TL;DR
This study reveals that the visual thalamus (dLGN) integrates not only visual input but also crucial modulatory signals from the cortex and the animal's behavioral state. They found that these modulatory inputs are particularly important for neurons with weak visual responses and that their impact is stronger when visual stimulation is minimal, suggesting a dynamic interplay that fine-tunes visual processing.
What Problem Does This Paper Address?
The dorsolateral geniculate nucleus (dLGN) is a critical relay station in the visual pathway, receiving direct input from the retina and also significant feedback from the visual cortex (corticothalamic, or CT, feedback). Additionally, the dLGN's activity is known to be influenced by the animal's behavioral state, such as locomotion and arousal (indicated by pupil size). However, how these multiple inputs – stimulus-driven, CT feedback, and behavioral state – jointly shape dLGN responses, especially during complex naturalistic stimuli, remained poorly understood. Specifically, the relative importance of these modulatory inputs and how they interact with the strength of the visual stimulus itself was unclear.
Methods
The researchers employed a sophisticated approach combining electrophysiology with optogenetics and advanced computational modeling. They recorded from dLGN neurons in awake, head-fixed mice viewing a movie stimulus. Crucially, they used optogenetics to selectively suppress CT feedback from V1. Simultaneously, they tracked the mice's locomotion and pupil size to capture behavioral state.
The core analytical tool was a spline-based generalized linear model (GLM). This model allowed them to predict dLGN neuron firing rates based on a combination of the visual stimulus, CT feedback suppression, running speed, and pupil size. The use of splines enabled flexible and smooth estimation of the receptive fields (RFs) and the modulatory effects. To assess the significance of each input, they used a permutation test, comparing the model's performance with actual data against performance when one input was randomized.
Datasets
The primary dataset consisted of extracellular recordings from dLGN neurons in awake mice while they viewed a movie stimulus. This movie stimulus was composed of a random sequence of clips from various films, presented at 30 frames per second. For comparison and validation of receptive field properties, they also utilized sparse noise stimuli.
Key Insights and Ideas
- Modulatory Inputs are Crucial for "Weakly Visual" Neurons: The most significant finding is that CT feedback and behavioral state didn't just universally boost responses; they disproportionately improved the prediction accuracy for neurons that were poorly responsive to the visual stimulus alone. This suggests these modulatory signals are essential for bringing less visually sensitive neurons into the processing stream.
- CT Feedback Shines in the Absence of Strong Visual Input: CT feedback's influence was found to be stronger, faster, and more prevalent during periods of blank screen stimulation compared to when the mice were viewing the movie. This implies that CT feedback might play a more dominant role in shaping dLGN responses when the visual scene is less informative or dynamic.
- CT Feedback Enhances Stimulus Discriminability (Under Certain Conditions): Counterintuitively, suppressing CT feedback improved the ability of a decoder to distinguish between short movie fragments based on single-neuron responses, but only for neurons that were significantly modulated by CT feedback. This suggests that intact CT feedback might introduce noise or variability that, while potentially useful for other functions (like attention), can hinder simple stimulus decoding at the single-neuron level.
Key Results
- Including CT feedback and behavioral state predictors in the GLM significantly improved model performance compared to a stimulus-only model.
- This improvement was most pronounced for a subpopulation of neurons poorly explained by the visual stimulus alone.
- CT feedback suppression led to a stronger reduction in firing rates during blank periods (45% of neurons modulated) compared to movie periods (18% modulated).
- The effect of CT feedback suppression was faster during blank periods than during movie presentation.
- A decoder could better discriminate between movie clips when CT feedback was suppressed, particularly for neurons sensitive to CT feedback. The percentage of correct decisions increased from 60.27% (control) to 68.34% (CT feedback suppressed).
Limitations and Future Work
- Significance Thresholds: The authors acknowledge that the fraction of neurons showing statistically significant modulation by CT feedback, running, or pupil size was relatively small. They attribute this partly to the conservative statistical methods used and suggest that alternative approaches (e.g., different optogenetic stimulation patterns or analysis of extreme conditions) might reveal more widespread effects.
- Model Simplifications: While powerful, the GLM is a linear model with a non-linear output. The authors note that it faced challenges in capturing fast modulations and response peaks, a common limitation of traditional GLMs. They also suggest future extensions to the model could incorporate more complex thalamic mechanisms like adaptation, interactions between inputs, and tonic/burst firing dynamics.
- Future Directions: The authors propose several avenues for future research:
- Investigating CT feedback effects with more varied stimulus types (contrast, spatial structure) and pathway-specific manipulations.
- Performing simultaneous recordings from V1 and dLGN to better understand the origin and interaction of feedback.
- Exploring how CT feedback-mediated synaptic noise might influence population-level decoding and sensitivity to weak inputs.
- Extending the GLM to capture more complex neural dynamics and interactions.
- Utilizing deep neural networks (DNNs) for potentially better capture of non-linear dynamics, while striving to maintain interpretability.
Summary updated