A computational framework linking synaptic adaptation to circuit behaviors in the early visual system

Liuyuan He, Yutao He, Kehuan Lun, Lei Ma, Kai Du, Tiejun HuangbioRxiv 2022 · DOI

TL;DR

This paper introduces a new computational framework, the Ribbon Adaptive Block (RAB) and hierarchical Linear-Nonlinear-Synapse (hLNS), to model synaptic adaptation in the early visual system. This framework accurately reproduces experimental data across various cell types and species, offering a powerful tool to understand how ribbon synapses contribute to visual adaptation.

What Problem Does This Paper Address?

The early visual system, particularly the retina, is constantly bombarded with complex visual information. To process this information effectively, it needs to adapt to changing conditions, a phenomenon known as visual adaptation. Ribbon synapses, a crucial component of the retina, play a key role in this adaptation by compressing visual signals. However, the precise mechanisms by which these synapses achieve this compression and contribute to circuit-level adaptation have remained poorly understood. Existing computational models often rely on data fitting without strong biophysical constraints, limiting their ability to generalize to new stimuli. This paper aims to bridge this gap by developing a physiologically constrained computational framework that can capture both synaptic and circuit-level adaptive behaviors.

Methods

The core of the proposed framework is the Ribbon Adaptive Block (RAB), a two-state kinetic model inspired by the Hodgkin-Huxley formalism. This module models vesicle release and resupply processes within ribbon synapses. The RAB module is then integrated into a larger hierarchical Linear-Nonlinear-Synapse (hLNS) model, which represents the layered structure of the retina (photoreceptor-bipolar cell-ganglion cell circuit).

Key methodological aspects include:

  • Physiologically Constrained Module (RAB): Unlike purely data-driven models, the RAB module's parameters are derived from biophysical processes, ensuring interpretability. It analytically links vesicle dynamics to adaptive properties like time constants and gain control.
  • Hierarchical Structure (hLNS): This model captures the multi-layered processing in the retina, with each cell (photoreceptor, bipolar cell) having an LN block for signal transduction and an N-RAB block for synaptic adaptation.
  • Parameter Selection Pipeline: A systematic approach is used to determine optimal parameters and functional forms for the RAB module by fitting experimental data and incorporating biological constraints.
  • Gain Control Analysis: The authors develop an analytical method to quantify synaptic gain, allowing them to investigate how synapses adjust their sensitivity to input signals.

Datasets

The study utilizes existing experimental datasets for training and evaluating their models. These include:

  • Vesicle release data from salamander cone synapses.
  • Postsynaptic responses from rat rod bipolar cell synapses.
  • Light-stimulated responses from mouse cone bipolar cells.
  • Paired-pulse depression experiments on various retinal synapses.

The paper does not introduce new datasets but rather leverages existing experimental recordings to validate their computational framework.

Key Insights and Ideas

  • RAB's Biophysical Basis: The RAB module's strength lies in its biophysical grounding, allowing it to capture intrinsic synaptic adaptation mechanisms rather than just fitting data. This leads to better generalization.
  • Diverse Adaptive Time Constants: The hLNS model reveals a rich and diverse range of adaptive time constants in ribbon synapses within the retinal circuit, suggesting specialized roles for different cell types.
  • Frequency-Sensitive Gain Control: A surprising finding is the prediction of a frequency-sensitive gain-control strategy in the PR-Cx BC synapse, which differs from the traditional contrast-based adaptation. This suggests that the synapse dynamically adjusts its sensitivity based on the temporal characteristics of the input.
  • Robustness to Unseen Stimuli: The models demonstrate strong robustness to stimuli not used during training, indicating that they capture fundamental adaptive mechanisms.

Key Results

  • The RAB module successfully reproduces a wide range of experimental recordings of synaptic and circuit-level adaptive behaviors across different cell types and species.
  • The N-RAB model achieved excellent performance in fitting adaptive behaviors in RBC-AII synapses, with feature errors comparable to or better than existing LNK models.
  • The hLNS model accurately captured the complex adaptive behaviors of mouse bipolar cells under 'chirp' stimuli, with a high correlation coefficient (89 ± 5%) between model and experimental responses.
  • The model predicts that the PR-Cx BC synapse exhibits a frequency-sensitive gain control, with gain decreasing significantly at higher frequencies. This was experimentally verified.
  • The adaptive time constants of PR-BC synapses were found to be larger than those of BC-GC synapses, and the PR-Cx BC synapse showed a particularly large adaptive time constant at hyperpolarized membrane potentials.

Limitations and Future Work

  • Simplification of Biophysical Processes: While the RAB module is biophysically constrained, it simplifies complex processes like calcium dynamics and vesicle resupply. Future work could incorporate more detailed biophysical models for these aspects.
  • Ignoring Interneurons: The current hLNS model does not explicitly include the role of horizontal cells, which can provide inhibitory feedback. Incorporating these could lead to more comprehensive circuit models.
  • Spatial Input: The current model processes input as a single value, neglecting spatial information. Future development should aim to handle spatiotemporal stimuli to better explain visual pathway interactions.
  • Extensibility: The authors highlight the flexibility of their framework, suggesting future work could add components to represent other relevant mechanisms in synaptic adaptation, such as Calcium-Induced Calcium Release (CICR) or different types of calcium channels. They also propose extending the model to more complex circuits with multiple neurons and connections.

Summary updated

Abstract

Abstract Retina ribbon synapses are the first synapses in the visual system. Unlike the conventional synapses in the central nervous system triggered by action potentials, ribbon synapses are uniquely driven by graded membrane potentials and are thought to transfer early sensory information faithfully. However, how ribbon synapses compress the visual signals and contribute to visual adaptation in retina circuits is less understood. To this end, we introduce a physiologically constrained module for the ribbon synapse, termed Ribbon Adaptive Block (RAB), and an extended “hierarchical Linear-Nonlinear-Synapse” (hLNS) framework for the retina circuit. Our models can elegantly reproduce a wide range of experimental recordings on synaptic and circuit-level adaptive behaviors across different cell types and species. In particular, it shows strong robustness to unseen stimulus protocols. Intriguingly, when using the hLNS framework to fit intra-cellular recordings from the retina circuit under stimuli similar to natural conditions, we revealed rich and diverse adaptive time constants of ribbon synapses. Furthermore, we predicted a frequency-sensitive gain-control strategy for the synapse between the photoreceptor and the CX bipolar cell, which differ from the classic contrast-based strategy in retina circuits. Overall, our framework provides a powerful analytical tool for exploring synaptic adaptation mechanisms in early sensory coding.
BibTeX
@misc{he2022computational,
  author = {Liuyuan He and Yutao He and Kehuan Lun and Lei Ma and Kai Du and Tiejun Huang},
  title = {{A computational framework linking synaptic adaptation to circuit behaviors in the early visual system}},
  year = {2022},
  doi = {10.1101/2022.08.27.505287},
  howpublished = {bioRxiv},
}
Figure