The Functional Logic of Odor Information Processing in the Drosophila Antennal Lobe

Aurel A. Lazar, Tingkai Liu, Chung-Heng YehbioRxiv 2021 · DOI

TL;DR

This paper proposes that the fruit fly's antennal lobe (AL) processes olfactory information by separating odor identity from concentration, using parallel neural pathways that act like Divisive Normalization Processors. This separation allows for robust odor recognition and associative learning by signaling "ON-OFF" odor events.

What Problem Does This Paper Address?

The olfactory system faces a challenge: odor signals are a mix of what the odor is (identity) and how much of it there is (concentration). This confounding of "semantic" (identity) and "syntactic" (concentration) information at the sensory neuron level needs to be disentangled for effective downstream processing like recognition and learning. Specifically, the authors investigate how the Drosophila antennal lobe (AL) achieves concentration-invariant odor identity representation while still capturing the dynamic changes (onset and offset) of odor concentration. They aim to uncover the functional logic and computational mechanisms within the AL that enable this crucial separation.

Methods

The researchers employed a computational approach, building and simulating various models of the AL circuit. They exhaustively explored different configurations of Local Neurons (LNs) – presynaptic (Pre-LNs) and postsynaptic (Post-eLNs, Post-iLNs) – and their connectivity patterns (local vs. global, feedback vs. feedforward).

Key techniques include:

  • Differential Divisive Normalization Processors (DNPs): The authors propose that the LN pathways can be modeled as parallel DNPs, which are known for their ability to perform normalization and feature extraction.
  • Two-Step Optimization Procedure: Models were optimized by minimizing the distance between simulated and physiological responses (using L2 norm) and also by minimizing the angular distance between steady-state responses and odorant affinity vectors. This involved Latin Hypercube Sampling (LHS) for initial parameter exploration and Differential Evolution (DE) for fine-tuning.
  • Decomposition into Steady-State and Transient Components: Neuron responses were analyzed by separating them into steady-state (concentration-invariant) and transient (contrast-boosting) components to understand the roles of different LN types.

Datasets

The study primarily uses physiological data from previous work for comparison and validation. Specifically, they refer to:

  • DM4 PN PSTH data: Used to evaluate the temporal responses of their AL circuit models.
  • Or59b OSN PSTH data: Used for comparison with PN responses.
  • Acetone odorant waveforms: Used as input stimuli for the models.
  • Odorant affinity vectors: Derived from previous studies [2, 25] and used to evaluate the recovery of odorant identity.

No new datasets were introduced in this paper.

Key Insights and Ideas

  • Separation of Identity and Concentration: The AL's core function is to decouple odor identity from concentration. This is achieved by having different LN pathways specialize in these tasks: presynaptic LNs for concentration invariance (identity) and postsynaptic LNs for capturing temporal dynamics (concentration changes).
  • DNP as a Unifying Framework: The authors propose that the complex interactions within the AL can be elegantly described by parallel Divisive Normalization Processors (DNPs). This provides a powerful algorithmic abstraction for understanding the AL's computational logic.
  • "ON-OFF" Odorant Identity Recovery: The combination of these specialized pathways leads to an "ON-OFF" processing paradigm. The AL signals the timing of odor events (ON/OFF) and simultaneously recovers the odor's identity, enabling rapid and robust recognition.
  • Global Inhibition is Key for Identity Recovery: For spatio-temporal processing, global inhibition by Pre-LNs across all glomeruli is crucial for robustly recovering odorant identity, outperforming local inhibition.

Key Results

  • Single-Channel Temporal Processing: Presynaptic LNs significantly improve concentration invariance, while postsynaptic LNs (Post-eLN and Post-iLN) are essential for contrast boosting (capturing ON/OFF transients).
  • Spatio-Temporal Processing: Global Pre-LN inhibition (feedback or feedforward) dramatically improves odorant identity recovery compared to local inhibition.
  • Robustness: The proposed DNP models demonstrate robustness in capturing odorant identity across various concentration levels and waveform profiles. Post-LNs act as reliable ON/OFF event detectors.
  • Model Performance: The study compared 20 different spatio-temporal circuit architectures, finding that global inhibition models generally outperformed local ones in odorant identity recovery.

Limitations and Future Work

  • Experimental Validation: While the models show strong computational evidence, direct experimental validation of the spatio-temporal AL processing across all glomeruli with simultaneous PN recordings is ideal but challenging.
  • Model Complexity: The current models focus on uni-glomerular PNs and specific LN types (pan-glomerular or uni-glomerular). The authors acknowledge that many LNs fall in between these categories and suggest that the models can be extended to incorporate these additional neuron types.
  • Future Directions: The authors suggest that the proposed "ON-OFF odorant identity recovery" paradigm provides a framework for understanding rapid odor recognition and associative memory in downstream brain regions. Further experimental work is needed to fully validate these hypotheses.

Summary updated

Abstract

Abstract The early olfactory system of the fruit fly, while sensing a complex odorant landscape, encodes the odorant object identity (semantic information) and the odorant concentration waveform (syntactic information) into a combinatorial neural code. Single-channel physiology recordings at the output of the Antenna Lobe (AL) exhibit concentration-invariance and contrast-boosting properties, indicating a decoupling of the odorant object identity from the concentration waveform in steady-state while responding strongly to odorant concentration onset and offset in transient states. Through exhaustive computational explorations of the AL circuits, we show that the steady-state and transient response features of the AL are, respectively, due to presynaptic and postsynaptic Local Neurons (LNs). Theoretical analysis reveals that the LN pathways can be modeled as parallel differential Divisive Normalization Processors (DNPs). Differential DNPs robustly extract odorant identity (semantic information) and ON/OFF odorant event-timing (syntactic information), thereby providing for the AL the functional logic of ON-OFF odorant identity recovery .
BibTeX
@misc{lazar2021functional,
  author = {Aurel A. Lazar and Tingkai Liu and Chung-Heng Yeh},
  title = {{The Functional Logic of Odor Information Processing in the <i>Drosophila</i> Antennal Lobe}},
  year = {2021},
  doi = {10.1101/2021.12.27.474306},
  howpublished = {bioRxiv},
}
Figure