Strategies for dynamic vision in the Drosophila peripheral visual system

Abstract

Listen

Translate article

Changes in luminance over space and time drive visual behaviors across species. Thus, sensitivity to luminance changes or contrast is fundamental to visual perception. Visual perception has to work in many different conditions, and these conditions can change rapidly. For perception unaffected by viewing conditions, contrast sensitivity must remain constant despite changing visual statistics, such as changes in mean luminance. Across species, photoreceptor gain control maintains the system sensitive enough to capture light changes over vastly varying illumination, however the gain control often does not meet the goal of constant contrast sensitivity, especially in dynamic conditions. Yet, animals across the evolutionary tree seem to reliably interpret visual cues when they navigate environments, suggesting a role of post-receptor visual circuitry in revising contrast sensitivity through additional layers of luminance gain control. Furthermore, gain correction must occur across parallel divisions of visual processing hierarchies, such as in both the ON and OFF pathways, within their differential circuit architectures and physiology. How the post-receptor gain control operates towards a robust contrast sensitivity evident in behavior, and how this is organized across parallel pathways is not understood in any visual system. In this thesis, I study how the peripheral visual system of fruit flies Drosophila melanogaster organizes information processing to support robust contrast computation and guide behavior. In the first manuscript, we show that fly behavior is luminance invariant in dynamic conditions, although peripheral contrast computation is not. Two first-order interneurons L2 and L3 in parallel convey luminance and contrast information respectively, and the luminance information is key to revise contrast computation in downstream circuitry. Luminance information scales up contrast signals in sudden dim light, when photoreceptors and the first-order interneurons do not increase their gain sufficiently fast. Thus, the peripheral processing stages themselves do not achieve a fixed contrast sensitivity in dynamic conditions, but they preserve and relay the forms of information required for contrast refinement in downstream circuitry. Here, parallel processing of the features contrast and luminance is the strategy that helps tackle the challenge of processing visual cues dynamically (Ketkar et al., 2020). I next explore how ON and OFF pathways compare in the ways they implement parallel feature processing and enable luminance invariance, especially considering their anatomical and physiological distinctions. Whereas the OFF pathway in Drosophila receives both luminance and contrast inputs through L2 and L3, the ON pathway was known to rely on input from a single contrast-sensitive first-order interneuron, L1. In the second manuscript, we show that the interneurons L1, L2 and L3 do not form ON- or OFF-specific inputs, but rather specialize in encoding contrast and luminance differently. Both ON and OFF pathways then benefit from this diversified luminance information as they compute luminance-invariant contrast in further processing stages. Therefore, the peripheral fly visual circuitry seems to first filter the photoreceptor signals differentially and then distribute them across ON and OFF pathways. In both pathways, luminance is required for behaviorally relevant contrast correction downstream of L1-L3. Finally in the third manuscript, I perform a comprehensive analysis of behaviorally relevant luminance gain across many different contrast and luminance conditions. This work demonstrates that post-receptor gain correction is a two-way process that can both enhance and reduce gain to restore the behavioral relevance of different contrasts. The correction is generally required across both fast and slow illumination changes. L2 contrast sensitivity scales with luminance even in fully adapted conditions, and the fixed contrast sensitivity seen in behavior is a result of post-L2 gain control. L3-mediated gain correction plays a dichotomous role, wherein particularly dim stimuli are amplified, and contrast signals in bright conditions are either scaled up or down to make up for the nature of contextual gain deficit. Based on behavioral data, an algorithmic model proposes a multi-channel circuitry to explain the multitude of gain correction operations led by luminance signals. Therefore, the circuitry seems to adopt a strategy of parallel computations up to an advanced circuit element, where the features must eventually be integrated. In sum, these findings reveal how behavioral responses to ON and OFF contrasts can achieve constancy beyond the first two processing stages, owing to downstream luminance gain control strategies. This work sheds light on the neural correlates of contrast constancy in a more proximal circuitry than photoreceptors, and the genetic access to cell types in Drosophila allows to dissect neural circuitry while directly correlating cell type physiology with behavioral output. Considering the common environmental challenges faced by different species and shared computational principles around which the systems have evolved, the involvement of gain control hierarchy in behavior is likely similar across visual systems, including our own.