Nature

Large language models embedded in autonomous agents process trusted instructions and untrusted data in one context window, leaving them open to direct and indirect prompt injection. In healthcare this is not hypothetical: a 2025 JAMA Network Open study found commercial medical LLMs followed injected instructions in 94.4% of simulated patient encounters, including life threatening recommendations . Yet the clinically decisive problem we quantify here is different. Most real clinical threats protected health information PHI exfiltration, cross patient access, bulk export, out of scope advice are fluent, legitimate looking requests that carry no attack signal, so even a state of the art injection detector passes them. Existing runtime guardrails trade safety against latency: model based auditors are accurate but add hundreds of milliseconds of Python inference, while lexical filters are fast but blind to obfuscated or semantically disguised payloads. We present QFIRE, an inline, provider agnostic prompt firewall implemented as a single self contained Rust toolchain proxy, CLI, and benchmark harness. QFIRE combines three mechanisms: (i) positive security scope constraints, which restrict a model call to a declared natural language purpose and block out of scope drift even when no overt attack token is present; (ii) an asynchronous detector graph that runs N rules and their detector nodes concurrently, cheapest checks first; and (iii) a de obfuscation pass that decodes Base64 hex ROT13, folds homoglyphs and leetspeak, and strips zero width characters before detection. QFIRE ships 106 versioned firewall rules and a dedicated HIPAA Safe Harbor 18 identifier PHI panel, and runs a local DeBERTa v3 injection classifier via embedded ONNX Runtime. On 1968 public prompt injection and jailbreak prompts QFIREs deterministic hybrid attains F1 0.86, statistically tied with Metas state of the art PromptGuard 2 0.86 and above protectai DeBERTa v3 0.83; lexical baselines lag 0.16 to 0.50. Our central result is on QFIRE HealthBench, a new 2000 prompt healthcare benchmark we build and release with real garak and Microsoft PyRIT payloads. There the same PromptGuard-2 recovers only 0.40 recall DeBERTa v3 0.57, because most clinical threats carry no injection signal; QFIREs combined scope plus PHI chain reaches 0.83 recall F1 0.87 at a calibrated 0.08 false positive rate. Generic injection detection, even state of the art, is therefore necessary but not sufficient for healthcare agents. A bare LLM judge also closes most of this static corpus gap F1 0.90; QFIREs contribution beyond static accuracy is auditable determinism, bounded latency, and adaptive robustness, where the bare judge falls to 34 to 59% recall section 5.5. End to end, placing QFIRE in front of a tool using agent over a mock EHR sandbox cuts the agents harmful action rate from 0.38 to 0.00 at a 0.13 benign utility cost. All code, rules, corpora snapshots, and scripts are released, and every table regenerates from a single make paper target against local models with no paid API keys.

Bacterial, plant, and animal cells synthesize nucleotide immune signals as a conserved strategy to defend against viral infection1-4. In bacteria, Thoeris anti-phage defense systems convert nicotinamide adenine dinucleotide (NAD+) into the cyclic ADP-ribose signals 2'cADPR and 3'cADPR to activate downstream effectors and restrict viral replication5-8. Phage proteins can bind and sequester Thoeris signals6,9-13, but no mechanisms are known to degrade the exceptionally stable 2'cADPR and 3'cADPR molecules and terminate immune activation. Here we use a forward biochemical screen to discover the mycobacteriophage protein RyDEP as the founding member of an enzyme family that cleaves 2'cADPR and 3'cADPR to inactivate Thoeris defense. We show that RyDEP is a glycosidase that cleaves the ribose-ribose linkage in 2' and 3' cADPR immune signals to both inactivate host defense and enable direct restoration of NAD+. A crystal structure of the RyDEP-3'cADPR complex in the post-cleavage state explains the molecular basis of immune signal degradation and reveals surprising homology with the Repeat12 domain of animal ryanodine receptors (RyRs) that control calcium flux and muscle contraction14,15. We demonstrate that diverse phage RyDEP proteins tune RyR-domain activity to either degrade or sequester immune signals. Our results define RyR-domain proteins as regulators of nucleotide immune signaling and explain how viruses subvert host antiviral defense.

The endosymbiotic evolution of plastids and mitochondria was central to the origin and success of eukaryotes. One of the most prominent molecular machineries thought to have disappeared early in eukaryote evolution is the multi-subunit bacterial DNA polymerase III (DNApol-III), which is the principal enzyme complex supporting DNA replication in bacteria. Here, we combined worldwide metagenomics and cultivation to characterise the mosaic genomic landscape of abundant phytoplankton lineages of Teleaulax (Cryptophyceae), which contain an endosymbiotically-derived nucleomorph genome. Unexpectedly, the nuclear, plastid and nucleomorph genomes of Teleaulax contain ubiquitously expressed genes for plastid-targeted DNApol-III subunits. These genes shed light on the functioning of Teleaulax genomes when sequestered by the ciliate Mesodinium during its kleptoplastidic photosynthetic activity1-3. In particular, the alpha subunit gene (encoding the polymerase activity), which resides in the nucleomorph genome, is continuously expressed in Mesodinium in controlled laboratory experiments. This provides a mechanistic explanation for the replication of Teleaulax plastid genomes weeks after the nuclear genome is lost4. Beyond Teleaulax and close relatives, we also identified genes encoding plastid-targeted DNApol-III subunits (including alpha) in nuclear genomes of unicellular and multicellular lineages of Archaeplastida that form, along with those of Cryptophyceae, monophyletic clades firmly positioned within Cyanobacteria. Together, our results reveal a previously overlooked retention of bacterial DNA replication machinery from plastid primary endosymbiosis in Archaeplastida, its acquisition by Cryptophyceae during secondary endosymbiosis, and its direct role in contemporary plankton as a facilitator of kleptoplastidic photosynthetic activity by heterotrophic ciliates.

Innate immune systems detect molecular signatures of infection to initiate antiviral defence1-3, yet the identity of pathogen-associated signals that distinguish phage from host nucleic acids remains incompletely understood. While recent work has shown that nucleic acid structures can act as triggers for bacterial defense systems4-7, how these structural signals are coupled with immune activation remain unclear. Here we show that forked DNA structures activate a helicase-nuclease immune complex in type III Druantia through a processing-dependent mechanism. Using cryo-electron microscopy and biochemical reconstitution, we find that the exonuclease DruH processes 3' DNA termini to generate 5' overhangs that recruit and activate the helicase-nuclease DruE at duplex-single-stranded DNA junctions. Structural analysis of the DruE-DruH complex reveals how substrate-dependent assembly remodels an autoinhibited helicase dimer into an active DNA degradation complex. Functional assays demonstrate that coordinated nuclease and helicase activities enable efficient degradation of forked DNA substrates and mediate phage defense without detectable host toxicity. Together, our findings define a mechanism in which enzymatic processing of replication-associated DNA structures licenses immune activation, providing a framework for how nucleic acid architecture is coupled to effector activation in bacterial immunity.

Animals use the location of visual stimuli to select appropriate actions1-5, and the upper and lower visual field often carry different ecological and behavioral meaning6-9. In mice, the superior colliculus is a key central hub that transforms visual input into orienting, defensive, and approach behaviors3,10-13. Its superficial layers receive retinotopically organized input from the retina and contain genetically defined cell types with distinct downstream projections, including wide-field neurons that project to the lateral posterior thalamus and narrow-field neurons that target the parabigeminal nucleus and deeper collicular layers14-18. These features raise the question of whether circuits of the superior colliculus are repeated across visual space or exhibit visual-field-dependent specializations. Here, we show that the mouse superficial superior colliculus contains visual-field-dependent circuit modules. Dual-color rabies tracing revealed that wide-field and narrow-field neurons receive input from a largely shared set of brain regions, whereas upper- and lower-field domains differ in how they sample those inputs. Some source regions preferentially innervate one visual-field domain, producing biased regional input strength, while others contain topographically segregated projecting neurons that target upper- or lower-field domains. MAPseq showed that most superficial collicular neurons project to single downstream targets, with upper- and lower-field populations differing in target probability. Two-photon calcium imaging further showed that wide-field neurons in upper- and lower-field domains differ in stimulus selectivity. Together, these findings reveal a visual-field-dependent wiring logic that biases how the superior colliculus samples inputs and routes signals to downstream pathways. HighlightsO_LIWide- and narrow-field neurons receive broadly overlapping inputs C_LIO_LIUpper- and lower-field domains differ in input strength and topographic organization C_LIO_LIMost superficial collicular neurons project to a single target C_LIO_LIVisual field position biases downstream target probability C_LI

For the brain to compute, electrical signals must propagate over the membranes of individual neurons, connecting synaptic inputs to synaptic outputs1. Complex neuronal morphologies coupled with the spatial organization of synaptic inputs and outputs enable diverse voltage transformations that underlie cell-type specific computations2,3. However, measuring these transformations in vivo has remained challenging, leaving a crucial gap in our mechanistic understanding of single neuron computation. Here, we develop ASAP7y, a genetically encoded voltage indicator with unprecedented subthreshold sensitivity and expanded excitation compatibility in both mice and flies. We leveraged ASAP7y combined with two-photon random-access microscopy to record sensory stimulus-evoked voltage dynamics with millisecond, subcellular, and subthreshold resolution along the neurites of individual neurons in Drosophila. We found remarkable heterogeneity in voltage propagation across cell-types, delineating a fundamental axis of electrical diversity. Leveraging a nanoscale EM reconstruction of the visual system4, we modeled the electrotonic properties of single neurons spanning 717 cell types, revealing how morphology shapes voltage transformations. Finally, we demonstrate that confined voltage propagation creates substrates for local computation, producing subcellular domains with distinct feature selectivity across multiple cell types. These results provide mechanistic insight into how critical single neuron computations arise and reveal parallel processing in single neurons.

Mounting evidence from surgical type II focal cortical dysplasia (FCD) tissues and mouse models have recently shown that dysmorphic neurons carrying MTOR mutations (DNs) in FCD exhibit hallmarks of cellular senescence. Building on pioneering work from the Baulac group identifying cellular senescence as a feature of mTOR-pathway FCD, a recent study by Ribierre et al. (2024) [1] proposed oral dasatinib and quercetin (DQ) as a therapy that partially decreases the load of mutant, senescent neurons and thus reduces seizure occurrence in FCD mice. Using a different mouse strain and a different gain-of-function mutation in MTOR, our data confirm the presence of senescence hallmarks in FCD mice, but do not support one of the conclusions of Ribierre et al.--that DQ acts as a senolytic in an FCD mouse model--and we propose an alternative interpretation. We longitudinally tracked individual cell fate using two-photon microscopy and complemented these data with EEG monitoring and immunohistochemistry. Immunohistochemical analyses were performed within the same sections using multiple markers, allowing direct identification of mutant neurons and assessment of senescence-associated labeling. While we observed a detectable reduction in a senescence-associated marker, consistent with a senomorphic effect, it did not translate into a change in seizure phenotype, despite treatment timing and dosing matching those in the original study. For detailed materials and methods, see Extended Methods.

Summary paragraphWhen cells enter S phase, bidirectional DNA replication is initiated through the kinase-regulated recruitment of three activators (Cdc45, GINS and Pol epsilon) to a duplex DNA-loaded double hexamer of MCM ATPases. Together these proteins form two CMGE helicases that establish divergent replication forks as they become separated1. To understand CMGE biogenesis, we reconstituted the pre-Initiation Complex with purified yeast proteins. The cryo-EM structure shows a set of firing factors caught in the act of assembling two symmetric CMGEs. We show how stepwise complex formation reshapes MCM in preparation for DNA opening and we explain how ATP promotes firing-factor ejection and CMGE maturation. While we find that Sld2 promotes GINS recruitment to MCM as expected, it also aids efficient separation of the CMGE dimer, and it is essential for lagging strand ejection from MCM. These findings have direct implications for our understanding of the metazoan Sld2 ortholog, RECQL4, pointing to a replication-fork establishment mechanism conserved across eukaryotes.

Brains rely on internal models of the world to interpret sensory input and to simulate the future. In the mammalian hippocampal-entorhinal network, environments are represented by cognitive maps that contain spatially selective neurons such as place, grid, head direction and object-vector cells1-6. Neurons with allocentric spatial tuning have recently been discovered also in non-mammalian organisms including larval zebrafish7 but it remains unclear to what extent these neurons establish internal cognitive representations. We measured neuronal activity in telencephalic area Dc of head-fixed adult zebrafish exploring a novel, richly structured virtual reality. Neurons were sharply tuned to one or multiple locations and collectively represented environmental space. Activity fields exhibited neuron-specific associations to visual landmarks, indicating a prominent vectorial component in spatial representations. Population activity evolved and became increasingly informative as fish explored the environment. When landmarks were removed after familiarization, landmark-associated activity partially persisted and subsets of neurons reported prediction errors, implying that activity was in part driven by an internal representation. Strong functional coupling among neuronal ensembles and winner-take-all dynamics suggest that representations evolve by refinements of pre-structured networks. The teleost brain therefore generates internal models of structured environments that are optimized by experience and enable cognitive inference and prediction.

The anteroposterior collinear expression of Hox genes is a hallmark of animal development that underpins the diversification of body plans1 and life cycles2. However, the origin and drivers of this coordinated expression remain elusive: while vertebrates rely on complex cluster-wide Hox gene regulation3-8, insects define gene-specific, sub-cluster regulatory domains9-11. Here, we discover a new mode of Hox gene regulation in segmented worms (Annelida). By combining chromatin conformation data with histone modifications profiling in Owenia fusiformis, we show that a large distal enhancer forms developmentally regulated, long-range contacts across the Hox cluster, and its activation coincides with the consolidation of a cluster-wide topologically associating domain (TAD), loss of Polycomb-mediated repression, and Hox gene transcription. This chromatin structure also occurs in the annelids Dimorphilus gyrociliatus and Capitella teleta, the latter showing additional subTAD structures that correlate with Hoxs temporal collinearity12. Moreover, related phyla with intact, organised Hox clusters and spatial collinearity, such as nemerteans and chitons, show annelid-like chromatin organisations, whereas phyla with disorganised13 Hox clusters do not. Coordinated Hox gene regulation from a "global control region" is thus ancestral to Lophotrochozoa, indicating that complex regulatory logics based on cluster-wide, long-range chromatin interactions with distal enhancers evolved convergently in vertebrates and spiralians.

Whether closely related species are repurposed for similar biotechnologies -- a phylogenetic signal in human technological interest -- has lacked a tractable test at scale. We built GRAFT (Graph of Relatedness, Applications, Families and Taxonomy), a Neo4j knowledge graph linking the Open Tree of Life synthetic taxonomy (4.53 x 106 taxa)1 to multilingual common names2,3 and to a Google Patents BigQuery patent layer from a single 257 GB SQL scan, recovering 22,876 species in 759,182 patents with all CPC and IPC class definitions resolved. Treating each species CPC-subclass profile as a binary application vector, we tested the correlation between pairwise topological phylogenetic distance and pairwise Jaccard distance of patent profiles by Mantel test6 (999 permutations, n = 9,944 species at [≥]5 patents, 49,436,596 pairs). The global correlation was significant (Pearson r = +0.188, one-sided p = 0.001), with Bonferroni-significant phylogenetic signal in every close-distance bin from sister-species through within-class. The signal is not an artefact of the unweighted topology: re-expressing phylogenetic distance as time-calibrated divergence from the TimeTree of Life confirms Bonferroni-significant signal in every bin out to [~]500 Myr of divergence. The same graph supports a predictive query that returns sister-species bioprospecting candidates for any application: ten Angelica congeners are unflagged for medicinal preparations while A. sinensis (Chinese angelica) already carries 86,814 such edges. GRAFT is an openly extensible scaffold linking phylogeny, ecology and the global IP record.

Microglia are the resident hematopoietic cells of the central nervous system1. In mice, microglia seed the brain during embryogenesis and can be maintained throughout life with minimal input from adult hematopoiesis2-4. The origins of human microglia are less clear, but recent evidence suggests that marrow-derived cells may be able to supplement the human microglial pool in certain individuals5,6. Here, to investigate the ontogeny of human microglia, we develop a method that uses the collection of accumulated somatic mutations which uniquely labels each clone of cells to track the infiltration of marrow-derived cells into the human brain. Applying this method to 20 aged individuals, we find evidence of an influx of marrow-derived cells into the brain in all examined individuals. Single cell analysis, including single cell lineage tracing using mitochondrial DNA variants, demonstrates that these infiltrating cells are nearly identical to microglia and can comprise a large fraction of the microglial pool. Analysis of large-scale sequencing cohorts demonstrates a protective association between most types of clonal hematopoiesis and Alzheimers disease. In sum, this work uncovers a widespread influx of myeloid cells into the healthy human brain which serves to reinforce the pool of human microglia and becomes common with aging.

Gammaherpesvirus reactivation drives a collapse of host mRNA splicing fidelity that extends across the transcriptome, with exon skipping affecting up to [~]57% of expressed genes, exceeding the effects of depletion of any of 186 splicing factors. Combining five Epstein-Barr virus (EBV) and Kaposis sarcoma-associated herpesvirus (KSHV) reactivation systems across B cell and epithelial models with deep poly(A)+ RNA sequencing of purified lytic cells, we find that most induced isoforms are predicted to undergo nonsense-mediated decay or to lose conserved protein domains, broadly compromising cell cycle, innate immune and RNA-processing pathways. The phenotype arises independently of viral DNA replication, indicating early host remodeling. A screen of EBV early genes identifies BRRF1 as a key driver: through a CIY(Y/E) motif conserved in KSHV ORF49, BRRF1 engages the nuclear CCR4-NOT complex through its CNOT9 and CNOT1 subunits, hijacking this canonically cytoplasmic deadenylation hub for nuclear disruption of host splicing.

Inflammatory bowel diseases (IBD) are modulated by microbiota composition, host genetics, and environmental factors1. Humans and other mammals are colonized by multiple strains of Escherichia coli, a species that expands in abundance in IBD patients2. The state of chronic inflammation characteristic of IBD is expected to intensify selective pressures on the gut microbial community, leading to distinct adaptive trajectories among its constituents. Here we couple in vivo experimental evolution with short- and long-read sequencing to test this hypothesis at the level of mutation and horizontal gene transfer (HGT). By colonizing IL10KO mice, a model of IBD3, and healthy wild-type mice with two strains of E. coli, we show that the tempo and mode of evolution are strain-specific and strongly shaped by host inflammatory status. Unique mutations associate with host inflammatory status independently of microbiota composition, and rates of transfer are diagnostic of chronic inflammation. Extensive transduction events occur in the inflamed gut, giving rise to hybrid clones that form a new genetic lineage, one that becomes dominant in this disease context. The high levels of recombination between prophages uncovered here point to a critical role of HGT and viral evolution in IBD.

The functional organization of human cortex has been mapped in considerable detail1-5, yet the axonal connections linking these areas remain largely unknown. This gap precludes inferring either the computational pathways through cortical networks or the principles governing why each area connects to its particular target regions. Here, in patients undergoing intracranial monitoring for epilepsy, we used concurrent electrical stimulation and functional magnetic resonance imaging (es-fMRI), validated previously against tracer studies in nonhuman primates6, to map anatomical connectivity of cortical sites across the whole brain, and combined these maps with task and resting-state fMRI in the same individuals. Es-fMRI revealed four main findings. First, connectivity followed an asymmetric functional-similarity principle: connected sites tended to share similar functional profiles, but, because connectivity is sparse, most pairs of functionally similar sites were not connected. Second, although connectivity also declines with distance7, the functional profile explained three times as much variance in connectivity as distance did, and long-range connections were the most functionally specific. Third, es-fMRI revealed direct monosynaptic links between established functional regions, including between the fusiform face area (FFA)2 and the temporoparietal junction (TPJ)3 for social cognition. Fourth, resting-state functional connectivity (rsFC) of a stimulation site was only weakly correlated with that sites es-fMRI connectivity. Together, these results provide a first principled link between anatomical connectivity and functional organization in the human cortex, and establish es-fMRI as a scalable approach to building a tracer-grade connectome of the human brain.

The pyruvate dehydrogenase complex (PDHc)1 links glycolysis to the Krebs cycle by catalyzing the oxidative decarboxylation of pyruvate to acetyl-CoA and CO2, a process essential for life2,3. PDHc is formed by structural proteins (E3-binding protein, E3BP)4-7, enzymatic subunits (E1, E2, E3)4,6, and mobile lipoyl domains (LDs), the latter shuttling intermediates across active sites6,8. Although numerous details regarding pyruvate oxidation steps have been elucidated9, the precise organization of the entire PDHc remains unknown due to its large size and dynamic heterogeneity. Here, we employ in silico, in vitro, and in situ methods to propose a multi-scale model of PDHc that includes approximately one million atoms and to visualize multiple conformational states. This model reveals a [~]40-50 nm nested shell structure, formed by flexible linkers that spatially coordinate the E1 and E3 enzyme complexes around the E2-E3BP core scaffold. This structure acts as a molecular sieve, selectively guiding lipoyl arms while maintaining enzyme positioning with sub-nm precision. During catalysis, the nested shell structure expands and adopts a mechanically reinforced state comparable in magnitude to viral assemblies10. Our findings provide structural context for the textbook "link reaction"11, building on decades of biochemical knowledge, are transferable to functional aspects of other -ketoacid dehydrogenase complexes, and, ultimately, expand our understanding of primary metabolism as a whole. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=195 SRC="FIGDIR/small/724543v1_ufig1.gif" ALT="Figure 1"> View larger version (70K): org.highwire.dtl.DTLVardef@1615612org.highwire.dtl.DTLVardef@159bc8dorg.highwire.dtl.DTLVardef@69f8c1org.highwire.dtl.DTLVardef@14a83a3_HPS_FORMAT_FIGEXP M_FIG C_FIG

Dopamine is a neurotransmitter essential for cognition, and its dysregulation is associated with neurological diseases1,2. Historically, dopamine has been understood to signal exclusively through metabotropic receptors3. Delta-type ionotropic glutamate receptors (GluDs), which have recently been established as ligand-gated ion channels4,5, are fundamental for synaptic maintenance, are implicated in neurological disorders, and co-localize with dopaminergic machinery. Here, we report that dopamine is a direct agonist of GluDs, eliciting ionotropic activity, as visualized by cryo-electron microscopy (cryo-EM), bilayer recordings, mutagenesis, and patch clamp recordings. Dopamine binds to the GluD ligand binding domain, inducing clamshell closure and channel activation through a distinct molecular interface. GluD channel activity is tightly regulated by G-proteins, which act as molecular switches to tune GluD activity: free G{beta}{gamma} inhibits ligand-gating, while G or inactive G-protein heterotrimers enable dopamine-induced GluD currents. This tuning of GluD activity by G-proteins is uncoupled in a point mutation associated with neurodegeneration. These findings expand mechanisms of neuronal dopaminergic signaling, uncover how G-proteins tune GluD channel activity, and provide a framework for targeting GluDs in neurological diseases.

Alternative splicing generates extraordinary transcriptomic complexity in the human brain, yet the full-length isoform landscape across human cortical cell types remains uncharted. Combining fluorescence-activated nuclei sorting with long- and short-read RNA sequencing, we generated isoform-resolved transcriptomes for five major lineages of the adult human prefrontal and orbitofrontal cortex: GABAergic neurons, glutamatergic neurons, oligodendrocytes, astrocytes, and microglia. We cataloged over 220,000 full-length isoforms, [~]35-56% previously unannotated; novel transcripts were longer, more exon-rich, and predominantly protein-coding. Contrary to the neuron-centric view of cortical complexity, glial lineages, particularly oligodendrocytes and microglia, emerged as the most isoform-diverse populations in the cortex. Differential transcript usage and dominant isoform switching defined cell identity, with [~]59-62% of differentially regulated transcripts absent from current annotations. Critically, pathogenic variants were enriched >2-fold at novel splice boundaries within disease genes including POGZ, TARDBP, and PLP1, establishing isoform selection as a primary axis of cortical identity and exposing a layer of pathogenic variation invisible to canonical gene annotations.

Neurons in the hippocampal formation are part of the brains cognitive map, representing the spatial structure of the environment through coordinated activity across place cells, grid cells, border cells, head-direction cells and others1-5. Although remapping between environments has been extensively documented6-9, it remains unknown whether transitions between maps reflect unconstrained reorganization or obey a systematic transformation principle. To address this question, we recorded large neuronal populations from the subicular complex and entorhinal cortex in awake, behaving mice navigating environments spanning a wide range of geometries. We asked whether spatial representations across rooms could be related through a shared class of coordinate transformations. Despite pronounced heterogeneity and apparent randomness in single-cell remapping, population-level decoding across environments demonstrated a consistent low-dimensional affine transformation of coordinates, comprising rotation, scaling, shear, reflection, and translation. Thus, what appears as complex remapping at the level of individual neurons reduces to a compact geometric rule at the level of neural assemblies. These results indicate that the hippocampal formation maintains a structured internal coordinate template that is flexibly tailored to environmental geometry. This may serve as the organisms internal model of space.

Environmental exposures are increasingly examined in relation to mental health, yet large-scale epidemiological analyses remain constrained by fragmented geospatial data, heterogeneous spatial and temporal resolutions, and privacy-preserving linkage requirements, limiting systematic investigation of multiple environmental domains at the population level. We present environMAP, a harmonised set of analysis-ready environmental exposure layers derived from open, global sources. environMAP spans the built environment, green and blue spaces, light exposure (solar radiation and night-time light), terrain, weather and extremes, and air pollution. We document data provenance, spatial buffers, preprocessing, projection alignment, and metadata, and provide a reproducible workflow for privacy-preserving linkage to cohort residential locations. To demonstrate utility, we linked environMAP to >200,000 adults in the German National Cohort (NAKO) and summarised self-reported lifetime doctor-diagnosed depression across exposure gradients using sex-stratified descriptive analyses. Gradients were interpretable and broadly consistent with prior evidence, supporting feasibility, scalability, and hypothesis generation. The framework is adaptable to other outcomes, cohorts, and regions.