Excitatory local interneurons enhance tuning of sensory information

Abstract

Neurons in the insect antennal lobe represent odors as spatiotemporal patterns of activity that unfold over multiple time scales. As these patterns unspool they decrease the overlap between odor representations and thereby increase the ability of the olfactory system to discriminate odors. Using a realistic model of the insect antennal lobe we examined two competing components of this process -lateral excitation from local excitatory interneurons, and slow inhibition from local inhibitory interneurons. We found that lateral excitation amplified differences between representations of similar odors by recruiting projection neurons that did not receive direct input from olfactory receptors. However, this increased sensitivity also amplified noisy variations in input and compromised the ability of the system to respond reliably to multiple presentations of the same odor. Slow inhibition curtailed the spread of projection neuron activity and increased response reliability. These competing influences must be finely balanced in order to decorrelate odor representations.

Figure 1

a. Schematic diagram of the antennal lobe network consisting of projection neurons (PNs), inhibitory local interneurons (LNs) and excitatory local interneurons (eLNs). (See text for connection probabilities between different neuron types). A random sampling of neurons receives external input (red arrows). b. Input to the neurons. Each PN, LN, and eLN receives external input with amplitude chosen from a truncated Gaussian distribution. The similarity between odors can be varied by changing the overlap in the input profile to individual neurons. The Gaussian intensity profile of similar odors (compare the blue lines) show a large overlap whereas dissimilar odors (compare blue and red lines) show very little overlap. (Note that the red profile “wraps around” from right to left.).

Figure 2

a. Raster plots showing the effect of lateral excitation and slow inhibition on network dynamics. In the absence of lateral excitation, the neuronal activity is largely driven by the input (top left). Large values of lateral excitation allow input to explosively and unrealistically recruit the entire network (top right). Adding slow inhibition curtails this activity (bottom right). The figure shows both the PNs (n = 300) and the eLNs (n = 50). The width of input to the eLNs is a scaled down from that of the PNs. The trace on the left of the plot shows the maximum amplitude of the input to PNs and eLNs. The onset of the input is indicated by the gray bars along the time axis. Two odor presentations, each lasting 1000 ms, are shown. b. Left panels. Traces show the temporal evolution of the first three principal components generated from peri–stimulus time histograms as a function of increasing lateral excitation (top panel excitation increases from blue to red,  = 0.0002 is constant) and slow inhibition (bottom panel inhibition increases from blue to red,  = 0.0002 is constant). Middle Panels. Normalized amplitude of the traces shown in the left panels as a function of time. Different color traces correspond to different values of lateral excitation (top panel) or slow inhibition (bottom panel). The gray bar indicates an odor presentation. Right panels. The amplitude of the traces shown in the left panels as a function of time. Here the amplitude is not normalized and the responses are shown following odor offset. c. Response distribution of PNs. The proportion of PNs generating a given number of spikes during odor stimulation is shown for different values of lateral excitation and slow inhibition.

Figure 3

a. Each group of panels shows the activity of a representative set of three neurons. The image map shows the spike activity evolving over the duration of the stimulus presentation (500 to 1500 ms) in response to an array of 21 odors averaged over ten repetitions of each odor. The average normalized activity over this duration is shown by the red trace in the panel above. The blue trace shows the normalized amplitude of the input to the neuron for the set of 21 odors. The responses to different combinations of lateral excitation and slow inhibition are shown. b. Complexity of PN responses. The temporal spiking patterns for individual neurons over an array of 21 odors (see panel a for examples) were chosen and the number principal components required to explain 80% of the variance of each such pattern was calculated. For each value of lateral excitation and slow inhibition, this generated 300PNs×10trials = 3000 such numbers estimating the complexity of each PNs response to the array of odors. The normalized distribution of this number is shown as a function of lateral excitation and slow inhibition.

Figure 4

a. The correlation between the 300-D activity vector at the onset of the odor stimulus and the activity at subsequent 50 ms epochs decreases progressively (red trace). The blue trace shows the correlation between the input to the network at the onset of the odor and subsequent points in time. b. Average correlation between 300-D activity vectors over time for similar odors (left panels) and dissimilar odors (right panels) for two values of slow inhibition ( = 0 and 0.0002 µS) and a range of values of lateral excitation. Lateral excitation was required to observe measurable decorrelation. Odor stimulation was applied at 500 msec. Note that correlations start to change before t = 500 msec, because 50 msec time bins were used (see methods). Representative time interval of the correlation coefficient change during odor stimulation includes t = [500 msec, 1500 msec]. c. Average correlation between 300-D activity vectors over time for similar (left panels) and dissimilar (right panels) odors for two values of lateral excitation ( = 0 and 0.0002 µS) and a range of values of slow inhibition.

Figure 5

a. The mean Euclidean distance between the representations of similar odors shown as a function of lateral excitation and slow inhibition. b. Time series of the Euclidean distance and correlation coefficient between similar odors arranged in order of the excitation-to-inhibition (E/I) ratio. The leftmost panel shows the value of the E/I ratio (x-axis) for different parameter sets (y-axis). When the denominator was zero we set the value of the ratio to 5 and arranged it according to increasing strength of lateral excitation. The right panels show the change in the distance and the correlation coefficient from odor onset (500 ms) to the end of the trial (3000 ms). The odor was presented from 500 ms to 1500 ms. We calculated the time series for 36 E/I ratios. Note, that the decrease of the correlation coefficient during the odor duration is significant but it is masked by limited dynamical range of the graphs that also shows low correlations after odor offset.

Figure 6

a. Top panels show the correlation coefficient as a function of lateral excitation for two different values of slow inhibition denoted by the colored arrows in the bottom panels. b. Image maps of the time averaged correlation coefficient as a function of lateral excitation and slow inhibition. The quantity to be maximized is calculated by summing the first two panels.

Figure 7

a. Average correlation coefficient between two odors. Each odor was repeatedly presented 20 times. The diagonal blocks show the correlation between trials of the same odor. The off–diagonal blocks with lower correlation coefficients provide a measure of the similarity between trials associated with different odors. b. Classification of responses by a hierarchical clustering algorithm. The bottom leaves of the tree represent each trial (40 trials across 2 odors). c. Mean error in classification of two similar odors.