Anatomical prediction of synaptic connectivity between filiform sensory afferents and interneurons. (A) A reconstruction of interneuron right (R)10-2 is shown in yellow. Afferent arbors from 12 different filiform hair receptors are shown in other colors. The color of each afferent corresponds to its direction of peak activation. These 12 classes span the range of all different classes of receptor directional sensitivities. Inset cartoon shows the color code indicating the preferred stimulus direction with respect to the cricket body coordinates. (B) Composite view (saggital) of 11 different sensory afferents from the left cercus illustrating the continuous representation of direction selectivity within the nervous system. Cells with similar directional tuning arborize near each other and those tuned to other directions are spatially segregated showing their color. (C) Image of the afferent map of air current direction, from both cerci, with an image of the compartmental model of interneuron 10-2 imbedded in the map. Each directional class of afferent arbors is transformed into a `statistical cloud' corresponding to the density of synaptic terminals for that stimulus direction. This provides a direct demonstration of the neural map of direction. The overlap between the sensory interneuron with the afferent map of air current direction predicts synaptic connectivity from the afferents onto that interneuron. Here we just mask the interneuron dendrites with the color corresponding to the statistical cloud of afferent synapses in that region. (D) Image of the distribution of synaptic varicosities of the population of sensory afferents from the left cercus tuned to different air current directions from the left cercus. Same view as in B. The varicosities form a continuous three-dimensional structure in the neuropil. Note that the peak directional tuning of the varicosities changes continuously with location around the structure. Starting at the top of the structure (pink) and moving clockwise [red, yellow (out of view), green and blue].

Directional tuning curve for interneuron 10-2a. (A) Single puffs of air from eight different directions relative to the cricket (top traces) elicit various patterns of spiking activity (bottom traces) in an interneuron of class R10-2a. Scale bar: x 10 ms, y 875 mm s^–1 (stimulus)/10 mV (intracellular membrane potential). (B) To generate the tuning curve the same cell as in A was presented with 10 stimuli from each of 24 different directions in the horizontal plane (15° separation between samples). The number of spikes elicited in the 60 ms window following stimulus onset was counted for each trial, and mean and s.d. across trials is shown as a function of stimulus direction. The spontaneous firing rate of the cell was also determined, and the gray broken line shows the expected number of spontaneous spikes in a 60 ms window. Note that stimuli from angles –15° to 105° inhibit the firing activity of this cell below the spontaneous rate, which can also be seen as a slight hyperpolarization in the membrane potentials of A. (C) The mean values from B, plotted in polar instead of Cartesian coordinates.

Directional tuning curves for ventral giant interneurons (vGIs) and dorsal giant interneurons (dGIs). (A) Mean Cartesian tuning curves for interneurons with axons in the ventral group, with amplitude normalized to maximal firing rate. The shaded background represents ±1 s.d. across the populations of specified neurons. Ai: 8-1a (medial giant interneuron, MGI); Aii: 9-1a (lateral giant interneuron, LGI); Aiii: 9-1b; Aiv: 10-1a. (B) Mean Cartesian tuning curves for dGIs, grouped into potential functional units (data format as in A). Bi: 7-2a and 8-2a; Bi: 9-2a and 9-3a; Bi: 10-2a and 10-3a. (C) Representation of peak directional selectivity of all GIs with unimodal directional tuning in relation to the cricket. R, right; L, left.

Stimulus reconstruction and coherence measurements. (A) A 500 ms recording of 10–200 Hz band-passed Gaussian White Noise (GWN; r.m.s.=73 mm s^–1) stimulation (lower panel) and elicited response (upper panel) in a right 10-2a interneuron (same cell as in Fig. 3). (B) Linear kernel obtained from a full 100 s of simultaneously recorded stimulus and response data. (C) Stimulus from A (black) and best linear estimate obtained from stimulus reconstruction using kernel in B (broken blue line). The upper panel shows the full stimulus and stimulus estimate; the lower panel shows both after low-pass filtering below 50 Hz. (D) Upper panel: stimulus–response coherence mean (black line) ±1 s.d. (gray background), calculated over 10 repeats of stimulus. Lower panel: power spectra of stimulus (upper and lower panels calculated from data in A). (E) Stimulus reconstruction using kernel from B on a test data set where the stimulus was drawn from the same statistical distribution as the stimulus in A (upper and lower panels same convention as in C). (F) Simultaneous recording for 500 ms of R10-2a (blue, not the same cell as A) and L10-3a (green) in response to a 10–300 Hz band-passed GWN stimulus (lower trace, r.m.s.=43 mm s^–1). (G) Estimated reconstruction of stimulus in F using combined kernel from R10-2a and L10-3a (upper and lower panels same convention as in C and E). (H) Upper panel: coherence curves from data in F obtained using only cell R10-2a (blue), only cell L10-3a (green), and both cells together as a functional unit (red). Lower panel: power spectrum of stimulus from F.