Resting state fMRI connectivity is sensitive to laminar connectional architecture in the human brain

Abstract

Previous invasive studies indicate that human neocortical graymatter contains cytoarchitectonically distinct layers, with notable differences in their structural connectivity with the rest of the brain. Given recent improvements in the spatial resolution of anatomical and functional magnetic resonance imaging (fMRI), we hypothesize that resting state functional connectivity (FC) derived from fMRI is sensitive to layer-specific thalamo-cortical and cortico-cortical microcircuits. Using sub-millimeter resting state fMRI data obtained at 7 T, we found that: (1) FC between the entire thalamus and cortical layers I and VI was significantly stronger than between the thalamus and other layers. Furthermore, FC between somatosensory thalamus (ventral posterolateral nucleus, VPL) and layers IV, VI of the primary somatosensory cortex were stronger than with other layers; (2) Inter-hemispheric cortico-cortical FC between homologous regions in superficial layers (layers I–III) was stronger compared to deep layers (layers V–VI). These findings are in agreement with structural connections inferred from previous invasive studies that showed that: (i) M-type neurons in the entire thalamus project to layer-I; (ii) Pyramidal neurons in layer-VI target all thalamic nuclei, (iii) C-type neurons in the VPL project to layer-IV and receive inputs from layer-VI of the primary somatosensory cortex, and (iv) 80% of collosal projecting neurons between homologous cortical regions connect superficial layers. Our results demonstrate for the first time that resting state fMRI is sensitive to structural connections between cortical layers (previously inferred through invasive studies), specifically in thalamo-cortical and cortico-cortical networks.

Acquisition and processing methods of whole-brain layer-fMRI VASO and BOLD: The Kenshu dataset

“…There are indications of feedback signatures in: V1-V2, V1-V3, V1-V4, and V1-MT. Also overall, we find the largest similarities between superficial layers compared to deeper layers (Guidi et al 2020; Huber et al 2017; Deshpande, Wang, and Robinson 2022; Pais-Roldán et al 2020). This might indicate that feedback signals dominate the temporal signal fluctuations during repeated movie watching.…”

“…Early attempts of layer-fMRI connectivity studies have been somewhat limited to relatively small field of views, constrained to individual brain systems (Polimeni et al 2010; Guidi et al 2020; Huber et al 2017; Huber, Finn, et al 2020). More recent advancements in data sampling approaches, MR-contrast generation strategies, and confidence of the laminar signal interpretability allowed proof-of-principle extensions of layer-fMRI connectivity to larger FOV (Sharoh et al 2019; Huber et al 2021; Deshpande, Wang, et al 2022; Pais-Roldán et al 2020; Yun et al 2022). In this work, we aim to help the layer-fMRI community in building tools to make such whole-brain layer-fMRI connectivity protocols usable for neuroscience application studies.…”

Submillimeter fMRI Acquisition Techniques for Detection of Laminar and Columnar Level Brain Activation

“…To facilitate the various analysis approaches suggested for rs‐fMRI (eg regional homogeneity, functional connectivity density or independent component analysis), whole‐brain coverage is demanded as the first requirement in rs‐fMRI. In early work, several submillimeter protocols were proposed to investigate subcortical functional profiles for rs‐fMRI: Guidi 2020 (1.15 mm 3 ; 10 slices; 1.65 s), 48 Huber 2021 (0.51 mm 3 ; 72 ~ 104 slices; 6.5 ~ 8 s), 47 Deshpande 2022 (1.08 mm 3 ; 37 slices; 3 s), 30 and Yun 2022 (0.25 mm 3 ; 123 slices; 3.5 s), 9 where the parameters presented in the parenthesis show the achieved single‐voxel volume, slices and temporal resolution provided in each study.…”

Noninvasive Characterization of Functional Pathways in Layer-Specific Microcircuits of the Human Brain Using 7T fMRI

“…Therefore, we performed blind hemodynamic deconvolution of raw BOLD time series to obtain latent neural signals and used them in DGC estimation. This approach has been employed and validated in multiple previous studies [ 93 , 94 , 95 , 96 , 97 , 98 ]. We employed a recently validated framework based on the Cubature Kalman filter and smoother to invert a nonlinear hemodynamic model [ 50 ].…”

“…The traditional method to resolve this issue is to interpolate the fMRI signal at certain cortical depths and then average the surface profiles [ 43 , 46 , 47 ]. The fact that we performed hemodynamic deconvolution before connectivity analysis also helps in removing some of the contributions of vasculature [ 98 ]. However, we need to come up with a better model to extract laminar signals to increase the spatial specificity of fMRI.…”