Communications Chemistry

Linker chemistry and conformation are central determinants of PROTAC activity, shaping ternary-complex geometry, cooperativity, target-lysine presentation and cellular permeability. Existing linker generators often lack explicit control over linker flexibility, require predefined attachment sites and linker lengths, or produce structures that demand substantial geometric correction, limiting their utility in practical PROTAC design. Here we introduce FlexiTAC, a Bayesian flow network that jointly generates linker atom types and coordinates from the warhead and E3-ligase-ligand contexts. We also assemble PROTAC-3D, a quality-controlled collection of 63,554 component-resolved PROTAC structures for model training, and PROTAC-Bench, which covers molecular quality, fragment preservation, geometric fidelity, conformational stability, fragment awareness, rediscovery and sampling efficiency. Compared to the best 3D baseline models, FlexiTAC improves validity by 12.0-12.7% and achieves the highest PoseBusters pass rate of 79.5%-80.0%. A differentiable guidance module shifted generated linkers along a conformational ensemble-derived rigidity axis without retraining the generator. In silico case studies further show that the model can accept crystal-derived, redocked or predicted structural inputs. Together, FlexiTAC, PROTAC-3D and PROTAC-Bench establish an integrated and reproducible framework for data-driven PROTAC linker design, combining controllable structure-conditioned generation with standardized training data and evaluation protocols. This framework expands the linker chemical and conformational space accessible to computational exploration, provides a foundation for future method development and enables the systematic generation of structure-conditioned linker designs with tunable conformational flexibility.

Cereblon (CRBN) is widely used in targeted protein degradation, but its ligand space has remained dominated by a narrow set of cyclic imide chemotypes. Here, we show that the accessible CRBN ligand space extends substantially beyond this canonical space. A high-throughput screen of > 40,000 compounds, followed by orthogonal biophysical validation, X-ray crystallography and SAR analyses, identified several chemically distinct ligand classes. These include linear acetyl-based motifs, a phthalide-derived scaffold, steroidal compounds, and a range of bicyclic ligands. They engage CRBN through distinct recognition modes, several of which deviate from the canonical hydrogen-bonding pattern. Steroidal scaffolds were particularly notable: cortisone binds the human CRBN thalidomide-binding domain with an affinity comparable to thalidomide, with its A-ring occupying the tri-tryptophan pocket in a glutarimide-like orientation despite lacking the canonical imide NH donor. SAR within this series showed substantial tolerance for chemical modification and scaffold simplification, raising the possibility that endogenous steroidal metabolites may contribute to the physiological ligand landscape of CRBN. Bicyclic lactams additionally provided synthetically accessible scaffolds with tunable affinity and promising sites for linker attachment. Across the identified ligand classes, none of the tested representatives induced detectable degradation of canonical CRBN neosubstrates, and several showed largely clean proteomic profiles. Together, these findings broaden the chemical, mechanistic and potential physiological landscape of CRBN recognition and provide diverse starting points for alternative, potentially neosubstrate-sparing CRBN recruiters.

Online structural characterization during microfluidic lipid self-assembly is important for understanding and controlling the formation of nonlamellar liquid crystalline nanodispersions. Here, we report a 3D-printed, X-ray-compatible hydrodynamic flow-focusing microfluidic chip with variable channel dimensions, integrated with synchrotron small-angle X-ray scattering (SAXS), for position-resolved SAXS-on-chip monitoring of Ca2+-triggered hexosome formation. Hexosomes were produced under continuous flow by mixing ethanolic solutions of docosahexaenoic acid monoglyceride (MAG-DHA), the negatively charged phosphatidylglycerol DOPG, and -tocopherol with Ca2+-containing PIPES buffer. Online SAXS-on-chip measurements detected three Bragg reflections characteristic of the internal inverse hexagonal (H2) phase on a tens-of-milliseconds residence-time scale, revealing rapid structural evolution during microfluidic mixing. Complementary ex situ SAXS identified the DOPG/Ca2+ molar ratio as a key parameter modulating the direct vesicle-to-hexosome transformation and the compactness of the internal H2 nanostructures. Dynamic light scattering showed that the flow-rate ratio modulated nanoparticle size, yielding hexosomes with mean hydrodynamic diameters in the range of approximately 120-175 nm and polydispersity index values down to 0.14 at a total flow rate of 200 {micro}L min-1. Cryo-TEM revealed coexistence of hexosomes and vesicular nanostructures, highlighting morphological heterogeneity, while Coarse-Grained Molecular Dynamics simulations supported a central role of Ca2+-DOPG association in promoting a direct lamellar-H2 phase transition. Overall, this work shows that 3D-printed SAXS-compatible microfluidics can integrate continuous production with online structural characterization, providing a basis for future formulation and process optimization of drug-loaded cubosomes, hexosomes, and related nonlamellar liquid crystalline nanodispersions.

Deep learning has accelerated drug discovery, yet most existing models are trained using in vitro affinity datasets and consequently remain disconnected from the cellular context in which functional ligand-protein interactions occur. This limitation hinders the ability to reflect the complexity of native interactomes and characterize biological responses to molecular perturbation. Here we introduce C-PLANK (Chemi-Proteome Language Attention NetworK), a deep learning framework trained on fragment-protein interactions profiled directly in living cells using fully functionalized fragment (FFF) chemoproteomics. C-PLANK combines physicochemical embeddings with a bilinear attention network (BAN) to model both global cellular context and local residue-atom interactions, generating interpretable interaction fingerprints. Particularly, C-PLANK incorporates Cellular Interaction State Index (CISI), a systems-level evidential metric that contextualizes the biological plausibility of each predicted interaction against the global cellular interaction landscape. Across 431 ligand interactomes curated from eight independent chemoproteomic studies, C-PLANK consistently outperformed current state-of-the-art interaction prediction frameworks under both random and cold-protein evaluation settings. The inferred interaction fingerprints aligned with orthogonal evidence from structure-based pocket predictions, co-crystal structures, and cellular binding-site annotations. C-PLANK further generalized to unseen ligands. In a cellular target-focused discovery campaign, C-PLANK identified a previously unrecognized ligand that was subsequently advanced into an active chemical probe acting as a SIRT3 agonist in cellular assays. By learning directly from cellular chemoproteomics, C-PLANK moves beyond isolated interaction prediction toward cellular interaction-state modelling, establishing a computational foundation for future digital-twin frameworks in drug discovery.

Disulfide-rich peptides possess exceptional potency and selectivity but are often limited by the instability and synthetic challenges associated with native disulfide bonds. Here, we report the design, synthesis, pharmacological evaluation, and structural characterisation of triazole-based peptidomimetics of the -GI conotoxin, a selective antagonist of the muscle-type nicotinic acetylcholine receptor (nAChR). A series of 1,4- and 1,5-disubstituted triazole analogues were prepared entirely on resin using CuAAC and RuAAC chemistry to replace the native Cys3/13 disulfide bridge. Functional evaluation against human muscle nAChRs revealed that 1,5-triazole analogues retained low-nanomolar potency, with the lead mimetic exhibiting activity comparable to native -GI. Cryo-electron microscopy of the lead compound bound to the muscle-type nAChR provided the first structure of a disulfide-isostere peptidomimetic in complex with a membrane receptor. The structure demonstrates that the 1,5-triazole reproduces the native peptide fold with high fidelity while contributing receptor-facing interactions not available to the native disulfide bridge. Molecular dynamics simulations further revealed conserved hydration networks and similar conformational sampling between the native peptide and lead mimetic. Together, these findings establish triazoles as effective disulfide surrogates and provide a structural framework for the rational design of stabilised conotoxin therapeutics.

Simultaneous quantification of monoglycosphingolipid stereoisomers is required to monitor changes in defective enzymatic pathways linked to diseases such as Gaucher Disease, Parkinson's Disease, and Krabbe Disease. Resolution of beta-glucosyl and beta-galactosyl epimers cannot be achieved by standard liquid chromatography, electrospray ionization, tandem mass spectrometry (LC-ESI-MS/MS). Separation becomes possible when field asymmetric ion mobility spectrometry (FAIMS), also known as differential mobility mass spectrometry (DMS), is added as an orthogonal separation technique to LC. FAIMS/DMS separates epimeric ion clusters in a high versus low electric field (separation voltage, SV) then redirects the target epimeric ions to the mass spectrometer through the application of a direct current (compensation voltage, CoV). Resolving SVs and CoVs must be manually determined for each lipid. Manual derivation is a labour-intensive process that requires pure synthetic standards, limiting the number of stereoisomers a user can include in an assay. To address this problem, we introduce here intelligent DMS (iDMS). iDMS is an in silico supervised neural network model that learns the ion mobility relationships between SV and CoV and the monoglycosphingolipid structural features of sugar headgroup, N-acyl chain length, and N-acyl degree of unsaturation. iDMS predicts the SV and CoV combinations capable of resolving any stereoisomer pair from a training dataset of composed of measured signal intensities across a range of SVs and CoVs of 12 lipids. This machine learning alternative to manual DMS optimization promises to accelerate the deployment of multiple-reaction-monitoring mode (MRM) RPLC-ESI-DMS-MS/MS assays for the routine and rapid quantification of biologically relevant monoglycosphingolipid stereoisomers.

Small interfering RNA (siRNA) enables precise post-transcriptional gene silencing for refractory diseases, yet its clinical translation remains limited by the lack of safe and efficient delivery vectors. Inspired by the dissymmetric alkyl chain architecture of natural membrane phospholipids, we designed and synthesized 34 novel ionizable lipids with dissymmetric hydrophobic tails and formulated them into lipid nanoparticles (LNPs). Through systematic physicochemical and biological assessments, we established clear structure-activity relationships and identified two lead LNPs (O14-LNP, H18a-LNP) with superior endosomal escape capacity, enhanced in vivo gene silencing potency, and favorable biosafety relative to the clinical benchmark MC3-LNP. In both streptozotocin-induced and spontaneous db/db type 2 diabetes (T2D) mouse models, lead LNPs delivering ferroptosis-related siRNAs effectively ameliorated glucose and lipid metabolic disorders, restored islet function, and alleviated hepatic steatosis. This study not only lays a theoretical foundation for the rational design of novel ionizable lipids, but also validates the therapeutic potential of siRNA therapy targeting ferroptosis, providing a versatile delivery platform and targeted therapeutic strategy for the treatment of T2D.

Lysine acylations, including lactylation (Klac), are pivotal regulators of cellular physiology. However, their analysis is currently bottlenecked by antibody enrichment strategies that suffer from sequence bias and require milligram-scale protein inputs, severely precluding the profiling of scarce clinical biopsies and rare cell populations. Here we present ChemIntelligence, an acyl-NHS chemistry-empowered derivatization strategy that rapidly generates unprecedented acylation-specific spectral libraries, exemplified by over 2.5x10^9 human Klac peptides, enabling cross-species reference atlases. Integrated with Prosit-based rescoring, these libraries substantially increase Klac identifications across diverse proteomic datasets. Leveraging this spectral resource, we devised ChemIntelligence Scope, a reproducible, multiplexed parallel reaction monitoring (PRM) platform that quantifies hundreds of Klac peptides per injection from as little as ~200 ng of cell lysates, clinical biopsies, and even true single cells - revealing functional Klac signatures inaccessible to conventional methods. The ChemIntelligence pipeline also extends seamlessly to lysine nicotinylation, underscoring its broad adaptability for discovering and profiling new acylations. Together, these chemical and computational advances establish a scalable, antibody-free framework for acyl-proteome mapping that overcomes input constraints and enables deep functional insights from otherwise intractable biological samples.

Recent advances in protein structure prediction, exemplified by AlphaFold, have largely addressed the determination of static structures, one aspect of the protein folding problem. However, predicting folding pathways, by which proteins reach their native states, remains a significant challenge. Here, we present PathFold, a deep learning framework that predicts protein folding pathways directly from sequence information. PathFold leverages an AlphaFold-based module to extract structural information from the sequence and generates a progressive folding trajectory from an extended conformation using a diffusion model. By modeling the full trajectory, it enables prediction of folding intermediates and transition pathways, analogous to those observed in steered molecular dynamics (SMD) simulations. The predicted pathways reveal well-defined intermediates and sequential folding events, and show agreement with experimental folding data, including measured {Phi}-values.

Proteins are dynamic molecules existing in diverse conformational states underlying their biological functions. Although recent approaches have enabled diverse conformational sampling by emulating molecular dynamics simulations, perturbing evolutionary information, or steering internal mechanisms of structure prediction models, predicting conformations resulting from major domain motions or motions that occur over long timescales still remains a challenge. To this end, we introduce GNMCADS, a conformational sampling strategy that enhances the diversity of protein diffusion models by selectively annealing the conditioning signal guided by the intrinsic dynamical organization of the sampled protein. Further, we implement GNMCADS in the diffusion module of AlphaFold3, enabling the generation of diverse protein conformations. When benchmarked across 92 proteins that include 54 class A GPCRs, 15 transporters, and 23 proteins with major domain movements, GNMCADS exhibits improved sampling diversity compared to other current conformational sampling methods.

Amyloid fibrils are highly ordered protein aggregates characterized by a conserved cross-{beta}-sheet architecture despite originating from structurally diverse precursor proteins. Growing evidence suggests that interactions between different amyloidogenic proteins can modulate aggregation pathways through heterologous cross-seeding; however, the influence of seed polymorphism on the structure and biological properties of cross-seeded fibrils remains poorly understood. Here, we investigated the cross-seeding of native human insulin by two structurally distinct polymorphs of hen egg-white lysozyme (HEWL): flexible fibrils (FFs) and rigid fibrils (RFs). Native insulin remained stable under physiological conditions and underwent spontaneous fibrillation only under acidic conditions. In contrast, both HEWL polymorphs efficiently induced insulin aggregation at physiological pH, bypassing the nucleation barrier. Thioflavin T fluorescence, circular dichroism spectroscopy, and transmission electron microscopy revealed that lysozyme FFs templated the formation of insulin flexible fibrils (IFFs), whereas lysozyme RFs produced insulin rigid fibrils (IRFs), demonstrating that the structural characteristics of the parental HEWL polymorphs were propagated during heterologous cross-seeding. The toxicity of the resulting insulin fibrils was evaluated in SH-SY5Y neuronal cells and CCF-STTG1 astrocytes. IFFs exhibited minimal cytotoxicity and only subtle morphological alterations, whereas IRFs caused modest reductions in cell viability accompanied by more pronounced cellular damage. These findings demonstrate that the structural polymorphism of HEWL fibrils governs both the architecture and biological activity of cross-seeded insulin fibrils, highlighting amyloid polymorphism as an important determinant of heterologous amyloid propagation and a potential design principle for engineering functional amyloid-based biomaterials and protein delivery platforms.

Protein synthesis supports most biological processes. In the brain in particular, protein synthesis plays a critical role in physiological and pathological states. Here, we describe Tellurophene-Alkyne Cycloaddition-mediated Amino acid Tagging (TeACAT), a versatile strategy for fast, facile, and flexible tagging of newly synthesized proteins in mice. TeACAT is based on metabolic incorporation of the non-canonical amino acid TePhe into proteins by the endogenous protein synthesis machinery. Due to their high similarity, TePhe can efficiently replace canonical Phe without dietary or genetic manipulation. The subsequent bio-orthogonal reaction of TePhe with either fluorescent dyes or affinity handles enables both visualization and affinity enrichment of proteins synthesized during TePhe exposure. TeACAT is compatible with immunofluorescence for cell-type specific visualization of protein synthesis with subcellular resolution and can be used in conjunction with routine proteomics to identify and quantify newly synthesized proteins. Robust incorporation into the mouse proteome was observed on the scale of hours to days, allowing the interrogation of various biological processes. In summary, TeACAT enables the visualization and quantification of protein synthesis with minimal perturbation for biological discoveries.

Rational design of covalent protein-labeling reagents in complex biological environments requires a molecular-level understanding of how the protein microenvironment governs chemical reactivity; yet, such mechanistic details remain inaccessible to experimental methods alone. In living neurons, Ligand-Directed Acyl Imidazole (LDAI) chemistry has been used to label AMPA receptors as a traceless, affinity-based protein labeling method. Although LDAI labeling reagents have been optimized in the lab, the atomic details of their interactions with the protein and the underlying mechanism remain elusive. In this work, we combined Quantum Mechanical (QM) calculations and molecular dynamics (MD) simulations to propose a detailed reaction mechanism for AMPAR labeling by LDAI reagents and to clarify how the protein microenvironment governs reactivity. Although Lys residues are usually protonated at physiological pH and therefore less nucleophilic in water, our QM results show that Lys labeling is energetically more favorable than competing reactions with Ser or water. MD simulations reveal that PFQX ---the LDAI reagent precursor--- binds dynamically to the GluA2 AMPAR as an antagonist, inducing conformational changes that reshape the local environment of the acyl imidazole (AI) warhead, underscoring that ligand identity strongly affects labeling outcomes. We also identified intra and intermolecular hydrogen bond networks that may contribute to further immobilize and pre-organize the LDAI reagent. Moreover, the probe's chemical nature shapes its interactions with the Ligand Binding Domain (LBD), offering a plausible rationale for the previously experimentally observed ligand-dependent fluorescent response. Taken together, our results establish design principles for exploiting the reagent geometry and binding pocket hydrogen-bonding networks for the rational design of LDAI reagents.

Biological aging is accompanied by systematic changes in epigenetic modifications and chromatin organization. The reversal of the effects of aging, rejuvenation, is experimentally achieved by the transient induction of factors that modify these marks in cells and organisms. Here, we show that key features of rejuvenation experiments emerge from the biophysical interplay between dynamic epigenetic marks and the three-dimensional conformation of chromatin. Using a minimal field theory and molecular dynamics simulations, we show that the system responds in three distinct temporal regimes. The intermediary regime fulfills necessary conditions for successful rejuvenation. In this regime, the system spends time near a separatrix, allowing for high epigenetic plasticity, while memory retained in the chromatin conformation enables restoration of the original epigenetic correlations. Analysis of sequencing data further supports the predicted coupling between chromatin compaction and epigenetic correlations. Our results provide a physical explanation for how rejuvenation may remodel age-associated epigenetic states without irreversibly erasing cellular identity. We identify a general mechanism by which memory stored in a slow structural variable permits reversible remodeling of a faster internal state.

Carbapenems are the most potent {beta}-lactams, key antibiotics for healthcare-associated infections by Gram-negative bacteria and evade hydrolysis by most {beta}-lactamases, but are increasingly threatened by emergence of enzymes exhibiting hydrolytic activity towards them. Of the four recognised {beta}-lactamase subclasses, class A (active-site serine enzymes that hydrolyse {beta}-lactams via a covalent acylenzyme intermediate) is the most widely disseminated and, while the majority of such enzymes react with carbapenems to form long-lasting acylenzyme complexes, several possess carbapenem-hydrolyzing activity (carbapenemases). Here, we investigate the basis for these differences in a panel of class A {beta}-lactamases using molecular dynamics (MD) simulations of the respective acylenzyme complexes and tetrahedral intermediates (TI). The simulations reveal multiple features associated with catalytic activity across the spectrum of enzymes tested, including more extensive interactions of the carbapenem acylenzyme carbonyl and generally increased lifetimes of active site water molecules positioned for deacylation. Analysis of the dynamic trajectories shows carbapenemases to have reduced root mean-squared fluctuation (RMSF) differences between the acylenzyme and TI, that are not limited to the active site, indicating that the acylenzyme complex is pre-organised for reaction in carbapenemases but not in carbapenem-inhibited enzymes. Similarly, Principal Component Analysis (PCA) of acylenzyme and TI dynamics shows greater overlap between the two states in carbapenemases, providing further evidence for acylenzyme pre-organisation. Such simulations may represent an effective computational assay able to identify enzymes with carbapenemase activity at relatively modest computational cost.

Reading both physical and chemical properties of a microenvironment from a single fluorophore remains a challenge. Here we demonstrate that two coexisting molecular equilibria within one near-infrared cyanine, CyC4, encode two mechanistically distinct ratiometric reporting channels. A meso-amino group and a pendant carboxylate form a tunable intramolecular hydrogen bond that toggles the dye between closed (700 nm) and open (780 nm) emissive conformers. Time-dependent density functional theory (TD-DFT) calculations show that the hydrogen bond raises the LUMO and blue-shifts the emission, establishing the 700/780 emission ratio as a local reporter of hydrogen bonding and polarity. Independently, the chromophore self-associates under crowding- and cosolvent-rich conditions into an aggregate with a blue-shifted, H-type absorption signature near 530-540 nm and a distinct emission near 610 nm upon 540 nm excitation. The intensity of this aggregate band relative to the monomer emission (Ra) serves as a ratiometric reporter of crowding and self-association. Because the two channels arise from distinct molecular equilibria (intramolecular hydrogen bonding vs. intermolecular self-association) they are largely decoupled: a glycerol titration series confirms that the self-association channel (Ra) can be moved while the hydrogen-bonding channel stays essentially fixed. Applied to protein-PEG biomolecular condensates, the two ratios move oppositely with increasing salt, showing that the interior's chemical (polarity, hydrogen bonding) and physical (packing, self-association) environments co-vary across the salt series; a single CyC4 measurement thereby maps this coupled microenvironment, providing a general strategy for multiparametric, ratiometric sensing of crowded microenvironments.

Protein aggregation is a context-dependent process in which the molecular environment can influence the properties of the resulting assemblies. In biological systems, these interactions can occur sequentially, as aggregates formed in one cellular or tissue context may encounter different molecular partners and act as seeds in subsequent aggregation events. Here, we used sequential seeding as a controlled experimental model of this temporal and contextual complexity to investigate how prion-like sequences from the gut microbiome modulate amyloid-{beta} aggregation across successive aggregation cycles. Combining kinetic, biophysical, conformational, and toxicity analyses, we show that early interactions with exogenous peptides modify the properties of first-generation A{beta}40- and A{beta}42-derived seeds, resulting in propagated A{beta}42 assemblies with distinct molecular and functional properties. These findings support an Interaction History model in which exogenous sequences bias the emergence of aggregate populations whose properties and subsequent propagation depend on the molecular contexts experienced during earlier aggregation events. Overall, our results present A{beta} aggregation as a history-dependent process and suggest that single-step assays may fail to capture aggregate diversity that emerges across successive aggregation cycles.

Deep learning-based protein structure prediction methods that leverage evolutionary information from multiple sequence alignments (MSAs), exemplified by AlphaFold2, have achieved remarkable accuracy. However, existing methods still struggle to predict challenging proteins, particularly those with novel folds or limited evolutionary information, and to recover alternative conformational states. Here we show that structure prediction models trained under different MSA-depth distributions corresponding to different levels of evolutionary information exhibit complementary generalization behaviors, and that a model trained on a mixture of these distributions can combine their complementary generalization strengths. Building on this insight, we developed ProtMonomer, a deep learning framework trained on MSA-depth distributions representing a broad range of evolutionary information levels to improve structure prediction. Across benchmarks comprising CASP15 targets, non-redundant experimentally determined structures, orphan proteins, and short peptides, ProtMonomer performed comparably to or better than leading methods, including AlphaFold2 and AlphaFold3, with particularly strong performance on challenging targets. For fold-switching proteins, ProtMonomer also recovered alternative conformational states more accurately than AlphaFold2 and AlphaFold3 across diverse homologous sequence sampling strategies. In addition to improving predictive accuracy, ProtMonomer substantially reduced inference cost through an efficient architecture, enabling high-throughput applications. Together, these findings provide insights into the generalization of evolution-informed structure prediction models and support ProtMonomer as an accurate and efficient framework for protein structure prediction.

High-quality experimental datasets that link protein-ligand structures with binding affinity data are essential for developing and evaluating structure-based machine learning methods. To help address this need, we established OpenBind as an open-science initiative to generate large-scale experimental datasets for structure-based AI and molecular discovery. Here, we describe the first public OpenBind release, which, to the best of our knowledge, is the largest public single-target experimental structure-affinity dataset. The dataset focuses on enteroviral 2A protease, comprising 925 crystallographic binding events from 699 compounds and associated affinity measurements for 601 compounds. It combines structures from an initial fragment screen and follow-on molecules, together with affinity data, linking experimentally determined protein-ligand binding modes to biophysical measurements within a coherent antiviral discovery campaign. We used this dataset to evaluate protein-ligand structure prediction, binding-affinity prediction, and virtual screening using representative structure-based methods, including docking and cofolding. This exposed several challenges that are central to practical structure-based modelling: docking performance depends strongly on binding-pocket conformation, poses are difficult to rank, and structure-based affinity prediction remains challenging. Fine-tuning OpenFold3-p2 on the fragment-screen structures substantially improved pose prediction and virtual screening for related follow-on compounds, demonstrating how early-stage experimental structures can support target-specific model adaptation.

Molecular staging of Alzheimer's disease (AD) increasingly defines transition boundaries along single-cell pseudo-progression trajectories, yet whether such boundaries reproduce across brain regions, cohorts and molecular modalities is rarely tested. We present a permutation-controlled audit that combines nine boundary-detection algorithms with a fixed marker panel and four orthogonal reproducibility axes-algorithmic consensus, region, cohort and modality. On synthetic data with planted ground-truth boundaries the audit reaches 100% sensitivity and 94% specificity, rejecting four distinct artefact classes each by a different axis. Applied to the Seattle Alzheimer's Disease Brain Cell Atlas middle temporal gyrus, it localizes a transition that is robust across algorithms and recovered in most cell types but does not generalize: its leading marker is attenuated or absent in prefrontal cortex, entorhinal cortex and cerebrospinal fluid, and an apparent cross-region conservation of glial metabolic genes proves to be a global-expression offset rather than a shared program. The same audit nonetheless certifies an externally validated marker (astrocytic PTGDS) as reproducible across regions and modalities, showing that it separates generalizable anchors from dataset-specific ones rather than rejecting all signals. We provide this four-axis audit as a transferable, code-available standard to apply before a trajectory boundary is read as a biological stage, in AD and other progressive proteinopathies.