BCI study of neural constraints on learning in macaque motor cortex, 2014

An animal study of whether the structure of an existing neural network limits which activity patterns a subset of its neurons can produce. Rhesus macaques controlled a computer cursor through an intracortical brain-computer interface, and the experimenters changed how neural activity mapped to cursor velocity. Authors include Patrick T. Sadtler, Steven M. Chase and Byron M. Yu (Carnegie Mellon) and Aaron P. Batista (University of Pittsburgh); Yu and Batista contributed equally as senior authors. Published in Nature, 27 August 2014.

Implant and animals

• Two male rhesus macaques (Macaca mulatta), aged 7 and 8 years.
• 96-channel microelectrode arrays (Blackrock Microsystems) recording from the proximal arm region of primary motor cortex (M1). Animals were head-fixed in a primate chair. The paper says S. I. Ryu and E. C. Tyler-Kabara implanted the arrays.
• Spike threshold set at 3.0 times the RMS voltage per channel; one “neural unit” is the threshold crossings on one electrode. Each day used 85 to 91 neural units.
• Procedures were approved by the University of Pittsburgh Institutional Animal Care and Use Committee.
• The paper does not name the array as a Utah array in the parts read; the catalog’s Utah microelectrode array sheet is the closest device entry.

Results as stated

The animals could readily learn to control the cursor with neural activity patterns that were within the intrinsic manifold, a low-dimensional subspace of the characteristic population activity, but were less able to learn patterns outside it. The authors conclude that on a timescale of hours it is difficult to generate activity patterns inconsistent with the existing network structure.

Limits

Two animals. This is a basic-science BCI learning paradigm, not an assistive device or a human study.

Primary sources

• Sadtler PT et al., Neural constraints on learning, Nature (2014), DOI 10.1038/nature13665; author manuscript PMC4393644.