Fig 1.
Schematic diagram of the models.
(A) This circuit, based on results obtained in vivo, contains feedback inhibition between the KCs and GGN. (B) This hypothetical circuit contains feed-forward inhibition from the PNs to GGN inhibiting the KCs. (C) Left: Intensity of odor input received by each of the 300 PNs for representative odor concentrations (higher concentrations activate more PNs). Right: Different odors activate different sets of PNs; the two solid lines represent similar odors (activating largely overlapping subsets of PNs), and the dashed line represents a very different odor (activating a largely separate subset of PNs).
Fig 2.
Population activity elicited by different odor concentrations, low (red) and high (green), in FB and FF models.
Activity plots show the number of population action potentials except for the nonspiking GGN, which shows graded intracellular membrane potential. (A) PN population. (B) LHN population. (C) KC population. (D) GGN.
Fig 3.
GGN activity and its inhibitory input to the LHNs across a range of concentrations.
Colors indicate the type of the motif: feed-forward (FF) inhibition (blue) and feedback (FB) inhibition (red). (A) The activity of GGN calculated by integrating GGN’s membrane depolarization over the stimulus duration. (B) Inhibitory input to the LHNs from GGN for different odor concentrations.
Fig 4.
Network activity and response sparseness generated by different inhibitory motifs.
Colors indicate these motifs: no giant GABAergic neuron inhibition (noGGN) (green), feed-forward (FF) inhibition (blue), and feedback (FB) inhibition (red). (A) The average numbers of action potentials (APs) per trial for Kenyon cells (KCs). The average numbers of action potentials (APs) per trial for lateral horn neurons (LHNs). (B) The sparseness of odor representation across all Kenyon cells. (C) Frequency distribution of KC and LHN response intensity (total number of spikes elicited by a 1-s odor presentation) for different odor concentrations. (D) The number of cells generating an action potential in response to stimuli. Very high numbers of APs and unrealistically active KCs elicited by “high” odor concentrations suggest that the physiological range usually explored in vivo corresponds to “low-medium” range of the model.
Fig 5.
Effect of integration time on odor classification. For each motif (FB or FF) are shown: similar odors of low concentration (dotted magenta); similar odors of high concentration (magenta); different odors of low concentration (dotted blue); different odors of high concentration (blue).
(A) The average radius of the “cloud” representing multiple trials for high and low odor concentrations. (A1) KCs. (A2) LHNs. (B) Euclidean distance between centers of “clouds” representing multiple trials of two odors. (C) Classification error across multiple trials with two odors.
Fig 6.
Phase locking across odor concentrations.
(A) Circular phase graphs; length of the red arrow indicates the strength of phase locking; direction of the arrow indicated mean spike phase with respect to the field potential (peak of average activity in PNs, defined as zero phase). (Left) Feedback motif, PNs (top), KCs (middle), and LHNs (bottom), across low (left) and high (right) odor concentration. (Right) Feed-forward motif. (B) Schematic diagram comparing preferred firing phase of different cell types in FB model (left), FF model (right) and recordings made in vivo from locust (middle). (C) Phase locking across odor concentrations. (Left) KCs show stable phase locking (minimal phase change across concentrations) with more stability in the FB condition (red) than the FF condition (blue). (Middle) The phase of LHNs firing advances as odor concentration increases. The FB model (red) generated an almost linear shift with increasing odor concentration. The FF model (blue) produced a drastic phase shift between low and medium concentration, which levels out to no change at high odor concentrations. (Right) GGN responds much later in the FB model than in the FB model. Results from the FB model match observations made in vivo.
Fig 7.
Simplified model with sine-wave input and single cells representing each cell population illustrates phase locking across a range of inhibitory synaptic strengths.
Single cells: projection neuron (PN, green), Kenyon cell (KC, orange), lateral horn neuron (LHN, blue), giant GABAergic neuron (GGN, red). (A) Structure of the circuit. (B, Left) Feedback (FB) inhibitory model shows cells fired in the same order regardless of inhibitory synaptic strengths, although the KC, LHN, and GGN do fire slightly later as the inhibitory synapses were strengthened. (B, Right) Feed-forward (FF) inhibitory model: since GGN responses were determined by PN and not KCs, GGN does not phase shift. KC and LHN responses changed phase as inhibitory synaptic strength varied. KCs and LHNs fired after GGN for all values of inhibition except the weakest ones.