The Golgi apparatus is best known as the processing and dispatch center of the secretory pathway. Proteins manufactured in the endoplasmic reticulum pass through its stacked membrane cisternae, where they are chemically modified, most notably through glycosylation, the attachment of sugar chains, and then sorted to their correct destinations, whether the plasma membrane, secretory vesicles, or lysosomes. The organelle also manufactures key lipids such as sphingomyelin and glycosphingolipids, participates in autophagosome formation, and runs its own quality-control degradation pathways. In other words, the Golgi sits at the crossroads of protein and lipid homeostasis, which is precisely why its failure could ripple through nearly every cellular process implicated in neurodegeneration.

That failure is not hypothetical. Postmortem studies of Alzheimer’s brains have long revealed fragmented, atrophied Golgi structures in affected neurons. More strikingly, when researchers expressed the amyloid precursor protein Swedish mutant in cells, the Golgi fragmented, and restoring Golgi architecture by expressing the structural proteins GRASP65 and GRASP55 dramatically reduced amyloid-beta production. In induced pluripotent stem cell-derived neurons from patients with sporadic Alzheimer’s disease, the fragmented Golgi phenotype persisted and, crucially, appeared before amyloid and tau pathology developed, suggesting that Golgi breakdown may be one of the earliest cellular events in the disease rather than a late consequence of it.

The strongest new argument for the Golgi’s relevance comes from large-scale human data. When the authors surveyed genetic studies, a striking pattern emerged: multiple independent analyses keep nominating Golgi-related genes as Alzheimer’s risk factors. An RNA-sequencing analysis found that Golgi-related gene ontology terms were elevated both in patients and in people with high polygenic risk scores for the disease, particularly in the temporal cortex. An X-chromosome-wide association study identified SLC9A7, a sodium/proton exchanger concentrated in the trans-Golgi network that maintains the organelle’s acidity, as the only genome-wide significant risk locus. Because Golgi luminal pH governs protein modification, glycosylation, and trafficking, its disruption could derail amyloid precursor protein processing among many other consequences.

Other risk genes reinforce the same theme. SORL1 and SORCS1, both well-established genetic risk factors, encode sortilin-family receptors that shepherd the amyloid precursor protein back to the trans-Golgi network, keeping it away from the endosomes where amyloidogenic cleavage predominates. Adaptor protein 4 complex genes AP4M1 and AP4E1, flagged by enhancer mapping of genome-wide association loci, regulate the precursor protein’s delivery to endosomes. A protective variant in NSF, a gene essential for Golgi reassembly and vesicle fusion, was identified in whole-exome sequencing of patients lacking the ApoE4 allele, hinting that Golgi integrity itself may confer resilience. Machine learning analysis of European GWAS data proposed COG7, a subunit of the complex that maintains Golgi structure and glycosylation, as a new risk gene, and multiancestry proteomic analyses independently nominated COG7, the trans-Golgi metalloprotease CPD, and the retrograde trafficking proteins SNX1, SNX32, and STX6 as candidate causal proteins.

Proteomics adds a second layer of evidence. A landmark 2022 study of more than 500 human brain samples found that Golgi and protein transport module signatures were reduced in asymptomatic Alzheimer’s brains and diminished further in symptomatic cases, correlating with cognitive scores. Perhaps most intriguingly, the extracellular matrix protein repertoire, the matrisome, showed the most prominent proteomic changes in the disease that were invisible at the RNA level, implying that post-translational modifications are the real driver. Since matrix proteins such as proteoglycans and collagen are extensively processed and glycosylated within the Golgi before secretion, Golgi dysfunction offers a mechanistically coherent explanation for this transcriptome-proteome discordance, and matrisome proteins are known to colocalize with amyloid plaques.

A 2024 glycomics study sharpened the picture further by mapping the N-glycan landscape of Alzheimer’s brains. The most altered sugar structures, sialylated and highly branched forms, are built inside the Golgi, and the affected glycoproteins clustered at synapses and the cell surface. Glycoform signatures tied to disease status involved L1CAM and NPTX1, markers of synaptic plasticity and synaptic loss, as well as ATP1B1, a subunit of the sodium-potassium pump essential for neural signaling. Expression of several Golgi-resident glycosylation enzymes, including MAN2A1, MGAT1, and members of the B3GALT, B4GALT, ST6GAL, and ST8SIA families, was altered across multiple brain regions, consistent with a systematic breakdown of Golgi-dependent protein modification that could directly undermine synapses.

Lipid biology supplies a third pillar. The Golgi is the primary site of ceramide processing through glucosylceramide synthase, whose levels fall in Alzheimer’s brains in correlation with ceramide accumulation, a lipid that promotes amyloid-beta production, oxidative stress, and neuronal death. The Golgi-localized enzyme neutral sphingomyelinase 2 drives the secretion of extracellular vesicles that can spread pathological tau between cells; inhibiting it reduced tau propagation and neurodegeneration in mouse models, and neuronal deletion of the gene reduced amyloid-beta production. Meanwhile, the oxysterol-binding protein OSBP1, which exchanges cholesterol and PI4P between the trans-Golgi network and the endoplasmic reticulum at membrane contact sites, regulates cholesterol distribution, mitochondrial fission, and retrograde trafficking, and its knockdown increased amyloidogenic processing of the amyloid precursor protein, linking Golgi-managed cholesterol homeostasis directly to plaque biology.

The authors synthesize these threads into four candidate pathogenic mechanisms. First, Golgi fragmentation may disrupt the normal segregation of the amyloid precursor protein and its processing enzymes, increasing their contact and boosting amyloid-beta generation, possibly even within the Golgi itself. Second, impaired trafficking between the Golgi and the endolysosomal system would trap the precursor protein in endosomes while starving lysosomes of properly sorted enzymes, simultaneously raising amyloid production and lowering its clearance. Third, defective glycosylation would alter countless proteins, including the amyloid precursor protein itself, whose O-glycosylation influences its trafficking and cleavage. Fourth, lipid dyshomeostasis would feed back on Golgi membrane integrity, creating a self-reinforcing vicious cycle that also damages mitochondria and the endoplasmic reticulum. Upstream triggers, including aging, DNA damage, oxidative stress, and neuronal hyperexcitability, as well as amyloid-beta and tau themselves, can all induce Golgi fragmentation, closing the loop.

Therapeutically, the possibilities are still largely conceptual but tantalizing. Restoring Golgi structural integrity, for example through nonphosphorylatable GRASP55/65 variants, could in principle ameliorate multiple disease processes at once. More targeted options include nSMase2 inhibitors, which have already shown benefits in Alzheimer’s mouse models; overexpression of SORL1, which redirects the amyloid precursor protein to the Golgi and lowers amyloid-beta output; and reinforcement of glycosylation or lipid balance through COG7, glucosylceramide synthase, or OSBP1. The authors are candid about the caveats: causality between Golgi dysfunction and Alzheimer’s pathogenesis has not been rigorously established, most evidence comes from familial disease models rather than the sporadic form that accounts for roughly 90 percent of cases, and because the Golgi handles so much cellular traffic, interventions could carry pleiotropic side effects. Still, with anti-amyloid antibodies delivering only modest clinical benefit, the review makes a compelling case that keeping this cellular post office running smoothly may be one of the most underexplored strategies for halting the disease, and that the humble Golgi apparatus deserves a central seat at the Alzheimer’s research table.

Subject of Research: The role of Golgi apparatus dysfunction in Alzheimer's disease pathogenesis and therapy

Article Title: Golgi dysfunction in Alzheimer disease: from human multiomic signatures to therapeutic targets

Article References: Song, J., Lee, S.-J., & Mook-Jung, I. (2026). Golgi dysfunction in Alzheimer disease: from human multiomic signatures to therapeutic targets. Experimental & Molecular Medicine. https://doi.org/10.1038/s12276-026-01851-8

Image Credits: AI Generated

DOI: 10.1038/s12276-026-01851-8

Keywords: Alzheimer's disease, Golgi apparatus, amyloid-beta, tau, glycosylation, lipid metabolism, SORL1, trans-Golgi network, multiomics, neurodegeneration, endolysosomal trafficking, therapeutic targets

The scale of the problem is growing. Pediatric-onset inflammatory bowel disease is rising in incidence worldwide, and children tend to carry a more aggressive disease phenotype than adults, facing decades of potential inflammation, hospitalization and surgery. Biologic agents such as infliximab and adalimumab have transformed outcomes, but long-term cohort data show that a substantial fraction of children lose response over time, and drug approval pathways for young patients remain constrained. Against this backdrop, nutritional therapy occupies a unique position: it is the only induction strategy recommended by European guidelines as first-line for pediatric Crohn’s disease that simultaneously improves inflammation, supports growth and avoids pharmacological side effects. The catch, as the review emphasizes, is that remission rates vary considerably across studies depending on research design, treatment population and how remission is defined in the first place.

The authors frame the field’s central challenge as a transition away from a one-treatment-suits-all strategy toward a tailor-made, precision-based approach. In principle, a clinician armed with a reliable treatment-responsiveness phenotype could match each child to the therapy most likely to work before the first formula feed or the first injection. The benefits would cascade through the healthcare system: fewer months spent on ineffective regimens, less exposure to unnecessary immunosuppression, earlier escalation for the children who need it, and reduced lifetime costs of a disease whose economic burden is among the highest of any chronic inflammatory condition. The obstacle is that, despite decades of research, the evidence base for definitive predictors of response to nutritional therapy remains insufficient, a conclusion the review states plainly rather than glossing over.

What does the existing evidence actually suggest? Among the most consistently reported signals are early clinical improvement and a reduction in faecal calprotectin, the stool-based marker of intestinal inflammation that has become a workhorse of pediatric IBD monitoring. Serial measurements during exclusive enteral nutrition have been shown to track with disease activity and treatment response, and falling calprotectin concentrations during therapy have been linked to better outcomes and to the prediction of clinical relapse after treatment ends. Early mucosal healing, visible at endoscopy in newly diagnosed children, is also associated with improved long-term outcomes. But the picture is not clean: at least one prospective pediatric study concluded that faecal calprotectin is not a clinically useful marker for predicting early non-response to exclusive enteral nutrition, and the review notes that the rapid fall in calprotectin seen during treatment is lost quickly once food is reintroduced. Timing, it seems, matters as much as the marker itself.

Disease location offers another thread of evidence, though a tangled one. Classic work from the early 2000s suggested that colonic Crohn’s disease without ileal involvement responds less well to enteral nutrition, hinting that the intestinal segment where inflammation resides shapes the therapy’s effectiveness. Later studies complicated this view, with some reporting that disease phenotype does not influence clinical remission at all, and others examining whether specific scoring systems can predict short- and long-term mucosal healing across different gastrointestinal segments. The review also highlights practical determinants that have nothing to do with biology: adherence. Real-world studies of children starting exclusive enteral nutrition have identified factors associated with higher adherence and better response, and parent-perspective research underscores how difficult the diet’s social and psychological demands can be for families. Whether the formula is polymeric or elemental, and whether it is delivered orally or by tube, appears to have little effect on outcomes, which at least simplifies one clinical decision.

The most scientifically vibrant frontier, and the one the review treats as the field’s future horizon, lies in the gut microbiome. Nutritional therapy works, at least in part, by reshaping the microbial communities that drive intestinal inflammation, and several prospective studies have now reported microbial signatures that predict response. Children whose baseline gut communities contain particular bacterial taxa and functional profiles have been shown more likely to achieve sustained remission after exclusive enteral nutrition, and early shifts in microbial community structure during treatment have been associated with durable remission. Metabolomic analyses add another layer, with gut metabolome signatures alongside microbiota profiles predicting response in prospective pediatric cohorts. Even a child’s pre-diagnosis diet has been found to correlate with the microbiome and to predict response to enteral nutrition, suggesting that long-term dietary habits may prime the gut environment for success or failure before therapy even begins.

These findings align with a broader conceptual shift captured in a 2024 position paper from the European Society for Paediatric Gastroenterology, Hepatology and Nutrition, which called for precision nutrition in pediatric IBD. The logic is compelling: if exclusive enteral nutrition remodels the microbiome and its metabolic output, then the composition of a child’s starting microbial community should, in theory, determine how far that remodeling can go. Multidisciplinary consensus work on microbiome-based biomarkers published in The Lancet Microbe has begun to define the standards such biomarkers would need to meet before clinical deployment, and the review argues that integrating clinical data with multiomic information, spanning genomics, microbiomics and metabolomics, into a unified decision-making framework could eventually enable genuinely personalized treatment selection. Machine learning approaches developed for targeted therapies elsewhere in medicine offer a template for how such heterogeneous data streams might be combined.

The authors are careful, however, to temper enthusiasm with rigor. Anticipating treatment response, they write, transcends simple stratification based on clinical characteristics, meaning that no checklist of age, disease location, severity or laboratory values currently suffices to forecast outcome. The literature they surveyed provides insufficient evidence for definitive predictors, and many of the most promising microbiome and metabolome studies are small, single-center and retrospective, vulnerable to the confounding and overfitting that plague biomarker research. The path forward, they suggest, involves prospective validation of candidate predictors in large, well-phenotyped cohorts, harmonized definitions of remission and response, and ultimately the kind of integration of clinical and multiomic data that precision medicine initiatives in other fields have begun to demonstrate at scale.

For families facing a new diagnosis, the practical message of the review is one of cautious optimism. Nutritional therapy remains a first-line, evidence-based option, whether as exclusive enteral nutrition or as partial enteral nutrition combined with the Crohn’s disease exclusion diet, and randomized trial evidence shows the dietary approach can induce rapid, sustained remission in children. Early clinical improvement and falling faecal calprotectin during the first weeks of treatment are reasonable indicators that the therapy is working, and clinicians can use them to guide decisions about continuing or switching. What the field cannot yet do is predict, before the first sip of formula, which child will respond. Closing that gap, the review concludes, may be a small but crucial step toward individualized therapy for every child with Crohn’s disease, transforming a treatment that sometimes works brilliantly into one that works brilliantly for the right patients, chosen in advance.

Subject of Research: Predictors of response to nutritional therapy in pediatric Crohn's disease

Article Title: Current perspectives and future horizons in predicting response to nutrition therapy in children with Crohn’s disease – a narrative review

Article References: Krawiec, P., & de Ridder, L. (2026). Current perspectives and future horizons in predicting response to nutrition therapy in children with Crohn’s disease – a narrative review. Pediatric Research. https://doi.org/10.1038/s41390-026-05559-3

Image Credits: AI Generated

DOI: 10.1038/s41390-026-05559-3

Keywords: Crohn's disease, exclusive enteral nutrition, pediatric IBD, precision nutrition, faecal calprotectin, gut microbiome, Crohn's disease exclusion diet, personalized medicine, multiomics, biomarkers, mucosal healing, narrative review

The research team, led by Feixiong Cheng of the Cleveland Clinic Genome Center together with Yuan Hou, Yichen Li and Pengyue Zhang as co-first authors, started from a well-established observation: genetically supported drug targets roughly double the probability of success in drug development. Alzheimer’s disease is highly heritable, with heritability estimates of about 90 percent for early-onset disease and 58 to 79 percent for late-onset disease. Yet the enormous volumes of genetic and genomic data generated by projects such as the Alzheimer’s Disease Sequencing Project and the Alzheimer’s Disease Neuroimaging Initiative have not been fully exploited to guide therapeutic development. The team set out to change that by screening 1,229 druggable proteins, defined as proteins targeted by FDA-approved drugs or investigational molecules with binding affinity below one micromolar.

The core statistical engine of the study was Mendelian randomization, a technique that uses naturally occurring genetic variants as instrumental variables to test whether altered levels of a protein or gene product causally influence disease risk, rather than merely correlating with it. The researchers drew instrumental variables from three protein quantitative trait locus datasets and nine expression quantitative trait locus datasets spanning five brain regions, including resources from the Religious Orders Study and Memory and Aging Project, the MetaBrain consortium and the Mayo Clinic Biobank. These were tested against seven Alzheimer’s genome-wide association study datasets comprising 275,540 cases and 1.55 million controls of European and African ancestry. Five complementary Mendelian randomization models were applied to strengthen rigor, and the team distilled the results into a consensus metric they call the alzMR score, which aggregates evidence across datasets and methods. Simulations and a Cauchy combination test confirmed that the score reliably ranks genes by the strength of their causal evidence, even when participant cohorts partially overlap.

The analysis yielded 19 high-confidence drug targets for individuals of European ancestry and seven suggestive targets for African ancestry. Targets were classified as antagonistic, where inhibition could be beneficial, or agonistic, where activation could help, based on the direction of the genetic signal. Among the strongest antagonistic targets were EPHX2, EGFR, PRKCB, FOLH1 and ADAM10, while ACE and eukaryotic translation initiation factor 4E emerged as leading agonistic targets. Colocalization and fine-mapping analyses, which test whether the same genetic variant drives both protein levels and disease risk, lent further support to targets including ACE, ADAM10, CTSH and EGFR. Encouragingly, the method independently recovered targets already under clinical investigation, such as BACE2 and APH1B in the amyloid pathway and CHRNE and GABBR1 in neuromodulation, suggesting the pipeline captures biologically meaningful signals. Notably, many of the prioritized targets act on inflammation, metabolism and vascular pathways, extending well beyond the classical amyloid and tau biology that has dominated the field.

The standout target was EPHX2, which encodes soluble epoxide hydrolase, an enzyme that breaks down protective lipid metabolites called epoxy fatty acids and promotes neuroinflammation and synapse damage. Elevated EPHX2 protein levels in the brain were strongly associated with increased Alzheimer’s risk across multiple datasets. Crucially, the team identified a protective missense variant, p.Arg287Gln, carried at the rs751141 locus, which reduces EPHX2 protein levels and protects against the disease. Functional work in induced pluripotent stem cell-derived neurons and cerebral organoids from patients with Alzheimer’s showed that this variant lowers levels of phosphorylated tau, a hallmark pathology. Using prime editing to introduce the protective mutation directly into patient cells, the researchers confirmed that it reduces EPHX2 protein and activates neuroprotective, neurogenesis and anti-inflammatory gene programs, including the PI3K-Akt pathway that restrains GSK-3β, a major tau kinase.

To test whether pharmacological inhibition could deliver the same benefit, the team treated 5xFAD mice, a transgenic model of Alzheimer’s disease, with EC5026, a picomolar-potency EPHX2 inhibitor already in phase I trials. Over eight months of treatment, the drug significantly reduced amyloid-beta plaque burden and microglia-associated inflammation, enhanced lysosomal function and preserved synapse-related pathways in excitatory neurons and oligodendrocytes, as revealed by single-nucleus RNA sequencing. Most strikingly, treated mice performed better on tests of learning and memory, including novel object recognition, contextual fear conditioning and the Morris water maze. Pharmacokinetic analysis confirmed that EC5026 reached therapeutic concentrations in the cortex with roughly 15 percent brain penetration, and mass spectrometry verified on-target enzyme inhibition. A phenome-wide scan of 624 disease phenotypes in the UK Biobank found no adverse associations with the EPHX2 genetic instruments, offering early reassurance about safety.

The third pillar of the study brought the findings into the clinic through drugome-wide association studies, a pharmacoepidemiologic approach that asks whether people taking drugs against genetically prioritized targets show altered disease incidence. Drawing on electronic health data from 111,680 patients with mild cognitive impairment, the team evaluated 210 frequently prescribed drugs using four complementary exposure models with propensity score matching and adjustment for age, sex, race and comorbidities. Twelve drugs were consistently associated with reduced Alzheimer’s incidence, including trazodone, an ADRA1A antagonist linked to a 14 percent risk reduction in a dose-dependent manner and a 22 percent reduction against the active comparator escitalopram in an independent claims database, as well as baclofen, a GABBR1 agonist, and amlodipine, a CACNA1D inhibitor associated with a 12 percent risk reduction. Half of these drugs acted on targets with significant alzMR scores, providing converging genetic and real-world evidence.

The study also made a deliberate effort to address a long-standing gap in Alzheimer’s research: the underrepresentation of people of African ancestry, who face different disease prevalence, risk factors and clinical courses yet have been largely excluded from genetic studies and trials. Using an African ancestry GWAS dataset of 2,784 cases and 5,222 controls, the team identified seven suggestive targets, including DHFR, SLC7A11, TRPV3, ABL1 and SETD7 as antagonistic targets and ALDH2 and PPARG as agonistic targets, with more than half tied to metabolic pathways. DHFR expression rose with Braak stage and cognitive decline, and methotrexate, an approved DHFR inhibitor, was associated with reduced Alzheimer’s risk in African ancestry patients with mild cognitive impairment in a dose-dependent fashion, but not in European ancestry patients. The authors argue that such ancestry-specific findings are scientifically and medically essential for populations underserved by existing genetic datasets, and they plan to extend the approach using the All of Us Research Program, which includes 45 percent racial and ethnic minorities.

The authors are candid about limitations. Overlapping participants across European ancestry GWAS cohorts can inflate statistical evidence, a problem they mitigated with the Cauchy combination test, and diagnoses in real-world claims data are rarely confirmed by PET imaging or cerebrospinal fluid biomarkers. Individual variation, including APOE genotype and the protective EPHX2 variant itself, will likely modulate drug response and should inform patient stratification, much as APOE ε4 homozygosity shapes the risk of amyloid-related imaging abnormalities with lecanemab. Even so, the study delivers something the field has lacked: a compendium of drug targets backed simultaneously by human genetics, experimental validation and population-scale clinical data, with EPHX2 standing out as a genetically indicated, experimentally confirmed anti-inflammatory target whose inhibitors already have established safety profiles. If the framework scales as promised, the long, costly road from Alzheimer’s genetic discovery to effective medicine may finally be getting a reliable map.

Subject of Research: Identification and validation of druggable Alzheimer's disease targets using Mendelian randomization, multiomics data and electronic health records

Article Title: Alzheimer’s disease target and drug discovery by leveraging multiomics and electronic health data

Article References: Hou, Y., Li, Y., Zhang, P., Lorincz-Comi, N., Gohel, D., Fan, F., Qiu, Y., Yang, J., Chen, X., Song, W., Yang, X., Tan, Z., Liu, Z., Fang, X., Rivera Paz, I., Martin, W., Feng, Y., Zhou, Y., Xu, J., … Cheng, F. (2026). Alzheimer’s disease target and drug discovery by leveraging multiomics and electronic health data. Nature Neuroscience. https://doi.org/10.1038/s41593-026-02472-0

Image Credits: AI Generated

DOI: 10.1038/s41593-026-02472-0

Keywords: Alzheimer's disease, Mendelian randomization, drug repurposing, EPHX2, multiomics, electronic health records, GWAS, neuroinflammation, tau pathology, precision medicine, drug targets, 5xFAD mice

The study, led by Wanda Bi, Jianxin Jiang, Saiying Hou and Li Li, combined an unusually broad arsenal of approaches: patient transcriptomic datasets, a machine-learning pipeline spanning twelve algorithms, single-cell and spatial transcriptomics, molecular docking simulations, and classical animal and cell experiments. The researchers began by treating mice in which pulmonary fibrosis had been induced with bleomycin, a chemotherapy drug that reliably scars the lungs. Starting one day after injury, the animals received intraperitoneal injections of uridine every 48 hours for three weeks. The results were striking. Survival at day 21 rose from 40 percent in untreated fibrotic mice to 84.6 percent in those given the higher, 4 mg/mL dose of uridine. Body weight recovered, lung architecture was better preserved, and collagen deposition—the hallmark of fibrosis—fell dramatically, with the collagen volume fraction dropping from 47.79 percent to 12.22 percent of tissue area.

Crucially, the treatment appeared safe. Serum biochemistry measuring kidney and liver function, along with histological examination of the kidney, liver, spleen, heart and lung, revealed no detectable toxicity. That safety profile matters because uridine is an endogenous molecule, already present in the body as a central player in pyrimidine metabolism and RNA biosynthesis, and prior work has suggested it carries regenerative and anti-inflammatory properties across multiple organs. But knowing that a molecule helps is not the same as knowing how, and this is where the study’s computational machinery came into play.

The team mined a public dataset of interstitial lung disease transcriptomes, corrected for batch effects, and identified more than a thousand genes altered in diseased lungs. Weighted gene coexpression network analysis then grouped these genes into modules correlated with clinical traits such as lung diffusion capacity and disease status. In parallel, three pharmacology databases—ChEMBL, SwissTargetPrediction and PharmMapper—were used to predict which human proteins uridine might physically engage. Intersecting the two lists yielded 28 candidate targets, and a machine-learning framework that cross-combined feature-selection and classification algorithms into 128 distinct models whittled the field down to five core genes: MMP7, CA4, MME, PTPRB and TYMS. Each of these individually distinguished patients from healthy controls with an area under the curve above 0.85, and a diagnostic nomogram built from all five achieved an AUC of 0.925 in validation cohorts.

MMP7 stood out. Interpretability analysis using SHAP values ranked it as the most influential predictor in the model, and its expression correlated strongly with clinical reality: the more MMP7 a patient expressed, the worse their diffusing capacity for carbon monoxide, and patients with high MMP7 in bronchoalveolar fluid survived significantly less long, with a hazard ratio of 2.93. Single-cell RNA sequencing of more than half a million cells from 127 human lung samples localized the signal with precision. MMP7 was highly expressed in basal and secretory epithelial cells, and its expression in secretory cells climbed steadily as fibrosis advanced from limited to severe. Spatial transcriptomics of human lung tissue reinforced the picture, showing that the proportion of MMP7-positive spots correlated tightly with histologic fibrosis scores, and that MMP7-rich hotspots colocalized with regions of dense immune infiltration, particularly macrophages, monocytes and neutrophils.

That spatial coupling between MMP7 and immune cells is the conceptual heart of the paper. MMP7 has long been regarded mainly as a biomarker of extracellular matrix degradation in IPF, but the new data suggest it may do something more active: reshape the immune microenvironment. Cell-cell communication analysis using CellChat showed that, in fibrotic lungs, secretory cells dramatically increase their signaling with dendritic cells, B cells, macrophages and T cells, activating pathways involved in antigen presentation, integrin-mediated adhesion and cytotoxic T-cell responses. In the bleomycin model, MMP7-positive secretory cells sat in close proximity to CD45-positive immune cells, F4/80-positive macrophages and Ly6G-positive neutrophils, and these populations expanded dramatically after injury.

Could uridine directly engage MMP7? The team built the case at the molecular level. Molecular docking predicted a favorable binding energy of −6.5 kcal/mol between uridine and MMP7, and a 100-nanosecond molecular dynamics simulation showed a stable complex with persistent hydrogen bonds and the residue GLU219 emerging as an energetic hotspot. Surface plasmon resonance on a Biacore system then confirmed concentration-dependent binding between uridine and recombinant human MMP7, with an equilibrium dissociation constant of 6.73 micromolar. A biotin-labeled uridine pull-down assay in human bronchial epithelial cells enriched MMP7, and free uridine competed that signal away—three independent lines of evidence for a physical interaction.

Functional experiments then connected the dots. In BEAS-2b airway epithelial cells, bleomycin raised MMP7 protein levels, and uridine brought them back down. When the researchers knocked down MMP7 with siRNA, the pro-inflammatory cytokine macrophage migration inhibitory factor, or MIF, fell while midkine, or MDK—a molecule associated with tissue repair—rose. Uridine produced the same directional shift, and adding recombinant MMP7 back into the cultures partially reversed uridine’s effects on MIF and antigen-presentation genes. The authors are careful to frame this as a hypothesis-generating mechanism rather than a fully resolved causal chain, but the pattern was consistent: MMP7 appears to bias the epithelial niche toward inflammatory signaling, and uridine seems to tip it back toward repair. In living mice, uridine treatment cut the total immune cell population in bronchoalveolar lavage fluid by roughly 80 percent, with significant reductions in neutrophils, macrophages and T cells, while messenger RNA levels of IL-1β, IL-6 and TGF-β1 all dropped. Bulk RNA sequencing of treated lungs confirmed dampened inflammatory, innate immune and adaptive immune signatures, and Mmp7 expression positively tracked with neutrophil and macrophage infiltration scores.

The translational implications are intriguing but guarded. Broad-spectrum matrix metalloproteinase inhibitors have historically stumbled over toxicity, which makes the idea of modulating a single MMP through an endogenous, well-tolerated metabolite attractive. Because MMP7 is also implicated in kidney fibrosis, the approach could theoretically extend beyond the lung. Yet substantial hurdles remain. Circulating uridine is rapidly degraded in the liver by uridine phosphorylase, severely limiting oral bioavailability, which is why the team chose injection; whether effective pulmonary exposure can be achieved clinically, and by which route, is unknown. The study used an early intervention window, whereas patients typically present with established disease, and the authors themselves note that definitive proof will require epithelial-specific MMP7 knockout models, humanized systems such as precision-cut lung slices from IPF patients, and careful dose-ranging work. Some discrepancies—such as divergent PTPRB patterns between mice and humans, and an unexplained rise in B cells despite MMP7 suppression—underscore that parallel pathways are at play.

Even with those caveats, the convergence of evidence is what makes the paper notable. A candidate molecule emerged not from a single screen but from multiple independent analytical layers—coexpression networks, machine learning, single-cell atlases, spatial maps, structural simulations and wet-lab validation—all pointing at the same node. If future work confirms that uridine, or derivatives engineered to survive first-pass metabolism, can safely suppress MMP7-high epithelial programs in human lungs, it could open a genuinely new front against a disease that has stubbornly resisted them. For now, the study offers something IPF research has badly needed: a testable, mechanistically grounded hypothesis linking metabolism, epithelial injury and immune remodeling in the fibrotic lung.

Subject of Research: Uridine as a candidate modulator of MMP7-mediated immune cell infiltration in pulmonary fibrosis

Article Title: Integrative multiomics and machine learning identify uridine as a candidate modulator of MMP7-mediated immune cell infiltration in pulmonary fibrosis

Article References: Bi, W., Zhou, Y., Gong, J., Liu, W., Li, J., Du, J., Nie, H., Luo, H., Zhang, H., Cai, Q., Gao, C., Zeng, L., Jiang, J., Hou, S., & Li, L. (2026). Integrative multiomics and machine learning identify uridine as a candidate modulator of MMP7-mediated immune cell infiltration in pulmonary fibrosis. Journal of Advanced Research. https://doi.org/10.1016/j.jare.2026.10.008

Image Credits: AI Generated

DOI: 10.1016/j.jare.2026.10.008

Keywords: pulmonary fibrosis, idiopathic pulmonary fibrosis, uridine, MMP7, multiomics, machine learning, single-cell RNA sequencing, spatial transcriptomics, immune cell infiltration, molecular docking, bleomycin model, antifibrotic therapy

The core problem Hydra tackles is one that has haunted computational biologists for years. Single-cell datasets are sparse and noisy, with each cell described by thousands of molecular features, most of which are irrelevant to distinguishing one cell type from another. This high dimensionality buries the key molecular signatures that define cellular identity. Feature selection, the process of identifying which genes or genomic regions truly matter, has therefore become a critical preprocessing step. But existing tools struggle on three fronts: most are built exclusively for transcriptomic data and cannot handle newer multimodal assays; they systematically overlook small cell populations; and the deep learning methods that do perform well tend to operate as inscrutable black boxes.

Hydra’s architecture is built around an ensemble of variational autoencoders, or VAEs, a class of neural networks that learn compressed probabilistic representations of complex data. Each VAE in the ensemble is paired with a cell type classification head, and the whole system is trained jointly with a loss function that balances reconstructing the input data against correctly classifying cell types. The framework operates through two connected modules. The first performs ensemble feature ranking, producing a consensus list of cell-type-specific markers. The second is an annotation module that deploys an ensemble of simple neural network classifiers, trained on the selected features, to automatically assign cell type labels to new query datasets.

The most ingenious element of the design is how Hydra confronts class imbalance, the persistent bias that causes algorithms to favor abundant cell types while ignoring rare ones. Because a VAE learns the probability distribution underlying the data, it can generate synthetic cells that faithfully mimic underrepresented populations. Hydra combines this generative augmentation with random downsampling of dominant cell types, producing balanced training sets for each member of the ensemble. Each refined model is then interrogated using Integrated Gradients, a post hoc attribution technique that traces each prediction back to the individual input features that drove it, accumulating gradients along a path from a baseline input to the actual data point.

The ensemble approach proved decisive for reliability. When the researchers perturbed lung transcriptomic datasets through stratified subsampling and measured the consistency of feature importance scores using Pearson correlations, ensemble models were markedly more stable than single models. Stability improved with ensemble size up to 25 members, after which gains plateaued, so the team adopted 25 as the default configuration. Integrated Gradients also outperformed three alternative attribution methods, Saliency, GradientSHAP, and DeepLIFT, delivering higher feature stability and lower variability across all cell types, a property the authors argue is essential for identifying reproducible markers that generalize across studies and sequencing platforms.

Benchmarking was extensive. The team evaluated Hydra on 21 datasets spanning unimodal and multimodal single-cell technologies, comparing it against 13 state-of-the-art methods. On a subsampled Mouse Cell Atlas containing 20 cell types with a severe 100-to-2 imbalance between major and minor populations, Hydra’s selected features clustered biologically related cell types together, correctly grouping naive B cells with late pro-B cells and classical monocytes with promonocytes, distinctions that statistical methods such as Welch’s t test, the Wilcoxon rank-sum test, and Limma-Voom failed to make. For kidney proximal convoluted tubule epithelial cells, the top five genes Hydra identified, including GPX3, TIMP3, and FTH1, were all highly and specifically expressed in that population.

The functional relevance of these selections was confirmed through gene ontology enrichment analysis. The top features for naive B cells were enriched for B-cell activation and B-cell receptor signaling, neutrophil features pointed to migration, chemotaxis, and inflammatory response programs, and mesenchymal cell features captured extracellular matrix organization and skeletal system development. In cell type prediction tasks across 13 transcriptomic datasets, Hydra achieved the highest balanced accuracy in both intra-dataset validation on prostate urethra and colon data, at 68.86 percent and 86.90 percent respectively, and in inter-dataset benchmarking across 22 train-test pairs spanning kidney, lung, peripheral blood mononuclear cells, and retina, where it reached 73.81 percent balanced accuracy and a 66.77 percent macro F1-score.

Hydra’s multimodal capabilities may prove even more consequential. Modern assays can simultaneously measure gene expression, chromatin accessibility, and surface protein abundance in the same cell, and Hydra’s architecture processes each modality through dedicated encoder layers before merging them in a shared latent space of 100 neurons. Across seven multiome technologies, including SHARE-seq, SNARE-seq, sciCAR, CITE-seq, and the trimodal TEA-seq, Hydra outperformed seven competing integration methods, among them MOFA+, totalVI, MultiVI, scGLUE, scJoint, UMINT, and scMoMaT, achieving the highest balanced accuracy and macro F1-score in both intra-dataset and inter-dataset evaluations across 28 train-test splits.

The framework’s most striking demonstration came in Alzheimer’s disease. Using a previously published dataset profiling the transcriptome and epigenome of the medial frontal cortex, the team trained Hydra on healthy brain tissue containing 27 distinct cell populations and asked whether it could transfer those annotations to diseased tissue, where molecular changes can obscure cellular identity. Hydra maintained balanced accuracy of roughly 85 percent on healthy held-out samples, about 85 percent on early-stage Alzheimer’s samples, and 84 percent on late-stage samples. Critically, it preserved disease-relevant signals: its predictions captured cell type proportion shifts between early and late disease stages and maintained strong Spearman correlations with differential gene expression signatures derived from expert annotations, including for rare populations that competing methods failed to resolve.

Practicality matters too, and Hydra completes training in under ten minutes even on datasets of ten thousand cells, though its ensemble design does demand higher peak GPU memory than baseline methods. The authors are candid about limitations: as a supervised framework, Hydra can only predict cell types present in its training reference and cannot discover novel populations, and its performance depends on the quality of reference labels. They also note that multiome benchmarks relying on RNA-derived ground truth may understate the value of methods that genuinely integrate additional modalities. Even so, with code and documentation publicly available through the Sydney BioX repository, Hydra arrives at a moment when single-cell multiomics is becoming routine, offering researchers a tool that is simultaneously powerful, balanced toward the rare cells that often matter most, and interpretable enough to trust.

Subject of Research: Interpretable deep generative ensemble learning for feature selection and cell type annotation in single-cell omics

Article Title: Interpretable deep generative ensemble learning for single-cell omics with Hydra

Article References: Wagle, M. M., Liu, C., Liu, Z., Wang, Y., Kellis, M., Patrick, E., & Yang, P. (2026). Interpretable deep generative ensemble learning for single-cell omics with Hydra. Molecular Systems Biology, 22(7), 1161-1179. https://doi.org/10.1038/s44320-026-00208-7

Image Credits: AI Generated

DOI: 10.1038/s44320-026-00208-7

Keywords: single-cell omics, deep learning, variational autoencoder, cell type annotation, feature selection, multimodal integration, Integrated Gradients, class imbalance, Alzheimer's disease, rare cell populations, multiomics, ensemble learning

The scale and breadth of the data collection is what sets this study apart. Rather than measuring a handful of immune markers, the team assembled four complementary omics layers from the same longitudinal samples: transcriptomics, which captures gene expression patterns in circulating immune cells; metabolomics, which profiles the small molecules that fuel and regulate cellular chemistry; lipidomics, which maps the fat-based signaling molecules and membrane components that shift during inflammation; and cytokine measurements, which quantify the soluble messengers that immune cells use to communicate. Integrating these layers allowed the investigators to move beyond the question of whether a vaccine produces antibodies and toward a systems-level picture of how vaccination reorganizes host physiology. This approach, often called systems vaccinology, treats the immune response not as an isolated event but as a coordinated, body-wide program of gene regulation, metabolism and cellular signaling.

The headline finding is that the two vaccine platforms produce pronounced and persistent differences in their multiomic fingerprints. BNT162b2 induced far more profound changes across the omics landscape than CoronaVac, with particularly marked alterations in adaptive immunity and in metabolomic function. That is consistent with the mechanistic distinction between the platforms: an mRNA vaccine delivers genetic instructions that cause the recipient’s own cells to manufacture the SARS-CoV-2 spike protein, triggering a potent innate sensing response and driving strong T-cell help and B-cell maturation. An inactivated virus vaccine, by contrast, presents a killed, non-replicating virus that primarily stimulates antibody production through more contained innate signaling. The molecular consequences of those different entry points into the immune system, the study shows, ripple outward into gene expression programs and metabolic pathways that remain measurable for months.

Perhaps the most striking observation is durability. At six months after the second dose, recipients of BNT162b2 still displayed a durable immunological imprint, characterized by persistent modifications in immune gene expression and ongoing metabolic reprogramming. In other words, the mRNA vaccine did not simply provoke a transient burst of activity that faded back to baseline; it left a lasting signature in the transcriptome and metabolome of vaccinated individuals. The authors link this prolonged imprint to the vaccine’s known immunogenicity and to the durability of protection it confers. CoronaVac recipients, meanwhile, showed a more subdued and less persistent omics response, which aligns with clinical observations that inactivated-virus primary series generally elicit lower and more rapidly waning neutralizing antibody levels than mRNA vaccines, particularly against evolving variants of concern.

One of the study’s technical contributions lies in its analysis of crosstalk between omics layers. The researchers found substantially greater interplay between transcriptomic, metabolomic and lipidomic features in the BNT162b2 group, suggesting that the mRNA platform engages a more integrated, multi-system response. Crosstalk of this kind matters because immune outcomes rarely depend on a single molecule. Gene expression changes drive metabolic shifts; metabolites in turn feed back on signaling pathways that regulate cytokine production and lymphocyte differentiation. A vaccine that coordinates these layers coherently may be more effective at shepherding naive immune cells through the full program of activation, clonal expansion, germinal center reaction and memory formation. The enhanced crosstalk observed after BNT162b2 vaccination, the authors argue, is related to the platform’s stronger immunogenicity and longer-lasting protection.

Beyond describing group-level differences, the team asked whether a person’s baseline molecular state—measured before any vaccination—could predict how well they would respond to a given vaccine. Using predictive modeling on the multiomics data, they found that pre-vaccination profiles could forecast vaccine-specific performance with area under the curve values of 0.73 to 0.79. In practical terms, that means the metabolic and transcriptional state of an individual before vaccination carries meaningful information about the strength and character of the immune response they will mount. An AUC in this range is well above chance and, while not yet accurate enough for standalone clinical decision-making, it demonstrates that the concept of predicting vaccine responsiveness from a baseline blood draw is scientifically viable. It also hints at why vaccine responses vary so widely between individuals: differences in baseline metabolism, immune tone and inflammatory state shape how the same vaccine is interpreted by different bodies.

The study’s setting gives it particular epidemiological weight. Hong Kong was one of the few places where mRNA and inactivated-virus vaccines were deployed side by side in large numbers during the primary vaccination campaign, creating a natural comparative cohort that would have been difficult to assemble elsewhere. Because both groups lived through the same pandemic conditions, were exposed to the same circulating variants and were recruited through the same health system, confounding from environment and exposure history is reduced. The 553-participant cohort, with samples spanning half a year, therefore provides an unusually clean head-to-head comparison of the two dominant vaccine platforms used across Asia and much of the world, where CoronaVac and similar inactivated vaccines reached billions of arms.

The implications extend beyond COVID-19. The finding that mRNA vaccination produces a prolonged metabolic reprogramming and persistent immune gene modification adds to a growing body of literature on trained immunity, the concept that innate immune cells undergo long-term epigenetic and metabolic rewiring after stimulation, altering their responsiveness to future challenges. If mRNA vaccines induce a durable trained-immunity-like state, that could partly explain their robust performance and could inform the design of future mRNA vaccines against influenza, respiratory syncytial virus, cancer and other targets. Conversely, understanding the molecular reasons why inactivated vaccines elicit narrower and less durable signatures could guide adjuvant choices or prime-boost regimens that combine platforms to broaden and prolong protection.

The authors are explicit about the translational direction of the work. They advocate integrating multiomics data to tailor vaccination strategies to individual immune profiles, with the goal of enhancing public health outcomes. In a future shaped by this kind of systems vaccinology, a pre-vaccination blood test could, in principle, identify who is likely to respond poorly to a standard regimen and who might benefit from an alternative platform, an earlier booster or an adjusted dose. The predictive models reported here, with their moderate but real discriminatory power, are early proof of concept for that vision. Before such tests reach the clinic, they would need validation in independent cohorts, standardization of assays across laboratories and demonstration that acting on a prediction actually improves outcomes. The cost of multiomics profiling, while falling, also remains a barrier to population-scale deployment.

Limitations should temper interpretation of the results. The study captured the primary two-dose series and did not follow participants through boosters or breakthrough infections, both of which reshape the omics landscape further. Multiomics datasets are high-dimensional, and associations between molecular signatures and protection, however biologically plausible, are correlational rather than proof of mechanism. The published version of the article is also subject to final editorial processing, as it was shared early under Springer Nature’s open-access policy with a permanent DOI. Even so, the study stands as a landmark demonstration that two vaccines against the same pathogen, both highly effective at reducing severe disease, inscribe fundamentally different and enduring patterns on human biology. As vaccine development accelerates for emerging pathogens, the lesson is clear: the platform matters not only for how strong the response is, but for what kind of molecular memory it leaves behind.

Subject of Research: Longitudinal multiomic comparison of host immune responses to mRNA and inactivated virus COVID-19 vaccines

Article Title: Differential prolonged multiomic responses to mRNA and inactivated virus COVID-19 vaccines

Article References: Tun, H. M., Zhao, S., Chen, C., Chen, Y., Peng, Y., Lv, H., Zhu, J., Tang, Y. S., Chan, K. K. P., Ip, B. Y., Sun, Y., Liu, X., Yiu, K., Hui, D. S. C., & Mok, C. K. P. (2026). Differential prolonged multiomic responses to mRNA and inactivated virus COVID-19 vaccines. Cellular and Molecular Life Sciences. https://doi.org/10.1007/s00018-026-06455-z

Image Credits: AI Generated

DOI: 10.1007/s00018-026-06455-z

Keywords: COVID-19 vaccines, mRNA vaccine, BNT162b2, CoronaVac, inactivated virus vaccine, multiomics, transcriptomics, metabolomics, lipidomics, cytokines, systems vaccinology, immune durability

The core idea is elegantly simple. The researchers designed photocleavable hashtag oligonucleotides, or PHOs, short synthetic DNA molecules conjugated to cholesterol so that they insert themselves into the fatty membranes of living cells. Each PHO carries two distinct barcode sequences, an inner and an outer tag, joined through a chemical bond that snaps apart when illuminated with ultraviolet light. Because the tags sit on the cell surface, they do not perturb the cell’s internal biochemistry, and because they are cleaved by light, they can be erased selectively wherever a researcher chooses to shine a beam.

To convert this chemistry into spatial information, the team used a digital micromirror device, the same light-engine found in many projectors, to cast user-defined patterns of ultraviolet light onto cultures of PHO-labeled cells. Cells bathed in light lose their outer tag; cells kept in shadow retain it. After several minutes of patterned exposure, the researchers dissociate the sample and run it through standard single-cell sequencing workflows, including droplet-based platforms such as 10x Genomics. In the resulting data, the ratio of outer to inner barcode reads recovered from each cell encodes how much light that cell received, and therefore where it sat in the original culture. In effect, the light pattern is stamped onto the cells as a molecular memory.

The precision of this stamping is tunable. Because photocleavage scales monotonically with light dose, a smoothly graded illumination pattern produces a continuous gradient of barcode cleavage across the sample, allowing cells to be assigned positions along an axis rather than sorted into coarse bins. Sequential rounds of labeling push resolution further: by applying different photomasks in succession with distinct PHO species, each carrying its own fluorophore and barcode, the team generated combinatorial codes that distinguish multiple spatial quadrants within a single dish. A refined second-generation tag, held in place by bridging oligonucleotides that prevent cleaved fragments from washing away, preserves earlier labeling patterns through subsequent rounds, expanding the encoding capacity exponentially with each additional exposure.

Critical to any method that douses living cells with ultraviolet light is demonstrating that the light itself does not distort the biology being measured. The researchers addressed this directly. Comparing transcriptomes of illuminated and non-illuminated cells, they found correlations approaching unity, with principal component analysis showing no separation between the two groups and differential expression testing revealing essentially no genes significantly altered by saturating UV exposure. They also showed that PHOs linger on cell membranes with a half-life of roughly nine hours, defining a practical window for live-cell experiments, and that the method works equally well on fixed samples, broadening its applicability to precious clinical specimens that cannot be kept alive.

Where scSTAMP-seq truly distinguishes itself is in its modularity. Rather than reading out gene expression alone, the team extended the approach to scSTAMP-MAT-seq, which jointly profiles messenger RNA, DNA accessibility, and DNA methylation from the same individual cell, all while retaining the spatial barcode. The workflow combines the light-stamped tags with a methyltransferase that marks open chromatin and a methylation-sensitive restriction enzyme that fragments the genome at methylated sites. Benchmarking showed that the multiomic version recovers tens of thousands of accessible GpC sites and endogenous methylated CpG sites per cell, comparable to the non-spatial predecessor, meaning spatial resolution came at little cost to data quality.

To showcase the method’s power, the researchers applied it to two-dimensional human gastruloids, micropatterned colonies of induced pluripotent stem cells that recapitulate the spatial patterning of the early embryo. During gastrulation, the embryonic disc organizes itself into domains fated to become different germ layers, and understanding how position and gene regulation intertwine during this process is a central question in developmental biology. Using a radial light gradient that decayed nonlinearly from the colony edge toward the center, the team encoded continuous positional information across each gastruloid and then read out the transcriptome of thousands of individual cells.

The resulting spatial maps revealed that canonical lineage markers such as CDX2, SOX17, EOMES, and DNMT3B each occupied characteristic radial positions, reproducing the epiblast-like, primitive streak-like, and extraembryonic-like domains seen in real gastrulating embryos. More strikingly, when the researchers grouped genes by their spatial expression profiles and examined the behavior of chromatin regulators, the enzymes and binding proteins that chemically modify and interpret histones and DNA, they found that these regulators themselves were spatially patterned. Histone editors, histone readers, DNA modifiers, ATP-dependent remodelers, and Polycomb group proteins each showed distinct distributions across the radial axis, suggesting that the epigenetic machinery governing cell fate is itself organized by position within the emerging tissue.

This observation speaks to a long-standing challenge in genomics. Studies from the ENCODE project and numerous single-cell atlases have catalogued gene expression and epigenetic states across cell types, but disentangling whether epigenetic differences drive spatial organization or merely reflect it requires measurements that capture both simultaneously in an intact spatial context. By coupling spatial barcoding with joint multiomic readout, scSTAMP-seq offers a way to ask these questions directly, correlating chromatin state, DNA methylation, and transcription as functions of position within a developing or diseased tissue.

The practical accessibility of the method may prove as influential as its scientific results. Unlike purpose-built spatial transcriptomics instruments that require specialized arrays or imaging systems, scSTAMP-seq piggybacks on standard single-cell sequencing platforms and a digitally addressable light source found in many microscopy facilities. The team has deposited its sequencing data in the Gene Expression Omnibus and released its analysis code on GitHub, lowering the barrier for other laboratories to adopt the approach. With a patent application pending and demonstrated compatibility with both live and fixed cells, plate-based and droplet-based workflows, and transcriptomic plus epigenomic readouts, light-stamped barcoding stands to bring spatial reasoning to single-cell genomics labs that could never justify a dedicated spatial platform, potentially reshaping how developmental biology, neuroscience, and cancer research map the molecular landscapes of complex tissues.

Subject of Research: A light-encoded barcoding method for spatially resolved single-cell transcriptomics and epigenomics

Article Title: Photolabile oligonucleotides with topological light gradients enable spatially resolved single-cell transcriptomics and epigenomics

Article References: Piscopio, R. A., Chialastri, A., Wang, C., Godzik, M., Heom, K. A., Wang, W., Li, L. J., Saxena, N., Wilson, M. Z., & Dey, S. S. (2026). Photolabile oligonucleotides with topological light gradients enable spatially resolved single-cell transcriptomics and epigenomics. Nature Biotechnology. https://doi.org/10.1038/s41587-026-03328-5

Image Credits: AI Generated

DOI: 10.1038/s41587-026-03328-5

Keywords: scSTAMP-seq, spatial transcriptomics, single-cell sequencing, photocleavable oligonucleotides, epigenomics, DNA methylation, chromatin accessibility, gastrulation, human gastruloids, digital micromirror device, multiomics, Nature Biotechnology

Mass spectrometry imaging has become one of the most powerful tools in modern biology because it can map hundreds of molecules directly within a tissue slice, revealing where lipids, metabolites, proteins and other biomolecules reside without labels or stains. Yet the technique has long been constrained by its spatial resolution. Conventional instruments raster a laser across a tissue section with a defined step size, and features smaller than that step blur together. Boosting resolution usually demands costly instrumentation such as specialized lasers, transmission-mode optics or next-generation ion optics, placing subcellular molecular imaging out of reach for many laboratories.

TEMI takes a different path. Instead of refining the instrument, the team enlarges the tissue itself. Drawing on principles from expansion microscopy, a technique introduced more than a decade ago, the researchers embed tissue in a water-rich hydrogel polymer network. When the gel swells, it carries the tissue along with it, spreading the molecules apart and effectively magnifying the sample before any mass spectrometry takes place. A laser raster that would have been too coarse for the original tissue becomes fine enough to resolve individual cells in the expanded version, yielding more than a 3.5-fold improvement in effective imaging resolution with standard equipment.

The critical challenge was chemistry. Traditional expansion protocols often rely on harsh denaturation conditions, including high heat and strong detergents, to homogenize tissue and allow gels to stretch uniformly. Those conditions would destroy the very molecules mass spectrometry aims to detect. The TEMI workflow solves this by using a re-embedding strategy that expands tissue under mild, harsh-condition-free conditions, preserving the chemical integrity of lipids, metabolites, N-glycans, peptides and proteins. The result is a sample that is both physically enlarged and chemically faithful to its original molecular composition.

The published protocol walks readers through every stage of the process. It begins with constructing a gelation chamber, then describes hydrogel-based tissue gelation and expansion, followed by cryosectioning of the expanded tissue-hydrogel composite. Because expanded samples are soft and hydrated, cutting them into thin sections requires careful handling, and the protocol provides the specific parameters that make reproducible sectioning possible. Detailed guidance covers matrix application, data acquisition and visualization pipelines, along with troubleshooting tips accumulated through the team’s extensive experimentation.

One of the most striking capabilities of TEMI is multiomics mapping on a single tissue section. The protocol describes a sequential workflow in which N-glycans are released in situ by treatment with the enzyme PNGase F, after which proteins are digested with trypsin and imaged, all following lipid and metabolite analysis on the same expanded section. This means a single slice of brain tissue can yield spatially resolved maps of multiple molecular classes, each anchored to the same anatomical context, an efficiency that conventional workflows struggle to match.

Quality control receives particular attention. The protocol includes a dedicated workflow for measuring deformation maps and expansion factors, which quantify how uniformly the gel has expanded. Uneven expansion would distort molecular maps, so the researchers provide computational tools, released through a public code repository, that quantify tissue expansion non-uniformity. They also demonstrate that analyte delocalization during sample preparation is minimal: lipid and small-metabolite signals were detected exclusively in tissue regions, with no corresponding signals in adjacent blank hydrogel areas, confirming that molecules stay where they belong during the swelling process.

The demonstration experiments showcase the method’s power on the mouse cerebellum, a tissue with exquisitely organized layers of cells. Comparing an unexpanded control cerebellum imaged with a 50-micrometer laser raster against a double-embedded, expanded sample at the same step size reveals a dramatic difference in molecular detail. Pushing further, the team performed TEMI after three cycles of gel embedding and expansion with a 10-micrometer raster step, resolving biomolecular heterogeneity that remains invisible in unexpanded tissue. The protocol also demonstrates mapping of small metabolites, N-glycans and proteins across cerebellar structures, illustrating the breadth of the multiomics capability.

The work builds on the team’s primary research paper published in Nature Methods, which introduced TEMI and established its foundational performance, and on a rich history of expansion microscopy methods developed by collaborators at the Howard Hughes Medical Institute’s Janelia Research Campus, including protein-retention expansion microscopy and ten-fold robust expansion. Raw datasets from the protocol’s figures are publicly available in the MassIVE repository, and the deformation-measurement code is hosted on GitHub, lowering barriers for laboratories that want to adopt or adapt the approach.

The implications extend across basic biology and medicine. Single-cell spatial resolution for untargeted molecular imaging could illuminate how metabolic gradients shape tissue function, how lipid compositions vary between neighboring cells in the nervous system, and how disease states such as cancer or neurodegeneration alter molecular architecture at scales previously accessible only to antibody-based imaging. Because TEMI requires no instrument modifications, it promises broad accessibility: laboratories with standard mass spectrometry imaging systems can now reach a resolution regime once reserved for a handful of specialized facilities. The protocol’s authors, spanning the University of Wisconsin-Madison and HHMI Janelia, and supported in part by the National Institutes of Health, position the method as a practical bridge between the spatial-omics revolution and the everyday mass spectrometry lab, inviting a wide research community to stretch its view of tissue, quite literally, into sharper focus.

Subject of Research: Tissue expansion combined with mass spectrometry imaging for high-spatial-resolution multiomics molecular mapping

Article Title: Tissue expansion mass spectrometry imaging (TEMI) for high-spatial-resolution multiomics molecular mapping

Article References: Tissue expansion mass spectrometry imaging (TEMI) for high-spatial-resolution multiomics molecular mapping. (n.d.). https://doi.org/10.1038/s41596-026-01427-w

Image Credits: AI Generated

DOI: 10.1038/s41596-026-01427-w

Keywords: mass spectrometry imaging, tissue expansion, TEMI, spatial omics, single-cell imaging, lipidomics, metabolomics, N-glycans, proteomics, hydrogel expansion, Nature Protocols, multiomics