Abstract
Visual motion is produced both by an organism’s movement through the world, and by objects moving in the world such as potential predators. Choosing appropriate behaviour therefore requires organisms to distinguish these sources of visual motion. Here we asked how mice integrate self-movement with looming visual motion by combining virtual reality and neural recordings from superior colliculus (SC), a brain area important in visually-guided approach and avoidance behaviours. We first measured locomotion behaviour and neural activity while animals approached an object in virtual reality, and while the same object loomed at them. In both cases, vision dominated activity in superficial layers (SCs), while locomotion had more influence on activity in intermediate layers (SCim). In addition, animals instinctively slowed their locomotion when nearing the object, or when the object neared them. To directly test animals’ ability to distinguish self-from object motion we replayed the visual images generated during object approach. Locomotion behaviour often changed during replay, showing animals are able to establish if visual motion is matched to their self-movement. Further, decoders trained on locomotion behaviour, or on population activity in SC, particularly in SCim, were able to reliably discriminate epochs of replay and object approach. We conclude that both mouse behaviour and SC activity encode whether looming visual motion arises from self-or object movement, with implications for understanding sensorimotor coordination in dynamic environments.
Introduction: The Ambiguity of Sensory Activity
Imagine walking through a forest and seeing a tree appear to grow larger. Is the tree falling toward you, or are you simply walking toward it? This "ambiguity of sensory activity" is a fundamental challenge for any mobile organism. Identical retinal patterns—such as a looming object—can be caused by an external threat or by the animal's own movement. Distinguishing between these sources is critical for survival, dictating whether an animal should approach a resource or flee a predator. This study investigates how the superior colliculus (SC), a multi-sensory hub in the midbrain, integrates locomotion with visual signals to resolve this ambiguity.
Experimental Framework
Researchers utilized a virtual reality (VR) setup where head-restrained mice ran on a treadmill. This allowed for precise control over visual stimuli while recording neural activity from two distinct regions of the SC: the superficial layers (SCs), which primarily receive visual inputs, and the intermediate layers (SCim), which are more closely linked to motor control and non-visual inputs.
The experimental design included three key conditions:
- Object Loom: A dark object moves toward a stationary or running mouse.
- Object Approach: The mouse runs toward a stationary virtual object, creating self-generated looming.
- Replay: The visual sequence from a previous "Object Approach" is replayed, creating a potential mismatch between the visual loom and the mouse's actual treadmill speed.
Layered Specialization and Sensorimotor Integration
The study revealed that the SC is not a uniform structure. The SCs functions as a "vision-first" layer, where neural activity is dominated by the distance to the object. In contrast, the SCim acts as an integration hub. Locomotion consistently increased activity in the SCim (mean increase of 0.23 z-score), whereas its effect on the SCs was variable and net-zero.
Mice also exhibited an "instinctive sensorimotor response" to looming stimuli, slowing their locomotion as an object neared, regardless of whether the motion was self-generated or external. Crucially, when visual feedback did not match their movement (the Replay condition), mice often stopped entirely, indicating an awareness of the sensorimotor conflict.
Decoding Context from Neural Activity
Using population activity, the researchers found that the SCim was significantly more effective at discriminating between "self-motion" (VR) and "external-motion" (Replay) than the SCs. Logistic regression decoders trained on SCim activity improved classification performance by 56.1% over behavioral data alone, compared to only a 28.8% improvement using SCs activity. This suggests that the SCim specifically encodes the context of visual stimuli—whether they are expected based on the animal's own actions.
Limitations and Future Directions
While the VR setup provides excellent control, head-restraint limits the mouse's natural behavioral repertoire, such as turning or jumping. Furthermore, the stimuli were restricted to frontal looms. Future research could investigate how these context-dependent signals are communicated to other brain regions, such as the motor cortex or pulvinar, to coordinate complex survival behaviors in three-dimensional environments.
Summary updated