Intrinsic functional connectivity mapping via resting-state BOLD fMRI (rsfMRI) has highlighted the presence of spatially-correlated spontaneous oscillations defining reproducible brain network systems underlying known as well as less understood brain functions. However, classic rsfMRI approaches typically rely on correlational connectivity measurements intrinsically insensitive to the direction of information flow among affected regions¹. The recent development of computational approaches enabling a description of rsfMRI networks in directional terms has highlighted the possibility of effectively describe how brain region interact and communicate in the resting-state, with encouraging initial results. Multivariate Granger Causality (GC)² in humans highlighted putative causal interactions between distributed rsfMRI networks, and revealed robust directional insular and middle temporal to prefrontal connectivity as well as a major receiving role from hippocampal formation areas³.

By employing rigorous control of motion and physiological artefacts, we have recently shown the presence of robust and reproducible rsfMRI networks in the mouse brain, including SN and a putative default mode network (DMN)^4,5. Here, we applied GC to probe group- and subject level directional functional connectivity among key nodes of distributed mouse rsfMRI networks. In particular, we first validated our GC methodology by demonstrating its ability to reveal directed connectivity in the ventro-hippocampal-prefrontal network, a neural system well studied in rodents and humans^6,7 and characterised by mono-directional underlying axonal connectivity^8,9. We then used this methodology to examine directional functional connectivity in key nodes of the mouse brain insular-prefrontal (“salience”-like) and DMN, two distributed systems characterized by complex anatomical connectivity patterns⁸, and that could putatively be related to analogous human networks.

Experiments were performed on male C57BL/6J (B6) mice (n=41) as previously described^4,5. RsfMRI timeseries were acquired under halothane anaesthesia (0.7%). Image acquisition: All experiments were performed by a 7.0 Tesla MRI scanner, using a single-shot EPI sequence with TR/TE 1200/15 ms, flip angle 30°, matrix 100 × 100, field of view 2 × 2 cm², 24 coronal slices, slice thickness 0.50 mm and a total rsfMRI acquisition time of 6 min. Image preprocessing: images were motion corrected, spatially normalized, smoothed band-pass filtered and regression of motion traces and the mean ventricular signal was applied. Granger causality (GC) analysis: We defined multiple sets of seed ROIs within the mouse Hippocampus-PFC network, SN & DMN that can be related to analogous key subsystems of the human. For each set of network we computed unconditional GC( UGC)² and conditional GC(CGC)²(one node conditional GC(1C-GC) and 2C-GC for 4 node network probing) along with dominant directional causality¹⁰ defined as the difference between causality in each direction. Subject level analysis: we used iMAAFT¹¹ non-parametric surrogates’ construction methods to generate 1000 surrogates for each pair of regions and selected significance as p<0.05, FDR corrected. We first fine-tuned GC analysis by applying over different GC measures to find similar directionality patterns following the anatomical monodirectional wiring of the ventro-hippocampal-prefrontal(vHC-PFC) network at both population and subject level (Fig. 1). CGC revealed robust and reproducible directional connectivity patterns along direction of hippocampal axonal projections^8,9. Probing of the insular prefrontal system highlighted the presence of dominant antero-posterior functional connectivity patterns reminiscent of directional features recently described for the human salience network^12,13(Fig. 2). To understand the effect of limited data size in four node network, we carried out simulation on different sample size of the data using full VAR² model fitted to all nodes in vHC-Rs-PFC network and found evidence that 1C-GC was the most efficient compared to other GC strategies (Fig. 3). We applied this approach to probe a four key nodes of the mouse DMN and observed the presence of robust directional functional connectivity between temporal associative and prefrontal regions, together with a “sink effect” of retrosplenial cortex with respect of temporal cortical areas, recapitulating analogous human findings ^14,15(Fig. 4). All the effects were reproduced with different sets of ROIs and also in data added with pink noise to homogenize signal throughout the brain.We demonstrate robust directional connectivity patterns in distributed rsfMRI networks of the mouse brain, including converging dominant connectivity between lateral cortical areas and medial prefrontal and cingulate regions, replicating topological features of known human rsfMRI networks, and consistent with a higher integrative function subserved by these areas. These results provide a novel interpretative window on the intrinsic functional architecture of the mouse brain at the macroscale.