Linking changes in sulcal morphometry to cognitive development from childhood to adolescence

Abstract

Despite dynamic sulcal changes during youth paralleling skill development, the link between extended postnatal development and cognition remains underexplored. This study analyzes structural MRI data from 307 children (6-14 years), with longitudinal data (inter-scan interval ~1 year) available for a subset. Results reveal widespread cortical thinning, sulcal widening, and reductions in adjusted area and depth, following a chronological gradient where earliest-forming sulci undergo the most profound change. Longitudinal remodeling, rather than baseline morphometry, predicts cognitive gains. Specifically, working memory improvements are predicted by widening of the left calcarine and posterior intralingual sulci. In contrast, attention network maturation involves global changes, though executive control is specifically linked to left intraparietal sulcus widening. Gene enrichment analysis links these changes to synaptic processes. This study advances our understanding of the association between sulcal morphometry and cognitive function, elucidating potential mechanisms underlying brain development from childhood to adolescence.

Sulcal morphometric remodeling during childhood and adolescence, characterized by cortical thinning and widening, predicts cognitive gains in working memory and attention, with gene enrichment linking these changes to synaptic processes.

Introduction

In gyrencephalic mammals like humans, the cerebral cortex folds into sulci and gyri, forming intricate morphological configurations essential for brain function¹. These cortical folds emerge through sophisticated biological and biomechanical mechanisms, with genetic factors playing a key role in regulating their developmental trajectory^2–8. Notably, different sulci appear in a relatively stable temporal sequence across individuals, despite minor variations observed in Magnetic Resonance Imaging (MRI) and anatomical studies^9–15. This developmental timeline spans the fetal period. Based on this sequence, sulci are categorized as primary (starting ~26 gestational weeks, GW), secondary (~32 GW), or tertiary (~term or shortly after)^16–18. This genetically controlled temporal emergence suggests that sulci formed at different fetal stages undergo distinct morphological changes during childhood and adolescence. Investigating these trajectories—potentially reflecting continued genetic programmatic control—could provide valuable insights into the mechanisms of brain structural development.

While qualitative sulcal patterns remain longitudinally stable from childhood to adulthood¹⁹, sulcal morphometry undergoes dynamic remodeling alongside micro- and macrostructural organization across the postnatal lifespan²⁰. Following initial neurogenesis and migration (12-20 GW)²¹, the brain undergoes continuous microstructural remodeling through axonal myelination and synaptic reorganization, with synaptic pruning eliminating >40% of synapses during childhood to adolescence²². These processes are accompanied by macrostructural changes: cortical volume follows an inverted U-shaped trajectory²³, surface area also peaks and then declines²⁴, thickness decreases regionally (most in parietal, least in occipital regions)²⁵, sulcal depth decreases while width increases²⁶, and overall folding degree (measured by Gyrification Index) rises in early childhood before decreasing²⁷. Yet systematic studies characterizing the morphometric developmental changes of different sulci in children and adolescents are lacking.

Empirical evidence demonstrates that sulcal morphological changes occur concurrently with cognitive improvements²⁸, with age-related gyrification shifts consistently linked to cognitive development²⁹. For instance, the morphological variability of the paracingulate sulcus has been associated with executive control³⁰. In the domain of numeracy, the sulcal pattern of the left intraparietal sulcus (IPS)—specifically the presence of a sectioned IPS—is significantly related to arithmetic performance in children and adolescents³¹. Furthermore, recent studies highlight the role of late-developing structures in higher-order cognition, showing that the depth and presence of tertiary sulci in the lateral prefrontal cortex, such as the para-intermediate frontal sulcus, are predictive of reasoning abilities^32–35. These findings align with timing-function relationships where early-developing linear sulci in unimodal cortices associate with basic sensorimotor processing, while later-developing tertiary sulci in heteromodal regions link to complex cognitive functions^8,16,17. However, cognitive development likely emerges from coordinated maturation across multiple sulcal regions rather than isolated changes in specific areas³⁶, necessitating systematic whole-brain investigations of sulcal-cognitive relationships that remain largely underexplored.

The significance of studying sulcal morphometric changes also lies in understanding the molecular basis of development, where macroscopic morphological features reflect underlying microstructural changes driven by gene expression-related processes^4,37–39. Cortical surface area is driven by radial glia proliferation and relates to minicolumn number and myelination processes⁴⁰, while thickness is determined by intermediate progenitor cells and associated with synaptic pruning and myelination⁴¹. Regional differences in cellular proliferation, coupled with axonal tension and biomechanical mechanisms, influence sulcal patterns, including folding degree, depth, length, and width⁴². Crucially, deviations in these sulcal morphometric trajectories are closely linked to atypical neural connectivity and have been identified as potential biomarkers for neurodevelopmental disorders, such as autism spectrum disorder and schizophrenia^43,44. Since these processes vary temporally and spatially, changes in sulcal morphometry throughout development should reflect corresponding gene expression patterns, providing a molecular basis for understanding postnatal cortical development.

Based on this evidence, we hypothesize that: (1) sulci emerging earlier in fetal development will exhibit more profound morphometric remodeling during childhood and adolescence compared to later-emerging sulci; (2) longitudinal sulcal remodeling will be predictive of cognitive gains, distinct from baseline morphometric associations; and (3) the changes reflect specific cellular processes and gene expression patterns identifiable through enrichment analysis. To test these hypotheses, we characterized whole-brain sulcal developmental trajectories aged 6-14 years from the first batch of Children School Functions and Brain Development Project in China (CBD, Beijing Cohort) using generalized additive mixed models (GAMMs)⁴⁵. We systematically investigated brain–behavior associations by distinguishing between baseline and longitudinal effects, using partial correlation (controlling for age and sex), Partial Least Squares (PLS) correlation⁴⁶, and Least Absolute Shrinkage and Selection Operator (LASSO) regression⁴⁷. Finally, we conducted gene enrichment analysis using whole-brain expression data to explore underlying biological processes. Our results reveal that sulcal widening and deepening follow distinct, region-specific non-linear patterns that significantly predict individual differences in cognitive development. These findings provide a multi-scale link between cortical folding patterns, gene expression, and cognitive maturation, highlighting the pivotal role of sulcal morphology as a biomarker for neurodevelopmental progress.

Results

Age-related development of sulcal morphometry from childhood to adolescence

The current analysis examined developmental trajectories of sulcal morphometry in a longitudinal cohort of 307 typically developing children aged 6.1 to 13.9 years (141 females, 166 males), contributing a total of 490 MRI scans across three time points with an inter-scan interval of approximately one year (N = 43 with 3 scans, N = 97 with 2 scans, and N = 167 with 1 scan; see Supplementary Figure 1). As shown in Fig. 1a, following HCP pipeline preprocessing and cortical surface reconstruction, sulcal identification and morphometric quantification were performed using the Morphologist toolbox within BrainVISA, which employed the Statistical Parametric Anatomy model(SPAM) for automated sulcal labeling^48–52. After rigorous quality control procedures, including manual inspection for extraction artifacts and exclusion of sulci with success rates below 75%, 53 whole-brain sulci were retained for analysis. Six morphometric features were extracted for each sulcus: surface area (SA), cortical thickness (CT), sulcal width (SW), maximum depth (maxD), mean depth (meanD), and sulcus length (SL). To control for brain size variations, SA, SL, maxD, and meanD underwent total intracranial volume (TIV) adjustment using robust MM-estimator regression. Generalized Additive Mixed Models (GAMMs) with thin-plate regression splines were employed to model non-linear age-related changes while accounting for the longitudinal data structure, with nested model comparisons and false discovery rate correction applied to identify robust developmental effects across sulcal regions and morphometric features.

Fig. 1. Workflow for sulcal morphometry analysis.

a Preprocessing phase: T1-weighted structural MRI (sMRI) images are initially processed using the Human Connectome Project (HCP) pipeline. The resulting cortical ribbon and T1-weighted images are then input into the BrainVISA Morphologist pipeline, which includes steps for brain segmentation, gray-white matter differentiation, and the generation of pial and white matter meshes. These processes facilitate the construction of the cortical folding graph, resulting in automatically labeled graphs. Key sulcal metrics are extracted during this process and are further analyzed in relation to cognitive outcomes, developmental modeling, and genetic analysis. b Sulcal selection: Based on extraction stability and existing literature, we selected 53 sulci throughout the brain.

Morphometric changes of each sulcus throughout the brain

Statistical screening using GAMMs with Benjamini-Hochberg False Discovery Rate (FDR) correction identified 120 models with significant age-related effects across sulcal regions and morphometric features (see Table 1, Supplementary Data 1). The overall magnitude of age-related changes was modest, with a mean Δadj R² of −0.017 (range: −0.099 to 0.114). Consistent with previous studies, we observed systematic morphometric changes characterized by decreases in adjusted surface area, adjusted maximum and mean depths, and adjusted sulcal length, accompanied by cortical thinning and sulcal widening across the brain^26,53. To illustrate the most prominent trends, Fig. 2a presents the sulci with the greatest developmental changes for each feature, which serve as examples of the patterns observed across almost all sulci. Cortical thickness emerged as the most developmentally sensitive measure, with 34 sulci showing consistent thinning, primarily in parietal, lateral temporal, and ventral prefrontal regions. Sulcal width increased in 21 sulci, predominantly in lateral temporal and parietal areas. Notably, TIV adjustment revealed stronger and more consistent age-related effects compared to raw measures, demonstrating that controlling for global brain size enhances the detection of specific sulcal developmental trajectories.

Table 1.

Age-related developmental changes in sulcal morphometry

Label	SA (adj)	maxD (adj)	meanD (adj)	SL (adj)	CT	SW
F.C.L._left	−0.045*	0.002	−0.005	0.001	−0.072*	0.058*
F.C.L._right	−0.040*	0.001	−0.001	−0.012	−0.034	0.018
F.C.M.ant._left	−0.013*	−0.015	−0.028	0.003	−0.037*	0.023
F.C.M.ant._right	−0.004	−0.017	−0.003	0.002	−0.029*	0.000
F.C.M.post._left	−0.004	−0.012*	−0.012*	−0.002	−0.027*	0.012
F.C.M.post._right	0.001	−0.019*	−0.019*	−0.002	−0.047*	0.030*
F.Cal.ant.-Sc.Cal._left	−0.027*	−0.014	−0.01	−0.013*	−0.059*	0.040
F.Cal.ant.-Sc.Cal._right	−0.002*	−0.009	−0.018*	−0.003	−0.024	0.017
F.Coll._left	−0.017*	−0.035*	−0.024	0.002	−0.030*	0.013
F.Coll._right	−0.015*	−0.032*	−0.031*	0.000	−0.030	0.010
F.I.P._left	−0.041*	−0.023*	0.005	−0.030*	−0.091*	0.030*
F.I.P._right	−0.013*	−0.017*	−0.01	0.001	−0.071*	0.033*
F.I.P.r.int.2_left	0.003	−0.002	−0.002	0.002	−0.025	0.004
F.P.O._left	−0.013*	−0.016	−0.012	−0.002	−0.081*	0.024*
F.P.O._right	−0.025*	−0.006*	−0.004	0.002	−0.099*	0.041*
OCCIPITAL_left	−0.002	0.003	0.004	0.001	−0.034*	0.022*
OCCIPITAL_right	−0.009	−0.009	−0.005	0.001	−0.032	0.018
S.C._left	−0.008*	−0.011	0.002	0.001	−0.010	0.006
S.C._right	−0.013*	−0.012*	0.005	−0.003	−0.013	0.007
S.Call._left	−0.005*	−0.013	−0.02	0.006	−0.018	0.010
S.Call._right	0.001	0.004	0.003	−0.001	−0.024	−0.001
S.Cu._left	−0.010*	−0.011	−0.007	−0.002	−0.058*	0.022
S.Cu._right	−0.015*	0.001	0.003	−0.012*	−0.052*	0.030
S.F.inf._left	−0.004	0.005	0.003	−0.002	−0.069*	0.033
S.F.inf._right	−0.004	−0.002	0.001	0.000	−0.051*	0.054*
S.F.int._left	−0.006	−0.003	−0.003	−0.004	−0.005	0.010
S.F.int._right	−0.008	−0.005	0.002	−0.01	−0.027	0.005
S.F.inter._left	−0.001	−0.001*	0.000	0.003	−0.036*	0.015
S.F.inter._right	0.008	0.003	−0.008	0.004	−0.054*	0.016*
S.F.orbitaire._left	0.000	−0.001	−0.011	0.003	−0.011	0.033
S.F.sup._left	−0.012*	−0.031*	−0.009	−0.006	−0.013	0.011
S.F.sup._right	−0.002*	−0.001	−0.003	0.005	−0.018*	0.024*
S.Li.ant._left	−0.003	−0.002	−0.003	−0.002	−0.026	0.005
S.Li.ant._right	0.002	0.002	0.002	0.002	−0.025	0.015
S.Li.post._left	−0.007	−0.006	−0.019	0.002	−0.008	0.015
S.Li.post._right	−0.009	−0.021	−0.005	0.002	−0.036*	0.009
S.O.T.lat._left	−0.019*	−0.012*	−0.011*	−0.005	−0.025*	0.037*
S.O.T.lat._right	−0.011*	0.000	−0.008	−0.019*	−0.051*	−0.004
S.Olf._left	−0.005	−0.027	−0.036*	0.001	−0.083*	0.053*
S.Olf._right	−0.028*	−0.045*	−0.040*	0.000	−0.073*	0.080*
S.Or._left	−0.023*	−0.007	−0.016	−0.002	−0.048*	0.011
S.Or._right	−0.037*	−0.023*	−0.015*	−0.009	−0.040*	0.031*
S.Pa.int._left	−0.009*	0.000	0.015*	−0.003	−0.081*	0.026*
S.Pa.int._right	−0.040*	−0.017*	−0.007*	−0.031*	−0.057*	0.022*
S.Pa.t._right	−0.004	−0.004	−0.005	−0.01	−0.023	0.000
S.Pe.C._left	−0.017*	0.001*	−0.01	−0.012*	−0.001	0.005
S.Pe.C._right	0.003*	0.018*	0.012	−0.002	−0.009	0.007
S.Po.C._left	−0.009*	−0.016*	−0.003	0.001	−0.057*	0.018
S.Po.C._right	0.001	0.004	−0.002	0.000	−0.047*	0.022*
S.T.i._left	−0.008*	−0.002	0.000	−0.006	−0.053*	0.114*
S.T.i._right	−0.009*	0.007	0.015	−0.011*	−0.047*	0.067*
S.T.s._left	0.001	0.002	−0.003	−0.006	−0.065*	0.037*
S.T.s._right	−0.004*	0.000	−0.001	−0.006	−0.039*	0.034*

Note: Values represent the change in $\mathrm{\Delta }\text{adj}{R}^2$ derived from GAMMs. To isolate specific regional morphometric changes from global brain growth, Surface Area, Sulcal Length, Max Depth, and Mean Depth were analyzed using TIV-adjusted residuals. Cortical Thickness and Sulcal Width are presented as raw measures.

adj TIV-adjusted residuals, SA surface area, maxD maximum depth, meanD mean depth, SL sulcal length, SW sulcal width, CT cortical thickness.

^*Indicate significant age-related effects that survived FDR correction.

Fig. 2. Quantitative analysis of age-related changes in sulcal morphometry.

a Developmental trajectories of key sulcal features. The developmental characteristics for each sulcus were modeled using a GAMM. To represent the most prominent developmental trends, this panel displays the trajectories of the sulci showing the largest developmental effects (maximum absolute Δ adj R²) for each of the six morphometric features (Surface Area, Maximum Depth, Mean Depth, Sulcal Length, Sulcal Width, and Cortical Thickness). The Δadj R² and p-values from the GAMM are provided, with blue indicating males and red indicating females. b Regional distribution of developmental effects $\mathrm{\Delta }\text{adj}{R}^2$ values associated with age-related changes across six features for various sulci. All sulci displayed in the figure exhibit significant age-related developmental effects ($p<0.05$, $p_{fdr}$ $<0.05$).

Regional analysis (Fig. 2b) revealed heterogeneous patterns. For TIV-adjusted surface area, decreases were found in bilateral lateral fissures, left intraparietal sulcus, right orbital sulcus, and right internal parietal sulcus. TIV-adjusted maximum depth reductions were most prominent in the left superior frontal sulcus, bilateral intraparietal sulci, bilateral collateral fissures, and right olfactory sulcus. Similarly, adjusted mean depth showed marked changes in bilateral olfactory sulci and the right collateral fissure. Cortical thickness reduction was widespread, showing the strongest effects in the left intraparietal sulcus and right parieto-occipital fissure. Detailed statistics regarding specific developmental effects for these regions are provided in Table 1. A total of 9 models demonstrated significant main effects of sex (Supplementary Table 2). Females predominantly exhibited greater sulcal width, a finding consistent with allometric scaling principles, suggesting that larger brains (typically males) undergo tighter cortical folding, thereby resulting in narrower sulci. Overall, sex differences in sulcal metrics were relatively modest.

Association between sulcal emergence timing and developmental magnitude

To investigate whether the chronological order of sulcal formation organizes postnatal development, we examined the relationship between the timing of sulcal emergence in utero and the magnitude of age-related changes observed in our cohort^16,54. We hypothesized that primary folds are the first to emerge under strong genetic control, establishing stable anatomical landmarks that persist from birth^55–57. In contrast, secondary and tertiary folds emerge later and are increasingly susceptible to mechanical and environmental influences²⁸.To investigate whether the chronological order of sulcal formation organizes childhood development, we examined the relationship between the timing of sulcal emergence and the magnitude of age-related changes observed in our cohort. Sulcal emergence was parameterized using three complementary metrics (detailed in Methods): $T_{app}$ (the gestational week of visual appearance based on anatomical consensus, Fig. 3a)^9–13,16, and $T_{10}$ / $T_{50}$ (the gestational ages marking the onset and midpoint of folding, derived from curvature modeling)⁸.

Fig. 3. Impact of sulcal emergence timing on postnatal morphometric remodeling.

a Surface map visualization of Sulcal Appearance Time ($T_{app}$), defined as the gestational week when a sulcus first becomes visually identifiable based on anatomical consensus (see “Methods”). b Scatter plots illustrating the relationship between sulcal emergence ($T_{app}$ and $T_{10}$) and the magnitude of postnatal development for adjusted surface area. The y-axis represents the signed $\mathrm{\Delta }\text{adj}{R}^2$, where more negative values indicate a stronger age-related reduction. The significant positive correlations indicate that earlier-emerging sulci (lower $T_{app}$) exhibit more profound morphometric reductions (larger negative effect sizes) and higher statistical significance during childhood and adolescence.

After correction for multiple comparisons, we identified robust associations between emergence timing and developmental magnitude. For adjusted surface area, we observed significant positive correlations between $T_{app}$ and the Signed $\mathrm{\Delta }\text{adj}{R}^2$ ($r=0.36$, $p_{fdr}=0.031$, Fig. 3b). Similarly, T₁₀ showed a positive correlation with the Signed $\mathrm{\Delta }\text{adj}{R}^2$ ($r=0.39,p_{fdr}=0.031$, Fig. 3b). Since surface area generally decreases with age (yielding negative signed $\mathrm{\Delta }\text{adj}{R}^2$ values), this positive correlation indicates that earlier-emerging sulci exhibit significantly larger negative values compared to later-emerging sulci, which show effects closer to zero. Similar patterns were observed for adjusted maximum depth, where earlier emergence was significantly correlated with stronger age-related changes (r = 0.39, p_fdr = 0.034). These findings suggest a chronological gradient where the earliest-forming primary sulci undergo the most profound and consistent morphological remodeling during childhood, whereas later-emerging sulci exhibit more attenuated group-level changes.

Association between sulcal morphometry changes and cognitive performance improvements

Absence of associations between baseline sulcal morphometry and cognitive abilities

To distinguish developmental plasticity from pre-existing individual differences, we first examined whether baseline sulcal morphometry predicted cognitive abilities. Neither mass-univariate nor multivariate analyses identified robust associations between baseline sulcal metrics and cognitive scores. These results suggest that baseline sulcal morphometry alone is insufficient to account for the variance in cognitive performance in this cohort, highlighting the necessity of investigating longitudinal remodeling.

Associations between sulcal morphometry changes and working memory gains

We examined the associations between longitudinal changes in sulcal morphometry and gains in working memory (WM) (N = 138). We observed distinct patterns between the two hemispheres regarding these structure-behavior associations.

In the left hemisphere, structural plasticity exhibited a highly focal pattern. The LASSO regression successfully identified a robust predictive model for 2-back performance gains (Test set Spearman r = 0.53, p = 0.004; see Fig. 4a). This prediction relied on two specific sulci: the widening of the left calcarine fissure (F.Cal.ant.Sc.Cal, $\beta =0.07$) and the posterior intralingual sulcus (S.Li.post, $\beta =0.06$). These findings were rigorously cross-validated by Multivariate PLS analysis, which confirmed a significant latent mode of covariance specifically for sulcal width changes (Overall p = 0.003). Furthermore, univariate partial correlation analyses (controlling for age and sex) also provided independent validation of the LASSO feature selection. Notably, the posterior intralingual sulcus also exhibited concurrent increases in adjusted sulcal depth ($r_{partial}=0.26,p_{fdr}=0.002$) and adjusted length ($r_{partial}=0.23,p_{fdr}=0.002$), suggesting that improved working memory capacity corresponds to a multifaceted structural expansion in this critical brain region. To further validate the specificity of these features, we compared the LASSO model against a Ridge regression. The Ridge model failed to generalize (Test set r = 0.07, p = 0.731, supplementary Fig. 4), confirming that the association with WM gains in the left hemisphere is driven by specific topographic changes rather than a diffuse global effect.

Fig. 4. Longitudinal sulcal remodeling predicts gains in working memory and attention network efficiency.

a Predictive modeling of working memory gains. The LASSO sparse regression framework identifies specific sulci whose structural remodeling predicts improvements in 2-back task performance. Left panels: The evolution of LASSO coefficients for each left-hemisphere feature as a function of the regularization parameter λ (top), and the cross-validation error ($RMS{E}_{cv}$) curve (bottom), where the red vertical line indicates the optimal λ at minimum error. Top right: Brain map illustrating the topographic locations of the key sulci selected by the model. Bottom right: Scatter plot depicting the correlation between observed and model-predicted performance gains within the 20% independent test set ($Spearmanr=0.53,p=0.004$). b–d Structural correlates of Attention Network Test (ANT) sub-networks. Panels represent b Alerting, c Orienting, and d Executive Control networks. For each network, the brain maps display the two most significant sulcal correlates in each hemisphere ($p_{fdr}<0.05$). The hue identifies the specific morphometric feature associated with each sulcus: Blue (cortical thickness), Green (adjusted maximum depth), Red (adjusted mean depth), Purple (adjusted surface area), Gray (adjusted sulcal length), and Orange (sulcal width). The color saturation represents the absolute partial correlation magnitude ($\mid r_{partial}\mid$), controlling for age and sex. adj: TIV-adjusted residuals, SA: surface area, maxD: maximum depth, meanD: mean depth, SL: sulcal length, SW: sulcal width, CT: cortical thickness.

In contrast, LASSO failed to identify individual predictors of WM gains in the right hemisphere. However, multivariate PLS analysis detected a significant latent mode of covariance between right-hemisphere remodeling and WM gains (Overall p < 0.001). Univariate partial correlation analyses (controlling for age and sex) further dissected this pattern. Specifically, positive associations were prominent in posterior regions, characterized by increased adjusted surface area in the right transverse parietal sulcus (S.Pa.t, $r_{partial}=0.29,p_{fdr}=0.007$), greater adjusted sulcal length in the lateral fissure (F.C.L, $r_{partial}=0.25,p_{fdr}<0.001$), and greater adjusted maximum depth in the lateral occipito-temporal sulcus (S.O.T.lat, $r_{partial}=0.22,p_{fdr}=0.013$). Conversely, negative associations were observed in frontal and medial regions, most notably in the inferior frontal sulcus (S.F.inf) and cuneus sulcus (S.Cu), which exhibited consistent reductions across both adjusted surface area and adjusted sulcal length.

Associations between sulcal morphometry changes and attention network efficiency

We next investigated the structural correlates of the three attention sub-networks: Executive Control, Alerting, and Orienting (N = 142, Fig. 4b–d). Unlike the findings for left-hemisphere working memory, sparse regression models (LASSO) did not yield significant predictions for any attention network in either hemisphere. However, multivariate PLS analyses revealed a robust and ubiquitous signal: significant latent modes of covariance were detected across all six morphometric indices in both hemispheres, indicating that the maturation of attention networks involves a global, multi-modal cortical remodeling rather than focal changes. Univariate partial correlation analyses (controlling for age and sex) further dissected these patterns:

Executive control

Improved executive control was linked to the IPS and limbic regions. In the left hemisphere, the widening of the secondary intermediate ramus of the intraparietal sulcus (F.I.P.r.int.2) was the strongest predictor of reduced executive control cost ($r_{partial}=-0.28,p_{fdr}<0.001$). In the right hemisphere, significant associations were observed as increased cortical thickness in the olfactory sulcus (S.Olf, $r_{partial}=0.27,p_{fdr}<0.001$) and reduced sulcal width in the collateral fissure (F.Coll, $r_{partial}=-0.24,p_{fdr}<0.001$).

Alerting

Alerting efficiency was associated with distributed changes in fronto-parietal regions. In the left hemisphere, where PLS showed the most extensive involvement of cortical thickness (18 significant loadings), the orbital frontal sulcus (S.F.orbitaire) emerged as a key hub, with increased adjusted surface area correlating with better alerting scores ($r_{partial}=0.26,p_{fdr}<0.001$). Conversely, in the right hemisphere, the intraparietal sulcus (F.I.P) showed a strong negative association with adjusted sulcal length ($r_{partial}=-0.34,p_{fdr}<0.001$).

Orienting

The orienting network exhibited fronto-posterior structural associations. In the left hemisphere, efficiency gains were associated with the widening of the orbital frontal sulcus (S.F.orbitaire, $r_{partial}=0.25,p_{fdr}<0.001$) and narrowing of the parieto-occipital fissure (F.P.O, $r_{partial}=-0.23,p_{fdr}<0.001$). In the right hemisphere, PLS results highlighted adjusted maximal sulcal depth as the dominant feature (12 significant loadings), a finding closely mirrored by the univariate analysis: the deepening of the right anterior intralingual sulcus (S.Li.ant, $r_{partial}=0.30,p_{fdr}<0.001$) and the shallowing of the superior frontal sulcus (S.F.sup, $r_{partial}=-0.26,p_{fdr}<0.001$) were the primary drivers.

Sulcus development in children aged 6–14 years is associated with gene expression profiles

To investigate the molecular basis underlying sulcal morphometric changes, we analyzed transcriptomic data from 53 brain sulcal regions using the Allen Human Brain Atlas database integrated through the abagen toolbox^58,59 (version 0.1.3, https://github.com/rmarkello/abagen). We examined correlations between regional gene expression patterns and the magnitude of age-related sulcal changes (Δadj R²) across these anatomical regions. Given the inherent spatial structure of brain data, we applied a rigorous statistical approach using BrainSmash-generated null distributions (10,000 permutations) to account for spatial autocorrelation effects⁶⁰. This analysis revealed a subset of genes whose expression levels were significantly associated with developmental sulcal changes after spatial correction. These candidate genes were then analyzed for functional pathways, molecular functions, and biological processes using Metascape enrichment analysis⁶¹ (Fig. 5a).

Fig. 5. Gene enrichment analysis workflow and results.

a Whole-brain gene expression data were organized according to the BrainVISA atlas to obtain gene expression levels in different brain regions. We manually converted the probabilistic atlas into a discrete atlas containing 53 sulcal parcels through binarization and other processing methods. Correlation analyses were conducted with the Δadj R² of various morphometry features. After performing a permutation test, a significant subset of genes correlating expression levels and Δadj R² was identified for enrichment analysis. b The figure displays a bar chart of biological processes significantly enriched for sulcal width and adjusted maximum depth, alongside their corresponding p-values.

Gene enrichment analysis across different sulcal morphometric features identified meaningful associations for SW and adjusted maxD with neurodevelopmental processes (Fig. 5b). All significantly correlated genes exhibited positive associations with the morphometric changes; therefore, no distinction was made between positively and negatively correlated gene sets. For SW, significantly enriched terms included “synaptic signaling” (GO:0099536, –logP = 4.93, enrichment = 3.30), “structural constituent of presynaptic active zone” (GO:0098882, –log P = 4.33, enrichment = 38.45), and “axon” (GO:0030424, –log P = 3.72, enrichment = 2.57), all of which are critical for neural communication and synaptic function in the brain. For adjusted maxD, strongly enriched terms included “postsynaptic intermediate filament cytoskeleton organization” (GO:0099185, –log P =  5.24, enrichment = 55.69), “regulation of membrane potential” (GO:0042391, –log P = 4.71, enrichment = 2.70), and “dendrite” (GO:0030425, –log P = 4.51, enrichment = 2.40). Notably, both SW and adjusted maxD results highlighted terms related to axons, dendrites, synapses, and neuronal structure, emphasizing the role of synaptic architecture and cytoskeletal organization in childhood sulcal development.

Discussion

Leveraging large-scale longitudinal neuroimaging data combined with transcriptomics and cognitive behavioral assessments, this study comprehensively characterizes the developmental landscape of sulcal morphometry from childhood to adolescence (ages 6–14). Our analyses reveal three core findings: First, sulcal morphometric remodeling during this period follows a global pattern of gradual structural refinement. After controlling for total intracranial volume to account for global growth effects, we observed widespread cortical thinning, reduced adjusted sulcal surface area, and sulcal widening^20,25,26. Second, we found that sulcal developmental trajectories adhere to a consistent “temporal principle”: primary sulci, which emerge earliest during the fetal period, exhibit the most significant population-level developmental effects (Signed $\mathrm{\Delta }\text{adj}{R}^2$) during childhood. This finding is consistent with the “Canalization” theory of development⁶². Finally, by integrating the Allen Human Brain Atlas (AHBA) gene atlas and cognitive data, we elucidated the biological underpinnings of sulcal remodeling. We found that morphometric changes in specific sulci, such as branches of the left intraparietal sulcus, specifically predict improvements in working memory and attention network functions. These findings highlight sulcal morphometry as a sensitive biomarker for tracking brain maturation and cognitive development.

Consistent with earlier findings, our results reveal changes in sulcal features, including reductions in adjusted surface area, adjusted maximum and mean depths, and adjusted sulcal length, alongside cortical thinning and widening across the brain²⁶. In contrast to the dramatic cortical expansion observed in infancy, the sulcal changes in school-age children were modest in magnitude (annual rate < 1%) and effect size ($\mathrm{\Delta }\text{adj}{R}^2=0.017$). These results align with recent large-scale neuroimaging evidence indicating that reproducible brain-phenotype associations are typically characterized by small effects⁶³. These subtle changes reflect cortical optimization rather than simple atrophy. Specifically, sulcal widening may be mechanically coupled with cortical thinning via tension-based morphogenesis⁵. Our transcriptomic analysis supports this mechanical hypothesis: we found that sulcal width variations are enriched for genes related to synaptic signaling. As synaptic pruning reduces gray matter volume and dendritic density, the consequent decline in local axonal tension allows the sulcal walls to relax, manifesting as widening. This process accommodates the maturation of short-range U-fibers⁶⁴, marking a shift from generalized to specialized neural connectivity⁶⁵. Crucially, this remodeling holds functional significance during the peak of structural-functional coupling⁶⁶. We postulate that increased sulcal width reflects neural reorganization aimed at optimizing transmission efficiency, providing the biophysical basis for efficient cognitive networks. Consequently, future longitudinal studies are warranted to further investigate how macroscopic sulcal width variations reflect underlying microstructural processes in vivo. Finally, observed sex differences (deeper sulci in males vs. wider in females) largely vanished after correcting for total intracranial volume. This supports the “Allometry” hypothesis⁶⁷, indicating that sexual dimorphism arises from physical constraints—where larger brains require more compact folding—rather than divergent neurodevelopmental programs⁶⁸.

Our study reveals that primary sulci formed earliest in the fetal period (e.g., central and calcarine sulci) exhibit stronger age-related effects (larger signed $\mathrm{\Delta }\text{adj}{R}^2$) during childhood. This finding is consistent with the principle of developmental canalization⁶², whereby essential phenotypes are buffered against variation. Early sulci are governed by strong genetic programs⁶⁹ and subserve fundamental functions with shared developmental milestones. This commonality minimizes within-group variance, yielding high signal-to-noise ratios in population analyses. By contrast, later-developing association sulci show less pronounced group-level changes⁷⁰, as they are highly sensitive to unique environmental experiences and cognitive abilities³³. This sensitivity generates substantial inter-individual variability that dilutes average trends. Thus, strong effects in early sulci reflect developmental conservation, while the seemingly weak effects in late sulci underscore the brain’s high environmental adaptability. This dissociation implies that future research could explore the distinct roles of these regions, with primary sulci potentially reflecting conserved maturation and association sulci offering insights into neuroplasticity.

Sulcal morphology is not randomly distributed; rather, it is shaped by an underlying physiological organization tightly coupled with genetic factors. Evidence from animal models indicates that deviations in the size of somatosensory-motor areas during embryonic development—whether expansion or contraction—result in significant behavioral deficits⁷¹. This implies an evolutionarily ‘optimal size’ for cortical regions. Furthermore, fetal gene expression gradients are known to correlate with the orientation of early sulci, potentially by imposing spatial constraints on cortical folding^8,72,73. To uncover the molecular mechanisms distinct to specific morphologies, we leveraged AHBA transcriptomic data to analyze gene expression patterns across 53 sulcal regions. Enrichment analysis revealed distinct biological processes underlying different morphometric metrics, highlighting a molecular divergence between sulcal width and adjusted depth. Variations in sulcal width were preferentially enriched for gene sets related to synaptic signaling, axonal structure, and the presynaptic active zone. Given the positive association between these genes and sulcal morphometry, sulcal widening may reflect the maturation of neural circuitry and the enhancement of presynaptic transmission efficiency. In contrast, variations in maximum adjusted sulcal depth were primarily characterized by terms involving the postsynaptic compartment, specifically cytoskeletal organization and membrane potential regulation. Unlike the presynaptic signature of SW, sulcal depth appears to rely heavily on the structural integrity of the dendritic cytoskeleton. Mechanistically, these findings suggest a model where the intermediate filament cytoskeleton provides the rigid scaffolding necessary to maintain the deep structural integrity of the fold against mechanical stress⁷⁴ (linked to adjusted maximum depth), whereas axonal and presynaptic density likely modulate the superficial tension regulating the opening width. Thus, sulcal width and adjusted maximum depth may represent biological proxies for presynaptic and postsynaptic architectures, respectively. Future multi-scale studies should integrate these morphological metrics with proteomic data to further decode the biological semantics of cortical shape changes.

In the left hemisphere, the association between changes in sulcal morphometry and gains in working memory performance exhibited a distinct, localized pattern. Crucially, the LASSO model, which imposes a sparsity constraint to select only the most relevant features, identified specific longitudinal predictors—specifically, the widening of the calcarine sulcus and the posterior intralingual sulcus. In contrast, the Ridge regression model, which assumes diffuse whole-brain contributions, showed lower predictive power (Supplementary Figure 4). Anatomically, the calcarine sulcus is situated at the periphery of the primary visual cortex; its longitudinal widening likely reflects the structural optimization of these peripheral visual areas over time. Furthermore, the posterior lingual gyrus serves as a critical cortical anchor for the Inferior Longitudinal Fasciculus (ILF)^75,76. Given that the ILF is a core pathway connecting the occipital visual cortex to the anterior temporal memory system, supporting visual object recognition and short-term storage^77,78, we postulate that the morphometric remodeling (e.g., widening) of these specific ILF-anchored nodes may facilitate more efficient neural transmission of visual representations into memory systems. In contrast, the right hemisphere exhibited a distinct pattern of structure-WM associations. Anatomically, the right transverse parietal sulcus and inferior frontal sulcus correspond to the dorsal and ventral core nodes of the right Fronto-parietal Attention Network, respectively⁷⁹. We observed that gains in WM performance were associated with an expansion of adjusted surface area in the posterior parietal region, concurrent with a contraction in the anterior inferior frontal region, manifested as reductions in adjusted surface area and adjusted sulcal length. This pattern of inverse structural changes across anterior and posterior regions may reflect a redistribution of neural resources within the attention network architecture during cognitive maturation. Crucially, these findings underscore the potential of using specific, localized sulcal features as sensitive predictors for tracking individual trajectories of working memory development.

Efficiency gains in Attention Networks exhibited a more spatially distributed pattern of structural-functional covariance across both hemispheres. This observation aligns with the network’s intrinsic reliance on large-scale, long-range connectivity and global resource allocation⁸⁰. Within the Executive Control Network, the widening of the left F.I.P.r.int.2 (corresponding to the Jensen Sulcus) emerged as the most robust predictor of improvements in executive control. This region resides within the Multiple-demand network, a system critical for complex task processing and problem-solving⁸¹. This functional relevance is supported by recent evidence showing that the sulcal morphometry of the IPS is significantly associated with individual differences in cognitive domains, including symbolic number skills⁸² and language and memory performance⁸³. The Jensen Sulcus serves as a key anatomical landmark delineating the Supramarginal Gyrus (SMG) from the Angular Gyrus (AG). While acknowledging that this sulcus exhibits inter-individual variability, cytoarchitectonic evidence indicates that when morphologically present, it marks the precise functional transition within its depth⁸⁴. Consequently, we postulate that the widening of this sulcus signifies an enhanced “anatomical segregation” between the SMG and AG⁸⁵. This structural adaptation aligns with our transcriptomic findings: the widening likely results from a localized reduction in synaptic tension, which allows the sulcal walls to relax and separate. While the Superior Longitudinal Fasciculus may exert the necessary mechanical constraint to maintain the fold’s deep integrity^82,83, the superficial widening acts as a physical barrier to reduce neural interference between adjacent functional areas. Future research combining functional connectivity MRI with sulcal morphology could validate whether these structural segregations directly correlate with reduced functional crosstalk between the SMG and AG. Gains in alerting network efficiency were primarily associated with adjusted surface area expansion in the left orbital frontal sulcus. As a critical node within the Salience Network, the expansion of the orbital cortex likely corresponds to an enlargement of the cortical projection zones of the Anterior Thalamic Radiation. This structural change may reflect an enhanced processing capacity for thalamic arousal signals within this region⁸⁶. Interestingly, while alerting efficiency was linked to surface area expansion in the left orbital frontal sulcus, the widening of this same region predicted gains in the Orienting network. This morphometric dissociation suggests a multi-dimensional remodeling process: adjusted surface area expansion may support the metabolic demand for maintaining tonic alertness by increasing the number of functional cortical columns^54,87, whereas sulcal widening—consistent with our transcriptomic evidence linking increased width to synaptic signaling efficiency—might optimize the rapid neural transmission required for spatial reorientation⁸⁰.

Limitations

In contrast to traditional vertex-wise analyses, the sulcal morphometry strategy employed in this study is anchored to topologically invariant anatomical units⁸⁸, thereby promoting the biological homology of the observed developmental trajectories. However, several limitations warrant consideration when interpreting these results. First, despite the substantial total sample size, the longitudinal sampling density was limited, with the majority of subjects contributing only 2-3 time points (>75%). This constrains our capacity to fully characterize complex, non-linear developmental trajectories. Second, the observed effect sizes were modest. While this is consistent with typical patterns in large-scale neuroimaging association studies, it warrants caution regarding the immediate translation of these findings to clinical contexts. Third, the metrics employed to define the timing of sulcal emergence (i.e., T₁₀, T₅₀, and T_app) were derived from heterogeneous literature sources and were largely based on visual inspection standards or cross-sectional inferences. Crucially, these heuristic thresholds currently lack rigorous biological validation regarding developmental onset and have not been verified through individual-level longitudinal tracking. Consequently, they may represent statistical approximations rather than precise anatomical milestones, highlighting the need for future studies to establish more physiologically grounded standards. Finally, the gene enrichment analysis relied on the adult AHBA atlas. Although the spatial topology of cortical gene expression is relatively conserved postnatally⁸⁹, developmental-specific dynamic expression patterns may be overlooked. Future studies should integrate pediatric transcriptomic data and Diffusion Tensor Imaging to further validate the direct links between sulcal morphological remodeling, white matter microstructure, and specific gene expression profiles.

Conclusion

In summary, by integrating large-scale longitudinal neuroimaging, cognitive-behavioral data, and transcriptomics, this study systematically characterizes the developmental patterns of sulcal morphometry from childhood to adolescence at the whole-brain level. Our results indicate that sulcal morphometric remodeling is characterized by cortical thinning and sulcal widening, consistent with ongoing structural refinement of the cortex. Furthermore, the magnitude and consistency of these changes appear to be associated with sulcal formation order, with early-forming primary folds exhibit robust developmental trajectories. Furthermore, we revealed an intrinsic link between these macroscopic structural changes and the expression of genes related to synaptic signaling and the cytoskeleton. We also confirmed that specific sulcal morphometric features serve as potential morphometric signatures for predicting gains in higher-order cognitive functions, such as working memory and attention control. These findings provide biological insights into structural brain maturation during typical development.

Methods

Participants

Structural MR images were collected from the first batch of Children School Functions and Brain Development Project in China (CBD, Beijing Cohort), which includes a longitudinal dataset of 360 participants (163 females and 197 males). After performing comprehensive data quality control procedures to exclude participants with cognitive abnormalities(history of neurological disorders, mental disorders, head injuries, or physical illnesses)⁹⁰, the final sample comprised 307 typically developing children aged between 6.1 and 13.9 years (141 females, 166 males, see Supplementary Fig. 1). For longitudinal participants, the inter-scan interval was approximately one year. Specifically, three MRI scans were available for 43 children (30 females and 13 males), two MRI scans were available for 97 children (47 females and 50 males), and one MRI scan was available for 167 children (64 females and 103 males), resulting in a total of 490 MRI scans. All the children were recruited from primary schools in Beijing based on their performance on a standardized cognitive ability test administered in Chinese. Written informed consent was obtained from each child’s parent or guardian, and the study protocol was approved by the Ethics Committee of Beijing Normal University⁹¹. All ethical regulations relevant to human research participants were followed.

Data acquisition

Image acquisition

High-resolution structural MR data were acquired with 3 T Siemens Prisma scanners at Peking University, Beijing, China⁹¹. For each subject, T1-weighted structural images were acquired using a MPRAGE sequence with the following parameters: repetition time (TR) = 2530 ms; echo time (TE) = 2.98 ms; inversion time (TI) = 1100 ms; flip angle (FA) = 7°; field of view (FOV) = 256 × 224 mm²; number of slices = 192; slice thickness = 1 mm; bandwidth (BW) = 240 Hz/Px; and scan time = 5 min and 58 s.

Cognitive assessment

The participants’ cognitive performance was assessed via the classic numerical N-back WM task and a child-friendly version of the ANT⁹².

In the WM task, participants completed 12 blocks of tasks under three different workload conditions: 0-back, 1-back, and 2-back. In the 0-back condition, participants were instructed to identify whether the current digit was the number 1. In the 1-back and 2-back conditions, participants had to determine whether the current digit matched the one presented one or two steps earlier in the sequence, respectively. The final WM dataset included 402 scans (212 females, 190 males, see Supplementary Figure 1). The metric resid2-back was computed by regressing 2-back hits on 0-back hits.

Additionally, ANT performance was measured across 411 scans (213 females, 198 males, see Supplementary Fig. 1). Network efficiency scores were calculated using orthogonal subtractions of mean response times (RT) across conditions. Specifically, the Alerting effect was calculated as $R{T}_{no-cue}-R{T}_{double-cue}$, representing the benefit of a temporal warning cue. The Orienting effect was calculated as $R{T}_{center-cue}-R{T}_{spatial-cue}$, representing the benefit of a spatial orienting cue. The Executive Control effect was calculated as $R{T}_{incongruent}-R{T}_{congruent}$, representing the performance cost induced by flanker conflict. Note that for Alerting and Orienting, higher scores indicate greater efficiency (better performance), whereas for Executive Control, a higher score indicates a larger executive control cost (worse performance).

MRI processing

As illustrated in Fig. 1a, all T1-weighted images were preprocessed using the HCP pipeline⁵¹, which is primarily based on FreeSurfer (version 6.0, http://surfer.nmr.mgh.harvard.edu/)⁵⁰. This preprocessing included brain extraction, gray and white matter segmentation, and cortical surface reconstruction. Subsequently, the HCP pipeline resampled and registered the reconstructed cortical surfaces to a template. The resulting segmented cortical structures and deformation field information were then directly imported into the Morphologist toolbox integrated within BrainVISA (http://brainvisa.info)^48,49. This toolbox utilizes the Statistical Parametric Anatomy model (SPAM) to automatically identify cortical sulci, employs supervised learning to label the sulci, and quantitatively measures the morphometry of the sulci.^1,48,49,52.

Sulcal selection and quality control

To ensure the reliability of the statistical results, the following quality control steps were implemented: (1) All extracted brain sulci were manually inspected to confirm the absence of extraction artifacts, including partial regional interruptions and missing sulcal segments. (2) Based on established anatomical references and MRI-based sulcal definitions^9–13,16, certain sulcal branches were systematically merged to create standardized sulcal segments, resulting in the initial selection of 56 sulci (28 per hemisphere, Supplementary Fig. 2). (3) Sulci with extraction success rates below 75% were excluded to maintain adequate sample sizes for subsequent statistical analyses. Following this quality control procedure, three sulci (F.I.P.r.int.2_right, S.Pa.t._left, and S.F.orbitaire._right, Supplementary Fig. 3) were excluded due to insufficient extraction success rates, yielding a final dataset of 53 whole-brain sulci (Fig. 1b, Supplementary Table 1). For each included sulcus, six morphological metrics were computed in native space using built-in functions of the Morphologist toolbox: cortical thickness (mm), sulcal width, surface area (mm²), maximum depth (mm), mean depth (mm), and sulcus length (mm)⁹³. Surface area was obtained by calculating the mesh area of sulcal regions. Cortical thickness was measured as the distance from the gray matter/white matter interface to the brain surface⁹⁴. Mean depth was computed as the average distance from all points within the sulcus to their corresponding surface points. Maximum depth was determined by identifying the deepest point within each sulcal region and calculating its distance to the brain surface. Sulcus length was defined as the length of the junction line between the sulcus and the external brain hull. Sulcal width was calculated as a proxy measure representing the ratio between CSF volume and surface area.

Definition of sulcal emergence timing

To characterize the chronological hierarchy of cortical folding, we defined sulcal emergence using three complementary metrics that integrate classical anatomical consensus with modern computational modeling (individual values for all sulci are listed in Supplementary Table 1). We first defined the Sulcal Appearance Time (T_app) as the gestational age (in weeks) at which a sulcus first becomes visibly identifiable on the cortical surface. Values for T_app were collated from a comprehensive review of established anatomical literature, including fetal autopsy studies¹⁰ and in utero MRI assessments^9,11,12, serving as a benchmark for the stage where primitive sulcal indentations become macroscopically consistent across the population. To complement these discrete visual observations, we utilized continuous quantitative milestones derived from the trajectory of surface curvature changes, as formally described by Snyder et al.⁸. In this framework, the maturation of mean curvature for each sulcus was modeled using a logistic growth function across gestation to extract two key physiological timepoints: the onset of folding (T₁₀), defined as the gestational age where local mean curvature reaches 10% of its maximum projected adult value, and the midpoint of folding (T₅₀), defined as the gestational age corresponding to 50% of curvature maturation. While T₁₀ serves as a quantitative proxy for the biological initiation of folding (often detecting subtle surface changes prior to stable visual identification). T₅₀ represents the mathematical inflection point of maximum folding acceleration, capturing the most dynamic phase of sulcal deepening.

Modeling the development of sulcal morphometry

All statistical analyses were conducted using R version 4.4.0 (https://www.r-project.org/). Prior to statistical modeling, comprehensive data preprocessing was performed to ensure data quality and reliability. Outliers were identified and removed using the median absolute deviation (MAD) method, where data points falling beyond ±3 MAD from the region-specific median were excluded. To control for individual differences in brain size, selected morphometric features, including SA, SL, maxD and meanD, underwent TIV adjustment. We employed robust regression using the MM-estimator method to derive TIV-adjusted metrics for sulcal width and cortical thickness, which exhibited little or no correlation with TIV, native values were analyzed without correction.

We employed GAMMs to model the developmental trajectories of sulcal morphometric features across the age range⁴⁵. This approach allows for the modeling of non-linear age-related changes while accounting for the hierarchical structure of longitudinal neuroimaging data. For each combination of sulcal region and morphometric feature, the GAMM was specified as:

$$Y_{ij}={\beta }_0+{\beta }_1\cdot Se{x}_i+f\left(Ag{e}_{ij}\right)+u_i+{\epsilon }_{ij}$$

where$Y_{ij}$ represents the morphometric measure for individual i at timepoint j, β₁ is the fixed effect of sex, $f\left(Ag{e}_{ij}\right)$ is a non-parametric smooth function of age modeled using thin-plate regression splines (k = 3), $u_i~N\left(0,{\sigma }_u^2\right)$ is the random intercept for individual i, and ${\epsilon }_{ij}~N\left(0,{\sigma }_e^2\right)$ is the residual error. Model parameters were estimated using restricted maximum likelihood (REML) to obtain unbiased variance component estimates.

Model convergence was assessed immediately after model fitting by examining the availability of asymptotic parameter variance estimates. Models failing to converge were excluded from statistical testing. To assess the significance of age-related developmental changes, we employed a nested model comparison approach. For each morphometric feature and sulcal region, we compared a full model (including both sex and age effects) against a null model (sex effects only). The contribution of age to model fit was quantified using the difference in adjusted R² (∆adjR²) between models. Since GAMM smooth terms do not possess a single coefficient sign, the direction of the age effect (increase/decrease) was determined by the slope of a supplementary linear model ($feature~Age+Sex$). Statistical significance was evaluated using both the smooth age term within the full model and likelihood ratio tests comparing full vs. null models, with only models showing significance in both tests (p < 0.05) considered to demonstrate robust age effects. For sex effects, significance was assessed based on the fixed effect term (${\beta }_1$) in the full model (Wald t-test).

FDR correction was applied separately to three families of p-values to control for multiple comparisons: (1) smooth term p-values for age, (2) likelihood ratio test p-values for model comparison, and (3) fixed effect p-values for sex ($p_{fdr}$ < 0.05). Following the completion of modeling, we conducted correlation analyses between sulcal emergence timing and age-related effect sizes across different sulci. Multiple comparisons correction was performed using the Benjamini-Hochberg procedure to control the false discovery rate across all tested sulci ($p_{fdr}$ < 0.05).

Statistical analysis of emergence-development associations

To investigate the relationship between the timing of sulcal emergence ($T_{app}$, $T_{10}$, $T_{50}$) and the magnitude of postnatal remodeling, we performed Spearman correlation analyses across the 53 sulci. To capture both the strength and the directionality of developmental changes, we computed the Signed Δadjusted R² for each sulcus. This metric was calculated by assigning the sign of the age-related change (derived from the t-statistic of the smooth age term in the GAMM) to the effect size (Δadjusted $R^2$). Consequently, a large negative Signed Δadjusted R² indicates a strong age-related decrease, while values near zero indicate structural stability. We tested the hypothesis that earlier emergence times would predict larger developmental magnitudes (i.e., more negative signed Δadjusted R² values for decreasing metrics). All correlations were corrected for multiple comparisons using the FDR method ($p_{fdr}$ < 0.05).

Statistical analysis of brain–behavior associations

We investigated the relationship between sulcal morphometry and four cognitive indices: working memory performance ($resid2-back$) and three attention network components (Alerting, Orienting, and Executive Control). Prior to analysis, all morphometric and cognitive metrics were standardized (Z-score) to facilitate cross-scale comparisons. The statistical analysis proceeded in two stages: first, we examined baseline associations between sulcal morphometry and cognitive function; second, for participants with at least two time points, we quantified the coupling between longitudinal morphometric and cognitive changes. Longitudinal change ($\mathrm{\Delta }$) was defined as the difference between follow-up ($T_2$) and baseline ($T_1$) measurements (using TIV-adjusted metrics where applicable):

$$\mathrm{\Delta }{\text{Metric}}_{\mathrm{i}}={\text{Metric}}_{\mathrm{i},\mathrm{T}2}\left(\text{adj}\right)-{\text{Metric}}_{\mathrm{i},\mathrm{T}1}\left(\mathrm{adj}\right)$$

Baseline brain–behavior associations

Univariate associations were assessed using Pearson correlation coefficients. To account for demographic confounds, we performed partial correlation analyses (R package ppcor) controlling for age and gender. Multiple comparisons were corrected using the FDR method ($p_{fdr}<0.05$).

To identify global modes of covariance between high-dimensional brain morphometry (X) and cognitive performance (Y), we employed Partial Least Squares (PLS) correlation⁴⁶. PLS decomposes the covariance matrix into orthogonal Latent Variables (LVs). We focused on the primary mode of covariation (LV1). Significance was established via permutation testing (1000 permutations; $p<0.05$), and the contribution of specific brain regions was evaluated using bootstrap resampling (1000 iterations), with p-values FDR-corrected for multiple comparisons.

Associations between longitudinal structural and cognitive changes

Covariance analysis of developmental changes

To examine the coupling between structural and cognitive trajectories, we applied statistical protocols analogous to the baseline analysis to the longitudinal change scores ($\mathrm{\Delta }$). Univariate associations were assessed via Pearson and partial correlations. Crucially, partial correlation models for the longitudinal data were adjusted for baseline age, gender, and the inter-scan time interval to strictly isolate developmental effects. Multivariate patterns were explored using PLS correlation on the change matrices ($\mathrm{\Delta }X$ and $\mathrm{\Delta }Y$), adhering to the same permutation (significance) and bootstrapping (feature stability) protocols as the cross-sectional analysis.

Predictive modeling via LASSO regression

To identify specific morphometric features predictive of cognitive gains, we employed LASSO regression⁴⁷. LASSO minimizes the residual sum of squares subject to an $L_1$ penalty, driving coefficients of irrelevant features to zero to prevent overfitting:

$$LASSO:min\beta \sum {\left(y_i-x_i^T\beta \right)}^2+\lambda \Sigma \mid {\beta }_j\mid$$

where λ ≥ 0 is the regularization parameter controlling penalty strength. The L1 penalty term $\lambda \Sigma \mid {\beta }_j\mid$ serves two functions: (1) regularization to prevent overfitting by constraining coefficient magnitude, and (2) feature selection by driving irrelevant coefficients to zero³³.

All predictive analyses were implemented using scikit-learn⁹⁵. The dataset was randomly partitioned into training (80%) and independent testing (20%) sets. Features were standardized (Z-score) based on training set statistics to prevent data leakage. The regularization parameter $\mathrm{\lambda }$ was optimized via 5-fold cross-validation on the training set, performing a grid search across 50 logarithmically spaced values (${10}^{-5}$ to ${10}^0$) to minimize the cross-validated Root Mean Squared Error (RMSE).

Model evaluation

The optimized LASSO model was evaluated on the held-out test set using Spearman’s correlation ($r$) between predicted and observed cognitive changes. To explicitly validate the benefit of feature selection, we performed a nested comparison against a “Full Model” (standard linear regression utilizing all predictors without regularization). This comparison assessed whether the sparse LASSO solution offered superior generalizability compared to a model retaining full complexity. Statistical significance for predictive performance was established using FDR correction ($p_{fdr}<0.05$).

Gene enrichment analysis

As illustrated in Fig. 5a, the genomic expression data were sourced from the Allen Human Brain Atlas⁵⁹ (AHBA, https://human.brain-map.org), a comprehensive whole-brain transcriptome atlas derived from microarray expression data. This dataset includes information on over 20,000 genes obtained from 3,702 tissue samples taken from distinct locations across the brains of six neurotypical adults (three white males, two African American males, and one white female; mean age = 42.5 years). Although derived from adults, the spatial topography of gene expression gradients is established early in development and remains relatively conserved into adulthood^89,96. The standard probabilistic sulcal atlas provided by the Morphologist toolbox was used to define sulcal regions through a two-step procedure. Step 1: The original four-dimensional probabilistic atlas was converted to a three-dimensional discrete label atlas using a probability threshold approach. For each voxel, the sulcal label with the highest probability was assigned, provided this maximum probability exceeded a threshold of 0.2 to define “core” sulcal regions and exclude ambiguous boundaries. Step 2: Based on anatomical considerations and research objectives, selected sulcal regions were systematically merged according to predetermined grouping criteria. Regions deemed irrelevant to the analysis were excluded from the final atlas, ultimately yielding 53 whole-brain regions with their Montreal Neurological Institute (MNI) coordinates.

The resulting 3D NIFTI file was processed using the abagen toolbox⁵⁸ (version 0.1.3; https://github.com/rmarkello/abagen) to generate regional gene expression profiles. A rigorous filtering process was applied to explicitly exclude probes not belonging to the target sulcal regions. Specifically, out of 3,702 probes initially mapped to the vicinity of our regions of interest, 2821 (approximately 76.2%) were excluded as they fell onto the background or gyral regions. Consequently, 881 probes located directly within the sulci were retained for analysis, ensuring that the expression profiles accurately reflected sulcal folding geometry rather than gyral. This procedure ultimately yielded a final 53 × 15,633 (region × gene) matrix, representing the expression levels of 15,633 genes across the 53 atlas regions (Supplementary Methods 1).

For each gene, the Pearson correlation coefficient and its corresponding p-value were calculated between the gene expression data and the age-related morphometric change effect sizes ($\mathrm{\Delta }\text{adj}{R}^2$). To account for spatial autocorrelation in brain data, a distance matrix was generated based on the Euclidean coordinates of each brain region using the BrainSmash Python tool⁶⁰ (version 0.11.0, https://github.com/murraylab/brainsmash). For each gene, we performed permutation testing by generating 1000 spatially-constrained null maps that preserved the spatial autocorrelation structure of the original expression data while randomizing gene expression values across the 53 brain regions. The observed correlation coefficient between gene expression and age-related morphometric changes (Δadj R²) was then compared against the null distribution derived from these 1000 permutations to calculate spatial autocorrelation-corrected p-values. Ultimately, we identified a set of genes that exhibited statistically significant correlations (p < 0.05) between expression levels and age-related morphometric changes across the 53 brain regions after correcting for spatial autocorrelation effects using this 1000-permutation approach. These spatially-validated genes were then subjected to functional enrichment analysis via Metascape⁶¹ (https://metascape.org). Both Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses were conducted to identify enriched biological processes, molecular functions, cellular components, and metabolic pathways^97–99. Metascape’s comprehensive analysis pipeline provided hierarchical clustering of enriched terms, protein-protein interaction network analysis, and identification of representative terms for each gene cluster to elucidate the biological significance of the identified gene sets⁹⁸.

Statistics and reproducibility

Sample size and data exclusion

The final sample of 307 participants (N = 490 scans) was determined by the available high-quality data from the CBD Beijing Cohort. Outliers were strictly removed using the MAD method with a threshold of $\pm 3$. Sulcal parcels with an automated extraction success rate below 75% were excluded to ensure the reliability of downstream morphological quantification.

Developmental modeling

Age-related changes were modeled using GAMMs with thin-plate regression splines (k = 3) to capture non-linear trajectories. To control for global brain growth, SA, maxD, meanD, and SL were adjusted for TIV using robust MM-estimator regression. Statistical significance was defined as passing both the smooth term Wald test and the likelihood ratio test (Full vs. Null model) at p < 0.05 after FDR correction across all sulci and features.

Predictive model stability

To ensure the robustness of the LASSO regression in predicting cognitive gains, we employed a nested 5-fold cross-validation on the training set (80%) to optimize the regularization parameter $\mathrm{\lambda }$. The final model performance was evaluated on a completely independent test set (20%). To assess whether the sparse LASSO solution outperformed a diffuse global model, we compared its predictive power against a Ridge regression and a “Full” linear regression model. Significance for the prediction was established through Spearman correlation between predicted and observed scores, with p-values calculated via 10,000 permutations.

Reproducibility of gene enrichment

Gene expression associations were validated using 1000 spatially-constrained null maps generated by BrainSmash to account for spatial autocorrelation in cortical data. This ensures that identified gene-morphometry correlations are not simply products of spatial proximity. The source data underlying all main figures (Figs. 2, 3, 4, and 5) and supplementary tables are provided as a Source Data file or have been deposited in the Github database under accession code (https://github.com/yijinshan1/sulcus2025).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Conceptualization: Y.S., Y.H., S.L.; Investigation: Y.S.; Formal analysis: Y.S.; Methodology: Y.S., S.L.; Visualization: Y.S.; Writing—original draft: Y.S.; Writing—review and editing: H.Q., Y.R.H., L.C., X.X.D., S.L.; Data curation: D.Z., T.Z., X.L., X.C., Y.X., T.L., L.S., W.M., Y.W., D.W., M.H., Z.P., S.T., J.G., S.Q., S.T., Q.D., R.C., L.M., X.R.; Supervision: Y.H., S.L.; Funding Acquisition: Q.D., Y.H., S.L.

Peer review

Peer review information

Communications Biology thanks Federica Santacroce and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary handling editors: Sahar Ahmad and Jasmine Pan. A peer review file is available.

Data availability

The raw structural MRI data are not publicly available due to institutional restrictions regarding participant privacy and the protection of minors, but can be requested from the corresponding authors (S.L. or Y.H.) upon reasonable academic request. Significant sulcal morphometric developmental models are available in Supplementary Data 1. Processed cognitive scores used for behavioral analyses and generating figures are available at: https://github.com/yijinshan1/sulcus2025.

Code availability

The customized R and Python scripts used for GAMM modeling, PLS correlation, LASSO prediction, and gene enrichment analysis are available at: https://github.com/yijinshan1/sulcus2025. Standard toolboxes used in this study include BrainVISA (http://brainvisa.info), abagen (https://github.com/rmarkello/abagen), and BrainSmash (https://github.com/murraylab/brainsmash).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s42003-026-09956-6.