AI News Archive: June 26, 2026 — Part 14
Sourced from 500+ daily AI sources, scored by relevance.
- Satisfanly
AI DM operating system for creators
- SnapSweep
AI screenshot cleanup that runs on your iPhone
- LuaCoder
Build Lua scripts faster with AI.
- Magggic
Ready-to-use AI image prompts and remixes
- Figma Motion
Figma Motion
- PaioClaw
PaioClaw: Secure Managed Platform for OpenClaw AI Agents PaioClaw simplifies OpenClaw deployment through a secure, fully managed platform. It removes the technical complexity and high infrastructure costs linked to self-managed environments. Private Clawspace With Fast OpenClaw Deployment Each account receives a private Clawspace designed for AI agents, complete with automatic updates, strict security controls, and […]
- Facefame
Facefame
- CinLink
CinLink
- VidTranslate
VidTranslate
- ImaginPrompt
ImaginPrompt
- TubeRepeat
TubeRepeat
- PhenoXtract: combining Large Language Model and Knowledge Graph embedding to extract phenotypes from clinical descriptions
Motivation: Standardized phenotypic descriptions are essential for accurate diagnosis, yet clinicians and researchers face challenges in manually extracting and mapping phenotypes from scientific literature or patient clinical records to the Human Phenotype Ontology. Recent advances in deep learning offer new opportunities for automation. We developed PhenoXtract, a novel phenotype extraction approach that combines Large Language Models and Knowledge Graph embedding. PhenoXtract is a multistep pipeline that takes clinical descriptions as input, extracts candidate phenotype entities using large language models, and maps them to terms from an enriched version of the Human Phenotype Ontology, processed as a knowledge graph. Results: Evaluation against expert-curated ground-truth datasets show a recall of 0.70 and precision of 0.85 for PhenoXtract, demonstrating concordance with manually extracted phenotypes, with a computation time of 10-20 seconds for each text analyzed. Moreover, PhenoXtract surpasses rule-based and deep learning-based state-of-the-art tools in two out of the three ground-truth datasets evaluated. These results suggest that hybrid approaches combining Large Language Models and Knowledge Graph embeddings represent a promising direction for automated clinical phenotyping at scale.
- Structural basis of multimodal adsorption and infection initiation by Vibrio phage Peru-2
Phage Peru-2, isolated during the 1993 cholera outbreaks in Peru, is distinct from the three ICP phage lineages typically associated with epidemic Vibrio cholerae. The molecular basis of Peru-2 adsorption and infection initiation has remained unknown. Here, we combine single-particle cryo-electron microscopy (cryo-EM) and cryo-electron tomography (cryo-ET) to define the architecture and infection mechanisms of Peru-2 at high resolution. The mature virion comprises an icosahedral capsid decorated with minor capsid proteins and a short tail apparatus surrounded by six structurally flexible tailspikes. These tailspikes are enzymatically active in mediating phage attachment to the Vibrio polysaccharide (VPS), a key component of biofilms. Three internal core proteins form a disordered core adjacent to the portal, positioning them for release before genome ejection during infection initiation. Structural analyses further resolve pre-ejection, genome-ejection, and post-ejection intermediates of the tail apparatus, while cryo-ET imaging of infected cells reveals a multimodal adsorption strategy during infection initiation.
- PlantGeneAnn: a strand-specific genome foundation model for ab initio gene structure annotation of plant genomes
High-quality plant genome assemblies are rapidly increasing, but accurate structural annotation remains reliant on transcript and homology evidence, limiting applications in newly sequenced and non-model species. Here, we present PlantGeneAnn, a plant-optimized, strand-specific genome foundation model for ab initio gene structure annotation. Fine-tuned on only nine high-quality model plant annotations, PlantGeneAnn outperformed a multi-species model trained on 42 species, showing that annotation quality is more important than token volume. On a stringent 13-species benchmark covering rosids, asterids, and monocots, PlantGeneAnn surpassed four state-of-the-art baselines across five evaluation levels, from base-level classification to complete transcript recovery. It achieved higher intron precision and better captured complex gene structures. In zero-shot variant effect prediction, PlantGeneAnn identified cryptic splice donors and premature stop codons in maize and rice, with saturation mutagenesis confirming single-nucleotide, context-dependent sensitivity. It also retained generalizability for epigenomic track prediction, highlighting its value for pan-genomics, crop improvement, and non-model plant research.
- Learning Perturbation Effects Through Contrastive Alignment of Multimodal Biological Embeddings
Multimodal single cell perturbation screens offer a scalable approach for characterizing the effects of genetic and chemical interventions on cellular state. However, most existing representation learning methods are tailored to a single perturbation modality and fail to explicitly incorporate external semantic knowledge, which limits their ability to generalize across datasets and perturbation types. Here, we introduce PertOmni, a CLIP style multimodal representation learning framework that aligns transcriptomic perturbation signatures with text derived embeddings of curated genes and compound descriptions, as well as image derived embeddings from cell paintings. PertOmni jointly trains a shared transcriptomic encoder and dataset specific text encoders using a masked contrastive objective that emphasizes within cell type discrimination while mitigating confounding effects arising from cell type heterogeneity. We evaluate the produced joint embedding space on bidirectional retrieval, drug gene interaction inference, and perturbation prediction across both small molecule and CRISPRi perturbation datasets, and demonstrate consistent improvements over strong baseline methods.
- Robust neural face identity codes in the Super-Recogniser brain
Super-Recognisers show exceptional ability in face recognition, providing a natural model of how perceptual systems optimise for individuating visually similar stimuli in variable viewing conditions. However, the neural representations supporting this extreme perceptual expertise are unknown. Here, we tested whether Super-Recognisers (n = 23) differed from typical recognisers (n = 21) in the dimensional organisation of neural face identity coding. We recorded 64-channel electroencephalography while participants viewed random and rapidly-presented sequences containing 10 naturally varying images of 40 unfamiliar identities. Using time-resolved representational similarity analysis we measured the geometry of identity representations, their consistency across observers, and how clearly they specified face identity. Although neural expression of identity information was robust in both groups, we found three key differences between Super-Recognisers and typical recognisers. First, the geometry of face identity representations differed between groups. Second, Super-Recognisers showed greater inter-individual consistency in representational geometry. Third, Super-Recognisers' neural signals discriminated between face identities more strongly than those of typical recognisers. Differences in the coding of broader face categories (sex, age, ethnicity) were notably weaker, suggesting that the observed group differences reflected fine-scale differences in identity coding rather than global reshaping of representational geometry. Strikingly, all three differences emerged within a common mid-latency interval (~300-500ms), implicating higher-stages of face processing associated with representations that are sensitive to face familiarity and link between perceptual and semantic domains. Together, these findings indicate that individual differences in face recognition ability reflect higher-level differences in neural identity coding, rather than enhanced early sensory processing.
- Consistent consensus-based annotation of spatial adaptive immune receptor repertoires from long-read sequencing using LongAIRR
The combination of spatial transcriptomics with long-read sequencing enables spatial characterization of full-length transcripts within solid tissue sections. However, standardized computational analysis frameworks are lacking, and it remains unclear whether available long-read sequencing platforms from Oxford Nanopore Technologies and Pacific Biosciences yield comparable results. Here, we present a computational strategy for spatial full-length transcript analysis, focusing on the spatial profiling of adaptive immune receptor repertoires (AIRR). Our approach introduces an adaptive filtering strategy that dynamically refines read selection and significantly improves consensus accuracy, enabling high-confidence sequence reconstruction independent of platform-specific sequencing error profiles. We further derive evidence-based guidelines tailored to the consistent and robust analysis of spatial AIRR data. The resulting software LongAIRR is modular and interoperable with existing spatial transcriptomics and AIRR analysis frameworks. This work establishes a methodological foundation for spatial immunology, enabling precise mapping of immune repertoires within their native tissue microenvironments.
- Coalescing nephron and ureteric bud progenitors potentiates nephrogenesis in recellularized kidney scaffolds
Bioengineered, transplantable kidney tissue using decellularized scaffolds offers a promising strategy to overcome the shortage of donor kidneys that limits organ transplantation for patients with end stage renal disease. These kidney scaffolds retain essential extracellular matrix architecture, providing a biologically active niche for recellularization. Successful generation of bioengineered kidney tissues includes enhanced patent vasculature and mature, functional nephrons with collecting ducts. Here, we report the development of engineered kidney tissue consisting of reconstituted kidney scaffolds and human pluripotent stem cell-derived nephron and ureteric bud progenitors. Structural analysis of recellularized kidney scaffolds showed advanced nephron structures that became more mature and exhibited interconnected nephron and collecting ducts. In vivo engraftment of reconstituted kidney scaffolds in mice led to vascularization, maturation, and secretory function. Notably, mouse-graft vascular anastomosis was evident with erythrocytes present in vasculature and nephron-secreted proteins detected in mouse urine, indicating functional integration. This approach demonstrates the feasibility to generate advanced bioengineered kidney tissues that offer a versatile platform for disease modeling, drug screening, and regenerative medicine.
- Consistent gut bacterial microbiota in European sea bass fed aquafeeds containing sustainable plant and invasive fish-based ingredients
The aim of this study was to evaluate the impact of two sustainable dietary protein sources on the structure and composition of the gut microbiota in European sea bass (Dicentrarchus labrax) juveniles. These protein sources were incorporated to the aquafeeds containing (a) Lupinus albus meal, treated with either exogenous enzymes (Solid state hydrolysis-SSH) or fermented with Saccharomyces cerevisiae (Solid state fermentation, SSF) and (b) Lagocephalus sceleratus meal. In the first case (a), the control aquafeed simulated a standard commercial diet, containing soybean meal whereas in the rest of the diets soybean meal was partially or totally replaced by hydrolysed or fermented Lupin meal. In the second case (b) the fish were fed Lagocephalus sceleratus unprocessed fishmeal as well as treated at different temperatures to deactivate tetrodotoxin (TTX). A control diet with 30% commercial fish meal was also fed as a reference diet. Both diets in all inclusion levels did not cause any significant gut microbiota change, suggesting their neutral role in this aspect. However, the gut bacterial communities of the fish fed with 12.5% lupin meal inclusion, had increased amino acid biosynthetic pathways suggesting a beneficial effect.
- Rapid assessment of nitrification inhibitor efficacy, competitiveness, and specificity using microrespirometry
Nitrification inhibitors are applied to reduce nitrogen losses and greenhouse gas emissions from fertilized agricultural ecosystems. However, their characterization is typically focused on determining effective inhibitor concentrations from growth or substrate conversion assays that are time-intensive and provide limited mechanistic resolution. Here, we present a microrespirometry (MR)-based workflow for rapid mechanistic characterization of nitrification inhibitors using oxygen consumption as a real-time readout for metabolic activity. The workflow enables the simultaneous assessment of inhibitor efficacy, competitiveness, and enzyme specificity within a single experimental setup, as sequential substrate and inhibitor additions enable direct discrimination between competitive and non-competitive inhibition and between ammonia monooxygenase-specific and broader respiratory inhibition. As a proof of concept, we evaluated three known nitrification inhibitors phenylacetylene (PA), nitrapyrin (NP), and dicyandiamide (DCD) using the ammonia-oxidizing bacteria Nitrosomonas europaea and Nitrosospira multiformis, the complete ammonia oxidizer Nitrospira inopinata, and the nitrite oxidizer Nitrospira moscoviensis. We also compared the results from the MR-based inhibition workflow with those from a conventional growth-based approach and observed a poor correlation between results for inhibitors that are not fully enzyme specific. In conclusion, this work establishes MR as a rapid and versatile platform for the mechanistic screening of novel potential nitrification inhibitors. MR assays reproduce known inhibitory responses while substantially reducing experimental time and increasing mechanistic resolution compared to other assays types. Additionally, we provide the first pure-culture characterization of PA, NP, and DCD efficacy and inhibition mechanisms in a complete ammonia oxidizer, N. inopinata.
- Mass-transfer-constrained thermodynamics links fluid motion to the preferential use of hydrogen and formate in syntrophic propionate oxidation
Syntrophic propionate oxidation in methanogenic environments depends on interspecies electron transfer through hydrogen and formate, yet the physical factors governing the relative use of these carriers remain poorly understood. Here, we examined how fluid motion alters electron-transfer energetics and pathway expression in the obligate syntrophic propionate oxidizer Pelotomaculum schinkii grown in coculture with Methanospirillum hungatei. A mass-transfer-constrained thermodynamic model was used to estimate H2 and formate concentrations at the P. schinkii cell surface and calculate the corresponding Gibbs free-energy change of H2- and formate-mediated propionate oxidation under different mixing conditions and growth stages. Transcriptomic analysis was used to assess expression of electron-transfer pathways. Under unmixed conditions, formate-mediated propionate oxidation was more thermodynamically favorable than the H2-mediated pathway, consistent with highly expressed genes involved in formate production. Mixing altered coculture activity and pathway energetics. H2 was more sensitive to mixing and certain conditions shifted the energetic advantage toward H2. Expression of the major hydrogenases and formate dehydrogenases generally tracked these pathway-specific energetic changes. These results show that fluid motion reshapes the near-cell thermodynamic favorability and enables condition- and growth-stage-dependent use of alternative electron-transfer pathways. Fluid motion should therefore be considered an ecological and engineering control on syntrophic metabolism.
- Scaling of Neuronal Growth and Excitability Through Separable mTORC1 and mTORC2 Pathways
As neurons grow, they must regulate intrinsic excitability to maintain an appropriate level of spiking based on synaptic inputs. Using gene knockouts, phosphoproteomics, and electrophysiology we show PTEN regulates neuronal growth and intrinsic excitability through separable downstream mechanisms. Pten loss induces cellular hypertrophy, increased excitatory synaptic input, reduced fast afterhyperpolarization, and burst firing. Deleting the mTORC1 scaffold, Raptor, rescues overgrowth and synaptic input but fails to normalize firing, while deleting Akt or the mTORC2 scaffold, Rictor, restores firing without rescuing growth. This dissociation identifies an AKT-mTORC2 mechanism that regulates voltage-gated calcium and BK potassium channels to set spike repolarization and burst firing. In vivo, Pten knockout produces altered network synchrony, lethal seizures, and impaired object and location behavior; Raptor co-deletion display non-lethal hyperexcitability with improved object-location coupling. The biological and pathophysiological significance of these mechanisms is demonstrated by overlap of the PTEN-regulated phosphoproteome with ASD and epilepsy.
- Coherent scene context accelerates and reshapes neural object representations
Coherent scenes facilitate object recognition, but the representational basis of this facilitation and its temporal evolution in the brain remain unclear. We tested this question using EEG and multivariate pattern analysis while 15 participants categorized objects from five semantic categories after a 500-ms preview of either an intact rendered scene or a phase-scrambled version of the same background. Reliable object decoding emerged earlier in intact scenes than scrambled scenes (142 {+/-} 5 vs. 162 {+/-} 10 ms), with higher decoding for intact scenes from 124 to 268 ms after object onset. Cross-condition decoding object information that generalized across scene formats, whereas subtracting cross-condition from within-condition decoding identified an earlier and stronger context-dependent component when scene structure was coherent. Cross-temporal representational similarity analysis (RSA) further showed that representational structure established during late scene preview generalized to early object processing only for intact scenes, linking contextual facilitation to anticipatory scene-derived representations. Finally, model-to-brain RSA showed that a language-aligned model explained neural representational geometry in intact scenes better than vision-only models, an advantage attenuated by scene scrambling. These findings indicate that coherent scene context shapes object coding by accelerating object-selective processing and contributing context-dependent representational structure beyond a context-invariant object code.
- LRP1 is an entry receptor for the botulinum toxin complex in the gut
Botulinum neurotoxin (BoNT) is an etiologic agent of food poisoning caused by Clostridium botulinum. The large progenitor toxin complex (L-PTC) crosses the intestinal epithelial barrier to deliver BoNT to target neurons; however, it is not clearly understood how BoNT enters the host. Here, we identified low-density lipoprotein receptor-related protein 1 (LRP1) as a major enterocyte transcytosis receptor for the hyper-oral-toxic L-PTC serotype B-Okra (L-PTC/BOkra). We found that hemagglutinin (HA), a neurotoxin-associated protein within the L-PTC/BOkra complex, binds to LRP1 via N-glycans. HA/BOkra co-localized with LRP1 within the internalized vesicles in cultured cells and enterocytes. LRP1 deletion inhibited the apical-to-basal transcytosis of L-PTC/BOkra in an intestinal epithelial cell line, and this effect was rescued by LRP1 re-expression. Finally, intestinal epithelial cell-specific LRP1-deficient mice displayed reduced susceptibility to toxicity caused by oral administration of L-PTC/BOkra. Taken together, these results indicate that N-glycosylated LRP1 mediates L-PTC/BOkra transcytosis via enterocytes, enabling BoNT to traverse the intestinal epithelial barrier.
- Bacterial Mercury Resistance Reveals a Robust Species-Structured Human Antimicrobial Mobilome
Antimicrobial resistance (AMR) is increasingly recognized as an ecological and evolutionary phenomenon that extends beyond clinical environments. Despite the predominant focus on antibiotics, bacterial metal resistance genes are among the oldest and most widespread adaptive antimicrobial systems, yet their distribution within human-associated microbial communities remains poorly characterized. Here, we investigated the ecology of mercury (Hg) resistance in children from a birth cohort with relatively high Hg exposure. Fecal samples from 234 children aged 4-8 years were analyzed using culture-based screening, whole-genome sequencing, and comparative analyses of metal- and antibiotic-resistance determinants and associated mobile genetic elements (MGEs). Hg-resistance was detected in 79.7% of samples, with HgR Enterobacterales isolated from 57% of children. Hair mercury concentrations were not associated with carriage. Sequencing revealed a phylogenetically diverse collection dominated by Escherichia spp. (61%). Hg resistance was mediated by 79 mer operons primarily associated with Tn21, Tn1696, and Tn5053 families circulating on chromosomes and a highly diverse plasmidome. Both rare and globally distributed plasmids related to foodborne, animal, and clinical Enterobacterales were identified. Metal resistance determinants exhibited strong taxonomic structuring, with Escherichia enriched in iron-uptake systems and siderophores whereas non-Escherichia taxa carried multimetal resistance operons. These findings indicate that Hg-resistance is shaped by ecological interactions and MGEs, becoming partially decoupled from contemporary Hg-exposure and bacterial community composition. The human gut therefore serves as an important reservoir linking environmental metal resistance to the broader evolution of AMR and provides insight into the baseline resistome and plasmidome of human populations.
- OmrA sRNA Inhibits Translation of Phosphoenolpyruvate Carboxylase to Impair TCA-Cycle Flux
Small RNAs (sRNAs) rarely cause strong growth phenotypes upon overexpression, complicating efforts to link regulatory interactions to physiological outcomes. Here, we report that high levels of the Escherichia coli sRNA OmrA, but not its sibling OmrB, severely inhibit growth in glucose minimal medium. Genetic, biochemical, and physiological analyses indicate that OmrA-dependent toxicity results from reduced flux through the tricarboxylic acid (TCA) cycle. A UV-based suppressor screen identified mutations in the gene encoding Hfq, the RNA-chaperone that aids sRNA-mRNA interactions. Secondly, three independent mutations clustered in the ribosome-binding site of ppc, encoding phosphoenolpyruvate carboxylase, a key anaplerotic enzyme. OmrA directly inhibits Ppc translation via Hfq-dependent base-pairing in the ppc 5' UTR, including the mutated nucleotides obtained in the genetic screen. OmrA is significantly more effective than OmrB in ppc repression in vivo and in vitro, consistent with sequence divergence in their central regions. Supplementation with glutamate, glutamine, or downstream TCA cycle metabolites fully restores growth, linking reduced Ppc levels to metabolic limitation. These results identify ppc as a physiologically relevant OmrA target and suggest how RNA toxicity can uncover central metabolic nodes used by sRNAs to modulate bacterial physiology.
- Anaerobic conditions increase plasmid transfer rates across Escherichia coli strains
Bacterial conjugation is the primary mechanism by which antibiotic resistance genes spread in microbial populations, yet our understanding of this process has been largely based on experiments conducted under aerobic conditions. This creates a fundamental disconnect: environments that are considered hotspots for gene exchange (e.g., the gut, abscesses, chronic wounds, and wastewater systems) are predominantly anaerobic. In this study, we investigate whether oxygen availability influences the transfer rate of a set of common ESBL-IncI1- and qnrS1-IncF plasmids in commensal Escherichia coli strains from chickens. We found that oxygen availability significantly shapes conjugation dynamics in a recipient strain-specific manner, with anaerobic conditions promoting higher ESBL-IncI1- plasmid transfer rates to commensal E. coli recipients. Conjugation rates of the ESBL-IncI1- plasmids to a laboratory strain of E. coli were several orders of magnitude higher and independent of oxygen level, while two qnrS1-IncF plasmids showed higher anaerobic rates. Our study reveals critical "oxygen blind spots" in conventional conjugation assays and suggests that conventional aerobic conjugation assays underestimate plasmid transfer rates in natural environments such as the chicken caeca. These findings highlight the importance of aligning experimental conditions with the physiological and ecological environments in which gene exchange naturally occurs. Tailoring these variables is essential for generating results that accurately reflect, predict, and potentially intervene in the horizontal spread of antimicrobial resistance.
- Early rhizosphere assembly during the onset of photosynthesis reveals inoculum-constrained succession with increased phylogenetic clustering, filtering and diversity
The rhizosphere microbiome, one of the most diverse and metabolically active microbial ecosystems known, plays fundamental roles in plant health and productivity. However, the ecological dynamics occurring during the transition between germination and the establishment of the first true leaves, a developmental window associated with the onset of active photosynthesis and rapid root expansion, remain poorly understood. Here, we investigated rhizosphere microbiome assembly during the first four weeks of tomato development by sampling communities arising from seven distinct natural soil inocula twice weekly to obtain fine-scale temporal resolution. Bacterial load, richness, evenness and phylogenetic diversity all increased significantly during plant development, indicating progressive increases in rhizosphere ecosystem complexity. In addition, diverse initial microbial communities differentially influenced both host plant development and the bacterial carrying capacity of the resulting rhizosphere ecosystem. Although temporal effects on rhizosphere microbiome composition were significant, assembly trajectories remained strongly constrained by the initial inoculum. Temporal analysis nevertheless revealed significant taxonomic turnover despite limited global compositional restructuring. In particular, Proteobacteria and Pseudomonadaceae decreased over time, whereas Actinobacteria, Acidobacteria and Streptomycetaceae increased. However, communities did not become progressively more similar or divergent over time. Altogether, our results indicate that early rhizosphere microbiome assembly involves rapid ecological succession within inoculum-constrained compositional trajectories, with early copiotrophic Proteobacteria progressively giving rise to more diverse and phylogenetically structured communities. These findings suggest that the first weeks of plant development may represent a critical ecological window for microbiome-based manipulation strategies in agriculture.
- Plasmid-encoded anatoxin biosynthesis in benthic cyanobacteria
Anatoxin-producing cyanobacteria pose growing ecological and public health risks, yet the genomic organization of anatoxin biosynthesis remains poorly resolved. Using complete genome assembly of a benthic cyanobacterium of the genus Microcoleus, we show that the entire anatoxin biosynthetic gene cluster is encoded on a small circular plasmid rather than the chromosome. This unexpected localization suggests enhanced mobility of anatoxin biosynthesis and has implications for toxin evolution, release, and environmental surveillance.
- Predictable induction responses of gut prophages
Temperate bacteriophages are dominant members of the human gut microbiome that can infect and lyse their bacterial hosts or integrate as prophages. During this integrated state, prophages exhibit extensive control over host physiology and lysis via induction. Here, we studied a diverse collection of Bacteroidales isolates, which are amongst the most abundant bacterial orders within the human gut, identifying 902 high-quality prophage genomes present within 305 isolates, 240 of which were poly-lysogens. Despite their prevalence, our understanding of the function and induction triggers of prophages is limited. To predict prophage induction, we employed an iterative profile Hidden Markov Model search across divergent bacterial hosts to identify prophage regulatory components. We found 197 Bacteroidales prophages encoding complete CI-like repressor proteins, which initiate induction upon DNA damage. We selected Bacteroides thetaiotaomicron strain Bt_806 to characterise further as it harboured six diverse prophages, including the prevalent and abundant prophage LoVE, which was the only integrated prophage encoding a complete CI-like repressor. Transcriptomics revealed phage LoVE was routinely induced upon DNA damage, while the five co-habiting prophages remained stably integrated yet exhibited transcriptionally active genes associated with regulation, prophage maintenance, and uncharacterised functions. Finally, we selected an additional eleven Bacteroidales poly-lysogens, confirming that integrated prophages encoding complete CI-like repressors were reliably induced upon DNA damage. Together, we demonstrate that mechanistic understanding of prophage induction linked with identification of regulatory genes enables selective and predictable induction of gut prophage species as a potential tool to modulate the microbiome.
- Spatially mediated interactions shape founder-cell fitness and community assembly in multi-species soil bacteria
Microbial communities spontaneously colonize pristine environments, yet how species growth kinetics and individual cell variation shape community assembly remains poorly understood. Here, we use time-lapse microscopy imaging to track the division of individual founder cells in communities composed of up to seven soil bacterial isolates grown on nutrient surfaces. With cell lineage tracking, we quantify species-specific absolute biomass formation and growth kinetics from early growth through stationary phase. The reproductive success of individual founder cells depended on the timing of their first cell division, which determined their access to the primary substrate and their maximum growth rates. In mixed-species communities, founder cell success also depended on species-specific, substrate-dependent growth rates and yields. In addition, spatial factors such as cell positioning, distances to non-kin neighbours, and identities of co-occurring species, further influenced outcomes. In spatially structured communities, interspecific interactions were globally governed by competition for primary substrates. We also observed cross-feeding of leaked metabolites, reflected in fluctuating paired interaction strengths and interaction signs. Species-pair interactions differed locally, with cells within distances of less than 15 mum exhibiting opposite interaction behaviours. Global pairwise interactions predicted from monoculture growth kinetics were observed in approximately half of the measured pairs, whereas measured paired interactions generally weakened in combinations of three or more species. Using a spatially explicit agent-based Monod growth model that includes interspecific interactions, we accurately predicted the compositions of seven-member communities. Overall, our results indicate that emergent, spatially mediated interspecific interactions between cells of different bacterial species primarily drive local and temporal changes in individual cell growth rates, which in turn determine final biomass formation. Because most natural microbial habitats are spatially structured, stochastic founder-cell positioning and fitness differences are key determinants of locally formed interaction patterns and species coexistence.
- Microglia-Specific Molecular Magnetic Resonance Imaging Probe Enables Noninvasive Separation of Parkinsonian Mice from Controls
Neuroinflammation mediated by reactive microgliosis is a central driver of Parkinsons disease (PD) pathogenesis. This inflammatory process unfolds years before clinical symptoms, creating an opportunity for early intervention. In vivo imaging technologies that could detect and quantify microglial reactivity are therefore essential for early diagnosis, patient stratification, and evaluating emerging immunomodulatory therapies that target this fundamental driver of PD progression. Yet no standardized, sensitive, and specific technology currently achieves this goal. Molecular magnetic resonance imaging (mMRI) is uniquely suitable to address this problem because it integrates inherent high spatial resolution and soft tissue contrast of conventional MRI with molecularly targeted contrast agents, enabling simultaneous acquisition of anatomical detail and functional/biological information at submillimeter isotropic resolution. Here we present a novel mMRI probe designed to specifically target colony stimulating factor-1 receptor, expressed primarily on microglia in the brain. In silico data show that the targeting ligand binds the extracellular Ig domain of the receptor. In vitro cell uptake studies with both murine and human microglia cell lines show that the probe binds the receptor triggering active cell uptake and in vivo MRI enabled effective separation of the A53T mouse model of PD from control mice using radiomics-assisted MR image analysis. Ex-vivo immunohistochemical analysis showed signal from the probe largely in the cytosolic compartment of IBA-1 reactive cells, confirming that the observed in vivo MRI signal is due primarily to retention of the agent by microglia. This novel technology has the potential to interrogate the regional presentation of microglial activation in PD.
- RNAPII and NER stall loop extrusion at UV lesions, shaping the 3D genome during repair
The three-dimensional (3D) genome architecture is highly elastic, adapting to nuclear processes such as transcription and the DNA damage response (Dekker & Mirny 2016; Carre-Simon & Fabre 2021). Nucleotide excision repair (NER) acts within this chromatin context to detect and repair mutagenic lesions induced by ultraviolet (UV) irradiation (Sancar 2016). UV irradiation has been shown to induce restructuring of the 3D genome across multiple scales, including chromatin compartments, domains, and loops. However, the extent to which NER activity contributes to this remodelling is unresolved, as the only prior study tracking such UV-induced changes was limited to repair-proficient cells (Kaya & Adebali 2025). Here, by combining genome-wide chromatin profiling of repair-deficient human cells with loop extrusion simulations, we show that lesion-stalled RNA polymerase II (RNAPII) and repair-associated barriers constrain loop extrusion. These events counter the loop-lengthening effects of UV-induced transcriptional shutdown, leading to shorter chromatin loops and reinforced chromatin domains that facilitate efficient lesion recognition and repair. The contribution by NER machinery underscores 3D genome reorganisation as an active mechanism both initiated and harnessed by DNA repair, rather than a passive consequence of DNA damage. The contribution by RNAPII extends its role beyond activating transcription-coupled repair to promoting a genome-wide repair-permissive state. Together, these findings advance our understanding of how nuclear processes coordinate on a shared chromatin substrate to preserve genome integrity.
- Ferroptosis is a Physiologic Vulnerability of Iron-Recycling Macrophages
Iron deficiency anemia affects one-third of the global human population. Paradoxically, the daily iron required to fuel the production red blood cell (RBC) and prevent anemia is provided through its recycling from senescent RBC. This is achieved by splenic red pulp macrophages (RPM) that extract iron from the heme groups of hemoglobin (Hb). How these professional erythrophagocytic macrophages prevent intracellular iron flux from inducing cell death via ferroptosis is unknown. Here we show that SPI-C, the master transcriptional regulator of the erythrophagocytic lineage, orchestrates two redundant anti-ferroptosis pathways. One supports glutathione synthesis, via NF-E2-related factor 2 (NRF2), and the other relies on bilirubin production by biliverdin reductase A (BVRA). Genetic ablation of both pathways, but not either alone, sensitizes erythrophagocytic macrophages to ferroptosis, depletes RPM and increases the severity of iron deficiency anemia in mice. These findings reveal a central physiologic role of ferroptosis in the control of macrophage function, iron homeostasis and iron-deficiency anemia.
- TDP-43 subtypes shape transcriptomic signatures in Alzheimer's disease
TAR DNA-binding protein 43 (TDP-43) pathology frequently co-occurs with Tau neurofibrillary tangles (NFTs) and amyloid {beta} plaques in Alzheimer's disease (AD), driving significant clinical heterogeneity. Whether TDP-43 engages autonomous molecular programs or instead amplifies Tau-driven neurodegeneration remains difficult to resolve, largely because these pathologies often co-occur. To separate these overlapping signatures, we generated regionally resolved transcriptomic profiles from cognitively normal controls (Controls), neuropathologically defined cohorts of AD, AD with limbic-predominant age-related TDP-43 encephalopathy (AD/LATE), and frontotemporal lobar degeneration (FTLD-TDP), categorizing them by their distinct TDP-43 subtypes (types and {beta} for AD/LATE; types A and B for FTLD-TDP). By integrating transcriptomic profiles with quantitative measures of phosphorylated TDP-43 (pTDP-43) and Tau (pTau), we separated pathology-associated signals within mixed disease contexts. We found that TDP-43 is linked to distinct transcriptomic programs in AD/LATE that are largely uncoupled from Tau burden and diverge from those observed in FTLD-TDP. These signatures showed regional specificity, with transcriptomic remodeling occurring in the amygdala across both diseases, whereas frontal cortex alterations were largely restricted to FTLD-TDP. Furthermore, by stratifying cases by TDP-43 morphological subtype, we unmasked specific biological trajectories, from immune activation to unique cellular vulnerabilities, that are not apparent in unstratified cohorts. Together, our findings provide a framework for decoupling mixed proteinopathies and demonstrate that TDP-43 shapes autonomous, subtype-dependent transcriptional landscapes in AD.
- Machine Learning-based Prediction of Preterm Birth Using Genetic Data
The leading cause of mortality and morbidity in children under the age of 5 is preterm birth. The timing of birth is influenced by both genetic and environmental factors, but the underlying mechanisms remain poorly understood, making its prediction difficult. In this study, we investigated the potential of using machine learning models to predict preterm birth based on genetic data from the Norwegian Mother, Father and Child Cohort Study (MoBa). We trained and evaluated several classification algorithms on individual-level genetic data from over 15,000 mothers and children. Our results indicate that the predictive capacity of maternal gestational duration-associated loci for preterm birth is limited, with the highest AUC values around 0.57. Additionally, incorporating more SNPs within the associated loci did not improve prediction performance. As expected, the contribution of the maternal genome to preterm birth prediction was found to be larger than that of the fetal genome. Overall, our findings suggest that while genetic testing provides some information about an individual's risk for preterm birth, further research incorporating additional factors is necessary to enhance predictability.
- Clinically aligned rationale generation for glaucoma subtype classification via a knowledge-distilled language model
Automated glaucoma subtype classification from clinical notes remains clinically unactionable without subspecialty-aligned explanations supporting clinician-facing deployment. We extended our Ci-SSGAN with a GPT-5.2-to-Qwen3-8B teacher-distilled reasoning module, fine-tuning Qwen3-8B on 2,660 de-identified ophthalmology notes using expert-reviewed rationales. On 294 notes, the fine-tuned model achieved ROUGE-L 0.792 and BERTScore F1 0.955, surpassing eight zero-shot comparators including GPT-4o and GPT-4.1, establishing privacy-preserving distillation as a path to interpretable AI.
- Response consistency of ChatGPT-4o for Type 2 Diabetes Nutrition and Physical-activity Recommendations: A Pilot NLP-based Assessment of GPT outputs
Generative AI tools such as ChatGPT are increasingly used by the public to seek guidance on diet and physical activity for type 2 diabetes (T2D) prevention and management. However, the consistency of model outputs across different users and disease-stage scenarios remains insufficiently characterized. This pilot study aims to evaluate the word-level and semantic-level consistency of GPT-4os diet and physical activity responses for type 2 diabetes prevention and management. We designed 12 prompts covering four categories: prediabetes, diagnosed type 2 diabetes (T2D), diagnosed T2D with complications, and general questions that did not specify dysglycemia stage. Word-level similarity was quantified with Term Frequency-Inverse Document Frequency (TF-IDF) cosine scores; sentence-level semantic similarity was measured using large language models (LLMs) - DeBERTa-v3 MNLI to calculate the entailment probabilities. The results showed that mean cosine similarity across users was moderate (0.44-0.66), whereas mean entailment similarity was higher (0.68-0.81). Across stages, word-level similarity was low to moderate (0.28-0.63) and entailment similarity remained moderate to high (0.63-0.80). Low similarity commonly referenced distinct food choices, operational details, safety warnings, and stage-specific suggestions. GPT-4o generated semantically consistent but variably detailed responses and the moderate semantic variation suggested limited differentiation of response content across diabetes-related stages in this pilot consistency assessment.
- Forget the Apple tax, this is the AI tax
Apple’s decision to raise prices in response to memory cost increases is not unique to the company. If Apple has to do it , everyone else will as well. Apple announced stiff price increases Thursday — up to 25% in some cases — that extended across most products, including refurbished Macs and iPads (which saw prices increase up to $330). Apple might have left iPhones out of the mix for now, but they’ll likely see price increases when new models appear this fall. Omdia believes the memory price crisis spells the end of low-cost smartphones . The price hikes begin “We had assumed a price hike of $100 to Pro and ProMax iPhones, and $50 hike to base models,” wrote IDC Senior Director Nabila Popal. “However, seeing the price hikes to iPads and Macs going as high as $300 for some models, my personal instinct says the hike to iPhones may be even higher than what we assumed — perhaps even $200 to the Pro/Pro Max models. I think the days of $50 price increases are over.” What’s driving all this? The answer, increasingly, is AI. The buildout of large language model (LLM) infrastructure requires vast quantities of high-bandwidth memory and that appetite is accelerating faster than manufacturers can respond. Will Apple suffer? Maybe, though perhaps not too much. Popal notes that the introduction of Siri AI will give a large number of existing users a reason to upgrade. Given most of these devices are sold on installment plans, even a $200 price increase over 36-months might not pose a major barrier to sales. IDC expects iPhone sales to decline, but only by 5% — while the rest of the industry shrinks. The worst is yet to come This may not be the worst of it. Major players now warn that the supply-demand gap could continue to widen in the coming years. The higher costs that result will spread across all types of electronic devices, so anything with memory, a processor, or storage will become more expensive. Microsoft announced its own price increase, adding up to $150 to the cost of Xbox this week. And Lenovo made its own contribution when it warned that high memory prices will become the new normal into 2030, and prices might never return to early-2025 levels. Micron, Samsung, and SK Hynix describe the situation as beyond their control, saying they’re having difficulty meeting demand even for their top customers. These companies are generating massive profits all the while. In Lenovo’s case, the company claims that while it is introducing additional manufacturing capacity, this is not making a dent in the supply-demand imbalance. SK Hynix is expected to expand its manufacturing capability by accelerating its original 2040 expansion plans to 2030, at which time it should have tripled output; even that might be insufficient to meet demand. A long journey ahead “We currently do not have line of sight as to when memory supply will be able to catch up with increasing demand,” Micron CEO Sanjay Mehrotra said this week . With no end in sight, Wedbush Securities analyst Dan Ives says he believes Apple is attempting to get ahead of the inflationary spiral , speculating that the latest price increases bake future increases into their model. Memory prices began to spiral out of control last fall, with prices today reaching levels no one anticipated. They rose as much as 98% in the first quarter of 2026 and are set to jump by another 58% to 63% in the current quarter, according to TrendForce . This consequential uncertainty is affecting tech stocks. Asian stock markets fell sharply on Friday, led by a sell-off in technology firms. Trading on South Korea’s Kospi was temporarily suspended as a result — for the third time this week. Apple shares are down again, and the sell-off is expected to continue. So, who’s winning? The beneficiaries of this memory squeeze are not hard to identify. While consumers face higher prices for phones, laptops and games consoles, the companies driving memory demand — the hyperscalers and AI firms building out server farms at extraordinary scale — are posting record revenues. The costs flow down; the profits flow up. Some, including Naomi Klein , see it as a kind of strip mining of human intellect and ingenuity, a cultural wealth transfer in which human innovation is packaged and resold as competing chatbots, for a fee. There are signs this may not be sustainable . A UBS Group survey finds approximately 60% of companies have started curbing AI spending because of token costs, shifting to low-cost and open-source models. But even if AI-driven memory demand softens, there is little reason to expect consumer tech prices to follow. History suggests they rarely do. Truncated dreaming For Apple watchers, it was pleasant enjoying a few months in which the dream of a sub-$500 Mac was realized, only for that happy reverie to be dashed by the VC-funded race to deploy AI server farms for the benefit of those with pockets deep enough for the tokens to feed them. Please join me on social media at BlueSky , LinkedIn , or Mastodon , even better, please subscribe to The Core for your daily collection of human-curated Apple News.
- Trump’s Burdensome AI Regulation; Apple’s Price Hike
Trump’s Burdensome AI Regulation; Apple’s Price Hike The Information
- Webbin
From Google Maps to a professional Website in seconds.
- Single-Axis Fairness Interventions Produce Asymmetric Cross-Axis Effects in Clinical Prediction
Objective: To systematically evaluate whether single-axis demographic balancing introduces cross-axis fairness trade-offs in clinical prediction, and to characterize the directional asymmetry of these effects across model architectures and balancing strategies. Materials and Methods: We evaluated cross-axis fairness effects of demographic balancing in one-year all-cause mortality prediction using the MIMIC-IV database (N=64,427). Seven machine learning architectures were trained under six balancing strategies targeting either gender or race, with performance assessed via outcome-stratified 80-20 train-test splits repeated 30 times across both targeted and non-targeted axes using AUC, TPR, and Brier score. Results: Gender-targeting interventions largely preserved race fairness, while race-targeting consistently disrupted gender fairness across all methods and the majority of architectures. This asymmetry was invisible to same-axis evaluation alone. Race-targeting also incurred greater performance costs and calibration loss, with observed fairness gains potentially reflecting leveling down rather than genuine improvement. The same intervention could appear successful under TPR but fail under AUC evaluation. Discussion: The asymmetry likely reflects differential category complexity: binary gender balancing requires modest distributional shifts, whereas multi-category race balancing necessitates aggressive reweighting that propagates to correlated axes. Cross-axis fairness effects are directionally dependent and metric-sensitive, indicating that single-metric, single-axis evaluation is insufficient. Conclusion: Single-axis fairness optimization cannot guarantee cross-dimensional equity. Cross-axis, multi-metric fairness evaluation should be integrated into pre-deployment auditing of healthcare artificial intelligence (AI) models.
- A language model framework for sequence modeling of EHR audit logs to characterize clinician-EHR interactions
Objective: Electronic health record (EHR) audit logs capture clinician-EHR interaction patterns, but most audit log research relies on aggregated measures (e.g., total time). We investigated how audit logs could be modeled using large language model (LLM) architectures to learn underlying workflow sequences during clinician-EHR interactions. Materials and Methods: Using >295 million EHR-based audit log actions from inpatient settings spanning 2019 to 2024, we fine-tuned Llama-3-8B under two encodings: (1) symbolic field-based tokens and (2) semantic natural language audit log action descriptions. A first-order Markov model, which used only the immediate prior action, served as baseline minimal context comparator. Model representation was assessed using next-action prediction accuracy in an early-period test set and two temporally distinct out-of-sample (OOS) periods. Results: In the early period test set, the semantic LLM achieved the highest accuracy (0.7418, 95%CI [0.7415 - 0.7420]) compared to the symbolic LLM (0.3838, 95% CI: 0.3836 - 0.3840) and Markov baseline (0.4553, 95%CI [0.4551 - 0.4555]). The semantic approach was also robust to temporal drift in EHR interaction patterns seen in the two OOS periods (semantic LLM vs Markov accuracy, OOS-1: 0.6509 vs 0.3169; OOS-2: 0.6232 vs 0.2648). Discussion and Conclusions: Semantic LLM relying on audit log action descriptions yielded the highest next-action prediction accuracy, and demonstrated robustness to temporal drift, suggesting that longer sequential context and semantic action descriptions may improve audit log-based sequence modeling. These findings support further development of semantic sequence models for audit log research, including task identification, automated workflow characterization, and safety-focused analyses of clinician-EHR interaction patterns.
- Performance of Google NotebookLM for AI-assisted data extraction and consensus statement generation in a heterogenous systematic review on inflammatory bowel disease, obesity, and cardiometabolic comorbidities: A Methodological Report
Background: Large language models (LLMs) offer promise for systematic review data extraction, but performance in complex multidisciplinary domains and utility for clinical statement generation remain insufficiently described. Objectives: To evaluate Google NotebookLM for AI-assisted data extraction and RAND/UCLA consensus statement generation in a systematic review of IBD, obesity, and cardiometabolic comorbidities. Methods: Studies were organized into domain-specific notebooks; structured prompts generated standardized evidence tables. Two independent reviewers validated outputs against full-text articles using a four-category error classification. Cell-level accuracy and critical accuracy (cells free of major factual errors) were the primary metrics; workflow time was compared against a published conventional extraction benchmark. Concordance between AI-generated and expert-finalized statements was assessed. Results: Across 57 articles, 1,710 data cells were extracted; 151 (8.83%) were flagged, yielding 91.17% cell-level accuracy. Major factual errors occurred in only 4 cells (0.23%), for a critical accuracy of 99.77%. Most errors were minor omissions (59.6%) or incomplete extractions (30.5%); domain error rates ranged from 7.08% to 11.33%. The pipeline required 17.7 versus a projected 165.1 person-hours (89.3% reduction). PICO-structured prompting generated 70 candidate statements; 58 of 112 finalized panel statements (51.8%) were AI-derived, and 85.7% were retained in the finalized set. Conclusion: Google NotebookLM demonstrates feasibility as a primary extraction and synthesis tool in a multidisciplinary systematic review, with extractive incompleteness as the principal limitation and substantial time savings over conventional approaches. Its novel application to RAND/UCLA consensus statement generation extends AI-assisted evidence synthesis to clinical consensus generation workflow.
- COMPASS: A Clinically-Optimized Multimodal Prediction Architecture with Survival Strategy for AD Prognosis
Precise early diagnosis and progression prediction of Alzheimer's Disease (AD) are critical for optimizing clinical intervention. However, current methodologies often suffer from the passive utilization of clinical priors and rigid modal fusion strategies, failing to capture the heterogeneous variations of imaging biomarkers. Furthermore, predicting the precise time-to-conversion from Mild Cognitive Impairment (MCI) to AD remains a formidable challenge. To address these limitations, we propose COMPASS, a clinical-guided multi-modal framework that unifies diagnosis with a comprehensive survival strategy. Specifically, we instigate a paradigm shift to "clinical-prior-driven" learning by incorporating Clinical-Guided Spatial Attention (CGSA), which actively transforms clinical states into visual signals to modulate neural focus on pathological regions. To bridge the semantic gap between modalities, we introduce Reciprocal Semantic Interaction (RSI) via cross-attention, while a Disease-Stage-Aware Modal Fusion (DSAMF) module dynamically adjusts modal weights based on inferred disease severity to mimic clinical reasoning. Moreover, we specifically design a Dual-Head Joint Survival Risk and Time Prediction Network (DH-Net) to jointly perform quantitative conversion time prediction and patient risk stratification. Extensive experiments demonstrate that COMPASS outperforms state-of-the-art methods, achieving 83.19% accuracy in pMCI vs. sMCI classification, an MAE of 7.96 months for conversion time prediction, and a C-index of 0.819. Furthermore, we conducted in-depth neurobiological interpretability analyses, revealing right hippocampal dominance and synergistic regional impairment patterns, thereby providing new biological insights for early AD diagnosis and subtype identification.
- AI-enabled proteomic profiling identifies interpretable diagnostic biomarkers for gastric-type adenocarcinoma of the uterine cervix
Gastric-type adenocarcinoma of the uterine cervix (GAS) is a rare, aggressive cervical cancer unrelated to HPV infection that is frequently misdiagnosed because it closely resembles both benign cervical glands and HPV-related cervical adenocarcinoma. This diagnostic confusion can lead to inappropriate treatment, highlighting the need for objective molecular markers. Here, we performed the systematic multi-center proteomic analysis of GAS, profiling 407 cervical tissue samples to map its molecular landscape. To overcome limited sample size and biological noise, we developed WEDGE. First, generative AI synthesizes realistic artificial proteomic profiles to augment the training data. Biologically informed network analysis then leverages known biological relationships to surface diagnostically meaningful patterns. WEDGE identified a two-protein signature, Pepsinogen C (PGC) and DNA Methyltransferase 1 (DNMT1), that distinguished GAS from HPV-related cervical cancer with 93% accuracy in the test cohort and 97% accuracy in an external proteomic cohort, outperforming existing biomarker-discovery methods. Tissue staining of an IHC validation cohort confirmed the expression patterns and reached a diagnostic accuracy of 87.9%. Beyond diagnosis, PGC independently predicted patient outcomes, and combining PGC with routine clinical features improved risk prediction (C-index 0.701). Together, these results establish an AI-driven framework for biomarker discovery and provide clinically relevant candidate tools for diagnosing and prognosticating for GAS.
- Origin
Free Font
- Fairness-aware, explainable clinical decision support for opioid use disorder risk stratification: development and internal validation of a dual-layer AI system
Opioid use disorder (OUD) remains a leading cause of preventable death in the United States, yet the tools used to assess OUD risk rely on episodic self-report, produce binary output, and exhibit documented performance disparities across demographic groups that can widen existing inequities in care. Machine-learning models for OUD risk are seldom evaluated for demographic fairness or designed for the transparency clinicians need to trust and act on them. We present a fairness-aware, explainable clinical decision support system for four-tier OUD risk stratification, developed and internally validated on a large electronic health record-derived cohort accessed via Mayo Clinic Platform_Discover. The system pairs an XGBoost classifier with a transparent Clinical Rules Engine that attributes risk across six clinical domains, providing clinician-interpretable explanations alongside each prediction. To address demographic disparity directly, we applied an iterative bias-mitigation strategy combining age-balanced resampling, removal of race as a model input, and cost-sensitive reweighting, and measured its effect using group-fairness metrics (demographic parity, equal opportunity, equalized odds, and calibration within groups). On a held-out internal test set, mitigation reduced the White-Black gap in high-risk detection from 30.3 to 7.4 percentage points (a 76% relative reduction) and the age-based accuracy gap from 6.6 to 2.7 percentage points (59% reduction), raising high-risk detection for Black patients from 58.3% to 75.0%, at a cost of fewer than two percentage points of overall accuracy; gender differences remained below three points. The system was independently qualified through Mayo Clinic Platform_Solutions Studio. This work offers an implementable, transparent blueprint for operationalizing fairness and explainability in clinical AI for high-risk prescribing, with external and prospective validation as the clear next steps.
- Gadget prices have fallen for decades. Then AI happened.
The race to build AI data centers is leading to a global shortage of memory chips, driving up the cost of personal electronics.
- OpenAI Limits Release of New Model Under Pressure From US
OpenAI is rolling out a preview version of a more capable new artificial intelligence model to select partners before making it available more widely in the coming weeks, following pressure from the Trump administration to stagger the release.