Sparse distributed representation of odors in a large-scale olfactory bulb circuit

Abstract

In the olfactory bulb, lateral inhibition mediated by granule cells has been suggested to modulate the timing of mitral cell firing, thereby shaping the representation of input odorants. Current experimental techniques, however, do not enable a clear study of how the mitral-granule cell network sculpts odor inputs to represent odor information spatially and temporally. To address this critical step in the neural basis of odor recognition, we built a biophysical network model of mitral and granule cells, corresponding to 1/100th of the real system in the rat, and used direct experimental imaging data of glomeruli activated by various odors. The model allows the systematic investigation and generation of testable hypotheses of the functional mechanisms underlying odor representation in the olfactory bulb circuit. Specifically, we demonstrate that lateral inhibition emerges within the olfactory bulb network through recurrent dendrodendritic synapses when constrained by a range of balanced excitatory and inhibitory conductances. We find that the spatio-temporal dynamics of lateral inhibition plays a critical role in building the glomerular-related cell clusters observed in experiments, through the modulation of synaptic weights during odor training. Lateral inhibition also mediates the development of sparse and synchronized spiking patterns of mitral cells related to odor inputs within the network, with the frequency of these synchronized spiking patterns also modulated by the sniff cycle.

Figure 1. Schematic description of the model and its relation to in vivo experimental data.

A) Dorsal surface of the olfactory bulb from which experimental data were obtained; different letters identify different clusters of glomeruli responding to different molecular features of the odor molecules; in this paper we used glomeruli activated in clusters A–D. B) Spatial distribution of the 73 glomeruli activated by the 72 odors used in this work; the black line represents the projection track used to reduce the system to 1 dimension (see main text for details). C) Top view of the soma spatial locations for the 500 mitral cells (blue triangles) and 10000 granule cells (red circles) used in all simulations; the black line is used to reduce the system to 1 dimension, and the arrow indicates the viewpoint used for all simulation movies. D) The connectivity for an efficient lateral inhibition from non-topographical connectivity of the granule-mitral cell network (see text). The basic microcircuit is schematically represented with two mitral (M1 and M2), two granule cells (GC1 and GC2), and their dendrodendritic reciprocal synapses (open and closed circles represent excitatory and inhibitory synapses, respectively). The somatic action potential elicited in M1 by an odor (red trace close to M1 soma) backpropagates (bAP) along the M1 lateral dendrite (in red). This activates GC2 and a consequent local inhibitory potential (black trace) close to the soma of M2. In this way inhibition can be independent of distance, imposed locally by granule cells activated by backpropagating action potentials. E) In the current model, the ratio of granule to mitral cells is 20∶1, with a 10% probability of connection, as schematically represented in the figure (triangles, mitral cells; large circles, granule cells; small circles, reciprocal synapses) for two different mitral cells (green and red); the yellow circle represents a granule cell which receives input from both the red and green mitral cells, whereas unfilled symbols represent mitral, granule, and reciprocal synapses not connected with either the green or red mitral cells. F) To appreciate more clearly the network activity, a movie was generated from each simulation, with the mitral cell somas laid out using {X,Y} coordinates reflecting the spatial location of the real glomeruli in the dorsal surface investigated in the experiments, and granule cell somas distributed inside a 3D space schematically representing their spatial organization in the real olfactory bulb (initial arrangement panel). An opened-out view of the system was used as the viewpoint to display somatic spikes during a simulation (opened-up view panel); the gray line represents the projection track used to reduce the system to 1 dimension, and the arrow indicates the viewpoint used for all simulation movies. See movie S1 and its full HD version on the ModelDB database (acc.n. 144570).

Figure 2. Odor inputs used for all simulations.

A) Odor input to each mitral cell was implemented following the experimental findings for 73 individual glomeruli within 4 clusters (Y-axis) during presentation of 72 different odors (X-axis). The intrinsic image responses of glomeruli in clusters A, B, C, and D in the dorsal surface of the olfactory bulb were classified into 4 response levels (weak, moderate, strong, very strong), represented in the table with circles of different sizes. In our model, the 4 response levels defined the peak synaptic excitatory conductances in the distal tuft of the mitral cells activated by each odor. The arrows at the bottom of the table point to the three odors (#14, octanal; #29, guaiacol; and #54, (+)-Cvn) that are discussed in detail in the main text. B) Histograms represent the input signal to each mitral cell for the three odors indicated by the arrows in part A.

Figure 3. Spatial-temporal firing patterns of the large scale simulated network for three types of odor input.

A) Raster plot of mitral and granule cell spikes during the first 7 sec of a weak odor presentation (odor 54, (+)-Cvn); note the weak suppression of mitral cell firing surrounding the most active cells after the first few seconds of simulation (left panel), corresponding to the increase in firing of granule cells (right panel) around the most active mitral cells. B) Raster plot for mitral and granule cell spikes during the first 7 sec of an odor with strong glomerular activation (odor 14, octanal); note a strong suppression of mitral cell firing surrounding the most active cells (left panel) after the first two seconds of simulation, corresponding to the increase in firing of granule cells (right panel) around the most active mitral cells. C) Simulation findings for an odor with strong and spatially distributed glomerular activation, odor 39, k3-3; note a strong and widespread suppression of mitral cell firing surrounding the most active cells (left panel) after the first two seconds of simulation, corresponding to the increase in firing of granule cells (right panel) around the most active mitral cells.

Figure 4. Development of synaptic weights among mitral and granule cells underlying the spike responses.

In all panels, the top histogram represents the input strength of each mitral cell, the middle and bottom plots represent the normalized excitatory and inhibitory peak synaptic conductance after 10 sec of odor presentation, respectively; (dark purple: 0, white: 1). A) Without odor input. B) Results for odor 54, (+)-Cvn; note the extent of mitral-to-granule cell potentiated synapses in response to the glomerular input, due to action potential propagation in the lateral dendrites, and the consequent formation of distributed inhibitory columns of granule-to-mitral synapses. C) Results for odor 14 (octanal); note the greater extent of mitral-to-granule synapses in response to the glomerular input, caused by the larger group of activated glomeruli; D) Results for odor 39 (k3-3); note the spatial distribution of excitatory mitral-to-granule synapses in response to the distributed glomerular input, and the consequent wide distribution of inhibitory granule-to-mitral synapses. See also Fig. S1.

Figure 5. Odors with a strong input generate lateral inhibition on weaker components.

A) Two typical odors with different input structure; top panels represent the input to each mitral cell; bottom panels show the distribution of the normalized peak inhibitory (granule-to-mitral) conductance after odor learning (dark purple: 0, white: 1); (left) example of a weak odor, all activated mitral cells receive weak input; (right) example of a distributed odor, active mitral cells are spatially distributed and receive both weak and strong inputs. B) Average granule-to-mitral peak inhibitory conductance for each odor; in all cases the average peak conductance was calculated over the set of mitral cells receiving a weak input (input levels 1–2, red circles), or a strong input (input levels 3–4, black circles); note that a value of 0 for the peak inhibitory conductance would be expected from a weak input. C) Typical traces from different mitral cells, all receiving a weak input from three odors (octanal, gua, and eug), during 10 sec simulation of odor learning; note the powerful inhibition, from mitral cells activated by odors with a very strong component (e.g. octanal, see Fig. 4C) after the first few seconds of simulation, corresponding to the initial organization of the network.

Figure 6. Sparse granule-mitral cell connectivity modulates in a robust way a mitral cell response to an odor.

A) Inhibitory synaptic weights (granule-to-mitral) after 10 sec presentation of odor k3-3 using different levels of granule-mitral connection probability; the top histogram represents the input strength on each mitral cell. B) Average instantaneous firing rate of mitral cells 220–500 during odor k3-3 presentation, without granule cells (no GC) and with GCs connected to mitral cells with different probabilities (2-5-10-15%). C) Interspike interval (ISI) distribution of the most active mitral cells (odor strength>2) in the range 0–50 (left) and 50–200 ms (right) with (triangles) or without (circles) GCs with 10% connectivity; D) ISI in the range 50–200 ms with (triangles) or without (circles) GCs using different connection probabilities with mitral cells. In all cases, the first and the last second of 10 sec simulations were used to calculate firing rates and ISI probabilities without or with GCs, respectively.

Figure 7. A complementary role for mitral cell feedback and lateral inhibition.

A) Average instantaneous firing rate of mitral cells 420–500 during 5 sec presentation of odor k3-3 at different concentrations without granule cells in the network; note the rather uniform and unstructured mitral cell activity at low odor concentrations (below 0.5). In all simulations, synaptic plasticity was blocked and a synaptic weights configuration corresponding to the “no odor” condition (see Fig. 4A) was used in all cases. B) Average instantaneous firing rate as in A) but with granule cells in the network; in all simulations synaptic plasticity was blocked. The synaptic weights configuration obtained after presentation of odor k3-3 (see Fig. 4D) was used in all cases. Note the firing depression of mitral cells activated by low odor concentrations (odor concentration below 0.5, mitral cells 446–465 and 476–485), and the firing depression of flanking mitral cells for higher odor concentrations (odor strength above 0.5, mitral cells 420–445, 466–475, and 486–500).

Figure 8. Spike timing of individual mitral cells following odor onset depends on input strength.

A) Experimental findings (from , Reprinted by permission from Macmillan Publishers Ltd: Nat. Neurosci.) on the spike times of 6 mitral cells during presentation of the same odor, warped to the sniff cycle (left) and their post stimulus time histograms (PSTH) (right). B) Distribution of spike times from the stimulus onset in our model for octanal (strong odor, red), k3-3 (strong and distributed, black) and +(−)Cvn (weak, blue); light grey line represents the distribution of spike times from background activity. C) (left) raster plots of spike times from 70 simulated sniffs for mitral cells activated by odor k3-3, grouped according to the 4 input levels coding odor strength in the different glomeruli (see Fig. 2); (right) corresponding distribution of spike times from the stimulus onset. The gray line in the top three graphs represents the “no odor” condition; to aid in comparing the distribution of mitral cells with strong inputs (bottom graph) with those obtained under other conditions, the distributions for “no odor”, medium, and weak inputs are indicated by gray lines.

Figure 9. Emergence of distributed mitral-granule cell connectivity.

(Left) Typical experimental findings for pseudorabies virus staining patterns after olfactory bulb injection (adapted from Fig. 2D of with permission from the National Academy of Sciences, U.S.A); the photo shows a coronal section of the olfactory bulb, with labeling of columns of granule cells; GCL, granule cells layer; MCL, mitral cells layer; GL, glomerular layer; colored loops indicate regions of strong (purple), medium (blue), and weak (orange) cell activation. (Right) Model results after learning of odor k3-3. The top traces show bursts in the soma (right) that become single spikes as they propagate in a lateral dendrite (left) of mitral cell 457 at the end of the learning period (t = 8–9 sec). The red histogram represents the strength of glomerular input to the mitral cell tuft (see Fig. 2B), and the field shows the location and normalized peak inhibitory conductance from granule cells on the mitral cell lateral dendrites (see color bar: strong conductance in yellow; moderate in green; low or absent in black). Note that the locations of strong conductances line up primarily with sites of mitral cell responses but also occur in the intervening spaces (see text). Regions similar to the experimental patterns are indicated with different colors (purple, blue, orange).