Westlake News ACADEMICS

Study Sheds New Light on Sensation and Action During Social Interactions

06, 2022

Email: zhangchi@westlake.edu.cn
Phone: +86-(0)571-86886861
Office of Public Affairs

Arguably the most indispensable aspect of animal behavior, social interactions require the detection of conspecifics, or organisms belonging to the same species, as a prerequisite. How animals perceive conspecifics with complex structured features to inform social interactions is fundamental to understanding social behavior. However, mapping sensation to action has been difficult to delineate due to complexities in both the sensory stimuli and the motor responses. Low dimensional simplifications such as a blob offer convenience in research, and have fueled a series of important discoveries in courtship behavior. Yet whether and how animals leverage higher-order structured cues to guide actions during social interactions remains perplexing.

The recent paper "Behavioral Signatures of Structured Feature Detection During Courtship in Drosophila" published March 28 in Current Biology (https://doi.org/10.1016/j.cub.2022.01.024) by the Westlake University Lab of Systems Neuroscience and Neuroengineering led by Prof. Yi Sun is shedding new light on this important problem. Westlake graduate student Jing Ning from the lab is the first author, and Yi Sun is the corresponding author.

Social interactions are unique in that both sensory inputs and motor outputs are basically positions and motions of conspecifics. The team explored such sensorimotor duality, and developed machine learning tools to quantitatively measure the gestures of interacting pairs over time. With these measurements and after a coordinate transformation, they identified “circling” behavior where deeply engaging flies carry out stereotyped actions that in many ways mimic a courtship dance. Circling benefits from multisensory integration, yet it is highly visual. It turned out that this fly “waltz” provides important clues to understanding conspecific recognition. The team delved into the dissection of “circling”.

Stereotyped patterns of circling behavior

We become what we behold. Arguing that the detection of structured features would influence actions and thus should be embedded in sensation-action relationships, the team established a series of sensorimotor relations. These relations clearly show the detection of complex structured features, such as the positions and motions of specific body parts, and how they determine actions. For example, males select specific wing and leg actions based on the positions and motions of the females’ heads and tails.

Sensorimotor transformation functions of circling

Beyond what they are currently seeing, interacting animals also use the recent history of what they saw to guide their next move. Using various system identification methods including the generalized linear model (GLM) and reverse correlation, researchers obtained a set of visuomotor transformation functions, in which specific stimuli from specific time points in the past are quantitatively related to specific actions in the future. Predictions of actions are also more precise when taking into account information over an extended time window rather than at single time points. These transformation functions reveal how the spatiotemporal dynamics history of specific features affects action selection and initiation. In a dispatch published on the same issue of Current Biology entitled "Social Behavior: Using Visual Cues To Guide Dancing on the Fly", Prof. Andrew Gordus from Johns Hopkins University commented: “The detailed computational approaches employed by Ning and colleagues enabled the quantification of these relationships, and a deeper understanding of the visuomotor transformations the flies perform while engaging in social interactions.”

Animals are complex structures composed of different body parts. Detection of specific body parts might be integrated to serve higher-level perception. Wondering if flies can see the forest for the trees, the authors found that the actions of different body parts are coordinated, and as a whole confer stronger conspecific responses than individual parts. Besides, circling induces mutual synchronization between interacting pairs. In addition, transformation functions reveal that the combination of information of different body parts better predicts actions. These findings provide evidence for the holistic detection of conspecifics. Such holistic detection, together with feature detection of specific body parts, constitutes a dual-approach strategy for the detection of conspecifics with high-dimensional structures, both top-down and bottom-up.

The establishment of a circling paradigm provides a novel paradigm for studying high-dimensional structures. The discovery of structured feature detection based on a circling paradigm provides important clues on the computation and algorithm. Future studies of the neural circuit mechanisms in the genetically tractable model of Drosophila should provide mechanistic insight on how simple systems achieve complex object recognition, which is critical knowledge for understanding intelligent systems. As Prof. Gordus wrote, “The principles uncovered in understanding how these circuits coordinate such a complex behavior can help inform how similar social paradigms are encoded in other systems, including our own.”

This work also published a suite of tools and more than 10 GB of datasets to help facilitate other researchers in the field. This includes the social behavior analysis tool SoAL (https://github.com/SunLabWestlake/SoAL), the Drosophila social behavior training dataset SDPD-15k (https://doi.org/10.6084/m9.figshare.17624888), the camera control software MIAS (https://github.com/SunLabWestlake/MIAS), and the dataset of analyzed circling behavior (10.6084/m9.figshare.17624081). As Prof. Gordus put it, “A challenge with answering this question is reliably tracking not only the position of two animals, but also their orientation, as well as their limb and wing extensions.” These computational ethology tools and datasets are critical for quantitative study of social interactions.