Nature

Medical large language models (LLMs) have shown promise in medical knowledge retrieval and clinical reasoning. However, their capacity for quantitative reasoning over dense, longitudinal health data remains limited, largely due to the lack of effective training strategies that leverage large-scale, natural-language-based prospective cohorts. To address this, we present HealthFound, a medical LLM trained on 12.4 million self-constructed examples. These include 15-year longitudinal biomedical records from 501,936 participants, as well as datasets for medical knowledge and general reasoning. We propose a novel three-stage training framework comprising progressive curriculum supervised fine-tuning, rejection-sampled retention fine-tuning, and task-verifiable policy optimization. This enables the model to effectively parse heterogeneous quantitative follow-up data and perform robust quantitative reasoning. HealthFound achieves state-of-the-art performance across multiple newly constructed benchmarks. It attains the highest accuracy on eight public medical benchmarks for general medical competence. On 1,162 UK Biobank tasks predicting future disease onset from biomedical measurements, it outperforms the second-best model by over 10 percentage points on average. In two independent external validations (MIMIC-IV and NHANES), it demonstrates superior zero-shot generalization, achieving improvements of over 6% across 484 disease endpoints. The model also achieves top results on novel tasks for literature-derived question answering and quantitative indicator imputation. Furthermore, HealthFound generates interpretable reasoning rationales that are consistent with established biomedical knowledge. Collectively, our results demonstrate that HealthFound enables medical LLMs to evolve from static knowledge repositories into dynamic quantitative reasoning engines over longitudinal health data.

Kinship practices underpin all traditional societies, forming the basis for socially sanctioned reproductive unions, residence patterns and the inheritance of rights and property1,2. Although the relationship between biological relatedness and kinship is not always straightforward, ancient DNA studies are increasingly used to examine the extent to which biological relatedness underpinned social constructs of kinship in prehistoric societies3-6. Here, we report the analysis of genome-wide data for 534 individuals from the Arras Culture of Middle Iron Age northeast England (including 390 from Wetwang Slack, 100 from Pocklington, and 29 from Melton), finding evidence for communities with kinship systems structured along matrilineal lines. At Wetwang Slack, we reconstruct a 13-generation pedigree comprising 195 individuals structured around female-line connections: matrilineal transmissions greatly outnumbered patrilineal ones and male reproductive partners were largely absent from the cemetery, plausibly because they were buried in their own natal communities. Furthermore, the three main sites with robust sample sizes were characterised by non-overlapping dominant mitochondrial haplogroups, implying a maternal clan-based structure. Reproductive unions at Wetwang Slack suggest a recurrent alliance between two dominant maternal descent groups, with members of each group never reproducing with members of their own maternal lineage. Meanwhile, three individuals from lavishly furnished chariot burials at Wetwang Slack were close maternal relatives belonging to a lineage with consecutive generations of close kin unions, a pattern largely absent among other individuals at the site. These results indicate highly distinctive social practices among an elite group embedded in the wider kinship network of the Arras community.

Hepatitis B is a leading preventable cause of liver cancer, and in high-burden settings most chronic infection is acquired around birth. Nigeria carries Africas largest hepatitis B burden. Preventing mother-to-child transmission requires a chain of antenatal steps: testing pregnant women, acting on the result, and vaccinating the newborn. No nationally representative study has assembled these into a single prevention cascade. We built the first national hepatitis B mother-to-child prevention cascade for Nigeria and measured how far it falls from the World Health Organizations triple-elimination targets. We analysed the 2023-24 Nigeria Demographic and Health Survey, the first Nigerian survey to carry both a maternal hepatitis B module and infant birth-dose records. Among 13,341 women with a most recent live birth in the preceding three years, each linked to that childs immunisation record, we estimated coverage at each cascade step (antenatal attendance, hepatitis B testing in antenatal care, result received, and the infant birth dose) using complex-survey methods, and examined patterning by wealth, education, residence, travel time, and region. Although 72% of mothers attended antenatal care, only 12.3% (95% CI 11.4-13.2) were tested for hepatitis B there, and among the few tested almost all received a result: the loss fell at testing, not at result-return. The infant birth dose reached 55.0% (53.1-56.9), 35 points below target, and ran at a level unrelated to maternal testing, consistent with universal, non-targeted delivery; place of delivery, not maternal testing, was the dominant correlate of an infant receiving it. Both steps were steeply patterned: antenatal testing ranged from 2.6% in the poorest wealth quintile to 32.4% in the richest, and was lowest in the North-West, the zone with Nigerias highest prevalence. Nigeria already reaches most mothers through antenatal care but rarely uses that contact to test for hepatitis B, and the universal birth dose is only half delivered. The country is far from elimination, and furthest where the virus is commonest. Integrating hepatitis B testing into antenatal care, made free and weighted toward the highest-burden north, would convert contact already achieved into protection.

Metastasis causes most cancer deaths1,2, yet no recurrent mutation specifically drives it3,4, raising the possibility that metastatic potential is a non-genetic yet heritable cell state. Classic experiments established that metastatically predisposed subclones pre-exist within a tumor and that these predispositions are inherited over many cell divisions5, but what molecular states or factors underlie this predisposition remain unknown. While previous lineage recording studies6,7 mapped how tumors disseminate, their recording sites saturate too quickly to resolve when a lineage branched, or to attribute a state to its founder. Here we show, using a DNA Typewriter lineage recorder8 with nearly 1,000 recording sites in lung cancer cells, that metastatic potential is already present before dissemination, with colonization predicted by a pre-existing glycolytic state and further spread by expression of ENO1, a glycolytic enzyme that also moonlights as a cell-surface plasminogen receptor9. Profiling the pre-transplant cells and the post-transplantation tumors for both their transcriptomes and their lineage recordings, we reconstructed time-resolved lineage trees across three orthotopically transplanted mice. These trees trace each liver metastasis to a single founder of known pre-transplant state, dating each dissemination event from the primary lung. When every clone was scored before transplant against 349 genes recurrently heritable in vitro, both that set and the glycolytic state independently shifted a clones odds of colonizing the lung. At the gene level, sixteen genes were both heritable and predictive of colonization, and ENO1 alone also predicted which established clones spread further. Hypoxia, the program most strongly associated with phylogenetic fitness within the metastases, did not predict colonization when scored before transplant, separating niche-selected traits from the inherited cell state. Metastatic potential in this system is therefore transmitted along the lineage rather than acquired after seeding. Looking forward, we anticipate that time-resolved lineage recorders will enable the separation of the heritable and acquired components of the cellular heterogeneity seen in single-cell studies of tumor progression and drug tolerance.

Light-oxygen-voltage (LOV) domains are blue-light photoreceptors of plants, algae and fungi, and among the most widely used tools in optogenetics. They switch on by forming a covalent thioether bond between a conserved cysteine and their flavin chromophore, in a reaction that needs a proton to cross from the cysteine to the flavin through a pocket containing essentially no water. Its mechanism has been debated for two decades1, and because the chemistry is over within a microsecond its elementary steps have stayed hidden. Here we combine 10 time-resolved serial femtosecond crystallography snapshots and infrared spectroscopy with QM/MM calculations to resolve the entire sequence of events at 1.4 [A] resolution: from excited-state distortion of the flavin ring (10-100 ps), through hydration of a surface channel (10 ns) and a single ordered water reaching the active site as the reactive cysteine shifts between its conformations (100-500 ns), to the thioether bond itself, caught half-formed at 1 {micro}s (half the molecules reacted, half still poised) and complete at 10-100 {micro}s. That water bridges the cysteine and the flavin and shuttles the proton, lowering the barrier from [~]35 to [~]15 kcal/mol and accelerating the reaction by roughly fourteen orders of magnitude (without it, the half-life would be [~]237,000 years), then departs before the bond forms. Proteins can therefore hydrate a dehydrated active site transiently and on demand to overcome otherwise prohibitive reaction barriers, a catalytic strategy that reaches well beyond photoreceptors.

The complete cell lineage of C. elegans, mapped over four decades ago, was tractable because the animal is small, transparent, and lineage invariant1. Most animals are none of these, having orders of magnitude more cells, being opaque, and developing with substantial stochasticity2. Since 2016, genome editing-based lineage tracing has opened the door to dense cell lineage reconstruction in such organisms3-8, but delivering on that promise has proven technically challenging. Here we apply DNA Typewriter9-11, a prime editing-based recorder that writes stochastic symbolic insertions to an engineered genomic TAPE in strictly sequential order, to trace mouse development from zygote (E0) to late organogenesis (E13.5). We introduce constructs encoding a prime editor, engineered prime editing guide RNAs (epegRNAs), and Pol-III-driven circularized TAPE RNA (circTAPE) into wildtype zygotes by pronuclear injection (PNI), then assay embryos by single-nucleus transcriptional profiling (sci-RNA-seq3)12 with paired circTAPE recovery. From one E13.5 embryo bearing [~]7 integrations of a constitutively expressed prime editor and 11 integrations of 6-unit circTAPE, we recover [~]1.75M single-nucleus transcriptomes and reconstruct a time-calibrated phylogeny of 1,340,794 annotated cells with parsimony-based node support. The first cell division is marked unequivocally, and although the resulting blastomeres contribute asymmetrically to the embryo proper, they are fate-neutral and serve as internal replicates that reproduce every finding. A modest cohort of pre-gastrulation founders dominates the embryo, with inequality exceeding neutral expectation within one to two cell cycles of founder allocation, yet these founders remain broadly multipotent; a second phase of clonal dominance arises in specific lineages during organogenesis. At the finest scale, sibling cells share cell type 9-fold in excess of chance, reaching 68- to 107-fold for cell types arising from spatially restricted founder pools, while the recent differentiations of organogenesis are legible in the heterotypic structure of terminal clades. From clade co-occurrence alone, we recover germ-layer organization and a dated hierarchy of cell-type couplings with branch points from E8.5 onwards. Finally, by integrating these data with our single-cell time-series of mouse development13, we impute transcriptional states and annotations for the majority of internal nodes and recover established state histories for diverse cell types. All data are made freely available, together with NextCell, an interactive browser for this annotated cellular phylogeny of mouse development from zygote to late organogenesis.

The neural chain of events necessary for movement initiation is poorly understood, particularly for endogenous actions that are self-timed. Here, we introduce a new task that requires mice to self-initiate a movement with high temporal precision in the absence of external go signals. Brief, precisely timed flashes of optogenetic photoinhibition reveal that movement initiation is causally driven by rapid sequential recruitment of neurons in prefrontal cortex and lateral cerebellum.

Functional neural networks emerge as developing neurons form synaptic connections. Revealing how these connections are organized and perturbed in disease requires linking single-neuron connectivity to molecular state across thousands of networks, which is challenging with current methods. We developed COSCO (Connectomics by Sequencing in Cerebral Organoids), combining barcoded monosynaptic rabies tracing with single-cell RNA sequencing to reconstruct over 10,000 networks across organoids. We demonstrate that organoid circuitry self-organizes into stereotyped network motifs governed by cell-intrinsic wiring rules aligning with features of mouse cortical connectivity. Projecting network topology onto transcriptomic space reveals that wiring correlates with cell-adhesion and synaptic gene expression programs. In a Tuberous Sclerosis Complex model, pathological hyperexcitability coincides with topological rewiring: mutant networks become more heterogeneous, driven by the transcriptional distribution of upper layer excitatory cell states. This work establishes cerebral organoids as a model to study self-organization of human connectivity, and its disruption in neurodevelopmental disease. O_FIG O_LINKSMALLFIG WIDTH=199 HEIGHT=200 SRC="FIGDIR/small/755792v1_ufig1.gif" ALT="Figure 1"> View larger version (66K): org.highwire.dtl.DTLVardef@50ade8org.highwire.dtl.DTLVardef@c97320org.highwire.dtl.DTLVardef@1a53a47org.highwire.dtl.DTLVardef@9520ad_HPS_FORMAT_FIGEXP M_FIG C_FIG

During the Last Glacial Maximum (LGM) 26-19 thousand years ago (kya), Europe had three principal refugia not covered by ice: the Iberian, Italian, and Balkan peninsulas1,2. While the Iberian and Italian refugia have been studied with ancient DNA3-7, the legacy of the Balkan refugium has remained a mystery due to a lack of genetic data 30-12 kya. We present genome-wide data from nine newly reported individuals: three Epipaleolithic from Romania, one Epipaleolithic from Bosnia and Herzegovina, three Mesolithic from Greece, one Epipaleolithic from Southern Italy, and one Mesolithic from Sardinia. We find that the Epipaleolithic individuals from Romania and the Mesolithic individual from mainland Greece were from a previously unsampled population that was the primary source for later European hunter-gatherers. Westward expansion of Balkan Epipaleolithic people into the Italian Peninsula and mixture with a minority contribution from pre-LGM Italians, produced a population that then spread further west to become the primary ancestry of west-ern European Mesolithic people. Northeastward expansion, bypassing Italy, contributed the European ancestry source of Scandinavian and Eastern European hunter-gatherers. Southeastward expansion to the Aegean led to Anatolian-European mixtures before the spread of farming.

Pooling cBioPortal studies into one patient-by-gene matrix is the standard preliminary to pan-cancer analysis, and the natural join key is the patient. Three Memorial Sloan Kettering clonal-hematopoiesis studies deposit blood normals carrying the same patient identifiers as tumors deposited elsewhere, so a patient-keyed merge records a hematopoietic clone as a mutation in that patients solid tumor. We linked 1,282,483 coding calls across 140 genes from 483 studies to patients, classified every call by specimen type, and rebuilt the matrix with and without blood-derived calls. Blood normals supply 1.7% of calls but concentrate sharply: 52.8% of patients flagged DNMT3A-mutant, 21.3% of TET2 and 19.7% of CHEK2 have no tumor-derived call in that gene, against 0.4% for TP53. The affected calls carry a clonal-hematopoiesis signature: DNMT3A supplies 42.0% of blood-normal and 0.61% of tumor calls, at a median allele fraction of 0.4% against 33.4%. The contamination reaches 65 other study cohorts, almost all of them solid-tumor studies, affecting more than one patient in ten in fifteen of them. It is a property of cohort composition rather than disease: across 45 cancer types the distortion tracks the share of patients with a paired blood normal at r = 0.994, and where one disease carries two unharmonized labels the rates differ up to fivefold before correction and agree after it. Correction lowers the pan-cancer DNMT3A rate by 45% and moves survival estimates further than chance removal of the same patients would. Two filter lines, applied before aggregation, remove it.

Genetic introgression from archaic hominins has profoundly reshaped the genetic diversity and adaptive potential of modern humans, yet the full catalog of introgressed sequences, particularly those residing in structurally complex regions has remained elusive. Here, we present ASMaid (ASseMbly-based archaic introgression detector), a Hidden Markov Model-based framework that leverages haplotype-resolved pangenome assemblies to identify archaic-derived sequences with unprecedented completeness. By integrating both single-nucleotide genotype and structural variation (SV) signals, ASMaid captures significantly more intact archaic segments than conventional reference-based approaches. Applying ASMaid to a global panel of 610 phased human genome assemblies, we show that non-African individuals carry approximately 79.8 Mbp of Neanderthal and 8.3 Mbp of Denisovan sequences, representing substantial increases over previous estimates, respectively. Notably, we detected several centromere-spanning archaic segments, including EAS-specific calls on chromosomes 5 and 7. Our assembly-based approach uncovered 1,701 archaic-derived SVs, revealing a previously overlooked layer of archaic functional legacy. High-frequency introgressed loci are enriched in pathways associated with metabolism, immunity, and nervous system (e.g. CTNNA2 linked to early-onset schizophrenia risk), underscoring the fundamental role of introgression in modulating modern human traits. Notably, we identified dozens of loci potentially facilitating local adaptation, such as PRDM16 involved in adipocyte differentiation and cold tolerance, and CSGALNACT2 associated with chondroitin sulfate synthesis. Furthermore, our analysis delineates three distinct Denisovan introgression pulses in Eastern Eurasian genomes, in which the first two pulses are shared across East Eurasian and Oceanian populations, while the third remain primarily exclusive in East Asians. Reflecting these complex introgression events, 31 Denisovan-derived segments, including the TBX15-WARS2 locus, are inferred to have been introduced via at least two events. This comprehensive map of archaic introgression provides a fundamental resource for understanding how ancient gene flow continuously shapes human phenotypic diversity and adaptation.

In early sepsis the direction in which blood immune-cell transcriptional programmes move may carry prognostic information beyond a single baseline measurement, but whether such trajectory associations survive independent testing is unknown. We scored five immune modules, frozen before analysis, in three public longitudinal whole-blood microarray sepsis cohorts and fitted a logistic model ladder fixed in advance to the change per 24 hours in the two cohorts with mortality data (82 patients, 24 deaths), pooling by inverse-variance fixed-effect meta-analysis with Benjamini-Hochberg control. No association survived correction for multiple testing. The two leading signals were a rising CD4/NK lymphocyte trajectory associated with lower mortality (pooled odds ratio 0.53, 95% confidence interval 0.31 to 0.90) and a rising emergency-granulopoiesis trajectory associated with higher mortality (1.60, 0.92 to 2.79), both per one standard deviation. We then tested both in an independent transcriptomic cohort with serial sampling (63 patients, 15 deaths), scored by the identical frozen method, and against their cell-count analogues in an intensive-care database of 12,607 adults meeting Sepsis-3 criteria, of whom 744 to 4,206 had the serial measurements each analogue required. Independent testing separated the two signals, in the order opposite to the one discovery had suggested. The emergency-granulopoiesis association was reproduced in direction and effect size without reaching conventional significance on its own (validation odds ratio 1.72, 0.92 to 3.20, p=0.088; pooled 1.65, 1.09 to 2.50), was positive in all nine sensitivity analyses, each fixed before the estimates were examined, and was supported by two of its three analogues, including the neutrophil-to-lymphocyte ratio (1.31, 1.20 to 1.42). The CD4/NK association did not reproduce (1.06, 0.59 to 1.89), was null in the window most favourable to it, and received no support from an analogue well powered to detect the discovery effect. The discovery signal that looked most consistent failed independent testing.

Learning an adaptive behavior requires identifying which actions, in which contexts, lead to particular outcomes. This problem of credit assignment is fundamental to both biological and artificial learners1-3. Songbird vocal learning presents a particularly demanding credit assignment problem: singing is controlled by millisecond-precise activity of thousands of motor neurons4-7, but song quality is encoded by a diffuse, delayed dopamine signal with more than an order of magnitude less temporal precision8-12. It is unknown how precisely the songbird brain can drive changes in specific premotor neurons at precise times to improve song performance. Here we show that the cortico-basal ganglia circuit thought to underlie songbird vocal learning can learn with millisecond-scale temporal precision and single-neuron spatial precision. We find that playing disruptive auditory feedback contingent on the activity of a targeted neuron in a vocal variability-generating premotor nucleus at one time in the song causes adaptive changes in the activity of the targeted neuron with 3.2 ms temporal precision. Learned changes in firing rate are not observed in uncorrelated neighboring neurons, thus revealing single-neuron spatial precision. These findings challenge the prevailing view of dopamine-mediated reinforcement as slow and imprecise13-15, and redefine our understanding of the limits of credit assignment in the brain.

Vibriophage VP882 launches its lytic cascade upon detection of a quorum-sensing autoinducer produced by its bacterial host. This capability enables the phage to maximize transmission by transitioning from lysogeny to lysis only at high host cell density, when abundant host cells are present to infect. Here, we show that two different pathways can be triggered upon prophage induction. When the prophage is induced via quorum sensing, Qtip, an anti-repressor, inhibits the cI lysogeny maintenance protein, driving expression of the lysis genes. The phage can also launch a cascade that causes host growth arrest. In this case, cI derepresses two genes, one encoding a DksA homolog, TraRVP882, and one encoding a protein that we name QisA. TraRVP882 and QisA form a complex that causes host growth arrest, relying on RNAP-binding by TraRVP882. Phage VP882 quorum sensing also activates production of a protein we call QtiQ, which inactivates the QisA-TraRVP882 complex, reestablishing host cell growth. Thus, in the presence of QtiQ, the phage lysis program is enacted. To our knowledge, QtiQ is the first protein inactivator of a TraR or DksA-like homolog. By inducing growth arrest, phage VP882 may enable its host to survive under stress-inducing conditions, or the phage may delay host lysis until optimal conditions are met. In both cases, the phage thereby enhances its own prospects for spread.

DNA lesions impede RNA polymerase II, yet the transcriptional consequences of that stress are less understood than the lesions themselves. We reanalysed total RNA and matched whole genomes from 438 liver lesions across four divergent mouse backgrounds, one of them the separate species M. caroli, 428 of them induced by a single dose of diethylnitrosamine (DEN) and 10 arising spontaneously, together with untreated liver and, critically, histologically normal liver from treated, tumour-bearing animals (paired normal liver). Three post-transcriptional changes accompany transformation in every background. Canonical splicing efficiency falls by 3.6 to 4.8 percentage points, and in genes that acquire them, cancer-specific junctions account for a median of 20-25% of junctions and arise at weaker splice signals. Output shifts toward circular RNA: a five-method consensus pipeline identifies 56,418 recurrent backsplice junctions, 9,718 of them detected in all four backgrounds, with ~2-fold enrichment in circRNA abundance and an expanded repertoire of distal configurations. Adenosine-to-inosine editing sites fall 2.2- to 3.8-fold, most steeply at intronic sites. Exposure changes not only these outputs but their coordination. In untreated C57BL/6J liver, splicing efficiency, mean relative backsplice span and editing co-vary positively (pairwise {rho} +0.78 to +0.87). That coordination is lost in exposed tissue, and paired normal liver shows the same loss as tumour. Pooling conditions obscures the effect because opposing estimates cancel. In C57BL/6J, paired normal liver already shows tumour-like splicing, circRNA abundance and editing roughly eight months after one exposure, while backsplice architecture has begun to reorganise. These samples carry a median of 337 callable somatic single-nucleotide variants against 78,843 in tumours and show no elevation of Mki67 or cell-cycle programmes. The post-transcriptional phenotype is therefore a durable property of damaged tissue established before morphological transformation.

The central mechanisms through which glial cells regulate whole-body metabolism remain poorly understood. Here, we identify Notch signaling in hypothalamic oligodendrocyte lineage cells as a previously unrecognized regulator of systemic energy homeostasis. Pharmacological inhibition of the Notch ligands Jagged1 (Jag1) and Jagged2 (Jag2) induces rapid and reversible weight loss across diverse physiological and metabolic contexts independently of toxicity or caloric intake. Single-nucleus transcriptomic analyses identify hypothalamic oligodendrocyte precursor cells (OPCs) as the principal Notch-responsive population following systemic Jag1/2 inhibition and reveal expansion of a metabolically specialized GPR17 intermediate state characterized by enhanced oxidative metabolism and increased predicted communication with hypothalamic neurons. This glial remodeling is accompanied by fasting-like transcriptional reprogramming of AgRP neurons, reorganization of melanocortin-autonomic circuit activity, and activation of peripheral catabolic programs. Importantly, selective deletion of Notch1/2 in hypothalamic OPCs recapitulates the major physiological and metabolic effects of systemic Jag1/2 inhibition, establishing oligodendrocyte Notch signaling as a causal regulator of whole-body metabolism. Together, our findings establish Notch-dependent oligodendrocyte state transitions as a previously unrecognized mechanism linking glial plasticity to systemic energy homeostasis.

Extracellular recordings carry information about neuronal location, and pooling per-unit predictions across neighboring electrodes improves its readout, but whether this local pooling should be learned is untested. We present MYCEL (Message-passing Yields Cellular Embeddings of Location), a graph neural network whose nodes are spike-sorted units carrying waveform and spike-timing features selected on cell-type ground truth, and whose edges encode recording geometry. On the Allen Brain Observatory Visual Coding Neuropixels dataset, learned message passing outperforms fixed aggregation, with the largest gains where single units are least informative. On the International Brain Laboratory Brain Wide Map, MYCEL matches the strongest population model at one-tenth the pairwise interactions. On held-out animals, the embeddings recover laminar depth within visual cortex. Retrained with only the edge geometry changed, the same framework decodes anatomy in macaque visual cortex and human medial temporal lobe. Thus, how neighboring neurons are combined matters as much as the features that describe them.

Population-genetic interpretations of allele frequency commonly begin from a single-origin assumption: that a variant arose once and was subsequently shaped by selection, drift and demographic history. Here we show that for a substantial fraction of human variation this assumption fails, and fails predictably: ectopic gene conversion reintroduces the same nucleotide change generation after generation, at rates strongly structured by two fixed properties of genome architecture, homologous template length and donor-acceptor distance. Across 534 million gnomAD v4.1 variants, this recurrent input accounts for an estimated 4% of the rarest variants and more than 15% of common ones. De novo mutations in 11,963 trios show the process directly: the mutation rate is elevated up to 28-fold precisely where the same variant is already common, with the newly arising allele matching the paralogous donor in 94% of such cases at long templates. The effect is strongest within segmental duplications but extends along every chromosome. Affected positions show reduced linkage disequilibrium and, at common frequencies, are depleted by 30-40% among reported GWAS associations. For this fraction of human variation, allele frequency is encoded in genome sequence architecture rather than set by population-genetic processes alone.

Adaptive control of breathing is essential for life, yet the neural circuits that couple respiration to locomotion and internal physiological state remain poorly understood. Insects regulate gas exchange through spiracles: valve-like openings in the cuticle that allow oxygen uptake while minimizing water loss. We find that spiracle opening in Drosophila is dynamically matched to flight power and reduced by dehydration. We identify the motor neurons that innervate the spiracle muscles and show that their optogenetic activation closes all eighteen spiracles and rapidly limits flight power. Inhibitory interneurons transmit descending flight commands to the spiracle motor neurons, opening spiracles in proportion to metabolic demand. A parallel interoceptive pathway converges on the same interneurons to close the spiracles and suppress flight. Feedforward descending commands thus couple flight power to spiracle opening, while interoceptive feedback pathways close spiracles to balance oxygen supply against water loss. This compact circuit architecture may reflect a general solution for matching respiration to the competing demands of locomotion and water conservation.

Centromeres are essential for chromosome segregation and are epigenetically defined by CENH3/CENP-A nucleosomes. Centromere position along chromosomes varies within and between species, ranging from telocentric to metacentric architectures. Yet, how centromere position influences chromosome inheritance and karyotype evolution remains poorly understood. To reposition the centromere, we used CRISPR-Cas9 to break the centromeric satellite array of Arabidopsis thaliana chromosome 3. Fission chromosomes rapidly acquired telomeres, converting a metacentric into two stable telocentric neo-chromosomes. Neo-centromere formation involved genetic restructuring and de novo satellite higher-order repeat formation, together with epigenetic remodeling of CENH3 and DNA methylation. Crossing the six-chromosome fission line to five-chromosome wild type produced a meiosis-specific trivalent that mis-segregated and generated aneuploidy. Trivalent recombination doubled through two obligate crossovers, which were shifted towards the telomeres. Inheritance was strongly distorted in favor of the wild type centromere, as telocentrics segregated into inviable monosomic gametes. The reciprocal trisomic gametes were associated with centromere-proximal recombination, demonstrating that crossover position governs trivalent segregation. Distortion further increased when CENH3 was over-expressed, implicating centromere strength in the outcomes of trivalent meiosis. Our results reveal rapid centromere remodeling following karyotype change, and how trivalent centromere architecture distorts inheritance, with implications for hybrid incompatibility and synthetic chromosome design.