Disordered protein LAT encodes relative levels of signaling pathways in T cell activation

The disordered adapter protein LAT helps relay signals from T cell receptors to multiple downstream pathways, but how it coordinates these parallel outputs has been unclear. Researchers developed a single-cell screening method to systematically map LAT's functional regions, uncovering important segments including protein interaction motifs and negatively charged blocks. Surprisingly, disrupting any individual segment tended to impair all downstream pathways simultaneously rather than selectively affecting one. To explain this, the team combined molecular biology, computational modeling, and imaging, showing that losing one partner protein interaction indirectly disrupts others — likely because these partner proteins serve double duty as both downstream effectors and as bridging factors that physically link LAT molecules together. The result is a tightly coupled signaling architecture where perturbation anywhere propagates broadly. Beyond the biology, the study demonstrates a flexible screening framework for dissecting sequence-function relationships in disordered proteins with complex, multi-pathway activities.

Read the original article →

Toward life with a 19–amino acid alphabet through generative artificial intelligence design

Researchers asked whether life could function with fewer than the standard 20 amino acids by attempting to engineer *Escherichia coli* to survive without isoleucine (Ile). Computational analyses first suggested Ile might be dispensable, and the team then systematically replaced all Ile residues in essential proteins, including all 382 instances across ribosomal subunits. Protein language models and structure-based AI design tools were critical for successfully redesigning functional Ile-free versions of these proteins in most cases. Combining 21 redesigned ribosomal subunits at a native genomic locus produced a viable, evolutionarily stable bacterial cell. The work demonstrates that a living organism can, in principle, operate with only a 19–amino acid alphabet and offers a methodological blueprint for eventually creating a fully Ile-free organism—pushing back on the long-held assumption that the canonical set of 20 amino acids represents a hard biological minimum.

Read the original article →

Can AI simplify the alphabet of life?

Proteins in nature are built from 20 standard amino acids, but researchers asked whether that full chemical alphabet is truly necessary. Using generative AI models trained on natural protein sequences, they designed functional proteins while excluding one amino acid — cysteine — from the palette. The resulting 19-amino-acid proteins folded correctly and retained biological activity, suggesting the 20-amino-acid canon may not be a hard requirement for protein function. Beyond practical implications for protein engineering, this finding raises evolutionary questions about why life settled on its particular chemical toolkit. The work also demonstrates that AI design tools are sensitive enough to navigate tight compositional constraints without sacrificing functionality, opening possibilities for engineering proteins with non-standard or simplified chemistries that could be easier to synthesize or more stable in therapeutic and industrial applications.

Read the original article →

Rumen ciliates modulate methane emissions in ruminants

Rumen ciliates are known to contribute substantially to methane emissions from livestock, but the biological mechanisms have been unclear. Researchers built a catalog of 450 rumen ciliate genomes—87% newly sequenced—and used it alongside methane measurements from 100 cows and nearly 1,900 rumen metagenomic and metatranscriptomic datasets. This analysis revealed clear links between ciliate abundance, methanogen abundance, and methane output. The study identified a previously undescribed organelle, termed the hydrogenobody, responsible for hydrogen production in these ciliates. Distinct from the hydrogenosomes found in other protists, hydrogenobodies sit near ciliary basal bodies and carry specific hydrogenases and oxygen reductases. Crucially, Vestibuliferida ciliates possess far more hydrogenobodies than Entodiniomorphida ciliates, giving them greater hydrogen-producing and oxygen-scavenging capacity and making them stronger drivers of methanogenesis. These findings offer new targets for strategies aimed at reducing livestock greenhouse gas emissions.

Read the original article →

Performance of a large language model on the reasoning tasks of a physician

A study tested a large language model on complex clinical diagnostic reasoning cases — the longstanding benchmark for expert medical AI systems — comparing its performance against hundreds of physicians across five structured experiments. The same LLM was also evaluated in a real-world setting, where its second opinions were compared with those of human experts for randomly selected emergency room patients at a major academic hospital. In every experiment, the LLM outperformed the physician baseline and showed clear improvement over earlier generations of clinical AI. The findings suggest that current LLMs have surpassed most established benchmarks for clinical reasoning, making the case that prospective clinical trials are now urgently needed to understand how these systems might be responsibly deployed in patient care.

Read the original article →