New Marker Lets Scientists Pull Neuron Vesicles Out of Blood and Spinal Fluid

Extracellular vesicles, the tiny membrane-bound sacs that cells release into blood, cerebrospinal fluid and other biofluids, have long tantalized neuroscientists as a potential window into the living brain. If researchers could reliably pull out only the vesicles shed by neurons, they could in principle read out the molecular state of brain cells from a simple blood draw or spinal tap, without ever touching the organ itself. A new study published in Nature Neuroscience reports a major step toward that goal, presenting a systematic framework for identifying surface markers that distinguish vesicles from specific brain cell types and demonstrating the approach by isolating vesicles carrying the neuron-specific protein NRXN3 from human plasma and cerebrospinal fluid.

The challenge that the team, led by researchers affiliated with the Wyss Institute, Mass General Brigham, Harvard Medical School and the Broad Institute, set out to solve is one of specificity. A milliliter of plasma contains vesicles shed by many different cell types, plus free proteins such as albumin that are many orders of magnitude more abundant than the proteins carried on vesicles. Any attempt to isolate neuron-derived vesicles from this mixture depends on finding a surface protein that is genuinely displayed on the vesicle membrane, accessible to antibodies, and made almost exclusively by the cell type of interest. Candidate proteins that fail on either count can lead researchers to capture soluble protein fragments instead of vesicles, a pitfall the authors had previously exposed for L1CAM, a molecule widely used in earlier neuron vesicle studies.

The researchers distilled their marker criteria into three requirements: the protein must actually be present on vesicles in the biofluid of interest, it must be a transmembrane protein so that antibodies can grab it, and its gene must be expressed specifically in the target cell type. To test the first criterion, they needed a way to catalog vesicle proteins without the overwhelming background of free proteins masking the signal in mass spectrometry. Size exclusion chromatography, the workhorse method for vesicle purification, reduces free proteins by several orders of magnitude but leaves substantial albumin contamination behind, blunting the depth of proteomic analysis.

Their solution was a new purification method they call MMR slurry, built on Capto Core 700, a mixed-mode chromatography resin originally developed for viral purification. The beads have an inert outer shell with pores that admit molecules smaller than 700 kilodaltons, while a core lined with hydrophobic, positively charged octylamine ligands traps any protein that enters. Because vesicles exceed the size cutoff, they are excluded from the beads. The trick is simple: mix the resin directly with the biofluid, let free proteins diffuse into the beads, then spin the beads away, leaving purified vesicles in solution. By adjusting the ratio of resin to total protein, the team could tune the purity. Applied to cerebrospinal fluid in a single step, or to plasma after size exclusion chromatography in a two-step protocol, MMR slurry improved vesicle purity more than 100-fold over size exclusion chromatography alone, as measured with ultrasensitive single molecule array immunoassays comparing the vesicle tetraspanins CD9, CD63 and CD81 against albumin. For plasma, which carries roughly two orders of magnitude more free protein than cerebrospinal fluid, the team also showed that MMR slurry can follow dual-mode chromatography, which combines size exclusion with cation exchange resin to reduce the lipoprotein ApoB-100.

With this high-purity isolation in hand, the researchers generated reference vesicle proteomes from one milliliter each of pooled human cerebrospinal fluid and plasma, detecting 2,104 proteins in cerebrospinal fluid vesicles and 1,862 in plasma vesicles. They also profiled vesicles from ten individual cerebrospinal fluid and plasma samples. Comparing the combined methods with size exclusion chromatography alone on the same plasma samples, the new workflow allowed detection of fourfold more proteins, a substantial gain in proteomic depth for clinically relevant sample volumes.

The second half of the framework was computational. Starting from more than 20,000 human proteins, the team filtered for the roughly 5,200 annotated as containing transmembrane domains, then screened for cell type-specific expression using two complementary strategies. The first drew on bulk RNA sequencing of the five major brain cell types isolated by immunopanning, combined with organ-level data from the GTEx project, scoring each gene with the tau specificity index and requiring scores above 0.7 for both neuron-specific and brain-specific expression. This yielded 165 transmembrane proteins preferentially expressed in neurons and in brain. The second leveraged SCimilarity, a deep learning model applied to more than 22 million cells from 399 single-cell RNA sequencing studies, which assigned each gene an attribution score reflecting how strongly it distinguishes neurons from all other cell types; a cutoff of 0.001 yielded 405 neuron-specific genes, of which 180 encode transmembrane proteins. Crossing these expression screens with the vesicle proteomic data produced 28 candidate neuron-specific vesicle markers from the bulk analysis, 41 from the single-cell analysis and 13 supported by both. The same pipeline was applied to other brain cell types: bulk RNA sequencing identified 38 candidate transmembrane proteins for astrocytes, 21 for oligodendrocytes and 3 for microglia, the latter being relatively scarce because microglia share much of their expression program with other immune cells.

Not every candidate could be tested, so the team prioritized three on the basis of antibody availability: SLC12A5, SLC32A1 and NRXN3. They built custom single molecule array immunoassays for each, cross-testing every pair of commercially available antibodies. To validate the assays, they used induced neurons generated from a human induced pluripotent stem cell line that differentiates rapidly upon doxycycline-inducible expression of the transcription factors neurogenin 1 and 2. Only NRXN3, a member of the neurexin family of neuronal adhesion proteins, behaved as hoped: it appeared in the early, vesicle-containing fractions of size exclusion chromatography for the induced neuron vesicles, cerebrospinal fluid and plasma, whereas SLC12A5 was undetectable in vesicle fractions and SLC32A1 peaked only in the neuron culture vesicles but not in either biofluid.

Because NRXN3 is known to have soluble isoforms generated by alternative splicing and proteolytic cleavage, the researchers attacked the validation from multiple angles. They repeated size exclusion chromatography on individual samples, performed density gradient centrifugation showing NRXN3 co-fractionating with CD9 but not albumin in two batches of pooled plasma, and ran western blots on purified vesicles. Most tellingly, they turned to single-particle fluorescence imaging: vesicles from 18 individual plasma samples and 18 individual cerebrospinal fluid samples were captured on glass with antibodies against the tetraspanins, then probed with fluorescently labeled anti-NRXN3 antibody. Every one of the 36 samples showed NRXN3-positive particles well above isotype control background, including some in which the protein was not detectable by the immunoassay. Colocalization imaging suggested that roughly 4 percent of tetraspanin-positive particles in plasma carry NRXN3. Imaging of plasma from eight patients with Parkinson’s disease and seven controls revealed similar numbers of NRXN3-positive vesicles in both groups, and analysis of an independent single-cell brain dataset confirmed that NRXN3 expression is high across neuronal subtypes. Head-to-head comparisons against other recently proposed markers, ATP1A3 and APLP1, reinforced NRXN3’s advantages: ATP1A3 could not be detected in plasma vesicle fractions by immunoassay or density gradient analysis, and APLP1, while brain-enriched, is expressed in oligodendrocytes as well as neurons and appeared in both vesicle and non-vesicle fractions. In direct imaging comparisons of the four markers, only NRXN3 exceeded background.

The final piece was an optimized immuno-isolation protocol. The team systematically refined antibody conjugation strategy, bead and antibody quantity and their ratio, first in the K562 cell line using the ubiquitous vesicle protein CD81 and then with induced neuron vesicles mixed with vesicles from the parental stem cell line. Even when neuron vesicles made up only one-tenth of the mixture, the protocol captured them specifically. Applied directly to cerebrospinal fluid and plasma without prior vesicle purification, the method achieved nearly complete depletion of target tetraspanins from the flow-through, indicating high efficiency. When the researchers used an anti-NRXN3 antibody on cerebrospinal fluid and plasma, NRXN3 was enriched in the pulldown fraction by western blot and depleted from the flow-through by immunoassay, relative to both isotype and GFP control antibodies, demonstrating immuno-isolation of NRXN3-positive vesicles from both human biofluids.

The authors are careful to frame this as a proof of concept rather than a finished diagnostic tool. NRXN3’s many alternatively spliced isoforms include soluble forms, so targeted mass spectrometry mapping and techniques such as immunogold electron microscopy and super-resolution imaging will be needed to confirm that the epitope used for capture sits exclusively on vesicle-bound protein. It remains to be shown that vesicles isolated with NRXN3 are truly neuron-derived, by demonstrating enrichment for neuron-specific RNA or protein cargo, and that marker levels are robust across ages, disease states and individual differences in blood-brain barrier permeability. There are also topology questions to resolve, since some candidate markers also populate synaptic vesicles where they would face the opposite direction, and the fraction of total cargo carried by the isolated vesicle subset is unknown. Still, the framework itself, spanning computational marker discovery, high-purity vesicle proteomics and validated immuno-isolation, is general and modular. The authors note that the pipeline can be re-run with GPI-anchored proteins in place of transmembrane proteins, that markers of neuronal subtypes could be sought as reference datasets grow, and that the MMR slurry method, which they envision adapting to a high-throughput format, is well suited to proteomic profiling of vesicles from clinical samples for biomarker discovery across cell types and organs.

Subject of Research: Identification and validation of surface markers for isolating neuron-specific extracellular vesicles from human biofluids

Article Title: Identification of candidate markers for the isolation of neuron-specific extracellular vesicles

Article References: Ter-Ovanesyan, D., Whiteman, S., Gilboa, T., Kowal, E. J. K., Trieu, W., Iyer, S., Budnik, B., Babila, C. M., Han, C., Vance, D., Heimberg, G., Burgess, M. W., Keshishian, H., Carr, S. A., Regev, A., Church, G. M., & Walt, D. R. (2026). Identification of candidate markers for the isolation of neuron-specific extracellular vesicles. Nature Neuroscience. https://doi.org/10.1038/s41593-026-02466-y

Image Credits: AI Generated

DOI: 10.1038/s41593-026-02466-y

Keywords: extracellular vesicles, NRXN3, neurons, biomarkers, cerebrospinal fluid, plasma, proteomics, immuno-isolation, single-cell RNA sequencing, MMR slurry, neurodegeneration, liquid biopsy