How myelin rewrites the signal an electrode can see

What the grant proposes, and what it does not

This is an R21, an exploratory grant, which means a proposal rather than a result. It plans a microfluidic tri-culture that promotes directional axonal growth, myelin formation, and pre and post-synaptic compartmentalization, and it will assess axonal structure, axonal transport, nodal architecture, synaptic structure, and electrophysiological status by live imaging, confocal and electron microscopy, and microelectrode-array recordings; a second aim exposes the model to amyloid beta to induce an Alzheimer-like axonopathy.¹

One boundary matters before anything else. The abstract lists array recordings as one modality among several and assigns nodal architecture to microscopy, not to the array. Everything that follows about what an array will and will not detect from a myelinated axon is my own engineering reading from established biophysics, not a claim the grant makes. With no data yet, the useful move is not to grade the proposal but to ask what its stated goal, a myelinating human model read in part by electrodes, does to the instrument.

What myelin does to the current

Myelin is a lipid sheath that oligodendrocytes wrap around the internodes of an axon. Electrically it raises the membrane resistance and lowers the membrane capacitance along each internode, so the regenerative inward current is restricted to the nodes of Ranvier, the roughly one micron gaps between sheath segments. Excitation then jumps from node to node, called saltatory conduction, which is faster and metabolically cheaper than continuous conduction. The speed-up is diameter-dependent rather than a fixed multiplier: myelinated conduction velocity scales roughly with fiber diameter while unmyelinated velocity scales with its square root,³ so for the thin axons typical of human stem-cell cultures the advantage is a factor of a few, widening toward an order of magnitude only for larger fibers.

What an electrode senses follows from that. A microelectrode does not read transmembrane current directly; it reads the extracellular potential those currents set up in the surrounding conductive medium, weighted by distance.² For an unmyelinated axon the transmembrane current is spread along the membrane, so the extracellular signature is a smooth travelling waveform. For a myelinated axon the current is concentrated at the nodes, so the axonal source becomes spatially punctate: strong at nodes and weak between them. It is not exactly zero between nodes, because capacitive current still flows through the sheath, but it is node-dominated. Internode length scales with diameter, on the order of a hundred times the axon diameter, running from a couple of tenths of a millimeter to a couple of millimeters in mature tissue and typically much shorter and more immature in culture.

The strongest case for taking this seriously

The field genuinely is moving to add glia and myelin to human models, and that step changes a real, headline array capability rather than a peripheral one. High-density arrays earned their place partly by resolving axonal propagation and subcellular footprints, and those are exactly the axonal signals that myelination reshapes. So a maturation the field wants, more realistic white-matter biology, collides with a measurement the field prizes. That is a substantive tension and not a quibble, and it is worth working out before the models arrive rather than after.

Where a skeptic should push

Scope the threat honestly. The soma and the axon initial segment, where myelin does not begin, remain large distributed sources, so an array does not go blind; somatic spike detection is largely untouched, and somatic signals dominate array recordings in the first place. The capability that degrades is specifically axonal and propagation-level detection, and even in unmyelinated cultures those signals are already at the microvolt scale and lean on averaging and templates. Myelination makes an already-marginal capability more marginal; it does not abolish recording.

The load-bearing assumption behind the whole argument is that the model myelinates at all. Compact, reproducible myelination of human stem-cell-derived axons in vitro is notoriously difficult, and many such cultures show little or immature myelin. If this model does not myelinate to a meaningful degree, the biophysical concern is moot, so the honest reading is that this is a conditional threat gated on a hard, unproven experimental step. A second caution follows for any temptation to use the array as a myelin readout: conduction velocity depends not only on myelin but also on axon diameter, temperature, and ion-channel maturation, so a velocity change cannot be attributed to myelin without holding those constant.

What myelination hides from a planar array

The non-obvious implication is a registration problem. When axonal current collapses toward the nodes, detecting a myelinated axon is strongly favored by having a node within the small detection radius of a contact, and node positions are set by oligodendrocyte internode length, that is by biology, not by the electrode grid. Aligning axons over electrodes, which the grant's directional microfluidic geometry does well, fixes only the transverse coordinate: the axon runs over the electrode column. It does not by itself control the longitudinal question of whether a roughly one micron node lands over a contact. A rough scaling makes the regime dependence clear: the chance that a node falls within an electrode's detection zone is on the order of twice the detection radius divided by the internode length. For the short, immature internodes typical of in-vitro human cultures, tens of microns and comparable to the electrode pitch, that chance is not small and nodes will often fall near contacts. It is as internodes lengthen with maturation, toward the hundreds of microns and beyond of mature tissue, that registration becomes a genuine lottery and most electrodes come to sit under an electrically quiet internode. Clean single-axon velocity also needs at least two registered nodes. So axon alignment is not the same as nodal registration, and how much that distinction bites depends on how far the model matures.

The way out, and the genuine opportunity, is to stop demanding single-node registration. Microtunnel and microgroove arrays, the classic way to route aligned axons over defined electrodes, recover conduction velocity from bundles by averaging many axons crossing several electrodes, so the ensemble delay across the array yields a velocity without catching any one node. Read that way, a maturing myelinating model turns an array into a functional white-matter assay: conduction velocity measured across electrodes becomes an electrical biomarker of myelination state, complementary to the microscopy the grant already plans and directly relevant to screening remyelination therapies for the Alzheimer and demyelinating indications this work targets. Threat and opportunity share one mechanism: myelin concentrates and speeds the axonal current, which hurts naive single-electrode detection and helps an ensemble velocity measurement.

The threat that deserves a hardware response is drift. A neuron's extracellular footprint, the spatial pattern of amplitude and inter-electrode delay that spike sorting relies on, is not fixed while a culture myelinates. As the sheath forms, amplitudes redistribute toward the nodes and inter-electrode conduction delays shorten, so a footprint template learned early in maturation decays later. This is the same disease as sampling drift in a long recording, arriving from biology rather than from an algorithm: the instrument's model of the signal goes stale because the tissue changed underneath it. The conservative answer is periodic re-templating tied to maturation state, and treating footprint stability as a measured quantity rather than an assumption. Whether myelination raises or lowers the peak amplitude an electrode sees is genuinely geometry-dependent and unsettled; the redistribution toward nodes and the shortening of delays are the robust parts, and they are enough to break a static template.

The bottom line

The established biophysics is not in doubt: myelin insulates the internodes and confines transmembrane current to the nodes, making axonal conduction saltatory and faster and the axonal extracellular source discrete. What is not established here is that this particular human model will myelinate, and that any array signal change could be attributed to myelin rather than to diameter, temperature, or channel maturation. The claim I am making, and labelling as my own, is conditional: if the model myelinates, a planar array's axonal detection becomes increasingly a nodal registration problem as internodes lengthen and its footprint templates drift, while ensemble conduction velocity becomes a usable readout. What would confirm it is paired array and electron-microscopy data showing that nodal position governs axonal detectability, plus measured velocity tracking myelin over maturation. What would break it is a model whose array signals stay soma-dominated and stationary because myelination never reaches the degree that reshapes the extracellular field.