A possible neural basis of the 3D reference frame transformation for reaching

*G. BLOHM, G. P. KEITH, J. D. CRAWFORD
Centre for Vision Research, York University, Toronto, ON, CANADA.

For an accurate reaching movement, the brain transforms gaze-centered information about the hand and target location (as seen through the eyes) into a shoulder-centered reach plan. This transformation must take into account the complete 3D geometry of the eye-to-shoulder linkage. It remains unclear where and how the brain performs the 3D reference frame transformation for reaching.
To provide insight into the neural properties expected within brain regions performing this transformation, we designed a 3-layer feed-forward artificial neural network. The inputs included: two retinotopic maps providing hand and target direction, two maps of retinal disparity (= right - left eye positions), 3D head and (cyclopean) eye positions and an ocular vergence signal. The output of the network consisted of a 3D cosine-tuned population (125 units) with uniformly distributed preferred directions encoding the shoulder-centered movement plan.
We analyzed the reference frames of the HLU and output units in two different ways by investigating separately their input and output properties. First, we measured the units’ visual receptive fields (RF), and observed how the RF changed with different eye and head positions. Second, we simulated neural micro-stimulation (MS), i.e. we modified the activity of individual HLUs and observed the resulting movement plan while simultaneously changing eye and head position. We found that the reference frames of each HLU and output unit were different at the input (RF) or output (MS) level, e.g. the output units coded a shoulder-centered movement plan, whereas their RF reference frame showed a continuum between gaze- and shoulder-centered coding. Probing horizontal and vertical eye/head movements within the same unit could also lead to very different RF shift patterns, e.g. a horizontal eye movement could induce no horizontal change in RF but a purely vertical RF shift.
This neural network study suggests that the complex 3D reference frame transformation may take place at the level of the individual neuron. The individual units implement different intermediate reference frame transformations in their input/output relationships. These intermediate transformations are combined at the population level through gain-field like modulations of HLUs. Our results may also reconcile previous contradictory neurophysiological findings within the same neural structure examined through single unit recording vs. micro-stimulation.