The temporal structure of the inner retina at a single glance

Zhijian Zhao, David Klindt, André Maia Chagas, Klaudia P. Szatko, Luke E. Rogerson, Darío A. Protti, Christian Behrens, Deniz Dalkara, Timm Schubert, Matthias Bethge, Katrin Franke, Philipp Berens, Alexander S. Ecker, Thomas Euler

Abstract

The retina decomposes visual stimuli into parallel channels that encode different features of the visual environment. Central to this computation is the synaptic processing in the inner plexiform layer (IPL). Here, we extend a two-photon microscope with an electrically tunable lens allowing us to obtain optical vertical slices of the IPL, providing a complete picture of the response diversity of bipolar cells at a “single glance.” This approach allows us to isolate and investigate batch effects—inter-experimental variations that result in systematic differences in response speed. By developing a model that disentangles biological from experimental variability, we recovered the characteristic gradient of response speeds across the IPL with higher precision than before.

Introduction

The mammalian retina is far more than a simple biological camera. It acts as a sophisticated pre-processor, decomposing the visual world into dozens of parallel feature channels—such as motion, contrast, and color—before sending information to the brain. This complex computation primarily occurs within the Inner Plexiform Layer (IPL), a dense synaptic neuropil where various types of bipolar cells (BCs) relay excitatory signals from photoreceptors to amacrine and ganglion cells.

Understanding the "temporal structure" of the IPL—how different layers process visual information over time—is crucial for deciphering retinal function. However, traditional imaging techniques have struggled to capture this structure accurately due to the physical orientation of the tissue and the inherent variability of biological experiments.

The Challenge: Horizontal Constraints and Batch Effects

Standard two-photon microscopy typically utilizes "horizontal" (x-y) scanning, imaging one depth of the IPL at a time. While effective for mapping spatial organization, this approach is poorly suited for studying temporal dynamics across the entire depth of the retina. Because different layers are recorded in separate sessions, comparing the timing of signals becomes problematic.

Small, often unavoidable variations between experimental sessions—known as "batch effects"—can significantly distort data. Factors such as a 1°C fluctuation in tissue temperature or slight differences in the concentration of fluorescent sensors (like iGluSnFR) can alter the perceived speed of neural responses. In many cases, these experimental artifacts are large enough to mask the actual biological differences between cell types.

Methodology: Vertical Imaging and Encoding Models

To overcome these limitations, researchers developed a framework combining hardware innovation with computational modeling.

1. Axial (x-z) Scanning

The team modified a standard two-photon microscope by integrating an Electrically Tunable Lens (ETL). By rapidly modulating the lens's curvature, the focal plane can be shifted along the z-axis (depth) at high speeds. This allows for "vertical" or axial (x-z) scanning, capturing an optical slice that spans the entire thickness of the IPL (~50 µm) near-simultaneously. This "single glance" ensures that every recording contains a representative sample of all IPL layers, subjected to the exact same experimental conditions.

2. The Local Chirp Encoding Model

To analyze the resulting data, the researchers developed a "local chirp encoding model." This model operates on a key insight: the functional diversity of bipolar cells is largely a matter of temporal rescaling. Rather than treating every neuron as having a unique response profile, the model assumes a shared underlying response kernel that is simply stretched or compressed in time based on two factors: the cell's depth (biological) and the specific experimental session (batch).

Key Insights and Results

The study analyzed 5,379 regions of interest (ROIs) across 37 scan fields. By comparing their vertical imaging data with structural references from Electron Microscopy (EM) datasets, the researchers validated the density and distribution of their functional recordings.

Disentangling Noise from Biology

The model revealed that batch effects are surprisingly dominant, accounting for approximately 66% of the variance in signal timing, while biological depth accounted for 48%. By identifying these effects, the researchers could "clean" the data, revealing the true underlying biological patterns. The model estimated that the observed batch-to-batch speed variations were consistent with temperature fluctuations of just ±1°C, highlighting the extreme sensitivity of retinal timing.

The Speed Map of the IPL

Once corrected for experimental noise, a clear "speed map" of the IPL emerged. Visual responses are not uniform; they are fastest in the center of the IPL and slow down significantly toward the borders of the Inner Nuclear Layer (INL) and the Ganglion Cell Layer (GCL). The combined "depth + batch" model explained nearly half of the total variance (46-49%), a substantial improvement over models that only considered biological factors.

Discussion and Future Directions

The ability to capture the entire depth of a neural structure at once represents a significant step forward for retinal research. While the ETL-based method introduces some technical challenges—such as a 10ms "settling time" for the lens and slight laser power fluctuations across depths—these are manageable within the scale of the mouse retina.

This framework provides a more efficient and precise way to study how the retina’s parallel channels are organized. Beyond the retina, this "single glance" approach could be applied to other layered neural tissues, such as the cerebral cortex, or used with different functional sensors to explore the complex temporal symphony of the brain.

Summary updated

Abstract

Abstract The retina decomposes visual stimuli into parallel channels that encode different features of the visual environment. Central to this computation is the synaptic processing in a dense and thick layer of neuropil, the so-called inner plexiform layer (IPL). Here, different types of bipolar cells stratifying at distinct depths relay the excitatory feedforward drive from photoreceptors to amacrine and ganglion cells. Current experimental techniques for studying processing in the IPL do not allow imaging the entire IPL simultaneously in the intact tissue. Here, we extend a two-photon microscope with an electrically tunable lens allowing us to obtain optical vertical slices of the IPL, which provide a complete picture of the response diversity of bipolar cells at a “single glance”. The nature of these axial recordings additionally allowed us to isolate and investigate batch effects, i.e. inter-experimental variations resulting in systematic differences in response speed. As a proof of principle, we developed a simple model that disentangles biological from experimental causes of variability, and allowed us to recover the characteristic gradient of response speeds across the IPL with higher precision than before. Our new framework will make it possible to study the computations performed in the central synaptic layer of the retina more efficiently.
BibTeX
@inproceedings{zhao2020temporal,
  author = {Zhijian Zhao and David Klindt and André Maia Chagas and Klaudia P. Szatko and Luke E. Rogerson and Darío A. Protti and Christian Behrens and Deniz Dalkara and Timm Schubert and Matthias Bethge and Katrin Franke and Philipp Berens and Alexander S. Ecker and Thomas Euler},
  title = {{The temporal structure of the inner retina at a single glance}},
  booktitle = {Scientific},
  year = {2020},
  doi = {10.1038/s41598-020-60214-z},
}
Figure