2  ▏Nature

      iPEX Achieves Micrometre-Scale Untargeted Spatial Proteomics

      Multi-omics Dissection of Pig-to-Human Kidney Xenotransplantation: Mapping the Dynamics of Immune Rejection

      Individual Variation in the Gut Microbiome Mediates the Efficacy of Resistant Starch in MASLD

      From Microbiota to Metabolism: Bridging a Two-Decade Translational Gap

      From Static Snapshots to a Dynamic Movie: ESPRESSOEnables Spatiotemporal High-Dimensional Phenotyping at Single-Cell Resolution

      A New Tool for Precision Medicine: HapScoreDB Reveals the Cooperative Functional Effects of Genetic Variants

Designing entirely new genes (de novo genes) with defined biological functions from scratch has long been a major challenge in synthetic biology. Conventional methods depend on known protein structures or evolutionarily conserved sequences, which limits the space for innovation. Recently, the team of Brian L. Hie at Stanford University published a landmark study in Nature, proposing a new paradigm called "semantic design" that harnesses the genomic language model Evo to generate large numbers of brand-new genes with experimentally validated activity. The team found that by learning the "semantic relationships" among prokaryotic genes, Evo works like an intelligent input method: given a DNA prompt encoding a functional context, it automatically completes and generates novel sequences enriched for the target function. Remarkably, these generated genes — including anti-CRISPR proteins and type II/III toxin–antitoxin systems — share no significant sequence similarity with any protein known in nature, yet display high activity and a high success rate in experiments. Going further, the team used Evo to run millions of prompt-based generations and built the SynGenome database, which contains more than 120 billion base pairs of AI-generated genomic sequence, providing a powerful resource for function-oriented gene design. The method requires no structural priors, no evolutionary conservation and no task-specific fine-tuning, marking the arrival of generative genomics at a new stage that goes "beyond natural sequences."

▼

Scan the QR code to read the original article

Existing spatial omics technologies struggle to deliver both high coverage and micrometre-scale in situ localization of proteins. A team from the Chinese Academy of Sciences, working with MIT and other institutions, presented a breakthrough method in Nature: iPEX (in situ imaging proteomics via expansion). By innovatively combining isotropic tissue expansion with MALDI mass spectrometry imaging, iPEX raises spatial resolution to 1–5 micrometres while improving protein detection sensitivity 10–100 fold. Using the mouse retina as a model, the study produced a high-precision spatial protein atlas that clearly resolved single-cell layers and extracellular structures and enabled visualization of co-localized proteins. iPEX has been validated in brain, intestine, liver and organoid tissues, identifying 600–1,500 proteins in a single run. When applied to the 5xFAD model of Alzheimer's disease, iPEX detected mitochondrial dysfunction at an early disease stage and revealed that the peroxisomal protein ACAA1A (whose N392S variant is a monogenic AD risk factor) is significantly downregulated, blocking the synthesis of long-chain polyunsaturated fatty acids such as DHA. This phenotype could be rescued by wild-type ACAA1A but not by the mutant, suggesting that lipid metabolism disturbance is at work early in neurodegeneration. iPEX provides a powerful new tool for untargeted, high-resolution spatial proteomics.

▼

Multi-omics Dissection of Pig-to-Human Kidney Xenotransplantation: Mapping the Dynamics of Immune Rejection

Organ shortages have made gene-edited pig kidneys a new hope for transplantation, but immune rejection remains the key obstacle. Robert A. Montgomery and Brendan J. Keating of NYU Langone Health, together with more than 20 institutions, published in Nature the first multi-omics longitudinal study of a gene-edited pig kidney transplanted into a brain-dead human recipient and followed for 61 days. Using single-cell transcriptomics, immune repertoire analysis and proteomics, the team revealed B cell and NK cell activation on days 10–28 after surgery and antibody-mediated rejection (AbMR) on day 33; T cell infiltration then intensified, with combined antibody and cellular rejection by day 49. CXCL9⁺ macrophages accumulated in large numbers within the graft, indicating that IFN-γ-driven type I inflammation dominated the local response, and interactions between pig-derived macrophages and human immune cells were observed. Pro-fibrotic markers such as SPP1 and S100A6 appeared early, while complement activation could be partially blocked by inhibitors. This study provides the first systematic spatiotemporal map of the human immune response in xenotransplantation, identifying key targets such as CXCL9 and TRBV2/J1 clones and laying a mechanistic foundation for improving xenograft survival.

▼

Scan the QR code to read the original article

Resistant starch (RS) has shown promise as a prebiotic in the treatment of metabolic dysfunction-associated steatotic liver disease (MASLD), yet clinical outcomes differ substantially between individuals. The team of Academician Wei-Ping Jia at the Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, together with multicentre researchers, reported in Cell Metabolism for the first time that baseline gut microbiota composition is the core determinant of this heterogeneity in RS efficacy. A randomized, double-blind, placebo-controlled trial showed that about 30% of MASLD patients responded poorly to RS. Multi-omics analyses and faecal microbiota transplantation experiments confirmed that inter-individual differences in gut microbiota are the main cause of the divergence in efficacy. Further mechanistic work found that a high abundance of Prevotella suppresses beneficial bacteria capable of degrading RS (such as Ruminococcus), thereby weakening the metabolic benefit, whereas Bifidobacterium pseudocatenulatum RRP01 isolated from responders restored RS-degrading capacity and reversed the Prevotella-driven loss of efficacy. The predictive model built by the team from baseline microbiota plus clinical indicators (AUC = 0.74–0.87) can identify "potential non-responders." The study reveals the origin of heterogeneity in dietary intervention outcomes and opens a new path toward "microbiome-guided precision nutrition."

▼

Scan the QR code to read the original article

The human microbiome was once expected to become the next engine of precision medicine after the genome. Yet although the past two decades have produced a flood of studies linking gut microbiota to obesity, diabetes, inflammatory bowel disease and neurodegenerative disease, clinical translation has remained frustratingly slow. Professors Matthias Van Hul and Patrice D. Cani of KU Leuven in Belgium published a forward-looking review in Cell Metabolism that systematically dissects the "translational gap" now facing microbiome research: an over-reliance on correlational descriptions (such as "dysbiosis"), a lack of causal mechanistic validation, and highly heterogeneous effects of interventions (such as probiotics and faecal microbiota transplantation) across populations. The authors argue that the key to breaking the bottleneck is to move from reductionism to systems biology — integrating multi-omics data (metagenomics, metabolomics, host immunomics and more), focusing on the functional output of the microbiota rather than species composition alone, and building individualized predictive models with the help of artificial intelligence. The article places particular emphasis on the regulatory role of circadian rhythms and the gut clock in microbiota–host interactions, and calls for future research centred on "function – dynamics – precision," so as to truly cross the distance from "association discovery" to "causal intervention" and open new routes to personalized treatment for metabolic diseases.

▼

Scan the QR code to read the original article

From Static Snapshots to a Dynamic Movie: ESPRESSO Enables Spatiotemporal High-Dimensional Phenotyping at Single-Cell Resolution

Although genomics, transcriptomics and proteomics have greatly advanced our understanding of cellular heterogeneity, existing technologies generally lack the temporal dimension and subcellular spatial resolution needed to capture the dynamic evolution of cell states. To address this bottleneck, a team from the University of California, Irvine (UCI) and collaborators in Italy published a breakthrough methods paper in Nature Methods introducing ESPRESSO (Environmental Sensor Phenotyping RElayed by Subcellular Structures and Organelles) — a novel spatiotemporal multi-omics technology that fuses imaging with computation. Through fluorescent labelling, advanced microscopy and intelligent image analysis, ESPRESSO extracts the morphology, distribution and functional features of organelles at the single-cell level to build a high-dimensional "organelle phenotype." The method can distinguish cell types, trace differentiation trajectories, monitor stress responses and immune cell polarization, and can be combined with gene expression data to dissect the underlying mechanisms. Its generality has been validated across multiple cell systems and 3D models, opening new avenues for the dynamic study of complex systems such as organoids and the tumour microenvironment. Because ESPRESSO leaves the sample intact, it enables long-term, non-invasive, high-throughput monitoring of cell state, truly moving cell biology into the era of the "dynamic movie."

▼

Scan the QR code to read the original article

A New Tool for Precision Medicine: HapScoreDB Reveals the Cooperative Functional Effects of Genetic Variants

Deciphering the cooperative functional effects of multiple gene variants on a haplotype is an important challenge for precision medicine. A team at the University of Trento in Italy has released HapScoreDB in Nucleic Acids Research — the first haplotype functional score database built on protein language models (pLMs). The database integrates GENCODE/Ensembl transcript annotations with phased variant data from the 1000 Genomes Project, covering around 18,000 genes and 78,000 transcripts and constructing more than 130,000 protein haplotypes; advanced pLMs including ESM2, PoET and proSST are used to compute pseudo-log-likelihood (PLL) scores that quantify functional fitness. The analysis shows that haplotypes carrying cancer GWAS variants score significantly lower, and that the dispersion among haplotypes of key cancer genes is greater, indicating that the approach effectively captures functional perturbation. HapScoreDB offers an interactive web platform (https://bcglab.cibio.unitn.it/hapscoredb) supporting browsing, visualization and bulk download, providing an important resource for variant pathogenicity assessment, transcript prioritization and population functional genomics research.

▼

Scan the QR code to read the original article