The brain always predicts what the eye will see next, relying on an internal feedback network that physically rewires it to match the patterns it encounters in the world. Experiments manipulating vision in mice have revealed that our visual system behaves like an active learning machine, providing a biological basis for how the mind generates situational expectations. The results of the animal experiment were published in Current Biology.
When light hits the retina, information travels through the back of the brain directly to the primary visual cortex. This region identifies basic visual elements such as primitive edges and simple geometric shapes. From there, the data travels to higher-order brain regions that stitch together simple inputs to process more abstract concepts, such as entire moving objects or complex scenes.
Information does not flow only in one direction, upwards. Higher order areas constantly send dense bundles of neural signals back to the primary visual cortex. These descending connections are known as feedback paths.
This feedback architecture helps the brain provide the necessary context for what the eye currently perceives. If part of the shape is hidden in shadow, the feedback network uses statistical probabilities learned from nature to infer the missing part. This system uses past visual scenes to predict what’s in your blind spot.
Biologists have recognized that early visual experiences shape how these feedback networks mature in young animals. Whether this visual input simply functions as a general switch that initiates a precise biological sequence has remained largely unknown. An alternative hypothesis was that the feedback network receives direct instructions from the visual world and shapes its physical architecture to mimic certain statistical properties of the native environment.
Radhika Rajan and Rodrigo F. Diaz from the Champalimaud Neuroscience Program in Portugal led a research team to resolve this uncertainty. They designed an experiment to examine what happens within the brain’s physical wiring when the visual world is artificially restricted during the first few weeks of life.
The researchers fitted custom-designed steel goggles to young mice, which were about 45 days old. These goggles had special cylindrical lenses that distorted incoming light before it reached the animal’s eyes. The lens restricted the animal’s vision to a narrow range in a particular direction, somewhat similar to seeing the world through a series of stationary parallel bars.
One group of mice wore goggles that only allowed them to see at a full 45-degree angle. Another group of mice wore goggles that restricted their vision to a full 135-degree angle. A third control group wore goggles with flat, unaltered plastic lenses that provided normal vision.
The mice lived for more than a month wearing the goggles in a specially enriched environment. The cage had patterned walls and sufficient space to explore, allowing the animal to continuously engage with the changing visual environment. The goal was to completely immerse the animal in an environment completely dominated by a single geometric angle.
To observe how this limited image uptake affects neural circuits, the scientists used a technique called two-photon imaging. The method involves introducing a fluorescent protein into a targeted group of brain cells. These special proteins light up briefly each time a cell experiences a spike of electrical activity, allowing researchers to watch living brain cells communicate in real time.
The scientists focused their microscope on the precise connection point where nerves from a higher-order area called the lateral medial area terminate in cells in the primary visual cortex. They recorded microscopic flashes of light while lightly anesthetized mice watched moving geometric patterns displayed on a laboratory monitor.
The researchers measured the responses of individual nerve endings to see which visual patterns evoked the most electrical activity. Brain cells usually indicate a preferred direction. This means that it fires fastest when presented with a line drawn at a certain angle. They also have receptive fields, specific small patches of visual space that are responsible for monitoring.
In mice wearing goggles, cells in the primary visual cortex completely changed their orientation preferences. The cells became highly sensitive to specific angles forced by the optical lens. This means that mice raised to look only at 45-degree angles developed large amounts of extra brain cells dedicated solely to 45-degree angles.
Feedback nerves descending from higher visual cortex perfectly reflected this sensory change. The entire feedback population adjusted its adjustments to the angle of orientation the mouse experienced within the cage. The actual physical parameters of the receptive fields of these feedback nerves also changed their shape.
These receptive fields are usually somewhat rounded and symmetrical. In laboratory mice, the boundaries of the receptive field shape were stretched and lengthened. They became tilted ellipses aligned in the exact direction of the forced optical experience.
This experiment also revealed structural changes in the spatial mapping of descending feedback networks. Nerves that send top-down signals typically overlap precisely with primary visual cells located directly beneath the cortical tissue. Scientists expected to receive comprehensive feedback covering key areas.
Instead, they found that visual experience changes how different neural sets distribute information about the surrounding visual background. Feedback nerves that were actively responding to patterns on the monitor would be less likely to send visual data from the spatial axis corresponding to a particular angle of the goggles. The feedback system was reorganized around the excess geometric signals.
To understand how these physical changes occur at the biological level, researchers turned to computer modeling. They built a mathematical simulation of the visual cortex to test different governing rules for how the brain’s synapses adapt over time. Synapses are small chemical connections between individual brain cells through which electrical signals pass.
Computer models showed that this change likely results from two competing learning behaviors occurring simultaneously in the same brain region. The ascending path from the primary visual cortex appeared to follow rules known as Hevian plasticity.
Hebbian plasticity is the principle by which neurons that fire together wire together. Consistent visual exposure to a single specific angle strengthened the chemical bonds between ascending cells that were repeatedly activated by that angle. This mechanism explains why the lower brain has developed an exaggerated sensory preference for a single geometric direction.
Conversely, descending feedback pathways coming from higher regions likely utilize a mechanism called anti-Hebian plasticity. In this scenario, when a feedback synapse fires at exactly the same time as the primary visual cell it targets, the feedback synapse actively weakens and regresses. This system essentially eliminates redundancy.
This anti-Hevian rule acts as an active decorrelation mechanism. This prevents higher regions of the brain from simply echoing the exact same visual information back to lower regions. Feedback networks selectively amplify new visual information that deviates from what the animal normally sees by reducing redundant and overlapping signals.
Although experimental results provide a mechanistic explanation of how visual expectations are formed, this approach has limitations. Mathematical modeling demonstrates what is computationally possible based on known biological laws, but it does not directly measure the microscopic molecular changes occurring at individual synapses. Proving the existence of anti-Hebian processes in living animals is notoriously difficult.
This methodology also focused only on vertical connections between different layers and areas of the visual brain. The researchers acknowledge that horizontal connections between adjacent cells within the very same brain layer likely contribute to shaping these variable responses. Currently, the model treats neural layers as isolated vertical columns.
Future research will need to elucidate how horizontal mapping and vertical feedback networks interact during normal development. Moving beyond screen-based geometric patterns will help researchers understand how these wiring rules apply to active visual perception in the wild.
The study, “Visual experience plays a beneficial role in cortical feedback input to the primary visual cortex,” was authored by Radhika Rajan, Rodrigo F. Dias, Nikos Malakasis, Margarida Baeta, Xinyun Zhang, Julijana Gjorgjieva, and Leopoldo Petreanu.
