Abstract
From infancy through adulthood, the human brain undergoes profound structural and functional maturation that supports the development of complex cognitive, social, and behavioral abilities. The advent of multi-modal neuroimaging techniques has enabled non-invasive mapping of the developing structural and functional connectivity, namely the developmental connectome. Recent advances in large-scale, high-resolution, and multi-site neuroimaging have ushered developmental connectomics into the era of big data. This shift is characterized by large sample sizes, both longitudinal and cross-sectional designs, and the integration of cognitive, biological, and environmental measures. These data-rich resources have not only overcome previous limitations but also expanded upon earlier findings, advancing the field beyond descriptive observations towards mechanistic insights into brain development. In this review, we highlight recent advances in developmental connectomics from the prenatal period to early adulthood, with a focus on the big data perspective enabled by multi-modal magnetic resonance imaging. We first introduce major large-scale neuroimaging datasets that provide comprehensive, multi-dimensional data on brain development. Next, we review cutting-edge connectome-based approaches, including graph-based and network communication models, along with statistical methods such as growth curve modeling and multivariate analysis. Finally, we summarize key findings on developmental principles, derived from both prior studies and recent large-scale efforts, and their associations with cognitive, behavioral, and genetic factors, and outline emerging challenges and future directions in the field.
Introduction
Human brain development is a complex, prolonged, and dynamic process that unfolds from the prenatal stages through adulthood (Vértes & Bullmore, 2015). It begins in the second trimester of gestation with rapid neuronal proliferation and axonal growth (Kostovic et al., 2019; Webb et al., 2001), followed by extensive synaptic pruning and axonal myelination from infancy to adolescence (Tau & Peterson, 2010). These neurodevelopmental processes are intricately regulated by interactions between genetic and environmental factors (Gao et al., 2019) and occur in parallel with the emergence of cognitive and behavioral functions (Cao et al., 2017b; Gilmore et al., 2018). Notably, this prolonged developmental trajectory also gives rise to critical windows of vulnerability (Patel et al., 2021). For example, the prenatal period is characterized by heightened susceptibility to environmental insults and increased risk for neurodevelopmental disorders, whereas adolescence is marked by a peak onset of many psychiatric disorders (Paus et al., 2008). Deviations from typical developmental trajectories have been linked to increased risk for disorders such as autism spectrum disorder, attention-deficit/hyperactivity disorder, and schizophrenia (Bu et al., 2021; Duan & Chen, 2022; Sun & Xia, 2025), highlighting the importance of understanding the normative principles of brain development.
The emerging field of “developmental connectomics” has provided a novel framework for understanding brain maturation from a network perspective (Cao et al., 2017b). Advanced neuroimaging techniques, including multi-modal magnetic resonance imaging (MRI), functional near-infrared spectroscopy (fNIRS), electroencephalography (EEG), and magnetoencephalography (MEG), allow for non-invasive mapping of brain structure and function (Hagiwara et al., 2025; Huang & Roberts, 2021). By integrating these techniques with network science, researchers can construct structural and functional brain networks and explore how these networks organize and reconfigure across development (Cao et al., 2017b). Specifically, structural and diffusion MRI can be used to construct structural networks based on cortical morphological similarity and white matter tractography, while functional MRI, fNIRS, EEG, and MEG can be used to construct functional networks by measuring the statistical dependence of hemodynamic or neural activity between brain regions. Beyond capturing developmental changes in connectivity, developmental connectomics also elucidates how brain network development relates to individual differences in cognition and behavior. These advances have greatly enhanced our understanding of both typical and atypical brain development, shedding light on the neural substrates of cognitive and behavioral maturation (Uddin, 2021) and the pathophysiology of neuropsychiatric disorders (Sydnor et al., 2021).
Over the past decade, developmental connectomics has witnessed a rapid expansion in both the scale and richness of neuroimaging and behavioral datasets, enabling multi-scale investigations of typical and atypical brain development. This progress is marked by key advances: (1) larger sample sizes (
Building upon prior reviews (Cao et al., 2014b, 2016, 2017b; Collin & van den Heuvel, 2013; Gilmore et al., 2018; Vértes & Bullmore, 2015; Zhang et al., 2019; Zhao et al., 2019c), this review primarily focuses on recent progress in developmental connectomics using MRI modalities, given that most large-scale publicly available datasets are derived from multi-modal MRI (Figure 1). Distinct from earlier reviews, we further integrate insights from emerging large-scale, data-driven investigations to provide a more comprehensive understanding of brain network development. We begin by introducing major publicly available neuroimaging datasets, then outline recent advances in network analysis approaches. We then summarize typical developmental trajectories and their associations with genetic and neurocognitive factors based on findings from both small- and large-scale studies, with particular emphasis on novel insights enabled by big data. Finally, we discuss emerging challenges and outline directions for future research.
Schematic illustration of the developmental connectomics framework based on neuroimaging big data, including brain network construction, connectome-based modeling, growth curve modeling, and associations with behavior and genetics. (A) Brain network construction. The morphological covariance network was generated by calculating the similarity of morphometric features, such as cortical thickness and gray matter volume, based on structural MRI. The white matter network was derived via white matter fiber tractography based on diffusion MRI. The functional network was obtained by calculating Pearson's correlation coefficients between the time series of all pairs of regions based on functional MRI. (B) Connectome-based modeling. For network segregation, the clustering coefficient quantifies the tendency of local aggregation of a network. For example, the neighboring nodes of node
Gantt chart of the developmental processes and corresponding segregation and integration of the brain network during the first two decades. The color bar intensity reflects the extent of the network segregation and integration. Triangles, circles, and squares indicate the early presence of small world properties, modular architecture, and rich club organization, respectively. Notably, the timing of emergence shown here is based on reports to date and does not represent the exact timing, as earlier prenatal imaging data are unavailable. This schematic provides an overview based on current findings (Ball et al., 2014; Baum et al., 2017; Cao et al., 2017a; Fan et al., 2021; Fan et al., 2011; Fenchel et al., 2020; Feng et al., 2023; Gao et al., 2011; Gu et al., 2015; Huang et al., 2015; Khundrakpam et al., 2013; Li et al., 2024a; Li et al., 2024b; Meng et al., 2015; Song et al., 2017; Thomason et al., 2014; van den Heuvel et al., 2015; Wang et al., 2024; Yin et al., 2025; Zhao et al. 2019a). Source: Adapted with permission from Zhao et al. (2019c).
Large-Scale Imaging Datasets for Developmental Connectomics
Traditional neuroimaging studies of brain network development have been constrained by limited imaging quality, small sample sizes, and a scarcity of longitudinal data. These challenges are particularly pronounced in early childhood, where acquiring high-quality neuroimaging data, especially from preschool-aged children, remains technically difficult (Turesky et al., 2021). Recent advances in imaging technologies and the emergence of large-scale neuroimaging initiatives have begun to overcome these barriers. Improvements in imaging hardware and acquisition protocols have significantly enhanced image quality and reliability. For example, the Developing Human Connectome Project developed a neonatal brain imaging system (NBIS) featuring a 32-channel neonatal head coil and custom positioning devices, enabling higher signal-to-noise ratios and improved image quality (Hughes et al., 2017). In addition, novel sequences like multi-band echo-planar imaging have increased spatial and temporal resolution in MRI acquisition (Feinberg et al., 2010; Makropoulos et al., 2018). These advancements have laid the technical foundation for more precise and reproducible characterization of early brain maturation, paving the way for data-driven mapping of developmental trajectories.
Several large-scale neuroimaging initiatives have collected multi-modal data, including structural MRI, diffusion MRI, and resting state- and/or task-based functional MRI, from individuals aged 0 to 20 years. These imaging datasets are enriched with diverse developmental measures, such as demographic, behavioral, cognitive, medical, and genomic information, enabling comprehensive investigations of brain maturation and its genetic and environmental underpinnings.
In the perinatal period, the dHCP project captures rapid brain development from 28 to 44 gestational weeks in 783 neonates and 273 fetuses using specialized neonatal MRI systems (Edwards et al., 2022). During infancy and early childhood, the BCP project provides densely sampled longitudinal neuroimaging and behavioral assessments from 358 infants aged 0–5 years (Howell et al., 2019). The Growing Up in Singapore Towards Healthy Outcomes (GUSTO) study offers a comprehensive Asian dataset that links prenatal and early childhood conditions to later development outcomes (Soh et al., 2014). Bridging infancy and middle childhood, the Healthy Brain and Child Development Study (HBCD) is following 7,500 mother-child dyads from the prenatal period through early childhood to chart typical brain and behavioral trajectories and identify early-life influences on development (Nelson et al., 2024).
From school-age through adolescence, several large-scale cohorts with overlapping age ranges and complementary research aims provide valuable resources for cross-cohort validation and replication. The Philadelphia Neurodevelopmental Cohort (PNC) integrates neuroimaging with cognitive and psychopathological data from 1,445 youths aged 8–21 year (Satterthwaite et al., 2014). The HCP-D employs standardized imaging protocols to chart normative brain development in individuals aged 5–21 years (Somerville et al., 2018). The IMAGEN study, the first large-scale European imaging-genetics cohort, investigates the neural basis of traits such as impulsivity in approximately 2,000 adolescents (Schumann et al., 2010). Notably, the Adolescent Brain Cognitive Development (ABCD) study, the largest longitudinal cohort to date, is tracking nearly 12,000 U.S. children from age 9 onward, incorporating mobile and wearable technologies to study substance use and environmental influences (Casey et al., 2018). In China, emerging efforts offer culturally specific perspective: the developing Chinese Color Nest Project (devCCNP) aims to construct normative neurodevelopmental atlases for Asian populations (Liu et al., 2021), while the Children Brain Development Project (CBDP) explores the relationship between school performance and brain maturation in 770 school-aged children (Zhao et al., 2019b).
Collectively, these large-sample, high-quality, and multi-dimensional neuroimaging datasets provide a critical foundation for charting developmental trajectories of brain structure and function, as well as for elucidating their relationship with cognition and behavior. The integration of neuroimaging with genetic, environmental, cognitive, and behavioral data offers a comprehensive framework for understanding brain-behavior relationships and their potential molecular basis. Importantly, open access to these large-scale datasets promotes research transparency, reproducibility, and cross-cohort synergy, accelerating progress in developmental neuroscience. A brief overview of these representative datasets is presented in Table 1 and the Supplement.
Representative datasets for developmental connectomics.
Note: dHCP, Developing Human Connectome Project (Edwards et al., 2022); BCP, Baby Connectome Project (Howell et al., 2019); GUSTO, Growing Up in Singapore Towards healthy Outcomes (Soh et al., 2014); HBCD, HEALthy Brain and Child Development Study (Nelson et al., 2024); HCP-D, Lifespan Human Connectome Project in Development (Somerville et al., 2018); PNC, Philadelphia Neurodevelopmental Cohort (Satterthwaite et al., 2016); devCCNP, developing Chinese Color Nest Project (Fan et al., 2023); CBDP, Children School Functions and Brain Development Project (Zhao et al., 2019b) ; ABCD, Adolescent Brain Cognitive Development Study (Casey et al., 2018); IMAGEN, Imaging Genetics in Adolescents (Schumann et al., 2010).
Computational Models and Statistical Methods in Developmental Connectomics
The shift toward a big data perspective in developmental connectomics has enabled the integration of multi-modal, multi-dimensional data with diverse analysis approaches, deepening our understanding of brain development and its functional significance. In this section, we introduce commonly used connectome-based models and statistical methods employed to characterize developmental trajectories and their association with behavior.
Connectome-Based Models
The introduction of the human brain “connectome” concept in 2005 has spurred widespread use of connectome modeling to investigate the organization principles of brain structure and function from a network perspective (Sporns et al., 2005). By applying network neuroscience techniques to neuroimaging data, the human brain can be modeled as a macroscale network comprising brain nodes (e.g., brain regions) and edges (e.g., white matter tracts, morphological similarity, or functional correlations). This framework enables the examination of age-related changes in network topology and interregional information communication across development.
Graph Theory-Based Models
Graph theory-based complex network analysis is widely used to characterize the topological organization of structural and functional brain networks (Bullmore & Bassett, 2011; Bullmore & Sporns, 2009; He & Evans, 2010; Liao et al., 2017). By analyzing interregional connections, researchers can infer how brain regions interact and how information is processed and transmitted across the brain.
Functional segregation and integration are two fundamental principles of brain network organization (Rubinov & Sporns, 2010; Sporns, 2013). Segregation refers to the specialization of functions within densely connected clusters or communities, typically measured by metrics such as the clustering coefficient, local efficiency of nodes, and modularity. Integration reflects the brain's capacity for efficient communication across spatially distributed regions, measured by metrics like characteristic path length, global efficiency, or intermodular connectivity. A well-balanced trade-off between segregation and integration yields a small-world architecture, which combines high local clustering and short characteristic path lengths, allowing for efficient information processing and global communication (Liao et al., 2017).
At the regional level, nodal degree quantifies the number of direct connections, while other nodal centrality metrics, such as nodal efficiency, betweenness centrality, and participation coefficient, capture different aspects of nodal roles in network integration (van den Heuvel & Sporns, 2011). Nodes with high centrality are typically defined as network hubs, which can be further classified into provincial hubs or connector hubs depending on their connectivity profiles within and between modules. Rich-club organization refers to a subnetwork of hub regions that are more densely interconnected with each other than would be expected by chance. Connections associated with these hubs can be classified into three types: rich-club links (between hub nodes), feeder links (between hub and non-hub nodes), and local links (between non-hub nodes). Each type plays distinct roles in supporting global communication and network integration.
Notably, the choice of brain parcellation scheme plays a crucial role in deriving reliable and interpretable network metrics, as node definitions directly affect network resolution and comparability across studies (Zalesky et al., 2010; Zhao et al., 2015b). This issue is particularly pronounced in developmental research, where age-specific structural and functional atlases for infants and children remain limited. However, recent advances in neuroimaging technology and computational power have enabled high-resolution, voxel- or vertex-level network construction (∼10,000 nodes), allowing for finer-grained analysis without relying on predefined parcellations.
Network Communication Models
Brain structure serves as a foundational substrate supporting the functional organization (Seguin et al., 2023b). Structure-function coupling has been examined to elucidate how brain function is sculpted by the underlying structure during development, typically assessed as Pearson's correlations between connectivity profiles at both global and nodal levels (Fotiadis et al., 2024). More recently, network communication models have been introduced to specify how information is transmitted across structural networks to support functional coordination, providing novel insights into the structure-function relationship (Avena-Koenigsberger et al., 2018; Seguin et al., 2023a, 2023b). Extended from the correlation analysis between structural and functional connectivity profiles, these models focus on communication dynamics, identifying potential strategies and key pathways for information transmission, especially between brain regions without direct anatomical connections.
Several models have been proposed to characterize how structural connectivity supports function interactions, each based on distinct assumptions about transmission strategies. The shortest path routing model assumes that information travels along the most efficient path between two brain regions. This model underlies commonly used metrics such as shortest path length and global efficiency, which quantify the ease of communication across all possible node pairs in the network (Rubinov & Sporns, 2010). In contrast, navigation communication models were based on local optimization rules, where each step in the communication path is chosen to minimize the spatial distance to the target region using only locally available information (Seguin et al., 2023b). The corresponding measure, navigation efficiency, reflects how effectively signals can be routed based on such decentralized strategies. Communicability models take a different approach, conceptualizing signal transmission as a diffusion-like process, whereby information can spread along multiple possible paths or via random walks. The communicability measure captures the extent to which two nodes can influence each other through all possible walks in the network (Masuda et al., 2017).
Collectively, these models offer comprehensive perspectives on how structural connectivity supports functional coordination (Avena-Koenigsberger et al., 2018; Seguin et al., 2023b), each highlighting different aspects of the structural-functional coupling.
Statistical Methods
In the context of age-dependent connectome-based modeling, statistical methods are essential for characterizing developmental trajectories of brain network properties and for examining the associations with cognitive and behavioral outcomes. In this section, we review several statistical methods that are commonly used in developmental connectomics, with an emphasis on methods suited for large-scale, multi-site neuroimaging data.
Growth Curve Modeling
In pediatric clinical practice, growth charts for basic physical characteristics such as height, weight, and head circumference serve as fundamental tools. Analogously, modeling the developmental trajectories of brain network is crucial for characterizing both typical and atypical neurodevelopment. Traditional statistical models, including general linear models (GLMs), generalized linear mixed models (GLMMs), and generalized additive mixed models (GAMMs), have been widely used to fit age-related growth curves of connectome features, using both cross-sectional and longitudinal neuroimaging data from single-center dataset (Rigby & Stasinopoulos, 2005; Stasinopoulos et al., 2018; Zuo et al., 2017). With the increasing availability of large-scale, multi-site neuroimaging datasets, it is now possible to estimate more robust and generalizable growth curves at the population level.
The generalized additive model for location, scale, and shape (GAMLSS) provides a flexible statistical framework capable of modeling heteroskedasticity, non-linear effects, and hierarchical structures in data (Stasinopoulos et al., 2017). This method shows high flexibility in capturing temporal changes across different distribution moments, making it well-suited for modeling brain network development using large-scale longitudinal datasets (Bethlehem et al., 2022; Rutherford et al., 2022, 2023; Sun et al., 2025). Of note, normative modeling frameworks, such as GAMLSS-based growth charts, enable the estimation of population-level developmental norms while preserving individual-level variation (Marquand et al., 2016). This dual-level capacity supports the identification of individual deviations from typical developmental trajectories, enabling individualized risk stratification and subtypes classification in neuropsychiatric conditions, such as major depression disorder (Sun et al., 2023) and attention-deficit/hyperactivity disorder (Bu et al., 2024). Thus, normative growth modeling offers a powerful bridge between population neuroscience and personalized medicine, while enriching our understanding of both typical and atypical brain development.
Multivariate Association Analysis
Multivariate analysis aims to uncover complex, multi-dimensional relationships between brain structure, function, and behavior and genetics. Unlike univariate approaches that examine one-to-one associations, multivariate models leverage joint variability across multiple brain features to identify patterns linked to cognitive or behavioral outcomes.
Canonical correlation analysis (CCA) and partial least squares (PLS) are commonly used in developmental neuroimaging to capture these relationships. CCA identifies correlated modes between two sets of variables (Zhuang et al., 2020), while PLS extracts latent components that maximize shared covariance between them (Krishnan et al., 2011). The key distinction is that CCA optimizes correlation, whereas PLS maximizes covariance (Krishnan et al., 2011).
Beyond correlation-based analyses, multivariate predictive models have been increasingly applied to predict cognitive or behavioral outcomes from neuroimaging data (Scheinost et al., 2023). These models often incorporate machine learning algorithms, such as support vector regression, ridge regression, or elastic-net algorithm, to predict individual-level outcomes based on multiple brain features. The typical workflow includes: (1) feature selection to identify brain measures that are correlated with outcomes of interest, (2) model training using these features as input and cognitive/behavioral scores as outputs, and (3) application of the trained model to predict outcomes in independent test set. In addition to estimating prediction accuracy, these models can evaluate the relative contribution of specific brain features, thereby shedding light on which neural substrates most strongly influence cognitive or behavioral variation.
To evaluate model generalization, cross-validation techniques, such as
Typical Development of Healthy Brain Connectomics
The human brain undergoes substantial changes in both structural and functional connectomes throughout the first two decades of life, supporting the emergence of increasingly complex cognitive and behavioral functions. Numerous research has applied graph-theoretical analysis to explore the topological properties of morphological, white matter, and functional brain networks (Vértes & Bullmore, 2015) . The structure-function coupling has been quantified with Pearson's correlation analysis and the communication models (Fotiadis et al., 2024).
In this section, we summarize key findings on normative developmental trajectories from both small- and large-scale developmental connectomics studies, especially emphasis studies that leveraging the datasets and analysis approaches mentioned above. We first review age-related changes in the structural connectome, followed by functional connectome, and conclude with insights into the structure-function relationship. These findings highlight how the connectome evolves across developmental stages, from infancy and childhood through adolescence and adulthood (Figure 2). For discussions on neurodevelopmental disorders such as attention-deficit/hyperactive disorder, autistic spectrum disorder, and early-onset schizophrenia, refer to other reviews (Cao et al., 2014a; Duan & Chen, 2022; Sun & Xia, 2025).
Developmental Trajectories of Brain Structural Connectomes
The structural connectome reflects the anatomical wiring between brain regions and comprises two primary components: the morphological brain connectome (MBC), derived from interregional similarities in structural morphology obtained from structural MRI, and white matter connectome (WMC), reconstructed from white matter tracts using diffusion MRI (Hagiwara et al., 2025).
Morphological Brain Connectome
Structural MRI provides high-resolution morphological features, such as cortical thickness, surface area, sulcal depth, and gray matter volume, for constructing morphological networks. Traditional population-level MBCs measures inter-regional covariance of morphological features across individuals (Alexander-Bloch et al., 2013), while more recent methods enable the construction of individualized MBCs, capturing the statistical interdependence of local morphological features between brain regions within a single subject (Wang & He, 2024).
Across both population- and individual-levels, MBCs show a clear trajectory of early topological optimization. In full-term neonates, MBCs already exhibits the small-world topology and modular organization consistent with known functional networks (e.g., sensory-motor, limbic, and association regions) and cytoarchitectonic classes (Fan et al., 2011; Fenchel et al., 2020; Meng et al., 2015; Wang et al., 2024). Based on a large sample of 427 scans of preterm and term infants aged from 30–44 postmenstrual weeks, Wang et al. (2024) found that both network integration and segregation increase before 40 postmenstrual weeks, followed by declines in the network segregation. During the first two years of life, the global and local efficiency, clustering coefficient and modularity increases with age, while the shortest path length decreases with age, which reflect enhanced network segregation and integration (Fan et al., 2011; Meng et al., 2015). In early infancy, the hub regions were relatively consistent located in the sensorimotor cortex, the medial posterior parietal, and temporal regions (Fan et al., 2011). By late childhood and adolescence, these hubs progressively shift toward association cortices, particularly the prefrontal and parietal regions (Khundrakpam et al., 2013; Vijayakumar et al., 2021), reflecting a reorganization from primary to higher-order systems.
Using the large sample dataset from dHCP project, Fenchel et al. (2020) found that the posterior regions became more morphometrically similar with development, whereas the peri-cingulate and medial temporal regions became more divergent around the term age (from 37–44 postmenstrual weeks). The increased nodal connectivity strength is mainly observed in primary sensory cortices (e.g., sensorimotor, auditory, and visual regions), while decreases were primarily found in the cingulate gyrus and limbic cortex (Fenchel et al., 2020; Galdi et al., 2020). Notably, connectivity within primary cortices increases at a faster rate than that of the higher-order association cortical regions (Wang et al., 2024). From early childhood to adolescence, the global efficiency and number of connectors continue to increase, while local clustering progressively declines (Khundrakpam et al., 2013). Meanwhile, network modules becomes increasingly spatially distributed, indicating greater functional differentiation (Li et al., 2024a). Collectively, these findings suggest that MBCs follow a developmental trajectory marked by early backbone emergence followed by gradual refinement, characterized by spatially heterogeneous maturation that prioritizes primary sensory regions early on, and higher-order association areas later.
However, some studies have reported inconsistent findings. For example, from birth to 2 years of old, age-related increases in global efficiency but decreases in local efficiency have been found in MBCs derived from cortical thickness; while opposite trends were observed for MBCs derived from curvedness (Nie et al., 2014). Similarly, from age 3 to 5 years, global efficiency decreases while local efficiency increases in MBCs constructed from either cortical thickness or curvedness (Nie et al., 2013). These variations may be attributed to methodological differences in morphological feature selection and network construction and analysis strategies. To address these limitations, emerging approaches propose constructing of multimodal morphological similarity network that integrate multiple morphological features, offering a more holistic view of interregional developmental coordination (Sebenius et al., 2023; Seidlitz et al., 2018; Wang & He, 2024).
White Matter Connectome
The white matter connectome captures anatomical connectivity by modeling axonal pathways between grey matter regions via diffusion MRI (Zhang et al., 2022). During the perinatal period, adult-like topological architecture, such as small-worldness, modular, and rich club organization have been observed (Brown et al., 2014; Song et al., 2017; Tymofiyeva et al., 2013; van den Heuvel et al., 2015; Zhao et al., 2019a; Zheng et al., 2023), suggesting that the white matter connectome is already highly efficient and specialized around term age. Before birth, the global and local efficiency, clustering coefficient, and modularity increased, while characteristic shortest path length decreased with development (Ball et al., 2014; Batalle et al., 2017; Brown et al., 2014; Tymofiyeva et al., 2013; van den Heuvel et al., 2015). After birth, the white matter connectome continues to reorganize, primarily enhancing integration and long-range communication. With development, global and local efficiency, nodal strength, number of modules and connectors increased linearly, while nodal clustering stable or decreased, modularity, and shortest path length decreased with age (Feng et al., 2023; Hagmann et al., 2010; Huang et al., 2015; Zhao et al., 2015a). Meanwhile, network modules become more internally coherent, showing stronger within-module and weaker inter-module connections (Baum et al., 2017).
At the region level, network hubs in preterm infants are primarily located in the superior and medial frontal, superior parietal, and sensorimotor cortices, which are largely overlap with hubs observed in adults (Ball et al., 2014; Pandit et al., 2014; van den Heuvel et al., 2015; Zhao et al., 2019a). By term age, hub regions further expanded into inferior frontal and insular regions (Ball et al., 2014; Zhao et al., 2019a). During toddlerhood, additional hubs emerge in the left anterior cingulate gyrus and left superior occipital gyrus (Huang et al., 2015). Nodal efficiency in the precuneus and cuneus continues to increase into childhood (Huang et al., 2015).
Of note, recent modeling advances have begun to directly link white matter connectivity with cortical morphogenesis. Leveraging a network-based diffusion model, Liang et al. (2024) demonstrated that nodal diffusion profiles in the white matter connectome can robustly predict the maturation pattern of regional cortical thickness across childhood and adolescence, especially in the lateral frontal and parietal regions. These findings highlight a potential mechanism that connected neighbors in white matter shape the development of cortical morphology. Future studies should integrate high-resolution multimodal datasets and individualized connectome models to elucidate how cortical morphology and white matter connectivity co-develop during the early and childhood stages of brain maturation.
Developmental Trajectories of Brain Functional Connectomes
The functional network, usually constructed from resting-state fMRI by assessing temporal correlations between BOLD signals across regions, offer critical insights into the maturation of functional architecture. Several studies have shown that even during the middle to late gestational period, the functional network has exhibits non-trivial properties, such as hubs, small-worldness, modular, and rich club organization (Cao et al., 2017a; Thomason et al., 2014; Turk et al., 2019), suggesting an early emergence of efficient and specialized architecture.
During prenatal development, functional networks exhibit a strengthening of short-range connections within local primary regions, leading to increased network segregation. Network metrics such as global and local efficiency, clustering coefficient, nodal degree, and small-worldness increase with age (Eyre et al., 2021; Fenn-Moltu et al., 2023), while modularity and characteristic path length decrease (Nazari & Salehi, 2023). After birth, both the global and local efficiency continue to increase, reflecting enhanced network integration. Using large-scale cohorts, two recent studies have further demonstrated the non-linear changes of network topologies during the first 1000 days (or 28 months) of life (Jiang et al., 2023; Li et al., 2024b). Specifically, the functional segregation (i.e., clustering coefficient) exhibits an inverted U-shaped trajectory, while functional integration (i.e., nodal efficiency) follows a U-shaped trajectory. These developmental patterns are spatially heterogeneous, varying along the anterior-posterior axis of the cerebral cortex. Prior research and recent large-scale analyses have consistently revealed a progressive shift in functional hubs from primary to higher-order association cortices (Gao et al., 2011; Li et al., 2024b). At birth, the hub regions are primarily located in the sensorimotor and visual regions (Fransson et al., 2011; Gao et al., 2011). Around two years of age, these hub gradually move towards to the medial superior frontal gyrus (Gao et al., 2011; Li et al., 2024b).
The reconfiguration of modular architecture has attracted great attention in developmental functional connectomics, emphasizing the dynamic reconfiguration between functional specialization and integration. Age-related changes have been observed in the spatial layout of modules, intra- and inter-module interactions, and their alignment with their cortical functional hierarchy (Fan et al., 2021; Gu et al., 2015; Pines et al., 2022). In neonates, adult-like modular patterns are evident within the primary sensory networks (e.g., sensorimotor, visual, and auditory), while association networks remain largely immature (Eyre et al., 2021; Jiang et al., 2023). The default mode and salience networks only approximate adult patterns around 1 years of age (Gao et al., 2015). Leveraging 1,091 resting-state functional MRI scans from typically developing children aged birth to 6 years, Yin et al. (2025) further revealed that rapid maturation occurs in the visual, limbic, and default-mode networks, followed by more prolonged development of the frontoparietal and control networks. During mid-childhood (6 to 14 years), both global and local efficiency within the default-mode network increased with age, indicating improved within-network functional integration and segregation (Fan et al., 2021). From late childhood to adolescence (8 to 22 years), integration between the default-mode network and other functional networks increases with age, while integration involving higher-order cognitive and subcortical networks decreases (Gu et al., 2015). Personalized network analyses further suggest that inter-module interaction develops in a manner that refines the functional hierarchy along the sensorimotor-association axis (Pines et al., 2022). Extending these findings (Fan et al., 2021; Gu et al., 2015; Pines et al., 2022), a large-scale study combing four independent datasets (total n = 3355; ages 5–23 years) demonstrated age-related increases in functional connectivity strength primarily in sensorimotor regions and decreases in association cortices, reinforcing the spatial organization of the functional hierarchy (Luo et al., 2024). These developmental trends are consistent with a gradual reorganization of functional connectivity gradients, shifting from an early anchor in the unimodal cortex to an adult-like pattern characterized by separation between the default-mode and primary networks (Dong et al., 2021; Xia et al., 2022).
Recent studies has further explored the potential functional diversity of brain regions from the perspective of spatial overlap in functional module affiliations (Faskowitz et al., 2020; Lei et al., 2024). In children, the degree of module overlap varies across regions, with higher overlap observed in the ventral attention, somatomotor, and subcortical regions, and lower overlap in the visual and the default-mode regions (Lei et al., 2024). From childhood through adolescence (6–14 years), age-related increases in overlap are mainly located in the lateral prefrontal and parietal cortices while decreases are observed in the dorsomedial and ventral prefrontal cortex and the putamen. Beyond static (i.e., time-constant) measures, recent work has also examined developmental changes in temporal dynamics of functional module switching. During the third trimester, the dynamic switching between the brain modules was significantly decreased, primarily located in the lateral precentral gyrus, medial temporal lobe, and subcortical areas (Xu et al., 2024). From birth to 2 years of age, the variability of whole brain functional connectivity linearly increases (Wen et al., 2020), and the neural flexibility, reflecting functional module switching, also increases with age (Yin et al., 2020). Both age-related changes are prominent in primary and high-order functional networks/regions. During childhood and adolescence, the temporal switching of brain regions among functional modules decreases with age, primarily involving the default-mode, frontoparietal, and somatomotor systems, suggesting a progressive stabilization of brain dynamics (Lei et al., 2022).
Overall, the maturation of functional connectomes is characterized by a hierarchical and spatially heterogeneous progression from local, sensory-dominant architecture to globally integrated association networks (Gao, 2025).
Development of Structure-Function Relationship
Beyond single-modality investigations, increasing attention has focused on the developmental trajectories of structure-function relationship, typically assessed by structural-functional connectivity coupling (Baum et al., 2020; Feng et al., 2024; Hong et al., 2023; Wang et al., 2025). Previous studies have demonstrated significant correlations between whole-brain structural and functional connectivity profiles (Hagmann et al., 2008), which strengthen substantially with development (Hagmann et al., 2010; Uddin et al., 2011; van den Heuvel et al., 2015). More recent work has shifted from global analyses to regional assessments of structure-function coupling. The most common approach evaluates the spatial similarity between structural and functional connectivity profiles of brain regions using Pearson's or Spearman correlations (Baum et al., 2020). Alternative approaches quantify coupling by predicting functional connectivity patterns based on multiple communication models (e.g., Euclidean distance, shortest path, and communicability) applied to the white matter structural network (Feng et al., 2024; Vázquez-Rodríguez et al., 2019). The coupling strength is defined as the adjusted goodness-of-fit R2 of the linear regression model, denoting how well the functional connectivity is predicted from structural features (Feng et al., 2024).
Accumulating evidence indicates that structure-function coupling exhibits similar spatial patterns from preterm and term infants through toddlers to children, resembling adult-like configurations. Across developmental stages, stronger coupling is generally observed in unimodal cortices (e.g., visual and somatomotor areas), supporting efficient sensory processing, whereas weaker coupling predominates in transmodal cortices, potentially facilitating multimodal integration and cognitive flexibility (Fotiadis et al., 2024; Vázquez-Rodríguez et al., 2019). Nevertheless, coupling strength changes with age in a spatially heterogeneous manner, producing subtle spatial reconfigurations. During the prenatal period, longitudinal changes occur primarily in the bilateral visual, prefrontal, and temporal cortices, as well as the bilateral cingulate gyrus and the left sensorimotor cortex (Wang et al., 2025). After birth, infants aged one month exhibit increased coupling in auditory, lateral prefrontal and inferior parietal cortices (Tooley et al., 2025). By early childhood, frontal, medial parietal, and occipital regions exhibit relatively strong coupling, while lateral temporal and parietal regions show weaker coupling (Hong et al., 2023). From childhood to adolescent, age-related increases in coupling are evident in the frontoparietal, dorsal attention, default-mode networks (Feng et al., 2024), as well as temporoparietal junction, and prefrontal cortex (Baum et al., 2020), whereas decreases are mainly observed in the visual, motor, and insular cortices (Baum et al., 2020). Notably, the spatial pattern of structure-function coupling and its developmental changes mirror evolutionary expansion and the principal gradient of functional networks (Baum et al., 2020; Feng et al., 2024). These findings suggest that the heterogeneous organization of coupling is rooted in evolutionary processes and aligns with the macroscale functional hierarchy.
Computational modeling studies construct dynamic models to replicate large-scale brain activity, providing a novel approach to link local morphology and white matter connectivity to functional organization (Seguin et al., 2023a, 2023b). Recent studies further highlighted the potential role of excitation-inhibition balance in shaping structure-function coupling during the development (Saberi et al., 2025; Zhang et al., 2024). Using data from 1,000 participants across GUSTO and PNC cohorts and biophysically based cortical circuit models, Zhang et al. (2024) revealed that the excitation-inhibition ratio declines with age in a spatially heterogeneous way during youth, with larger reductions in sensorimotor systems relative to association areas. Similarly, Saberi et al. (2025) employed individualized biophysical network modeling based on rs-fMRI data from both cross-sectional (PNC) and longitudinal (IMAGEN) cohorts and observed age-related decreases in association areas, but reported different developmental changes (increase or lack of changes) in sensorimotor regions. Nevertheless, the extent to which local brain morphology and microbiological factors (e.g., neurotransmitters) contribute to the developing structure-function coupling remain poorly understood and warrant further investigation.
Relationship with Behavior and Genetics in Developmental Connectomics
Accumulating evidence suggests that the development of structural and functional brain networks may lay the foundation for later cognitive and behavioral outcomes, as shown by findings from both regression analyses and machine learning approaches. For example, regression analyses revealed that regional network efficiency at birth in the left temporal occipital fusiform and bilateral occipital fusiform gyri were positively associated with cognitive abilities at 28 months (Jiang et al., 2023). Similarly, SC-FC coupling within early reward network at 4.5 years is associated with executive function at ages 7 and 8.5 years (Chan et al., 2022). Extending these associations, machine learning models have enabled individualized prediction of developmental outcomes. For example, the functional connectivity strength, nodal efficiency, and modularity at birth predict cognitive and language scores at 18 months via support vector regression (Li et al., 2024b). Similarly, elastic-net regression models have revealed that SC-FC coupling predicts individual general intelligence, with particularly strong contributions from the frontoparietal and default-mode networks (Feng et al., 2024). Leveraging large-scale PNC cohort data, several studies have demonstrated that the maturation of structural and functional networks, particularly in term of segregation and integration, is tightly associated with cognitive development across childhood and adolescence (Keller et al., 2023). For example, the SC-FC coupling in the rostrolateral prefrontal cortex (Baum et al., 2020), modular segregation of white matter networks (Baum et al., 2017), and functional segregation within the sensorimotor and default-mode networks (Gu et al., 2015) have each been associated with individual executive performance. Moreover, inter-modular functional connectivity can predict general cognitive functioning (Pines et al., 2022), while individualized functional topography of association networks can predict individual differences in executive function (Cui et al., 2020). Collectively, these studies converge to suggest that the brain network reorganization supports the emergence of cognitive abilities and may serve as potential biomarkers for predicting cognitive development.
Beyond behavioral associations, brain networks development is shaped by complex interactions between genetic and environmental factors (Gao et al., 2019).Two complementary approaches have been used to study these interactions. The first approach combines neuroimaging data with individual genomics from large-scale cohorts (i.e., dHCP and GUSTO) to link genomic variation and prenatal environmental exposures to early brain morphology such as cortical thickness and volume. Prenatal maternal stress and healthy conditions may significantly influence neonatal brain development through the expression of specific genotypes. For example, the
The second approach integrates developmental connectomics with transcriptomics using datasets such as the Allen Human Brain Atlas (Arnatkevic˘iūtė et al., 2019) and the BrainSpan Developing Brain Atlas (Miller et al., 2014) to uncover the molecular underpinning of the developing connectome. For example, the functional connectome maturation in the first three years of life is associated with gene expression involved in microstructural development (e.g., dendrite development, synaptogenesis, neuron differentiation, neuron migration, axon development, and myelination) and energy metabolism (e.g., aerobic glycolysis and oxidative phosphorylation) (Li et al., 2024b). During childhood and adolescence, the progressive stabilization of functional module dynamics is linked to expression of genes related to ion transport and nucleobase-containing compound transport (Lei et al., 2022), while spatially heterogeneous development of SC-FC coupling correlates positively with oligodendrocyte-related gene expression and negatively correlated with astrocyte-related genes (Feng et al., 2024). Notably, heteromodal regions exhibit higher expression of genes related to gray matter maturation (e.g., synaptic and dendritic development) and lower expression of genes involved in white matter maturation (e.g., axon and myelin development) (Liang et al., 2024).
These integrative behavioral, genetic, and transcriptomic studies underscore the multiscale mechanisms through which early brain network architecture emerges to support developing cognition, which are also valuable for developing biologically grounded framework for early risk identification and individualized intervention. Building on this foundation, computational modeling of brain dynamics holds promise for constructing multiscale models that bridge genes, brain structure, neural dynamics, and behavior, thereby advancing our understanding of the mechanistic pathways linking biology to cognition.
Challenges and Future Directions
The rise of developmental connectomics big data, along with advances in data analysis and mining techniques, has created unprecedented opportunities to understand brain development. However, several critical challenges remain to be addressed.
First, acquiring high-quality imaging data in pediatric populations is challenging. The small head size of infants requires specialized head coils to enhance spatial resolution (Hughes et al., 2017). Moreover, young children often struggle to remain still during scanning, resulting in substantial motion artifacts and reduced data quality (Spann et al., 2023). These challenges can be alleviated by developing child-friendly imaging protocols, such as shorter scan duration, sequence optimization, and MRI-compatible entertainment, as well as applying improved motion correction strategies, including both prospective (real-time tracking) and retrospective (post-hoc correction) approaches. Beyond MRI, other imaging modalities provide complementary advantages, such as higher temporal resolutions (EEG/MEG) or more child-friendly designs (EEG/fNIRS). These features enable the capture of fine-grained temporal dynamics, naturalistic behaviors (Bagdasarov et al., 2025), and even multi-person interactions through hyper-scanning (Reindl et al., 2018). Integrating these modalities with MRI holds promise for a more comprehensive understanding of developmental connectomics.
Second, integrating multi-center datasets poses major challenges for data preprocessing and analysis, particularly when these datasets differ in demographics and neuroimaging protocols. Traditional manual quality control approaches lack scalability for large cohorts and multiple-center studies, underscoring the need for automatic, standardized, and multi-site compatible pipelines, potentially enhanced by supervised machine learning (Alfaro-Almagro et al., 2018). Moreover, preprocessing pipelines developed for adults cannot be directly applied to infants due to lower tissue contrast and rapid brain development (Turesky et al., 2021). Several pediatric-specific pipelines, such as iBEAT 2.0 (Wang et al., 2023) and Infant FreeSurfer (Zöllei et al., 2020), have been developed to enhance image segmentation and registration in early childhood. Even after preprocessing, site heterogeneity in demographics and scanning protocols complicates cross-site comparisons, which can be mitigated by using statistical harmonization methods like ComBat (Fortin et al., 2018) and deep learning-based approaches (Tian et al., 2022). Longitudinal studies face additional challenges, particularly participant attrition in long-term designs. These issues can be addressed by optimizing study designs (e.g., shorter sessions and flexible scheduling) and analytically through mixed-effects models, likelihood-based methods, and multiple-imputation (Matta et al., 2018; Telzer et al., 2018). Collectively, these challenges highlight the urgent need for unified computational platforms that integrate quality control, preprocessing, analysis, and statistical modeling across the lifespan.
Third, recent advances in artificial intelligence algorithms offer promising avenues for developmental connectomics, especially when leveraging large cohort datasets. For example, deep convolutional neural networks have been applied to improve infant brain image segmentation (Zhang et al., 2015), predict neurodevelopmental outcomes (Kawahara et al., 2017) and brain age (Chen et al., 2022), and facilitate early diagnosis of autism spectrum disorder (Uddin et al., 2024). Beyond supervised approaches, unsupervised deep learning methods may capture spatial heterogeneity in brain development by automatically extracting complex latent representations from large-scale, high-dimensional, and multimodal developmental data (Kazemivash & Calhoun, 2022; Li et al., 2023), thereby uncovering nonlinear and hierarchical patterns that are often inaccessible to traditional clustering techniques. While these methods have improved the characterization of developmental trajectories, our understanding regarding how brain networks emerge and evolve under biological constraints remains limited. Moving forward, research should prioritize generative and mechanistic modeling frameworks capable of simulating the formation and maturation of brain networks across developmental stages (Vértes, 2023), bridging the gap between descriptive analyses and mechanistic understanding.
Fourth, incorporating multi-center datasets, particularly those encompassing diverse populations, is essential for enhancing the generalizability of developmental connectomics findings (Zhang & Xue, 2025). Such integration, however, raises important considerations regarding across-site reproducibility. On one hand, including participants from different countries, ethnicities, and cultural backgrounds reduces biases inherent in homogeneous samples. On the other hand, assessing reproducibility across sites facilitates cross-cultural comparisons by revealing how sociocultural and environmental factors influence developmental trajectories, thus helping to identify both universal and population-specific aspects of brain development. Beyond increasing sample diversity, constructing large-scale, lifespan-spanning datasets is invaluable for establishing precise and robust normative developmental trajectories and for detecting individualized deviations, both critical for personalized diagnostic and targeted interventions (Bethlehem et al., 2022; Sun et al., 2025). When combined with genetic and environmental data, such datasets enable the construction of multilevel models that capture the complex interactions among factors such as genetics, socioeconomic status, education, nutrition, and early life experiences, elucidating their collective influence on the maturation of brain networks (Gao et al., 2019).
Supplemental Material
sj-pdf-1-pac-10.1177_18344909251392757 - Supplemental material for Developmental Connectomics from a “Big Data” Perspective
Supplemental material, sj-pdf-1-pac-10.1177_18344909251392757 for Developmental Connectomics from a “Big Data” Perspective by Yuehua Xu, Tengda Zhao, Mingrui Xia and Xuhong Liao in Journal of Pacific Rim Psychology
Footnotes
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the National Natural Science Foundation of China (Grant No. 82102131) and the Tang Scholar Award of Beijing Normal University.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Supplemental Material
Supplemental material for this article is available online.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
