Abstract

Introduction: The hippocampus (HP) is a key region for important cognitive functions including episodic memory and spatial navigation¹. Hitherto, these functions have been extensively studied at macroscale^2,3. However, the underlying circuit-level mechanisms are poorly understood. Recent advancements in functional magnetic resonance imaging (fMRI) at submillimeter scale in combination with the availability of new analysis approaches and software packages have enabled reliable measurement of depth-dependent activity in isocortex^4,5. Nonetheless, the application of such recordings in allocortex has been limited⁶ primarily due to methodological challenges. Harnessing the current developments in the field, we aimed to dissociate the contribution of laminar activations in HP subregions to self-selected navigation strategies. Methods: Thirty-nine healthy volunteers (mean age = 25.22 years, 17 females) underwent structural and functional MRI in 2-day sessions using a 7T MAGNETOM-Terra scanner. During the first day, an anatomical image was acquired with 3D-MP2RAGE sequence (isotropic spatial resolution of 0.75 mm³, TR/TE/TI1/TI2 = 6000/1.85/800/2700 ms, flip angles = 4/5°, matrix size = 256 x 340 x 340, bandwidth = 290 Hz/px). This was followed by the acquisition of two functional runs using a 3D GE-EPI sequence with BOLD contrast (isotropic voxel size of 0.8 mm³, phase-encoding direction = A-P, TR/TE = 2500/28.40 ms, matrix size = 240 x 240 x 40, flip angle = 14°, phase partial Fourier = 7/8, bandwidth = 1096 Hz/px, acceleration factor = 4 with GRAPPA reconstruction). On the second day, six additional functional runs were obtained. During each run, the participants were asked to navigate to six hidden objects distributed across random locations in a circular virtual arena⁷. Movement in the arena was enabled using a button box. Each participant completed 18 trials per run (three repetitions per object) consisting of a retrieval and a subsequent re-encoding phase (see Figure 1). Given that the paradigm was self-paced, the functional data had variable duration, ranging between 240–330 volumes per run (~10-13 minutes). Moreover, an auxiliary functional scan with 13 volumes and opposite phase-encoding direction was acquired prior to acquisition of every fMRI scan to facilitate retrospective correction of susceptibility-induced distortion artefacts. The structural and functional data were preprocessed using an in-house developed pipeline based on ANTS, FSL, SPM and ‘PreSurfer’ packages⁸. Briefly, the preprocessing steps included bias field correction of the anatomical and functional data, motion and distortion correction of the fMRI data and their subsequent alignment with the anatomical images. Afterwards, HP was segmented into subregions including cornu ammonis (CA 1-4) and dentate gyrus using the Hippunfold package⁹. Furthermore, three folded surfaces were generated using an equivolume model, representing inner/outer sections and mid-thickness of the HP gray matter⁸ (Figure 2). These surfaces were then transformed into the space of the corresponding anatomical image and equidistantly sampled into 20 depth bins using an in-house MATLAB script. This allowed the extraction of the BOLD signal profile across the bins per subregion, time-point and hemisphere in the fMRI data, from which the motion estimates were regressed out. The bins were then grouped into the inner and outer depths, corresponding to superficial and deep layers of HP, respectively. While the preprocessing steps and HP segmentation have been completed in all participants, the extraction of laminar profiles in subregions is currently ongoing (N = 7 until now). Additionally, we quantified putative navigation strategies via the following metrics, illustrated in Figure 3: straightness index (SI; the ratio between the optimal path length towards the drop location and the length of the actual taken path), and median deviation to the boundary (MDB; the difference between the median distance of the optimal and the observed paths from the boundary). Subsequently, we performed linear mixed-effect models to assess the association of subregional laminar activity with navigation strategies. Note that for this analysis, the BOLD signal was extracted only from the timepoints when the subjects were navigating in the arena.
Results: We observed an inverse U-shaped profile of the BOLD signal across the depths of CA1 whereas the BOLD profile in CA3 demonstrated an almost monotonous decrease from inner to outer depths. This pattern was consistent across participants (Figure 4). Our analyses on preliminary data further suggest a negative association between higher SI and lower activity in inner and outer depths of CA3 (β = -6.6 & -7.7, p < 0.001, respectively).
Conclusions: These preliminary results demonstrate the feasibility of laminar-resolution recordings in human HP and suggest that distinct navigation strategies are related to subregion-specific laminar activities.