---
title: "Companion Diagnostic Development for Blood-Based CNS Biomarkers"
url: "https://neurodex.co/companion-diagnostic-blood-based-cns-biomarkers/"
type: "post"
published: "2026-09-30T16:56:47+00:00"
modified: "2026-09-30T16:57:37+00:00"
author: "Elena Karpukhina"
---

# Companion Diagnostic Development for Blood-Based CNS Biomarkers

In oncology, a companion diagnostic usually starts with a piece of the tumor. The pathologist knows the tissue came from the lesion the drug is meant to treat, so the assay only has to answer one question: is the target there? CNS drug development has no equivalent. No sponsor will biopsy the brain to decide trial eligibility, and serial lumbar puncture is too burdensome to screen large populations. The test that decides who receives a disease-modifying CNS drug will almost certainly rely on blood-based CNS biomarkers.

That changes what a companion diagnostic has to prove. A non-invasive CNS biomarker used for patient selection must show that the protein it measures is present, and also that the signal reflects what is happening in the brain rather than in platelets, muscle, liver, or peripheral nerve. Sensitivity gets a blood test through analytical validation. Knowing where the signal came from is what carries it through clinical validation and onto a drug label.

### Key Takeaways

- Blood-based CNS biomarkers used as companion diagnostics are predictive tests tied to one therapy’s label, which makes their evidence burden heavier than that of exploratory trial biomarkers.
- Oncology companion diagnostics benefited from tissue access: the sample’s origin was never in question. Blood tests for CNS targets have to establish origin as part of the evidence.
- Copathology is common in neurodegeneration, so a blood test that detects the targeted protein can still select patients whose decline is driven by something else.
- Cutoffs, pre-analytical handling, reference standards, and assay bridging all have to be settled before the pivotal readout, not after it.
- Neuron-derived extracellular vesicles offer a way to condition a blood measurement on cell of origin, but the capture method itself has to be validated as part of the diagnostic package.

## Can blood-based CNS biomarkers support a companion diagnostic?

Yes, but only if they clear a higher bar than detection. A companion diagnostic is a predictive test tied to one therapy’s label. Blood-based CNS biomarkers used this way must show that the signal reflects brain pathology, predicts treatment response at a locked cutoff, and performs reproducibly across sites.

The distinction matters because most blood-based CNS biomarkers in neurodegeneration trials today serve a different purpose. They characterize a cohort, track disease progression, or show that a drug engaged its target. None of those uses requires the test to decide, patient by patient, who is treated.

### Predictive, not only diagnostic

The FDA–NIH Biomarkers, EndpointS, and other Tools (BEST) framework separates biomarker categories by the decision they support [[1](https://doi.org/10.1177/1535370217750088)]. A diagnostic biomarker confirms that a disease or subtype is present. A prognostic biomarker estimates the likelihood of a clinical event. A predictive biomarker identifies people who are more likely than others to respond, favorably or unfavorably, to a specific medical product. A companion diagnostic is the device form of a predictive biomarker, co-developed with the drug and referenced in its label. Its job is to separate likely responders from likely non-responders, and the trial that supports the label must show that the separation holds.

This is a harder bar than disease detection. A blood test can classify amyloid status accurately and still fail as a companion diagnostic if amyloid positivity does not predict who benefits from a given mechanism.

### Where a companion diagnostic differs from a pharmacodynamic marker

Pharmacodynamic markers answer a different question: did the drug reach and modify its target? They are measured after dosing, compared within the same patient over time, and are allowed to be exploratory through much of development. A companion diagnostic is measured before dosing, compared against a fixed cutoff, and carries regulatory consequences from the day the cutoff is locked. The same blood-based CNS biomarker can serve both roles, but the validation packages are different and should be planned separately. For the pharmacodynamic side of this question, see our analysis of [measuring CNS target engagement from blood]().

## Oncology’s advantage: the tissue told you where the signal came from

The companion diagnostic model matured in oncology around markers such as EGFR, HER2, and PD-L1, where predictive testing now shapes treatment selection across several tumor types [[2](https://doi.org/10.1002/mco2.70638)]. Those tests are frequently described as triumphs of assay design, and they were. But they also rested on an advantage that is easy to overlook: the sample came from the disease.

An immunohistochemistry slide cut from a tumor biopsy carries its own provenance. When the assay detects receptor overexpression, no one has to argue about whether the signal came from the tumor or from somewhere else in the body. Even oncology’s move to liquid biopsy kept much of that advantage. Circulating tumor DNA carries somatic mutations that exist only in the tumor, so the analyte itself encodes its origin.

Blood-based CNS biomarkers rarely have that property. Tau, alpha-synuclein, TDP-43, and neurofilament are not tumor-specific mutations. They are normal proteins, several of them expressed outside the brain, whose pathological forms differ from healthy forms in conformation, phosphorylation, or location rather than in sequence. A blood measurement of these proteins arrives with no built-in address.

## Why blood-based CNS biomarkers lack a built-in address

Blood is a mixture. Extracellular vesicles and soluble proteins enter circulation from nearly every tissue, and the brain’s contribution is small relative to peripheral sources. Vesicles do preserve molecular information from the cells that released them [[3](https://doi.org/10.1002/jex2.70106)], which is why they are attractive as a diagnostic substrate. But in whole plasma, that information is diluted by everything else the body sheds.

Alzheimer’s disease shows how far a well-chosen analyte can go despite that dilution. Plasma p-tau217 has emerged as the most specific of the blood-based CNS biomarkers for Alzheimer’s pathology. In three cohorts totaling 786 participants, one p-tau217 immunoassay identified abnormal amyloid with an area under the curve (AUC) between 0.92 and 0.96, performance comparable to CSF biomarkers [[4](https://doi.org/10.1001/jamaneurol.2023.5319)]. In May 2025, the FDA cleared the first blood test intended to aid in identifying amyloid pathology in symptomatic adults, a practical lumbar puncture alternative for that one clinical question.

That test is a diagnostic aid, not a companion diagnostic. It helps a clinician evaluate whether amyloid pathology is present. It does not tell anyone which therapy a patient will respond to, and it does not address the pathologies that alpha-synuclein-targeted or TDP-43-targeted programs need to measure.

### Why copathology is a label risk

The deeper problem is that neurodegenerative diseases rarely arrive alone. In a neuropathological study of 766 autopsied individuals across the spectrum of neurodegenerative disease, Robinson and colleagues found alpha-synuclein copathology in 41–55% of Alzheimer’s cases and TDP-43 copathology in 33–40% [[5](https://doi.org/10.1093/brain/awy146)]. The authors noted the implications for trials of monotherapies aimed at a single protein.

For companion diagnostic development, copathology creates a specific failure mode. A blood test can correctly detect the protein a drug targets and still enroll patients whose clinical decline is driven by a different pathology. The drug may clear its target in those patients and produce no measurable benefit. The trial reports a diluted effect, and the diagnostic appears to have failed at prediction even though it performed well at detection. This is one of the quieter mechanisms behind patient selection failure in neurodegenerative trials, and it is covered from the enrollment side in our article on [why patient stratification keeps failing in ALS trials]().

Blood-based CNS biomarkers built for this environment need two things a detection assay does not. They need to reflect CNS burden rather than peripheral pools. And they need enough context, whether from additional analytes or from knowing the cell type the signal came from, to indicate when the targeted pathology is unlikely to be the driver.

## What clinical validation requires of blood-based CNS biomarkers

Analytical validation asks whether the assay measures what it claims, precisely and reproducibly. Clinical validation asks whether the result predicts the outcome that matters, in the intended population, at a fixed cutoff. For blood-based CNS biomarkers intended as companion diagnostics, five elements tend to decide whether the second question can be answered.

### A cutoff locked before the pivotal readout

A companion diagnostic cannot be tuned after unblinding. The cutoff, and the rules for handling results near it, have to be pre-specified. The p-tau217 data illustrate why this is more than a formality: the assay in that study used a three-range reference, with an intermediate zone between clearly negative and clearly positive results, and that structure held across all three cohorts [[4](https://doi.org/10.1001/jamaneurol.2023.5319)]. A sponsor adopting a similar design has to decide in advance whether intermediate patients are excluded, confirmed with a second test, or enrolled and analyzed separately. Each choice changes both the trial’s statistical power and the language that ends up in the label.

The practical implication is that cutoff selection belongs in Phase II. Running the candidate marker across all Phase II participants, including those enrolled without biomarker gating, produces the effect-by-marker data needed to set a defensible threshold.

### A plan to bridge the trial assay to the final device

The assay used to enroll a pivotal trial is often not the exact assay that is later commercialized. Reagent lots change, platforms are transferred, and protocols are simplified for routine use. Each change requires a bridging study that shows concordance between the clinical trial assay and the final device, run on samples from the trial population. That is only possible if the sponsor banked enough plasma from enrolled participants, under consistent handling, to retest later. Sample retention decisions made in Phase I determine whether bridging is feasible in Phase III.

### Pre-analytical control from the first draw

Blood-based CNS biomarkers are sensitive to handling. The Standardization of Alzheimer’s Blood Biomarkers working group tested the effects of collection tube type, delayed centrifugation, centrifugation temperature, aliquot volume, delayed storage, and freeze–thaw cycles across a panel of Alzheimer’s-related plasma markers, and used the results to build a standardized handling procedure [[6](https://doi.org/10.1002/alz.12510)]. Vesicle-based assays add variables of their own, including particle stability through storage and the consistency of capture chemistry. A companion diagnostic needs one locked pre-analytical protocol applied at every site, and deviations need to be documented so their effect can be tested.

### A reference standard that is honest about its own limits

Clinical validation requires comparison to something. For CNS pathology, the candidates are autopsy, PET imaging, and CSF assays, and each has limits. The CSF alpha-synuclein seed amplification assay is the strongest current reference for synucleinopathy. In the Parkinson’s Progression Markers Initiative cohort, it showed 87.7% sensitivity for Parkinson’s disease and 96.3% specificity against healthy controls [[7](https://doi.org/10.1016/S1474-4422(23)00109-6)]. But positivity fell to 67.5% in LRRK2-associated Parkinson’s disease [[7](https://doi.org/10.1016/S1474-4422(23)00109-6)]. A blood test validated against that reference in a genetic cohort would inherit its blind spots. Sponsors running programs in GBA or LRRK2 carriers need to decide in advance how discordant results will be adjudicated.

### Reproducibility across sites, operators, and lots

A companion diagnostic will be run in laboratories far from the one where it was developed, on patients who were never part of the discovery cohort. Multi-site precision, lot-to-lot consistency, and operator independence all have to be demonstrated under a regulated laboratory framework. Isolation-dependent assays carry a particular burden here, because variation in the enrichment step propagates into every downstream measurement.

## Neuron-derived EVs: tying blood-based CNS biomarkers to their cell of origin

Neuron-derived extracellular vesicles (NDEVs), often called neuron-derived exosomes in the clinical trials literature, approach the provenance problem from the sample side rather than the analyte side. Vesicles carry membrane proteins from the cell that released them. Capturing vesicles by neuronal surface proteins before measuring their cargo conditions the measurement on cell type: the result describes alpha-synuclein, tau, or TDP-43 in vesicles of neuronal origin, rather than the total pool in plasma. In principle, this restores some of what tissue access gave oncology.

### What the clinical data show

The strongest clinical evidence to date comes from synucleinopathy. Jiang and colleagues analyzed 664 serum samples from three European cohorts and found that neuron-derived vesicular alpha-synuclein was roughly twofold higher in prodromal and clinical Parkinson’s disease than in multiple system atrophy, controls, or other neurodegenerative diseases [[8](https://doi.org/10.1136/jnnp-2019-322588)]. Performance held across a training group of 314 and a validation group of 105 (AUC 0.86). Combining vesicular alpha-synuclein with clusterin separated Parkinson’s disease from other proteinopathies with an AUC of 0.98 and from multiple system atrophy with an AUC of 0.94 [[8](https://doi.org/10.1136/jnnp-2019-322588)].

For companion diagnostic purposes, the differential result matters more than the headline accuracy. A synuclein-targeted drug needs a test that distinguishes neuronal synucleinopathy from oligodendroglial disease and from non-synuclein proteinopathies, because those patients would dilute a trial. That is a selection question, not a detection question.

### The capture method is part of the evidence

The field has also learned that neuronal origin cannot be assumed. In 2021, Norman and colleagues reported that a surface protein widely used for neuronal vesicle capture was found predominantly in soluble form in plasma and CSF, rather than on vesicles, and recommended against its use for NDEV isolation [[9](https://doi.org/10.1038/s41592-021-01174-8)]. The finding did not show that neuronal vesicles are absent from blood. It showed that any capture antigen has to be validated for vesicle association and neuronal specificity before its results can support a clinical claim.

For a companion diagnostic, that validation becomes part of the submission. Regulators will ask what the test captures, how the sponsor knows, and how consistent the capture is across lots and sites.

Subsequent work addressed the question directly. A 2023 study using ExoSORT™ immunocapture screened multiple neuronal antigens and selected alternative capture targets, then confirmed vesicle identity by electron microscopy, western blotting, and proteomics [[10](https://doi.org/10.20517/evcna.2023.13)]. One result from that study captures the provenance argument in a single data point. Vesicular proBDNF was lower in early Alzheimer’s disease than in controls (P = 0.002), while the same comparison in unprocessed plasma showed no difference [[10](https://doi.org/10.20517/evcna.2023.13)]. The disease-relevant signal was present in the neuronal fraction and invisible in the whole-plasma measurement. The study also reported that proBDNF measured in two independently stored aliquots of the same samples correlated with R² = 0.7, an early indication of measurement stability.

In 2026, NeuroDex co-authored a protocol with investigators at Columbia University, Johns Hopkins, and the Harvard T.H. Chan School of Public Health describing immunoaffinity enrichment of tissue-specific vesicles from human plasma across nervous system cells, lung alveolar cells, and hepatocytes [[11](https://doi.org/10.1002/jex2.70171)]. The work confirmed tissue specificity for vesicle populations derived from neurons, astrocytes, and microglia. The authors were also explicit about a limit: vesicles from a given tissue are heterogeneous, so any immunocapture strategy enriches a subset of that tissue’s vesicles rather than all of them [[11](https://doi.org/10.1002/jex2.70171)].

### What has not been shown yet

No NDEV marker has yet been validated prospectively as a predictive biomarker in an interventional trial, with a locked cutoff and a treatment-by-marker interaction as the primary evidence. The questions that remain are the ones this article has outlined: longitudinal stability within patients, precision across sites, bridging between assay versions, and concordance against imperfect reference standards. These are answerable questions, but they are answered only in drug programs that plan for them.

## Building blood-based CNS biomarkers into a drug program from Phase I

Most companion diagnostic problems in CNS development are sequencing problems. The decisions that determine whether blood-based CNS biomarkers can support a label are made early, often before anyone is thinking about a label. A sponsor considering biomarker-selected enrollment should settle five things while the program is still small.

**Define the intended use.** Specify who will be tested, at what disease stage, and what decision the result drives. A test to enrich a prodromal population faces different evidence requirements than a test to exclude copathology in symptomatic patients.

**Lock sample handling at first-in-human.** Every plasma sample collected from Phase I onward should follow one written protocol, with enough volume retained for later retesting and bridging.

**Measure the candidate marker in every Phase II participant.** Even when enrollment is not biomarker-gated, retrospective measurement across the whole population produces the effect-by-marker data needed to choose a cutoff.

**Measure the competing pathologies too.** A panel that reads the targeted protein alongside the most common copathologies, in the same cell-type-specific fraction, gives the sponsor a way to identify patients unlikely to respond for reasons unrelated to the drug.

**Engage regulators before the pivotal design is fixed.** Pre-submission discussions on intended use, reference standard, and cutoff strategy are far less costly than a post-hoc argument about a diagnostic that was never designed to carry a label. For the isolation-method side of these requirements, see our overview of [what Phase II CNS trials require from an NDEV isolation method]().

ExoSORT™ was built for the part of this problem that blood-based CNS testing cannot avoid: isolating a neuronal fraction from plasma so that downstream measurements describe the cell type the drug is meant to reach. Sponsors working through companion diagnostic planning can discuss sample strategy and panel design with our team at any stage of development.

The first companion diagnostic built on blood-based CNS biomarkers is likely to appear in a program where copathology is common, tissue is inaccessible, and the drug’s mechanism is specific enough that the wrong patients would erase its effect. Assay sensitivity will not be the deciding factor in that program, because several platforms can already detect CNS proteins in blood. The deciding factor will be whether the sponsor can show that the signal came from the brain, and that it came from the pathology the drug was designed to treat.

#### References

1. Califf RM. Biomarker definitions and their applications. *Exp Biol Med (Maywood)*. 2018;243(3):213–221. [https://doi.org/10.1177/1535370217750088](https://doi.org/10.1177/1535370217750088)
2. Wu Y, Xue R, Luo X, et al. Companion diagnostics in clinical therapy: current applications and future directions. *MedComm*. 2026;7(3):e70638. [https://doi.org/10.1002/mco2.70638](https://doi.org/10.1002/mco2.70638)
3. Pierri B, Eitan E, Witwer KW, Re DB, Baccarelli AA, Wu H. Tissue-specific extracellular vesicles enriched from circulation: exploring the liquid biopsy perspective. *J Extracell Biol*. 2026;5(2):e70106. [https://doi.org/10.1002/jex2.70106](https://doi.org/10.1002/jex2.70106)
4. Ashton NJ, Brum WS, Di Molfetta G, et al. Diagnostic accuracy of a plasma phosphorylated tau 217 immunoassay for Alzheimer disease pathology. *JAMA Neurol*. 2024;81(3):255–263. [https://doi.org/10.1001/jamaneurol.2023.5319](https://doi.org/10.1001/jamaneurol.2023.5319)
5. Robinson JL, Lee EB, Xie SX, et al. Neurodegenerative disease concomitant proteinopathies are prevalent, age-related and APOE4-associated. *Brain*. 2018;141(7):2181–2193. [https://doi.org/10.1093/brain/awy146](https://doi.org/10.1093/brain/awy146)
6. Verberk IMW, Misdorp EO, Koelewijn J, et al. Characterization of pre-analytical sample handling effects on a panel of Alzheimer’s disease–related blood-based biomarkers: results from the Standardization of Alzheimer’s Blood Biomarkers (SABB) working group. *Alzheimers Dement*. 2022;18(8):1484–1497. [https://doi.org/10.1002/alz.12510](https://doi.org/10.1002/alz.12510)
7. Siderowf A, Concha-Marambio L, Lafontant DE, et al. Assessment of heterogeneity among participants in the Parkinson’s Progression Markers Initiative cohort using α-synuclein seed amplification: a cross-sectional study. *Lancet Neurol*. 2023;22(5):407–417. [https://doi.org/10.1016/S1474-4422(23)00109-6](https://doi.org/10.1016/S1474-4422(23)00109-6)
8. Jiang C, Hopfner F, Katsikoudi A, et al. Serum neuronal exosomes predict and differentiate Parkinson’s disease from atypical parkinsonism. *J Neurol Neurosurg Psychiatry*. 2020;91(7):720–729. [https://doi.org/10.1136/jnnp-2019-322588](https://doi.org/10.1136/jnnp-2019-322588)
9. Norman M, Ter-Ovanesyan D, Trieu W, et al. L1CAM is not associated with extracellular vesicles in human cerebrospinal fluid or plasma. *Nat Methods*. 2021;18(6):631–634. [https://doi.org/10.1038/s41592-021-01174-8](https://doi.org/10.1038/s41592-021-01174-8)
10. Eitan E, Thornton-Wells T, Elgart K, et al. Synaptic proteins in neuron-derived extracellular vesicles as biomarkers for Alzheimer’s disease: novel methodology and clinical proof of concept. *Extracell Vesicles Circ Nucleic Acids*. 2023;4:133–150. [https://doi.org/10.20517/evcna.2023.13](https://doi.org/10.20517/evcna.2023.13)
11. Pierri B, Jackson GL, Gololobova O, et al. Immunoaffinity-based protocol to enrich nervous system cell-, lung alveolar cell-, and hepatocyte-derived extracellular vesicles from human plasma. *J Extracell Biol*. 2026;5:e70171. [https://doi.org/10.1002/jex2.70171](https://doi.org/10.1002/jex2.70171)
12.  
13.