Modeling and Analysis - Connectivity

Brain network organization is constrained by a trade-off between the energetic cost of forming connections and the advantages they confer on network function[1]. Computational models have shown this cost-benefit trade-off can explain many-but not all-network properties[2,3]. During gestation, brain development proceeds according to a precise spatiotemporal pattern defined by a series of morphogen gradients[4]. Cortical areas display heterochronicity, differential timing of key developmental events, which induces spatial patterns that persist in later life as smoothly varying gradients in cytoarchitecture, neuronal connectivity, and functional activation[5]. Therefore, heterochronicity may provide an additional constraint on the formation of cortical connectivity[6,7]. While the putative role of developmental timing on cortical wiring has been modelled abstractly[8,9], prior studies have not accounted for the brain's geometry, a critical factor in modeling the potential role of diffusing molecular gradients. To address this, we developed a framework to examine how heterochronicity and synchronicity across cortical areas can shape human brain networks. 

Glucose metabolism is essential for providing the energetic supply that supports the neural dynamics vital for brain function. Abnormal glucose metabolism, such as hypo- and hypermetabolism, is frequently reported in patients with neurological diseases, including Alzheimer's disease and epilepsy. It has become clear that regional metabolic changes are not isolated incidents; rather, local inhibitory circuitry or diaschisis, due to connections from pathological regions, may also affect local metabolism. These remote impacts, which can be characterized by metabolic covariance networks (MCNs), have been found to be indicative of brain pathologies [1]. However, the biological substrates of such covariance-specifically, the basis on which this covariance emerges-remain to be elucidated. Based on multimodal imaging data from a group of temporal lobe epilepsy (TLE) patients, we explore the fundamental determinants of MCN, including intrinsic functional connectivity (FC), white matter pathways (structural connectivity, SC), morphometric similarity (structural covariance network, SCN), geometric proximity (Euclidean distance, ED), as well as neurochemical and genetic underpinnings. 

At any moment, neuromodulators bind and unbind from neurons, driving brain activity through neuroreceptors and transporters. How the coordination of neuromodulators drives brain activity is nontrivial to quantify. Traditional in vivo methods, such as fluorescent probes (Marvin et al., 2013), microdialysis (Watson et al., 2006), and electrochemistry (Bucher & Wightman, 2015), provide limited insight, measuring only a few neuromodulators or brain regions at a time. The invasive nature of these methods further limits their application in humans, highlighting the need for a non-invasive approach.

Neuroreceptor density, crucial in mediating neuromodulator effects, can be estimated non-invasively using PET imaging. Neuroreceptor maps have been linked to brain-wide communication measured by the BOLD signal, suggesting that combining PET-derived neuroreceptor maps with BOLD signals could uncover how neuromodulators drive human brain activity and elucidate mechanisms of cognition and behavior (Salvan et al., 2023). 

Understanding and predicting signal propagation of Transcranial Magnetic Stimulation (TMS) through the human connectome would be immensely useful across research and clinical contexts. To date, neural mass and field models as well as graph-theoretic measures have been employed (Bortoletto et al., 2015; Gollo et al., 2017; Momi et al., 2021; Seguin et al., 2023), but they face challenges in balancing spatial resolution and computational efficiency, particularly when applied to large-scale brain networks. Here, we developed a novel Resistor-Capacitor (RC) circuit model to predict TMS-induced signal propagation across the human connectome. We compared cortical maps of model-predicted TMS-induced activation patterns to those measured in a previous interleaved TMS-fMRI experiment in which triplets of 10Hz stimulation pulses were targeted to the dorsolateral prefrontal cortex (DLPFC) (Tik et al., 2023). Our model provides dynamic, time-resolved cortical maps of the stimulation-induced brain activity at high spatial resolution, which accord with fMRI measurements. 

How do inter-areal connections shape brain function? One theoretical framework posits that intrinsic (i.e., stimulus-independent) inter-areal connectivity patterns, or a region's connectivity receptive field (CRF), is as a major determinant of its functional localization (Passingham et al., 2002). In our previous study (Oh et al., 2024), we introduced the integrated effective connectivity (iEC) – a biologically grounded measure that reliably estimates directed intrinsic connectivity using empirical resting-state fMRI data. Here, we found that CRFs derived from task-free iEC (i) predicts the task tuning curves of diverse cognitive tasks in an independent task dataset, (ii) identifies functionally localized clusters of brain regions, and (iii) systematically vary along the cytoarchitectural hierarchy. 

The cerebral cortex takes the form of a sheet of interconnected neurons that interact both through local intracortical connections, and through a complex network of long-range connections (LRCs) that facilitate rapid communication between distant neural populations. LRCs play a well-established functional role in supporting information processing across distributed neural systems, and underpin the prevalent conceptualization of the brain as a communication network (or connectome) of functionally specialized regions. However, recent results have challenged this network-based view, showing that key properties of resting-state fMRI dynamics can be accurately captured by simple geometric models that neglect the LRCs measured from connectomics^1,2. A key open question thus remains: if LRCs are crucial for cortical function, why do they appear to play a minimal role in capturing key dynamical properties of resting-state fMRI?
Here we address this question through a range of investigations using a novel mathematical model of cortical dynamics, in which neural populations interact both through a connectome of LRCs and through a sheet of local connections. For a large variety of connectome topologies, we demonstrate that, in spontaneous settings and on long timescales (as per resting-state fMRI data), simulated brain dynamics increasingly resemble that of a geometric model that excludes the connectome. Our results thus provide a plausible account for the role of LRCs in shaping cortical dynamics on different length and timescales, and explain why LRCs (which predominantly shape fast information processing of precise input stimuli) have a minimal role in shaping resting-state fMRI dynamics.