Neuron-driven actuators: sourcing a working neuromuscular junction

Every other actuator on this site is a muscle that you command directly. This one is a muscle commanded by neurons, which you then command — and the extra layer is the entire point.

A neuron-driven actuator gives you a biological control stage sitting between your signal and your force. That stage does things electrodes cannot: it integrates inputs, it has its own dynamics, and it is the natural interface if you eventually want the device to sense as well as act. It is also substantially harder to build than anything else in this hub, and there is no product you can buy that contains one.

What has actually been built

The 2025 result is the one that matters most for sourcing, and it is easy to overlook because it comes from disease modelling rather than robotics. It demonstrates junction formation in 12 days from cryopreserved cells — which is to say, from things you can buy in a vial rather than differentiate yourself. That is the first version of this route that looks like procurement instead of a research programme.

The curare test, and why you should insist on it

Aydin and colleagues’ most important methodological contribution is not the swimmer. It is the control experiment.

Muscle in a dish twitches. It twitches spontaneously, it twitches from stray field, and it twitches when neurons nearby happen to depolarise it non-synaptically. Observing that muscle moves when you stimulate neurons does not establish that a neuromuscular junction exists, and a co-culture with neurons sitting next to muscle is not a motor unit.

The proof is pharmacological. 25 µM curare abolished the contraction. Curare blocks nicotinic acetylcholine receptors at the junction; if the muscle stops responding when you add it, the signal was travelling through a synapse. If it keeps responding, it never was.

Two practical consequences:

If you contract this work out, put the curare test — or an equivalent junction blocker control — in the acceptance criteria. Otherwise your deliverable is “muscle moved when we stimulated neurons”, which is not the thing you paid for.

If you build it yourself, run it before you believe your own result. It is a cheap experiment that prevents a very expensive misunderstanding, and it is the single clearest example on this site of an acceptance criterion that is both meaningful and easy to specify.

What you have to source

Note where the difficulty concentrates. The cells are purchasable and the prices are known — motor neurons are among the better-priced iPSC products, with a real spread from about $333 to $739 per million depending on supplier and pack size, covered in full on the motor neuron sourcing page. The device is quote-only but exists as a real catalogue item.

What you cannot buy is the co-culture protocol, and that is where these projects actually fail.

The medium problem

Motor neurons and skeletal myotubes want different media. Neurons want neurobasal formulations with specific supplements; differentiating myotubes want low-serum myogenic medium. Run either cell type in the other’s medium for long and it declines.

This is the reason compartmentalised devices dominate the field. The OMEGA-NMJ architecture, and the Uzel design before it, separate the neuron chamber from the muscle chamber and connect them through a channel that axons can cross but bulk medium largely does not. That lets each compartment have its own formulation while still permitting the junction to form. It is a genuinely elegant answer to a problem that has no chemical solution.

The consequence for sourcing: do not attempt this in an undivided well to save money on the device. The device architecture is doing load-bearing biological work, not just holding fluid.

Honest difficulty ranking

Set against the other routes in this hub:

The forces column is the honest caution. NMJ-driven biohybrid work reports device behaviour — swimming, contraction, aligned fibre movement — far more often than it reports normalised construct force. That is not an oversight by those authors; the contribution is the neural control, not the output. But it means that if your requirement is a force number, this route has the thinnest supporting literature to specify against, and you should know that before you commit. The force specification reference has what exists.

Whether to build or contract

Contracting is unusually attractive here, more so than for plain muscle work, because the failure modes are subtle and expensive to discover yourself.

eNUVIO advertises contract research services explicitly listing NMJ models (2D-3D) alongside 3D muscle cultures — the same organisation that makes the compartmentalised device. That is the shortest path from nothing to a validated junction, and the natural first call. The broader shortlist is in the CRO directory.

Three things to settle in the statement of work before signing:

1. The junction must be demonstrated pharmacologically, not just observed. Name curare or an equivalent blocker and require the negative control.
2. State whether the deliverable is data or a physical construct. As everywhere in this field, the default deliverable is a report, and a living construct is a separate negotiation.
3. Specify which cells. If you need human motor neurons of a particular genotype, that is a procurement line with a known price and it should be named — not left to the provider’s default, which will usually be whatever they already have growing.

How we can help

We source the motor neurons, the muscle cells and the compartmentalised device against one specification, and pull quotes from the contract laboratories that will attempt the co-culture — and we will get the junction acceptance criterion written into the quote rather than left implicit, because “we co-cultured neurons with muscle” and “we demonstrated a functional neuromuscular junction” are different deliverables that look identical on an invoice.