Adolescent maturation of cortical excitation-inhibition ratio based on individualized biophysical network modeling

Abstract

The excitation-inhibition ratio is a key functional property of cortical microcircuits which changes throughout an individual's lifespan. Adolescence is considered a critical period for maturation of excitation-inhibition ratio. This has primarily been observed in animal studies. However, there is limited human in vivo evidence for maturation of excitation-inhibition ratio at the individual level. Here, we developed an individualized in vivo marker of regional excitation-inhibition ratio in human adolescents, estimated using large-scale simulations of biophysical network models fitted to resting-state functional imaging data from both cross-sectional (n = 752) and longitudinal (n = 149) cohorts. In both datasets, we found a widespread decrease in excitation-inhibition ratio in association areas, paralleled by an increase or lack of change in sensorimotor areas. This developmental pattern was aligned with multiscale markers of sensorimotor-association differentiation. Although our main findings were robust across alternative modeling configurations, we observed local variations, highlighting the importance of methodological choices for future studies.

Fig. 1.. Overview.

Individualized BNM simulation-optimization (A to C) was performed to derive the subject/session–specific regional measures of the E-I ratio, defined as time-averaged in silico input current to the excitatory neurons, $⟨I_i^{\mathrm{E}}⟩$ (D). The model consists of coupled excitatory and inhibitory neuronal pools in each node, where the excitatory neuronal pools of brain nodes are interconnected through the structural connectome of the given subject/session [(A), left]. The model is controlled by a global parameter $G$ , which adjusts interregional coupling, in addition to regional parameters $w_i^{\text{EE}}$ , $w_i^{\text{EI}}$ , and $w_i^{\text{IE}}$ , which characterize the connection weights between excitatory and inhibitory neuronal pools within each node. In each simulation, $G$ , $w_i^{\text{EE}}$ , and $w_i^{\text{EI}}$ are set by the optimizer, while $w_i^{\text{IE}}$ is determined on the basis of the FIC algorithm (B). The covariance matrix adaptation-evolution strategy was used to optimize model parameters given empirical data of a subject/session (C). The optimization goal was to maximize the goodness of fit by tuning 15 free parameters, including $G$ , as well as bias and coefficient terms used to determine $w_i^{\text{EE}}$ and $w_i^{\text{EI}}$ based on six fixed biological cortical maps (fig. S1). The goodness of fit of each simulation to the empirical functional data [(A), right] was assessed as the correlation of FC matrices subtracted by their absolute mean difference and the KS distance of FCD matrices derived from the simulated or empirical BOLD signal. After completion of two optimization runs, the optimal simulation with the best goodness of fit to the empirical functional data of the target subject/session was selected (C). Last, the in silico input current to the excitatory neuron of each node $I_i^{\mathrm{E}}$ was averaged across simulation time, resulting in an E-I ratio map for each subject/session (D).

Fig. 2.. Cross-sectional effect of age on the E-I ratio during adolescence.

(A) Effect of age on the E-I ratio, showing its significant age-related decrease (blue) and increase (red) during adolescence in the PNC dataset, after removing outliers and controlling for the goodness of fit, sex, and in-scanner rs-fMRI motion, corrected for multiple comparisons using FDR. (B) Unthresholded map of the effect of age on the E-I ratio. (C) Distribution of correlation coefficients between E-I ratio age effect maps of all pairs of subsamples across 100 half-split subsamples of the dataset.

Fig. 3.. Longitudinal effect of age on the E-I ratio during adolescence.

(A) Longitudinal effect of age on the E-I ratio, showing its significant decrease (blue) and increase (red) through adolescence, using a mixed-effects model with random intercepts for each subject, after removing outliers and controlling for goodness of fit, sex, in-scanner rs-fMRI motion, and site, corrected for multiple comparisons using FDR. (B) Unthresholded effect of age on the E-I ratio. (C) Distribution of correlation coefficients between E-I ratio age effect maps of all pairs of subsamples across 100 half-split subsamples of the dataset. (D) Conjunction of regions showing significant decreases in E-I ratio associated with age in the PNC and IMAGEN datasets. (E) Spatial coalignment [Pearson correlation (r) or cosine similarity (cos)] of longitudinal effects of age on the E-I ratio in IMAGEN with a cross-sectional effect of age on the E-I ratio in PNC.

Fig. 4.. Embedding of the E-I developmental pattern in the PNC dataset along the sensorimotor-association axis.

(A) Spatial correlation of the E-I ratio maturation map in the PNC dataset with the maps of the sensorimotor-association cortical axis based on Sydnor et al. (18) (fig. S6). Colored diamonds show statistically significant (P_spin < 0.05) positive (red) and negative (blue) spatial correlations. (B) Distribution of the E-I ratio maturation map across the canonical resting-state networks (F = 13.85, P_spin < 0.001). Post hoc tests (Bonferroni-corrected) showed significantly more positive age effects in the visual (VIS) and somatomotor (SMN) compared to the limbic (LIM), frontoparietal (FPN), and default mode networks (DMN), in addition to more positive age effects in the dorsal attention network (DAN) compared to DMN. (C) Bottom: Mean expression of the top 500 genes associated with the E-I ratio maturation map, split into sets of negatively associated (n = 187, blue) and positively associated (n = 313, red) genes. Top: Specific expression analysis of the two sets of genes across developmental stages in the cortex. The y axis shows the negative log of FDR-corrected P values. Asterisks denote significantly enriched developmental stages compared to null genes based on spin surrogate maps (1000 permutations) and after FDR adjustment. SA, sensorimotor association; Evo., evolutionary; CMR, cerebral metabolic rate; Glu., glucose; CBF, cerebral blood flow; NeuroSynth PC1, principal component of NeuroSynth meta-analytical maps; LTC G1, principal gradient of laminar thickness covariance.

Fig. 5.. Sensitivity analyses.

The unthresholded effect of age on the E-I ratio observed in a random subsample of the PNC dataset (n = 200) using the default configurations (A) compared to age effects observed using alternative configurations, including the following: (B) using a fixed template SC based on the MICs dataset, (C) definition of nodes based on a Schaefer parcellation with higher granularity of 200 nodes, (D) using alternative subsets of biological maps to determine the heterogeneity of regional parameters, (E) including the interhemispheric connections in the goodness of fit, and (F) using an alternative Gaussian noise seed. In (B) to (F), the statistics indicate spatial coalignment [Pearson correlation (r) or cosine similarity (cos)] of each map with the E-I ratio maturation map observed using the default configurations (A). (G) Pooled partial correlation of age with the E-I ratio (controlled for goodness of fit, sex, and in-scanner rs-fMRI motion) across (A) to (F) based on random-effects meta-analyses.

Fig. 6.. Optimal model parameter interrelation and association with age in the PNC dataset.

(A) Left: Pearson correlation of model parameter $G$ and brain-averaged values of regional parameters $w^{\text{EE}}$ , $w^{\text{EI}}$ , and $w^{\text{IE}}$ across subjects are shown. Asterisks denote statistically significant correlations. Right: Interrelation of regional values of parameters $w^{\text{EE}}$ , $w^{\text{EI}}$ , and $w^{\text{IE}}$ across nodes and subjects based on a linear mixed-effects model with random intercepts and slopes per each node. (B) Left: Effect of age on optimal parameter $G$ . Points represent the residual of $G$ for each subject after removing confounds. Right: Unthresholded effect of age on regional parameters $w^{\text{EE}}$ , $w^{\text{EI}}$ , and $w^{\text{IE}}$ . Age in years (y)