We propose a new method, the Functionnectome, to investigate the functional involvement of brain circuits during task-related cerebral processes. Applying our method to a high-quality functional neuroimaging dataset revealed - for the first time - the white matter circuits supporting motor, working memory, and language activations. Results also indicated a higher reproducibility of the Functionnectome maps compared to classical task-related activation methods suggesting that the functioning of the brain is mediated by interactions via anatomical circuits rather than the isolated contribution of brain regions. To support the broad uptake of this method and facilitate its application to a wide range of datasets, including in the clinic, we provide a GUI and terminal-based companion software. This toolbox allows the implementation of the Functionnectome to any previously acquired fMRI dataset and beyond. While the toolbox is flexible and users can integrate their own priors, the current release and the high quality priors accompanying it open up novel avenues for research on the integrative function of white matter.
Obtaining significant activation of brain circuits with our method requires the functional involvement of the brain areas at both ends of this circuit. As first examples of such functional exploration of white matter, we investigated the brain circuits supporting prominent activation tasks including motor, memory and language functions. Our knowledge of the fine circuitry of the motor system is mostly derived from animal and lesion studies. The Functionnectome applied to motor tasks offer the unique opportunity to explore these circuits in the healthy human brain. In that regard, our findings match the pathways suggested by the literature and were replicated with both hands and feets twice. We confirmed the involvement of the internal capsule, which is a well established31 part of the ascending and descending motor pathways that include the cortico-subcortical motor loops32,33 and the cortico-cerebellar pathways. The latter has been long known to be part of the motor system34, with relays in subcortical nuclei, but has never been directly shown before. The cortico-ponto-cerebellar pathways35 connect the primary motor cortex of one hemisphere with ipsilateral pontine nuclei and the contralateral anterior lobe of the cerebellum, passing through the internal capsule and the pons. The involvement of the corpus callosum in the motor tasks cannot be directly identified with fMRI but has long been considered essential to integrate the left and right motor systems36,37,. The Functionnectome maps distinctly isolated these circuits.
Similarly, the patterns of white matter activation from the working memory task confirm and complement the literature. For example, the involvement of the superior longitudinal fasciculus reported with the Functionnectome has been well described38. Our results also confirm the importance of the cerebellum in working memory35,39 but extend this insight by mapping the full circuitry supporting its involvement. Similarly, the fronto-frontal tracts involvement in the working memory Functionnectome supports the hypothesis of the frontal lobe working hierarchically8,20,40. Additionally, classical working memory activation tasks41 and clinical studies24 unveiled the importance of both hemispheres for working memory, but researchers could only speculate about the underlying anatomical circuitry. Here, the Functionnectome revealed the exact portion of the corpus callosum that integrates bilateral contribution to working memory.
The language circuitry, and more so its semantic system, offers an even greater challenge as it cannot be explored in animal studies26. The Functionnectome confirmed, for the first time in the healthy human brain, the structural-functional circuitry supporting semantic processes originally suggested by lesion studies25 or intraoperative stimulation26 in humans (i.e. the uncinate fasciculus, the inferior fronto-occipital fasciculus, the middle longitudinal fasciculus, and the posterior segment of the arcuate fasciculus). The Functionnectome also revealed the involvement of the anterior corpus callosum in story comprehension. Whilst some language processes require the integration of the left and right hemispheres via the posterior corpus callosum42, the anterior corpus callosum has been implicated in semantic disorders (e.g. semantic dementia)43. The Functionnectome result also suggests an involvement of the fornix. As a limbic pathway, it may play an important role in the colouring of the story comprehension with emotions and memories30. Our results thus prompt a closer inquiry into the role of the anterior corpus callosum and fornix in semantic processes and offer a non-invasive tool to study its involvement in healthy participants. In sum, our application of the Functionnectome to classical fMRI allows the confirmation and the exploration of the involvement of circuits for specific tasks for the first time in the healthy human brain.
As reproducibility of findings is at utmost importance in science44 we verified whether our activation maps were consistent across different acquisitions. The replication of our analysis confirmed the high reproducibility of the results highlighted by our method. Importantly, the Functionnectome results were more reproducible than classical task-related activation methods. While the two methods are not identical and not perfectly comparable with regards to filtering, the observed differences also emerge from the fact that they rely on different assumptions for the assessment of the functioning of the brain. While classical fMRI computes differences between regions of the brain independently, the Functionnectome associates their circuits to brain function. Higher reproducibility for the Functionnectome would then suggest that the functioning of the brain is mediated by interactions via anatomical circuits rather than the isolated contribution of brain regions.
To upscale the validation of this network view of brain functioning, crowdsource analysis of additional data is possible. In this regard, we provide an open-source software that will allow easy application of the Functionnectome method to revisit already acquired fMRI datasets, either private or publicly available (e.g. HCP45, UK Biobank46, ABIDE47). The use of the Functionnectome is also not bound to the activation paradigm and can be combined with advanced fMRI statistics7 to reveal the dynamic causal interaction between brain circuits. Additionally, the Functionnectome can leverage the wealth of MRI modalities to explore the involvement of white matter circuits in different aspects of brain dynamics. For example, it could be applied to resting-state functional connectivity or cortical thickness to open up new perspectives in the study of functional synchronisation, cortical changes during development, and brain pathologies.
The Functionnectome is a new and promising method that relies on anatomical priors to determine how to project the functional signal onto the white matter. The current best way to obtain a complete 3D map of the white matter pathways in the living human brain is tractography. Although this method has been successfully applied to explore the relationship between white matter structure, and brain functions and dysfunctions10,16, tractography is still facing limitations48. Nonetheless, great progress has been made towards the resolution of these problems in the last decade13. Future developments in this area will likely improve upon the quality of current tractograms. These improvements will be implemented in the Functionnectome as priors can easily be replaced (see supplementary material) in the future to incorporate novel advances in tractography. Similarly, although we provide high-quality anatomical priors based on best white matter mapping derived from the Human Connectome Project 7T49 in the software, other research teams are welcome to use homemade priors. Future developments of the priors might include a separation in interhemispheric (i.e. commissural circuits), cortico-subcortical (i.e. projection circuits) and cortico-cortical connectivity (i.e. association circuits) to better disentangle brain circuitries.
Finally, the computation of the Functionnectome using every brain voxel can be very computationally expensive and thus time-consuming. While we recommend using this procedure, we acknowledge that not all research teams have access to the computing power required to compute the “voxel-wise” Functionnectome of several subjects in a reasonable amount of time. To circumvent this potential constraint, we provide an option within the software allowing the use of an atlas and its parcels instead of all voxels, for a less computationally intensive “region-wise” analysis. The anatomical priors come with this alternative option using the recently published multimodal parcellation atlas50, should you choose to use this option (see supplementary materials).
Overall, we introduced and demonstrated the potential of the Functionnectome method, opening the field of in-vivo study of the function of white matter in healthy humans. In this context, the Functionnectome promotes a paradigm shift in the study of the brain, focusing on the interaction of brain regions in the support of a brain function, rather than the fractionated contribution of independent regions.