Researchers at the Champalimaud Foundation, in Lisbon, have for the first time managed to identify, with an imaging technique, whether nervous impulses in the brain of rats are flowing in a “bottom-up” (feedforward), carrying information about visual input, or a “top-down” (feedback) direction, carrying information about expectations or predictions on a given task or about the perception of the world around us. Their results, which are published today (May 4th 2026) in the journal Nature Communications, could have important implications for understanding changes in the brains of people with hallucinations, Alzheimer’s, schizophrenia, autism and other conditions.
Joana Carvalho, first author of the new study, who at the time was working in the Preclinical MRI lab, led by senior author Noam Shemesh (she has since become a group leader at Coimbra University), “came up with the ideas, did the experiments and analysed the results. I just brought the MRI expertise”, says Shemesh good-humouredly. Co-author Koen V. Haak, from Tilburg University (Netherlands), gave assistance with the computational models and the others helped with the experiments.
The team showed that spontaneous feedforward and feedback nervous impulses in these rodents (the brain never sleeps) each have a unique, distinct signature, which can be detected by using a method they developed, called uFLARE (UltraFast Layer-Resolved Encoding), a neuroimaging technique designed to map brain activity with unprecedented high temporal and spatial resolutions.
Achieving the ultrafast (tens of milliseconds) acquisition of submillimeter-resolution images that was needed to distinguish between feedforward and feedback nervous information flow would not have been possible without the ultra-high field experimental MRI scanner in Shemesh’s lab, which generates a magnetic field of 9.4 Tesla (in comparison, medical MRI machines usually generate fields of one to three Tesla). Thanks to this machine, the team was able to perform ultrafast functional MRI (or fMRI), and to measure the activity of cortical circuits at high speed, capturing rapid changes.
However, in order to interpret the data, “a second ingredient was needed besides these very high temporal and spatial resolution images”, says Shemesh. “What was needed was a computational, bidirectional model of how information flows between different parts of the brain.”
The team’s work focused on the visual cortex, which is organised in six distinct layers, and in which target regions of nervous impulses in each direction are well known and distinct. Namely, feedforward signals arriving into the visual cortex from the sensory organs target primarily layer IV of the visual cortex, while feedback connections from higher order visual areas to the primary visual cortex preferentially project to layers I and VI.
A directional model of the visual cortex
Accordingly, the model the researchers used to correctly discriminate feedforward and feedback pathways in the visual cortex was a so-called “layer‑based connective field model”. This model of neuronal communication predicts a target neuron's activity in a specific layer based on neuronal signals from another part of the brain.
In this model, the size of the connective field of a group of neurons “is a measure of the extent of the information that is being sent or received by different neurons”, Carvalho explains. “If we have a large connective field size, that means that a particular population of neurons receives information from a large group of neurons from another area”, she adds.
This means that the variation of the size of the connective field across layers differs systematically for ascending (feedforward) and descending (feedback) pathways. Indeed, for feedforward signals, the model predicts that this size reaches a maximum in the intermediate layers of the visual cortex, whereas for feedback connections it is maximal for both the most superficial and deepest layers.
When the researchers extracted the relevant information about the size and distribution of the connective fields from the huge sequences of fMRI images they had obtained, it matched the size and distribution predicted by the model. In other words, the signature for feedforward signals extracted from the images was radically distinct from the signature for feedback signals, correctly inferring the directionality of information flow.
The researchers further showed that the methodology also applies to other brain areas, such as the somatosensory and motor systems, suggesting a general pattern of cortical communication.
The results could have important implications for understanding human disease. “In autism or Parkinson's or Alzheimer's, and also in many types of diseases where there is a lesion in the brain”, Shemesh explains, “all we can say at the moment is that the pathways of feedward and feedback activity change, but we don't know how this happens nor do we understand what is different. With this new method and this new analysis we can study the changes in directional information flow. It's a new way of looking into activity in the brain.”
“We are now trying to see if we can replicate these patterns of feedback and feedforward connections in humans”, says Carvalho. “That’s where it becomes interesting: we may then be able to see how a lesion influences the feedback and the feedforward signals that are actually causing the changes in the brain. In people with autism, we know that their perception is different because they have different integrative mechanisms, and we may be able to see if they concern feedback or feedforward signals. We can now try to answer these questions.”