Biology subjects

Bourdenx, M.

Publications and source records attributed to Bourdenx, M..

9 recordsLinked to original sources

bioRxiv2026

Trustworthy agentic genomics through versioned skill libraries

Genomics is adopting autonomous AI agents that interpret genomes from natural-language instructions faster than it is building the means to trust them. We report the first large-scale controlled evaluation of where, in an agentic genomic pipeline, correctness must reside for the system to be trustworthy at clinical scale. Using pharmacogenomics, a domain where errors are measurable and sometimes lethal, we benchmarked nine frontier large language models across 44,550 scored evaluations on 110 pharmacogenomic cases, and tested model interpretation of real star-allele diplotypes from more than 7,000 individuals in three ancestrally diverse populations. Trustworthiness proved to be a property of pipeline architecture, not of the model. Letting the model reason was stochastic and unsafe, and grounding it in the correct guidelines by retrieval paradoxically increased lethal-class errors. Encoding the validated decision logic as a versioned skill and executing it as code made the pharmacogenomic mapping exact, auditable and identical across models, confining all residual error to a single input-interpretation step. On individual genomes, unguarded model interpretation degraded along an ancestry gradient; execution removes this gradient from the clinical mapping, relocating it to the auditable completeness of the input caller. This establishes a generalisable, auditable architecture for trustworthy agentic genome interpretation at scale. HighlightsO_LICorrectness must be executed, not reasoned or retrieved, to be trustworthy C_LIO_LIRetrieval raises phenotype accuracy yet increases lethal-class errors; skills do not C_LIO_LIExecution makes the clinical mapping exact and model-invariant; error stays at input C_LIO_LIA deterministic input caller is the predicted route to all-correct emitted answers C_LI In briefCorpas and colleagues show that trustworthy agentic genome interpretation comes not from making language models reason correctly about biology, but from confining them to interpreting input while versioned, validated skills do the reasoning as executed code. Across nine large language models and 110 pharmacogenomics cases, executing the skill makes the clinical mapping deterministic, auditable and model-invariant. SignificanceGenomics is adopting autonomous, language-model-mediated agents faster than it is building the standards needed to trust them. On a pharmacogenomic benchmark with lethal-class consequences, we show that an agents trustworthiness is not a property of the model but of how the agent is constrained: correctness must be moved out of the stochastic model into a versioned skill executed as code, with the model confined to interpreting heterogeneous input. This gives the field a transferable architecture for trustworthy agentic genome interpretation, a predicted route to deploying it so that every emitted answer is correct (execute the validated skill, call the input deterministically, and abstain on the irreducible residual), and a way to develop genomic skills as validated, executable, versioned units rather than prompts. Following a validation framework described elsewhere, we use clinical-grade to mean determinism, auditability, traceability to versioned components and population-invariant performance, all achieved under skill-constrained execution. We distinguish two senses of population performance: the executed clinical mapping is population-invariant by construction, verified across European, Latin American and East African origin individuals, whereas the models interpretation of real, ancestrally diverse diplotypes is not, degrading along an ancestry gradient, which is precisely why the mapping must be executed rather than reasoned. We do not claim full clinical validation, which would additionally require non-canonical inputs, real-world genomic and clinical data, human comparators and multi-site concordance.

genomics↗

bioRxiv2026

Spatially resolved transcriptomic identification of thousands of neurons recorded in vivo.

Transcriptomics has transformed our understanding of the brain, but assigning transcriptomic identities to neurons recorded in vivo remains challenging at scale. Existing platforms can pair transcriptomic identity with two-photon calcium imaging in small populations of approximately 100 neurons, but they require recorded cells to be sparse and therefore cannot be applied to large population recordings. Here, we present coppaFISH 3D, a spatially resolved transcriptomics method, and CASTalign, an in silico alignment framework, which together enable transcriptomic identification of thousands of simultaneously recorded cells. coppaFISH 3D detects hundreds of genes in thick 50m fixed sections while preserving tissue integrity, enabling both 3D registration to in vivo imaging and integration with immunofluorescence labelling. The platform is fully powered by open chemistry and open source software, runs on commodity hardware, and can be performed at very low cost per section. It therefore enables transcriptomic identification of recorded neurons at scale, making it possible to study how transcriptomic identity shapes activity in neural populations.

neuroscience↗

bioRxiv2026

Spatial Single-Cell Proteomics Reveals Molecular Trajectories Of Tangle-Bearing Neurons In Alzheimer's Disease

Neurofibrillary tangles composed of hyperphosphorylated tau are a defining pathological hallmark of Alzheimers disease (AD); however, the pathways and mechanisms associated with the transition from physiological tau to tangle pathology remain unclear. Here, we integrate laser microdissection of post-mortem, fixed human AD brain tissue labelled with an antibody recognizing tangle-associated phospho-tau (AT8) with mass spectrometry-based proteomics, applied to individual neurons and to small neuronal pools. This approach identified [~]2,000 and [~]5,000 proteins, respectively, and enabled direct detection of disease-associated tau phosphorylation sites without prior enrichment. A layered analysis of the proteome of tangle-positive and tangle-negative neurons revealed heterogeneous disease-associated states. Pseudotime analysis, combined with an AI-driven analytical framework, indicates that neurons do not segregate into discrete classes but instead organize along a continuum of proteomic changes that correlate with tau abundance. This organization enabled the construction of a trajectory of pathological neuronal responses that can be resolved within an individual brain. Early stages of this trajectory are characterized by coordinated remodeling of proteostasis networks, including reduced proteasome component abundance and increased lysosomal acidification machinery, followed by disruption of synaptic pathways. Notably, despite extensive proteomic remodeling, neurons bearing tangles show little evidence of activated cell-death programs, suggesting prolonged molecular adaptation rather than acute degeneration. Together, these findings establish a framework for single-cell-resolved proteome analysis of human brain disease in situ and define a continuum of neuronal states underlying tau pathogenesis, revealing early vulnerabilities and adaptive responses during AD progression.

neuroscience↗

bioRxiv2026

Distinct mechanistic pathways of early tauopathy revealed by MAPT mutations

Tau pathology underlies a broad spectrum of neurodegenerative disorders, collectively termed tauopathies, yet these diseases exhibit striking heterogeneity in their biological mechanisms and clinical outcomes. The basis for this heterogeneity remains poorly understood. Here, we address this question using knock-in mouse models expressing two distinct frontotemporal dementia-associated tau mutations to define how different tau variants drive divergent pathogenic programs in vivo. We find that the two mutations give rise to fundamentally different trajectories of tau pathogenesis. One trajectory is marked by progressive tau hyperphosphorylation and cytoskeletal destabilization occurring in the absence of detectable tau seed formation. In contrast, an alternative trajectory is characterized by tau hypophosphorylation, early seed formation, and alterations in nucleotide metabolism and chromatin organization, without overt cytoskeletal disruption. With aging, tau in this latter pathway transitions to a hyperphosphorylated state and forms mature fibrillar aggregates. Genetic enhancement of {beta}-amyloid selectively accelerates fibril formation, particularly in the model exhibiting early seeding. Together, these findings demonstrate that distinct tau mutations can engage separable pathogenic mechanisms, providing a biological framework for the heterogeneity observed across tauopathies, and highlighting the need for mechanism-informed therapeutic strategies and patient stratification. O_FIG O_LINKSMALLFIG WIDTH=199 HEIGHT=200 SRC="FIGDIR/small/705716v1_ufig1.gif" ALT="Figure 1"> View larger version (46K): org.highwire.dtl.DTLVardef@11eb652org.highwire.dtl.DTLVardef@153cc53org.highwire.dtl.DTLVardef@8b670eorg.highwire.dtl.DTLVardef@1afaad_HPS_FORMAT_FIGEXP M_FIG C_FIG Graphical abstractSchematic representation comparing two trajectories of tau pathology. The S305N mutation promotes a 4R isoform shift, cytoskeletal damage and synapse loss, and accumulation of soluble hyperphosphorylated tau. Tau remains soluble even at old ages. In contrast, the P301S mutation generates hypophosphorylated, seed-competent tau that forms fibrils. The effect of amyloid is slow in the S305N, but results in accelerated acceleration of pathology in the P301S. Figure made with Biorender.com.

neuroscience↗

bioRxiv2025

Scalable human neuronal models of tauopathy producing endogenous seed-competent 4R tau

The accumulation of pathological four-repeat (4R) tau is central to several frontotemporal dementia (FTD) subtypes, but human neuronal models amenable to high-throughput screening of 4R tau-targeting therapies remain very limited. To address this, we developed iPSC-derived i3Neuron (i3N) lines expressing >75% 4R tau, driven by FTD splice-shifting mutations (S305N or S305N/IVS10+3). These neurons develop hyperphosphorylated tau and demonstrate somatodendritic mis-localization. Unlike other stem cell models of tauopathy, these i3N neurons develop endogenous seed-competent tau and present pFTAA-positive tau assemblies after 28 days in culture. For scalable screening, we CRISPR-engineered a HiBiT luminescence tag at the endogenous MAPT locus into the S305N/IVS10+3 iPSC line, enabling precise quantification of tau levels and pharmacological responses. The model responded predictably to compounds affecting tau clearance, demonstrating its suitability for drug discovery. Overall, this i3N platform recapitulates key features of 4R tauopathy and provides a robust system to identify therapeutic modulators of pathological tau.

neuroscience↗

bioRxiv2024

Selective suppression of oligodendrocyte-derived amyloid beta rescues neuronal dysfunction in Alzheimer's Disease

Reduction of amyloid beta (A{beta}) has been shown to be effective in treating Alzheimers Disease (AD), but the underlying assumption that neurons are the main source of pathogenic A{beta} is untested. Here we challenge this prevailing belief by demonstrating that oligodendrocytes are an important source of A{beta}, and play a key role in promoting abnormal neuronal hyperactivity in AD. We show that selectively suppressing oligodendrocyte A{beta} production improves AD brain pathology and restores neuronal function in vivo. Our findings suggest that targeting oligodendrocyte A{beta} production could be a promising therapeutic strategy for treating AD.

neuroscience↗

bioRxiv2022

mTORC2 couples fasting to mitochondrial fission

Fasting triggers diverse cellular and metabolic adaptations to facilitate organismal survival1,2. During nutrient deprivation, increases in circulating fatty acids support mitochondrial respiration2. The mechanisms driving mitochondrial adaptations and respiratory sufficiency during nutrient deprivation remain incompletely understood. Here we show that extended periods of fasting, or lipid availability stimulates mTORC2 activity. Activation of mTORC2 and phosphorylation of its target NDRG13 at S336 sustains mitochondrial fission and respiratory sufficiency. Timelapse imaging reveals that wildtype NDRG1, but not phosphorylation-deficient NDRG1S336A mutant, engages with mitochondria to facilitate its scission. Using proteomics, and an siRNA screen, we show that mTORC2-phosphorylated NDRG1 cooperates with the small GTPase Cdc424 and Cdc42-specific effectors and regulators to orchestrate fission. Accordingly, RictorKO, NDRG1S336A mutants, and Cdc42-deficient cells each display mitochondrial phenotypes reminiscent of fission failure. During nutrient surplus, mTOR complexes perform anabolic functions5; however, paradoxical reactivation of mTORC2 during fasting plays an unexpected role in driving mitochondrial fission and respiration.

cell biology↗

bioRxiv2020

CB1 receptor-mediated inhibitory LTD triggers presynaptic remodeling via protein synthesis and ubiquitination

Long-lasting forms of postsynaptic plasticity commonly involve protein synthesis-dependent structural changes of dendritic spines. However, the relationship between protein synthesis and presynaptic structural plasticity remains unclear. Here, we investigated structural changes in cannabinoid-receptor 1 (CB1)-mediated long-term depression of inhibitory transmission (iLTD), a form of presynaptic plasticity that requires protein synthesis and involves a long-lasting reduction in GABA release. We found that CB1-iLTD in acute rat hippocampal slices was associated with protein synthesis-dependent presynaptic structural changes. Using proteomics, we determined that CB1 activation in hippocampal neurons resulted in increased ribosomal proteins and initiation factors, but decreased levels of proteins involved in regulation of the actin cytoskeleton, such as Arp2/3, and presynaptic release. Moreover, while CB1-iLTD increased ubiquitin/proteasome activity, ubiquitination but not proteasomal degradation was critical for structural and functional presynaptic CB1-iLTD. Thus, CB1-iLTD relies on both protein synthesis and ubiquitination to elicit structural changes that underlie long-term reduction of GABA release.

neuroscience↗

bioRxiv2019

Identification of distinct pathological signatures induced by patient-derived α-synuclein structures in non-human primates

AO_SCPLOWBSTRACTC_SCPLOWDopaminergic neuronal cell death, associated with intracellular -synuclein (-syn)-rich protein aggregates (termed Lewy bodies), is a well-established characteristic of Parkinsons disease. Much evidence, accumulated from multiple experimental models has suggested that -syn plays a role in PD pathogenesis, not only as a trigger of pathology but also as a mediator of disease progression through pathological spreading. Here we have used a machine learning-based approach to identify unique signatures of neurodegeneration in monkeys induced by distinct -syn pathogenic structures derived from PD patients. Unexpectedly, our results show that, in non-human primates, a small amount of singular -syn aggregates is as toxic as larger amyloid fibrils present in the LBs, thus reinforcing the need for preclinical research in this species. Furthermore, our results provide evidence supporting the true multifactorial nature of PD as multiple causes can induce similar outcome regarding dopaminergic neurodegeneration.

neuroscience↗