Ultra-high resolution fMRI reveals origins of feedforward and feedback activity within laminae of human ocular dominance columns

Abstract

Ultra-high field MRI can functionally image the cerebral cortex of human subjects at the submillimeter scale of cortical columns and laminae. Here, we investigate both in concert, by, for the first time, imaging ocular dominance columns (ODCs) in primary visual cortex (V1) across different cortical depths. We ensured that putative ODC patterns in V1 (a) are stable across runs, sessions, and scanners located in different continents (b) have a width (~1.3 mm) expected from post-mortem and animal work and (c) are absent at the retinotopic location of the blind spot. We then dissociated the effects of bottom-up thalamo-cortical input and attentional feedback processes on activity in V1 across cortical depth. Importantly, the separation of bottom-up information flows into ODCs allowed us to validly compare attentional conditions while keeping the stimulus identical throughout the experiment. We find that, when correcting for draining vein effects and using both model-based and model-free approaches, the effect of monocular stimulation is largest at deep and middle cortical depths. Conversely, spatial attention influences BOLD activity exclusively near the pial surface. Our findings show that simultaneous interrogation of columnar and laminar dimensions of the cortical fold can dissociate thalamocortical inputs from top-down processing, and allow the investigation of their interactions without any stimulus manipulation. Significance StatementThe advent of ultra-high field fMRI allows for the study of the human brain non-invasively at submillimeter resolution, bringing the scale of cortical columns and laminae into focus. De Hollander et al imaged the ocular dominance columns and laminae of V1 in concert, while manipulating top-down attention. This allowed them to separate feedforward from feedback processes in the brain itself, without resorting to the manipulation of incoming information. Their results show how feedforward and feedback processes interact in the primary visual cortex, highlighting the different computational roles separate laminae play.

Correcting for Superficial Bias in 7T Gradient Echo fMRI

“…We also applied the six correction methods to the GE-fMRI 7T data acquired on the same paradigm. The uncorrected data showed the characteristic increase in BOLD signal towards superficial layers (Hollander et al, 2020; Kok et al, 2016). This superficial profile remained after Z-scoring or applying SVM or LDC, while the Voxel Ratio produced an extremely variable estimate with apparently-greatest modulation in the middle layer.…”

“…These limitations of alternative sequences have resulted in continued popularity for BOLD-GE laminar fMRI (Hollander et al, 2020;Kok et al, 2016;Lawrence et al, 2019;Liu et al, 2020). There is instead increasing interest in using post-processing analysis techniques to remove the bias in the BOLD signal towards superficial layers (Fracasso et al, 2018;Polimeni et al, 2010;Yacoub et al, 2013), thus mitigating the main drawback of BOLD-GE sequences.…”

“…Meanwhile, the leakage model attempts to capture the effect of draining veins on the fMRI data as it simulates the propagation of signal from the deep layers to the superficial layers, mimicking the effect of draining veins carrying oxygenated blood towards the superficial layers. In terms of methods that seek to correct bias in empirical studies, proposed solutions include Z-scoring timecourses (Lawrence et al, 2019), L2 normalization of beta estimates across conditions (Kay et al, 2019), taking the ratio of activations in two experimental conditions (Kashyap et al, 2017;Liu et al, 2020) and decoding using multi-voxel classification techniques (Hollander et al, 2020;Muckli et al, 2015).…”

Investigation of layer specific BOLD in the human visual cortex during visual attention

“… Muckli et al ( 2015 ), Kok et al ( 2016 ), Lawrence et al ( 2018 )). Two-by-two experimental designs are surprisingly rare in layer-specific analysis and have only been recently employed by de Hollander et al ( 2020 ). We emphasise that a multifactorial design is important to account for changes due to, for example, cortical depth and signal leakage per se, as opposed to true layer-based changes in activity due to the experimental manipulations.…”

“…The thickness of the cerebral cortex varies between 1 and 4.5 millimetres ( Zilles, 1990 ; Fischl and Dale, 2000 ), giving sufficient resolution to characterise activity across the individual layers. FMRI is now often used to try and measure layer specific activation in, for example, the visual system ( Muckli et al, 2015 ; Kok et al, 2016 ; Lawrence et al, 2018 ; de Hollander et al, 2020 ), the motor system ( Huber et al, 2018 ), working memory tasks ( Finn et al, 2019 ), and to find directional connectivity between language areas ( Sharoh et al, 2019 ). If layer specific analysis can make good on its promise of reliably discerning layer specific signals, it can be useful for answering questions in a wide range of cognitive domains ( Lawrence et al, 2019 ) and for questions of directional connectivity and cognitive network neuroscience ( Huber et al, 2020 ), including research questions involving spatial attention.…”