From the Interface

From neurons in dishes to behaving insects to primates, how do we trade off accessibility and iteration speed against clinical relevance? What model is best suited to which research question?

Photo a micro-electrode array and a neural culture from Potter et al.

For newcomers to BCI research, it might feel like no man's land between Neuralink's implantables and Emotiv's (or openBCI's) wearables. But there are many ways to contribute to BCI, or to learn about the field.

There's a lot that happens at the interface between the wet grey gloop of neurons and the ever-finer wires that prod them listening for signals to pass outwards. The number of components that have to be developed, tested, and perfected before you can build a functional BCI is immense.

This is where model systems come in. The better we understand and develop these components — basic neurobiology, materials and biocompatibility, imaging techniques, stimulation and recording methods, and signal decoding — the faster and safer the path to BCI.

Scientists have done fascinating work with two-dimensional cultures of neurons. In one of the more well-known applications, a group trained a cultured dish of rat neurons to fly a plane in simulations of various weather conditions. Gimmicks aside, growing rat or human neurons on multielectrode arrays provides a way to study the activity of neural networks — albeit artificial ones. A neuronal culture is an elaborate and spontaneously active living neural network, but it is not likely that its organization has any correlations with real-life networks. The neuron culture technique has been standardized down to kits, with open-source software to help analyze the recordings.

This paradigm even has direct commercial potential, with 'wetware' startups like Koniku using these little dishes of lab-grown neural networks for odor detection for security, military, and agricultural applications.

Neurons grown in 3 dimensions using 3D scaffolds or a gel matrix allow even more surface area for recording. This is where these cultures start to be called 'brain organoids', bringing to mind science fiction scenarios. Every model system raises a new set of ethical questions, and these 'brain organoids' come with a particularly large list of issues to consider before you embark.

Human midbrain organoids from Sun et al. 2018

Some techniques that are relevant to today's BCI have been developed in the so-called 'lower animals', notable voltage-sensitive dyes that allow imaging of neuronal activity. Nematodes (Caenorhabditis Elegans) are 1-mm-long worms whose nervous system is completely mapped out -- providing a system to study a rudimentary yet complete network of neurons. The Openworm project is an effort towards creating a fully virtual nervous system. Leeches have also been used for similar experiments, for their accessible, simple, and well-mapped out nervous systems.

Insects models of neuronal stimulation are perhaps more captivating. Backyard brains offer a Roboroach set for high-school-level science experiments. In research labs, simultaneous multiple motor neuron activity has been studied in tethered cockroaches, but it is unclear how much of this research can be translated to humans. Other insects like bees are also been used to study the relationship between behavior and neuronal activity, primarily odor.

The RoboRoach Bundle from Backyard Brains

Rodent models are usually the first stop for testing invasive BCI prototypes— whether they are head-mounted recorders, or sono-optogenetics stimulation systems. Systems that are too large for rodents, like vascular stent electrodes are typically tested in sheep.

Research in primates is pre-clinical. Primate studies are one step away from human implantation, so implants have to be in near-final shape before they reach this stage. Human BCI aside, many seminal demonstrations of motor control, have been done in primates and much of our understanding of the motor and sensory cortices come from research in monkeys.

Model systems relevant to BCI research

Model System	Applications	Relevance to BCI research	Limitations
Single-unit neuron recordings	Study of cell- and molecular-level neurobiological processes in concert with electrical activity.	Fundamental method of recording neuronal activity and form the basis for classic experiments underpinning the neuron doctrine.	Single-neuron recordings are a ‘method’ rather than a ‘model system’
2D neural cell cultures	Study of neurons acting as a collective.	Computational neuroscience starts to get interesting at this level Testing for material biocompatibility, co-adaptation of new materials with neural tissue Early R&D on new stimulation/recording techniques like optogenetics	The organization of an in vitro networks is very different from real-life neural networks. The effects of any perturbations, whether genetic, pharmaceutical or even electrical stimuli are not translatable
3D cell culture	More intricate and longer lasting than 2D cultures, with more neural cell types in the mix	The 3D structure allows more complex versions of the studies done with 2D cultures as there is more than 1 plane/surface for recording	Start to capture popular imagination as ‘brain organoids’, even though there are few similarities to the real brain.
Slices of brain tissue from animals or even humans (ex vivo)	Study of intact neural circuits. Each neuron in the circuit can be studied.	Most useful for studies of the hippocampus, retina or other areas with a known architecture, where many relevant cells circuit can be found in the slice	Only preserves connections along one plane. Important input connections may be lost, forming an incomplete picture
Invertebrates (Nematode worms, leeches, etc.)	Contribute to single-neuron studies or simple networks of a few well-characterized neurons operating in loops	Early R&D on new stimulation/recording techniques like optogenetics Have been instrumental in the development of techniques such as voltage-sensitive dyes	Nervous system is highly accessible, but neuronal behavior is significantly different from humans, research is rarely translatable
Insects (Arthropods)	Most complex models that are available to DIY-ers, or science labs without institutional oversight. Allow studies in ‘behaving’ organisms	Have been used to study the relationship between behavior and neuronal activity – (primarily odor) and stimulus-driven changes. Basic experiments on motor behavior, as insects walk while tethered	Nervous system is highly accessible, but neuronal behavior is significantly different from humans, research is rarely translatable
Rodents	Probably the most widely used model system for testing BCI	Proof of concept/feasibility studies for BCI prototypes, research into materials and stimulation/recording techniques Pre-clinical studies of safety and recording bandwidth Practically every BCI implant is first tested in rats – Utah array, Neuralink and Paradromics (links) Also used for spine and motor cortex interface research	Limited utility for the study of BCI function. As rodent models are relatively accessible within academic settings, neuroscience has a history of over-reliance on these for pre-clinical research, often spending graduate student time on developing hypotheses that are not translatable.
Mammals	Sheep are large enough to fit bulky prototypes	Proof-of-concept to demonstrate recording bandwidth, and long-term studies for of durability Pre-clinical safety studies	Limited utility in the study of BCI functional efficacy
Non-human primates	Precedes human implants, allows studying BCI function through 2-way communication	Pre-clinical studies of motor and sensory cortex, potentially other brain areas	Primates are not used for research until an advanced, functional interface is developed

Our understanding of neuroscience is inseparable from the models and measures we choose. If we look at single-cell recordings we're likely to think on the scale of individual neurons. If we use EEG headsets, we're going to think at the level of surface-accessible brain rhythms and terms of large brain regions. Models dictate the speed at which we iterate, and therefore the rate at which we progress.

Although none of these models will independently amount to a useful BCI, we need them to make parallel progress on all the components of an interface. The BCI community is growing, and each model system is a way for new researchers to get a foothold in the field.