September 14, 2017
Predicting what we'll see
Neurobiologists in Georg Keller's group at the FMI have identified a neuronal circuit in cortex that integrates signals from movement with visual input: a projection from the motor areas of cortex to the visual cortex conveys predictions of visual flow based on movement. Neural activity in this projection adapts as mice learn to navigate a left-right inverted virtual maze; and artificial activation of the projection creates an illusory visual flow, influencing the mice?s motor behavior.
Let's start with a little experiment: shut your eyes and conjure up a mental image of the scene in front of you – the computer monitor, the keyboard, maybe a coffee mug. Now, in this mental image, shift your gaze to the right, without moving either your head or your eyes. Then – keeping your eyes closed – try again while turning your head in the appropriate direction. Most people find it much easier to "look around" a mental image while actually performing the movements that would be necessary if their eyes were open.
Georg Keller, Group leader at the Friedrich Miescher Institute for Biomedical Research (FMI) in Basel, explains: "We believe there's an integration of signals generated by movement and vision in the brain, and we postulate that signals generated during movement constitute predictions of the visual input based on movement. For example, the motor command for a right turn would lead to a prediction of visual flow to the left."
But how can this be tested experimentally? The neurobiologists first checked whether the brain region that computes the input from movement – the motor cortex – is connected to the region that processes visual information – the visual cortex. In mice, they found that there is indeed a dense, topographically organized projection from motor regions of the cortex, which targets most neurons in layer 2/3 of the visual cortex. The scientists subsequently showed that the activity of these neurons is strongly correlated with movement.
However, if different cortical areas work together to generate predictions, and if these are continuously updated based on sensory input, then, says Keller, "we should also see a change in activity as the visual input changes." To demonstrate this, mice were trained to navigate a left-right inverted virtual environment. Marcus Leinweber, a postdoctoral fellow in Keller's lab, explains: "In this environment, the visual feedback is uncoupled from the motor input. The mice had to learn to turn right if the visual scene moved to the right. And, with learning, the activity of the same neuronal connections dynamically adapted to the new coupling." Finally, by stimulating these neurons, the scientists were able to elicit illusory visual flows – and observed concomitant changes in behavior.
Keller concludes: "Predictions play an important role in perception. In the theory of predictive coding, the cortex generates and maintains an internal model of the world by continuously updating and comparing this model to actual sensory input. Here, we identified a first set of neuronal connections in cortex that convey predictions – in this case, concerning visual flow."
Leinweber M, Ward DR, Sobczak JM, Attinger A, Keller GB (2017) A sensorimotor circuit in mouse cortex for visual flow predictions. Neuron 95:1420-1432
About Georg Keller
Georg Keller is a Junior group leader at the FMI. He and his group aim to elucidate the key principles underlying sensory processing in the visual cortex.
» More about Georg Keller