Biophysically relevant network model of the piriform cortex predicts odor frequency encoding using network mechanisms

Technical Summary: Biophysically Relevant Network Model of the Piriform Cortex Predicts Odor Frequency Encoding

Problem Statement
Natural odor plumes are characterized by rapid, turbulent concentration fluctuations that carry crucial spatial and temporal information about odor source location. While recent studies have established that the olfactory bulb (OB) encodes these temporal features, specifically odor frequencies up to 20 Hz, it remains unknown whether this temporal information is preserved and encoded in the higher olfactory cortical area, the piriform cortex (PCx). Historically, olfaction was thought to lack high-frequency temporal resolution due to the slow kinetics of GPCR-based receptors. Although the PCx is known for encoding odor identity, intensity, and performing pattern separation/completion via its recurrent network, its specific capacity to encode odor frequency fluctuations has not been detailed. This study investigates whether the PCx recurrent circuitry can encode and discriminate between different odor frequency stimuli (specifically 2 Hz and 20 Hz).

Methodology
The authors developed a biophysically constrained network model of the piriform cortex using the NEURON simulation environment. The model architecture included:

• Cell Types: 500 excitatory pyramidal neurons (PYRs), 50 feedforward inhibitory interneurons (FFINs), and 50 feedback inhibitory interneurons (FBINs).
• Connectivity: Connections were randomly assigned based on experimental probabilities from rodent data. PYRs received inputs from mitral/tufted (M/T) cells and exhibited recurrent excitation. FFINs and FBINs provided distal and somatic inhibition to PYRs, respectively, with additional intra-population recurrent connections.
• Inputs: The model was driven by previously recorded in vivo spiking activity of M/T cells in response to three odor mixtures (two binary mixtures and a combined mixture) presented at 2 Hz (basal sniffing range) and 20 Hz (upper physiological limit).
• Analysis:
  • Single-Cell vs. Population: Firing rates were analyzed across 336 trials using Mann–Whitney U tests and Cliff's delta to assess discrimination at the single-neuron and population levels.
  • Decoding: One-dimensional convolutional neural networks (1D-CNNs) and logistic regression classifiers were trained on binned spike trains (20 ms windows) to decode odor frequency (2 Hz vs. 20 Hz).
  • Virtual Knockouts: To determine the role of specific circuit motifs, the authors systematically removed specific synaptic connections (e.g., FFIN→PYR, FBIN→PYR, PYR→PYR, intra-interneuron connections) and re-evaluated decoding performance.
  • Control for Excitability: To distinguish between circuit motif effects and general changes in baseline firing rates, control variants were simulated by modulating baseline noise to match the firing rates of knockout models.

Key Results

1. Population-Level Discrimination: Individual pyramidal neurons were largely ineffective at discriminating between 2 Hz and 20 Hz frequencies; very few cells showed significant differences even with relaxed statistical thresholds. However, the trial-averaged population activity of PYRs significantly discriminated between the two frequencies across all odor stimuli.
2. Distributed Encoding: Decoding accuracy using 1D-CNNs reached up to 95% for distinguishing 2 Hz from 20 Hz. Analysis of log-likelihood scores revealed that this discrimination capability was highly distributed across the PYR population, with no single "master" neuron driving the classification.
3. Circuit Motif Effects (Knockouts):
   • Enhanced Discrimination: Surprisingly, eliminating direct feedforward (FFIN→PYR) or feedback (FBIN→PYR) inhibition onto pyramidal neurons improved frequency decoding accuracy across all odor conditions compared to the control network.
   • Impaired Discrimination: Conversely, removing recurrent excitation among PYRs (PYR→PYR) or simultaneously removing recurrent inhibition within both interneuron populations (FFIN→FFIN and FBIN→FBIN) degraded decoding performance.
   • Minimal Impact: Removing intra-population connections within a single interneuron type (e.g., only FFIN→FFIN) had minimal effects on decoding accuracy.
4. Mechanism Specificity: The improvement in decoding upon removing direct inhibition was not merely a result of increased baseline firing rates. Control simulations where baseline firing rates were artificially matched to the knockout levels failed to reproduce the enhanced decoding accuracy, indicating that the specific synaptic architecture (circuit motifs), rather than global excitability, drives the frequency encoding.

Significance and Claims
The paper claims that the piriform cortex possesses the capacity to encode and process the temporal frequency of odor stimuli, a feature previously unconfirmed in this cortical region. The study predicts that odor frequency discrimination in the PCx is a network-level phenomenon that emerges from distributed activity across the pyramidal population rather than single-cell tuning.

A central finding is the bidirectional modulation of odor frequency encoding by specific circuit motifs. The authors propose that the balance of excitation and inhibition in the PCx shapes temporal encoding in a task-dependent manner. Specifically, the model suggests that reducing direct inhibition onto pyramidal neurons enhances the network's ability to discriminate temporal frequencies, likely by increasing response gain and altering correlation structures, whereas recurrent excitation is necessary to maintain the stability required for this discrimination.

The authors conclude that their biophysically relevant model predicts that the PCx can differentiate between naturalistic odor frequencies (2 Hz vs. 20 Hz) using distributed network mechanisms, and that this capability is sculpted by the interplay of specific synaptic connections rather than simple changes in neuronal excitability. They note that these findings offer experimentally testable predictions regarding how specific interneuron subtypes contribute to temporal odor processing.