDA), enabling rapid, time-resolved, kinetic PAM profiling across many candidates in parallel. One Cas9 effector, E10, showed PAM-specific activity in vitro and was catalytically active in E. coli. Together, these findings demonstrate a scalable, long-read sequencing-based framework for discovering and biochemically characterizing novel Cas9 orthologs from metagenomic data.

Long-read amplicon sequencing profiles communities at high resolution, but taxonomic accuracy depends on the reference, classifier and the parameters used. Here, we present HiFiTaxa, a reproducible, single command Nextflow pipeline for PacBio HiFi 16S rRNA gene and fungal ITS amplicons. It builds versioned references from current GTDB and UNITE releases and supports Bayesian lowest common ancestor (BLCA) and naive Bayes (NB) classifiers and the read level profilers Emu (16S rRNA gene) and EMITS (ITS). We benchmarked them on mock communities and on holdouts with 10% of GTDB or UNITE databases sequences withheld. On the 16S rRNA gene mock of 20 species, BLCA detected all 20 species and reported two false positives, NB detected 17 and reported none, and Emu detected all 20 and reported eight. In the GTDB holdout, for taxa represented in the reference, BLCA reached 80.0% species level accuracy against 48.1% for the exact matching species step of NB. However, NB produced fewer incorrect assignments for taxa absent from the reference. On fungal ITS, BLCA had slightly higher holdout accuracy at species level(54.7%) against 52.2% for NB. NB was more conservative for novel taxa, and EMITS detected the most mock species. HiFiTaxa therefore offers a single reproducible workflow that uses current GTDB and UNITE releases and enables users to choose between different classifier algorithms depending on their needs.

The methanotroph Methylococcus capsulatus encodes the carbon dioxide (CO2)-fixing enzyme ribulose-1,5-bisphosphate carboxylase/oxygenase (RubisCO), which, along with CO2 and methane (CH4), are required for growth of this bacterium. This unique, dual one-carbon metabolism is present in phylogenetically diverse methanotrophs and is likely widespread in nature, but the connection between organic and inorganic carbon metabolism in these methanotrophs is unclear. Here we used a systems approach to investigate the requisite role of CO2 in M. capsulatus metabolism and physiology. We show that atmospheric CO2 levels are sufficient for M. capsulatus cultivation, and that CO2 assimilation is linked to cellular redox balance. In the absence of CO2, M. capsulatus elicits a global transcriptional response that includes induction of RubisCO- and other inorganic carbon assimilation-encoding genes with a concomitant decrease in ribosome biogenesis and nitrogen assimilation genes typical of a bacterial stringent response. This CO2 starvation response balances biosynthesis with carbon availability. Genetic and biochemical evidence show that a LysR-type transcriptional regulator (CbbR) activates RubisCO and other CO2 assimilation gene transcription via direct promoter binding in response to elevated intracellular RubisCO substrate, ribulose-1,5-bisphosphate, in CO2-starved cells. These results offer insights into the essentiality of CO2 for M. capsulatus metabolism and establish a function for CbbR-family regulators in methanotrophs beyond their canonical association with CO2 fixation in autotrophic bacteria.

Antifungal peptides are promising drug candidates to fight fungal infections in the clinics and agriculture. However, recent data suggest that antifungal peptides might also play a role within their own producing organism to survive nutrient limiting conditions. We have therefore studied the function of the antifungal AnAFP in Aspergillus niger in more detail. To achieve this, we established a Tet-on controlled anafp expression system, which allowed us to study a null and an overexpression phenotype in the same isolate. We observed that increased intracellular AnAFP expression reduces growth of A. niger and prematurely activates autophagy. Comparative transcriptome analyses of glucose-starving mycelium demonstrated that increased anafp expression strongly impacts expression of genes important for cell wall integrity and remodeling, as well as genes with a predicted function in metabolism and transport of carbohydrates, proteins, and lipids. Notably, genes encoding regulators of conidiophore development such as flbC and flbD became induced upon anafp overexpression. Fluorescent analyses of a Tet-on driven AnAFP::eGFP fusion protein congruently unraveled that AnAFP localizes to cell walls and septa of A. niger. Moreover, AnAFP::eGFP expression is spatially restricted to selected compartments only and affected cells displayed a sudden reduction in hyphal diameter. From these data we conclude that AnAFP is important to drive vegetative growth and sporulation in A. niger during nutrient limitation through autophagic recycling. We predict that AnAFP drives nutrient mobilization through selective cell lysis to ensure the survival of the whole colony during phases of starvation.

In vivo protein-RNA interactions are mapped almost exclusively by CLIP-seq, which profiles one protein in one cell type per experiment and needs a specific antibody; only about 10% of the roughly 2,000 human RNA-binding proteins have been assayed, and the Protein Data Bank holds around 5,000 protein-RNA complexes, a quarter of them ribosomes. Sequence-based prediction could extend coverage to unprofiled, newly discovered, or designed molecules, but existing predictors are not built for that case. One family trains a separate model per protein and cannot move beyond proteins with CLIP data; the other scores pairs but targets a single RNA class. Most are evaluated with random splits and randomly paired negatives that leave near-identical sequences on both sides of the split, and many require structure or read only short windows, discarding the context around a binding site. Their accuracy falls sharply when whole molecules are held out. We address this with two contributions. First, a strict benchmark of 2.1 million examples built from 11.6 million peaks of CLIP experiments targeting 209 proteins, taken from four public databases, in which each RNA is a spliced transcript with nucleotide-level labels, and sequence-similarity clusters are assigned whole to training, validation, and test sets, giving separate slices for unseen RNAs, unseen proteins, and unseen pairs. Negatives are matched to positives in length and in the number of other proteins binding the transcript, removing the shortcut by which a model predicts binding from how popular a transcript is without reading the protein. Second, RPI-Zero, a sequence-only model that reads any full-length protein and RNA jointly in up to 4,096 tokens, so the RNA representation is conditioned on the protein, and describes every protein as a mixture of shared motif prototypes, placing unseen proteins in the same terms as training ones. One model covers all proteins and RNA classes, uses no structure, and needs no interaction data for a new molecule. Compared with five state-of-the-art methods, it achieves the highest score in eight different metrics on unseen RNAs, unseen proteins, and unseen pairs. The model and data are available at https://github.com/ciceklab/RPI-Zero and https://zenodo.org/records/23748141, respectively.

Protein sequence databases are growing far faster than experimentally curated annotations, limiting functional interpretation of newly identified proteins and coding variants. We developed PALM, a protein annotation language model that jointly predicts domain and functional-site occurrence, while preserving overlapping and nested annotations through a soft-target objective. PALM integrates ESM2 with low-rank adaptation (LoRA), learned layer aggregation, and a bidirectional long short-term memory (BiLSTM) encoder. PALM was trained on ~15 million InterPro-annotated UniRef50 sequences spanning 17,951 domain and 1,383 site classes. Across 2.3 million held-out proteins, PALM achieved domain- and site-occurrence F1 scores of 0.941 and 0.877, respectively, and retained substantial performance in proteins with <20% sequence identity to the training and validation data. PALM also substantially outperformed ESM3 annotation baseline. Without training on variant-pathogenicity labels, PALM embedding disruption distinguished 151,823 ClinVar missense variants more accurately than ESM2 embedding disruption (AUROC 0.835 versus 0.710), and combining PALM embedding disruption with the ESM2 likelihood-ratio signal reached an AUROC of 0.877. The same combined zero-shot score outperformed AlphaMissense on a benchmark dataset on 128,161 missense variants, that were derived from literature but not present in ClinVar database. At the protein level, PALM embeddings showed stronger CATH-Gene3D superfamily organization than frozen ESM2 and ESM3 embeddings, and improved prediction of experimental activation and repression effects when combined with ESM2. In a downstream case study, we demonstrated the ability to use PALM-derived embedding as proxy of deep mutational scanning (DMS), showcasing the broad application of PALM. Together, PALM converts curated residue-level knowledge into transferable and mechanistically interpretable representations that connect protein sequence, function, and variant consequence.

Obtaining sufficient quantities of purified proteins in their intended molecular states is essential for biological research and protein engineering. Advances in computational protein design have expanded the functional capabilities of de novo proteins and accelerated candidate generation. However, reliable methods for ranking expected monomeric yield of candidate de novo designs, including molecular binders, fluorescent proteins and enzymes, have yet to be established. Standardized measurements of monomeric yield across diverse sequences remain scarce, limiting our understanding of its determinants and the development of accurate predictors for candidate selection. Here we developed a scalable pooled assay combining peptide barcodes, size-exclusion chromatography and mass spectrometry to quantify monomeric yield for over 39,000 natural protein domain variants, de novo proteins and nanobodies under uniform and standardized E. coli production conditions. We used this dataset to (1) quantify the relationship between folding stability and monomeric yield, (2) benchmark inverse-folding methods, (3) evaluate the utility of structure-prediction confidence metrics in selecting high-yield design candidates, and (4) identify sequence and structural patterns in nanobodies associated with monomeric yield and monomer-to-dimer yield ratios. We further developed ESMC-MonoYield, a protein language model fine-tuned to predict monomeric yield from amino acid sequence alone, and prospectively demonstrated enrichment of high-yield candidates among BindCraft-designed binders. Our approach enables standardized, large-scale measurement of monomeric yield across diverse proteins, and establishes monomeric yield as a sequence-encoded property that can be learned, predicted and ultimately designed.

Hyperuricemia is associated with renal injury and disturbances in gut microbial ecology. Intestinal urate clearance by engineered probiotics is limited by antibiotic-free functional maintenance and the oxygen requirement of urate oxidase (UOX). We reprogrammed the native cryptic plasmids of E. coli Nissle 1917 (EcN) to construct M12U, a marker-free dual-plasmid strain. pMUT2 carried an HucR-regulated urate-responsive UOX module, while pMUT1 encoded a low-oxygen-responsive UOX module and Tat-targeted Vitreoscilla haemoglobin (tat-VHB). M12U degraded urate in vitro and retained activity under low-oxygen conditions. In hyperuricemic mice (n = 5 per group), oral M12U reduced serum urate, creatinine, blood urea nitrogen, inflammatory cytokines and renal histopathological injury. 16S rRNA profiling showed increased microbial richness and a partial shift in community composition toward healthy controls. Shotgun metagenomics revealed enrichment of Lactobacillus johnsonii and Bacteroides acidifaciens, together with increased metabolic potential related to the pentose phosphate pathway and glutathione metabolism. Faecal metabolomics showed reduced uric acid abundance and increased thiamine monophosphate and argininosuccinic acid. These results support further evaluation of dual-native-plasmid architectures for gut-directed urate clearance.