Figure 5. Neural mechanisms for edge orientation processing.

(a) Schematic of a 5 × 5 mm square area on the skin surface. The gray lines and circles represent papillary ridges and mechanoreceptive end organs, respectively. Three colors of filled dots represent the mechanoreceptors (e.g. Meissner corpuscles) innervated by one of three first-order tactile neurons, the shaded area behind subsets of these mechanoreceptors represent subfields and the color-matched contour represents that neuron’s receptive field boundary. (b) Top: Same format as (a) but showing color-coded subfields for 10 first-order tactile neurons. Note the high amount of receptive field overlap and subfield intermingling and that, in practice, even this representation is simplified as any point on the fingertip skin would activate ~36 of the relevant first-order tactile neurons (Vallbo and Johansson, 1984) (20 fast-adapting type 1: FA-1; 16 slow-adapting type 1: SA-1). The two edges (2 mm long) are superimposed on the layout are centered at the same location but differ in orientation by 20˚. Bottom: Activation pattern of the population of neurons in the cartoon above. Neurons are filled if the edge touches any of its subfields and unfilled otherwise. Arrows point to two neurons that change their state for the two edge orientations. (c) Output of our model, relating subfields to the neuronal populations’ ability to signal edge orientation (ordinate) as a function of edge length (abscissa). Here we directly contrast two synthetic populations where: (1) each unit has a uniform receptive field by virtue of being connected to one receptive element the same size as its receptive field and (2) each unit has subfields by virtue of being connected to a random number (2–64) of receptor elements (each 250 μm in diameter). Each simulation was repeated 100 times for each edge length. The lines indicate the mean and the shaded areas represent the 95% confidence interval.

Figure 5—figure supplement 1. Schematic and flow chart of discrimination model.

The top panel shows how the receptive field maps were constructed. For both model types, the receptive field diameters (dRF) were chosen from the same lognormal distribution (see Materials and methods). The model with subfields it was populated by receptive elements (E) each with a diameter of 250 microns. The first receptive element was placed randomly on the boundary of the receptive field and the second receptive element was placed opposite the first element, also on the receptive field boundary. The rest of the receptive elements (total number of receptive elements was random between 2 and 64) were placed randomly in the receptive field and could overlap. The model without subfields had only one receptive element co-aligned perfectly with the receptive field. The middle panel shows how the population was constructed. For both models, receptive fields were randomly placed over a virtual patch of skin such that the average hexagonal distance between receptive fields was ~1 mm. Blue lines show example stimuli both initial and rotated. The red ring illustrates how we quantified the normalization factor (Nc), essentially the sum of all receptive fields within the potential reach of the given stimulus. The bottom panel presents a flow chart of the discrimination algorithm, see Equations 1–3 in Materials and methods for further details.