Fine-scale computations for adaptive processing in the human brain

Elisa Zamboni, Vg Kemper, Nr Goncalves, Ke Jia, Vm Karlaftis, Samuel J. Bell, Jj Giorgio, Reuben Rideaux, Rainer Goebel, Zoe KourtzibioRxiv 2020 · DOI

TL;DR

This study uses ultra-high field fMRI to investigate how the brain adapts to repetitive visual stimuli at a fine-grained, layer-by-layer level within the cortex. They found that adaptation primarily occurs in superficial layers of the visual cortex, driven by recurrent processing, and is further modulated by top-down feedback from parietal areas, revealing a sophisticated circuit for efficient information processing.

What Problem Does This Paper Address?

The brain is remarkably adept at processing the constant stream of sensory information we encounter. A key mechanism for this efficiency is sensory adaptation, where the brain reduces its response to repeated or predictable stimuli. While we know adaptation happens, the precise neural computations and circuits involved, especially at the fine-grained level of cortical layers, remain largely mysterious. Standard fMRI struggles to resolve these deep cortical structures, leaving a gap in our understanding of how feedforward (bottom-up) and feedback (top-down) processing contribute to this crucial adaptive mechanism. This paper aims to bridge that gap by using advanced imaging techniques to peer into the layered structure of the brain.

Methods

The core of this study lies in the application of ultra-high field (UHF) 7 Tesla (7T) fMRI. This technique offers sub-millimeter resolution, allowing researchers to distinguish signals from different cortical layers, a feat not possible with standard fMRI. They employed a laminar fMRI approach, specifically analyzing BOLD (Blood-Oxygen-Level-Dependent) signals across three distinct cortical layers: deeper, middle, and superficial.

The experimental paradigm involved an orientation adaptation task. Participants viewed sinewave gratings. In "adaptation" blocks, the same grating orientation was repeatedly presented, while in "non-adaptation" blocks, orientations varied. This allowed researchers to measure the difference in brain activity (fMRI adaptation) between predictable and unpredictable stimuli.

To ensure the observed signals were truly from neuronal activity and not just blood vessel artifacts (a common challenge in fMRI, especially at high fields), they performed several control analyses. These included using a different fMRI sequence (3D GRASE) known to be less susceptible to large draining veins, and applying signal unmixing techniques to isolate superficial signals.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Superficial layers are key players in adaptation: The most striking finding is that fMRI adaptation (a decrease in BOLD signal for repeated stimuli) is significantly stronger in the superficial layers of the visual cortex compared to middle and deeper layers. This suggests that the brain's adaptation mechanisms are particularly active at the "output" stage of cortical processing.
  • Recurrent processing drives adaptation in V1: The observed adaptation in superficial layers points towards recurrent processing within the visual cortex itself, rather than solely relying on input from earlier stages. This means the visual cortex is actively re-processing and suppressing familiar information.
  • Top-down control from parietal cortex: The study also reveals a crucial role for feedback from the posterior parietal cortex (involved in expectation and attention) to deeper layers of the visual cortex. This suggests that our expectations about what we're about to see actively shape how we process incoming visual information.
  • Dissociable connectivity for adaptation: The research highlights distinct pathways for adaptation: enhanced feedforward connectivity within the visual cortex (V1 to V4) for processing the adapted stimulus, and feedback connectivity from the parietal cortex to V1 for top-down modulation.

Key Results

  • Visual Cortex Adaptation: fMRI adaptation (reduced BOLD signal for repeated stimuli) was observed across all cortical layers in visual areas (V1-V4), but was significantly stronger in superficial layers compared to middle and deeper layers.
  • Posterior Parietal Cortex Adaptation: Similar adaptation effects were found in the posterior parietal cortex (IPS1, IPS2) across its cortical layers.
  • Functional Connectivity:
    • Feedforward: Connectivity between superficial V1 layers and middle V4 layers was enhanced during adaptation compared to non-adaptation, suggesting stronger feedforward processing.
  • Feedback: Connectivity between deeper V1 layers and IPS1 was enhanced during adaptation compared to non-adaptation, indicating stronger feedback processing.
  • Control Analyses: The observed laminar differences in adaptation were robust and not attributable to vascular confounds, as confirmed by comparisons with a 3D GRASE sequence and signal unmixing techniques.

Limitations and Future Work

The authors acknowledge that while UHF imaging provides unprecedented detail, GE-EPI sequences can still be limited by vasculature-related signals, even with their correction methods. Future work could further explore these dynamics using other advanced imaging techniques like cerebral blood volume (CBV) imaging (VASO) to enhance spatial specificity.

They also suggest that combining laminar fMRI with electrophysiological recordings (e.g., EEG, MEG) would be a powerful next step to unravel the temporal dynamics of these recurrent and feedback circuits, particularly the role of different oscillatory frequencies (gamma for feedforward, alpha/beta for feedback) in adaptive processing. Understanding these circuit dynamics is highlighted as a key challenge for deciphering fast brain plasticity mechanisms.

Summary updated

Abstract

Abstract Adapting to the environment statistics by reducing brain responses to repetitive sensory information is key for efficient information processing. Yet, the fine-scale computations that support this adaptive processing in the human brain remain largely unknown. Here, we capitalize on the sub-millimetre resolution afforded by ultra-high field imaging to examine BOLD-fMRI signals across cortical depth and discern competing hypotheses about the brain mechanisms (feedforward vs. feedback) that mediate adaptive visual processing. We demonstrate suppressive recurrent processing within visual cortex, as indicated by stronger BOLD decrease in superficial than middle and deeper layers for gratings that were repeatedly presented at the same orientation. Further, we show dissociable connectivity mechanisms for adaptive processing: enhanced feedforward connectivity within visual cortex, while feedback occipito-parietal connectivity, reflecting top-down influences on visual processing. Our findings provide evidence for a circuit of local recurrent and feedback interactions that mediate rapid brain plasticity for adaptive information processing.
BibTeX
@misc{zamboni2020fine,
  author = {Elisa Zamboni and VG Kemper and NR Goncalves and Ke Jia and VM Karlaftis and Samuel J. Bell and JJ Giorgio and Reuben Rideaux and Rainer Goebel and Zoe Kourtzi},
  title = {{Fine-scale computations for adaptive processing in the human brain}},
  year = {2020},
  doi = {10.1101/2020.02.14.947895},
  howpublished = {bioRxiv},
}
Figure