Skip to main content
NCBI home page

This is a preprint.

It has not yet been peer reviewed by a journal.

The National Library of Medicine is running a pilot to include preprints that result from research funded by NIH in PMC and PubMed.

bioRxiv logoLink to bioRxiv
[Preprint]. 2024 Mar 28:2023.06.22.546023. [Version 2] doi: 10.1101/2023.06.22.546023

In-vivo whole-cortex marker of excitation-inhibition ratio indexes cortical maturation and cognitive ability in youth

Shaoshi Zhang 1,2,3,4,5,*, Bart Larsen 6,7,8,*, Valerie J Sydnor 6,7,8,*, Tianchu Zeng 1,2,3,5, Lijun An 1,2,3,5, Xiaoxuan Yan 1,2,3,4,5, Ru Kong 1,2,3,5, Xiaolu Kong 1,2,3,5,9, Ruben C Gur 7,8,10, Raquel E Gur 7,8,10, Tyler M Moore 7,8, Daniel H Wolf 7, Avram J Holmes 11,12, Yapei Xie 1,2,3,5, Juan Helen Zhou 1,2,4,5, Marielle V Fortier 13,14, Ai Peng Tan 14,15, Peter Gluckman 16, Yap Seng Chong 14,17, Michael J Meaney 14,18, Gustavo Deco 19,20, Theodore D Satterthwaite 6,7,8,+, BT Thomas Yeo 1,2,3,4,5,21,+
PMCID: PMC10996460  PMID: 38586012

Abstract

A balanced excitation-inhibition ratio (E/I ratio) is critical for healthy brain function. Normative development of cortex-wide E/I ratio remains unknown. Here we non-invasively estimate a putative marker of whole-cortex E/I ratio by fitting a large-scale biophysically-plausible circuit model to resting-state functional MRI (fMRI) data. We first confirm that our model generates realistic brain dynamics in the Human Connectome Project. Next, we show that the estimated E/I ratio marker is sensitive to the GABA-agonist benzodiazepine alprazolam during fMRI. Alprazolam-induced E/I changes are spatially consistent with positron emission tomography measurement of benzodiazepine receptor density. We then investigate the relationship between the E/I ratio marker and neurodevelopment. We find that the E/I ratio marker declines heterogeneously across the cerebral cortex during youth, with the greatest reduction occurring in sensorimotor systems relative to association systems. Importantly, among children with the same chronological age, a lower E/I ratio marker (especially in association cortex) is linked to better cognitive performance. This result is replicated across North American (8.2 to 23.0 years old) and Asian (7.2 to 7.9 years old) cohorts, suggesting that a more mature E/I ratio indexes improved cognition during normative development. Overall, our findings open the door to studying how disrupted E/I trajectories may lead to cognitive dysfunction in psychopathology that emerges during youth.

1. Introduction

Healthy brain function requires a delicate balance between neural excitation (E) and inhibition (I) (1–4). This balance – the E/I ratio – is refined during critical developmental periods of heightened experience-dependent plasticity (5,6). E/I imbalances during critical developmental periods are thought to contribute to the etiology of many psychiatric disorders (7,8) and confer vulnerability to cognitive deficits (9,10). Here we capitalize on advances in biophysically plausible large-scale circuit models to chart the normative development of cortex-wide E/I ratio and uncover links to cognition.

Human cortical development unfolds hierarchically – sensory systems mature earlier, while association systems follow a more protracted developmental course extending through adolescence (11,12). A potential mechanism driving this hierarchical development might be the temporal progression of critical plasticity periods along the sensorimotor-to-association axis (13–16). More specifically, the maturation of GABAergic inhibitory circuitry involving parvalbumin positive (PV) interneurons suppresses stimulus-irrelevant activity, yielding a higher signal-to-noise ratio (13). The maturation of PV interneurons also modulates longterm potentiation by enforcing a narrower spike integration window (17). Overall, the maturation of the inhibitory circuitry facilitates the experience-dependent pruning of excitatory pyramidal neuronal connections via the Hebbian mechanism, triggering a critical plasticity period (18–20). Therefore, a hallmark feature of the critical period development is a reduction in the E/I ratio (21,22). While the hierarchical progression of inhibitory development is documented in animal models (14,23), it is unclear if the same mechanisms exist in humans, extend to the evolutionarily expanded association cortex and impact cognitive ability.

Studying E/I ratio development in-vivo in humans is challenging due to limitations in non-invasive neuroimaging techniques. MR spectroscopy studies suggest changes in the balance of excitatory and inhibitory neurotransmitter levels in single brain regions during development (24,25). A recent study used a machine learning marker trained with pharmacological-fMRI data to provide evidence of E/I ratio reduction in the association cortex during development (26). However, these past studies were limited to partial portions of the cortex, so normative development of cortex-wide E/I ratio remains unclear. Indirect estimates of whole-cortex E/I ratio have provided insights into autism spectrum disorder in adults and Alzheimer’s Disease (27–29), but these approaches mostly lack a direct mapping to an underlying biophysically-plausible model of excitatory and inhibitory dynamics.

Biophysically-plausible large-scale circuit models of coupled brain regions have provided mechanistic insights into spontaneous brain dynamics (30–32). However, most large-scale circuit models assume that local synaptic properties are spatially uniform across brain regions (9,27,33), which lacks biological plausibility. Indeed, spatial heterogeneity in excitatory and inhibitory cell types (34–36) might be a driver of large-scale brain dynamics (37,38). Studies have shown that incorporating spatial heterogeneity across local synaptic parameters generates more realistic spontaneous brain dynamics (39,40). Our previous study (41) demonstrated that parameterizing local synaptic parameters with anatomical and functional gradients led to dramatically more realistic brain dynamics in adults. However, we utilized a large-scale circuit model (42) that did not differentiate among excitatory and inhibitory neural populations, so the E/I ratio could not be derived.

Here we investigate the development of cortical E/I ratio over youth and its association with cognitive ability. We apply our previous approach (41) to the feedback inhibition control (FIC) model with coupled excitatory and inhibitory neuronal populations (33). The resulting parameteric FIC (pFIC) model is used to derive a potential marker of E/I ratio. We first confirm that the pFIC model yields realistic brain dynamics in healthy young adults from the Human Connectome Project (HCP; 43). Using a pharmacological fMRI dataset (44), we show that the E/I ratio marker is sensitive to E/I ratio reduction induced by the GABA-agonist alprazolam. Then, using the Philadelphia Neurodevelopmental Cohort (PNC; 45,46), we find that the E/I ratio declines across the cortex during youth. Furthermore, a lower E/I ratio indexes greater cognitive ability, with the strongest relationships observed in association cortex. We generalize the link between E/I ratio and cognitive ability in a younger GUSTO (Growing Up in Singapore with Healthy Outcomes) cohort (47). Overall, our study suggests that E/I ratio maturation might be a driver of healthy neurocognitive development during youth.

2. Results

2.1. Overview

We first evaluated the optimization of the spatially heterogeneous parametric feedback inhibition control (pFIC) model in the HCP dataset (Figure 1A). The biological plausibility of the estimated marker of excitation-inhibition (E/I) ratio was then evaluated using pharmacological fMRI involving GABAergic benzodiazepine alprazolam. Finally, we investigated developmental changes and cognitive effects of E/I ratio in the PNC dataset. Assocations with cognition were replicated in the GUSTO cohort.

Figure 1. Workflow and schematic of the parametric feedback inhibition control (pFIC) model.

Figure 1.

Open in a new tab

(A) Young adults from the Human Connectome Project (HCP) were used to evaluate the optimization of the spatially heterogeneous pFIC model. Pharmacological fMRI with benzodiazepine alprazolam was then used to evaluate the biological plausibility of the estimated E/I ratio. Next, the pFIC model was used to investigate the development of cortex-wide E/I ratio and its association with cognitive ability in the PNC dataset. Cognitive associations were replicated in a sample of seven-year-olds from the GUSTO cohort. HCP logo is used with permission from the HCP team. (B) The FIC model (33) is a neural mass model obtained by mean field reduction of spiking neuronal network models. The FIC model consists of differential equations at each cortical region governing the neural dynamics of excitatory and inhibitory neuronal populations (“E” and “I” respectively in the right panel). A red triangle indicates an excitatory connection. A blue circle indicates an inhibitory connection. “wxy” indicates the connection strength from neuronal population x to neuronal population y. For example, “wIE” indicates the connection strength from the inhibitory population to the excitatory population. The regional models are connected by excitatory connections parameterized by a structural connectivity (SC) matrix. For a given set of model parameters, time courses of excitatory (SE) and inhibitory (SI) synaptic gating variables (representing the fraction of open channels) can be simulated. The excitation-inhibition ratio (E/I ratio) was defined as the ratio between the temporal average of SE and SI. Local synaptic parameters were estimated using the same approach as our previous study (41). We refer to the resulting model as the pFIC model.

2.2. Optimization of the parametric feedback inhibition control (pFIC) model

We randomly divided 1004 HCP participants (43,48) into 3 non-overlapping training (N = 335), validation (N = 335), and test (N = 334) sets. The Desikan–Killiany anatomical parcellation (49) with 68 cortical regions of interest (ROIs) was used to generate group-level structural connectivity (SC), static functional connectivity (FC), and functional connectivity dynamics (FCD) from the training, validation, and test sets separately. To compute FCD for each fMRI run, a 68 × 68 FC matrix was computed for each sliding window of length ~60 seconds. The 68 × 68 FC matrices were then correlated across the 1118 windows, yielding a 1118 × 1118 FCD matrix (41). The FCD matrix has been shown to reflect temporal fluctuations in resting-state FC that are not captured by static FC (50,51). See Supplementary Methods S2 for details.

The FIC model (33) is a neural mass model obtained by mean field reduction of a spiking neuronal network model (52,53). The model comprises ordinary differential equations (ODEs) at each cortical region describing the dynamics of excitatory and inhibitory neuronal populations (Figure 1B top panel). The local dynamics are driven by recurrent connections within separate excitatory and inhibitory populations, as well as connections between excitatory and inhibitory populations. Greater excitatory-to-excitatory recurrent strength (wEE) and smaller inhibitory-to-excitatory connection strength (wIE) amplify synaptic currents of the excitatory population. Similarly, greater excitatory-to-inhibitory connection strength (wEI) and smaller inhibitory-to-inhibitory recurrent strength (wII) amplify synaptic currents of the inhibitory population. Neuronal noise in each cortical region is controlled by the noise amplitude σ. Finally, the excitatory populations of the regional local models are connected via the SC matrix, scaled by a global constant G.

Following previous studies (33,39), wII was set to one and wIE was automatically set to maintain a uniform baseline excitatory firing rate of around 3Hz. Excitatory-to-excitatory recurrent strength (wEE), excitatory-to-inhibitory connection strength (wEI), regional noise amplitude (σ) and the SC scaling constant (G) were estimated using our previous approach (41). More specifically, wEE, wEI and σ were parameterized as a linear combination of the principal resting-state FC gradient (54) and T1w/T2w myelin estimate (55), resulting in 9 unknown linear coefficients and 1 unknown parameter G. We refer to the resulting model as the parametric FIC (pFIC) model.

The 10 pFIC parameters were estimated using the covariance matrix adaptation evolution strategy (CMA-ES) (56) by minimizing the difference between simulated and empirical fMRI data. More specifically, agreement between empirical and simulated FC matrices was defined as the Pearson’s correlation (r) between the z-transformed upper triangular entries of the two matrices. Larger r indicates more similar static FC. However, Pearson’s correlation does not account for scale difference, so we also computed the absolute difference (d) between the means of the empirical and simulated FC matrices (39). A smaller d indicates more similar static FC. The inclusion of d was necessary to prevent overly synchronized fMRI signals (Figure S1). Finally, we do not expect the brain states of two participants to be the same at any given timepoint t during the resting-state, i.e., there is no temporal correspondence between participants in the resting-state. Because FCD was computed based on sliding window FC, there was similarly no temporal correspondence between simulated and empirical FCD matrices. Therefore, disagreement between the simulated and empirical FCD matrices was defined as the Kolmogorov–Smirnov (KS) distance, following previous studies (41,50). The KS distance was defined as the maximum distance between the cumulative distribution functions obtained by collapsing the upper triangular entries of simulated and empirical FCD matrices, so no temporal correspondence was assumed (more details in Supplementary Methods S10). The overall cost was defined as (1 – r) + d + KS. A smaller cost indicates better agreement between simulated and empirical fMRI.

2.3. The pFIC model generates realistic fMRI dynamics

We first demonstrate that the parametrization of the local synaptic parameters with T1w/T2w and FC gradient led to more realistic brain dynamics than spatially homogeneous parameters (Figure 2A). We applied CMA-ES to the HCP training set to generate 500 candidate model parameter sets. The 500 parameter sets were evaluated in the HCP validation set. The top 10 parameter sets from the HCP validation set were used to simulate FC and FCD using SC from the HCP test set, which were then compared with empirical FC and FCD from the HCP test set. A strong agreement between simulated and empirical FC (as well as between simulated and empirical FCD) would suggest that the pFIC model was able to generate realistic brain dynamics.

Figure 2. The parametric feedback inhibition control (pFIC) model generates more realistic fMRI dynamics than the spatially homogeneous FIC model.

Figure 2.

Open in a new tab

(A) The CMA-ES algorithm (41,56) was applied to the HCP training set to generate 500 sets of model parameters. The top 10 parameter sets from the validation set were evaluated in the test set. (B) Agreement (Pearson’s correlation) between empirical and simulated static FC in the HCP test set. (C) Empirical FCD from an HCP test participant. (D) Simulated FCD from the pFIC model using the best model parameters (from the validation set) and SC from the test set. (E) Total test cost of the pFIC model compared with three control conditions: (1) local synaptic parameters parameterized by only principal resting-state FC gradient, (2) local synaptic parameters parametrized by only T1w/T2w ratio map and (3) local synaptic parameters constrained to be spatially uniform. The boxes show the inter-quartile range (IQR) and the median. Whiskers indicate 1.5 IQR. Black crosses represent outliers. * indicates that the pFIC model achieved statistically better (lower) test cost.

Figure 2B visualizes the correlation between empirical and simulated FC in the HCP test set (based on the best model parameters from the validation set). Across the 10 best parameter sets from the validation set, correlation between empirical and simulated static FC was 0.71 ± 0.005 (mean ± std) in the test set. As a reference, correlation between empirical FC and SC in the test set was 0.48. On the other hand, the absolute difference between the means of the empirical and simulated FC matrices was 0.11 ± 0.015 in the test set. This suggests that the pFIC model was able to generate realistic FC.

Figure 2C shows the empirical FCD from a single run of a representative HCP test participant. Figure 2D shows the simulated FCD using the best model parameters (from the validation set) and SC from the test set. Off-diagonal red blocks in both empirical and simulated FCD indicated recurring FC patterns that were not simply due to temporal autocorrelation. Similarity in the amount of off-diagonal red blocks between empirical and simulated FCD suggests that the pFIC model was able to generate realistic FCD. Across the 10 best candidate sets from the validation set, the Kolmogorov-Smirnov (KS) distance between empirical and simulated FCD was 0.18 ± 0.028 in the HCP test set. Disagreement between simulated and empirical fMRI appeared more pronounced in posterior regions, but the pattern of disagreement was not correlated with the RSFC gradient or the T1w/T2w ratio map (Figures S2 and S3).

Overall, the pFIC model was able to generate realistic FC and FCD, yielding an overall cost of 0.58 ± 0.018 in the HCP test set (Figure 2E). Parameterizing model parameters with only the principal FC gradient or only T1w/T2w ratio map led to worse (higher) cost in the HCP test set (Figure 2E). Most large-scale circuit model studies assume spatially homogeneous parameters. When local synaptic parameters (wEE, wEI and σ) were constrained to be uniform across brain regions (33,57) and optimized by CMA-ES, the cost was poor in the test set (Figure 2E). These results emphasize the importance of parameterizing local synaptic parameters with spatial gradients that smoothly varied from sensory-motor to association cortex. Consistent with our previous study (41), the T1w/T2w and FC gradient were complementary in the sense that combining the two spatial maps led to more realistic fMRI dynamics.

2.4. Estimated E/I ratio is sensitive to the effect of benzodiazepine alprazolam

In the previous section, we showed that the pFIC model could be effectively optimized to generate realistic fMRI dynamics. Here, we evaluated the biological plausibility of the estimated E/I ratio in a pharmacological-fMRI dataset (44) comprising 45 participants, who completed a placebo-controlled double-blind fMRI study with benzodiazepine alprazolam. Alprazolam is a benzodiazepine that enhances GABAergic signaling at GABAA receptor sub-units, including α1,2,3,5 and γ1−3 (58,59). Alprazolam enhances GABAergic inhibitory signaling via positive allosteric modulation, thus reducing E/I ratio (60). Therefore, we hypothesized that the E/I ratio estimated with the pFIC model would be lower during the alprazolam condition compared with the placebo condition.

The 45 participants were equally divided into training, validation and test sets. Group-level SC, first principal FC gradient, and T1w/T2w ratio maps from the HCP dataset were used in the following analysis. For each experimental condition (placebo or alprazolam), 250 candidate parameter sets were generated from the condition’s training set. The top 10 parameter sets from the validation set were evaluated in the test set. The costs of the 10 parameter sets generalized well to the test set (Figure S4), suggesting that there was no overfitting in the validation set.

One challenge in analyzing this dataset was that the fMRI data was acquired with a limited field of view (FOV). Therefore, 26 out of 68 Desikan-Killiany ROIs with less than 50% coverage (Figure S5) were not considered during the estimation of the model parameters. The estimated model parameters were extrapolated to the entire cortex (see Supplementary Methods S11 for details) and used to simulate the excitatory (SE) and inhibitory (SI) time courses (Figure 1B). Motivated by rodent studies, the E/I ratio was defined as the ratio between the temporal average of SE and SI (61).

An E/I ratio contrast was computed by subtracting the E/I ratio estimated during the drug (alprazolam) session from the E/I ratio estimated during the placebo session. Since alprazolam is expected to reduce E/I ratio, we hypothesized the E/I ratio would be lower during the alprazolam condition, yielding a positive E/I ratio contrast. Consistent with our hypothesis, the E/I ratio contrasts of all regions were positive (Figure 3B). 67 out of 68 regions exhibited E/I ratio contrasts statistically different from zero after correcting for multiple comparisons with a false discovery rate (FDR) of q < 0.05. We note that there was no motion difference between the drug and placebo fMRI sessions (p > 0.1). These results suggest that the E/I ratio estimated by the pFIC model was sensitive to the pharmacological enhancement of inhibitory activities.

Figure 3. E/I ratio estimate is sensitive to the effect of benzodiazepine alprazolam.

Figure 3.

Open in a new tab

(A) Seven resting-state networks (64). (B) Left: Regional E/I ratio contrast overlaid with the boundaries (black) of seven resting-state networks. 67 out of 68 regions showed significant E/I ratio difference between placebo and drug sessions after FDR correction (q < 0.05). E/I ratio difference was greater than zero for all regions, consistent with lower E/I ratio during the alprazolam session. Right: E/I ratio differences exhibited a spatial gradient with higher differences in sensory-motor regions compared with regions in the control and default networks. The boxes show the inter-quartile range (IQR) and the median. Whiskers indicate 1.5 IQR. Black crosses represent outliers. (C) Spatial distribution of BZR density (pmol/ml) from in-vivo positron emission tomography in a separate group of participants (62). (D) Higher regional BZR density was associated with larger E/I ratio changes during the drug session (r = 0.52, two-tail spin test p = 0.016).

Since the distribution of benzodiazepine receptors (BZR) density is not spatially uniform (62), we hypothesized that the E/I ratio contrast would also not be spatially uniform, and would align with BZR density. Supporting this, we found that the E/I ratio contrast exhibited a spatial gradient with strongest effects in sensory-motor networks and weakest effects in control and default networks (Figure 3B). Figure 3C shows the spatial distribution of benzodiazepine receptors (BZR) density estimated from in-vivo positron emission tomography in a separate group of participants (62). Regions with greater BZR density exhibited greater reduction in E/I ratio during the drug session (r = 0.52; two-tail spin test p = 0.016; Figure 3D). Therefore, the spatial distribution of E/I ratio contrast was biologically plausible.

To evaluate robustness, we repeated the above analyses 5 times with different random splits of the 45 participants into training, validation and test sets. The results were similar across the 5 splits (Figures S6 and S7). Results from the most representative split were shown in Figure 3. Using this most representative split, we performed several additional sensitivity analyses. In the previous analyses, the acceptable excitatory firing rate was constrained to be between 2.7Hz and 3.3Hz. Relaxing the thresholds to between 2.5Hz and 3.5Hz yielded similar results (Figure S8). Changing the ROI coverage threshold from 50% to 60% also yielded similar results (Figure S9). We repeated the analysis using a 100-region homotopic functional parcellation (63), which also yielded similar results (Figure S10). Pairwise comparisons between the control analyses are found in Figure S11. Similar results were obtained with log-transformation or square root of BZR density (Figure S12).

2.5. The E/I ratio declines with development in youth

Having demonstrated that the E/I ratio estimates were sensitive to the alprazolam-induced enhancement of inhibitory activities, we next explored how the E/I ratio changes during development in the Philadelphia Neurodevelopmental Cohort (PNC; 45,46). We hypothesized that the estimated E/I ratio would decline with age.

After data preprocessing and quality control, we obtained a sample of 885 participants ages 8–23 years old (Figure 4A). Participants were sorted according to age and evenly divided into 29 age groups, so each group comprised 30 or 31 participants. Within each age group, 15 participants were randomly selected as the validation set, while the remaining participants were assigned to the training set. For each age group, 250 candidate model parameter sets were generated from the group’s training set using CMA-ES and evaluated in the group’s validation set; the parameter set with the lowest validation cost was used to estimate regional E/I ratio across the cortex.

Figure 4. E/I ratio continuously declines throughout child and adolescent development.

Figure 4.

Open in a new tab

(A) Age distribution of 885 PNC participants (mean = 15.66, std = 3.36, min = 8.17, max = 23). (B) Participants in older age groups exhibited lower E/I ratio (r = −0.51, p = 0.004). Participants were divided into 29 non-overlapping age groups. There are 29 dots in the scatter plot, corresponding to the 29 age groups. The shaded area depicts 95% confidence interval of the linear relationship. (C) Spatial distribution of linear regression slope between regional E/I ratio and age. The values represent the rate of E/I ratio changes during development. All slopes were negative and significant (FDR q < 0.05). (D) The slopes exhibited a spatial gradient with sensory-motor networks showing the fastest E/I ratio reduction and association networks showing slower E/I ratio reduction. The boxes show the inter-quartile range (IQR) and the median. Whiskers indicate 1.5 IQR. Black crosses represent outliers.

We performed linear regression between age and mean cortical E/I ratio (i.e., E/I ratio averaged across the whole cortex), as well as between age and regional E/I ratio. Mean cortical E/I ratio declined throughout child and adolescent development (r = −0.51, p = 0.004; Figure 4B). This E/I ratio reduction was statistically significant for all cortical regions (FDR q < 0.05; Figure 4C). Furthermore, the rate of E/I ratio decrease exhibited a spatial gradient with sensory-motor regions exhibiting greater rate of E/I ratio decrease (i.e., more negative slope) compared with association networks (Figure 4D).

To evaluate the robustness of these effects, the PNC analyses were repeated 5 times with different splits of the participants (within each age group) into training and validation sets. The results were similar across the 5 random splits of the data (Figures S13 and S14). We conducted several additional sensitivity analyses using the most representitve split (which was shown in Figure 4). Relaxing the firing rate thresholds to between 2.5Hz and 3.5Hz yielded similar results (Figure S15), as did using a 100-region homotopic parcellation (63) (Figure S16). Pairwise comparisons between the control analyses are found in Figure S17. Finally, consistent with the literature, younger participants exhibited higher head motion during the fMRI scan. Therefore, as a control analysis, we regressed out mean framewise displacement from the E/I ratio estimates of each age group, yielding similar results (Figure S18).

2.6. Lower E/I ratio is associated with better cognition within the same age group

Having shown that older children exhibited lower E/I ratio (Figure 3B), we next evaluated the cognitive implications of such a decline in the E/I ratio as part of normative development. We hypothesize that a lower E/I ratio would be associated with better cognition. To test this hypothesis, we compared the E/I ratio of PNC participants who were matched on age but differed in cognitive performance.

Participants in the PNC completed the Penn Computerized Neurocognitive Battery (CNB), a 12-task battery that has been previously summarized using an overall (domain-general) measure of accuracy as well as three domain-specific factor scores (65). Participants were divided into 14 high-performance groups and 14 low-performance groups based on the overall accuracy measure. Each high-performance group was age-matched to a low-performance group (Figure 5A). Each low-performance or high-performance group comprised 31 or 32 participants. For each group, 15 participants were randomly assigned to the validation set, while the remaining participants were assigned to the training set. For each group, 250 candidate parameter sets were generated from the training set and the top parameter set from the validation set was used to estimate the E/I ratio; we compared the E/I ratio between the high and low performance groups.

Figure 5. Lower E/I ratio is associated with better cognitive performance within the same age group in the PNC cohort.

Figure 5.

Open in a new tab

(A) Boxplots of age, (B) ‘overall accuracy’ and (C) mean cortical E/I ratio of high-performance and low-performance (overall accuracy) groups. The mean cortical E/I ratio of the high-performance group was significantly lower than that of the low-performance group (FDR q < 0.05). (D) Spatial distribution of effect size of regional E/I ratio difference between high-performance and low-performance groups. All regions were significant after FDR correction with q < 0.05. (E) Effect size of E/I ratio differences in cognition is larger in control and default networks compared with sensory-motor regions. The boxes show the inter-quartile range (IQR) and the median. Whiskers indicate 1.5 IQR. Black crosses represent outliers. (F) ROI rankings based on the sensorimotor-association (S-A) axis (66). Lower ranks were assigned to ROIs that were more towards the sensorimotor pole; higher ranks were assigned to ROIs that were more towards the association pole. (G) Agreement between effect size of E/I ratio difference and S-A axis rank. Spearman’s correlation r = 0.87, two-tailed spin test p < 0.001.

The high-performance group exhibited lower mean cortical E/I ratio than the low-performance group (two-tailed t-test p = 1.2 × 10−3; Figure 5C). There was no motion difference between high-performance and low-performance groups (p > 0.2). To test for domain specificity, we also compared the E/I ratio for the three domain-specifc factor scores (complex reasoning, memory and social cognition), but observed no statistical difference after correcting for multiple comparisons (Figure S19).

Having found global differences in the E/I ratio between the high and low cognitive performance groups, we next evaluated regional effects (Figure 5D). We found that E/I ratio differences between low-performance and high-performance groups were larger in control and default networks, compared with sensory-motor regions (Figure 5E; all FDR q <0.05). Notably, the effect sizes of these regional differences in the E/I ratio aligned well with the sensorimotor-association (S-A) axis of cortical organization (66), such that effect sizes were lowest at the sensorimotor pole and largest at the association pole (Figure 5F). Spearman’s correlation between effect sizes and S-A axis ranks was r = 0.87 (two-tailed spin test p < 0.001; Figure 5G). Overall, these results suggest that a more mature E/I ratio – especially in higher-order association cortex – is linked to more mature cognition.

To evaluate the robustness of these results, we repeated these analyses 5 times with different random training-validation splits of participants within each high-performance group and each low-performance group. The results were similar across the 5 splits (Figures S20 to S24; the most representative split is displayed in Figure 5). Within the most representative split, we found that relaxing the thresholds to between 2.5Hz and 3.5Hz yielded similar results (Figure S25) as did use of a 100-region homotopic functional parcellation (63) (Figure S26). Pairwise comparisons between the control analyses are found in Figure S27.

2.7. Results generalize to a younger Asian cohort

As a final step, we evaluated whether the link between the E/I ratio and cognition generalized to a group of younger participants of different anscestry. This was motivated by recent concerns that relationships between resting-fMRI and behavior may not generalize well across ethnic groups (67). We utilized the Growing Up in Singapore with Healthy Outcomes (GUSTO) dataset (47), which included 154 participants (after quality control) with a mean age of 7.5 years. An overall cognitive performance score was obtained by a principal component analysis of five cognitive tests. Participants were then divided into groups with high and low cognitive performance. The ages were well-matched between the high and low-performance groups (Figures 6A and 6B). There was no motion difference between the high and low-performance groups during the fMRI scans (p > 0.1).

Figure 6. Lower E/I ratio is associated with better cognitive performance within the same age group in the GUSTO dataset.

Figure 6.

Open in a new tab

(A) Boxplots of age for high-performance and low-performance groups. (B) Overall cognition of high-performance and low-performance groups. (C) Spatial distribution of effect size of E/I ratio difference between low-performance group and high-performance group. (D) Effect size of E/I ratio differences is larger in control and default networks compared with sensory-motor regions. The boxes show the inter-quartile range (IQR) and the median. Whiskers indicate 1.5 IQR. Black crosses represent outliers. (E) Agreement between effect size of E/I ratio difference and S-A axis rank. Spearman’s correlation r = 0.56, two-tailed spin test p = 0.01.

Replicating PNC results, we found that the high-performance group exhibited a lower E/I ratio in higher-order association cortex than the low-performance group (Figures 6C). Differences were largest in the default and control networks (Figure 6D). Statistical significance was evaluated using a permutation test, where the null distribution was constructed by randomly assigning participants into high or low-performance groups, and then re-estimating the E/I ratio. We note that only 29 (out of 68) regions were significant after FDR correction with q < 0.05. These 29 regions were all in association cortex. By contrast, differences in the E/I ratio between cognitive performance groups was largely not significant in sensory-motor networks. As in the PNC, we found the effect sizes of these cognitive differences aligned with the S-A axis (r = 0.56, two-tailed spin test p = 0.01; Figure 6E). These results in a younger Asian cohort emphasize the robustness and generalizability of our findings.

3. Discussion

We first established that the pFIC model could generate realistic fMRI dynamics in a large adult dataset. We then demonstrated that our E/I ratio marker was sensitive to increased inhibitory activity induced by benzodiazepine alprazolam. In a large developmental sample from North America, we found that the E/I ratio marker decreased with age. We also demonstrated that a lower E/I ratio marker – reflective of more mature cortex – was associated with better cognitive performance, particularly in transmodal association cortex. Critically, these findings generalized to a younger Asian cohort. Together, our findings provide evidence that refinements in the cortical E/I ratio persist into adolescence, suggesting that prolonged E/I-linked developmental plasticity in the association cortex supports continued neurocognitive development. We speculate that insufficient refinement of the E/I ratio during development may create a vulnerability to cognitive deficits, with potentially important implications for transdiagnostic psychopathology.

The E/I ratio is challenging to be non-invasively investigated in humans. Post-mortem studies have established how the expressions of E/I relevant genes vary across the cortex (35,68,69). On the other hand, there is a lack of a direct mapping of in-vivo neuroimaging signals with excitatory and inhibitory neurobiology, as well as constrained spatial coverage and specificity of available E/I techniques (24–26). Here, we capitalized on recent developments in biologically interpretable computational modeling of cortical circuits to gain insight into the E/I ratio from fMRI data. We fitted a large-scale circuit model with interacting excitatory and inhibitory populations to resting-state fMRI and calculated the E/I ratio from the time courses of excitatory and inhibitory synaptic gating variables. Our E/I ratio marker captured reductions in the E/I ratio induced by alprazolam, a positive allosteric modulator that increases the effectiveness of GABAergic signaling (70). Furthermore, the spatial pattern of benzodiazepine-related E/I reductions described by the model was correlated with the distribution of benzodiazepine-sensitive GABA receptors from positron emission tomography (62). Interestingly, one of the pharmacological targets of benzodiazepines is GABAA α1 receptors (70). Increases in GABA signaling at GABAA α1 receptors have been shown to trigger the onset of developmental critical periods in animal models (71), indicating that the pFIC model is well equipped to study development-linked changes in inhibitory signaling in the human brain.

We found that the E/I ratio decreased across the cortex throughout child and adolescent development. E/I ratio declined with age across all cortical systems, but the magnitude of decline varied along a unimodal-transmodal cortical hierarchy. Specifically, by age 22, the E/I ratio had declined the most in unimodal sensory territories, such as the visual and somatomotor systems, and the least in transmodal systems like the default and frontoparietal control systems. Differences in E/I development across these systems may be linked to differences in their maturational time courses. The development of cortical inhibitory circuitry is well-established as a central mechanism that controls the timing and progression of critical periods of development (21,72). Initially, the development of inhibitory circuitry lags behind that of excitatory pyramidal cells, leading to an early increase in the local E/I ratio (73). Later, experience and evoked activity stimulate the development of inhibitory circuitry (21) — particularly fast-spiking parvalbumin positive interneurons and GABAA α1 receptors — which begins to reduce the E/I ratio, facilitating experience-dependent plasticity and triggering the opening of the critical period window (22,71,72). As the critical period progresses, excitatory synapses are pruned, further reducing the E/I ratio (74,75). Finally, as inhibitory circuitry reaches maturity, a new set of plasticity braking factors are triggered, including the formation of intracortical myelin and perineuronal nets, which stabilize cortical circuits and close the critical period window (6,76). Consequently, an initial decrease in the E/I ratio can signify that a critical period has been triggered and the cortex is in a relatively immature, plasticity-permissive state. As the E/I ratio reduces further, it may signify that the cortex has reached a mature, plasticity-restrictive state, with pruned excitatory synapses, fully-developed inhibitory circuitry, and mature plasticity brakes that have closed the critical period (6). As such, the greater reduction in the E/I ratio we observe in sensory systems relative to association systems may reflect that sensory systems have reached a higher degree of maturity by the end of the adolescent years while association systems remain in a more immature, plasticity-permissive state. To test this hypothesis, future work could use multimodal neuroimaging that combines our pFIC approach with other markers of critical period closure - such as intracortical myelination - to evaluate biologically-relevent signatures of when windows of critical period plasticity open and close during youth.

Our findings align with a wealth of literature demonstrating differences in the development of sensorimotor and association cortices. Studies have shown that functional connectivity, functional topography, structure-function coupling, and intrinsic dynamics follow different developmental trajectories between sensory and associative cortical systems (16,77–79). For example, while the intrinsic fluctuation amplitude of sensory systems linearly decline with age, association systems follow curvilinear developmental trajectories that peak over adolescence before declining into adulthood (80). Importantly, other recent work has shown that the development of intracortical myelin, which functions as a brake on plasticity, also varies along the sensory-to-association axis (11,81). Specifically, the period of peak growth in intracortical myelin occurs during childhood in sensorimotor cortex, yet not until adolescence in association cortex. Coupled with our current findings of a greater reduction in the E/I ratio of sensory systems (versus a weaker reduction in association systems), this work jointly indicates that sensorimotor systems are more mature by the onset of adolescence whereas association cortex may remain more plastic during the adolescent period. This interpretation aligns with a recent study showing that an fMRI marker of functional plasticity peaks during early adolescence in association cortex but continuously declines throughout childhood and adolescence in sensorimotor cortex (80).

The protracted development of the E/I ratio throughout adolescence may facilitate healthy cognitive development. We found that better cognitive ability was associated with a lower E/I ratio across the cortex in groups of age-matched youth. Since the E/I ratio normatively decreased with age, this effect may indicate that more mature cognitive performance is associated with a more mature cortical E/I ratio. As such, the E/I ratio may capture aspects of development independent of chronological age. Importantly, the magnitude of the effect was not spatially uniform. The greatest effect sizes were observed in association cortex, while the weakest effect sizes were observed in sensory cortex. This pattern is consistent with prior work showing that functional properties of the association cortex are most strongly related to cognitive performance across development (79,82). Our findings also support theoretical predictions from a biophysically-based cortical circuit model of decision-making that a balanced E/I ratio supports optimal decision-making (9). Together, our results suggest that although E/I ratio continues to develop throughout the cortex during adolescence, the development of the E/I ratio in association cortex is particularly relevant to maturing cognition. Critically, we generalized associations between the E/I ratio and cognitive ability in an independent sample of youth collected from a different continent, demonstrating the robustness of these effects across both populations and recruitment sites.

Our findings have important implications for understanding the emergence of psychopathology during adolescence. Though a prolonged period of developmental plasticity in the association cortex may be essential to healthy cognitive development, it may also represent a period of vulnerability to atypical developmental outcomes. A growing body of work has begun to implicate a disrupted E/I ratio in prefrontal cortex as a central mechanism in neuropsychiatric disorders such as depression and psychosis spectrum disorders (83–85). These conditions are thought to involve an atypically high E/I ratio in prefrontal cortex (86–89). Future studies can use our model to understand how atypical development of E/I ratio in association cortex may lead to transdiagnostic cognitive dysfunction in developmental psychopathology.

Limitations and future work

Parameterization of local circuit parameters with the T1w/T2w ratio map and the FC gradient yielded more realistic fMRI dynamics than either gradient alone or if local circuit parameters were constrained to be spatially uniform. Future studies can explore more generic parameterizations, such as geometric eigenmodes (90).

The current study utilized parcellations with only 68 or 100 regions. Simulating the FIC model with a higher spatial resolution is computationally challenging because the number of inter-regional connections increases quadratically with the number of regions. Future work can explore more efficient algorithms. Furthermore, our analyses were limited to linear modeling of E/I ratio across a set of age bins. Future work in larger samples may facilitate the estimation of nonlinear developmental trajectories of E/I ratio.

Finally, our approach can also be used to study E/I ratio changes during cognitive tasks or during a naturalistic paradigm. When applying the pFIC model to a new dataset, dataset-specific SC, T1w/T2w ratio map and FC gradient can be used, although that might not be necessary. For example, SC, T1w/T2w ratio map and whole-cortex FC gradient were not available in the alprazolam dataset, so we utilized SC, T1w/T2w and whole-cortex FC gradient from the HCP dataset.

Conclusion

Our results underscore the utility of large-scale circuit models to provide insights into the mechanisms driving neurocognitive development. We find that an essential aspect of healthy brain function—the cortical E/I ratio—is refined during childhood and adolescence. We also provide new evidence that this hallmark critical period mechanism is associated with improved cognitive ability. Our findings pave the way for future work to investigate how disrupted E/I balance may lead to cognitive dysfunction in psychopathology that emerges during youth and is characterized by atypical development of association cortex that undergoes protracted maturation.

4. Methods

We utilized the HCP S1200 release (N = 1004; Figure 2), pharmacological (benzodiazepine alprazolam) fMRI dataset (N = 45; Figure 3), PNC dataset (N = 885; Figures 4 and 5) and the GUSTO cohort (N = 154; Figure 6). In the case of HCP, we used the publicly available ICA-FIX MSMAll resting-state fMRI data in fsLR surface space. For alprazolam and PNC datasets, we used preprocessed fMRI data from our previous study, which involved slice time correction, motion correction, field distortion correction and anatomical CompCor (26). In the case of the alprazolam dataset, no resting-state fMRI was available, so we used task-fMRI after regressing out the task regressors, following our previous study (26). To be consistent, the GUSTO dataset was also preprocessed in a similar fashion as the PNC dataset. More details can be found in the Supplementary Methods.

After preprocessing, static FC was computed using Pearson’s correlation for all datasets. FCD was computed using sliding window length of ~60 seconds, corresponding to windows of length 83, 20, 20 and 23 for the HCP, alprazolam, PNC and GUSTO datasets respectively. The window length followed best practice recommendations from previous studies (51,91). SC was computed based on the number of streamlines generated with probabilistic tractography using MRtrix3 (92). More details can be found in the Supplementary Methods.

The pFIC model (33) was fitted to the different datasets using the covariance matrix adaptation evolution strategy (CMA-ES) (56). The fitted pFIC model was used to simulate the synaptic gating variable time courses SE and SI of the excitatory and inhibitory populations respectively. The E/I ratio was defined as the ratio between the temporal average of SE and SI. More details can be found in the Supplementary Methods.

The HCP data is publicly available (https://www.humanconnectome.org/). The GUSTO dataset can be obtained via a data transfer agreement (https://www.gusto.sg/). The PNC dataset is publicly available in the Database of Genotypes and Phenotypes (dbGaP accession phs000607.v3.p2). All pharmacological imaging data necessary to evaluate the conclusions in the paper are available here (https://github.com/YeoPrivateLab/CBIG_private/tree/develop/stable_projects/fMRI_dynamics/Zhang2024_pFIC/replication/Alprazolam). Code for this study can be found here (https://github.com/ThomasYeoLab/CBIG/tree/master/stable_projects/fMRI_dynamics/Zhang2024_pFIC). Co-authors (TZ and LA) reviewed the code before merging into the GitHub repository to reduce the chance of coding errors.

Supplementary Material

1

Significance.

Healthy brain function requires a delicate balance of neural excitation (E) and inhibition (I). In animals, this balance – the E/I ratio – is known to decrease with the maturation of inhibitory circuitry during healthy development. However, in humans, the normative development of cortex-wide E/I ratio remains unclear. Here, we use a biophysical model and non-invasive brain scans to estimate a marker of E/I ratio. Spatial changes in our E/I ratio marker are consistent with a drug that decreases E/I ratio. We also find that our cortex-wide E/I ratio marker decreases during development. Furthermore, North American and Asian children with lower E/I ratio, especially in higher-order cortex, have better cognitive performance. Overall, E/I ratio is a potential index of healthy neurocognitive development.

Acknowledgements

Our research is supported by the NUS Yong Loo Lin School of Medicine (NUHSRO/2020/124/TMR/LOA), the Singapore National Medical Research Council (NMRC) LCG (OFLCG19May-0035), NMRC CTG-IIT (CTGIIT23jan-0001), NMRC STaR (STaR20nov-0003), Singapore Ministry of Health (MOH) Centre Grant (CG21APR1009), the Temasek Foundation (TF2223-IMH-01), and the United States National Institutes of Health (R01MH120080 & R01MH133334). Our computational work was partially performed on resources of the National Supercomputing Centre, Singapore (https://www.nscc.sg). Any opinions, findings and conclusions or recommendations expressed in this material are those of the authors and do not reflect the views of the Singapore NRF, NMRC, MOH or Temasek Foundation. In RIE2025, the GUSTO dataset was supported by funding from the NRF’s Human Health and Potential (HHP) Domain, under the Human Potential Programme. The alprazolam sample was funded by AstraZeneca Pharmaceuticals LP. DHW was also supported by NARSAD and the Sidney R. Baer, Jr. Foundation. BL was funded by NIH K99MH127293 and T32MH014654. VJS was supported by a National Science Foundation Graduate Research Fellowship (DGE-1845298). The PNC was funded via RC2 grants from the National Institute of Mental Health: MH089983 and MH089924. Additional support was provided by R01MH113550, R01MH120482, R01MH112847, R01EB022573, RF1MH116920, RF1MH121867, R37MH125829, T32MH019112, the AE Foundation, and the Penn-CHOP Lifespan Brain Institute. Data were in part provided by the Human Connectome Project, WU-Minn Consortium (Principal Investigators: David Van Essen and Kamil Ugurbil; 1U54MH091657) funded by the 16 NIH Institutes and Centers that support the NIH Blueprint for Neuroscience Research; and by the McDonnell Center for Systems Neuroscience at Washington University.

References

  • 1.Nelson S. Temporal interactions in the cat visual system. III. Pharmacological studies of cortical suppression suggest a presynaptic mechanism. J. Neurosci. 11, 369–380 (1991). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Atallah B. V. & Scanziani M. Instantaneous Modulation of Gamma Oscillation Frequency by Balancing Excitation with Inhibition. Neuron 62, 566–577 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Dorrn A. L., Yuan K., Barker A. J., Schreiner C. E. & Froemke R. C. Developmental sensory experience balances cortical excitation and inhibition. Nature 465, 932–936 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Sun Y. J. et al. Fine-tuning of pre-balanced excitation and inhibition during auditory cortical development. Nature 465, 927–931 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Turrigiano G. G. & Nelson S. B. Homeostatic plasticity in the developing nervous system. Nat Rev Neurosci 5, 97–107 (2004). [DOI] [PubMed] [Google Scholar]
  • 6.Takesian A. E. & Hensch T. K. Balancing Plasticity/Stability Across Brain Development. in Progress in Brain Research vol. 207 3–34 (Elsevier, 2013). [DOI] [PubMed] [Google Scholar]
  • 7.Rubenstein J. L. R. & Merzenich M. M. Model of autism: increased ratio of excitation/inhibition in key neural systems. Genes, Brain and Behavior 2, 255–267 (2003). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Bruining H. et al. Measurement of excitation-inhibition ratio in autism spectrum disorder using critical brain dynamics. Sci Rep 10, 9195 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Lam N. H. et al. Effects of Altered Excitation-Inhibition Balance on Decision Making in a Cortical Circuit Model. J. Neurosci. 42, 1035–1053 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Manyukhina V. O. et al. Globally elevated excitation–inhibition ratio in children with autism spectrum disorder and below-average intelligence. Molecular Autism 13, 20 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Paquola C. et al. Shifts in myeloarchitecture characterise adolescent development of cortical gradients. eLife 8, e50482 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Bethlehem R. A. I. et al. Brain charts for the human lifespan. Nature 604, 525–533 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Larsen B. & Luna B. Adolescence as a neurobiological critical period for the development of higher-order cognition. Neuroscience & Biobehavioral Reviews 94, 179–195 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Reh R. K. et al. Critical period regulation across multiple timescales. Proc. Natl. Acad. Sci. U.S.A. 117, 23242–23251 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Larsen B., Sydnor V. J., Keller A. S., Yeo B. T. T. & Satterthwaite T. D. A critical period plasticity framework for the sensorimotor–association axis of cortical neurodevelopment. Trends in Neurosciences 46, 847–862 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Sydnor V. J. & Satterthwaite T. D. Neuroimaging of plasticity mechanisms in the human brain: from critical periods to psychiatric conditions. Neuropsychopharmacol. 48, 219–220 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Pouille F. & Scanziani M. Enforcement of Temporal Fidelity in Pyramidal Cells by Somatic Feed-Forward Inhibition. Science 293, 1159–1163 (2001). [DOI] [PubMed] [Google Scholar]
  • 18.Iwai Y., Fagiolini M., Obata K. & Hensch T. K. Rapid Critical Period Induction by Tonic Inhibition in Visual Cortex. J. Neurosci. 23, 6695–6702 (2003). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Gu Y. et al. Neuregulin-Dependent Regulation of Fast-Spiking Interneuron Excitability Controls the Timing of the Critical Period. J. Neurosci. 36, 10285–10295 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Wu Y. K., Miehl C. & Gjorgjieva J. Regulation of circuit organization and function through inhibitory synaptic plasticity. Trends in Neurosciences 45, 884–898 (2022). [DOI] [PubMed] [Google Scholar]
  • 21.Hensch T. K. Critical period plasticity in local cortical circuits. Nat Rev Neurosci 6, 877–888 (2005). [DOI] [PubMed] [Google Scholar]
  • 22.Hensch T. K. & Fagiolini M. Excitatory–inhibitory balance and critical period plasticity in developing visual cortex. in Progress in Brain Research vol. 147 115–124 (Elsevier, 2005). [DOI] [PubMed] [Google Scholar]
  • 23.Stern E. A., Maravall M. & Svoboda K. Rapid Development and Plasticity of Layer 2/3 Maps in Rat Barrel Cortex In Vivo. Neuron 31, 305–315 (2001). [DOI] [PubMed] [Google Scholar]
  • 24.Silveri M. M. et al. Frontal Lobe γ-Aminobutyric Acid Levels During Adolescence: Associations with Impulsivity and Response Inhibition. Biological Psychiatry 74, 296–304 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Ghisleni C. et al. Subcortical Glutamate Mediates the Reduction of Short-Range Functional Connectivity with Age in a Developmental Cohort. Journal of Neuroscience 35, 8433–8441 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Larsen B. et al. A developmental reduction of the excitation:inhibition ratio in association cortex during adolescence. Sci. Adv. 8, eabj8750 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Trakoshis S. et al. Intrinsic excitation-inhibition imbalance affects medial prefrontal cortex differently in autistic men versus women. eLife 9, e55684 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Ranasinghe K. G. et al. Altered excitatory and inhibitory neuronal subpopulation parameters are distinctly associated with tau and amyloid in Alzheimer’s disease. eLife 11, e77850 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Fortel I. et al. Disrupted Excitation-Inhibition Balance in Cognitively Normal Individuals at Risk of Alzheimer’s Disease. JAD 95, 1449–1467 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Honey C. J., Kötter R., Breakspear M. & Sporns O. Network structure of cerebral cortex shapes functional connectivity on multiple time scales. Proc. Natl. Acad. Sci. U.S.A. 104, 10240–10245 (2007). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Ghosh A., Rho Y., McIntosh A. R., Kötter R. & Jirsa V. K. Noise during Rest Enables the Exploration of the Brain’s Dynamic Repertoire. PLoS Comput Biol 4, e1000196 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Zalesky A., Fornito A., Cocchi L., Gollo L. L. & Breakspear M. Time-resolved resting-state brain networks. Proc. Natl. Acad. Sci. U.S.A. 111, 10341–10346 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Deco G. et al. How Local Excitation-Inhibition Ratio Impacts the Whole Brain Dynamics. Journal of Neuroscience 34, 7886–7898 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Kim Y. et al. Brain-wide Maps Reveal Stereotyped Cell-Type-Based Cortical Architecture and Subcortical Sexual Dimorphism. Cell 171, 456–469.e22 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Burt J. B. et al. Hierarchy of transcriptomic specialization across human cortex captured by structural neuroimaging topography. Nat Neurosci 21, 1251–1259 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Anderson K. M. et al. Convergent molecular, cellular, and cortical neuroimaging signatures of major depressive disorder. Proc. Natl. Acad. Sci. U.S.A. 117, 25138–25149 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Cardin J. A. et al. Driving fast-spiking cells induces gamma rhythm and controls sensory responses. Nature 459, 663–667 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Deco G. et al. Dynamical consequences of regional heterogeneity in the brain’s transcriptional landscape. Sci. Adv. 7, eabf4752 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Demirtaş M. et al. Hierarchical Heterogeneity across Human Cortex Shapes Large-Scale Neural Dynamics. Neuron 101, 1181–1194.e13 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Wang P. et al. Inversion of a large-scale circuit model reveals a cortical hierarchy in the dynamic resting human brain. Sci. Adv. 5, eaat7854 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Kong X. et al. Sensory-motor cortices shape functional connectivity dynamics in the human brain. Nat Commun 12, 6373 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Deco G. et al. Resting-State Functional Connectivity Emerges from Structurally and Dynamically Shaped Slow Linear Fluctuations. Journal of Neuroscience 33, 11239–11252 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Van Essen D. C. et al. The WU-Minn Human Connectome Project: An overview. NeuroImage 80, 62–79 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Wolf D. H. et al. Amygdala abnormalities in first-degree relatives of individuals with schizophrenia unmasked by benzodiazepine challenge. Psychopharmacology 218, 503–512 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Satterthwaite T. D. et al. Neuroimaging of the Philadelphia Neurodevelopmental Cohort. NeuroImage 86, 544–553 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Calkins M. E. et al. The Philadelphia Neurodevelopmental Cohort: constructing a deep phenotyping collaborative. J Child Psychol Psychiatr 56, 1356–1369 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Soh S.-E. et al. Cohort Profile: Growing Up in Singapore Towards healthy Outcomes (GUSTO) birth cohort study. International Journal of Epidemiology 43, 1401–1409 (2014). [DOI] [PubMed] [Google Scholar]
  • 48.Glasser M. F. et al. The minimal preprocessing pipelines for the Human Connectome Project. NeuroImage 80, 105–124 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49.Desikan R. S. et al. An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. NeuroImage 31, 968–980 (2006). [DOI] [PubMed] [Google Scholar]
  • 50.Hansen E. C. A., Battaglia D., Spiegler A., Deco G. & Jirsa V. K. Functional connectivity dynamics: Modeling the switching behavior of the resting state. NeuroImage 105, 525–535 (2015). [DOI] [PubMed] [Google Scholar]
  • 51.Liégeois R., Laumann T. O., Snyder A. Z., Zhou J. & Yeo B. T. T. Interpreting temporal fluctuations in resting-state functional connectivity MRI. NeuroImage 163, 437–455 (2017). [DOI] [PubMed] [Google Scholar]
  • 52.Brunel N. & Wang X.-J. Effects of neuromodulation in a cortical network model of object working memory dominated by recurrent inhibition. Journal of Computational Neuroscience 11, 63–85 (2001). [DOI] [PubMed] [Google Scholar]
  • 53.Wong K.-F. & Wang X.-J. A Recurrent Network Mechanism of Time Integration in Perceptual Decisions. J. Neurosci. 26, 1314–1328 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54.Margulies D. S. et al. Situating the default-mode network along a principal gradient of macroscale cortical organization. Proc. Natl. Acad. Sci. U.S.A. 113, 12574–12579 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55.Glasser M. F. & Van Essen D. C. Mapping Human Cortical Areas In Vivo Based on Myelin Content as Revealed by T1- and T2-Weighted MRI. J. Neurosci. 31, 11597–11616 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56.Hansen N. The CMA Evolution Strategy: A Comparing Review. in Towards a New Evolutionary Computation (eds. Lozano J. A., Larrañaga P., Inza I. & Bengoetxea E.) vol. 192 75–102 (Springer Berlin Heidelberg, Berlin, Heidelberg, 2006). [Google Scholar]
  • 57.Kringelbach M. L. & Deco G. Brain States and Transitions: Insights from Computational Neuroscience. Cell Reports 32, 108128 (2020). [DOI] [PubMed] [Google Scholar]
  • 58.Sigel E. Mapping of the Benzodiazepine Recognition Site on GABAA Receptors. CTMC 2, 833–839 (2002). [DOI] [PubMed] [Google Scholar]
  • 59.Olsen R. W. & Sieghart W. GABAA receptors: Subtypes provide diversity of function and pharmacology. Neuropharmacology 56, 141–148 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 60.Verster J. C. & Volkerts E. R. Clinical Pharmacology, Clinical Efficacy, and Behavioral Toxicity of Alprazolam: A Review of the Literature. CNS Drug Reviews 10, 45–76 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 61.Xue M., Atallah B. V. & Scanziani M. Equalizing excitation–inhibition ratios across visual cortical neurons. Nature 511, 596–600 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 62.Nørgaard M. et al. A high-resolution in vivo atlas of the human brain’s benzodiazepine binding site of GABAA receptors. NeuroImage 232, 117878 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 63.Yan X. et al. Homotopic local-global parcellation of the human cerebral cortex from resting-state functional connectivity. NeuroImage 273, 120010 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 64.Yeo B. T. et al. The organization of the human cerebral cortex estimated by intrinsic functional connectivity. Journal of Neurophysiology 106, 1125–1165 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 65.Moore T. M., Reise S. P., Gur R. E., Hakonarson H. & Gur R. C. Psychometric properties of the Penn Computerized Neurocognitive Battery. Neuropsychology 29, 235–246 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 66.Sydnor V. J. et al. Neurodevelopment of the association cortices: Patterns, mechanisms, and implications for psychopathology. Neuron 109, 2820–2846 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 67.Li J. et al. Cross-ethnicity/race generalization failure of behavioral prediction from resting-state functional connectivity. Sci. Adv. 8, eabj1812 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 68.Gandal M. J. et al. Transcriptome-wide isoform-level dysregulation in ASD, schizophrenia, and bipolar disorder. Science 362, eaat8127 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 69.Goulas A. et al. The natural axis of transmitter receptor distribution in the human cerebral cortex. Proc. Natl. Acad. Sci. U.S.A. 118, e2020574118 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 70.Rudolph U. & Knoflach F. Beyond classical benzodiazepines: novel therapeutic potential of GABAA receptor subtypes. Nat Rev Drug Discov 10, 685–697 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 71.Katagiri H., Fagiolini M. & Hensch T. K. Optimization of Somatic Inhibition at Critical Period Onset in Mouse Visual Cortex. Neuron 53, 805–812 (2007). [DOI] [PubMed] [Google Scholar]
  • 72.Toyoizumi T. et al. A Theory of the Transition to Critical Period Plasticity: Inhibition Selectively Suppresses Spontaneous Activity. Neuron 80, 51–63 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 73.Lodato S. et al. Excitatory Projection Neuron Subtypes Control the Distribution of Local Inhibitory Interneurons in the Cerebral Cortex. Neuron 69, 763–779 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 74.Han K.-S., Cooke S. F. & Xu W. Experience-Dependent Equilibration of AMPAR-Mediated Synaptic Transmission during the Critical Period. Cell Reports 18, 892–904 (2017). [DOI] [PubMed] [Google Scholar]
  • 75.Xu W., Löwel S. & Schlüter O. M. Silent Synapse-Based Mechanisms of Critical Period Plasticity. Front. Cell. Neurosci. 14, 213 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 76.Willis A., Pratt J. A. & Morris B. J. Enzymatic Degradation of Cortical Perineuronal Nets Reverses GABAergic Interneuron Maturation. Mol Neurobiol 59, 2874–2893 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 77.Baum G. L. et al. Development of structure–function coupling in human brain networks during youth. Proc. Natl. Acad. Sci. U.S.A. 117, 771–778 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 78.Dong H.-M., Margulies D. S., Zuo X.-N. & Holmes A. J. Shifting gradients of macroscale cortical organization mark the transition from childhood to adolescence. Proc. Natl. Acad. Sci. U.S.A. 118, e2024448118 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 79.Pines A. R. et al. Dissociable multi-scale patterns of development in personalized brain networks. Nat Commun 13, 2647 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 80.Sydnor V. J. et al. Intrinsic activity development unfolds along a sensorimotor–association cortical axis in youth. Nat Neurosci 26, 638–649 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 81.Baum G. L. et al. Graded Variation in T1w/T2w Ratio during Adolescence: Measurement, Caveats, and Implications for Development of Cortical Myelin. J. Neurosci. 42, 5681–5694 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 82.Keller A. S. et al. Hierarchical functional system development supports executive function. Trends in Cognitive Sciences 27, 160–174 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 83.Anticevic A. & Murray J. D. Rebalancing Altered Computations: Considering the Role of Neural Excitation and Inhibition Balance Across the Psychiatric Spectrum. Biological Psychiatry 81, 816–817 (2017). [DOI] [PubMed] [Google Scholar]
  • 84.Sohal V. S. & Rubenstein J. L. R. Excitation-inhibition balance as a framework for investigating mechanisms in neuropsychiatric disorders. Mol Psychiatry 24, 1248–1257 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 85.Vinogradov S., Chafee M. V., Lee E. & Morishita H. Psychosis spectrum illnesses as disorders of prefrontal critical period plasticity. Neuropsychopharmacol. 48, 168–185 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 86.Foss-Feig J. H. et al. Searching for Cross-Diagnostic Convergence: Neural Mechanisms Governing Excitation and Inhibition Balance in Schizophrenia and Autism Spectrum Disorders. Biological Psychiatry 81, 848–861 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 87.Duman R. S., Sanacora G. & Krystal J. H. Altered Connectivity in Depression: GABA and Glutamate Neurotransmitter Deficits and Reversal by Novel Treatments. Neuron 102, 75–90 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 88.Gao W.-J., Yang S.-S., Mack N. R. & Chamberlin L. A. Aberrant maturation and connectivity of prefrontal cortex in schizophrenia—contribution of NMDA receptor development and hypofunction. Mol Psychiatry 27, 731–743 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 89.Yao H. K. et al. Reduced inhibition in depression impairs stimulus processing in human cortical microcircuits. Cell Reports 38, 110232 (2022). [DOI] [PubMed] [Google Scholar]
  • 90.Pang J. C. et al. Geometric constraints on human brain function. Nature (2023) doi: 10.1038/s41586-023-06098-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 91.Leonardi N. & Van De Ville D. On spurious and real fluctuations of dynamic functional connectivity during rest. NeuroImage 104, 430–436 (2015). [DOI] [PubMed] [Google Scholar]
  • 92.Tournier J.-D. et al. MRtrix3: A fast, flexible and open software framework for medical image processing and visualisation. NeuroImage 202, 116137 (2019). [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHPP2023.06.22.546023V2-supplement-1.pdf (4.3MB, pdf)

Articles from bioRxiv are provided here courtesy of Cold Spring Harbor Laboratory Preprints

RESOURCES