Mapping how local perturbations influence systems-level brain dynamics

Introduction

• How localized neural changes are globally integrated remains unresolved, including the specificity of global effects and the principles shaping altered communication.The mechanisms by which local insults may propagate to large-scale system deficits are also unknown.Source: Introduction · pp. 3–4
• The paper reviews empirical and computational studies of how local changes propagate to large-scale brain dynamics, then presents an original computational investigation.The investigation assesses local perturbation effects on within- and across-system dynamics.Source: Introduction · p. 4
• The review focuses on non-invasive perturbations and excludes broad-effect tDCS, which is described as modulating rather than perturbing local neural activity.Source: Introduction · p. 4
• The whole-brain stimulation map highlights cortical hierarchy and core-periphery organization as features explaining the effects of local activity changes.Source: Introduction · pp. 4–5

Part I: An overview on the system-level effects of local changes in brain activity

• FEF stimulation enhanced activity in distal visual areas and improved detection of peripheral relative to foveal stimuli.The remote activity changes suggest altered communication between the targeted FEF and early visual regions.Source: Part I: An overview on the… · pp. 5–6
• Focal TMS selectively alters functional and effective connectivity within related systems and can also change interactions between systems (Grefkes and Fink, 2011; Grefkes et al., 2010; Eldaief et al., 2011; Cocchi et al., 2015; Valchev et al., 2015; Watanabe et al., 2014).Source: Part I: An overview on the… · p. 6
• Inhibition of right primary motor cortex increased sensorimotor-system integration and reduced its communication with other systems, while rich-club connectivity patterns were preserved.Source: Part I: An overview on the… · p. 6
• Although structural connectivity may contribute to selective TMS effects, its role in determining functional-network changes remains unclear, making experiment planning and interpretation challenging.Source: Part I: An overview on the… · p. 6
• Whole-brain mean-field models use coupled neural masses or neural fields, suited to macroscopic dynamics and stimulation because noninvasive recording and stimulation affect large neuronal populations (Deco et al., 2008; Foster et al., 2016; Allen et al., 2007).Source: Part I: An overview on the… · p. 7
• Prior TMS models have explored spatially uniform field effects or only a few regions, limiting their extension to whole-brain perturbation mapping (Fung et al., 2013; Wilson et al., 2014; Fung and Robinson, 2013, 2014; Wilson et al., 2016a; Wilson et al., 2016b; Cona et al., 2011).A three-region coupled-neural-mass approach is difficult to scale to the whole brain.Source: Part I: An overview on the… · p. 8
• Structural-network spreading models identify hubs and backbones as routes for activity and associative areas as integrators, but derive global dynamics from connectivity without local dynamics.Structural connectivity also constrains the spread of dynamics.Source: Part I: An overview on the… · p. 9
• Simplified local-dynamics models address that limitation: Kuramoto models capture cortical oscillations, while normal-form models with connectome data explore patterns after focal stimulation (Spiegler et al., 2016).Normal-form models use few parameters to capture nonlinear phenomena such as bifurcations to oscillations.Source: Part I: An overview on the… · p. 9
• TMS parameters must avoid over-excitation that can trigger seizures (Rossi et al., 2009); seizure spread from focal onset is a network phenomenon that patient-specific modeling may inform therapeutically (Proix et al., 2016).Spatially extended neural models have studied focal seizure initiation and spread.Source: Part I: An overview on the… · pp. 9–10
• Modeling safe stimulation ranges and brain-dynamics stability is a research direction, alongside growing interest in controllability and noninvasive stimulation (Fung and Robinson, 2013; Wilson et al., 2016b; Betzel et al., 2016).Source: Part I: An overview on the… · p. 10

Part II: Modeling the global effects of focal changes in neural activity

• The simulations assess global connectivity changes by target-region connectivity, hierarchical effects on whole-brain connectivity, and within-system responses to local perturbations.Together, these analyses address how localized changes modulate large-scale dynamics.Source: Part II: Modeling the global effects… · p. 10
• The study uses a Kuramoto model to systematically test focal perturbations across the cortex, balancing simplicity with plausibility for whole-brain functional connectivity.The model incorporates anatomical constraints and macroscopic dynamics captured by neuroimaging.Source: Part II: Modeling the global effects… · pp. 10–11
• The model represents slow BOLD fluctuations in 513 brain regions as coupled oscillators, with coupling based on anatomical connections and intrinsic frequencies tuned to resting-state data.The coupling parameter was calibrated against resting-state fMRI, and the structural connectivity matrix came from a group-averaged connectome.Source: Part II: Modeling the global effects… · p. 11
• Simulations run for 10 minutes across 50 random initial conditions; after discarding the first two minutes, the study averages functional connectivity and partitions it into systems.The system partition follows the approach of Power et al. (2011).Source: Part II: Modeling the global effects… · p. 11
• Figure 1 summarizes the model’s structural connectivity, natural frequencies, rich-club organization, simulated functional connectivity, and functional-system assignments.The connectome uses a 513-node parcellation, and the natural-frequency distribution was tuned to reproduce resting-state connectivity patterns.Source: Figure 1 · p. 12

Anatomical connectivity

• Probabilistic tractography in 75 healthy adults estimated group-average white-matter tracks between 513 regions, yielding 262,656 connections.The resulting connectome was normalized so its maximum weight was 1.Source: Anatomical connectivity · pp. 12–13
• The rich club comprises the 75 cortical regions whose anatomical strength is at least one standard deviation above the mean.The figure depicts hub-to-hub connections within this group.Source: Anatomical connectivity · p. 13

A hierarchy of intrinsic timescales across the cortex

• Each region’s intrinsic frequency is assumed to depend on anatomical node strength, with frequencies spanning 0.01–0.1 Hz.The mean natural frequency is 0.0883 Hz, and this hierarchy maps onto a gradient of timescales.Source: A hierarchy of intrinsic timescales… · p. 13
• The natural-frequency distribution was tuned to reproduce resting-state fMRI connectivity, and coupling was calibrated to match baseline functional connectivity.Figure 2 displays natural frequencies, coupled frequencies, and coupling-induced shifts across regions ordered by anatomical connectivity strength.Source: A hierarchy of intrinsic timescales… · pp. 13–14, Figure 2 · p. 14
• After coupling, slower hub regions speed up while faster peripheral regions slow down.The coupling-induced shift is shown as a mean across trials.Source: A hierarchy of intrinsic timescales… · p. 13

Brain stimulation

• TMS-related frequency and power changes are represented by changing the stimulated region’s natural frequency from baseline.Inhibitory stimulation reduces the natural frequency; excitation increases it.Source: Brain stimulation · pp. 14–15

• The Kuramoto model incorporates heterogeneous natural frequencies, which network interactions slightly shift in the resulting dynamics.Source: Results · p. 15
• Natural frequencies depend on each region’s anatomical connectivity strength, linking the anatomical hierarchy with intrinsic oscillatory dynamics.Source: Results · p. 15
• Peripheral regions exhibit fast dynamics, whereas brain hubs operate in slower regimes.Source: Results · p. 15
• The authors systematically stimulate each of 513 brain regions in numerical experiments to characterize changes in brain dynamics.Source: Results · p. 15
• The framework may guide hypothesis-driven empirical work because experimentally mapping widespread effects of target-dependent local activity changes is challenging.Source: Results · p. 15

Tuning curves: Effects of local inhibition or excitation on whole-brain dynamics

• Inhibition and excitation produce opposing tuning curves (r = -0.6): inhibition increases ΔFC in weakly connected regions and decreases it in core regions, with the reverse pattern for excitation.The strongest and weakest connectivity regions are largely insensitive to simulation.Source: Tuning curves: Effects of local… · p. 16
• The spatial pattern after local inhibition consists of small islands of strong correlation increases or decreases, without a clear structure.Source: Tuning curves: Effects of local… · p. 16
• Stimulation strength changes tuning-curve amplitude rather than shape, while the largest regional ΔFC response grows nonlinearly and can reverse beyond a threshold.The region with the maximum ΔFC response differs across stimulation intensities, underscoring region-specific effects.Source: Tuning curves: Effects of local… · p. 16
• Figure 3 shows region-dependent ΔFC tuning curves: connectivity changes follow a pattern organized by node strength, with response propensity determined by each region’s natural frequency relative to the mean.Strongly connected regions have slower timescales and weakly connected regions faster timescales; their responses intersect at the mean natural frequency.Source: Tuning curves: Effects of local… · pp. 15–16, Figure 3 · p. 17
• Figure 4 indicates that the periphery-to-core axis better explains stimulation effects across the brain than rostro-caudal position, which shows a trend only in frontal regions.Outside the most rostral regions, dependence on rostro-caudal position is relatively flat.Source: Figure 4 · p. 18, Tuning curves: Effects of local… · p. 18
• Mean seed-to-whole-brain ΔFC can conceal large, balanced increases and decreases in connectivity that cancel in the net effect.Source: Tuning curves: Effects of local… · p. 19
• As stimulation intensity rises, absolute-change curves for excitation and inhibition diverge in peripheral and core regions: excitation reduces peripheral effects, whereas inhibition increases their connectivity with slower-regime regions.Core hubs show a substantially smaller divergence because stimulation effects there are reduced.Source: Tuning curves: Effects of local… · p. 20
• Figure 6 shows that absolute ΔFC changes are greater in peripheral than core regions, while regions near the mean natural frequency show relatively large magnitudes despite opposing signed effects.Inhibition and excitation yield similar absolute-change curves under weak stimulation.Source: Tuning curves: Effects of local… · p. 20, Figure 6 · p. 21
• Regions near the whole-brain mean natural frequency can produce balanced, equal-and-opposite functional-connectivity changes, while very fast or slow regions drive only small effects.Regions close to the mean are more likely to increase or decrease functional connectivity with other regions.Source: Tuning curves: Effects of local… · pp. 21–22
• Projecting functional influence onto the cortical surface provides a systematic map of inhibition and excitation effects on large-scale resting-state brain dynamics.Source: Tuning curves: Effects of local… · p. 22

Effects of local changes in brain activity on specialized functional brain systems

• External-region functional influence was consistent across systems, whereas internal-region influence varied substantially; system functional influence was anticorrelated with mean functional-system strength (r = -0.45) and strongest constituent-region strength (r = -0.55).Source: Effects of local changes in brain… · pp. 22–23
• Within-system connectivity changes depended on stimulation-site location, with neighboring regions generally showing similar effects and internal hotspots prominent in peripheral systems.All networks had internal and external hotspots; internal hotspots were dominant in visual and somatosensory networks relative to high-order associative systems.Source: Effects of local changes in brain… · p. 23
• Inhibition of hub regions produced weaker within-system Δ|FC| changes than inhibition of peripheral regions in both visual and cingulo-opercular systems.The magnitude and regional variability of stimulation effects were enhanced in peripheral regions.Source: Effects of local changes in brain… · p. 25, Figure 8 · p. 25
• Across systems, local-stimulation effects followed relatively smooth tuning curves linked to anatomical connectivity strength, with peripheral regions causing larger connectivity changes than core hubs.The authors attribute this pattern to peripheral regions’ flexible dynamics and similar timescales with many other regions, whereas hub dynamics are more stable and constrained.Source: Stimulation tuning curves emerge from… · pp. 26–27
• Inhibition and excitation have symmetric, opposite effects: inhibition tends to increase connectivity for faster peripheral regions and decrease it for slower core regions, with converse effects under excitation.Peripheral-region excitation saturates at greater intensity, typically weakening its effect on global connectivity; peripheral regions can have large absolute effects because their natural frequencies resemble those of many regions.Source: Stimulation tuning curves emerge from… · p. 28
• Functional influence is similar across stimulation sites outside the targeted system but varies substantially across sites within it, supporting selective changes through calibrated local modulation.The results define functional influence as the mean effect of local stimulation on connectivity within a system.Source: Selectively manipulating large-scale… · p. 29
• Cortical hotspots can significantly alter system connectivity; their influence decays smoothly around broad, imprecisely bounded areas, making small TMS coil-position errors less consequential.The resulting regional influence maps may help plan TMS studies targeting specific systems.Source: Selectively manipulating large-scale… · p. 30
• The left DLPFC is a valuable target for modulating connectivity with and within the default-mode network, while simulations predict stronger overall modulation from the right DLPFC.Within the left DLPFC, the rostro-dorsal portion is predicted to produce greater effects on connectivity between the DLPFC and default-mode regions.Source: Selectively manipulating large-scale… · p. 30
• Peripheral visual and somatosensory systems are more reactive to local activity changes than high-level cingulo-opercular systems, so stimulation intensity should account for both target region and system.Connectivity patterns are more likely to change when stimulation targets a region within a peripheral rather than a high-level system; this prediction requires empirical testing.Source: Selectively manipulating large-scale… · p. 31

From data to theory, and back again

• The study proposes an iterative theory–experiment loop: empirical findings inform models that generate predictions, whose testing can validate or refine the mechanisms and models.The reported tuning curves and results are predictions requiring future experimental tests.Source: From data to theory, and back again · p. 31

• The model is experimentally validated only for V1/V2 and FEF among 513 regions; effects at other sites remain predictions, motivating comprehensive experimental mapping.Source: From data to theory, and back again · p. 32
• Predicting network impact requires considering the intermingled functional and structural characteristics of targeted regions.Source: Conclusion · p. 33
• A periphery–core hierarchy determines the effects of local stimulation on large-scale brain systems.Source: Conclusion · p. 33
• The study’s maps can guide experiments testing model-generated mechanistic hypotheses and help identify sites for manipulating specific functional connections.Source: Conclusion · p. 33