RMH

In this study, we developed two novel cumate-inducible expression systems by engineering distinct cumate-responsive operators, CuO1 and CuO2, respectively derived from the cym and cmt operons in P. putida KT2440. We assessed the expression strength of the constructed systems and optimized inducer concentrations using eGFP fluorescence intensity as a reporter. Through systematic promoter optimization, operator architecture optimization, and quantitative benchmarking, we achieved substantial improvements in system performance. Finally, we applied the optimized system to bio-transform limonene into perillyl alcohol via a heterologous cytochrome P450 to achieve an increase in the production. Cumate-inducible expression system R2CuO1 demonstrated superior gene expression strength than the IPTG-inducible system Ilac. We successfully implemented our optimized cumate-responsive platform for perillyl alcohol bio-production from limonene, achieving a titer of 40.93 mg·L⁻¹, 2.1 times higher than that of the conventional lac expression system.

Förster resonance energy transfer (FRET) is commonly used to monitor protein-protein interactions in situ. The high spatiotemporal resolution and facile implementation inside complex molecular environments have spearheaded FRET's widespread adoption in biosensing. Despite these advantages, current FRET biosensors are largely restricted to the detection of the presence/absence of individual inputs and are thus unable to sense several multiplexable inputs simultaneously within complex milieu of biological environments. In this work, we introduce a generalizable strategy to construct genetically encoded protein-based FRET biosensors capable of recognizing multiple inputs following Boolean logic-type (YES/OR/AND) operations. These topologically specified FRET sensors powerfully expand the input capacity in sensing protein-protein interactions while providing a user-programmable platform for monitoring heterogeneous biological activities both in vitro and in living cells.

Accurate and fast prediction of protein binding sites remains essential for understanding molecular interactions and facilitating protein engineering. Here, we present SurfGraphPro, a geometric deep learning approach that bridges protein language models with surface-based structural representations for a rapid binding interface prediction. Our method operates on coarse-graphed triangulated protein surfaces, utilizing solvent-excluded surface meshes downsampled into amino acid residue centered patches. Unlike existing approaches that rely on extensive physicochemical feature engineering, we leverage evolutionary information directly from protein language model embeddings, eliminating the need for computationally expensive multiple sequence alignments and hand-crafted features. This integration achieves on average 18-28 times speedup for proteins of 100 to a few 1000s of amino acids over current state-of-the-art surface-based model while maintaining comparable accuracy on diverse binding interfaces, including challenging antibody-antigen complexes. To our knowledge, this represents the first approach to integrate protein language model embeddings with coarse geometric surface representations for binding site prediction, demonstrating that learned evolutionary features coupled with geometric transformers can replace traditional feature engineering without sacrificing performance.

Reliable constitutive promoters for non-model bacteria are lacking, resulting in a bottleneck for genetic tool development. Existing computational promoter predictors are trained on model Gammaproteobacteria and degrade on phylogenetically distant or compositionally atypical genomes. Here we present ProFinder, a computational method that integrates three biologically informed criteria for promoter prediction: a conserved 98-gene constitutive marker panel, an empirically derived intergenic search window, and a 72-motif σ70 motif catalogue built from diverse lineages. From an input bacterial assembly, ProFinder returns a ranked shortlist of putative constitutive promoters. It matches the precision of full-genome classifiers and is more robust to GC extremes. Applied to 56,742 bacterial genomes spanning 149 phyla, ProFinder returned candidate promoters for 99.76%, delivering a pan-bacterial catalogue of pre-computed constitutive promoter parts to accelerate chassis development across the bacterial tree. https://plabase.cs.uni-tuebingen.de/profinder/ 

TnpB nucleases, the evolutionary progenitors of CRISPR-associated Cas12 enzymes, are transposon-encoded, RNA-guided endonucleases found throughout bacteria. Independent of the trajectory towards adaptive immunity, TnpB nucleases have also recurrently given rise to TnpB-like nuclease-dead repressors (TldRs), a family of programmable RNA-guided transcription factors, though the physiological roles of most TldR clades remain unclear. Recently, we identified a TldR clade associated with bacterial ABC transporter operons, but this clade has not been characterized in any native host, leaving the biological significance of its predicted regulatory function unknown. Here, we demonstrate direct in vivo repression of the oligopeptide permease (opp) binding protein OppA in Enterococcus faecalis, implicating TldR in shaping the substrate repertoire of bacterial peptide import machinery. Phylogenetic and comparative genomic analyses reveal that bacterial genomes typically encode multiple, structurally conserved but functionally diversified OppA paralogs, suggesting that TldR-mediated repression could enable selective tuning of transporter composition. To define the structural basis of this regulatory specificity, we used cryo–electron microscopy to capture an oppF-associated TldR in multiple functional states, revealing a conserved bilobed architecture and a TAM recognition mechanism inherited from TnpB ancestors. Together, these findings define the structural, genetic, and evolutionary basis of a widespread RNA-guided regulatory system domesticated from mobile genetic elements to tune peptide transport in bacteria.

Reliable biological prediction by AI is most likely to emerge first in the simplest systems with rich datasets and the fastest opportunities for testing and refinement. Phages with small genomes are an ideal testbed. Decades of experimental work have created an unusually information-rich literature benchmark that sits largely outside the sequence repositories typically used to train genome language models (GLMs). Recent work has shown that GLMs such as Evo2 can generate viable whole bacteriophage genomes, demonstrating that genome-scale biological design is possible. The next question is more mechanistic: can such models correctly predict the effects of simple, local sequence changes? Here, we evaluated Evo2 against experimental mutation data from two model phages: the single-stranded RNA phage MS2 and the single-stranded DNA phage ΦX174. We first addressed whether Evo2 sequence log-likelihood scores could distinguish viable from nonviable mutations, and then whether models trained on Evo2 embeddings were predictive of function. Across both phages, Evo2 captured expected sequence-level constraints: stop codons were generally penalized, synonymous substitutions had higher likelihood than nonsynonymous substitutions. Evo2 distinguished among synonymous codons in ways only weakly explained by host codon usage. However, for MS2, these capabilities did not translate into robust biological predictions. Specifically, Evo2 ΔSLL failed to distinguish functional from nonfunctional mutations in the lysis gene, an overlapping viral region that appears to be a particularly hard test case. In ΦX174, performance was stronger but still modest, and much of the apparent signal could be explained by simple covariates such as nonsense mutations, genomic position, and nucleotide distance from the reference. Together, these results introduce phages as a tractable proving ground for stress-testing GLMs against experimentally grounded genotype-to-phenotype tasks. More broadly, they provide a durable framework for identifying what data and model ingredients are required for biologically reliable prediction.

Fungal bio-leaching, as a green and sustainable technology for metal recovery, demonstrates unique advantages in addressing growing global metal demand and utilizing secondary resources. This review systematically summarizes recent progress and technological advances in fungal bioleaching for metal recovery. Fungi solubilize metals through three core mechanisms, acidolysis, complexolysis, and redoxolysis, mediated by organic acids (citric, oxalic, gluconic), while cell wall functional groups enable biosorption. Most frequently used genera in modern bioleaching (Aspergillus, Penicillium, Trichoderma) have achieved remarkable recoveries from e-waste (Cu, Li, Co up to 100%), rare earth elements (consortia up to 76%), tailings, and spent catalysts. Optimization of key parameters (pH, temperature, pulp density) and application of intelligent strategies (response surface methodology, machine learning) have further enhanced leaching efficiency. Strain engineering and synthetic biology offer new avenues for constructing hyper-efficient strains. Despite remaining bottlenecks such as slow kinetics and scale-up difficulties, fungal bioleaching holds broad industrial prospects for advancing the circular metal economy and the green metallurgical transition.

Glucoamylase is essential for industrial starch saccharification, but the limited thermostability and near-neutral pH tolerance of fungal glucoamylases necessitate cooling and acidification of liquefied starch. Here, we developed an artificial intelligence-guided strategy to simultaneously improve the thermostability, pH tolerance, and catalytic activity of glucoamylase from Penicillium oxalicum (PoGA). Two property-specific machine-learning models, CASPE-T and CASPE-A, identified substitutions associated with thermostability and pH tolerance, respectively. Experimental screening identified beneficial substitutions in 11 of 21 CASPE-T and 12 of 22 CASPE-A candidates. Folding-energy-guided recombination integrated the two traits while maintaining structural compatibility. The optimal variant, PoGA T513E/Q305N, exhibited 2.21-fold higher specific activity than the wild type, with half-life extended from 22.3 to 57.9 min at 60 degrees C and from 16.6 to 64.7 min at pH 8.0. Molecular dynamics simulations attributed these improvements to reinforcement of high-occupancy hydrogen-bonding networks, suppression of conformational fluctuations in the linker and carbohydrate-binding module, enhanced long-range dynamic coordination, and preservation of a compact catalytic architecture. At 60 degrees C and pH 6.5 without acidification, PoGA T513E/Q305N produced 219.9 g/L glucose and achieved 89.1% starch conversion, 31.4% higher than the wild type. This work provides an efficient framework for multi-objective enzyme engineering and sustainable starch biorefining.

Recent advances in deep learning-based structure prediction, including AlphaFold3 and the open-source Boltz model family, have extended biomolecular modeling to joint prediction of protein-ligand, protein-nucleic acid, and multi-chain complexes with binding-affinity estimation. Boltz-2 is among the most feature-complete of these open models, but its practical use requires a local CUDA-capable GPU, command-line execution, and manually authored YAML configuration files, limiting accessibility for researchers without dedicated computational infrastructure. We developed Boltz2-Notebook, a Colab-native interface comprising four integrated stages - automated environment setup, interactive parameter-to-YAML generation, execution management, and automated confidence and affinity visualization - together with a manifest-driven batch mode for multi-target screening. All modelling capabilities are inherited unmodified from Boltz-2; Boltz2-Notebook's contributions are limited to accessibility, input construction, and workflow automation. Independent of the software, we curated a benchmark of 317 protein-ligand pairs (122 proteins, 277 ligands) from BindingDB and predicted binding affinity in triplicate using the Boltz-2 command-line engine on high-performance computing infrastructure. Predicted and experimental pIC50 values showed moderate correlation (Pearson r = 0.609 [95% CI 0.540-0.675]; Spearman ρ = 0.625; R2 = 0.371; MAE = 0.968 pIC50 units), with high triplicate reproducibility (pairwise r = 0.97) but a systematic compression of the predicted affinity range and no measurable relationship between Boltz-2's self-reported confidence metrics and prediction accuracy. Boltz2-Notebook is freely available as open-source software and provides external, reproducible evidence - including a specific confidence-calibration limitation - relevant to interpreting Boltz-2 affinity predictions responsibly.

Division of labor is commonly associated with cooperative behavior, yet one of its most extreme forms occurs during bacterial warfare, where a subset of cells undergoes suicidal lysis to release toxins. Why bacteria divide labor among a few cells rather than producing toxin uniformly remains unknown. Here, we combine timelapse microscopy and simulations, to understand the division of labor during bacterial warfare using bacteriocin (colicin) production by the gut bacterium E. coli as a model system. At the single-cell level, we find that lytic toxin production is a tightly regulated event: only cells that commit to lysis produce significant toxin and then lysis only occurs once a large amount of toxin has been made. This high threshold ensures each sacrifice delivers a large dose, which is released in a rapid burst from a lysing cell. While such burst-like release appears to provide no advantage over uniform labor in well-mixed conditions, we show it becomes extremely effective in spatially structured populations where the rapid release of toxin by one cell can generate lethal concentrations locally and eliminate competitors. Finally, we explain why the lysing fraction remains so small. While an increase in the producing cells increases toxin levels, it also increases the probability of local patch extinctions. The division of labor during bacterial warfare, therefore, enables powerful localized killing while safeguarding the population from self-destruction.

Protein language models (PLMs) have transformed our ability to learn from evolutionary sequence space, but protein engineering ultimately asks a different question: not what evolution selected, but what we should build next. Zero-shot likelihoods therefore provide useful, but not universal, measures of fitness and can misalign with engineering objectives. Experimental supervision redirects these priors towards properties of interest, enabling target-specific prediction and closed-loop optimization. PLMs thereby complement structure-based design: structural methods provide geometric control, while PLMs integrate experimental feedback to optimize functional and developability properties. Realizing this potential requires evaluation beyond retrospective predictive accuracy, towards extrapolation, multi-objective optimization, and prospective experimental success. We argue for recurring competitions combining scalable predictive benchmarks with prospective experimental challenges. Ultimately, progress should be measured not by predicting existing experiments, but by enabling successful new ones.