Companion Diagnostic Development on a Blood-Based CNS Biomarker

What does companion diagnostic development actually require?

A companion diagnostic is an in vitro diagnostic device that supplies information essential to the safe and effective use of a specific therapeutic product. Regulators expect it to be authorized contemporaneously with that product, under its own marketing application, with a locked assay, a defined context of use, and analytical validation completed before the pivotal trial reads out.

Each clause in that definition carries weight. FDA’s foundational guidance on in vitro companion diagnostic devices establishes that where a drug’s safe and effective use depends on a test, the agency will generally not authorize the drug without contemporaneously authorizing the test [1]. The co-development guidance goes further, treating the diagnostic and the therapeutic as parallel programs with separate submissions, even when a single sponsor runs both [2]. Most companion diagnostics are regulated as Class III devices and reach market through premarket approval rather than clearance, which sets the documentation burden.

In the European Union, Regulation 2017/746 defines companion diagnostics as devices essential to the safe and effective use of a corresponding medicinal product and places them in the highest risk classes, with notified body involvement and a required consultation with the relevant medicines authority [3]. A program building toward both markets is therefore managing two review processes against one locked assay.

The word that trips up CNS programs is essential. A biomarker that improves trial efficiency is not essential. A test becomes essential when the label says patients should not receive the drug, or should receive it differently, based on the result. That is a claim about a treatment decision, and it has to be defended all the way back to the measurement.

Why no blood-based companion diagnostic exists in neurodegeneration

The field’s regulatory precedents are surrogate and monitoring biomarkers, not companion diagnostics. The clearest example came from ALS. Tofersen missed its primary clinical endpoint in a randomized trial but produced a substantial reduction in plasma neurofilament light chain, and that reduction supported accelerated approval as a surrogate reasonably likely to predict clinical benefit [4]. This was a genuine advance in how a blood measurement can carry regulatory weight in neurodegeneration. It was not a companion diagnostic. Nothing in that label conditions treatment on a test result.

The distinction matters because the analytical requirements diverge. A surrogate endpoint has to move with treatment in a way the agency finds interpretable at the population level. A companion diagnostic has to classify an individual patient correctly, reproducibly, at a fixed cut-off, in whatever laboratory runs the sample. Neurofilament light chain illustrates why that second bar is harder. Blood levels correlate with serum creatinine in older adults, with reported coefficients of 0.50 and 0.56 in two independent groups, and the association survived adjustment for age, sex, and body mass index [5]. Blood volume and body mass shift the same number. A careful appraisal of the biomarker’s role in ALS therapy development described it as both a leading candidate and profoundly misunderstood, precisely because the field has applied it across contexts of use without asking which one the evidence supports [6].

Plasma NfL variability of this kind is manageable in a trial with covariate adjustment and within-patient change. It is not manageable in a labelled cut-off applied to one patient in a clinic, where the ordering physician does not know whether an elevated result reflects motor neuron loss or reduced glomerular filtration. The same logic applies to plasma p-tau specificity, where comorbid conditions shift measured concentrations enough to move patients across diagnostic thresholds [7].

Cerebrospinal fluid does not solve this. CSF biomarker limitations in late-stage trials are practical rather than analytical: two or three timepoints per patient, high screen-failure and consent-refusal rates, and site-level variance in collection technique. A lumbar puncture alternative is what most sponsors want. But swapping compartments without addressing provenance simply moves the specificity problem from an access constraint to a measurement one.

The provenance problem: which tissue produced this signal?

Consider what a reviewer has to accept for a blood-based CNS companion diagnostic claim to hold. The measured analyte must originate predominantly in the target tissue. Its concentration must respond to the pathological process the drug addresses. And the relationship must be stable enough across patients that a single threshold separates those who should receive the drug from those who should not.

Total plasma assays break the first link. Neurofilament light chain, α-synuclein, and TDP-43 all have peripheral sources. An α-synuclein plasma assay measuring the total circulating pool draws from red blood cells and platelets in quantities that dwarf any central contribution. Seed amplification assays circumvent this by measuring an aggregation property rather than abundance, and in cerebrospinal fluid they perform well — 87.7% sensitivity and 96.3% specificity across a 1,123-participant cohort [8]. That performance is compartment-dependent, and it demonstrates the general principle: specificity came from restricting the measurement, not from measuring more sensitively.

Restriction by cell of origin is the other route. Extracellular vesicles carry surface antigens inherited from their parent cell, which makes immunoaffinity capture against a cell-type-restricted surface antigen a way to define the compartment before the analyte is quantified. Peer-reviewed work using single and sequential immunoaffinity capture has demonstrated specificity, efficiency, and consistency in enriching vesicle subpopulations from nervous system cells, alveolar cells, and hepatocytes out of human plasma, characterized against canonical vesicle features and co-localization of tissue-associated and general surface markers [9]. That publication is deliberately careful about what enrichment means: it recovers antigen-positive populations most plausibly derived from the intended tissue, not every vesicle a tissue releases, and it states that conclusive assignment would require additional orthogonal evidence [9]. For regulatory purposes that honesty is an asset. A reviewer can evaluate a bounded claim. A reviewer cannot evaluate a claim that overreaches its data.

The clinical relevance of compartment choice is visible in the TDP-43 literature. In a cohort of 704 patients including 37 genetically and 31 neuropathologically confirmed cases, vesicle-associated TDP-43 levels and vesicle tau isoform ratios separated ALS, progressive supranuclear palsy, and behavioural variant frontotemporal dementia by underlying molecular pathology, with cut-off values that transferred across sub-cohorts [10]. That is a TDP-43 blood test performing a task total plasma measurement has not accomplished. It is also the kind of dataset a companion diagnostic argument can be built on, because the measurement is defined by where it came from.